projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: send: remove BUG_ON from name_cache_delete
[deliverable/linux.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"
37
38static int g_verbose = 0;
39
40#define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
41
42/*
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
48 */
49struct fs_path {
50 union {
51 struct {
52 char *start;
53 char *end;
31db9f7c
AB		 54
55 char *buf;
1f5a7ff9
DS		 56 unsigned short buf_len:15;
57 unsigned short reversed:1;
31db9f7c
AB		 58 char inline_buf[];
59 };
60 char pad[PAGE_SIZE];
61 };
62};
63#define FS_PATH_INLINE_SIZE \
64 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
65
66
67/* reused for each extent */
68struct clone_root {
69 struct btrfs_root *root;
70 u64 ino;
71 u64 offset;
72
73 u64 found_refs;
74};
75
76#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
77#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
78
79struct send_ctx {
80 struct file *send_filp;
81 loff_t send_off;
82 char *send_buf;
83 u32 send_size;
84 u32 send_max_size;
85 u64 total_send_size;
86 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
cb95e7bf 87 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
31db9f7c 88
31db9f7c
AB		 89 struct btrfs_root *send_root;
90 struct btrfs_root *parent_root;
91 struct clone_root *clone_roots;
92 int clone_roots_cnt;
93
94 /* current state of the compare_tree call */
95 struct btrfs_path *left_path;
96 struct btrfs_path *right_path;
97 struct btrfs_key *cmp_key;
98
99 /*
100 * infos of the currently processed inode. In case of deleted inodes,
101 * these are the values from the deleted inode.
102 */
103 u64 cur_ino;
104 u64 cur_inode_gen;
105 int cur_inode_new;
106 int cur_inode_new_gen;
107 int cur_inode_deleted;
31db9f7c
AB		 108 u64 cur_inode_size;
109 u64 cur_inode_mode;
16e7549f 110 u64 cur_inode_last_extent;
31db9f7c
AB		 111
112 u64 send_progress;
113
114 struct list_head new_refs;
115 struct list_head deleted_refs;
116
117 struct radix_tree_root name_cache;
118 struct list_head name_cache_list;
119 int name_cache_size;
120
31db9f7c 121 char *read_buf;
9f03740a
FDBM		 122
123 /*
124 * We process inodes by their increasing order, so if before an
125 * incremental send we reverse the parent/child relationship of
126 * directories such that a directory with a lower inode number was
127 * the parent of a directory with a higher inode number, and the one
128 * becoming the new parent got renamed too, we can't rename/move the
129 * directory with lower inode number when we finish processing it - we
130 * must process the directory with higher inode number first, then
131 * rename/move it and then rename/move the directory with lower inode
132 * number. Example follows.
133 *
134 * Tree state when the first send was performed:
135 *
136 * .
137 * |-- a (ino 257)
138 * |-- b (ino 258)
139 * |
140 * |
141 * |-- c (ino 259)
142 * | |-- d (ino 260)
143 * |
144 * |-- c2 (ino 261)
145 *
146 * Tree state when the second (incremental) send is performed:
147 *
148 * .
149 * |-- a (ino 257)
150 * |-- b (ino 258)
151 * |-- c2 (ino 261)
152 * |-- d2 (ino 260)
153 * |-- cc (ino 259)
154 *
155 * The sequence of steps that lead to the second state was:
156 *
157 * mv /a/b/c/d /a/b/c2/d2
158 * mv /a/b/c /a/b/c2/d2/cc
159 *
160 * "c" has lower inode number, but we can't move it (2nd mv operation)
161 * before we move "d", which has higher inode number.
162 *
163 * So we just memorize which move/rename operations must be performed
164 * later when their respective parent is processed and moved/renamed.
165 */
166
167 /* Indexed by parent directory inode number. */
168 struct rb_root pending_dir_moves;
169
170 /*
171 * Reverse index, indexed by the inode number of a directory that
172 * is waiting for the move/rename of its immediate parent before its
173 * own move/rename can be performed.
174 */
175 struct rb_root waiting_dir_moves;
176};
177
178struct pending_dir_move {
179 struct rb_node node;
180 struct list_head list;
181 u64 parent_ino;
182 u64 ino;
183 u64 gen;
184 struct list_head update_refs;
185};
186
187struct waiting_dir_move {
188 struct rb_node node;
189 u64 ino;
31db9f7c
AB		 190};
191
192struct name_cache_entry {
193 struct list_head list;
7e0926fe
AB		 194 /*
195 * radix_tree has only 32bit entries but we need to handle 64bit inums.
196 * We use the lower 32bit of the 64bit inum to store it in the tree. If
197 * more then one inum would fall into the same entry, we use radix_list
198 * to store the additional entries. radix_list is also used to store
199 * entries where two entries have the same inum but different
200 * generations.
201 */
202 struct list_head radix_list;
31db9f7c
AB		 203 u64 ino;
204 u64 gen;
205 u64 parent_ino;
206 u64 parent_gen;
207 int ret;
208 int need_later_update;
209 int name_len;
210 char name[];
211};
212
9f03740a
FDBM		 213static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
214
16e7549f
JB		 215static int need_send_hole(struct send_ctx *sctx)
216{
217 return (sctx->parent_root && !sctx->cur_inode_new &&
218 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
219 S_ISREG(sctx->cur_inode_mode));
220}
221
31db9f7c
AB		 222static void fs_path_reset(struct fs_path *p)
223{
224 if (p->reversed) {
225 p->start = p->buf + p->buf_len - 1;
226 p->end = p->start;
227 *p->start = 0;
228 } else {
229 p->start = p->buf;
230 p->end = p->start;
231 *p->start = 0;
232 }
233}
234
924794c9 235static struct fs_path *fs_path_alloc(void)
31db9f7c
AB		 236{
237 struct fs_path *p;
238
239 p = kmalloc(sizeof(*p), GFP_NOFS);
240 if (!p)
241 return NULL;
242 p->reversed = 0;
31db9f7c
AB		 243 p->buf = p->inline_buf;
244 p->buf_len = FS_PATH_INLINE_SIZE;
245 fs_path_reset(p);
246 return p;
247}
248
924794c9 249static struct fs_path *fs_path_alloc_reversed(void)
31db9f7c
AB		 250{
251 struct fs_path *p;
252
924794c9 253 p = fs_path_alloc();
31db9f7c
AB		 254 if (!p)
255 return NULL;
256 p->reversed = 1;
257 fs_path_reset(p);
258 return p;
259}
260
924794c9 261static void fs_path_free(struct fs_path *p)
31db9f7c
AB		 262{
263 if (!p)
264 return;
265 if (p->buf != p->inline_buf) {
e25a8122 266 if (is_vmalloc_addr(p->buf))
31db9f7c
AB		 267 vfree(p->buf);
268 else
269 kfree(p->buf);
270 }
271 kfree(p);
272}
273
274static int fs_path_len(struct fs_path *p)
275{
276 return p->end - p->start;
277}
278
279static int fs_path_ensure_buf(struct fs_path *p, int len)
280{
281 char *tmp_buf;
282 int path_len;
283 int old_buf_len;
284
285 len++;
286
287 if (p->buf_len >= len)
288 return 0;
289
290 path_len = p->end - p->start;
291 old_buf_len = p->buf_len;
292 len = PAGE_ALIGN(len);
293
294 if (p->buf == p->inline_buf) {
8be04b93 295 tmp_buf = kmalloc(len, GFP_NOFS | __GFP_NOWARN);
31db9f7c
AB		 296 if (!tmp_buf) {
297 tmp_buf = vmalloc(len);
298 if (!tmp_buf)
299 return -ENOMEM;
31db9f7c
AB		 300 }
301 memcpy(tmp_buf, p->buf, p->buf_len);
302 p->buf = tmp_buf;
303 p->buf_len = len;
304 } else {
e25a8122 305 if (is_vmalloc_addr(p->buf)) {
31db9f7c
AB		 306 tmp_buf = vmalloc(len);
307 if (!tmp_buf)
308 return -ENOMEM;
309 memcpy(tmp_buf, p->buf, p->buf_len);
310 vfree(p->buf);
311 } else {
312 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
313 if (!tmp_buf) {
314 tmp_buf = vmalloc(len);
315 if (!tmp_buf)
316 return -ENOMEM;
317 memcpy(tmp_buf, p->buf, p->buf_len);
318 kfree(p->buf);
31db9f7c
AB		 319 }
320 }
321 p->buf = tmp_buf;
322 p->buf_len = len;
323 }
324 if (p->reversed) {
325 tmp_buf = p->buf + old_buf_len - path_len - 1;
326 p->end = p->buf + p->buf_len - 1;
327 p->start = p->end - path_len;
328 memmove(p->start, tmp_buf, path_len + 1);
329 } else {
330 p->start = p->buf;
331 p->end = p->start + path_len;
332 }
333 return 0;
334}
335
b23ab57d
DS		 336static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
337 char **prepared)
31db9f7c
AB		 338{
339 int ret;
340 int new_len;
341
342 new_len = p->end - p->start + name_len;
343 if (p->start != p->end)
344 new_len++;
345 ret = fs_path_ensure_buf(p, new_len);
346 if (ret < 0)
347 goto out;
348
349 if (p->reversed) {
350 if (p->start != p->end)
351 *--p->start = '/';
352 p->start -= name_len;
b23ab57d 353 *prepared = p->start;
31db9f7c
AB		 354 } else {
355 if (p->start != p->end)
356 *p->end++ = '/';
b23ab57d 357 *prepared = p->end;
31db9f7c
AB		 358 p->end += name_len;
359 *p->end = 0;
360 }
361
362out:
363 return ret;
364}
365
366static int fs_path_add(struct fs_path *p, const char *name, int name_len)
367{
368 int ret;
b23ab57d 369 char *prepared;
31db9f7c 370
b23ab57d 371 ret = fs_path_prepare_for_add(p, name_len, &prepared);
31db9f7c
AB		 372 if (ret < 0)
373 goto out;
b23ab57d 374 memcpy(prepared, name, name_len);
31db9f7c
AB		 375
376out:
377 return ret;
378}
379
380static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
381{
382 int ret;
b23ab57d 383 char *prepared;
31db9f7c 384
b23ab57d 385 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
31db9f7c
AB		 386 if (ret < 0)
387 goto out;
b23ab57d 388 memcpy(prepared, p2->start, p2->end - p2->start);
31db9f7c
AB		 389
390out:
391 return ret;
392}
393
394static int fs_path_add_from_extent_buffer(struct fs_path *p,
395 struct extent_buffer *eb,
396 unsigned long off, int len)
397{
398 int ret;
b23ab57d 399 char *prepared;
31db9f7c 400
b23ab57d 401 ret = fs_path_prepare_for_add(p, len, &prepared);
31db9f7c
AB		 402 if (ret < 0)
403 goto out;
404
b23ab57d 405 read_extent_buffer(eb, prepared, off, len);
31db9f7c
AB		 406
407out:
408 return ret;
409}
410
31db9f7c
AB		 411static int fs_path_copy(struct fs_path *p, struct fs_path *from)
412{
413 int ret;
414
415 p->reversed = from->reversed;
416 fs_path_reset(p);
417
418 ret = fs_path_add_path(p, from);
419
420 return ret;
421}
422
423
424static void fs_path_unreverse(struct fs_path *p)
425{
426 char *tmp;
427 int len;
428
429 if (!p->reversed)
430 return;
431
432 tmp = p->start;
433 len = p->end - p->start;
434 p->start = p->buf;
435 p->end = p->start + len;
436 memmove(p->start, tmp, len + 1);
437 p->reversed = 0;
438}
439
440static struct btrfs_path *alloc_path_for_send(void)
441{
442 struct btrfs_path *path;
443
444 path = btrfs_alloc_path();
445 if (!path)
446 return NULL;
447 path->search_commit_root = 1;
448 path->skip_locking = 1;
449 return path;
450}
451
48a3b636 452static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
31db9f7c
AB		 453{
454 int ret;
455 mm_segment_t old_fs;
456 u32 pos = 0;
457
458 old_fs = get_fs();
459 set_fs(KERNEL_DS);
460
461 while (pos < len) {
1bcea355 462 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
31db9f7c
AB		 463 /* TODO handle that correctly */
464 /*if (ret == -ERESTARTSYS) {
465 continue;
466 }*/
467 if (ret < 0)
468 goto out;
469 if (ret == 0) {
470 ret = -EIO;
471 goto out;
472 }
473 pos += ret;
474 }
475
476 ret = 0;
477
478out:
479 set_fs(old_fs);
480 return ret;
481}
482
483static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
484{
485 struct btrfs_tlv_header *hdr;
486 int total_len = sizeof(*hdr) + len;
487 int left = sctx->send_max_size - sctx->send_size;
488
489 if (unlikely(left < total_len))
490 return -EOVERFLOW;
491
492 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
493 hdr->tlv_type = cpu_to_le16(attr);
494 hdr->tlv_len = cpu_to_le16(len);
495 memcpy(hdr + 1, data, len);
496 sctx->send_size += total_len;
497
498 return 0;
499}
500
95bc79d5
DS		 501#define TLV_PUT_DEFINE_INT(bits) \
502 static int tlv_put_u##bits(struct send_ctx *sctx, \
503 u##bits attr, u##bits value) \
504 { \
505 __le##bits __tmp = cpu_to_le##bits(value); \
506 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
507 }
31db9f7c 508
95bc79d5 509TLV_PUT_DEFINE_INT(64)
31db9f7c
AB		 510
511static int tlv_put_string(struct send_ctx *sctx, u16 attr,
512 const char *str, int len)
513{
514 if (len == -1)
515 len = strlen(str);
516 return tlv_put(sctx, attr, str, len);
517}
518
519static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
520 const u8 *uuid)
521{
522 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
523}
524
31db9f7c
AB		 525static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
526 struct extent_buffer *eb,
527 struct btrfs_timespec *ts)
528{
529 struct btrfs_timespec bts;
530 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
531 return tlv_put(sctx, attr, &bts, sizeof(bts));
532}
533
534
535#define TLV_PUT(sctx, attrtype, attrlen, data) \
536 do { \
537 ret = tlv_put(sctx, attrtype, attrlen, data); \
538 if (ret < 0) \
539 goto tlv_put_failure; \
540 } while (0)
541
542#define TLV_PUT_INT(sctx, attrtype, bits, value) \
543 do { \
544 ret = tlv_put_u##bits(sctx, attrtype, value); \
545 if (ret < 0) \
546 goto tlv_put_failure; \
547 } while (0)
548
549#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
550#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
551#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
552#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
553#define TLV_PUT_STRING(sctx, attrtype, str, len) \
554 do { \
555 ret = tlv_put_string(sctx, attrtype, str, len); \
556 if (ret < 0) \
557 goto tlv_put_failure; \
558 } while (0)
559#define TLV_PUT_PATH(sctx, attrtype, p) \
560 do { \
561 ret = tlv_put_string(sctx, attrtype, p->start, \
562 p->end - p->start); \
563 if (ret < 0) \
564 goto tlv_put_failure; \
565 } while(0)
566#define TLV_PUT_UUID(sctx, attrtype, uuid) \
567 do { \
568 ret = tlv_put_uuid(sctx, attrtype, uuid); \
569 if (ret < 0) \
570 goto tlv_put_failure; \
571 } while (0)
31db9f7c
AB		 572#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
573 do { \
574 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
575 if (ret < 0) \
576 goto tlv_put_failure; \
577 } while (0)
578
579static int send_header(struct send_ctx *sctx)
580{
581 struct btrfs_stream_header hdr;
582
583 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
584 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
585
1bcea355
AJ		 586 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
587 &sctx->send_off);
31db9f7c
AB		 588}
589
590/*
591 * For each command/item we want to send to userspace, we call this function.
592 */
593static int begin_cmd(struct send_ctx *sctx, int cmd)
594{
595 struct btrfs_cmd_header *hdr;
596
fae7f21c 597 if (WARN_ON(!sctx->send_buf))
31db9f7c 598 return -EINVAL;
31db9f7c
AB		 599
600 BUG_ON(sctx->send_size);
601
602 sctx->send_size += sizeof(*hdr);
603 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
604 hdr->cmd = cpu_to_le16(cmd);
605
606 return 0;
607}
608
609static int send_cmd(struct send_ctx *sctx)
610{
611 int ret;
612 struct btrfs_cmd_header *hdr;
613 u32 crc;
614
615 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
616 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
617 hdr->crc = 0;
618
0b947aff 619 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
31db9f7c
AB		 620 hdr->crc = cpu_to_le32(crc);
621
1bcea355
AJ		 622 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
623 &sctx->send_off);
31db9f7c
AB		 624
625 sctx->total_send_size += sctx->send_size;
626 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
627 sctx->send_size = 0;
628
629 return ret;
630}
631
632/*
633 * Sends a move instruction to user space
634 */
635static int send_rename(struct send_ctx *sctx,
636 struct fs_path *from, struct fs_path *to)
637{
638 int ret;
639
640verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
641
642 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
643 if (ret < 0)
644 goto out;
645
646 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
647 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
648
649 ret = send_cmd(sctx);
650
651tlv_put_failure:
652out:
653 return ret;
654}
655
656/*
657 * Sends a link instruction to user space
658 */
659static int send_link(struct send_ctx *sctx,
660 struct fs_path *path, struct fs_path *lnk)
661{
662 int ret;
663
664verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
665
666 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
667 if (ret < 0)
668 goto out;
669
670 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
671 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
672
673 ret = send_cmd(sctx);
674
675tlv_put_failure:
676out:
677 return ret;
678}
679
680/*
681 * Sends an unlink instruction to user space
682 */
683static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
684{
685 int ret;
686
687verbose_printk("btrfs: send_unlink %s\n", path->start);
688
689 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
690 if (ret < 0)
691 goto out;
692
693 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
694
695 ret = send_cmd(sctx);
696
697tlv_put_failure:
698out:
699 return ret;
700}
701
702/*
703 * Sends a rmdir instruction to user space
704 */
705static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
706{
707 int ret;
708
709verbose_printk("btrfs: send_rmdir %s\n", path->start);
710
711 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
712 if (ret < 0)
713 goto out;
714
715 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
716
717 ret = send_cmd(sctx);
718
719tlv_put_failure:
720out:
721 return ret;
722}
723
724/*
725 * Helper function to retrieve some fields from an inode item.
726 */
727static int get_inode_info(struct btrfs_root *root,
728 u64 ino, u64 *size, u64 *gen,
85a7b33b
AB		 729 u64 *mode, u64 *uid, u64 *gid,
730 u64 *rdev)
31db9f7c
AB		 731{
732 int ret;
733 struct btrfs_inode_item *ii;
734 struct btrfs_key key;
735 struct btrfs_path *path;
736
737 path = alloc_path_for_send();
738 if (!path)
739 return -ENOMEM;
740
741 key.objectid = ino;
742 key.type = BTRFS_INODE_ITEM_KEY;
743 key.offset = 0;
744 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
745 if (ret < 0)
746 goto out;
747 if (ret) {
748 ret = -ENOENT;
749 goto out;
750 }
751
752 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
753 struct btrfs_inode_item);
754 if (size)
755 *size = btrfs_inode_size(path->nodes[0], ii);
756 if (gen)
757 *gen = btrfs_inode_generation(path->nodes[0], ii);
758 if (mode)
759 *mode = btrfs_inode_mode(path->nodes[0], ii);
760 if (uid)
761 *uid = btrfs_inode_uid(path->nodes[0], ii);
762 if (gid)
763 *gid = btrfs_inode_gid(path->nodes[0], ii);
85a7b33b
AB		 764 if (rdev)
765 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
31db9f7c
AB		 766
767out:
768 btrfs_free_path(path);
769 return ret;
770}
771
772typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
773 struct fs_path *p,
774 void *ctx);
775
776/*
96b5bd77
JS		 777 * Helper function to iterate the entries in ONE btrfs_inode_ref or
778 * btrfs_inode_extref.
31db9f7c
AB		 779 * The iterate callback may return a non zero value to stop iteration. This can
780 * be a negative value for error codes or 1 to simply stop it.
781 *
96b5bd77 782 * path must point to the INODE_REF or INODE_EXTREF when called.
