projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix extent_map block_len after merging
[linux-2.6-block.git] / fs / btrfs / send.c
CommitLineData
31db9f7c
AB		 1/*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/bsearch.h>
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/sort.h>
23#include <linux/mount.h>
24#include <linux/xattr.h>
25#include <linux/posix_acl_xattr.h>
26#include <linux/radix-tree.h>
27#include <linux/crc32c.h>
a1857ebe 28#include <linux/vmalloc.h>
ed84885d 29#include <linux/string.h>
31db9f7c
AB		 30
31#include "send.h"
32#include "backref.h"
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;
54 char *prepared;
55
56 char *buf;
57 int buf_len;
35a3621b
SB		 58 unsigned int reversed:1;
59 unsigned int virtual_mem:1;
31db9f7c
AB		 60 char inline_buf[];
61 };
62 char pad[PAGE_SIZE];
63 };
64};
65#define FS_PATH_INLINE_SIZE \
66 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
67
68
69/* reused for each extent */
70struct clone_root {
71 struct btrfs_root *root;
72 u64 ino;
73 u64 offset;
74
75 u64 found_refs;
76};
77
78#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
79#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
80
81struct send_ctx {
82 struct file *send_filp;
83 loff_t send_off;
84 char *send_buf;
85 u32 send_size;
86 u32 send_max_size;
87 u64 total_send_size;
88 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
cb95e7bf 89 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
31db9f7c
AB		 90
91 struct vfsmount *mnt;
92
93 struct btrfs_root *send_root;
94 struct btrfs_root *parent_root;
95 struct clone_root *clone_roots;
96 int clone_roots_cnt;
97
98 /* current state of the compare_tree call */
99 struct btrfs_path *left_path;
100 struct btrfs_path *right_path;
101 struct btrfs_key *cmp_key;
102
103 /*
104 * infos of the currently processed inode. In case of deleted inodes,
105 * these are the values from the deleted inode.
106 */
107 u64 cur_ino;
108 u64 cur_inode_gen;
109 int cur_inode_new;
110 int cur_inode_new_gen;
111 int cur_inode_deleted;
31db9f7c
AB		 112 u64 cur_inode_size;
113 u64 cur_inode_mode;
16e7549f 114 u64 cur_inode_last_extent;
31db9f7c
AB		 115
116 u64 send_progress;
117
118 struct list_head new_refs;
119 struct list_head deleted_refs;
120
121 struct radix_tree_root name_cache;
122 struct list_head name_cache_list;
123 int name_cache_size;
124
31db9f7c
AB		 125 char *read_buf;
126};
127
128struct name_cache_entry {
129 struct list_head list;
7e0926fe
AB		 130 /*
131 * radix_tree has only 32bit entries but we need to handle 64bit inums.
132 * We use the lower 32bit of the 64bit inum to store it in the tree. If
133 * more then one inum would fall into the same entry, we use radix_list
134 * to store the additional entries. radix_list is also used to store
135 * entries where two entries have the same inum but different
136 * generations.
137 */
138 struct list_head radix_list;
31db9f7c
AB		 139 u64 ino;
140 u64 gen;
141 u64 parent_ino;
142 u64 parent_gen;
143 int ret;
144 int need_later_update;
145 int name_len;
146 char name[];
147};
148
16e7549f
JB		 149static int need_send_hole(struct send_ctx *sctx)
150{
151 return (sctx->parent_root && !sctx->cur_inode_new &&
152 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
153 S_ISREG(sctx->cur_inode_mode));
154}
155
31db9f7c
AB		 156static void fs_path_reset(struct fs_path *p)
157{
158 if (p->reversed) {
159 p->start = p->buf + p->buf_len - 1;
160 p->end = p->start;
161 *p->start = 0;
162 } else {
163 p->start = p->buf;
164 p->end = p->start;
165 *p->start = 0;
166 }
167}
168
924794c9 169static struct fs_path *fs_path_alloc(void)
31db9f7c
AB		 170{
171 struct fs_path *p;
172
173 p = kmalloc(sizeof(*p), GFP_NOFS);
174 if (!p)
175 return NULL;
176 p->reversed = 0;
177 p->virtual_mem = 0;
178 p->buf = p->inline_buf;
179 p->buf_len = FS_PATH_INLINE_SIZE;
180 fs_path_reset(p);
181 return p;
182}
183
924794c9 184static struct fs_path *fs_path_alloc_reversed(void)
31db9f7c
AB		 185{
186 struct fs_path *p;
187
924794c9 188 p = fs_path_alloc();
31db9f7c
AB		 189 if (!p)
190 return NULL;
191 p->reversed = 1;
192 fs_path_reset(p);
193 return p;
194}
195
924794c9 196static void fs_path_free(struct fs_path *p)
31db9f7c
AB		 197{
198 if (!p)
199 return;
200 if (p->buf != p->inline_buf) {
201 if (p->virtual_mem)
202 vfree(p->buf);
203 else
204 kfree(p->buf);
205 }
206 kfree(p);
207}
208
209static int fs_path_len(struct fs_path *p)
210{
211 return p->end - p->start;
212}
213
214static int fs_path_ensure_buf(struct fs_path *p, int len)
215{
216 char *tmp_buf;
217 int path_len;
218 int old_buf_len;
219
220 len++;
221
222 if (p->buf_len >= len)
223 return 0;
224
225 path_len = p->end - p->start;
226 old_buf_len = p->buf_len;
227 len = PAGE_ALIGN(len);
228
229 if (p->buf == p->inline_buf) {
8be04b93 230 tmp_buf = kmalloc(len, GFP_NOFS | __GFP_NOWARN);
31db9f7c
AB		 231 if (!tmp_buf) {
232 tmp_buf = vmalloc(len);
233 if (!tmp_buf)
234 return -ENOMEM;
235 p->virtual_mem = 1;
236 }
237 memcpy(tmp_buf, p->buf, p->buf_len);
238 p->buf = tmp_buf;
239 p->buf_len = len;
240 } else {
241 if (p->virtual_mem) {
242 tmp_buf = vmalloc(len);
243 if (!tmp_buf)
244 return -ENOMEM;
245 memcpy(tmp_buf, p->buf, p->buf_len);
246 vfree(p->buf);
247 } else {
248 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
249 if (!tmp_buf) {
250 tmp_buf = vmalloc(len);
251 if (!tmp_buf)
252 return -ENOMEM;
253 memcpy(tmp_buf, p->buf, p->buf_len);
254 kfree(p->buf);
255 p->virtual_mem = 1;
256 }
257 }
258 p->buf = tmp_buf;
259 p->buf_len = len;
260 }
261 if (p->reversed) {
262 tmp_buf = p->buf + old_buf_len - path_len - 1;
263 p->end = p->buf + p->buf_len - 1;
264 p->start = p->end - path_len;
265 memmove(p->start, tmp_buf, path_len + 1);
266 } else {
267 p->start = p->buf;
268 p->end = p->start + path_len;
269 }
270 return 0;
271}
272
273static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
274{
275 int ret;
276 int new_len;
277
278 new_len = p->end - p->start + name_len;
279 if (p->start != p->end)
280 new_len++;
281 ret = fs_path_ensure_buf(p, new_len);
282 if (ret < 0)
283 goto out;
284
285 if (p->reversed) {
286 if (p->start != p->end)
287 *--p->start = '/';
288 p->start -= name_len;
289 p->prepared = p->start;
290 } else {
291 if (p->start != p->end)
292 *p->end++ = '/';
293 p->prepared = p->end;
294 p->end += name_len;
295 *p->end = 0;
296 }
297
298out:
299 return ret;
300}
301
302static int fs_path_add(struct fs_path *p, const char *name, int name_len)
303{
304 int ret;
305
306 ret = fs_path_prepare_for_add(p, name_len);
307 if (ret < 0)
308 goto out;
309 memcpy(p->prepared, name, name_len);
310 p->prepared = NULL;
311
312out:
313 return ret;
314}
315
316static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
317{
318 int ret;
319
320 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
321 if (ret < 0)
322 goto out;
323 memcpy(p->prepared, p2->start, p2->end - p2->start);
324 p->prepared = NULL;
325
326out:
327 return ret;
328}
329
330static int fs_path_add_from_extent_buffer(struct fs_path *p,
331 struct extent_buffer *eb,
332 unsigned long off, int len)
333{
334 int ret;
335
336 ret = fs_path_prepare_for_add(p, len);
337 if (ret < 0)
338 goto out;
339
340 read_extent_buffer(eb, p->prepared, off, len);
341 p->prepared = NULL;
342
343out:
344 return ret;
345}
346
9ea3ef51 347#if 0
31db9f7c
AB		 348static void fs_path_remove(struct fs_path *p)
349{
350 BUG_ON(p->reversed);
351 while (p->start != p->end && *p->end != '/')
352 p->end--;
353 *p->end = 0;
354}
9ea3ef51 355#endif
31db9f7c
AB		 356
357static int fs_path_copy(struct fs_path *p, struct fs_path *from)
358{
359 int ret;
360
361 p->reversed = from->reversed;
362 fs_path_reset(p);
363
364 ret = fs_path_add_path(p, from);
365
366 return ret;
367}
368
369
370static void fs_path_unreverse(struct fs_path *p)
371{
372 char *tmp;
373 int len;
374
375 if (!p->reversed)
376 return;
377
378 tmp = p->start;