31db9f7c 783 */
924794c9 784static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 785 struct btrfs_key *found_key, int resolve,
786 iterate_inode_ref_t iterate, void *ctx)
787{
96b5bd77 788 struct extent_buffer *eb = path->nodes[0];
31db9f7c
AB		 789 struct btrfs_item *item;
790 struct btrfs_inode_ref *iref;
96b5bd77 791 struct btrfs_inode_extref *extref;
31db9f7c
AB		 792 struct btrfs_path *tmp_path;
793 struct fs_path *p;
96b5bd77 794 u32 cur = 0;
31db9f7c 795 u32 total;
96b5bd77 796 int slot = path->slots[0];
31db9f7c
AB		 797 u32 name_len;
798 char *start;
799 int ret = 0;
96b5bd77 800 int num = 0;
31db9f7c 801 int index;
96b5bd77
JS		 802 u64 dir;
803 unsigned long name_off;
804 unsigned long elem_size;
805 unsigned long ptr;
31db9f7c 806
924794c9 807 p = fs_path_alloc_reversed();
31db9f7c
AB		 808 if (!p)
809 return -ENOMEM;
810
811 tmp_path = alloc_path_for_send();
812 if (!tmp_path) {
924794c9 813 fs_path_free(p);
31db9f7c
AB		 814 return -ENOMEM;
815 }
816
31db9f7c 817
96b5bd77
JS		 818 if (found_key->type == BTRFS_INODE_REF_KEY) {
819 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
820 struct btrfs_inode_ref);
dd3cc16b 821 item = btrfs_item_nr(slot);
96b5bd77
JS		 822 total = btrfs_item_size(eb, item);
823 elem_size = sizeof(*iref);
824 } else {
825 ptr = btrfs_item_ptr_offset(eb, slot);
826 total = btrfs_item_size_nr(eb, slot);
827 elem_size = sizeof(*extref);
828 }
829
31db9f7c
AB		 830 while (cur < total) {
831 fs_path_reset(p);
832
96b5bd77
JS		 833 if (found_key->type == BTRFS_INODE_REF_KEY) {
834 iref = (struct btrfs_inode_ref *)(ptr + cur);
835 name_len = btrfs_inode_ref_name_len(eb, iref);
836 name_off = (unsigned long)(iref + 1);
837 index = btrfs_inode_ref_index(eb, iref);
838 dir = found_key->offset;
839 } else {
840 extref = (struct btrfs_inode_extref *)(ptr + cur);
841 name_len = btrfs_inode_extref_name_len(eb, extref);
842 name_off = (unsigned long)&extref->name;
843 index = btrfs_inode_extref_index(eb, extref);
844 dir = btrfs_inode_extref_parent(eb, extref);
845 }
846
31db9f7c 847 if (resolve) {
96b5bd77
JS		 848 start = btrfs_ref_to_path(root, tmp_path, name_len,
849 name_off, eb, dir,
850 p->buf, p->buf_len);
31db9f7c
AB		 851 if (IS_ERR(start)) {
852 ret = PTR_ERR(start);
853 goto out;
854 }
855 if (start < p->buf) {
856 /* overflow , try again with larger buffer */
857 ret = fs_path_ensure_buf(p,
858 p->buf_len + p->buf - start);
859 if (ret < 0)
860 goto out;
96b5bd77
JS		 861 start = btrfs_ref_to_path(root, tmp_path,
862 name_len, name_off,
863 eb, dir,
864 p->buf, p->buf_len);
31db9f7c
AB		 865 if (IS_ERR(start)) {
866 ret = PTR_ERR(start);
867 goto out;
868 }
869 BUG_ON(start < p->buf);
870 }
871 p->start = start;
872 } else {
96b5bd77
JS		 873 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
874 name_len);
31db9f7c
AB		 875 if (ret < 0)
876 goto out;
877 }
878
96b5bd77
JS		 879 cur += elem_size + name_len;
880 ret = iterate(num, dir, index, p, ctx);
31db9f7c
AB		 881 if (ret)
882 goto out;
31db9f7c
AB		 883 num++;
884 }
885
886out:
887 btrfs_free_path(tmp_path);
924794c9 888 fs_path_free(p);
31db9f7c
AB		 889 return ret;
890}
891
892typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
893 const char *name, int name_len,
894 const char *data, int data_len,
895 u8 type, void *ctx);
896
897/*
898 * Helper function to iterate the entries in ONE btrfs_dir_item.
899 * The iterate callback may return a non zero value to stop iteration. This can
900 * be a negative value for error codes or 1 to simply stop it.
901 *
902 * path must point to the dir item when called.
903 */
924794c9 904static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 905 struct btrfs_key *found_key,
906 iterate_dir_item_t iterate, void *ctx)
907{
908 int ret = 0;
909 struct extent_buffer *eb;
910 struct btrfs_item *item;
911 struct btrfs_dir_item *di;
31db9f7c
AB		 912 struct btrfs_key di_key;
913 char *buf = NULL;
914 char *buf2 = NULL;
915 int buf_len;
916 int buf_virtual = 0;
917 u32 name_len;
918 u32 data_len;
919 u32 cur;
920 u32 len;
921 u32 total;
922 int slot;
923 int num;
924 u8 type;
925
926 buf_len = PAGE_SIZE;
927 buf = kmalloc(buf_len, GFP_NOFS);
928 if (!buf) {
929 ret = -ENOMEM;
930 goto out;
931 }
932
31db9f7c
AB		 933 eb = path->nodes[0];
934 slot = path->slots[0];
dd3cc16b 935 item = btrfs_item_nr(slot);
31db9f7c
AB		 936 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
937 cur = 0;
938 len = 0;
939 total = btrfs_item_size(eb, item);
940
941 num = 0;
942 while (cur < total) {
943 name_len = btrfs_dir_name_len(eb, di);
944 data_len = btrfs_dir_data_len(eb, di);
945 type = btrfs_dir_type(eb, di);
946 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
947
948 if (name_len + data_len > buf_len) {
949 buf_len = PAGE_ALIGN(name_len + data_len);
950 if (buf_virtual) {
951 buf2 = vmalloc(buf_len);
952 if (!buf2) {
953 ret = -ENOMEM;
954 goto out;
955 }
956 vfree(buf);
957 } else {
958 buf2 = krealloc(buf, buf_len, GFP_NOFS);
959 if (!buf2) {
960 buf2 = vmalloc(buf_len);
961 if (!buf2) {
962 ret = -ENOMEM;
963 goto out;
964 }
965 kfree(buf);
966 buf_virtual = 1;
967 }
968 }
969
970 buf = buf2;
971 buf2 = NULL;
972 }
973
974 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
975 name_len + data_len);
976
977 len = sizeof(*di) + name_len + data_len;
978 di = (struct btrfs_dir_item *)((char *)di + len);
979 cur += len;
980
981 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
982 data_len, type, ctx);
983 if (ret < 0)
984 goto out;
985 if (ret) {
986 ret = 0;
987 goto out;
988 }
989
990 num++;
991 }
992
993out:
31db9f7c
AB		 994 if (buf_virtual)
995 vfree(buf);
996 else
997 kfree(buf);
998 return ret;
999}
1000
1001static int __copy_first_ref(int num, u64 dir, int index,
1002 struct fs_path *p, void *ctx)
1003{
1004 int ret;
1005 struct fs_path *pt = ctx;
1006
1007 ret = fs_path_copy(pt, p);
1008 if (ret < 0)
1009 return ret;
1010
1011 /* we want the first only */
1012 return 1;
1013}
1014
1015/*
1016 * Retrieve the first path of an inode. If an inode has more then one
1017 * ref/hardlink, this is ignored.
1018 */
924794c9 1019static int get_inode_path(struct btrfs_root *root,
31db9f7c
AB		 1020 u64 ino, struct fs_path *path)
1021{
1022 int ret;
1023 struct btrfs_key key, found_key;
1024 struct btrfs_path *p;
1025
1026 p = alloc_path_for_send();
1027 if (!p)
1028 return -ENOMEM;
1029
1030 fs_path_reset(path);
1031
1032 key.objectid = ino;
1033 key.type = BTRFS_INODE_REF_KEY;
1034 key.offset = 0;
1035
1036 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1037 if (ret < 0)
1038 goto out;
1039 if (ret) {
1040 ret = 1;
1041 goto out;
1042 }
1043 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1044 if (found_key.objectid != ino ||
96b5bd77
JS		 1045 (found_key.type != BTRFS_INODE_REF_KEY &&
1046 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1047 ret = -ENOENT;
1048 goto out;
1049 }
1050
924794c9
TI		 1051 ret = iterate_inode_ref(root, p, &found_key, 1,
1052 __copy_first_ref, path);
31db9f7c
AB		 1053 if (ret < 0)
1054 goto out;
1055 ret = 0;
1056
1057out:
1058 btrfs_free_path(p);
1059 return ret;
1060}
1061
1062struct backref_ctx {
1063 struct send_ctx *sctx;
1064
1065 /* number of total found references */
1066 u64 found;
1067
1068 /*
1069 * used for clones found in send_root. clones found behind cur_objectid
1070 * and cur_offset are not considered as allowed clones.
1071 */
1072 u64 cur_objectid;
1073 u64 cur_offset;
1074
1075 /* may be truncated in case it's the last extent in a file */
1076 u64 extent_len;
1077
1078 /* Just to check for bugs in backref resolving */
ee849c04 1079 int found_itself;
31db9f7c
AB		 1080};
1081
1082static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1083{
995e01b7 1084 u64 root = (u64)(uintptr_t)key;
31db9f7c
AB		 1085 struct clone_root *cr = (struct clone_root *)elt;
1086
1087 if (root < cr->root->objectid)
1088 return -1;
1089 if (root > cr->root->objectid)
1090 return 1;
1091 return 0;
1092}
1093
1094static int __clone_root_cmp_sort(const void *e1, const void *e2)
1095{
1096 struct clone_root *cr1 = (struct clone_root *)e1;
1097 struct clone_root *cr2 = (struct clone_root *)e2;
1098
1099 if (cr1->root->objectid < cr2->root->objectid)
1100 return -1;
1101 if (cr1->root->objectid > cr2->root->objectid)
1102 return 1;
1103 return 0;
1104}
1105
1106/*
1107 * Called for every backref that is found for the current extent.
766702ef 1108 * Results are collected in sctx->clone_roots->ino/offset/found_refs
31db9f7c
AB		 1109 */
1110static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1111{
1112 struct backref_ctx *bctx = ctx_;
1113 struct clone_root *found;
1114 int ret;
1115 u64 i_size;
1116
1117 /* First check if the root is in the list of accepted clone sources */
995e01b7 1118 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
31db9f7c
AB		 1119 bctx->sctx->clone_roots_cnt,
1120 sizeof(struct clone_root),
1121 __clone_root_cmp_bsearch);
1122 if (!found)
1123 return 0;
1124
1125 if (found->root == bctx->sctx->send_root &&
1126 ino == bctx->cur_objectid &&
1127 offset == bctx->cur_offset) {
ee849c04 1128 bctx->found_itself = 1;
31db9f7c
AB		 1129 }
1130
1131 /*
766702ef 1132 * There are inodes that have extents that lie behind its i_size. Don't
31db9f7c
AB		 1133 * accept clones from these extents.
1134 */
85a7b33b
AB		 1135 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1136 NULL);
31db9f7c
AB		 1137 if (ret < 0)
1138 return ret;
1139
1140 if (offset + bctx->extent_len > i_size)
1141 return 0;
1142
1143 /*
1144 * Make sure we don't consider clones from send_root that are
1145 * behind the current inode/offset.
1146 */
1147 if (found->root == bctx->sctx->send_root) {
1148 /*
1149 * TODO for the moment we don't accept clones from the inode
1150 * that is currently send. We may change this when
1151 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1152 * file.
1153 */
1154 if (ino >= bctx->cur_objectid)
1155 return 0;
e938c8ad
AB		 1156#if 0
1157 if (ino > bctx->cur_objectid)
1158 return 0;
1159 if (offset + bctx->extent_len > bctx->cur_offset)
31db9f7c 1160 return 0;
e938c8ad 1161#endif
31db9f7c
AB		 1162 }
1163
1164 bctx->found++;
1165 found->found_refs++;
1166 if (ino < found->ino) {
1167 found->ino = ino;
1168 found->offset = offset;
1169 } else if (found->ino == ino) {
1170 /*
1171 * same extent found more then once in the same file.
1172 */
1173 if (found->offset > offset + bctx->extent_len)
1174 found->offset = offset;
1175 }
1176
1177 return 0;
1178}
1179
1180/*
766702ef
AB		 1181 * Given an inode, offset and extent item, it finds a good clone for a clone
1182 * instruction. Returns -ENOENT when none could be found. The function makes
1183 * sure that the returned clone is usable at the point where sending is at the
1184 * moment. This means, that no clones are accepted which lie behind the current
1185 * inode+offset.
1186 *
31db9f7c
AB		 1187 * path must point to the extent item when called.
1188 */
1189static int find_extent_clone(struct send_ctx *sctx,
1190 struct btrfs_path *path,
1191 u64 ino, u64 data_offset,
1192 u64 ino_size,
1193 struct clone_root **found)
1194{
1195 int ret;
1196 int extent_type;
1197 u64 logical;
74dd17fb 1198 u64 disk_byte;
31db9f7c
AB		 1199 u64 num_bytes;
1200 u64 extent_item_pos;
69917e43 1201 u64 flags = 0;
31db9f7c
AB		 1202 struct btrfs_file_extent_item *fi;
1203 struct extent_buffer *eb = path->nodes[0];
35075bb0 1204 struct backref_ctx *backref_ctx = NULL;
31db9f7c
AB		 1205 struct clone_root *cur_clone_root;
1206 struct btrfs_key found_key;
1207 struct btrfs_path *tmp_path;
74dd17fb 1208 int compressed;
31db9f7c
AB		 1209 u32 i;
1210
1211 tmp_path = alloc_path_for_send();
1212 if (!tmp_path)
1213 return -ENOMEM;
1214
35075bb0
AB		 1215 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1216 if (!backref_ctx) {
1217 ret = -ENOMEM;
1218 goto out;
1219 }
1220
31db9f7c
AB		 1221 if (data_offset >= ino_size) {
1222 /*
1223 * There may be extents that lie behind the file's size.
1224 * I at least had this in combination with snapshotting while
1225 * writing large files.
1226 */
1227 ret = 0;
1228 goto out;
1229 }
1230
1231 fi = btrfs_item_ptr(eb, path->slots[0],
1232 struct btrfs_file_extent_item);
1233 extent_type = btrfs_file_extent_type(eb, fi);
1234 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1235 ret = -ENOENT;
1236 goto out;
1237 }
74dd17fb 1238 compressed = btrfs_file_extent_compression(eb, fi);
31db9f7c
AB		 1239
1240 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
74dd17fb
CM		 1241 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1242 if (disk_byte == 0) {
31db9f7c
AB		 1243 ret = -ENOENT;
1244 goto out;
1245 }
74dd17fb 1246 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
31db9f7c 1247
69917e43
LB		 1248 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1249 &found_key, &flags);
31db9f7c
AB		 1250 btrfs_release_path(tmp_path);
1251
1252 if (ret < 0)
1253 goto out;
69917e43 1254 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
31db9f7c
AB		 1255 ret = -EIO;
1256 goto out;
1257 }
1258
1259 /*
1260 * Setup the clone roots.
1261 */
1262 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1263 cur_clone_root = sctx->clone_roots + i;
1264 cur_clone_root->ino = (u64)-1;
1265 cur_clone_root->offset = 0;
1266 cur_clone_root->found_refs = 0;
1267 }
1268
35075bb0
AB		 1269 backref_ctx->sctx = sctx;
1270 backref_ctx->found = 0;
1271 backref_ctx->cur_objectid = ino;
1272 backref_ctx->cur_offset = data_offset;
1273 backref_ctx->found_itself = 0;
1274 backref_ctx->extent_len = num_bytes;
31db9f7c
AB		 1275
1276 /*
1277 * The last extent of a file may be too large due to page alignment.
1278 * We need to adjust extent_len in this case so that the checks in
1279 * __iterate_backrefs work.
1280 */
1281 if (data_offset + num_bytes >= ino_size)
35075bb0 1282 backref_ctx->extent_len = ino_size - data_offset;
31db9f7c
AB		 1283
1284 /*
1285 * Now collect all backrefs.
1286 */
74dd17fb
CM		 1287 if (compressed == BTRFS_COMPRESS_NONE)
1288 extent_item_pos = logical - found_key.objectid;
1289 else
1290 extent_item_pos = 0;
1291
31db9f7c
AB		 1292 extent_item_pos = logical - found_key.objectid;
1293 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1294 found_key.objectid, extent_item_pos, 1,
35075bb0 1295 __iterate_backrefs, backref_ctx);
74dd17fb 1296
31db9f7c
AB		 1297 if (ret < 0)
1298 goto out;
1299
35075bb0 1300 if (!backref_ctx->found_itself) {
31db9f7c
AB		 1301 /* found a bug in backref code? */
1302 ret = -EIO;
efe120a0 1303 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
31db9f7c 1304 "send_root. inode=%llu, offset=%llu, "
74dd17fb
CM		 1305 "disk_byte=%llu found extent=%llu\n",
1306 ino, data_offset, disk_byte, found_key.objectid);
31db9f7c
AB		 1307 goto out;
1308 }
1309
1310verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1311 "ino=%llu, "
1312 "num_bytes=%llu, logical=%llu\n",
1313 data_offset, ino, num_bytes, logical);
1314
35075bb0 1315 if (!backref_ctx->found)
31db9f7c
AB		 1316 verbose_printk("btrfs: no clones found\n");
1317
1318 cur_clone_root = NULL;
1319 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1320 if (sctx->clone_roots[i].found_refs) {
1321 if (!cur_clone_root)
1322 cur_clone_root = sctx->clone_roots + i;
1323 else if (sctx->clone_roots[i].root == sctx->send_root)
1324 /* prefer clones from send_root over others */
1325 cur_clone_root = sctx->clone_roots + i;
31db9f7c
AB		 1326 }
1327
1328 }
1329
1330 if (cur_clone_root) {
93de4ba8
FDBM		 1331 if (compressed != BTRFS_COMPRESS_NONE) {
1332 /*
1333 * Offsets given by iterate_extent_inodes() are relative
1334 * to the start of the extent, we need to add logical
1335 * offset from the file extent item.
1336 * (See why at backref.c:check_extent_in_eb())
1337 */
1338 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1339 fi);
1340 }
31db9f7c
AB		 1341 *found = cur_clone_root;
1342 ret = 0;
1343 } else {
1344 ret = -ENOENT;
1345 }
1346
1347out:
1348 btrfs_free_path(tmp_path);
35075bb0 1349 kfree(backref_ctx);
31db9f7c
AB		 1350 return ret;
1351}
1352
924794c9 1353static int read_symlink(struct btrfs_root *root,
31db9f7c
AB		 1354 u64 ino,
1355 struct fs_path *dest)
1356{
1357 int ret;
1358 struct btrfs_path *path;
1359 struct btrfs_key key;
1360 struct btrfs_file_extent_item *ei;
1361 u8 type;
1362 u8 compression;
1363 unsigned long off;
1364 int len;
1365
1366 path = alloc_path_for_send();
1367 if (!path)
1368 return -ENOMEM;
1369
1370 key.objectid = ino;
1371 key.type = BTRFS_EXTENT_DATA_KEY;
1372 key.offset = 0;
1373 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1374 if (ret < 0)
1375 goto out;
1376 BUG_ON(ret);
1377
1378 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1379 struct btrfs_file_extent_item);
1380 type = btrfs_file_extent_type(path->nodes[0], ei);
1381 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1382 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1383 BUG_ON(compression);
1384
1385 off = btrfs_file_extent_inline_start(ei);
514ac8ad 1386 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
31db9f7c
AB		 1387
1388 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
31db9f7c
AB		 1389
1390out:
1391 btrfs_free_path(path);
1392 return ret;
1393}
1394
1395/*
1396 * Helper function to generate a file name that is unique in the root of
1397 * send_root and parent_root. This is used to generate names for orphan inodes.
1398 */
1399static int gen_unique_name(struct send_ctx *sctx,
1400 u64 ino, u64 gen,
1401 struct fs_path *dest)
1402{
1403 int ret = 0;
1404 struct btrfs_path *path;
1405 struct btrfs_dir_item *di;
1406 char tmp[64];
1407 int len;
1408 u64 idx = 0;
1409
1410 path = alloc_path_for_send();
1411 if (!path)
1412 return -ENOMEM;
1413
1414 while (1) {
f74b86d8 1415 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
31db9f7c 1416 ino, gen, idx);
64792f25 1417 ASSERT(len < sizeof(tmp));
31db9f7c
AB		 1418
1419 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1420 path, BTRFS_FIRST_FREE_OBJECTID,
1421 tmp, strlen(tmp), 0);
1422 btrfs_release_path(path);
1423 if (IS_ERR(di)) {
1424 ret = PTR_ERR(di);
1425 goto out;
1426 }
1427 if (di) {
1428 /* not unique, try again */
1429 idx++;
1430 continue;
1431 }
1432
1433 if (!sctx->parent_root) {
1434 /* unique */
1435 ret = 0;
1436 break;
1437 }
1438
1439 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1440 path, BTRFS_FIRST_FREE_OBJECTID,
1441 tmp, strlen(tmp), 0);
1442 btrfs_release_path(path);
1443 if (IS_ERR(di)) {
1444 ret = PTR_ERR(di);
1445 goto out;
1446 }
1447 if (di) {
1448 /* not unique, try again */
1449 idx++;
1450 continue;
1451 }
1452 /* unique */
1453 break;
1454 }
1455
1456 ret = fs_path_add(dest, tmp, strlen(tmp));
1457
1458out:
1459 btrfs_free_path(path);
1460 return ret;
1461}
1462
1463enum inode_state {
1464 inode_state_no_change,
1465 inode_state_will_create,
1466 inode_state_did_create,
1467 inode_state_will_delete,
1468 inode_state_did_delete,
1469};
1470
1471static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1472{
1473 int ret;
1474 int left_ret;
1475 int right_ret;
1476 u64 left_gen;
1477 u64 right_gen;
1478
1479 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
85a7b33b 1480 NULL, NULL);
31db9f7c
AB		 1481 if (ret < 0 && ret != -ENOENT)
1482 goto out;
1483 left_ret = ret;
1484
1485 if (!sctx->parent_root) {
1486 right_ret = -ENOENT;
1487 } else {
1488 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
85a7b33b 1489 NULL, NULL, NULL, NULL);
31db9f7c
AB		 1490 if (ret < 0 && ret != -ENOENT)
1491 goto out;
1492 right_ret = ret;
1493 }
1494
1495 if (!left_ret && !right_ret) {
e938c8ad 1496 if (left_gen == gen && right_gen == gen) {
31db9f7c 1497 ret = inode_state_no_change;
e938c8ad 1498 } else if (left_gen == gen) {
31db9f7c
AB		 1499 if (ino < sctx->send_progress)
1500 ret = inode_state_did_create;
1501 else
1502 ret = inode_state_will_create;
1503 } else if (right_gen == gen) {
1504 if (ino < sctx->send_progress)
1505 ret = inode_state_did_delete;
1506 else
1507 ret = inode_state_will_delete;
1508 } else {
1509 ret = -ENOENT;
1510 }
1511 } else if (!left_ret) {
1512 if (left_gen == gen) {
1513 if (ino < sctx->send_progress)
1514 ret = inode_state_did_create;
1515 else
1516 ret = inode_state_will_create;
1517 } else {
1518 ret = -ENOENT;
1519 }
1520 } else if (!right_ret) {
1521 if (right_gen == gen) {
1522 if (ino < sctx->send_progress)
1523 ret = inode_state_did_delete;
1524 else
1525 ret = inode_state_will_delete;
1526 } else {
1527 ret = -ENOENT;
1528 }
1529 } else {
1530 ret = -ENOENT;
1531 }
1532
1533out:
1534 return ret;
1535}
1536
1537static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1538{
1539 int ret;
1540
1541 ret = get_cur_inode_state(sctx, ino, gen);
1542 if (ret < 0)
1543 goto out;
1544
1545 if (ret == inode_state_no_change ||
1546 ret == inode_state_did_create ||
1547 ret == inode_state_will_delete)
1548 ret = 1;
1549 else
1550 ret = 0;
1551
1552out:
1553 return ret;
1554}
1555
1556/*
1557 * Helper function to lookup a dir item in a dir.
1558 */
1559static int lookup_dir_item_inode(struct btrfs_root *root,
1560 u64 dir, const char *name, int name_len,
1561 u64 *found_inode,
1562 u8 *found_type)
1563{
1564 int ret = 0;
1565 struct btrfs_dir_item *di;
1566 struct btrfs_key key;
1567 struct btrfs_path *path;
1568
1569 path = alloc_path_for_send();
1570 if (!path)
1571 return -ENOMEM;
1572
1573 di = btrfs_lookup_dir_item(NULL, root, path,
1574 dir, name, name_len, 0);
1575 if (!di) {
1576 ret = -ENOENT;
1577 goto out;
1578 }
1579 if (IS_ERR(di)) {
1580 ret = PTR_ERR(di);
1581 goto out;
1582 }
1583 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1584 *found_inode = key.objectid;
1585 *found_type = btrfs_dir_type(path->nodes[0], di);
1586
1587out:
1588 btrfs_free_path(path);
1589 return ret;
1590}
1591
766702ef
AB		 1592/*
1593 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1594 * generation of the parent dir and the name of the dir entry.
1595 */
924794c9 1596static int get_first_ref(struct btrfs_root *root, u64 ino,
31db9f7c
AB		 1597 u64 *dir, u64 *dir_gen, struct fs_path *name)
1598{
1599 int ret;
1600 struct btrfs_key key;
1601 struct btrfs_key found_key;
1602 struct btrfs_path *path;
31db9f7c 1603 int len;
96b5bd77 1604 u64 parent_dir;
31db9f7c
AB		 1605
1606 path = alloc_path_for_send();
1607 if (!path)
1608 return -ENOMEM;
1609
1610 key.objectid = ino;
1611 key.type = BTRFS_INODE_REF_KEY;
1612 key.offset = 0;
1613
1614 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1615 if (ret < 0)
1616 goto out;
1617 if (!ret)
1618 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1619 path->slots[0]);
96b5bd77
JS		 1620 if (ret || found_key.objectid != ino ||
1621 (found_key.type != BTRFS_INODE_REF_KEY &&
1622 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1623 ret = -ENOENT;
1624 goto out;
1625 }
1626
96b5bd77
JS		 1627 if (key.type == BTRFS_INODE_REF_KEY) {
1628 struct btrfs_inode_ref *iref;
1629 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1630 struct btrfs_inode_ref);
1631 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1632 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1633 (unsigned long)(iref + 1),
1634 len);
1635 parent_dir = found_key.offset;
1636 } else {
1637 struct btrfs_inode_extref *extref;
1638 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1639 struct btrfs_inode_extref);
1640 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1641 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1642 (unsigned long)&extref->name, len);
1643 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1644 }
31db9f7c
AB		 1645 if (ret < 0)
1646 goto out;
1647 btrfs_release_path(path);
1648
96b5bd77 1649 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
85a7b33b 1650 NULL, NULL);
31db9f7c
AB		 1651 if (ret < 0)
1652 goto out;
1653
96b5bd77 1654 *dir = parent_dir;
31db9f7c
AB		 1655
1656out:
1657 btrfs_free_path(path);
1658 return ret;
1659}
1660
924794c9 1661static int is_first_ref(struct btrfs_root *root,
31db9f7c
AB		 1662 u64 ino, u64 dir,
1663 const char *name, int name_len)
1664{
1665 int ret;
1666 struct fs_path *tmp_name;
1667 u64 tmp_dir;
1668 u64 tmp_dir_gen;
1669
924794c9 1670 tmp_name = fs_path_alloc();
31db9f7c
AB		 1671 if (!tmp_name)
1672 return -ENOMEM;
1673
924794c9 1674 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
31db9f7c
AB		 1675 if (ret < 0)
1676 goto out;
1677
b9291aff 1678 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
31db9f7c
AB		 1679 ret = 0;
1680 goto out;
1681 }
1682
e938c8ad 1683 ret = !memcmp(tmp_name->start, name, name_len);
31db9f7c
AB		 1684
1685out:
924794c9 1686 fs_path_free(tmp_name);
31db9f7c
AB		 1687 return ret;
1688}
1689
766702ef
AB		 1690/*
1691 * Used by process_recorded_refs to determine if a new ref would overwrite an
1692 * already existing ref. In case it detects an overwrite, it returns the
1693 * inode/gen in who_ino/who_gen.
1694 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1695 * to make sure later references to the overwritten inode are possible.
1696 * Orphanizing is however only required for the first ref of an inode.
1697 * process_recorded_refs does an additional is_first_ref check to see if
1698 * orphanizing is really required.
1699 */
31db9f7c
AB		 1700static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1701 const char *name, int name_len,
1702 u64 *who_ino, u64 *who_gen)
1703{
1704 int ret = 0;
ebdad913 1705 u64 gen;
31db9f7c
AB		 1706 u64 other_inode = 0;
1707 u8 other_type = 0;
1708
1709 if (!sctx->parent_root)
1710 goto out;
1711
1712 ret = is_inode_existent(sctx, dir, dir_gen);
1713 if (ret <= 0)
1714 goto out;
1715
ebdad913
JB		 1716 /*
1717 * If we have a parent root we need to verify that the parent dir was
1718 * not delted and then re-created, if it was then we have no overwrite
1719 * and we can just unlink this entry.
1720 */
1721 if (sctx->parent_root) {
1722 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1723 NULL, NULL, NULL);
1724 if (ret < 0 && ret != -ENOENT)
1725 goto out;
1726 if (ret) {
1727 ret = 0;
1728 goto out;
1729 }
1730 if (gen != dir_gen)
1731 goto out;
1732 }
1733
31db9f7c
AB		 1734 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1735 &other_inode, &other_type);
1736 if (ret < 0 && ret != -ENOENT)
1737 goto out;
1738 if (ret) {
1739 ret = 0;
1740 goto out;
1741 }
1742
766702ef
AB		 1743 /*
1744 * Check if the overwritten ref was already processed. If yes, the ref
1745 * was already unlinked/moved, so we can safely assume that we will not
1746 * overwrite anything at this point in time.
1747 */
31db9f7c
AB		 1748 if (other_inode > sctx->send_progress) {
1749 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
85a7b33b 1750 who_gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 1751 if (ret < 0)
1752 goto out;
1753
1754 ret = 1;
1755 *who_ino = other_inode;
1756 } else {
1757 ret = 0;
1758 }
1759
1760out:
1761 return ret;
1762}
1763
766702ef
AB		 1764/*
1765 * Checks if the ref was overwritten by an already processed inode. This is
1766 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1767 * thus the orphan name needs be used.
1768 * process_recorded_refs also uses it to avoid unlinking of refs that were
1769 * overwritten.
1770 */
31db9f7c
AB		 1771static int did_overwrite_ref(struct send_ctx *sctx,
1772 u64 dir, u64 dir_gen,
1773 u64 ino, u64 ino_gen,
1774 const char *name, int name_len)
1775{
1776 int ret = 0;
1777 u64 gen;
1778 u64 ow_inode;
1779 u8 other_type;
1780
1781 if (!sctx->parent_root)
1782 goto out;
1783
1784 ret = is_inode_existent(sctx, dir, dir_gen);
1785 if (ret <= 0)
1786 goto out;
1787
1788 /* check if the ref was overwritten by another ref */
1789 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1790 &ow_inode, &other_type);
1791 if (ret < 0 && ret != -ENOENT)
1792 goto out;
1793 if (ret) {
1794 /* was never and will never be overwritten */
1795 ret = 0;
1796 goto out;
1797 }
1798
1799 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
85a7b33b 1800 NULL, NULL);
31db9f7c
AB		 1801 if (ret < 0)
1802 goto out;
1803
1804 if (ow_inode == ino && gen == ino_gen) {
1805 ret = 0;
1806 goto out;
1807 }
1808
1809 /* we know that it is or will be overwritten. check this now */
1810 if (ow_inode < sctx->send_progress)
1811 ret = 1;
1812 else
1813 ret = 0;
1814
1815out:
1816 return ret;
1817}
1818
766702ef
AB		 1819/*
1820 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1821 * that got overwritten. This is used by process_recorded_refs to determine
1822 * if it has to use the path as returned by get_cur_path or the orphan name.
1823 */
31db9f7c
AB		 1824static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1825{
1826 int ret = 0;
1827 struct fs_path *name = NULL;
1828 u64 dir;
1829 u64 dir_gen;
1830
1831 if (!sctx->parent_root)
1832 goto out;
1833
924794c9 1834 name = fs_path_alloc();
31db9f7c
AB		 1835 if (!name)
1836 return -ENOMEM;
1837
924794c9 1838 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
31db9f7c
AB		 1839 if (ret < 0)
1840 goto out;
1841
1842 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1843 name->start, fs_path_len(name));
31db9f7c
AB		 1844
1845out:
924794c9 1846 fs_path_free(name);
31db9f7c
AB		 1847 return ret;
1848}
1849
766702ef
AB		 1850/*
1851 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1852 * so we need to do some special handling in case we have clashes. This function
1853 * takes care of this with the help of name_cache_entry::radix_list.
5dc67d0b 1854 * In case of error, nce is kfreed.
766702ef 1855 */
31db9f7c
AB		 1856static int name_cache_insert(struct send_ctx *sctx,
1857 struct name_cache_entry *nce)
1858{
1859 int ret = 0;
7e0926fe
AB		 1860 struct list_head *nce_head;
1861
1862 nce_head = radix_tree_lookup(&sctx->name_cache,
1863 (unsigned long)nce->ino);
1864 if (!nce_head) {
1865 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
cfa7a9cc
TI		 1866 if (!nce_head) {
1867 kfree(nce);
31db9f7c 1868 return -ENOMEM;
cfa7a9cc 1869 }
7e0926fe 1870 INIT_LIST_HEAD(nce_head);
31db9f7c 1871
7e0926fe 1872 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
5dc67d0b
AB		 1873 if (ret < 0) {
1874 kfree(nce_head);
1875 kfree(nce);
31db9f7c 1876 return ret;
5dc67d0b 1877 }
31db9f7c 1878 }
7e0926fe 1879 list_add_tail(&nce->radix_list, nce_head);
31db9f7c
AB		 1880 list_add_tail(&nce->list, &sctx->name_cache_list);
1881 sctx->name_cache_size++;
1882
1883 return ret;
1884}
1885
1886static void name_cache_delete(struct send_ctx *sctx,
1887 struct name_cache_entry *nce)
1888{
7e0926fe 1889 struct list_head *nce_head;
31db9f7c 1890
7e0926fe
AB		 1891 nce_head = radix_tree_lookup(&sctx->name_cache,
1892 (unsigned long)nce->ino);
57fb8910
DS		 1893 if (!nce_head) {
1894 btrfs_err(sctx->send_root->fs_info,
1895 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
1896 nce->ino, sctx->name_cache_size);
1897 }
31db9f7c 1898
7e0926fe 1899 list_del(&nce->radix_list);
31db9f7c 1900 list_del(&nce->list);
31db9f7c 1901 sctx->name_cache_size--;
7e0926fe 1902
57fb8910
DS		 1903 /*
1904 * We may not get to the final release of nce_head if the lookup fails
1905 */
1906 if (nce_head && list_empty(nce_head)) {
7e0926fe
AB		 1907 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1908 kfree(nce_head);
1909 }
31db9f7c
AB		 1910}
1911
1912static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1913 u64 ino, u64 gen)
1914{
7e0926fe
AB		 1915 struct list_head *nce_head;
1916 struct name_cache_entry *cur;
31db9f7c 1917
7e0926fe
AB		 1918 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1919 if (!nce_head)
31db9f7c
AB		 1920 return NULL;
1921
7e0926fe
AB		 1922 list_for_each_entry(cur, nce_head, radix_list) {
1923 if (cur->ino == ino && cur->gen == gen)
1924 return cur;
1925 }
31db9f7c
AB		 1926 return NULL;
1927}
1928
766702ef
AB		 1929/*
1930 * Removes the entry from the list and adds it back to the end. This marks the
1931 * entry as recently used so that name_cache_clean_unused does not remove it.
1932 */
31db9f7c
AB		 1933static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1934{
1935 list_del(&nce->list);
1936 list_add_tail(&nce->list, &sctx->name_cache_list);
1937}
1938
766702ef
AB		 1939/*
1940 * Remove some entries from the beginning of name_cache_list.
1941 */
31db9f7c
AB		 1942static void name_cache_clean_unused(struct send_ctx *sctx)
1943{
1944 struct name_cache_entry *nce;
1945
1946 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1947 return;
1948
1949 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1950 nce = list_entry(sctx->name_cache_list.next,
1951 struct name_cache_entry, list);
1952 name_cache_delete(sctx, nce);
1953 kfree(nce);
1954 }
1955}
1956
1957static void name_cache_free(struct send_ctx *sctx)
1958{
1959 struct name_cache_entry *nce;
31db9f7c 1960
e938c8ad
AB		 1961 while (!list_empty(&sctx->name_cache_list)) {
1962 nce = list_entry(sctx->name_cache_list.next,
1963 struct name_cache_entry, list);
31db9f7c 1964 name_cache_delete(sctx, nce);
17589bd9 1965 kfree(nce);
31db9f7c
AB		 1966 }
1967}
1968
766702ef
AB		 1969/*
1970 * Used by get_cur_path for each ref up to the root.
1971 * Returns 0 if it succeeded.
1972 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1973 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1974 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1975 * Returns <0 in case of error.
1976 */
31db9f7c
AB		 1977static int __get_cur_name_and_parent(struct send_ctx *sctx,
1978 u64 ino, u64 gen,
9f03740a 1979 int skip_name_cache,
31db9f7c
AB		 1980 u64 *parent_ino,
1981 u64 *parent_gen,
1982 struct fs_path *dest)
1983{
1984 int ret;
1985 int nce_ret;
1986 struct btrfs_path *path = NULL;
1987 struct name_cache_entry *nce = NULL;
1988
9f03740a
FDBM		 1989 if (skip_name_cache)
1990 goto get_ref;
766702ef
AB		 1991 /*
1992 * First check if we already did a call to this function with the same
1993 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1994 * return the cached result.
1995 */
31db9f7c
AB		 1996 nce = name_cache_search(sctx, ino, gen);
1997 if (nce) {
1998 if (ino < sctx->send_progress && nce->need_later_update) {
1999 name_cache_delete(sctx, nce);
2000 kfree(nce);
2001 nce = NULL;
2002 } else {
2003 name_cache_used(sctx, nce);
2004 *parent_ino = nce->parent_ino;
2005 *parent_gen = nce->parent_gen;
2006 ret = fs_path_add(dest, nce->name, nce->name_len);
2007 if (ret < 0)
2008 goto out;
2009 ret = nce->ret;
2010 goto out;
2011 }
2012 }
2013
2014 path = alloc_path_for_send();
2015 if (!path)
2016 return -ENOMEM;
2017
766702ef
AB		 2018 /*
2019 * If the inode is not existent yet, add the orphan name and return 1.
2020 * This should only happen for the parent dir that we determine in
2021 * __record_new_ref
2022 */
31db9f7c
AB		 2023 ret = is_inode_existent(sctx, ino, gen);
2024 if (ret < 0)
2025 goto out;
2026
2027 if (!ret) {
2028 ret = gen_unique_name(sctx, ino, gen, dest);
2029 if (ret < 0)
2030 goto out;
2031 ret = 1;
2032 goto out_cache;
2033 }
2034
9f03740a 2035get_ref:
766702ef
AB		 2036 /*
2037 * Depending on whether the inode was already processed or not, use
2038 * send_root or parent_root for ref lookup.
2039 */
9f03740a 2040 if (ino < sctx->send_progress && !skip_name_cache)
924794c9
TI		 2041 ret = get_first_ref(sctx->send_root, ino,
2042 parent_ino, parent_gen, dest);
31db9f7c 2043 else
924794c9
TI		 2044 ret = get_first_ref(sctx->parent_root, ino,
2045 parent_ino, parent_gen, dest);
31db9f7c
AB		 2046 if (ret < 0)
2047 goto out;
2048
766702ef
AB		 2049 /*
2050 * Check if the ref was overwritten by an inode's ref that was processed
2051 * earlier. If yes, treat as orphan and return 1.
2052 */
31db9f7c
AB		 2053 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2054 dest->start, dest->end - dest->start);
2055 if (ret < 0)
2056 goto out;
2057 if (ret) {
2058 fs_path_reset(dest);
2059 ret = gen_unique_name(sctx, ino, gen, dest);
2060 if (ret < 0)
2061 goto out;
2062 ret = 1;
2063 }
9f03740a
FDBM		 2064 if (skip_name_cache)
2065 goto out;
31db9f7c
AB		 2066
2067out_cache:
766702ef
AB		 2068 /*
2069 * Store the result of the lookup in the name cache.
2070 */
31db9f7c
AB		 2071 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2072 if (!nce) {
2073 ret = -ENOMEM;
2074 goto out;
2075 }
2076
2077 nce->ino = ino;
2078 nce->gen = gen;
2079 nce->parent_ino = *parent_ino;
2080 nce->parent_gen = *parent_gen;
2081 nce->name_len = fs_path_len(dest);
2082 nce->ret = ret;
2083 strcpy(nce->name, dest->start);
31db9f7c
AB		 2084
2085 if (ino < sctx->send_progress)
2086 nce->need_later_update = 0;
2087 else
2088 nce->need_later_update = 1;
2089
2090 nce_ret = name_cache_insert(sctx, nce);
2091 if (nce_ret < 0)
2092 ret = nce_ret;
2093 name_cache_clean_unused(sctx);
2094
2095out:
2096 btrfs_free_path(path);
2097 return ret;
2098}
2099
2100/*
2101 * Magic happens here. This function returns the first ref to an inode as it
2102 * would look like while receiving the stream at this point in time.
2103 * We walk the path up to the root. For every inode in between, we check if it
2104 * was already processed/sent. If yes, we continue with the parent as found
2105 * in send_root. If not, we continue with the parent as found in parent_root.
2106 * If we encounter an inode that was deleted at this point in time, we use the
2107 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2108 * that were not created yet and overwritten inodes/refs.
2109 *
2110 * When do we have have orphan inodes:
2111 * 1. When an inode is freshly created and thus no valid refs are available yet
2112 * 2. When a directory lost all it's refs (deleted) but still has dir items
2113 * inside which were not processed yet (pending for move/delete). If anyone
2114 * tried to get the path to the dir items, it would get a path inside that
2115 * orphan directory.
2116 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2117 * of an unprocessed inode. If in that case the first ref would be
2118 * overwritten, the overwritten inode gets "orphanized". Later when we
2119 * process this overwritten inode, it is restored at a new place by moving
2120 * the orphan inode.
2121 *
2122 * sctx->send_progress tells this function at which point in time receiving
2123 * would be.