379 len = p->end - p->start;
380 p->start = p->buf;
381 p->end = p->start + len;
382 memmove(p->start, tmp, len + 1);
383 p->reversed = 0;
384}
385
386static struct btrfs_path *alloc_path_for_send(void)
387{
388 struct btrfs_path *path;
389
390 path = btrfs_alloc_path();
391 if (!path)
392 return NULL;
393 path->search_commit_root = 1;
394 path->skip_locking = 1;
395 return path;
396}
397
48a3b636 398static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
31db9f7c
AB		 399{
400 int ret;
401 mm_segment_t old_fs;
402 u32 pos = 0;
403
404 old_fs = get_fs();
405 set_fs(KERNEL_DS);
406
407 while (pos < len) {
1bcea355 408 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
31db9f7c
AB		 409 /* TODO handle that correctly */
410 /*if (ret == -ERESTARTSYS) {
411 continue;
412 }*/
413 if (ret < 0)
414 goto out;
415 if (ret == 0) {
416 ret = -EIO;
417 goto out;
418 }
419 pos += ret;
420 }
421
422 ret = 0;
423
424out:
425 set_fs(old_fs);
426 return ret;
427}
428
429static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
430{
431 struct btrfs_tlv_header *hdr;
432 int total_len = sizeof(*hdr) + len;
433 int left = sctx->send_max_size - sctx->send_size;
434
435 if (unlikely(left < total_len))
436 return -EOVERFLOW;
437
438 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
439 hdr->tlv_type = cpu_to_le16(attr);
440 hdr->tlv_len = cpu_to_le16(len);
441 memcpy(hdr + 1, data, len);
442 sctx->send_size += total_len;
443
444 return 0;
445}
446
447#if 0
448static int tlv_put_u8(struct send_ctx *sctx, u16 attr, u8 value)
449{
450 return tlv_put(sctx, attr, &value, sizeof(value));
451}
452
453static int tlv_put_u16(struct send_ctx *sctx, u16 attr, u16 value)
454{
455 __le16 tmp = cpu_to_le16(value);
456 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
457}
458
459static int tlv_put_u32(struct send_ctx *sctx, u16 attr, u32 value)
460{
461 __le32 tmp = cpu_to_le32(value);
462 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
463}
464#endif
465
466static int tlv_put_u64(struct send_ctx *sctx, u16 attr, u64 value)
467{
468 __le64 tmp = cpu_to_le64(value);
469 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
470}
471
472static int tlv_put_string(struct send_ctx *sctx, u16 attr,
473 const char *str, int len)
474{
475 if (len == -1)
476 len = strlen(str);
477 return tlv_put(sctx, attr, str, len);
478}
479
480static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
481 const u8 *uuid)
482{
483 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
484}
485
486#if 0
487static int tlv_put_timespec(struct send_ctx *sctx, u16 attr,
488 struct timespec *ts)
489{
490 struct btrfs_timespec bts;
491 bts.sec = cpu_to_le64(ts->tv_sec);
492 bts.nsec = cpu_to_le32(ts->tv_nsec);
493 return tlv_put(sctx, attr, &bts, sizeof(bts));
494}
495#endif
496
497static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
498 struct extent_buffer *eb,
499 struct btrfs_timespec *ts)
500{
501 struct btrfs_timespec bts;
502 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
503 return tlv_put(sctx, attr, &bts, sizeof(bts));
504}
505
506
507#define TLV_PUT(sctx, attrtype, attrlen, data) \
508 do { \
509 ret = tlv_put(sctx, attrtype, attrlen, data); \
510 if (ret < 0) \
511 goto tlv_put_failure; \
512 } while (0)
513
514#define TLV_PUT_INT(sctx, attrtype, bits, value) \
515 do { \
516 ret = tlv_put_u##bits(sctx, attrtype, value); \
517 if (ret < 0) \
518 goto tlv_put_failure; \
519 } while (0)
520
521#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
522#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
523#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
524#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
525#define TLV_PUT_STRING(sctx, attrtype, str, len) \
526 do { \
527 ret = tlv_put_string(sctx, attrtype, str, len); \
528 if (ret < 0) \
529 goto tlv_put_failure; \
530 } while (0)
531#define TLV_PUT_PATH(sctx, attrtype, p) \
532 do { \
533 ret = tlv_put_string(sctx, attrtype, p->start, \
534 p->end - p->start); \
535 if (ret < 0) \
536 goto tlv_put_failure; \
537 } while(0)
538#define TLV_PUT_UUID(sctx, attrtype, uuid) \
539 do { \
540 ret = tlv_put_uuid(sctx, attrtype, uuid); \
541 if (ret < 0) \
542 goto tlv_put_failure; \
543 } while (0)
544#define TLV_PUT_TIMESPEC(sctx, attrtype, ts) \
545 do { \
546 ret = tlv_put_timespec(sctx, attrtype, ts); \
547 if (ret < 0) \
548 goto tlv_put_failure; \
549 } while (0)
550#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
551 do { \
552 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
553 if (ret < 0) \
554 goto tlv_put_failure; \
555 } while (0)
556
557static int send_header(struct send_ctx *sctx)
558{
559 struct btrfs_stream_header hdr;
560
561 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
562 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
563
1bcea355
AJ		 564 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
565 &sctx->send_off);
31db9f7c
AB		 566}
567
568/*
569 * For each command/item we want to send to userspace, we call this function.
570 */
571static int begin_cmd(struct send_ctx *sctx, int cmd)
572{
573 struct btrfs_cmd_header *hdr;
574
fae7f21c 575 if (WARN_ON(!sctx->send_buf))
31db9f7c 576 return -EINVAL;
31db9f7c
AB		 577
578 BUG_ON(sctx->send_size);
579
580 sctx->send_size += sizeof(*hdr);
581 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
582 hdr->cmd = cpu_to_le16(cmd);
583
584 return 0;
585}
586
587static int send_cmd(struct send_ctx *sctx)
588{
589 int ret;
590 struct btrfs_cmd_header *hdr;
591 u32 crc;
592
593 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
594 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
595 hdr->crc = 0;
596
597 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
598 hdr->crc = cpu_to_le32(crc);
599
1bcea355
AJ		 600 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
601 &sctx->send_off);
31db9f7c
AB		 602
603 sctx->total_send_size += sctx->send_size;
604 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
605 sctx->send_size = 0;
606
607 return ret;
608}
609
610/*
611 * Sends a move instruction to user space
612 */
613static int send_rename(struct send_ctx *sctx,
614 struct fs_path *from, struct fs_path *to)
615{
616 int ret;
617
618verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
619
620 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
621 if (ret < 0)
622 goto out;
623
624 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
625 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
626
627 ret = send_cmd(sctx);
628
629tlv_put_failure:
630out:
631 return ret;
632}
633
634/*
635 * Sends a link instruction to user space
636 */
637static int send_link(struct send_ctx *sctx,
638 struct fs_path *path, struct fs_path *lnk)
639{
640 int ret;
641
642verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
643
644 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
645 if (ret < 0)
646 goto out;
647
648 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
649 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
650
651 ret = send_cmd(sctx);
652
653tlv_put_failure:
654out:
655 return ret;
656}
657
658/*
659 * Sends an unlink instruction to user space
660 */
661static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
662{
663 int ret;
664
665verbose_printk("btrfs: send_unlink %s\n", path->start);
666
667 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
668 if (ret < 0)
669 goto out;
670
671 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
672
673 ret = send_cmd(sctx);
674
675tlv_put_failure:
676out:
677 return ret;
678}
679
680/*
681 * Sends a rmdir instruction to user space
682 */
683static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
684{
685 int ret;
686
687verbose_printk("btrfs: send_rmdir %s\n", path->start);
688
689 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
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 * Helper function to retrieve some fields from an inode item.
704 */
705static int get_inode_info(struct btrfs_root *root,
706 u64 ino, u64 *size, u64 *gen,
85a7b33b
AB		 707 u64 *mode, u64 *uid, u64 *gid,
708 u64 *rdev)
31db9f7c
AB		 709{
710 int ret;
711 struct btrfs_inode_item *ii;
712 struct btrfs_key key;
713 struct btrfs_path *path;
714
715 path = alloc_path_for_send();
716 if (!path)
717 return -ENOMEM;
718
719 key.objectid = ino;
720 key.type = BTRFS_INODE_ITEM_KEY;
721 key.offset = 0;
722 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
723 if (ret < 0)
724 goto out;
725 if (ret) {
726 ret = -ENOENT;
727 goto out;
728 }
729
730 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
731 struct btrfs_inode_item);
732 if (size)
733 *size = btrfs_inode_size(path->nodes[0], ii);
734 if (gen)
735 *gen = btrfs_inode_generation(path->nodes[0], ii);
736 if (mode)
737 *mode = btrfs_inode_mode(path->nodes[0], ii);
738 if (uid)
739 *uid = btrfs_inode_uid(path->nodes[0], ii);
740 if (gid)
741 *gid = btrfs_inode_gid(path->nodes[0], ii);
85a7b33b
AB		 742 if (rdev)
743 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
31db9f7c
AB		 744
745out:
746 btrfs_free_path(path);
747 return ret;
748}
749
750typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
751 struct fs_path *p,
752 void *ctx);
753
754/*
96b5bd77
JS		 755 * Helper function to iterate the entries in ONE btrfs_inode_ref or
756 * btrfs_inode_extref.
31db9f7c
AB		 757 * The iterate callback may return a non zero value to stop iteration. This can
758 * be a negative value for error codes or 1 to simply stop it.
759 *
96b5bd77 760 * path must point to the INODE_REF or INODE_EXTREF when called.
31db9f7c 761 */
924794c9 762static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 763 struct btrfs_key *found_key, int resolve,
764 iterate_inode_ref_t iterate, void *ctx)
765{
96b5bd77 766 struct extent_buffer *eb = path->nodes[0];
31db9f7c
AB		 767 struct btrfs_item *item;
768 struct btrfs_inode_ref *iref;
96b5bd77 769 struct btrfs_inode_extref *extref;
31db9f7c
AB		 770 struct btrfs_path *tmp_path;
771 struct fs_path *p;
96b5bd77 772 u32 cur = 0;
31db9f7c 773 u32 total;
96b5bd77 774 int slot = path->slots[0];
31db9f7c
AB		 775 u32 name_len;
776 char *start;
777 int ret = 0;
96b5bd77 778 int num = 0;
31db9f7c 779 int index;
96b5bd77
JS		 780 u64 dir;
781 unsigned long name_off;
782 unsigned long elem_size;
783 unsigned long ptr;
31db9f7c 784
924794c9 785 p = fs_path_alloc_reversed();
31db9f7c
AB		 786 if (!p)
787 return -ENOMEM;
788
789 tmp_path = alloc_path_for_send();
790 if (!tmp_path) {
924794c9 791 fs_path_free(p);
31db9f7c
AB		 792 return -ENOMEM;
793 }
794
31db9f7c 795
96b5bd77
JS		 796 if (found_key->type == BTRFS_INODE_REF_KEY) {
797 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
798 struct btrfs_inode_ref);
dd3cc16b 799 item = btrfs_item_nr(slot);
96b5bd77
JS		 800 total = btrfs_item_size(eb, item);
801 elem_size = sizeof(*iref);
802 } else {
803 ptr = btrfs_item_ptr_offset(eb, slot);
804 total = btrfs_item_size_nr(eb, slot);
805 elem_size = sizeof(*extref);
806 }
807
31db9f7c
AB		 808 while (cur < total) {
809 fs_path_reset(p);
810
96b5bd77
JS		 811 if (found_key->type == BTRFS_INODE_REF_KEY) {
812 iref = (struct btrfs_inode_ref *)(ptr + cur);
813 name_len = btrfs_inode_ref_name_len(eb, iref);
814 name_off = (unsigned long)(iref + 1);
815 index = btrfs_inode_ref_index(eb, iref);
816 dir = found_key->offset;
817 } else {
818 extref = (struct btrfs_inode_extref *)(ptr + cur);
819 name_len = btrfs_inode_extref_name_len(eb, extref);
820 name_off = (unsigned long)&extref->name;
821 index = btrfs_inode_extref_index(eb, extref);
822 dir = btrfs_inode_extref_parent(eb, extref);
823 }
824
31db9f7c 825 if (resolve) {
96b5bd77
JS		 826 start = btrfs_ref_to_path(root, tmp_path, name_len,
827 name_off, eb, dir,
828 p->buf, p->buf_len);
31db9f7c
AB		 829 if (IS_ERR(start)) {
830 ret = PTR_ERR(start);
831 goto out;
832 }
833 if (start < p->buf) {
834 /* overflow , try again with larger buffer */
835 ret = fs_path_ensure_buf(p,
836 p->buf_len + p->buf - start);
837 if (ret < 0)
838 goto out;
96b5bd77
JS		 839 start = btrfs_ref_to_path(root, tmp_path,
840 name_len, name_off,
841 eb, dir,
842 p->buf, p->buf_len);
31db9f7c
AB		 843 if (IS_ERR(start)) {
844 ret = PTR_ERR(start);
845 goto out;
846 }
847 BUG_ON(start < p->buf);
848 }
849 p->start = start;
850 } else {
96b5bd77
JS		 851 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
852 name_len);
31db9f7c
AB		 853 if (ret < 0)
854 goto out;
855 }
856
96b5bd77
JS		 857 cur += elem_size + name_len;
858 ret = iterate(num, dir, index, p, ctx);
31db9f7c
AB		 859 if (ret)
860 goto out;
31db9f7c
AB		 861 num++;
862 }
863
864out:
865 btrfs_free_path(tmp_path);
924794c9 866 fs_path_free(p);
31db9f7c
AB		 867 return ret;
868}
869
870typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
871 const char *name, int name_len,
872 const char *data, int data_len,
873 u8 type, void *ctx);
874
875/*
876 * Helper function to iterate the entries in ONE btrfs_dir_item.
877 * The iterate callback may return a non zero value to stop iteration. This can
878 * be a negative value for error codes or 1 to simply stop it.
879 *
880 * path must point to the dir item when called.
881 */
924794c9 882static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 883 struct btrfs_key *found_key,
884 iterate_dir_item_t iterate, void *ctx)
885{
886 int ret = 0;
887 struct extent_buffer *eb;
888 struct btrfs_item *item;
889 struct btrfs_dir_item *di;
31db9f7c
AB		 890 struct btrfs_key di_key;
891 char *buf = NULL;
892 char *buf2 = NULL;
893 int buf_len;
894 int buf_virtual = 0;
895 u32 name_len;
896 u32 data_len;
897 u32 cur;
898 u32 len;
899 u32 total;
900 int slot;
901 int num;
902 u8 type;
903
904 buf_len = PAGE_SIZE;
905 buf = kmalloc(buf_len, GFP_NOFS);
906 if (!buf) {
907 ret = -ENOMEM;
908 goto out;
909 }
910
31db9f7c
AB		 911 eb = path->nodes[0];
912 slot = path->slots[0];
dd3cc16b 913 item = btrfs_item_nr(slot);
31db9f7c
AB		 914 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
915 cur = 0;
916 len = 0;
917 total = btrfs_item_size(eb, item);
918
919 num = 0;
920 while (cur < total) {
921 name_len = btrfs_dir_name_len(eb, di);
922 data_len = btrfs_dir_data_len(eb, di);
923 type = btrfs_dir_type(eb, di);
924 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
925
926 if (name_len + data_len > buf_len) {
927 buf_len = PAGE_ALIGN(name_len + data_len);
928 if (buf_virtual) {
929 buf2 = vmalloc(buf_len);
930 if (!buf2) {
931 ret = -ENOMEM;
932 goto out;
933 }
934 vfree(buf);
935 } else {
936 buf2 = krealloc(buf, buf_len, GFP_NOFS);
937 if (!buf2) {
938 buf2 = vmalloc(buf_len);
939 if (!buf2) {
940 ret = -ENOMEM;
941 goto out;
942 }
943 kfree(buf);
944 buf_virtual = 1;
945 }
946 }
947
948 buf = buf2;
949 buf2 = NULL;
950 }
951
952 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
953 name_len + data_len);
954
955 len = sizeof(*di) + name_len + data_len;
956 di = (struct btrfs_dir_item *)((char *)di + len);
957 cur += len;
958
959 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
960 data_len, type, ctx);
961 if (ret < 0)
962 goto out;
963 if (ret) {
964 ret = 0;
965 goto out;
966 }
967
968 num++;
969 }
970
971out:
31db9f7c
AB		 972 if (buf_virtual)
973 vfree(buf);
974 else
975 kfree(buf);
976 return ret;
977}
978
979static int __copy_first_ref(int num, u64 dir, int index,
980 struct fs_path *p, void *ctx)
981{
982 int ret;
983 struct fs_path *pt = ctx;
984
985 ret = fs_path_copy(pt, p);
986 if (ret < 0)
987 return ret;
988
989 /* we want the first only */
990 return 1;
991}
992
993/*
994 * Retrieve the first path of an inode. If an inode has more then one
995 * ref/hardlink, this is ignored.