2124 */
2125static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2126 struct fs_path *dest)
2127{
2128 int ret = 0;
2129 struct fs_path *name = NULL;
2130 u64 parent_inode = 0;
2131 u64 parent_gen = 0;
2132 int stop = 0;
9f03740a 2133 int skip_name_cache = 0;
31db9f7c 2134
924794c9 2135 name = fs_path_alloc();
31db9f7c
AB		 2136 if (!name) {
2137 ret = -ENOMEM;
2138 goto out;
2139 }
2140
9f03740a
FDBM		 2141 if (is_waiting_for_move(sctx, ino))
2142 skip_name_cache = 1;
2143
31db9f7c
AB		 2144 dest->reversed = 1;
2145 fs_path_reset(dest);
2146
2147 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2148 fs_path_reset(name);
2149
9f03740a 2150 ret = __get_cur_name_and_parent(sctx, ino, gen, skip_name_cache,
31db9f7c
AB		 2151 &parent_inode, &parent_gen, name);
2152 if (ret < 0)
2153 goto out;
2154 if (ret)
2155 stop = 1;
2156
9f03740a 2157 if (!skip_name_cache &&
03cb4fb9 2158 is_waiting_for_move(sctx, parent_inode))
9f03740a 2159 skip_name_cache = 1;
9f03740a 2160
31db9f7c
AB		 2161 ret = fs_path_add_path(dest, name);
2162 if (ret < 0)
2163 goto out;
2164
2165 ino = parent_inode;
2166 gen = parent_gen;
2167 }
2168
2169out:
924794c9 2170 fs_path_free(name);
31db9f7c
AB		 2171 if (!ret)
2172 fs_path_unreverse(dest);
2173 return ret;
2174}
2175
31db9f7c
AB		 2176/*
2177 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2178 */
2179static int send_subvol_begin(struct send_ctx *sctx)
2180{
2181 int ret;
2182 struct btrfs_root *send_root = sctx->send_root;
2183 struct btrfs_root *parent_root = sctx->parent_root;
2184 struct btrfs_path *path;
2185 struct btrfs_key key;
2186 struct btrfs_root_ref *ref;
2187 struct extent_buffer *leaf;
2188 char *name = NULL;
2189 int namelen;
2190
ffcfaf81 2191 path = btrfs_alloc_path();
31db9f7c
AB		 2192 if (!path)
2193 return -ENOMEM;
2194
2195 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2196 if (!name) {
2197 btrfs_free_path(path);
2198 return -ENOMEM;
2199 }
2200
2201 key.objectid = send_root->objectid;
2202 key.type = BTRFS_ROOT_BACKREF_KEY;
2203 key.offset = 0;
2204
2205 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2206 &key, path, 1, 0);
2207 if (ret < 0)
2208 goto out;
2209 if (ret) {
2210 ret = -ENOENT;
2211 goto out;
2212 }
2213
2214 leaf = path->nodes[0];
2215 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2216 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2217 key.objectid != send_root->objectid) {
2218 ret = -ENOENT;
2219 goto out;
2220 }
2221 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2222 namelen = btrfs_root_ref_name_len(leaf, ref);
2223 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2224 btrfs_release_path(path);
2225
31db9f7c
AB		 2226 if (parent_root) {
2227 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2228 if (ret < 0)
2229 goto out;
2230 } else {
2231 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2232 if (ret < 0)
2233 goto out;
2234 }
2235
2236 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2237 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2238 sctx->send_root->root_item.uuid);
2239 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
5a0f4e2c 2240 le64_to_cpu(sctx->send_root->root_item.ctransid));
31db9f7c
AB		 2241 if (parent_root) {
2242 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2243 sctx->parent_root->root_item.uuid);
2244 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
5a0f4e2c 2245 le64_to_cpu(sctx->parent_root->root_item.ctransid));
31db9f7c
AB		 2246 }
2247
2248 ret = send_cmd(sctx);
2249
2250tlv_put_failure:
2251out:
2252 btrfs_free_path(path);
2253 kfree(name);
2254 return ret;
2255}
2256
2257static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2258{
2259 int ret = 0;
2260 struct fs_path *p;
2261
2262verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2263
924794c9 2264 p = fs_path_alloc();
31db9f7c
AB		 2265 if (!p)
2266 return -ENOMEM;
2267
2268 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2269 if (ret < 0)
2270 goto out;
2271
2272 ret = get_cur_path(sctx, ino, gen, p);
2273 if (ret < 0)
2274 goto out;
2275 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2276 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2277
2278 ret = send_cmd(sctx);
2279
2280tlv_put_failure:
2281out:
924794c9 2282 fs_path_free(p);
31db9f7c
AB		 2283 return ret;
2284}
2285
2286static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2287{
2288 int ret = 0;
2289 struct fs_path *p;
2290
2291verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2292
924794c9 2293 p = fs_path_alloc();
31db9f7c
AB		 2294 if (!p)
2295 return -ENOMEM;
2296
2297 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2298 if (ret < 0)
2299 goto out;
2300
2301 ret = get_cur_path(sctx, ino, gen, p);
2302 if (ret < 0)
2303 goto out;
2304 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2305 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2306
2307 ret = send_cmd(sctx);
2308
2309tlv_put_failure:
2310out:
924794c9 2311 fs_path_free(p);
31db9f7c
AB		 2312 return ret;
2313}
2314
2315static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2316{
2317 int ret = 0;
2318 struct fs_path *p;
2319
2320verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2321
924794c9 2322 p = fs_path_alloc();
31db9f7c
AB		 2323 if (!p)
2324 return -ENOMEM;
2325
2326 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2327 if (ret < 0)
2328 goto out;
2329
2330 ret = get_cur_path(sctx, ino, gen, p);
2331 if (ret < 0)
2332 goto out;
2333 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2334 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2335 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2336
2337 ret = send_cmd(sctx);
2338
2339tlv_put_failure:
2340out:
924794c9 2341 fs_path_free(p);
31db9f7c
AB		 2342 return ret;
2343}
2344
2345static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2346{
2347 int ret = 0;
2348 struct fs_path *p = NULL;
2349 struct btrfs_inode_item *ii;
2350 struct btrfs_path *path = NULL;
2351 struct extent_buffer *eb;
2352 struct btrfs_key key;
2353 int slot;
2354
2355verbose_printk("btrfs: send_utimes %llu\n", ino);
2356
924794c9 2357 p = fs_path_alloc();
31db9f7c
AB		 2358 if (!p)
2359 return -ENOMEM;
2360
2361 path = alloc_path_for_send();
2362 if (!path) {
2363 ret = -ENOMEM;
2364 goto out;
2365 }
2366
2367 key.objectid = ino;
2368 key.type = BTRFS_INODE_ITEM_KEY;
2369 key.offset = 0;
2370 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2371 if (ret < 0)
2372 goto out;
2373
2374 eb = path->nodes[0];
2375 slot = path->slots[0];
2376 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2377
2378 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2379 if (ret < 0)
2380 goto out;
2381
2382 ret = get_cur_path(sctx, ino, gen, p);
2383 if (ret < 0)
2384 goto out;
2385 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2386 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2387 btrfs_inode_atime(ii));
2388 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2389 btrfs_inode_mtime(ii));
2390 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2391 btrfs_inode_ctime(ii));
766702ef 2392 /* TODO Add otime support when the otime patches get into upstream */
31db9f7c
AB		 2393
2394 ret = send_cmd(sctx);
2395
2396tlv_put_failure:
2397out:
924794c9 2398 fs_path_free(p);
31db9f7c
AB		 2399 btrfs_free_path(path);
2400 return ret;
2401}
2402
2403/*
2404 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2405 * a valid path yet because we did not process the refs yet. So, the inode
2406 * is created as orphan.
2407 */
1f4692da 2408static int send_create_inode(struct send_ctx *sctx, u64 ino)
31db9f7c
AB		 2409{
2410 int ret = 0;
31db9f7c 2411 struct fs_path *p;
31db9f7c 2412 int cmd;
1f4692da 2413 u64 gen;
31db9f7c 2414 u64 mode;
1f4692da 2415 u64 rdev;
31db9f7c 2416
1f4692da 2417verbose_printk("btrfs: send_create_inode %llu\n", ino);
31db9f7c 2418
924794c9 2419 p = fs_path_alloc();
31db9f7c
AB		 2420 if (!p)
2421 return -ENOMEM;
2422
1f4692da
AB		 2423 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2424 NULL, &rdev);
2425 if (ret < 0)
2426 goto out;
31db9f7c 2427
e938c8ad 2428 if (S_ISREG(mode)) {
31db9f7c 2429 cmd = BTRFS_SEND_C_MKFILE;
e938c8ad 2430 } else if (S_ISDIR(mode)) {
31db9f7c 2431 cmd = BTRFS_SEND_C_MKDIR;
e938c8ad 2432 } else if (S_ISLNK(mode)) {
31db9f7c 2433 cmd = BTRFS_SEND_C_SYMLINK;
e938c8ad 2434 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
31db9f7c 2435 cmd = BTRFS_SEND_C_MKNOD;
e938c8ad 2436 } else if (S_ISFIFO(mode)) {
31db9f7c 2437 cmd = BTRFS_SEND_C_MKFIFO;
e938c8ad 2438 } else if (S_ISSOCK(mode)) {
31db9f7c 2439 cmd = BTRFS_SEND_C_MKSOCK;
e938c8ad 2440 } else {
31db9f7c
AB		 2441 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2442 (int)(mode & S_IFMT));
2443 ret = -ENOTSUPP;
2444 goto out;
2445 }
2446
2447 ret = begin_cmd(sctx, cmd);
2448 if (ret < 0)
2449 goto out;
2450
1f4692da 2451 ret = gen_unique_name(sctx, ino, gen, p);
31db9f7c
AB		 2452 if (ret < 0)
2453 goto out;
2454
2455 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
1f4692da 2456 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
31db9f7c
AB		 2457
2458 if (S_ISLNK(mode)) {
2459 fs_path_reset(p);
924794c9 2460 ret = read_symlink(sctx->send_root, ino, p);
31db9f7c
AB		 2461 if (ret < 0)
2462 goto out;
2463 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2464 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2465 S_ISFIFO(mode) || S_ISSOCK(mode)) {
d79e5043
AJ		 2466 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2467 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
31db9f7c
AB		 2468 }
2469
2470 ret = send_cmd(sctx);
2471 if (ret < 0)
2472 goto out;
2473
2474
2475tlv_put_failure:
2476out:
924794c9 2477 fs_path_free(p);
31db9f7c
AB		 2478 return ret;
2479}
2480
1f4692da
AB		 2481/*
2482 * We need some special handling for inodes that get processed before the parent
2483 * directory got created. See process_recorded_refs for details.
2484 * This function does the check if we already created the dir out of order.
2485 */
2486static int did_create_dir(struct send_ctx *sctx, u64 dir)
2487{
2488 int ret = 0;
2489 struct btrfs_path *path = NULL;
2490 struct btrfs_key key;
2491 struct btrfs_key found_key;
2492 struct btrfs_key di_key;
2493 struct extent_buffer *eb;
2494 struct btrfs_dir_item *di;
2495 int slot;
2496
2497 path = alloc_path_for_send();
2498 if (!path) {
2499 ret = -ENOMEM;
2500 goto out;
2501 }
2502
2503 key.objectid = dir;
2504 key.type = BTRFS_DIR_INDEX_KEY;
2505 key.offset = 0;
2506 while (1) {
2507 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2508 1, 0);
2509 if (ret < 0)
2510 goto out;
2511 if (!ret) {
2512 eb = path->nodes[0];
2513 slot = path->slots[0];
2514 btrfs_item_key_to_cpu(eb, &found_key, slot);
2515 }
2516 if (ret || found_key.objectid != key.objectid ||
2517 found_key.type != key.type) {
2518 ret = 0;
2519 goto out;
2520 }
2521
2522 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2523 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2524
a0525414
JB		 2525 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2526 di_key.objectid < sctx->send_progress) {
1f4692da
AB		 2527 ret = 1;
2528 goto out;
2529 }
2530
2531 key.offset = found_key.offset + 1;
2532 btrfs_release_path(path);
2533 }
2534
2535out:
2536 btrfs_free_path(path);
2537 return ret;
2538}
2539
2540/*
2541 * Only creates the inode if it is:
2542 * 1. Not a directory
2543 * 2. Or a directory which was not created already due to out of order
2544 * directories. See did_create_dir and process_recorded_refs for details.
2545 */
2546static int send_create_inode_if_needed(struct send_ctx *sctx)
2547{
2548 int ret;
2549
2550 if (S_ISDIR(sctx->cur_inode_mode)) {
2551 ret = did_create_dir(sctx, sctx->cur_ino);
2552 if (ret < 0)
2553 goto out;
2554 if (ret) {
2555 ret = 0;
2556 goto out;
2557 }
2558 }
2559
2560 ret = send_create_inode(sctx, sctx->cur_ino);
2561 if (ret < 0)
2562 goto out;
2563
2564out:
2565 return ret;
2566}
2567
31db9f7c
AB		 2568struct recorded_ref {
2569 struct list_head list;
2570 char *dir_path;
2571 char *name;
2572 struct fs_path *full_path;
2573 u64 dir;
2574 u64 dir_gen;
2575 int dir_path_len;
2576 int name_len;
2577};
2578
2579/*
2580 * We need to process new refs before deleted refs, but compare_tree gives us
2581 * everything mixed. So we first record all refs and later process them.
2582 * This function is a helper to record one ref.
2583 */
2584static int record_ref(struct list_head *head, u64 dir,
2585 u64 dir_gen, struct fs_path *path)
2586{
2587 struct recorded_ref *ref;
31db9f7c
AB		 2588
2589 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2590 if (!ref)
2591 return -ENOMEM;
2592
2593 ref->dir = dir;
2594 ref->dir_gen = dir_gen;
2595 ref->full_path = path;
2596
ed84885d
AS		 2597 ref->name = (char *)kbasename(ref->full_path->start);
2598 ref->name_len = ref->full_path->end - ref->name;
2599 ref->dir_path = ref->full_path->start;
2600 if (ref->name == ref->full_path->start)
31db9f7c 2601 ref->dir_path_len = 0;
ed84885d 2602 else
31db9f7c
AB		 2603 ref->dir_path_len = ref->full_path->end -
2604 ref->full_path->start - 1 - ref->name_len;
31db9f7c
AB		 2605
2606 list_add_tail(&ref->list, head);
2607 return 0;
2608}
2609
ba5e8f2e
JB		 2610static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2611{
2612 struct recorded_ref *new;
2613
2614 new = kmalloc(sizeof(*ref), GFP_NOFS);
2615 if (!new)
2616 return -ENOMEM;
2617
2618 new->dir = ref->dir;
2619 new->dir_gen = ref->dir_gen;
2620 new->full_path = NULL;
2621 INIT_LIST_HEAD(&new->list);
2622 list_add_tail(&new->list, list);
2623 return 0;
2624}
2625
924794c9 2626static void __free_recorded_refs(struct list_head *head)
31db9f7c
AB		 2627{
2628 struct recorded_ref *cur;
31db9f7c 2629
e938c8ad
AB		 2630 while (!list_empty(head)) {
2631 cur = list_entry(head->next, struct recorded_ref, list);
924794c9 2632 fs_path_free(cur->full_path);
e938c8ad 2633 list_del(&cur->list);
31db9f7c
AB		 2634 kfree(cur);
2635 }
31db9f7c
AB		 2636}
2637
2638static void free_recorded_refs(struct send_ctx *sctx)
2639{
924794c9
TI		 2640 __free_recorded_refs(&sctx->new_refs);
2641 __free_recorded_refs(&sctx->deleted_refs);
31db9f7c
AB		 2642}
2643
2644/*
766702ef 2645 * Renames/moves a file/dir to its orphan name. Used when the first
31db9f7c
AB		 2646 * ref of an unprocessed inode gets overwritten and for all non empty
2647 * directories.
2648 */
2649static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2650 struct fs_path *path)
2651{
2652 int ret;
2653 struct fs_path *orphan;
2654
924794c9 2655 orphan = fs_path_alloc();
31db9f7c
AB		 2656 if (!orphan)
2657 return -ENOMEM;
2658
2659 ret = gen_unique_name(sctx, ino, gen, orphan);
2660 if (ret < 0)
2661 goto out;
2662
2663 ret = send_rename(sctx, path, orphan);
2664
2665out:
924794c9 2666 fs_path_free(orphan);
31db9f7c
AB		 2667 return ret;
2668}
2669
2670/*
2671 * Returns 1 if a directory can be removed at this point in time.
2672 * We check this by iterating all dir items and checking if the inode behind
2673 * the dir item was already processed.
2674 */
2675static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2676{
2677 int ret = 0;
2678 struct btrfs_root *root = sctx->parent_root;
2679 struct btrfs_path *path;
2680 struct btrfs_key key;
2681 struct btrfs_key found_key;
2682 struct btrfs_key loc;
2683 struct btrfs_dir_item *di;
2684
6d85ed05
AB		 2685 /*
2686 * Don't try to rmdir the top/root subvolume dir.
2687 */
2688 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2689 return 0;
2690
31db9f7c
AB		 2691 path = alloc_path_for_send();
2692 if (!path)
2693 return -ENOMEM;
2694
2695 key.objectid = dir;
2696 key.type = BTRFS_DIR_INDEX_KEY;
2697 key.offset = 0;
2698
2699 while (1) {
2700 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2701 if (ret < 0)
2702 goto out;
2703 if (!ret) {
2704 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2705 path->slots[0]);
2706 }
2707 if (ret || found_key.objectid != key.objectid ||
2708 found_key.type != key.type) {
2709 break;
2710 }
2711
2712 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2713 struct btrfs_dir_item);
2714 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2715
2716 if (loc.objectid > send_progress) {
2717 ret = 0;
2718 goto out;
2719 }
2720
2721 btrfs_release_path(path);
2722 key.offset = found_key.offset + 1;
2723 }
2724
2725 ret = 1;
2726
2727out:
2728 btrfs_free_path(path);
2729 return ret;
2730}
2731
9f03740a
FDBM		 2732static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2733{
2734 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2735 struct waiting_dir_move *entry;
2736
2737 while (n) {
2738 entry = rb_entry(n, struct waiting_dir_move, node);
2739 if (ino < entry->ino)
2740 n = n->rb_left;
2741 else if (ino > entry->ino)
2742 n = n->rb_right;
2743 else
2744 return 1;
2745 }
2746 return 0;
2747}
2748
2749static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2750{
2751 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2752 struct rb_node *parent = NULL;
2753 struct waiting_dir_move *entry, *dm;
2754
2755 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2756 if (!dm)
2757 return -ENOMEM;
2758 dm->ino = ino;
2759
2760 while (*p) {
2761 parent = *p;
2762 entry = rb_entry(parent, struct waiting_dir_move, node);
2763 if (ino < entry->ino) {
2764 p = &(*p)->rb_left;
2765 } else if (ino > entry->ino) {
2766 p = &(*p)->rb_right;
2767 } else {
2768 kfree(dm);
2769 return -EEXIST;
2770 }
2771 }
2772
2773 rb_link_node(&dm->node, parent, p);
2774 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2775 return 0;
2776}
2777
9f03740a
FDBM		 2778static int del_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2779{
2780 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2781 struct waiting_dir_move *entry;
2782
2783 while (n) {
2784 entry = rb_entry(n, struct waiting_dir_move, node);
2785 if (ino < entry->ino) {
2786 n = n->rb_left;
2787 } else if (ino > entry->ino) {
2788 n = n->rb_right;
2789 } else {
2790 rb_erase(&entry->node, &sctx->waiting_dir_moves);
2791 kfree(entry);
2792 return 0;
2793 }
2794 }
2795 return -ENOENT;
2796}
2797
9f03740a
FDBM		 2798static int add_pending_dir_move(struct send_ctx *sctx, u64 parent_ino)
2799{
2800 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2801 struct rb_node *parent = NULL;
2802 struct pending_dir_move *entry, *pm;
2803 struct recorded_ref *cur;
2804 int exists = 0;
2805 int ret;
2806
2807 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2808 if (!pm)
2809 return -ENOMEM;
2810 pm->parent_ino = parent_ino;
2811 pm->ino = sctx->cur_ino;
2812 pm->gen = sctx->cur_inode_gen;
2813 INIT_LIST_HEAD(&pm->list);
2814 INIT_LIST_HEAD(&pm->update_refs);
2815 RB_CLEAR_NODE(&pm->node);
2816
2817 while (*p) {
2818 parent = *p;
2819 entry = rb_entry(parent, struct pending_dir_move, node);
2820 if (parent_ino < entry->parent_ino) {
2821 p = &(*p)->rb_left;
2822 } else if (parent_ino > entry->parent_ino) {
2823 p = &(*p)->rb_right;
2824 } else {
2825 exists = 1;
2826 break;
2827 }
2828 }
2829
2830 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2831 ret = dup_ref(cur, &pm->update_refs);
2832 if (ret < 0)
2833 goto out;
2834 }
2835 list_for_each_entry(cur, &sctx->new_refs, list) {
2836 ret = dup_ref(cur, &pm->update_refs);
2837 if (ret < 0)
2838 goto out;
2839 }
2840
2841 ret = add_waiting_dir_move(sctx, pm->ino);
2842 if (ret)
2843 goto out;
2844
2845 if (exists) {
2846 list_add_tail(&pm->list, &entry->list);
2847 } else {
2848 rb_link_node(&pm->node, parent, p);
2849 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
2850 }
2851 ret = 0;
2852out:
2853 if (ret) {
2854 __free_recorded_refs(&pm->update_refs);
2855 kfree(pm);
2856 }
2857 return ret;
2858}
2859
2860static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
2861 u64 parent_ino)
2862{
2863 struct rb_node *n = sctx->pending_dir_moves.rb_node;
2864 struct pending_dir_move *entry;
2865
2866 while (n) {
2867 entry = rb_entry(n, struct pending_dir_move, node);
2868 if (parent_ino < entry->parent_ino)
2869 n = n->rb_left;
2870 else if (parent_ino > entry->parent_ino)
2871 n = n->rb_right;
2872 else
2873 return entry;
2874 }
2875 return NULL;
2876}
2877
2878static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
2879{
2880 struct fs_path *from_path = NULL;
2881 struct fs_path *to_path = NULL;
2882 u64 orig_progress = sctx->send_progress;
2883 struct recorded_ref *cur;
2884 int ret;
2885
2886 from_path = fs_path_alloc();
2887 if (!from_path)
2888 return -ENOMEM;
2889
2890 sctx->send_progress = pm->ino;
2891 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
2892 if (ret < 0)
2893 goto out;
2894
2895 to_path = fs_path_alloc();
2896 if (!to_path) {
2897 ret = -ENOMEM;
2898 goto out;
2899 }
2900
2901 sctx->send_progress = sctx->cur_ino + 1;
6cc98d90
JB		 2902 ret = del_waiting_dir_move(sctx, pm->ino);
2903 ASSERT(ret == 0);
2904
9f03740a
FDBM		 2905 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
2906 if (ret < 0)
2907 goto out;
2908
2909 ret = send_rename(sctx, from_path, to_path);
2910 if (ret < 0)
2911 goto out;
2912
2913 ret = send_utimes(sctx, pm->ino, pm->gen);
2914 if (ret < 0)
2915 goto out;
2916
2917 /*
2918 * After rename/move, need to update the utimes of both new parent(s)
2919 * and old parent(s).
2920 */
2921 list_for_each_entry(cur, &pm->update_refs, list) {
2922 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
2923 if (ret < 0)
2924 goto out;
2925 }
2926
2927out:
2928 fs_path_free(from_path);
2929 fs_path_free(to_path);
2930 sctx->send_progress = orig_progress;
2931
2932 return ret;
2933}
2934
2935static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
2936{
2937 if (!list_empty(&m->list))
2938 list_del(&m->list);
2939 if (!RB_EMPTY_NODE(&m->node))
2940 rb_erase(&m->node, &sctx->pending_dir_moves);
2941 __free_recorded_refs(&m->update_refs);
2942 kfree(m);
2943}
2944
2945static void tail_append_pending_moves(struct pending_dir_move *moves,
2946 struct list_head *stack)
2947{
2948 if (list_empty(&moves->list)) {
2949 list_add_tail(&moves->list, stack);
2950 } else {
2951 LIST_HEAD(list);
2952 list_splice_init(&moves->list, &list);
2953 list_add_tail(&moves->list, stack);
2954 list_splice_tail(&list, stack);
2955 }
2956}
2957
2958static int apply_children_dir_moves(struct send_ctx *sctx)
2959{
2960 struct pending_dir_move *pm;
2961 struct list_head stack;
2962 u64 parent_ino = sctx->cur_ino;
2963 int ret = 0;
2964
2965 pm = get_pending_dir_moves(sctx, parent_ino);
2966 if (!pm)
2967 return 0;
2968
2969 INIT_LIST_HEAD(&stack);
2970 tail_append_pending_moves(pm, &stack);
2971
2972 while (!list_empty(&stack)) {
2973 pm = list_first_entry(&stack, struct pending_dir_move, list);
2974 parent_ino = pm->ino;
2975 ret = apply_dir_move(sctx, pm);
2976 free_pending_move(sctx, pm);
2977 if (ret)
2978 goto out;
2979 pm = get_pending_dir_moves(sctx, parent_ino);
2980 if (pm)
2981 tail_append_pending_moves(pm, &stack);
2982 }
2983 return 0;
2984
2985out:
2986 while (!list_empty(&stack)) {
2987 pm = list_first_entry(&stack, struct pending_dir_move, list);
2988 free_pending_move(sctx, pm);
2989 }
2990 return ret;
2991}
2992
2993static int wait_for_parent_move(struct send_ctx *sctx,
2994 struct recorded_ref *parent_ref)
2995{
2996 int ret;
2997 u64 ino = parent_ref->dir;
2998 u64 parent_ino_before, parent_ino_after;
2999 u64 new_gen, old_gen;
3000 struct fs_path *path_before = NULL;
3001 struct fs_path *path_after = NULL;
3002 int len1, len2;
3003
3004 if (parent_ref->dir <= sctx->cur_ino)
3005 return 0;
3006
3007 if (is_waiting_for_move(sctx, ino))
3008 return 1;
3009
3010 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3011 NULL, NULL, NULL, NULL);
3012 if (ret == -ENOENT)
3013 return 0;
3014 else if (ret < 0)
3015 return ret;
3016
3017 ret = get_inode_info(sctx->send_root, ino, NULL, &new_gen,
3018 NULL, NULL, NULL, NULL);
3019 if (ret < 0)
3020 return ret;
3021
3022 if (new_gen != old_gen)
3023 return 0;
3024
3025 path_before = fs_path_alloc();
3026 if (!path_before)
3027 return -ENOMEM;
3028
3029 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3030 NULL, path_before);
3031 if (ret == -ENOENT) {
3032 ret = 0;
3033 goto out;
3034 } else if (ret < 0) {
3035 goto out;
3036 }
3037
3038 path_after = fs_path_alloc();
3039 if (!path_after) {
3040 ret = -ENOMEM;
3041 goto out;
3042 }
3043
3044 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3045 NULL, path_after);
3046 if (ret == -ENOENT) {
3047 ret = 0;
3048 goto out;
3049 } else if (ret < 0) {
3050 goto out;
3051 }
3052
3053 len1 = fs_path_len(path_before);
3054 len2 = fs_path_len(path_after);
5ed7f9ff
FDBM		 3055 if (parent_ino_before != parent_ino_after || len1 != len2 ||
3056 memcmp(path_before->start, path_after->start, len1)) {
9f03740a
FDBM		 3057 ret = 1;
3058 goto out;
3059 }
3060 ret = 0;
3061
3062out:
3063 fs_path_free(path_before);
3064 fs_path_free(path_after);
3065
3066 return ret;
3067}
3068
31db9f7c
AB		 3069/*
3070 * This does all the move/link/unlink/rmdir magic.
3071 */
9f03740a 3072static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
31db9f7c
AB		 3073{
3074 int ret = 0;
3075 struct recorded_ref *cur;
1f4692da 3076 struct recorded_ref *cur2;
ba5e8f2e 3077 struct list_head check_dirs;
31db9f7c 3078 struct fs_path *valid_path = NULL;
b24baf69 3079 u64 ow_inode = 0;
31db9f7c
AB		 3080 u64 ow_gen;
3081 int did_overwrite = 0;
3082 int is_orphan = 0;
3083
3084verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3085
6d85ed05
AB		 3086 /*
3087 * This should never happen as the root dir always has the same ref
3088 * which is always '..'
3089 */
3090 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
ba5e8f2e 3091 INIT_LIST_HEAD(&check_dirs);
6d85ed05 3092
924794c9 3093 valid_path = fs_path_alloc();
31db9f7c
AB		 3094 if (!valid_path) {
3095 ret = -ENOMEM;
3096 goto out;
3097 }
3098
31db9f7c
AB		 3099 /*
3100 * First, check if the first ref of the current inode was overwritten
3101 * before. If yes, we know that the current inode was already orphanized
3102 * and thus use the orphan name. If not, we can use get_cur_path to
3103 * get the path of the first ref as it would like while receiving at
3104 * this point in time.
3105 * New inodes are always orphan at the beginning, so force to use the
3106 * orphan name in this case.
3107 * The first ref is stored in valid_path and will be updated if it
3108 * gets moved around.
3109 */
3110 if (!sctx->cur_inode_new) {
3111 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3112 sctx->cur_inode_gen);
3113 if (ret < 0)
3114 goto out;
3115 if (ret)
3116 did_overwrite = 1;
3117 }
3118 if (sctx->cur_inode_new || did_overwrite) {
3119 ret = gen_unique_name(sctx, sctx->cur_ino,
3120 sctx->cur_inode_gen, valid_path);
3121 if (ret < 0)
3122 goto out;
3123 is_orphan = 1;
3124 } else {
3125 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3126 valid_path);
3127 if (ret < 0)
3128 goto out;
3129 }
3130
3131 list_for_each_entry(cur, &sctx->new_refs, list) {
1f4692da
AB		 3132 /*
3133 * We may have refs where the parent directory does not exist
3134 * yet. This happens if the parent directories inum is higher
3135 * the the current inum. To handle this case, we create the
3136 * parent directory out of order. But we need to check if this
3137 * did already happen before due to other refs in the same dir.
3138 */
3139 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3140 if (ret < 0)
3141 goto out;
3142 if (ret == inode_state_will_create) {
3143 ret = 0;
3144 /*
3145 * First check if any of the current inodes refs did
3146 * already create the dir.
3147 */
3148 list_for_each_entry(cur2, &sctx->new_refs, list) {
3149 if (cur == cur2)
3150 break;
3151 if (cur2->dir == cur->dir) {
3152 ret = 1;
3153 break;
3154 }
3155 }
3156
3157 /*
3158 * If that did not happen, check if a previous inode
3159 * did already create the dir.
3160 */
3161 if (!ret)
3162 ret = did_create_dir(sctx, cur->dir);
3163 if (ret < 0)
3164 goto out;
3165 if (!ret) {
3166 ret = send_create_inode(sctx, cur->dir);
3167 if (ret < 0)
3168 goto out;
3169 }
3170 }
3171
31db9f7c
AB		 3172 /*
3173 * Check if this new ref would overwrite the first ref of
3174 * another unprocessed inode. If yes, orphanize the
3175 * overwritten inode. If we find an overwritten ref that is
3176 * not the first ref, simply unlink it.
3177 */
3178 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3179 cur->name, cur->name_len,
3180 &ow_inode, &ow_gen);
3181 if (ret < 0)
3182 goto out;
3183 if (ret) {
924794c9
TI		 3184 ret = is_first_ref(sctx->parent_root,
3185 ow_inode, cur->dir, cur->name,
3186 cur->name_len);
31db9f7c
AB		 3187 if (ret < 0)
3188 goto out;
3189 if (ret) {
3190 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3191 cur->full_path);
3192 if (ret < 0)
3193 goto out;
3194 } else {
3195 ret = send_unlink(sctx, cur->full_path);
3196 if (ret < 0)
3197 goto out;
3198 }
3199 }
3200
3201 /*
3202 * link/move the ref to the new place. If we have an orphan
3203 * inode, move it and update valid_path. If not, link or move
3204 * it depending on the inode mode.
3205 */
1f4692da 3206 if (is_orphan) {
31db9f7c
AB		 3207 ret = send_rename(sctx, valid_path, cur->full_path);
3208 if (ret < 0)
3209 goto out;
3210 is_orphan = 0;
3211 ret = fs_path_copy(valid_path, cur->full_path);
3212 if (ret < 0)
3213 goto out;
3214 } else {
3215 if (S_ISDIR(sctx->cur_inode_mode)) {
3216 /*
3217 * Dirs can't be linked, so move it. For moved
3218 * dirs, we always have one new and one deleted
3219 * ref. The deleted ref is ignored later.
3220 */
d86477b3
FDBM		 3221 ret = wait_for_parent_move(sctx, cur);
3222 if (ret < 0)
3223 goto out;
3224 if (ret) {
9f03740a
FDBM		 3225 ret = add_pending_dir_move(sctx,
3226 cur->dir);
3227 *pending_move = 1;
3228 } else {
3229 ret = send_rename(sctx, valid_path,
3230 cur->full_path);
3231 if (!ret)
3232 ret = fs_path_copy(valid_path,
3233 cur->full_path);
3234 }
31db9f7c
AB		 3235 if (ret < 0)
3236 goto out;
3237 } else {
3238 ret = send_link(sctx, cur->full_path,
3239 valid_path);
3240 if (ret < 0)
3241 goto out;
3242 }
3243 }
ba5e8f2e 3244 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 3245 if (ret < 0)
3246 goto out;
3247 }
3248
3249 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3250 /*
3251 * Check if we can already rmdir the directory. If not,
3252 * orphanize it. For every dir item inside that gets deleted
3253 * later, we do this check again and rmdir it then if possible.
3254 * See the use of check_dirs for more details.
3255 */
3256 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
3257 if (ret < 0)
3258 goto out;
3259 if (ret) {
3260 ret = send_rmdir(sctx, valid_path);
3261 if (ret < 0)
3262 goto out;
3263 } else if (!is_orphan) {
3264 ret = orphanize_inode(sctx, sctx->cur_ino,
3265 sctx->cur_inode_gen, valid_path);
3266 if (ret < 0)
3267 goto out;
3268 is_orphan = 1;
3269 }
3270
3271 list_for_each_entry(cur, &sctx->deleted_refs, list) {
ba5e8f2e 3272 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 3273 if (ret < 0)
3274 goto out;
3275 }
ccf1626b
AB		 3276 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3277 !list_empty(&sctx->deleted_refs)) {
3278 /*
3279 * We have a moved dir. Add the old parent to check_dirs
3280 */
3281 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3282 list);
ba5e8f2e 3283 ret = dup_ref(cur, &check_dirs);
ccf1626b
AB		 3284 if (ret < 0)
3285 goto out;
31db9f7c
AB		 3286 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3287 /*
3288 * We have a non dir inode. Go through all deleted refs and
3289 * unlink them if they were not already overwritten by other
3290 * inodes.
3291 */
3292 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3293 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3294 sctx->cur_ino, sctx->cur_inode_gen,
3295 cur->name, cur->name_len);
3296 if (ret < 0)
3297 goto out;
3298 if (!ret) {
1f4692da
AB		 3299 ret = send_unlink(sctx, cur->full_path);
3300 if (ret < 0)
3301 goto out;
31db9f7c 3302 }
ba5e8f2e 3303 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 3304 if (ret < 0)
3305 goto out;
3306 }
31db9f7c
AB		 3307 /*
3308 * If the inode is still orphan, unlink the orphan. This may
3309 * happen when a previous inode did overwrite the first ref
3310 * of this inode and no new refs were added for the current
766702ef
AB		 3311 * inode. Unlinking does not mean that the inode is deleted in
3312 * all cases. There may still be links to this inode in other
3313 * places.
31db9f7c 3314 */
1f4692da 3315 if (is_orphan) {
31db9f7c
AB		 3316 ret = send_unlink(sctx, valid_path);
3317 if (ret < 0)
3318 goto out;
3319 }
3320 }
3321
3322 /*
3323 * We did collect all parent dirs where cur_inode was once located. We
3324 * now go through all these dirs and check if they are pending for
3325 * deletion and if it's finally possible to perform the rmdir now.
3326 * We also update the inode stats of the parent dirs here.
3327 */
ba5e8f2e 3328 list_for_each_entry(cur, &check_dirs, list) {
766702ef
AB		 3329 /*
3330 * In case we had refs into dirs that were not processed yet,
3331 * we don't need to do the utime and rmdir logic for these dirs.
3332 * The dir will be processed later.
3333 */
ba5e8f2e 3334 if (cur->dir > sctx->cur_ino)
31db9f7c
AB		 3335 continue;
3336
ba5e8f2e 3337 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 3338 if (ret < 0)
3339 goto out;
3340
3341 if (ret == inode_state_did_create ||
3342 ret == inode_state_no_change) {
3343 /* TODO delayed utimes */
ba5e8f2e 3344 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 3345 if (ret < 0)
3346 goto out;
3347 } else if (ret == inode_state_did_delete) {
ba5e8f2e 3348 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
31db9f7c
AB		 3349 if (ret < 0)
3350 goto out;
3351 if (ret) {
ba5e8f2e
JB		 3352 ret = get_cur_path(sctx, cur->dir,
3353 cur->dir_gen, valid_path);
31db9f7c
AB		 3354 if (ret < 0)
3355 goto out;
3356 ret = send_rmdir(sctx, valid_path);
3357 if (ret < 0)
3358 goto out;
3359 }
3360 }
3361 }
3362
31db9f7c
AB		 3363 ret = 0;
3364
3365out:
ba5e8f2e 3366 __free_recorded_refs(&check_dirs);
31db9f7c 3367 free_recorded_refs(sctx);
924794c9 3368 fs_path_free(valid_path);
31db9f7c
AB		 3369 return ret;
3370}
3371
3372static int __record_new_ref(int num, u64 dir, int index,
3373 struct fs_path *name,
3374 void *ctx)
3375{
3376 int ret = 0;
3377 struct send_ctx *sctx = ctx;
3378 struct fs_path *p;
3379 u64 gen;
3380
924794c9 3381 p = fs_path_alloc();
31db9f7c
AB		 3382 if (!p)
3383 return -ENOMEM;
3384
3385 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 3386 NULL, NULL);
31db9f7c
AB		 3387 if (ret < 0)
3388 goto out;
3389
31db9f7c
AB		 3390 ret = get_cur_path(sctx, dir, gen, p);
3391 if (ret < 0)
3392 goto out;
3393 ret = fs_path_add_path(p, name);
3394 if (ret < 0)
3395 goto out;
3396
3397 ret = record_ref(&sctx->new_refs, dir, gen, p);
3398
3399out:
3400 if (ret)
924794c9 3401 fs_path_free(p);
31db9f7c
AB		 3402 return ret;
3403}
3404
3405static int __record_deleted_ref(int num, u64 dir, int index,
3406 struct fs_path *name,
3407 void *ctx)
3408{
3409 int ret = 0;
3410 struct send_ctx *sctx = ctx;
3411 struct fs_path *p;
3412 u64 gen;
3413
924794c9 3414 p = fs_path_alloc();
31db9f7c
AB		 3415 if (!p)
3416 return -ENOMEM;
3417
3418 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 3419 NULL, NULL);
31db9f7c
AB		 3420 if (ret < 0)
3421 goto out;
3422
3423 ret = get_cur_path(sctx, dir, gen, p);
3424 if (ret < 0)
3425 goto out;
3426 ret = fs_path_add_path(p, name);
3427 if (ret < 0)
3428 goto out;
3429
3430 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3431
3432out:
3433 if (ret)
924794c9 3434 fs_path_free(p);
31db9f7c
AB		 3435 return ret;
3436}
3437
3438static int record_new_ref(struct send_ctx *sctx)
3439{
3440 int ret;
3441
924794c9
TI		 3442 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3443 sctx->cmp_key, 0, __record_new_ref, sctx);
31db9f7c
AB		 3444 if (ret < 0)
3445 goto out;
3446 ret = 0;
3447
3448out:
3449 return ret;
3450}
3451
3452static int record_deleted_ref(struct send_ctx *sctx)
3453{
3454 int ret;
3455
924794c9
TI		 3456 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3457 sctx->cmp_key, 0, __record_deleted_ref, sctx);
31db9f7c
AB		 3458 if (ret < 0)
3459 goto out;
3460 ret = 0;
3461
3462out:
3463 return ret;
3464}
3465
3466struct find_ref_ctx {
3467 u64 dir;
ba5e8f2e
JB		 3468 u64 dir_gen;
3469 struct btrfs_root *root;
31db9f7c
AB		 3470 struct fs_path *name;
3471 int found_idx;
3472};
3473
3474static int __find_iref(int num, u64 dir, int index,
3475 struct fs_path *name,
3476 void *ctx_)
3477{
3478 struct find_ref_ctx *ctx = ctx_;
ba5e8f2e
JB		 3479 u64 dir_gen;
3480 int ret;
31db9f7c
AB		 3481
3482 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3483 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
ba5e8f2e
JB		 3484 /*
3485 * To avoid doing extra lookups we'll only do this if everything
3486 * else matches.