996 */
924794c9 997static int get_inode_path(struct btrfs_root *root,
31db9f7c
AB		 998 u64 ino, struct fs_path *path)
999{
1000 int ret;
1001 struct btrfs_key key, found_key;
1002 struct btrfs_path *p;
1003
1004 p = alloc_path_for_send();
1005 if (!p)
1006 return -ENOMEM;
1007
1008 fs_path_reset(path);
1009
1010 key.objectid = ino;
1011 key.type = BTRFS_INODE_REF_KEY;
1012 key.offset = 0;
1013
1014 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1015 if (ret < 0)
1016 goto out;
1017 if (ret) {
1018 ret = 1;
1019 goto out;
1020 }
1021 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1022 if (found_key.objectid != ino ||
96b5bd77
JS		 1023 (found_key.type != BTRFS_INODE_REF_KEY &&
1024 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1025 ret = -ENOENT;
1026 goto out;
1027 }
1028
924794c9
TI		 1029 ret = iterate_inode_ref(root, p, &found_key, 1,
1030 __copy_first_ref, path);
31db9f7c
AB		 1031 if (ret < 0)
1032 goto out;
1033 ret = 0;
1034
1035out:
1036 btrfs_free_path(p);
1037 return ret;
1038}
1039
1040struct backref_ctx {
1041 struct send_ctx *sctx;
1042
1043 /* number of total found references */
1044 u64 found;
1045
1046 /*
1047 * used for clones found in send_root. clones found behind cur_objectid
1048 * and cur_offset are not considered as allowed clones.
1049 */
1050 u64 cur_objectid;
1051 u64 cur_offset;
1052
1053 /* may be truncated in case it's the last extent in a file */
1054 u64 extent_len;
1055
1056 /* Just to check for bugs in backref resolving */
ee849c04 1057 int found_itself;
31db9f7c
AB		 1058};
1059
1060static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1061{
995e01b7 1062 u64 root = (u64)(uintptr_t)key;
31db9f7c
AB		 1063 struct clone_root *cr = (struct clone_root *)elt;
1064
1065 if (root < cr->root->objectid)
1066 return -1;
1067 if (root > cr->root->objectid)
1068 return 1;
1069 return 0;
1070}
1071
1072static int __clone_root_cmp_sort(const void *e1, const void *e2)
1073{
1074 struct clone_root *cr1 = (struct clone_root *)e1;
1075 struct clone_root *cr2 = (struct clone_root *)e2;
1076
1077 if (cr1->root->objectid < cr2->root->objectid)
1078 return -1;
1079 if (cr1->root->objectid > cr2->root->objectid)
1080 return 1;
1081 return 0;
1082}
1083
1084/*
1085 * Called for every backref that is found for the current extent.
766702ef 1086 * Results are collected in sctx->clone_roots->ino/offset/found_refs
31db9f7c
AB		 1087 */
1088static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1089{
1090 struct backref_ctx *bctx = ctx_;
1091 struct clone_root *found;
1092 int ret;
1093 u64 i_size;
1094
1095 /* First check if the root is in the list of accepted clone sources */
995e01b7 1096 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
31db9f7c
AB		 1097 bctx->sctx->clone_roots_cnt,
1098 sizeof(struct clone_root),
1099 __clone_root_cmp_bsearch);
1100 if (!found)
1101 return 0;
1102
1103 if (found->root == bctx->sctx->send_root &&
1104 ino == bctx->cur_objectid &&
1105 offset == bctx->cur_offset) {
ee849c04 1106 bctx->found_itself = 1;
31db9f7c
AB		 1107 }
1108
1109 /*
766702ef 1110 * There are inodes that have extents that lie behind its i_size. Don't
31db9f7c
AB		 1111 * accept clones from these extents.
1112 */
85a7b33b
AB		 1113 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1114 NULL);
31db9f7c
AB		 1115 if (ret < 0)
1116 return ret;
1117
1118 if (offset + bctx->extent_len > i_size)
1119 return 0;
1120
1121 /*
1122 * Make sure we don't consider clones from send_root that are
1123 * behind the current inode/offset.
1124 */
1125 if (found->root == bctx->sctx->send_root) {
1126 /*
1127 * TODO for the moment we don't accept clones from the inode
1128 * that is currently send. We may change this when
1129 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1130 * file.
1131 */
1132 if (ino >= bctx->cur_objectid)
1133 return 0;
e938c8ad
AB		 1134#if 0
1135 if (ino > bctx->cur_objectid)
1136 return 0;
1137 if (offset + bctx->extent_len > bctx->cur_offset)
31db9f7c 1138 return 0;
e938c8ad 1139#endif
31db9f7c
AB		 1140 }
1141
1142 bctx->found++;
1143 found->found_refs++;
1144 if (ino < found->ino) {
1145 found->ino = ino;
1146 found->offset = offset;
1147 } else if (found->ino == ino) {
1148 /*
1149 * same extent found more then once in the same file.
1150 */
1151 if (found->offset > offset + bctx->extent_len)
1152 found->offset = offset;
1153 }
1154
1155 return 0;
1156}
1157
1158/*
766702ef
AB		 1159 * Given an inode, offset and extent item, it finds a good clone for a clone
1160 * instruction. Returns -ENOENT when none could be found. The function makes
1161 * sure that the returned clone is usable at the point where sending is at the
1162 * moment. This means, that no clones are accepted which lie behind the current
1163 * inode+offset.
1164 *
31db9f7c
AB		 1165 * path must point to the extent item when called.
1166 */
1167static int find_extent_clone(struct send_ctx *sctx,
1168 struct btrfs_path *path,
1169 u64 ino, u64 data_offset,
1170 u64 ino_size,
1171 struct clone_root **found)
1172{
1173 int ret;
1174 int extent_type;
1175 u64 logical;
74dd17fb 1176 u64 disk_byte;
31db9f7c
AB		 1177 u64 num_bytes;
1178 u64 extent_item_pos;
69917e43 1179 u64 flags = 0;
31db9f7c
AB		 1180 struct btrfs_file_extent_item *fi;
1181 struct extent_buffer *eb = path->nodes[0];
35075bb0 1182 struct backref_ctx *backref_ctx = NULL;
31db9f7c
AB		 1183 struct clone_root *cur_clone_root;
1184 struct btrfs_key found_key;
1185 struct btrfs_path *tmp_path;
74dd17fb 1186 int compressed;
31db9f7c
AB		 1187 u32 i;
1188
1189 tmp_path = alloc_path_for_send();
1190 if (!tmp_path)
1191 return -ENOMEM;
1192
35075bb0
AB		 1193 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1194 if (!backref_ctx) {
1195 ret = -ENOMEM;
1196 goto out;
1197 }
1198
31db9f7c
AB		 1199 if (data_offset >= ino_size) {
1200 /*
1201 * There may be extents that lie behind the file's size.
1202 * I at least had this in combination with snapshotting while
1203 * writing large files.