3487 */
3488 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3489 NULL, NULL, NULL);
3490 if (ret)
3491 return ret;
3492 if (dir_gen != ctx->dir_gen)
3493 return 0;
31db9f7c
AB		 3494 ctx->found_idx = num;
3495 return 1;
3496 }
3497 return 0;
3498}
3499
924794c9 3500static int find_iref(struct btrfs_root *root,
31db9f7c
AB		 3501 struct btrfs_path *path,
3502 struct btrfs_key *key,
ba5e8f2e 3503 u64 dir, u64 dir_gen, struct fs_path *name)
31db9f7c
AB		 3504{
3505 int ret;
3506 struct find_ref_ctx ctx;
3507
3508 ctx.dir = dir;
3509 ctx.name = name;
ba5e8f2e 3510 ctx.dir_gen = dir_gen;
31db9f7c 3511 ctx.found_idx = -1;
ba5e8f2e 3512 ctx.root = root;
31db9f7c 3513
924794c9 3514 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
31db9f7c
AB		 3515 if (ret < 0)
3516 return ret;
3517
3518 if (ctx.found_idx == -1)
3519 return -ENOENT;
3520
3521 return ctx.found_idx;
3522}
3523
3524static int __record_changed_new_ref(int num, u64 dir, int index,
3525 struct fs_path *name,
3526 void *ctx)
3527{
ba5e8f2e 3528 u64 dir_gen;
31db9f7c
AB		 3529 int ret;
3530 struct send_ctx *sctx = ctx;
3531
ba5e8f2e
JB		 3532 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3533 NULL, NULL, NULL);
3534 if (ret)
3535 return ret;
3536
924794c9 3537 ret = find_iref(sctx->parent_root, sctx->right_path,
ba5e8f2e 3538 sctx->cmp_key, dir, dir_gen, name);
31db9f7c
AB		 3539 if (ret == -ENOENT)
3540 ret = __record_new_ref(num, dir, index, name, sctx);
3541 else if (ret > 0)
3542 ret = 0;
3543
3544 return ret;
3545}
3546
3547static int __record_changed_deleted_ref(int num, u64 dir, int index,
3548 struct fs_path *name,
3549 void *ctx)
3550{
ba5e8f2e 3551 u64 dir_gen;
31db9f7c
AB		 3552 int ret;
3553 struct send_ctx *sctx = ctx;
3554
ba5e8f2e
JB		 3555 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3556 NULL, NULL, NULL);
3557 if (ret)
3558 return ret;
3559
924794c9 3560 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
ba5e8f2e 3561 dir, dir_gen, name);
31db9f7c
AB		 3562 if (ret == -ENOENT)
3563 ret = __record_deleted_ref(num, dir, index, name, sctx);
3564 else if (ret > 0)
3565 ret = 0;
3566
3567 return ret;
3568}
3569
3570static int record_changed_ref(struct send_ctx *sctx)
3571{
3572 int ret = 0;
3573
924794c9 3574 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
31db9f7c
AB		 3575 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3576 if (ret < 0)
3577 goto out;
924794c9 3578 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 3579 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3580 if (ret < 0)
3581 goto out;
3582 ret = 0;
3583
3584out:
3585 return ret;
3586}
3587
3588/*
3589 * Record and process all refs at once. Needed when an inode changes the
3590 * generation number, which means that it was deleted and recreated.
3591 */
3592static int process_all_refs(struct send_ctx *sctx,
3593 enum btrfs_compare_tree_result cmd)
3594{
3595 int ret;
3596 struct btrfs_root *root;
3597 struct btrfs_path *path;
3598 struct btrfs_key key;
3599 struct btrfs_key found_key;
3600 struct extent_buffer *eb;
3601 int slot;
3602 iterate_inode_ref_t cb;
9f03740a 3603 int pending_move = 0;
31db9f7c
AB		 3604
3605 path = alloc_path_for_send();
3606 if (!path)
3607 return -ENOMEM;
3608
3609 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3610 root = sctx->send_root;
3611 cb = __record_new_ref;
3612 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3613 root = sctx->parent_root;
3614 cb = __record_deleted_ref;
3615 } else {
4d1a63b2
DS		 3616 btrfs_err(sctx->send_root->fs_info,
3617 "Wrong command %d in process_all_refs", cmd);
3618 ret = -EINVAL;
3619 goto out;
31db9f7c
AB		 3620 }
3621
3622 key.objectid = sctx->cmp_key->objectid;
3623 key.type = BTRFS_INODE_REF_KEY;
3624 key.offset = 0;
3625 while (1) {
3626 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
e938c8ad 3627 if (ret < 0)
31db9f7c 3628 goto out;
e938c8ad 3629 if (ret)
31db9f7c 3630 break;
31db9f7c
AB		 3631
3632 eb = path->nodes[0];
3633 slot = path->slots[0];
3634 btrfs_item_key_to_cpu(eb, &found_key, slot);
3635
3636 if (found_key.objectid != key.objectid ||
96b5bd77
JS		 3637 (found_key.type != BTRFS_INODE_REF_KEY &&
3638 found_key.type != BTRFS_INODE_EXTREF_KEY))
31db9f7c 3639 break;
31db9f7c 3640
924794c9 3641 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
31db9f7c
AB		 3642 btrfs_release_path(path);
3643 if (ret < 0)
3644 goto out;
3645
3646 key.offset = found_key.offset + 1;
3647 }
e938c8ad 3648 btrfs_release_path(path);
31db9f7c 3649
9f03740a
FDBM		 3650 ret = process_recorded_refs(sctx, &pending_move);
3651 /* Only applicable to an incremental send. */
3652 ASSERT(pending_move == 0);
31db9f7c
AB		 3653
3654out:
3655 btrfs_free_path(path);
3656 return ret;
3657}
3658
3659static int send_set_xattr(struct send_ctx *sctx,
3660 struct fs_path *path,
3661 const char *name, int name_len,
3662 const char *data, int data_len)
3663{
3664 int ret = 0;
3665
3666 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3667 if (ret < 0)
3668 goto out;
3669
3670 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3671 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3672 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3673
3674 ret = send_cmd(sctx);
3675
3676tlv_put_failure:
3677out:
3678 return ret;
3679}
3680
3681static int send_remove_xattr(struct send_ctx *sctx,
3682 struct fs_path *path,
3683 const char *name, int name_len)
3684{
3685 int ret = 0;
3686
3687 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3688 if (ret < 0)
3689 goto out;
3690
3691 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3692 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3693
3694 ret = send_cmd(sctx);
3695
3696tlv_put_failure:
3697out:
3698 return ret;
3699}
3700
3701static int __process_new_xattr(int num, struct btrfs_key *di_key,
3702 const char *name, int name_len,
3703 const char *data, int data_len,
3704 u8 type, void *ctx)
3705{
3706 int ret;
3707 struct send_ctx *sctx = ctx;
3708 struct fs_path *p;
3709 posix_acl_xattr_header dummy_acl;
3710
924794c9 3711 p = fs_path_alloc();
31db9f7c
AB		 3712 if (!p)
3713 return -ENOMEM;
3714
3715 /*
3716 * This hack is needed because empty acl's are stored as zero byte
3717 * data in xattrs. Problem with that is, that receiving these zero byte
3718 * acl's will fail later. To fix this, we send a dummy acl list that
3719 * only contains the version number and no entries.
3720 */
3721 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3722 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3723 if (data_len == 0) {
3724 dummy_acl.a_version =
3725 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3726 data = (char *)&dummy_acl;
3727 data_len = sizeof(dummy_acl);
3728 }
3729 }
3730
3731 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3732 if (ret < 0)
3733 goto out;
3734
3735 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3736
3737out:
924794c9 3738 fs_path_free(p);
31db9f7c
AB		 3739 return ret;
3740}
3741
3742static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3743 const char *name, int name_len,
3744 const char *data, int data_len,
3745 u8 type, void *ctx)
3746{
3747 int ret;
3748 struct send_ctx *sctx = ctx;
3749 struct fs_path *p;
3750
924794c9 3751 p = fs_path_alloc();
31db9f7c
AB		 3752 if (!p)
3753 return -ENOMEM;
3754
3755 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3756 if (ret < 0)
3757 goto out;
3758
3759 ret = send_remove_xattr(sctx, p, name, name_len);
3760
3761out:
924794c9 3762 fs_path_free(p);
31db9f7c
AB		 3763 return ret;
3764}
3765
3766static int process_new_xattr(struct send_ctx *sctx)
3767{
3768 int ret = 0;
3769
924794c9
TI		 3770 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3771 sctx->cmp_key, __process_new_xattr, sctx);
31db9f7c
AB		 3772
3773 return ret;
3774}
3775
3776static int process_deleted_xattr(struct send_ctx *sctx)
3777{
3778 int ret;
3779
924794c9
TI		 3780 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3781 sctx->cmp_key, __process_deleted_xattr, sctx);
31db9f7c
AB		 3782
3783 return ret;
3784}
3785
3786struct find_xattr_ctx {
3787 const char *name;
3788 int name_len;
3789 int found_idx;
3790 char *found_data;
3791 int found_data_len;
3792};
3793
3794static int __find_xattr(int num, struct btrfs_key *di_key,
3795 const char *name, int name_len,
3796 const char *data, int data_len,
3797 u8 type, void *vctx)
3798{
3799 struct find_xattr_ctx *ctx = vctx;
3800
3801 if (name_len == ctx->name_len &&
3802 strncmp(name, ctx->name, name_len) == 0) {
3803 ctx->found_idx = num;
3804 ctx->found_data_len = data_len;
a5959bc0 3805 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
31db9f7c
AB		 3806 if (!ctx->found_data)
3807 return -ENOMEM;
31db9f7c
AB		 3808 return 1;
3809 }
3810 return 0;
3811}
3812
924794c9 3813static int find_xattr(struct btrfs_root *root,
31db9f7c
AB		 3814 struct btrfs_path *path,
3815 struct btrfs_key *key,
3816 const char *name, int name_len,
3817 char **data, int *data_len)
3818{
3819 int ret;
3820 struct find_xattr_ctx ctx;
3821
3822 ctx.name = name;
3823 ctx.name_len = name_len;
3824 ctx.found_idx = -1;
3825 ctx.found_data = NULL;
3826 ctx.found_data_len = 0;
3827
924794c9 3828 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
31db9f7c
AB		 3829 if (ret < 0)
3830 return ret;
3831
3832 if (ctx.found_idx == -1)
3833 return -ENOENT;
3834 if (data) {
3835 *data = ctx.found_data;
3836 *data_len = ctx.found_data_len;
3837 } else {
3838 kfree(ctx.found_data);
3839 }
3840 return ctx.found_idx;
3841}
3842
3843
3844static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3845 const char *name, int name_len,
3846 const char *data, int data_len,
3847 u8 type, void *ctx)
3848{
3849 int ret;
3850 struct send_ctx *sctx = ctx;
3851 char *found_data = NULL;
3852 int found_data_len = 0;
31db9f7c 3853
924794c9
TI		 3854 ret = find_xattr(sctx->parent_root, sctx->right_path,
3855 sctx->cmp_key, name, name_len, &found_data,
3856 &found_data_len);
31db9f7c
AB		 3857 if (ret == -ENOENT) {
3858 ret = __process_new_xattr(num, di_key, name, name_len, data,
3859 data_len, type, ctx);
3860 } else if (ret >= 0) {
3861 if (data_len != found_data_len ||
3862 memcmp(data, found_data, data_len)) {
3863 ret = __process_new_xattr(num, di_key, name, name_len,
3864 data, data_len, type, ctx);
3865 } else {
3866 ret = 0;
3867 }
3868 }
3869
3870 kfree(found_data);
31db9f7c
AB		 3871 return ret;
3872}
3873
3874static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3875 const char *name, int name_len,
3876 const char *data, int data_len,
3877 u8 type, void *ctx)
3878{
3879 int ret;
3880 struct send_ctx *sctx = ctx;
3881
924794c9
TI		 3882 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3883 name, name_len, NULL, NULL);
31db9f7c
AB		 3884 if (ret == -ENOENT)
3885 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3886 data_len, type, ctx);
3887 else if (ret >= 0)
3888 ret = 0;
3889
3890 return ret;
3891}
3892
3893static int process_changed_xattr(struct send_ctx *sctx)
3894{
3895 int ret = 0;
3896
924794c9 3897 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
31db9f7c
AB		 3898 sctx->cmp_key, __process_changed_new_xattr, sctx);
3899 if (ret < 0)
3900 goto out;
924794c9 3901 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 3902 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3903
3904out:
3905 return ret;
3906}
3907
3908static int process_all_new_xattrs(struct send_ctx *sctx)
3909{
3910 int ret;
3911 struct btrfs_root *root;
3912 struct btrfs_path *path;
3913 struct btrfs_key key;
3914 struct btrfs_key found_key;
3915 struct extent_buffer *eb;
3916 int slot;
3917
3918 path = alloc_path_for_send();
3919 if (!path)
3920 return -ENOMEM;
3921
3922 root = sctx->send_root;
3923
3924 key.objectid = sctx->cmp_key->objectid;
3925 key.type = BTRFS_XATTR_ITEM_KEY;
3926 key.offset = 0;
3927 while (1) {
3928 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3929 if (ret < 0)
3930 goto out;
3931 if (ret) {
3932 ret = 0;
3933 goto out;
3934 }
3935
3936 eb = path->nodes[0];
3937 slot = path->slots[0];
3938 btrfs_item_key_to_cpu(eb, &found_key, slot);
3939
3940 if (found_key.objectid != key.objectid ||
3941 found_key.type != key.type) {
3942 ret = 0;
3943 goto out;
3944 }
3945
924794c9
TI		 3946 ret = iterate_dir_item(root, path, &found_key,
3947 __process_new_xattr, sctx);
31db9f7c
AB		 3948 if (ret < 0)
3949 goto out;
3950
3951 btrfs_release_path(path);
3952 key.offset = found_key.offset + 1;
3953 }
3954
3955out:
3956 btrfs_free_path(path);
3957 return ret;
3958}
3959
ed259095
JB		 3960static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3961{
3962 struct btrfs_root *root = sctx->send_root;
3963 struct btrfs_fs_info *fs_info = root->fs_info;
3964 struct inode *inode;
3965 struct page *page;
3966 char *addr;
3967 struct btrfs_key key;
3968 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3969 pgoff_t last_index;
3970 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3971 ssize_t ret = 0;
3972
3973 key.objectid = sctx->cur_ino;
3974 key.type = BTRFS_INODE_ITEM_KEY;
3975 key.offset = 0;
3976
3977 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3978 if (IS_ERR(inode))
3979 return PTR_ERR(inode);
3980
3981 if (offset + len > i_size_read(inode)) {
3982 if (offset > i_size_read(inode))
3983 len = 0;
3984 else
3985 len = offset - i_size_read(inode);
3986 }
3987 if (len == 0)
3988 goto out;
3989
3990 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
3991 while (index <= last_index) {
3992 unsigned cur_len = min_t(unsigned, len,
3993 PAGE_CACHE_SIZE - pg_offset);
3994 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3995 if (!page) {
3996 ret = -ENOMEM;
3997 break;
3998 }
3999
4000 if (!PageUptodate(page)) {
4001 btrfs_readpage(NULL, page);
4002 lock_page(page);
4003 if (!PageUptodate(page)) {
4004 unlock_page(page);
4005 page_cache_release(page);
4006 ret = -EIO;
4007 break;
4008 }
4009 }
4010
4011 addr = kmap(page);
4012 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4013 kunmap(page);
4014 unlock_page(page);
4015 page_cache_release(page);
4016 index++;
4017 pg_offset = 0;
4018 len -= cur_len;
4019 ret += cur_len;
4020 }
4021out:
4022 iput(inode);
4023 return ret;
4024}
4025
31db9f7c
AB		 4026/*
4027 * Read some bytes from the current inode/file and send a write command to
4028 * user space.
4029 */
4030static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4031{
4032 int ret = 0;
4033 struct fs_path *p;
ed259095 4034 ssize_t num_read = 0;
31db9f7c 4035
924794c9 4036 p = fs_path_alloc();
31db9f7c
AB		 4037 if (!p)
4038 return -ENOMEM;
4039
31db9f7c
AB		 4040verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4041
ed259095
JB		 4042 num_read = fill_read_buf(sctx, offset, len);
4043 if (num_read <= 0) {
4044 if (num_read < 0)
4045 ret = num_read;
31db9f7c 4046 goto out;
ed259095 4047 }
31db9f7c
AB		 4048
4049 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4050 if (ret < 0)
4051 goto out;
4052
4053 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4054 if (ret < 0)
4055 goto out;
4056
4057 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4058 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
e938c8ad 4059 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
31db9f7c
AB		 4060
4061 ret = send_cmd(sctx);
4062
4063tlv_put_failure:
4064out:
924794c9 4065 fs_path_free(p);
31db9f7c
AB		 4066 if (ret < 0)
4067 return ret;
e938c8ad 4068 return num_read;
31db9f7c
AB		 4069}
4070
4071/*
4072 * Send a clone command to user space.
4073 */
4074static int send_clone(struct send_ctx *sctx,
4075 u64 offset, u32 len,
4076 struct clone_root *clone_root)
4077{
4078 int ret = 0;
31db9f7c
AB		 4079 struct fs_path *p;
4080 u64 gen;
4081
4082verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4083 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4084 clone_root->root->objectid, clone_root->ino,
4085 clone_root->offset);
4086
924794c9 4087 p = fs_path_alloc();
31db9f7c
AB		 4088 if (!p)
4089 return -ENOMEM;
4090
4091 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4092 if (ret < 0)
4093 goto out;
4094
4095 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4096 if (ret < 0)
4097 goto out;
4098
4099 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4100 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4101 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4102
e938c8ad 4103 if (clone_root->root == sctx->send_root) {
31db9f7c 4104 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
85a7b33b 4105 &gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 4106 if (ret < 0)
4107 goto out;
4108 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4109 } else {
924794c9 4110 ret = get_inode_path(clone_root->root, clone_root->ino, p);
31db9f7c
AB		 4111 }
4112 if (ret < 0)
4113 goto out;
4114
4115 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
e938c8ad 4116 clone_root->root->root_item.uuid);
31db9f7c 4117 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
5a0f4e2c 4118 le64_to_cpu(clone_root->root->root_item.ctransid));
31db9f7c
AB		 4119 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4120 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4121 clone_root->offset);
4122
4123 ret = send_cmd(sctx);
4124
4125tlv_put_failure:
4126out:
924794c9 4127 fs_path_free(p);
31db9f7c
AB		 4128 return ret;
4129}
4130
cb95e7bf
MF		 4131/*
4132 * Send an update extent command to user space.
4133 */
4134static int send_update_extent(struct send_ctx *sctx,
4135 u64 offset, u32 len)
4136{
4137 int ret = 0;
4138 struct fs_path *p;
4139
924794c9 4140 p = fs_path_alloc();
cb95e7bf
MF		 4141 if (!p)
4142 return -ENOMEM;
4143
4144 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4145 if (ret < 0)
4146 goto out;
4147
4148 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4149 if (ret < 0)
4150 goto out;
4151
4152 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4153 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4154 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4155
4156 ret = send_cmd(sctx);
4157
4158tlv_put_failure:
4159out:
924794c9 4160 fs_path_free(p);
cb95e7bf
MF		 4161 return ret;
4162}
4163
16e7549f
JB		 4164static int send_hole(struct send_ctx *sctx, u64 end)
4165{
4166 struct fs_path *p = NULL;
4167 u64 offset = sctx->cur_inode_last_extent;
4168 u64 len;
4169 int ret = 0;
4170
4171 p = fs_path_alloc();
4172 if (!p)
4173 return -ENOMEM;
4174 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4175 while (offset < end) {
4176 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4177
4178 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4179 if (ret < 0)
4180 break;
4181 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4182 if (ret < 0)
4183 break;
4184 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4185 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4186 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4187 ret = send_cmd(sctx);
4188 if (ret < 0)
4189 break;
4190 offset += len;
4191 }
4192tlv_put_failure:
4193 fs_path_free(p);
4194 return ret;
4195}
4196
31db9f7c
AB		 4197static int send_write_or_clone(struct send_ctx *sctx,
4198 struct btrfs_path *path,
4199 struct btrfs_key *key,
4200 struct clone_root *clone_root)
4201{
4202 int ret = 0;
4203 struct btrfs_file_extent_item *ei;
4204 u64 offset = key->offset;
4205 u64 pos = 0;
4206 u64 len;
4207 u32 l;
4208 u8 type;
28e5dd8f 4209 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
31db9f7c
AB		 4210
4211 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4212 struct btrfs_file_extent_item);
4213 type = btrfs_file_extent_type(path->nodes[0], ei);
74dd17fb 4214 if (type == BTRFS_FILE_EXTENT_INLINE) {
514ac8ad
CM		 4215 len = btrfs_file_extent_inline_len(path->nodes[0],
4216 path->slots[0], ei);
74dd17fb
CM		 4217 /*
4218 * it is possible the inline item won't cover the whole page,
4219 * but there may be items after this page. Make
4220 * sure to send the whole thing
4221 */
4222 len = PAGE_CACHE_ALIGN(len);
4223 } else {
31db9f7c 4224 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
74dd17fb 4225 }
31db9f7c
AB		 4226
4227 if (offset + len > sctx->cur_inode_size)
4228 len = sctx->cur_inode_size - offset;
4229 if (len == 0) {
4230 ret = 0;
4231 goto out;
4232 }
4233
28e5dd8f 4234 if (clone_root && IS_ALIGNED(offset + len, bs)) {
cb95e7bf
MF		 4235 ret = send_clone(sctx, offset, len, clone_root);
4236 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4237 ret = send_update_extent(sctx, offset, len);
4238 } else {
31db9f7c
AB		 4239 while (pos < len) {
4240 l = len - pos;
4241 if (l > BTRFS_SEND_READ_SIZE)
4242 l = BTRFS_SEND_READ_SIZE;
4243 ret = send_write(sctx, pos + offset, l);
4244 if (ret < 0)
4245 goto out;
4246 if (!ret)
4247 break;
4248 pos += ret;
4249 }
4250 ret = 0;
31db9f7c 4251 }
31db9f7c
AB		 4252out:
4253 return ret;
4254}
4255
4256static int is_extent_unchanged(struct send_ctx *sctx,
4257 struct btrfs_path *left_path,
4258 struct btrfs_key *ekey)
4259{
4260 int ret = 0;
4261 struct btrfs_key key;
4262 struct btrfs_path *path = NULL;
4263 struct extent_buffer *eb;
4264 int slot;
4265 struct btrfs_key found_key;
4266 struct btrfs_file_extent_item *ei;
4267 u64 left_disknr;
4268 u64 right_disknr;
4269 u64 left_offset;
4270 u64 right_offset;
4271 u64 left_offset_fixed;
4272 u64 left_len;
4273 u64 right_len;
74dd17fb
CM		 4274 u64 left_gen;
4275 u64 right_gen;
31db9f7c
AB		 4276 u8 left_type;
4277 u8 right_type;
4278
4279 path = alloc_path_for_send();
4280 if (!path)
4281 return -ENOMEM;
4282
4283 eb = left_path->nodes[0];
4284 slot = left_path->slots[0];
31db9f7c
AB		 4285 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4286 left_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 4287
4288 if (left_type != BTRFS_FILE_EXTENT_REG) {
4289 ret = 0;
4290 goto out;
4291 }
74dd17fb
CM		 4292 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4293 left_len = btrfs_file_extent_num_bytes(eb, ei);
4294 left_offset = btrfs_file_extent_offset(eb, ei);
4295 left_gen = btrfs_file_extent_generation(eb, ei);
31db9f7c
AB		 4296
4297 /*
4298 * Following comments will refer to these graphics. L is the left
4299 * extents which we are checking at the moment. 1-8 are the right
4300 * extents that we iterate.
4301 *
4302 * |-----L-----|
4303 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4304 *
4305 * |-----L-----|
4306 * |--1--|-2b-|...(same as above)
4307 *
4308 * Alternative situation. Happens on files where extents got split.
4309 * |-----L-----|
4310 * |-----------7-----------|-6-|
4311 *
4312 * Alternative situation. Happens on files which got larger.
4313 * |-----L-----|
4314 * |-8-|
4315 * Nothing follows after 8.
4316 */
4317
4318 key.objectid = ekey->objectid;
4319 key.type = BTRFS_EXTENT_DATA_KEY;
4320 key.offset = ekey->offset;
4321 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4322 if (ret < 0)
4323 goto out;
4324 if (ret) {
4325 ret = 0;
4326 goto out;
4327 }
4328
4329 /*
4330 * Handle special case where the right side has no extents at all.
4331 */
4332 eb = path->nodes[0];
4333 slot = path->slots[0];
4334 btrfs_item_key_to_cpu(eb, &found_key, slot);
4335 if (found_key.objectid != key.objectid ||
4336 found_key.type != key.type) {
57cfd462
JB		 4337 /* If we're a hole then just pretend nothing changed */
4338 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 4339 goto out;
4340 }
4341
4342 /*
4343 * We're now on 2a, 2b or 7.
4344 */
4345 key = found_key;
4346 while (key.offset < ekey->offset + left_len) {
4347 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4348 right_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 4349 if (right_type != BTRFS_FILE_EXTENT_REG) {
4350 ret = 0;
4351 goto out;
4352 }
4353
007d31f7
JB		 4354 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4355 right_len = btrfs_file_extent_num_bytes(eb, ei);
4356 right_offset = btrfs_file_extent_offset(eb, ei);
4357 right_gen = btrfs_file_extent_generation(eb, ei);
4358
31db9f7c
AB		 4359 /*
4360 * Are we at extent 8? If yes, we know the extent is changed.
4361 * This may only happen on the first iteration.
4362 */
d8347fa4 4363 if (found_key.offset + right_len <= ekey->offset) {
57cfd462
JB		 4364 /* If we're a hole just pretend nothing changed */
4365 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 4366 goto out;
4367 }
4368
4369 left_offset_fixed = left_offset;
4370 if (key.offset < ekey->offset) {
4371 /* Fix the right offset for 2a and 7. */
4372 right_offset += ekey->offset - key.offset;
4373 } else {
4374 /* Fix the left offset for all behind 2a and 2b */
4375 left_offset_fixed += key.offset - ekey->offset;
4376 }
4377
4378 /*
4379 * Check if we have the same extent.
4380 */
3954096d 4381 if (left_disknr != right_disknr ||
74dd17fb
CM		 4382 left_offset_fixed != right_offset ||
4383 left_gen != right_gen) {
31db9f7c
AB		 4384 ret = 0;
4385 goto out;
4386 }
4387
4388 /*
4389 * Go to the next extent.
4390 */
4391 ret = btrfs_next_item(sctx->parent_root, path);
4392 if (ret < 0)
4393 goto out;
4394 if (!ret) {
4395 eb = path->nodes[0];
4396 slot = path->slots[0];
4397 btrfs_item_key_to_cpu(eb, &found_key, slot);
4398 }
4399 if (ret || found_key.objectid != key.objectid ||
4400 found_key.type != key.type) {
4401 key.offset += right_len;
4402 break;
adaa4b8e
JS		 4403 }
4404 if (found_key.offset != key.offset + right_len) {
4405 ret = 0;
4406 goto out;
31db9f7c
AB		 4407 }
4408 key = found_key;
4409 }
4410
4411 /*
4412 * We're now behind the left extent (treat as unchanged) or at the end
4413 * of the right side (treat as changed).
4414 */
4415 if (key.offset >= ekey->offset + left_len)
4416 ret = 1;
4417 else
4418 ret = 0;
4419
4420
4421out:
4422 btrfs_free_path(path);
4423 return ret;
4424}
4425
16e7549f
JB		 4426static int get_last_extent(struct send_ctx *sctx, u64 offset)
4427{
4428 struct btrfs_path *path;
4429 struct btrfs_root *root = sctx->send_root;
4430 struct btrfs_file_extent_item *fi;
4431 struct btrfs_key key;
4432 u64 extent_end;
4433 u8 type;
4434 int ret;
4435
4436 path = alloc_path_for_send();
4437 if (!path)
4438 return -ENOMEM;
4439
4440 sctx->cur_inode_last_extent = 0;
4441
4442 key.objectid = sctx->cur_ino;
4443 key.type = BTRFS_EXTENT_DATA_KEY;
4444 key.offset = offset;
4445 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4446 if (ret < 0)
4447 goto out;
4448 ret = 0;
4449 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4450 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4451 goto out;
4452
4453 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4454 struct btrfs_file_extent_item);
4455 type = btrfs_file_extent_type(path->nodes[0], fi);
4456 if (type == BTRFS_FILE_EXTENT_INLINE) {
514ac8ad
CM		 4457 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4458 path->slots[0], fi);
16e7549f
JB		 4459 extent_end = ALIGN(key.offset + size,
4460 sctx->send_root->sectorsize);
4461 } else {
4462 extent_end = key.offset +
4463 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4464 }
4465 sctx->cur_inode_last_extent = extent_end;
4466out:
4467 btrfs_free_path(path);
4468 return ret;
4469}
4470
4471static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4472 struct btrfs_key *key)
4473{
4474 struct btrfs_file_extent_item *fi;
4475 u64 extent_end;
4476 u8 type;
4477 int ret = 0;
4478
4479 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4480 return 0;
4481
4482 if (sctx->cur_inode_last_extent == (u64)-1) {
4483 ret = get_last_extent(sctx, key->offset - 1);
4484 if (ret)
4485 return ret;
4486 }
4487
4488 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4489 struct btrfs_file_extent_item);
4490 type = btrfs_file_extent_type(path->nodes[0], fi);
4491 if (type == BTRFS_FILE_EXTENT_INLINE) {
514ac8ad
CM		 4492 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4493 path->slots[0], fi);
16e7549f
JB		 4494 extent_end = ALIGN(key->offset + size,
4495 sctx->send_root->sectorsize);
4496 } else {
4497 extent_end = key->offset +
4498 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4499 }
bf54f412
FDBM		 4500
4501 if (path->slots[0] == 0 &&
4502 sctx->cur_inode_last_extent < key->offset) {
4503 /*
4504 * We might have skipped entire leafs that contained only
4505 * file extent items for our current inode. These leafs have
4506 * a generation number smaller (older) than the one in the
4507 * current leaf and the leaf our last extent came from, and
4508 * are located between these 2 leafs.
4509 */
4510 ret = get_last_extent(sctx, key->offset - 1);
4511 if (ret)
4512 return ret;
4513 }
4514
16e7549f
JB		 4515 if (sctx->cur_inode_last_extent < key->offset)
4516 ret = send_hole(sctx, key->offset);
4517 sctx->cur_inode_last_extent = extent_end;
4518 return ret;
4519}
4520
31db9f7c
AB		 4521static int process_extent(struct send_ctx *sctx,
4522 struct btrfs_path *path,
4523 struct btrfs_key *key)
4524{
31db9f7c 4525 struct clone_root *found_clone = NULL;
57cfd462 4526 int ret = 0;
31db9f7c
AB		 4527
4528 if (S_ISLNK(sctx->cur_inode_mode))
4529 return 0;
4530
4531 if (sctx->parent_root && !sctx->cur_inode_new) {
4532 ret = is_extent_unchanged(sctx, path, key);
4533 if (ret < 0)
4534 goto out;
4535 if (ret) {
4536 ret = 0;
16e7549f 4537 goto out_hole;
31db9f7c 4538 }
57cfd462
JB		 4539 } else {
4540 struct btrfs_file_extent_item *ei;
4541 u8 type;
4542
4543 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4544 struct btrfs_file_extent_item);
4545 type = btrfs_file_extent_type(path->nodes[0], ei);
4546 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4547 type == BTRFS_FILE_EXTENT_REG) {
4548 /*
4549 * The send spec does not have a prealloc command yet,
4550 * so just leave a hole for prealloc'ed extents until
4551 * we have enough commands queued up to justify rev'ing
4552 * the send spec.
4553 */
4554 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4555 ret = 0;
4556 goto out;
4557 }
4558
4559 /* Have a hole, just skip it. */
4560 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4561 ret = 0;
4562 goto out;
4563 }
4564 }
31db9f7c
AB		 4565 }
4566
4567 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4568 sctx->cur_inode_size, &found_clone);
4569 if (ret != -ENOENT && ret < 0)
4570 goto out;
4571
4572 ret = send_write_or_clone(sctx, path, key, found_clone);
16e7549f
JB		 4573 if (ret)
4574 goto out;
4575out_hole:
4576 ret = maybe_send_hole(sctx, path, key);
31db9f7c
AB		 4577out:
4578 return ret;
4579}
4580
4581static int process_all_extents(struct send_ctx *sctx)
4582{
4583 int ret;
4584 struct btrfs_root *root;
4585 struct btrfs_path *path;
4586 struct btrfs_key key;
4587 struct btrfs_key found_key;
4588 struct extent_buffer *eb;
4589 int slot;
4590
4591 root = sctx->send_root;
4592 path = alloc_path_for_send();
4593 if (!path)
4594 return -ENOMEM;
4595
4596 key.objectid = sctx->cmp_key->objectid;
4597 key.type = BTRFS_EXTENT_DATA_KEY;
4598 key.offset = 0;
7fdd29d0
FDBM		 4599 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4600 if (ret < 0)
4601 goto out;
31db9f7c 4602
7fdd29d0 4603 while (1) {
31db9f7c
AB		 4604 eb = path->nodes[0];
4605 slot = path->slots[0];
7fdd29d0
FDBM		 4606
4607 if (slot >= btrfs_header_nritems(eb)) {
4608 ret = btrfs_next_leaf(root, path);
4609 if (ret < 0) {
4610 goto out;
4611 } else if (ret > 0) {
4612 ret = 0;
4613 break;
4614 }
4615 continue;
4616 }
4617
31db9f7c
AB		 4618 btrfs_item_key_to_cpu(eb, &found_key, slot);
4619
4620 if (found_key.objectid != key.objectid ||
4621 found_key.type != key.type) {
4622 ret = 0;
4623 goto out;
4624 }
4625
4626 ret = process_extent(sctx, path, &found_key);
4627 if (ret < 0)
4628 goto out;
4629
7fdd29d0 4630 path->slots[0]++;
31db9f7c
AB		 4631 }
4632
4633out:
4634 btrfs_free_path(path);
4635 return ret;
4636}
4637
9f03740a
FDBM		 4638static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4639 int *pending_move,
4640 int *refs_processed)
31db9f7c
AB		 4641{
4642 int ret = 0;
4643
4644 if (sctx->cur_ino == 0)
4645 goto out;
4646 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
96b5bd77 4647 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 4648 goto out;
4649 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4650 goto out;
4651
9f03740a 4652 ret = process_recorded_refs(sctx, pending_move);
e479d9bb
AB		 4653 if (ret < 0)
4654 goto out;
4655
9f03740a 4656 *refs_processed = 1;
31db9f7c
AB		 4657out:
4658 return ret;
4659}
4660
4661static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4662{
4663 int ret = 0;
4664 u64 left_mode;
4665 u64 left_uid;
4666 u64 left_gid;
4667 u64 right_mode;
4668 u64 right_uid;
4669 u64 right_gid;
4670 int need_chmod = 0;
4671 int need_chown = 0;
9f03740a
FDBM		 4672 int pending_move = 0;
4673 int refs_processed = 0;
31db9f7c 4674
9f03740a
FDBM		 4675 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4676 &refs_processed);
31db9f7c
AB		 4677 if (ret < 0)
4678 goto out;
4679
9f03740a
FDBM		 4680 /*
4681 * We have processed the refs and thus need to advance send_progress.
4682 * Now, calls to get_cur_xxx will take the updated refs of the current
4683 * inode into account.
4684 *
4685 * On the other hand, if our current inode is a directory and couldn't
4686 * be moved/renamed because its parent was renamed/moved too and it has
4687 * a higher inode number, we can only move/rename our current inode
4688 * after we moved/renamed its parent. Therefore in this case operate on
4689 * the old path (pre move/rename) of our current inode, and the
4690 * move/rename will be performed later.
4691 */
4692 if (refs_processed && !pending_move)
4693 sctx->send_progress = sctx->cur_ino + 1;
4694
31db9f7c
AB		 4695 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4696 goto out;
4697 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4698 goto out;
4699
4700 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
85a7b33b 4701 &left_mode, &left_uid, &left_gid, NULL);
31db9f7c
AB		 4702 if (ret < 0)
4703 goto out;
4704
e2d044fe
AL		 4705 if (!sctx->parent_root || sctx->cur_inode_new) {
4706 need_chown = 1;
4707 if (!S_ISLNK(sctx->cur_inode_mode))
31db9f7c 4708 need_chmod = 1;
e2d044fe
AL		 4709 } else {
4710 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4711 NULL, NULL, &right_mode, &right_uid,
4712 &right_gid, NULL);
4713 if (ret < 0)
4714 goto out;
31db9f7c 4715
e2d044fe
AL		 4716 if (left_uid != right_uid || left_gid != right_gid)
4717 need_chown = 1;
4718 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4719 need_chmod = 1;
31db9f7c
AB		 4720 }
4721
4722 if (S_ISREG(sctx->cur_inode_mode)) {
16e7549f
JB		 4723 if (need_send_hole(sctx)) {
4724 if (sctx->cur_inode_last_extent == (u64)-1) {
4725 ret = get_last_extent(sctx, (u64)-1);
4726 if (ret)
4727 goto out;
4728 }
4729 if (sctx->cur_inode_last_extent <
4730 sctx->cur_inode_size) {
4731 ret = send_hole(sctx, sctx->cur_inode_size);
4732 if (ret)
4733 goto out;
4734 }
4735 }
31db9f7c
AB		 4736 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4737 sctx->cur_inode_size);
4738 if (ret < 0)
4739 goto out;
4740 }
4741
4742 if (need_chown) {
4743 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4744 left_uid, left_gid);
4745 if (ret < 0)
4746 goto out;
4747 }
4748 if (need_chmod) {
4749 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4750 left_mode);
4751 if (ret < 0)
4752 goto out;
4753 }
4754
4755 /*
9f03740a
FDBM		 4756 * If other directory inodes depended on our current directory
4757 * inode's move/rename, now do their move/rename operations.
31db9f7c 4758 */
9f03740a
FDBM		 4759 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
4760 ret = apply_children_dir_moves(sctx);
4761 if (ret)
4762 goto out;
4763 }
4764
4765 /*
4766 * Need to send that every time, no matter if it actually
4767 * changed between the two trees as we have done changes to
4768 * the inode before.
4769 */
4770 sctx->send_progress = sctx->cur_ino + 1;
31db9f7c
AB		 4771 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4772 if (ret < 0)
4773 goto out;
4774
4775out:
4776 return ret;
4777}
4778
4779static int changed_inode(struct send_ctx *sctx,
4780 enum btrfs_compare_tree_result result)
4781{
4782 int ret = 0;
4783 struct btrfs_key *key = sctx->cmp_key;
4784 struct btrfs_inode_item *left_ii = NULL;
4785 struct btrfs_inode_item *right_ii = NULL;
4786 u64 left_gen = 0;
4787 u64 right_gen = 0;
4788
31db9f7c
AB		 4789 sctx->cur_ino = key->objectid;
4790 sctx->cur_inode_new_gen = 0;
16e7549f 4791 sctx->cur_inode_last_extent = (u64)-1;
e479d9bb
AB		 4792
4793 /*
4794 * Set send_progress to current inode. This will tell all get_cur_xxx
4795 * functions that the current inode's refs are not updated yet. Later,
4796 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4797 */
31db9f7c
AB		 4798 sctx->send_progress = sctx->cur_ino;
4799
4800 if (result == BTRFS_COMPARE_TREE_NEW ||
4801 result == BTRFS_COMPARE_TREE_CHANGED) {
4802 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4803 sctx->left_path->slots[0],
4804 struct btrfs_inode_item);
4805 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4806 left_ii);
4807 } else {
4808 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4809 sctx->right_path->slots[0],
4810 struct btrfs_inode_item);
4811 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4812 right_ii);
4813 }
4814 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4815 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4816 sctx->right_path->slots[0],
4817 struct btrfs_inode_item);
4818
4819 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4820 right_ii);
6d85ed05
AB		 4821
4822 /*
4823 * The cur_ino = root dir case is special here. We can't treat
4824 * the inode as deleted+reused because it would generate a
4825 * stream that tries to delete/mkdir the root dir.
4826 */
4827 if (left_gen != right_gen &&
4828 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
31db9f7c
AB		 4829 sctx->cur_inode_new_gen = 1;
4830 }
4831
4832 if (result == BTRFS_COMPARE_TREE_NEW) {
4833 sctx->cur_inode_gen = left_gen;
4834 sctx->cur_inode_new = 1;
4835 sctx->cur_inode_deleted = 0;
4836 sctx->cur_inode_size = btrfs_inode_size(
4837 sctx->left_path->nodes[0], left_ii);
4838 sctx->cur_inode_mode = btrfs_inode_mode(
4839 sctx->left_path->nodes[0], left_ii);
4840 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
1f4692da 4841 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4842 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4843 sctx->cur_inode_gen = right_gen;
4844 sctx->cur_inode_new = 0;
4845 sctx->cur_inode_deleted = 1;
4846 sctx->cur_inode_size = btrfs_inode_size(
4847 sctx->right_path->nodes[0], right_ii);
4848 sctx->cur_inode_mode = btrfs_inode_mode(
4849 sctx->right_path->nodes[0], right_ii);
4850 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
766702ef
AB		 4851 /*
4852 * We need to do some special handling in case the inode was
4853 * reported as changed with a changed generation number. This
4854 * means that the original inode was deleted and new inode
4855 * reused the same inum. So we have to treat the old inode as
4856 * deleted and the new one as new.
4857 */
31db9f7c 4858 if (sctx->cur_inode_new_gen) {
766702ef
AB		 4859 /*
4860 * First, process the inode as if it was deleted.
4861 */
31db9f7c
AB		 4862 sctx->cur_inode_gen = right_gen;
4863 sctx->cur_inode_new = 0;
4864 sctx->cur_inode_deleted = 1;
4865 sctx->cur_inode_size = btrfs_inode_size(
4866 sctx->right_path->nodes[0], right_ii);
4867 sctx->cur_inode_mode = btrfs_inode_mode(
4868 sctx->right_path->nodes[0], right_ii);
4869 ret = process_all_refs(sctx,
4870 BTRFS_COMPARE_TREE_DELETED);
4871 if (ret < 0)
4872 goto out;
4873
766702ef
AB		 4874 /*
4875 * Now process the inode as if it was new.
4876 */
31db9f7c
AB		 4877 sctx->cur_inode_gen = left_gen;
4878 sctx->cur_inode_new = 1;
4879 sctx->cur_inode_deleted = 0;
4880 sctx->cur_inode_size = btrfs_inode_size(
4881 sctx->left_path->nodes[0], left_ii);
4882 sctx->cur_inode_mode = btrfs_inode_mode(
4883 sctx->left_path->nodes[0], left_ii);
1f4692da 4884 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4885 if (ret < 0)
4886 goto out;
4887
4888 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4889 if (ret < 0)
4890 goto out;
e479d9bb
AB		 4891 /*
4892 * Advance send_progress now as we did not get into
4893 * process_recorded_refs_if_needed in the new_gen case.