1204 */
1205 ret = 0;
1206 goto out;
1207 }
1208
1209 fi = btrfs_item_ptr(eb, path->slots[0],
1210 struct btrfs_file_extent_item);
1211 extent_type = btrfs_file_extent_type(eb, fi);
1212 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1213 ret = -ENOENT;
1214 goto out;
1215 }
74dd17fb 1216 compressed = btrfs_file_extent_compression(eb, fi);
31db9f7c
AB		 1217
1218 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
74dd17fb
CM		 1219 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1220 if (disk_byte == 0) {
31db9f7c
AB		 1221 ret = -ENOENT;
1222 goto out;
1223 }
74dd17fb 1224 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
31db9f7c 1225
69917e43
LB		 1226 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1227 &found_key, &flags);
31db9f7c
AB		 1228 btrfs_release_path(tmp_path);
1229
1230 if (ret < 0)
1231 goto out;
69917e43 1232 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
31db9f7c
AB		 1233 ret = -EIO;
1234 goto out;
1235 }
1236
1237 /*
1238 * Setup the clone roots.
1239 */
1240 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1241 cur_clone_root = sctx->clone_roots + i;
1242 cur_clone_root->ino = (u64)-1;
1243 cur_clone_root->offset = 0;
1244 cur_clone_root->found_refs = 0;
1245 }
1246
35075bb0
AB		 1247 backref_ctx->sctx = sctx;
1248 backref_ctx->found = 0;
1249 backref_ctx->cur_objectid = ino;
1250 backref_ctx->cur_offset = data_offset;
1251 backref_ctx->found_itself = 0;
1252 backref_ctx->extent_len = num_bytes;
31db9f7c
AB		 1253
1254 /*
1255 * The last extent of a file may be too large due to page alignment.
1256 * We need to adjust extent_len in this case so that the checks in
1257 * __iterate_backrefs work.
1258 */
1259 if (data_offset + num_bytes >= ino_size)
35075bb0 1260 backref_ctx->extent_len = ino_size - data_offset;
31db9f7c
AB		 1261
1262 /*
1263 * Now collect all backrefs.
1264 */
74dd17fb
CM		 1265 if (compressed == BTRFS_COMPRESS_NONE)
1266 extent_item_pos = logical - found_key.objectid;
1267 else
1268 extent_item_pos = 0;
1269
31db9f7c
AB		 1270 extent_item_pos = logical - found_key.objectid;
1271 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1272 found_key.objectid, extent_item_pos, 1,
35075bb0 1273 __iterate_backrefs, backref_ctx);
74dd17fb 1274
31db9f7c
AB		 1275 if (ret < 0)
1276 goto out;
1277
35075bb0 1278 if (!backref_ctx->found_itself) {
31db9f7c
AB		 1279 /* found a bug in backref code? */
1280 ret = -EIO;
1281 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1282 "send_root. inode=%llu, offset=%llu, "
74dd17fb
CM		 1283 "disk_byte=%llu found extent=%llu\n",
1284 ino, data_offset, disk_byte, found_key.objectid);
31db9f7c
AB		 1285 goto out;
1286 }
1287
1288verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1289 "ino=%llu, "
1290 "num_bytes=%llu, logical=%llu\n",
1291 data_offset, ino, num_bytes, logical);
1292
35075bb0 1293 if (!backref_ctx->found)
31db9f7c
AB		 1294 verbose_printk("btrfs: no clones found\n");
1295
1296 cur_clone_root = NULL;
1297 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1298 if (sctx->clone_roots[i].found_refs) {
1299 if (!cur_clone_root)
1300 cur_clone_root = sctx->clone_roots + i;
1301 else if (sctx->clone_roots[i].root == sctx->send_root)
1302 /* prefer clones from send_root over others */
1303 cur_clone_root = sctx->clone_roots + i;
31db9f7c
AB		 1304 }
1305
1306 }
1307
1308 if (cur_clone_root) {
1309 *found = cur_clone_root;
1310 ret = 0;
1311 } else {
1312 ret = -ENOENT;
1313 }
1314
1315out:
1316 btrfs_free_path(tmp_path);
35075bb0 1317 kfree(backref_ctx);
31db9f7c
AB		 1318 return ret;
1319}
1320
924794c9 1321static int read_symlink(struct btrfs_root *root,
31db9f7c
AB		 1322 u64 ino,
1323 struct fs_path *dest)
1324{
1325 int ret;
1326 struct btrfs_path *path;
1327 struct btrfs_key key;
1328 struct btrfs_file_extent_item *ei;
1329 u8 type;
1330 u8 compression;
1331 unsigned long off;
1332 int len;
1333
1334 path = alloc_path_for_send();
1335 if (!path)
1336 return -ENOMEM;
1337
1338 key.objectid = ino;
1339 key.type = BTRFS_EXTENT_DATA_KEY;
1340 key.offset = 0;
1341 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1342 if (ret < 0)
1343 goto out;
1344 BUG_ON(ret);
1345
1346 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1347 struct btrfs_file_extent_item);
1348 type = btrfs_file_extent_type(path->nodes[0], ei);
1349 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1350 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1351 BUG_ON(compression);
1352
1353 off = btrfs_file_extent_inline_start(ei);
1354 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1355
1356 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
31db9f7c
AB		 1357
1358out:
1359 btrfs_free_path(path);
1360 return ret;
1361}
1362
1363/*
1364 * Helper function to generate a file name that is unique in the root of
1365 * send_root and parent_root. This is used to generate names for orphan inodes.
1366 */
1367static int gen_unique_name(struct send_ctx *sctx,
1368 u64 ino, u64 gen,
1369 struct fs_path *dest)
1370{
1371 int ret = 0;
1372 struct btrfs_path *path;
1373 struct btrfs_dir_item *di;
1374 char tmp[64];
1375 int len;
1376 u64 idx = 0;
1377
1378 path = alloc_path_for_send();
1379 if (!path)
1380 return -ENOMEM;
1381
1382 while (1) {
1383 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1384 ino, gen, idx);
1385 if (len >= sizeof(tmp)) {
1386 /* should really not happen */
1387 ret = -EOVERFLOW;
1388 goto out;
1389 }
1390
1391 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1392 path, BTRFS_FIRST_FREE_OBJECTID,
1393 tmp, strlen(tmp), 0);
1394 btrfs_release_path(path);
1395 if (IS_ERR(di)) {
1396 ret = PTR_ERR(di);
1397 goto out;
1398 }
1399 if (di) {
1400 /* not unique, try again */
1401 idx++;
1402 continue;
1403 }
1404
1405 if (!sctx->parent_root) {
1406 /* unique */
1407 ret = 0;
1408 break;
1409 }
1410
1411 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1412 path, BTRFS_FIRST_FREE_OBJECTID,
1413 tmp, strlen(tmp), 0);
1414 btrfs_release_path(path);
1415 if (IS_ERR(di)) {
1416 ret = PTR_ERR(di);
1417 goto out;
1418 }
1419 if (di) {
1420 /* not unique, try again */
1421 idx++;
1422 continue;
1423 }
1424 /* unique */
1425 break;
1426 }
1427
1428 ret = fs_path_add(dest, tmp, strlen(tmp));
1429
1430out:
1431 btrfs_free_path(path);
1432 return ret;
1433}
1434
1435enum inode_state {
1436 inode_state_no_change,
1437 inode_state_will_create,
1438 inode_state_did_create,
1439 inode_state_will_delete,
1440 inode_state_did_delete,
1441};
1442
1443static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1444{
1445 int ret;
1446 int left_ret;
1447 int right_ret;
1448 u64 left_gen;
1449 u64 right_gen;
1450
1451 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
85a7b33b 1452 NULL, NULL);
31db9f7c
AB		 1453 if (ret < 0 && ret != -ENOENT)
1454 goto out;
1455 left_ret = ret;
1456
1457 if (!sctx->parent_root) {
1458 right_ret = -ENOENT;
1459 } else {
1460 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
85a7b33b 1461 NULL, NULL, NULL, NULL);
31db9f7c
AB		 1462 if (ret < 0 && ret != -ENOENT)
1463 goto out;
1464 right_ret = ret;
1465 }
1466
1467 if (!left_ret && !right_ret) {
e938c8ad 1468 if (left_gen == gen && right_gen == gen) {
31db9f7c 1469 ret = inode_state_no_change;
e938c8ad 1470 } else if (left_gen == gen) {
31db9f7c
AB		 1471 if (ino < sctx->send_progress)
1472 ret = inode_state_did_create;
1473 else
1474 ret = inode_state_will_create;
1475 } else if (right_gen == gen) {
1476 if (ino < sctx->send_progress)
1477 ret = inode_state_did_delete;
1478 else
1479 ret = inode_state_will_delete;
1480 } else {
1481 ret = -ENOENT;
1482 }
1483 } else if (!left_ret) {
1484 if (left_gen == gen) {
1485 if (ino < sctx->send_progress)
1486 ret = inode_state_did_create;
1487 else
1488 ret = inode_state_will_create;
1489 } else {
1490 ret = -ENOENT;
1491 }
1492 } else if (!right_ret) {
1493 if (right_gen == gen) {
1494 if (ino < sctx->send_progress)
1495 ret = inode_state_did_delete;
1496 else
1497 ret = inode_state_will_delete;
1498 } else {
1499 ret = -ENOENT;
1500 }
1501 } else {
1502 ret = -ENOENT;
1503 }
1504
1505out:
1506 return ret;
1507}
1508
1509static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1510{
1511 int ret;
1512
1513 ret = get_cur_inode_state(sctx, ino, gen);
1514 if (ret < 0)
1515 goto out;
1516
1517 if (ret == inode_state_no_change ||
1518 ret == inode_state_did_create ||
1519 ret == inode_state_will_delete)
1520 ret = 1;
1521 else
1522 ret = 0;
1523
1524out:
1525 return ret;
1526}
1527
1528/*
1529 * Helper function to lookup a dir item in a dir.