4894 */
4895 sctx->send_progress = sctx->cur_ino + 1;
766702ef
AB		 4896
4897 /*
4898 * Now process all extents and xattrs of the inode as if
4899 * they were all new.
4900 */
31db9f7c
AB		 4901 ret = process_all_extents(sctx);
4902 if (ret < 0)
4903 goto out;
4904 ret = process_all_new_xattrs(sctx);
4905 if (ret < 0)
4906 goto out;
4907 } else {
4908 sctx->cur_inode_gen = left_gen;
4909 sctx->cur_inode_new = 0;
4910 sctx->cur_inode_new_gen = 0;
4911 sctx->cur_inode_deleted = 0;
4912 sctx->cur_inode_size = btrfs_inode_size(
4913 sctx->left_path->nodes[0], left_ii);
4914 sctx->cur_inode_mode = btrfs_inode_mode(
4915 sctx->left_path->nodes[0], left_ii);
4916 }
4917 }
4918
4919out:
4920 return ret;
4921}
4922
766702ef
AB		 4923/*
4924 * We have to process new refs before deleted refs, but compare_trees gives us
4925 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4926 * first and later process them in process_recorded_refs.
4927 * For the cur_inode_new_gen case, we skip recording completely because
4928 * changed_inode did already initiate processing of refs. The reason for this is
4929 * that in this case, compare_tree actually compares the refs of 2 different
4930 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4931 * refs of the right tree as deleted and all refs of the left tree as new.
4932 */
31db9f7c
AB		 4933static int changed_ref(struct send_ctx *sctx,
4934 enum btrfs_compare_tree_result result)
4935{
4936 int ret = 0;
4937
4938 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4939
4940 if (!sctx->cur_inode_new_gen &&
4941 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4942 if (result == BTRFS_COMPARE_TREE_NEW)
4943 ret = record_new_ref(sctx);
4944 else if (result == BTRFS_COMPARE_TREE_DELETED)
4945 ret = record_deleted_ref(sctx);
4946 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4947 ret = record_changed_ref(sctx);
4948 }
4949
4950 return ret;
4951}
4952
766702ef
AB		 4953/*
4954 * Process new/deleted/changed xattrs. We skip processing in the
4955 * cur_inode_new_gen case because changed_inode did already initiate processing
4956 * of xattrs. The reason is the same as in changed_ref
4957 */
31db9f7c
AB		 4958static int changed_xattr(struct send_ctx *sctx,
4959 enum btrfs_compare_tree_result result)
4960{
4961 int ret = 0;
4962
4963 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4964
4965 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4966 if (result == BTRFS_COMPARE_TREE_NEW)
4967 ret = process_new_xattr(sctx);
4968 else if (result == BTRFS_COMPARE_TREE_DELETED)
4969 ret = process_deleted_xattr(sctx);
4970 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4971 ret = process_changed_xattr(sctx);
4972 }
4973
4974 return ret;
4975}
4976
766702ef
AB		 4977/*
4978 * Process new/deleted/changed extents. We skip processing in the
4979 * cur_inode_new_gen case because changed_inode did already initiate processing
4980 * of extents. The reason is the same as in changed_ref
4981 */
31db9f7c
AB		 4982static int changed_extent(struct send_ctx *sctx,
4983 enum btrfs_compare_tree_result result)
4984{
4985 int ret = 0;
4986
4987 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4988
4989 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4990 if (result != BTRFS_COMPARE_TREE_DELETED)
4991 ret = process_extent(sctx, sctx->left_path,
4992 sctx->cmp_key);
4993 }
4994
4995 return ret;
4996}
4997
ba5e8f2e
JB		 4998static int dir_changed(struct send_ctx *sctx, u64 dir)
4999{
5000 u64 orig_gen, new_gen;
5001 int ret;
5002
5003 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5004 NULL, NULL);
5005 if (ret)
5006 return ret;
5007
5008 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5009 NULL, NULL, NULL);
5010 if (ret)
5011 return ret;
5012
5013 return (orig_gen != new_gen) ? 1 : 0;
5014}
5015
5016static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5017 struct btrfs_key *key)
5018{
5019 struct btrfs_inode_extref *extref;
5020 struct extent_buffer *leaf;
5021 u64 dirid = 0, last_dirid = 0;
5022 unsigned long ptr;
5023 u32 item_size;
5024 u32 cur_offset = 0;
5025 int ref_name_len;
5026 int ret = 0;
5027
5028 /* Easy case, just check this one dirid */
5029 if (key->type == BTRFS_INODE_REF_KEY) {
5030 dirid = key->offset;
5031
5032 ret = dir_changed(sctx, dirid);
5033 goto out;
5034 }
5035
5036 leaf = path->nodes[0];
5037 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5038 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5039 while (cur_offset < item_size) {
5040 extref = (struct btrfs_inode_extref *)(ptr +
5041 cur_offset);
5042 dirid = btrfs_inode_extref_parent(leaf, extref);
5043 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5044 cur_offset += ref_name_len + sizeof(*extref);
5045 if (dirid == last_dirid)
5046 continue;
5047 ret = dir_changed(sctx, dirid);
5048 if (ret)
5049 break;
5050 last_dirid = dirid;
5051 }
5052out:
5053 return ret;
5054}
5055
766702ef
AB		 5056/*
5057 * Updates compare related fields in sctx and simply forwards to the actual
5058 * changed_xxx functions.
5059 */
31db9f7c
AB		 5060static int changed_cb(struct btrfs_root *left_root,
5061 struct btrfs_root *right_root,
5062 struct btrfs_path *left_path,
5063 struct btrfs_path *right_path,
5064 struct btrfs_key *key,
5065 enum btrfs_compare_tree_result result,
5066 void *ctx)
5067{
5068 int ret = 0;
5069 struct send_ctx *sctx = ctx;
5070
ba5e8f2e 5071 if (result == BTRFS_COMPARE_TREE_SAME) {
16e7549f
JB		 5072 if (key->type == BTRFS_INODE_REF_KEY ||
5073 key->type == BTRFS_INODE_EXTREF_KEY) {
5074 ret = compare_refs(sctx, left_path, key);
5075 if (!ret)
5076 return 0;
5077 if (ret < 0)
5078 return ret;
5079 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5080 return maybe_send_hole(sctx, left_path, key);
5081 } else {
ba5e8f2e 5082 return 0;
16e7549f 5083 }
ba5e8f2e
JB		 5084 result = BTRFS_COMPARE_TREE_CHANGED;
5085 ret = 0;
5086 }
5087
31db9f7c
AB		 5088 sctx->left_path = left_path;
5089 sctx->right_path = right_path;
5090 sctx->cmp_key = key;
5091
5092 ret = finish_inode_if_needed(sctx, 0);
5093 if (ret < 0)
5094 goto out;
5095
2981e225
AB		 5096 /* Ignore non-FS objects */
5097 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5098 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5099 goto out;
5100
31db9f7c
AB		 5101 if (key->type == BTRFS_INODE_ITEM_KEY)
5102 ret = changed_inode(sctx, result);
96b5bd77
JS		 5103 else if (key->type == BTRFS_INODE_REF_KEY ||
5104 key->type == BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 5105 ret = changed_ref(sctx, result);
5106 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5107 ret = changed_xattr(sctx, result);
5108 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5109 ret = changed_extent(sctx, result);
5110
5111out:
5112 return ret;
5113}
5114
5115static int full_send_tree(struct send_ctx *sctx)
5116{
5117 int ret;
31db9f7c
AB		 5118 struct btrfs_root *send_root = sctx->send_root;
5119 struct btrfs_key key;
5120 struct btrfs_key found_key;
5121 struct btrfs_path *path;
5122 struct extent_buffer *eb;
5123 int slot;
5124 u64 start_ctransid;
5125 u64 ctransid;
5126
5127 path = alloc_path_for_send();
5128 if (!path)
5129 return -ENOMEM;
5130
5f3ab90a 5131 spin_lock(&send_root->root_item_lock);
31db9f7c 5132 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5f3ab90a 5133 spin_unlock(&send_root->root_item_lock);
31db9f7c
AB		 5134
5135 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5136 key.type = BTRFS_INODE_ITEM_KEY;
5137 key.offset = 0;
5138
31db9f7c 5139 /*
766702ef
AB		 5140 * Make sure the tree has not changed after re-joining. We detect this
5141 * by comparing start_ctransid and ctransid. They should always match.
31db9f7c 5142 */
5f3ab90a 5143 spin_lock(&send_root->root_item_lock);
31db9f7c 5144 ctransid = btrfs_root_ctransid(&send_root->root_item);
5f3ab90a 5145 spin_unlock(&send_root->root_item_lock);
31db9f7c
AB		 5146
5147 if (ctransid != start_ctransid) {
efe120a0 5148 WARN(1, KERN_WARNING "BTRFS: the root that you're trying to "
31db9f7c
AB		 5149 "send was modified in between. This is "
5150 "probably a bug.\n");
5151 ret = -EIO;
5152 goto out;
5153 }
5154
5155 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5156 if (ret < 0)
5157 goto out;
5158 if (ret)
5159 goto out_finish;
5160
5161 while (1) {
31db9f7c
AB		 5162 eb = path->nodes[0];
5163 slot = path->slots[0];
5164 btrfs_item_key_to_cpu(eb, &found_key, slot);
5165
5166 ret = changed_cb(send_root, NULL, path, NULL,
5167 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5168 if (ret < 0)
5169 goto out;
5170
5171 key.objectid = found_key.objectid;
5172 key.type = found_key.type;
5173 key.offset = found_key.offset + 1;
5174
5175 ret = btrfs_next_item(send_root, path);
5176 if (ret < 0)
5177 goto out;
5178 if (ret) {
5179 ret = 0;
5180 break;
5181 }
5182 }
5183
5184out_finish:
5185 ret = finish_inode_if_needed(sctx, 1);
5186
5187out:
5188 btrfs_free_path(path);
31db9f7c
AB		 5189 return ret;
5190}
5191
5192static int send_subvol(struct send_ctx *sctx)
5193{
5194 int ret;
5195
c2c71324
SB		 5196 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5197 ret = send_header(sctx);
5198 if (ret < 0)
5199 goto out;
5200 }
31db9f7c
AB		 5201
5202 ret = send_subvol_begin(sctx);
5203 if (ret < 0)
5204 goto out;
5205
5206 if (sctx->parent_root) {
5207 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5208 changed_cb, sctx);
5209 if (ret < 0)
5210 goto out;
5211 ret = finish_inode_if_needed(sctx, 1);
5212 if (ret < 0)
5213 goto out;
5214 } else {
5215 ret = full_send_tree(sctx);
5216 if (ret < 0)
5217 goto out;
5218 }
5219
5220out:
31db9f7c
AB		 5221 free_recorded_refs(sctx);
5222 return ret;
5223}
5224
66ef7d65
DS		 5225static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5226{
5227 spin_lock(&root->root_item_lock);
5228 root->send_in_progress--;
5229 /*
5230 * Not much left to do, we don't know why it's unbalanced and
5231 * can't blindly reset it to 0.
5232 */
5233 if (root->send_in_progress < 0)
5234 btrfs_err(root->fs_info,
5235 "send_in_progres unbalanced %d root %llu\n",
5236 root->send_in_progress, root->root_key.objectid);
5237 spin_unlock(&root->root_item_lock);
5238}
5239
31db9f7c
AB		 5240long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5241{
5242 int ret = 0;
5243 struct btrfs_root *send_root;
5244 struct btrfs_root *clone_root;
5245 struct btrfs_fs_info *fs_info;
5246 struct btrfs_ioctl_send_args *arg = NULL;
5247 struct btrfs_key key;
31db9f7c
AB		 5248 struct send_ctx *sctx = NULL;
5249 u32 i;
5250 u64 *clone_sources_tmp = NULL;
2c686537 5251 int clone_sources_to_rollback = 0;
896c14f9 5252 int sort_clone_roots = 0;
18f687d5 5253 int index;
31db9f7c
AB		 5254
5255 if (!capable(CAP_SYS_ADMIN))
5256 return -EPERM;
5257
496ad9aa 5258 send_root = BTRFS_I(file_inode(mnt_file))->root;
31db9f7c
AB		 5259 fs_info = send_root->fs_info;
5260
2c686537
DS		 5261 /*
5262 * The subvolume must remain read-only during send, protect against
5263 * making it RW.
5264 */
5265 spin_lock(&send_root->root_item_lock);
5266 send_root->send_in_progress++;
5267 spin_unlock(&send_root->root_item_lock);
5268
139f807a
JB		 5269 /*
5270 * This is done when we lookup the root, it should already be complete
5271 * by the time we get here.
5272 */
5273 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5274
2c686537
DS		 5275 /*
5276 * Userspace tools do the checks and warn the user if it's
5277 * not RO.
5278 */
5279 if (!btrfs_root_readonly(send_root)) {
5280 ret = -EPERM;
5281 goto out;
5282 }
5283
31db9f7c
AB		 5284 arg = memdup_user(arg_, sizeof(*arg));
5285 if (IS_ERR(arg)) {
5286 ret = PTR_ERR(arg);
5287 arg = NULL;
5288 goto out;
5289 }
5290
5291 if (!access_ok(VERIFY_READ, arg->clone_sources,
700ff4f0
DC		 5292 sizeof(*arg->clone_sources) *
5293 arg->clone_sources_count)) {
31db9f7c
AB		 5294 ret = -EFAULT;
5295 goto out;
5296 }
5297
c2c71324 5298 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
cb95e7bf
MF		 5299 ret = -EINVAL;
5300 goto out;
5301 }
5302
31db9f7c
AB		 5303 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5304 if (!sctx) {
5305 ret = -ENOMEM;
5306 goto out;
5307 }
5308
5309 INIT_LIST_HEAD(&sctx->new_refs);
5310 INIT_LIST_HEAD(&sctx->deleted_refs);
5311 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5312 INIT_LIST_HEAD(&sctx->name_cache_list);
5313
cb95e7bf
MF		 5314 sctx->flags = arg->flags;
5315
31db9f7c 5316 sctx->send_filp = fget(arg->send_fd);
ecc7ada7
TI		 5317 if (!sctx->send_filp) {
5318 ret = -EBADF;
31db9f7c
AB		 5319 goto out;
5320 }
5321
31db9f7c
AB		 5322 sctx->send_root = send_root;
5323 sctx->clone_roots_cnt = arg->clone_sources_count;
5324
5325 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5326 sctx->send_buf = vmalloc(sctx->send_max_size);
5327 if (!sctx->send_buf) {
5328 ret = -ENOMEM;
5329 goto out;
5330 }
5331
5332 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5333 if (!sctx->read_buf) {
5334 ret = -ENOMEM;
5335 goto out;
5336 }
5337
9f03740a
FDBM		 5338 sctx->pending_dir_moves = RB_ROOT;
5339 sctx->waiting_dir_moves = RB_ROOT;
5340
31db9f7c
AB		 5341 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5342 (arg->clone_sources_count + 1));
5343 if (!sctx->clone_roots) {
5344 ret = -ENOMEM;
5345 goto out;
5346 }
5347
5348 if (arg->clone_sources_count) {
5349 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5350 sizeof(*arg->clone_sources));
5351 if (!clone_sources_tmp) {
5352 ret = -ENOMEM;
5353 goto out;
5354 }
5355
5356 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5357 arg->clone_sources_count *
5358 sizeof(*arg->clone_sources));
5359 if (ret) {
5360 ret = -EFAULT;
5361 goto out;
5362 }
5363
5364 for (i = 0; i < arg->clone_sources_count; i++) {
5365 key.objectid = clone_sources_tmp[i];
5366 key.type = BTRFS_ROOT_ITEM_KEY;
5367 key.offset = (u64)-1;
18f687d5
WS		 5368
5369 index = srcu_read_lock(&fs_info->subvol_srcu);
5370
31db9f7c 5371 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
31db9f7c 5372 if (IS_ERR(clone_root)) {
18f687d5 5373 srcu_read_unlock(&fs_info->subvol_srcu, index);
31db9f7c
AB		 5374 ret = PTR_ERR(clone_root);
5375 goto out;
5376 }
2c686537
DS		 5377 clone_sources_to_rollback = i + 1;
5378 spin_lock(&clone_root->root_item_lock);
5379 clone_root->send_in_progress++;
5380 if (!btrfs_root_readonly(clone_root)) {
5381 spin_unlock(&clone_root->root_item_lock);
18f687d5 5382 srcu_read_unlock(&fs_info->subvol_srcu, index);
2c686537
DS		 5383 ret = -EPERM;
5384 goto out;
5385 }
5386 spin_unlock(&clone_root->root_item_lock);
18f687d5
WS		 5387 srcu_read_unlock(&fs_info->subvol_srcu, index);
5388
31db9f7c
AB		 5389 sctx->clone_roots[i].root = clone_root;
5390 }
5391 vfree(clone_sources_tmp);
5392 clone_sources_tmp = NULL;
5393 }
5394
5395 if (arg->parent_root) {
5396 key.objectid = arg->parent_root;
5397 key.type = BTRFS_ROOT_ITEM_KEY;
5398 key.offset = (u64)-1;
18f687d5
WS		 5399
5400 index = srcu_read_lock(&fs_info->subvol_srcu);
5401
31db9f7c 5402 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
b1b19596 5403 if (IS_ERR(sctx->parent_root)) {
18f687d5 5404 srcu_read_unlock(&fs_info->subvol_srcu, index);
b1b19596 5405 ret = PTR_ERR(sctx->parent_root);
31db9f7c
AB		 5406 goto out;
5407 }
18f687d5 5408
2c686537
DS		 5409 spin_lock(&sctx->parent_root->root_item_lock);
5410 sctx->parent_root->send_in_progress++;
5411 if (!btrfs_root_readonly(sctx->parent_root)) {
5412 spin_unlock(&sctx->parent_root->root_item_lock);
18f687d5 5413 srcu_read_unlock(&fs_info->subvol_srcu, index);
2c686537
DS		 5414 ret = -EPERM;
5415 goto out;
5416 }
5417 spin_unlock(&sctx->parent_root->root_item_lock);
18f687d5
WS		 5418
5419 srcu_read_unlock(&fs_info->subvol_srcu, index);
31db9f7c
AB		 5420 }
5421
5422 /*
5423 * Clones from send_root are allowed, but only if the clone source
5424 * is behind the current send position. This is checked while searching
5425 * for possible clone sources.
5426 */
5427 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5428
5429 /* We do a bsearch later */
5430 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5431 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5432 NULL);
896c14f9 5433 sort_clone_roots = 1;
31db9f7c
AB		 5434
5435 ret = send_subvol(sctx);
5436 if (ret < 0)
5437 goto out;
5438
c2c71324
SB		 5439 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5440 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5441 if (ret < 0)
5442 goto out;
5443 ret = send_cmd(sctx);
5444 if (ret < 0)
5445 goto out;
5446 }
31db9f7c
AB		 5447
5448out:
9f03740a
FDBM		 5449 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5450 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5451 struct rb_node *n;
5452 struct pending_dir_move *pm;
5453
5454 n = rb_first(&sctx->pending_dir_moves);
5455 pm = rb_entry(n, struct pending_dir_move, node);
5456 while (!list_empty(&pm->list)) {
5457 struct pending_dir_move *pm2;
5458
5459 pm2 = list_first_entry(&pm->list,
5460 struct pending_dir_move, list);
5461 free_pending_move(sctx, pm2);
5462 }
5463 free_pending_move(sctx, pm);
5464 }
5465
5466 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5467 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5468 struct rb_node *n;
5469 struct waiting_dir_move *dm;
5470
5471 n = rb_first(&sctx->waiting_dir_moves);
5472 dm = rb_entry(n, struct waiting_dir_move, node);
5473 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5474 kfree(dm);
5475 }
5476
896c14f9
WS		 5477 if (sort_clone_roots) {
5478 for (i = 0; i < sctx->clone_roots_cnt; i++)
5479 btrfs_root_dec_send_in_progress(
5480 sctx->clone_roots[i].root);
5481 } else {
5482 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5483 btrfs_root_dec_send_in_progress(
5484 sctx->clone_roots[i].root);
5485
5486 btrfs_root_dec_send_in_progress(send_root);
5487 }
66ef7d65
DS		 5488 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5489 btrfs_root_dec_send_in_progress(sctx->parent_root);
2c686537 5490
31db9f7c
AB		 5491 kfree(arg);
5492 vfree(clone_sources_tmp);
5493
5494 if (sctx) {
5495 if (sctx->send_filp)
5496 fput(sctx->send_filp);
5497
5498 vfree(sctx->clone_roots);
5499 vfree(sctx->send_buf);
5500 vfree(sctx->read_buf);
5501
5502 name_cache_free(sctx);
5503
5504 kfree(sctx);
5505 }
5506
5507 return ret;
5508}
Linux kernel - Deliverables
This page took 0.356396 seconds and 5 git commands to generate.