1530 */
1531static int lookup_dir_item_inode(struct btrfs_root *root,
1532 u64 dir, const char *name, int name_len,
1533 u64 *found_inode,
1534 u8 *found_type)
1535{
1536 int ret = 0;
1537 struct btrfs_dir_item *di;
1538 struct btrfs_key key;
1539 struct btrfs_path *path;
1540
1541 path = alloc_path_for_send();
1542 if (!path)
1543 return -ENOMEM;
1544
1545 di = btrfs_lookup_dir_item(NULL, root, path,
1546 dir, name, name_len, 0);
1547 if (!di) {
1548 ret = -ENOENT;
1549 goto out;
1550 }
1551 if (IS_ERR(di)) {
1552 ret = PTR_ERR(di);
1553 goto out;
1554 }
1555 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1556 *found_inode = key.objectid;
1557 *found_type = btrfs_dir_type(path->nodes[0], di);
1558
1559out:
1560 btrfs_free_path(path);
1561 return ret;
1562}
1563
766702ef
AB		 1564/*
1565 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1566 * generation of the parent dir and the name of the dir entry.
1567 */
924794c9 1568static int get_first_ref(struct btrfs_root *root, u64 ino,
31db9f7c
AB		 1569 u64 *dir, u64 *dir_gen, struct fs_path *name)
1570{
1571 int ret;
1572 struct btrfs_key key;
1573 struct btrfs_key found_key;
1574 struct btrfs_path *path;
31db9f7c 1575 int len;
96b5bd77 1576 u64 parent_dir;
31db9f7c
AB		 1577
1578 path = alloc_path_for_send();
1579 if (!path)
1580 return -ENOMEM;
1581
1582 key.objectid = ino;
1583 key.type = BTRFS_INODE_REF_KEY;
1584 key.offset = 0;
1585
1586 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1587 if (ret < 0)
1588 goto out;
1589 if (!ret)
1590 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1591 path->slots[0]);
96b5bd77
JS		 1592 if (ret || found_key.objectid != ino ||
1593 (found_key.type != BTRFS_INODE_REF_KEY &&
1594 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1595 ret = -ENOENT;
1596 goto out;
1597 }
1598
96b5bd77
JS		 1599 if (key.type == BTRFS_INODE_REF_KEY) {
1600 struct btrfs_inode_ref *iref;
1601 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1602 struct btrfs_inode_ref);
1603 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1604 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1605 (unsigned long)(iref + 1),
1606 len);
1607 parent_dir = found_key.offset;
1608 } else {
1609 struct btrfs_inode_extref *extref;
1610 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1611 struct btrfs_inode_extref);
1612 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1613 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1614 (unsigned long)&extref->name, len);
1615 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1616 }
31db9f7c
AB		 1617 if (ret < 0)
1618 goto out;
1619 btrfs_release_path(path);
1620
96b5bd77 1621 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
85a7b33b 1622 NULL, NULL);
31db9f7c
AB		 1623 if (ret < 0)
1624 goto out;
1625
96b5bd77 1626 *dir = parent_dir;
31db9f7c
AB		 1627
1628out:
1629 btrfs_free_path(path);
1630 return ret;
1631}
1632
924794c9 1633static int is_first_ref(struct btrfs_root *root,
31db9f7c
AB		 1634 u64 ino, u64 dir,
1635 const char *name, int name_len)
1636{
1637 int ret;
1638 struct fs_path *tmp_name;
1639 u64 tmp_dir;
1640 u64 tmp_dir_gen;
1641
924794c9 1642 tmp_name = fs_path_alloc();
31db9f7c
AB		 1643 if (!tmp_name)
1644 return -ENOMEM;
1645
924794c9 1646 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
31db9f7c
AB		 1647 if (ret < 0)
1648 goto out;
1649
b9291aff 1650 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
31db9f7c
AB		 1651 ret = 0;
1652 goto out;
1653 }
1654
e938c8ad 1655 ret = !memcmp(tmp_name->start, name, name_len);
31db9f7c
AB		 1656
1657out:
924794c9 1658 fs_path_free(tmp_name);
31db9f7c
AB		 1659 return ret;
1660}
1661
766702ef
AB		 1662/*
1663 * Used by process_recorded_refs to determine if a new ref would overwrite an
1664 * already existing ref. In case it detects an overwrite, it returns the
1665 * inode/gen in who_ino/who_gen.
1666 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1667 * to make sure later references to the overwritten inode are possible.
1668 * Orphanizing is however only required for the first ref of an inode.
1669 * process_recorded_refs does an additional is_first_ref check to see if
1670 * orphanizing is really required.
1671 */
31db9f7c
AB		 1672static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1673 const char *name, int name_len,
1674 u64 *who_ino, u64 *who_gen)
1675{
1676 int ret = 0;
ebdad913 1677 u64 gen;
31db9f7c
AB		 1678 u64 other_inode = 0;
1679 u8 other_type = 0;
1680
1681 if (!sctx->parent_root)
1682 goto out;
1683
1684 ret = is_inode_existent(sctx, dir, dir_gen);
1685 if (ret <= 0)
1686 goto out;
1687
ebdad913
JB		 1688 /*
1689 * If we have a parent root we need to verify that the parent dir was
1690 * not delted and then re-created, if it was then we have no overwrite
1691 * and we can just unlink this entry.
1692 */
1693 if (sctx->parent_root) {
1694 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1695 NULL, NULL, NULL);
1696 if (ret < 0 && ret != -ENOENT)
1697 goto out;
1698 if (ret) {
1699 ret = 0;
1700 goto out;
1701 }
1702 if (gen != dir_gen)
1703 goto out;
1704 }
1705
31db9f7c
AB		 1706 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1707 &other_inode, &other_type);
1708 if (ret < 0 && ret != -ENOENT)
1709 goto out;
1710 if (ret) {
1711 ret = 0;
1712 goto out;
1713 }
1714
766702ef
AB		 1715 /*
1716 * Check if the overwritten ref was already processed. If yes, the ref
1717 * was already unlinked/moved, so we can safely assume that we will not
1718 * overwrite anything at this point in time.
1719 */
31db9f7c
AB		 1720 if (other_inode > sctx->send_progress) {
1721 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
85a7b33b 1722 who_gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 1723 if (ret < 0)
1724 goto out;
1725
1726 ret = 1;
1727 *who_ino = other_inode;
1728 } else {
1729 ret = 0;
1730 }
1731
1732out:
1733 return ret;
1734}
1735
766702ef
AB		 1736/*
1737 * Checks if the ref was overwritten by an already processed inode. This is
1738 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1739 * thus the orphan name needs be used.
1740 * process_recorded_refs also uses it to avoid unlinking of refs that were
1741 * overwritten.
1742 */
31db9f7c
AB		 1743static int did_overwrite_ref(struct send_ctx *sctx,
1744 u64 dir, u64 dir_gen,
1745 u64 ino, u64 ino_gen,
1746 const char *name, int name_len)
1747{
1748 int ret = 0;
1749 u64 gen;
1750 u64 ow_inode;
1751 u8 other_type;
1752
1753 if (!sctx->parent_root)
1754 goto out;
1755
1756 ret = is_inode_existent(sctx, dir, dir_gen);
1757 if (ret <= 0)
1758 goto out;
1759
1760 /* check if the ref was overwritten by another ref */
1761 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1762 &ow_inode, &other_type);
1763 if (ret < 0 && ret != -ENOENT)
1764 goto out;
1765 if (ret) {
1766 /* was never and will never be overwritten */
1767 ret = 0;
1768 goto out;
1769 }
1770
1771 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
85a7b33b 1772 NULL, NULL);
31db9f7c
AB		 1773 if (ret < 0)
1774 goto out;
1775
1776 if (ow_inode == ino && gen == ino_gen) {
1777 ret = 0;
1778 goto out;
1779 }
1780
1781 /* we know that it is or will be overwritten. check this now */
1782 if (ow_inode < sctx->send_progress)
1783 ret = 1;
1784 else
1785 ret = 0;
1786
1787out:
1788 return ret;
1789}
1790
766702ef
AB		 1791/*
1792 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1793 * that got overwritten. This is used by process_recorded_refs to determine
1794 * if it has to use the path as returned by get_cur_path or the orphan name.
1795 */
31db9f7c
AB		 1796static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1797{
1798 int ret = 0;
1799 struct fs_path *name = NULL;
1800 u64 dir;
1801 u64 dir_gen;
1802
1803 if (!sctx->parent_root)
1804 goto out;
1805
924794c9 1806 name = fs_path_alloc();
31db9f7c
AB		 1807 if (!name)
1808 return -ENOMEM;
1809
924794c9 1810 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
31db9f7c
AB		 1811 if (ret < 0)
1812 goto out;
1813
1814 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1815 name->start, fs_path_len(name));
31db9f7c
AB		 1816
1817out:
924794c9 1818 fs_path_free(name);
31db9f7c
AB		 1819 return ret;
1820}
1821
766702ef
AB		 1822/*
1823 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1824 * so we need to do some special handling in case we have clashes. This function
1825 * takes care of this with the help of name_cache_entry::radix_list.
5dc67d0b 1826 * In case of error, nce is kfreed.
766702ef 1827 */
31db9f7c
AB		 1828static int name_cache_insert(struct send_ctx *sctx,
1829 struct name_cache_entry *nce)
1830{
1831 int ret = 0;
7e0926fe
AB		 1832 struct list_head *nce_head;
1833
1834 nce_head = radix_tree_lookup(&sctx->name_cache,
1835 (unsigned long)nce->ino);
1836 if (!nce_head) {
1837 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
cfa7a9cc
TI		 1838 if (!nce_head) {
1839 kfree(nce);
31db9f7c 1840 return -ENOMEM;
cfa7a9cc 1841 }
7e0926fe 1842 INIT_LIST_HEAD(nce_head);
31db9f7c 1843
7e0926fe 1844 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
5dc67d0b
AB		 1845 if (ret < 0) {
1846 kfree(nce_head);
1847 kfree(nce);
31db9f7c 1848 return ret;
5dc67d0b 1849 }
31db9f7c 1850 }
7e0926fe 1851 list_add_tail(&nce->radix_list, nce_head);
31db9f7c
AB		 1852 list_add_tail(&nce->list, &sctx->name_cache_list);
1853 sctx->name_cache_size++;
1854
1855 return ret;
1856}
1857
1858static void name_cache_delete(struct send_ctx *sctx,
1859 struct name_cache_entry *nce)
1860{
7e0926fe 1861 struct list_head *nce_head;
31db9f7c 1862
7e0926fe
AB		 1863 nce_head = radix_tree_lookup(&sctx->name_cache,
1864 (unsigned long)nce->ino);
1865 BUG_ON(!nce_head);
31db9f7c 1866
7e0926fe 1867 list_del(&nce->radix_list);
31db9f7c 1868 list_del(&nce->list);
31db9f7c 1869 sctx->name_cache_size--;
7e0926fe
AB		 1870
1871 if (list_empty(nce_head)) {
1872 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1873 kfree(nce_head);
1874 }
31db9f7c
AB		 1875}
1876
1877static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1878 u64 ino, u64 gen)
1879{
7e0926fe
AB		 1880 struct list_head *nce_head;
1881 struct name_cache_entry *cur;
31db9f7c 1882
7e0926fe
AB		 1883 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1884 if (!nce_head)
31db9f7c
AB		 1885 return NULL;
1886
7e0926fe
AB		 1887 list_for_each_entry(cur, nce_head, radix_list) {
1888 if (cur->ino == ino && cur->gen == gen)
1889 return cur;
1890 }
31db9f7c
AB		 1891 return NULL;
1892}
1893
766702ef
AB		 1894/*
1895 * Removes the entry from the list and adds it back to the end. This marks the
1896 * entry as recently used so that name_cache_clean_unused does not remove it.
1897 */
31db9f7c
AB		 1898static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1899{
1900 list_del(&nce->list);
1901 list_add_tail(&nce->list, &sctx->name_cache_list);
1902}
1903
766702ef
AB		 1904/*
1905 * Remove some entries from the beginning of name_cache_list.
1906 */
31db9f7c
AB		 1907static void name_cache_clean_unused(struct send_ctx *sctx)
1908{
1909 struct name_cache_entry *nce;
1910
1911 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1912 return;
1913
1914 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1915 nce = list_entry(sctx->name_cache_list.next,
1916 struct name_cache_entry, list);
1917 name_cache_delete(sctx, nce);
1918 kfree(nce);
1919 }
1920}
1921
1922static void name_cache_free(struct send_ctx *sctx)
1923{
1924 struct name_cache_entry *nce;
31db9f7c 1925
e938c8ad
AB		 1926 while (!list_empty(&sctx->name_cache_list)) {
1927 nce = list_entry(sctx->name_cache_list.next,
1928 struct name_cache_entry, list);
31db9f7c 1929 name_cache_delete(sctx, nce);
17589bd9 1930 kfree(nce);
31db9f7c
AB		 1931 }
1932}
1933
766702ef
AB		 1934/*
1935 * Used by get_cur_path for each ref up to the root.
1936 * Returns 0 if it succeeded.
1937 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1938 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1939 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1940 * Returns <0 in case of error.
1941 */
31db9f7c
AB		 1942static int __get_cur_name_and_parent(struct send_ctx *sctx,
1943 u64 ino, u64 gen,
1944 u64 *parent_ino,
1945 u64 *parent_gen,
1946 struct fs_path *dest)
1947{
1948 int ret;
1949 int nce_ret;
1950 struct btrfs_path *path = NULL;
1951 struct name_cache_entry *nce = NULL;
1952
766702ef
AB		 1953 /*
1954 * First check if we already did a call to this function with the same
1955 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1956 * return the cached result.
1957 */
31db9f7c
AB		 1958 nce = name_cache_search(sctx, ino, gen);
1959 if (nce) {
1960 if (ino < sctx->send_progress && nce->need_later_update) {
1961 name_cache_delete(sctx, nce);
1962 kfree(nce);
1963 nce = NULL;
1964 } else {
1965 name_cache_used(sctx, nce);
1966 *parent_ino = nce->parent_ino;
1967 *parent_gen = nce->parent_gen;
1968 ret = fs_path_add(dest, nce->name, nce->name_len);
1969 if (ret < 0)
1970 goto out;
1971 ret = nce->ret;
1972 goto out;
1973 }
1974 }
1975
1976 path = alloc_path_for_send();
1977 if (!path)
1978 return -ENOMEM;
1979
766702ef
AB		 1980 /*
1981 * If the inode is not existent yet, add the orphan name and return 1.
1982 * This should only happen for the parent dir that we determine in
1983 * __record_new_ref
1984 */
31db9f7c
AB		 1985 ret = is_inode_existent(sctx, ino, gen);
1986 if (ret < 0)
1987 goto out;
1988
1989 if (!ret) {
1990 ret = gen_unique_name(sctx, ino, gen, dest);
1991 if (ret < 0)
1992 goto out;
1993 ret = 1;
1994 goto out_cache;
1995 }
1996
766702ef
AB		 1997 /*
1998 * Depending on whether the inode was already processed or not, use
1999 * send_root or parent_root for ref lookup.
2000 */
31db9f7c 2001 if (ino < sctx->send_progress)
924794c9
TI		 2002 ret = get_first_ref(sctx->send_root, ino,
2003 parent_ino, parent_gen, dest);
31db9f7c 2004 else
924794c9
TI		 2005 ret = get_first_ref(sctx->parent_root, ino,
2006 parent_ino, parent_gen, dest);
31db9f7c
AB		 2007 if (ret < 0)
2008 goto out;
2009
766702ef
AB		 2010 /*
2011 * Check if the ref was overwritten by an inode's ref that was processed
2012 * earlier. If yes, treat as orphan and return 1.
2013 */
31db9f7c
AB		 2014 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2015 dest->start, dest->end - dest->start);
2016 if (ret < 0)
2017 goto out;
2018 if (ret) {
2019 fs_path_reset(dest);
2020 ret = gen_unique_name(sctx, ino, gen, dest);
2021 if (ret < 0)
2022 goto out;
2023 ret = 1;
2024 }
2025
2026out_cache:
766702ef
AB		 2027 /*
2028 * Store the result of the lookup in the name cache.
2029 */
31db9f7c
AB		 2030 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2031 if (!nce) {
2032 ret = -ENOMEM;
2033 goto out;
2034 }
2035
2036 nce->ino = ino;
2037 nce->gen = gen;
2038 nce->parent_ino = *parent_ino;
2039 nce->parent_gen = *parent_gen;
2040 nce->name_len = fs_path_len(dest);
2041 nce->ret = ret;
2042 strcpy(nce->name, dest->start);
31db9f7c
AB		 2043
2044 if (ino < sctx->send_progress)
2045 nce->need_later_update = 0;
2046 else
2047 nce->need_later_update = 1;
2048
2049 nce_ret = name_cache_insert(sctx, nce);
2050 if (nce_ret < 0)
2051 ret = nce_ret;
2052 name_cache_clean_unused(sctx);
2053
2054out:
2055 btrfs_free_path(path);
2056 return ret;
2057}
2058
2059/*
2060 * Magic happens here. This function returns the first ref to an inode as it
2061 * would look like while receiving the stream at this point in time.
2062 * We walk the path up to the root. For every inode in between, we check if it
2063 * was already processed/sent. If yes, we continue with the parent as found
2064 * in send_root. If not, we continue with the parent as found in parent_root.
2065 * If we encounter an inode that was deleted at this point in time, we use the
2066 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2067 * that were not created yet and overwritten inodes/refs.
2068 *
2069 * When do we have have orphan inodes:
2070 * 1. When an inode is freshly created and thus no valid refs are available yet
2071 * 2. When a directory lost all it's refs (deleted) but still has dir items
2072 * inside which were not processed yet (pending for move/delete). If anyone
2073 * tried to get the path to the dir items, it would get a path inside that
2074 * orphan directory.
2075 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2076 * of an unprocessed inode. If in that case the first ref would be
2077 * overwritten, the overwritten inode gets "orphanized". Later when we
2078 * process this overwritten inode, it is restored at a new place by moving
2079 * the orphan inode.
2080 *
2081 * sctx->send_progress tells this function at which point in time receiving
2082 * would be.
2083 */
2084static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2085 struct fs_path *dest)
2086{
2087 int ret = 0;
2088 struct fs_path *name = NULL;
2089 u64 parent_inode = 0;
2090 u64 parent_gen = 0;
2091 int stop = 0;
2092
924794c9 2093 name = fs_path_alloc();
31db9f7c
AB		 2094 if (!name) {
2095 ret = -ENOMEM;
2096 goto out;
2097 }
2098
2099 dest->reversed = 1;
2100 fs_path_reset(dest);
2101
2102 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2103 fs_path_reset(name);
2104
2105 ret = __get_cur_name_and_parent(sctx, ino, gen,
2106 &parent_inode, &parent_gen, name);
2107 if (ret < 0)
2108 goto out;
2109 if (ret)
2110 stop = 1;
2111
2112 ret = fs_path_add_path(dest, name);
2113 if (ret < 0)
2114 goto out;
2115
2116 ino = parent_inode;
2117 gen = parent_gen;
2118 }
2119
2120out:
924794c9 2121 fs_path_free(name);
31db9f7c
AB		 2122 if (!ret)
2123 fs_path_unreverse(dest);
2124 return ret;
2125}
2126
31db9f7c
AB		 2127/*
2128 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2129 */
2130static int send_subvol_begin(struct send_ctx *sctx)
2131{
2132 int ret;
2133 struct btrfs_root *send_root = sctx->send_root;
2134 struct btrfs_root *parent_root = sctx->parent_root;
2135 struct btrfs_path *path;
2136 struct btrfs_key key;
2137 struct btrfs_root_ref *ref;
2138 struct extent_buffer *leaf;
2139 char *name = NULL;
2140 int namelen;
2141
2142 path = alloc_path_for_send();
2143 if (!path)
2144 return -ENOMEM;
2145
2146 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2147 if (!name) {
2148 btrfs_free_path(path);
2149 return -ENOMEM;
2150 }
2151
2152 key.objectid = send_root->objectid;
2153 key.type = BTRFS_ROOT_BACKREF_KEY;
2154 key.offset = 0;
2155
2156 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2157 &key, path, 1, 0);
2158 if (ret < 0)
2159 goto out;
2160 if (ret) {
2161 ret = -ENOENT;
2162 goto out;
2163 }
2164
2165 leaf = path->nodes[0];
2166 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2167 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2168 key.objectid != send_root->objectid) {
2169 ret = -ENOENT;
2170 goto out;
2171 }
2172 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2173 namelen = btrfs_root_ref_name_len(leaf, ref);
2174 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2175 btrfs_release_path(path);
2176
31db9f7c
AB		 2177 if (parent_root) {
2178 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2179 if (ret < 0)
2180 goto out;
2181 } else {
2182 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2183 if (ret < 0)
2184 goto out;
2185 }
2186
2187 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2188 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2189 sctx->send_root->root_item.uuid);
2190 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2191 sctx->send_root->root_item.ctransid);
2192 if (parent_root) {
2193 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2194 sctx->parent_root->root_item.uuid);
2195 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2196 sctx->parent_root->root_item.ctransid);
2197 }
2198
2199 ret = send_cmd(sctx);
2200
2201tlv_put_failure:
2202out:
2203 btrfs_free_path(path);
2204 kfree(name);
2205 return ret;
2206}
2207
2208static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2209{
2210 int ret = 0;
2211 struct fs_path *p;
2212
2213verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2214
924794c9 2215 p = fs_path_alloc();
31db9f7c
AB		 2216 if (!p)
2217 return -ENOMEM;
2218
2219 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2220 if (ret < 0)
2221 goto out;
2222
2223 ret = get_cur_path(sctx, ino, gen, p);
2224 if (ret < 0)
2225 goto out;
2226 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2227 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2228
2229 ret = send_cmd(sctx);
2230
2231tlv_put_failure:
2232out:
924794c9 2233 fs_path_free(p);
31db9f7c
AB		 2234 return ret;
2235}
2236
2237static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2238{
2239 int ret = 0;
2240 struct fs_path *p;
2241
2242verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2243
924794c9 2244 p = fs_path_alloc();
31db9f7c
AB		 2245 if (!p)
2246 return -ENOMEM;
2247
2248 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2249 if (ret < 0)
2250 goto out;
2251
2252 ret = get_cur_path(sctx, ino, gen, p);
2253 if (ret < 0)
2254 goto out;
2255 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2256 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2257
2258 ret = send_cmd(sctx);
2259
2260tlv_put_failure:
2261out:
924794c9 2262 fs_path_free(p);
31db9f7c
AB		 2263 return ret;
2264}
2265
2266static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2267{
2268 int ret = 0;
2269 struct fs_path *p;
2270
2271verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2272
924794c9 2273 p = fs_path_alloc();
31db9f7c
AB		 2274 if (!p)
2275 return -ENOMEM;
2276
2277 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2278 if (ret < 0)
2279 goto out;
2280
2281 ret = get_cur_path(sctx, ino, gen, p);
2282 if (ret < 0)
2283 goto out;
2284 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2285 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2286 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2287
2288 ret = send_cmd(sctx);
2289
2290tlv_put_failure:
2291out:
924794c9 2292 fs_path_free(p);
31db9f7c
AB		 2293 return ret;
2294}
2295
2296static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2297{
2298 int ret = 0;
2299 struct fs_path *p = NULL;
2300 struct btrfs_inode_item *ii;
2301 struct btrfs_path *path = NULL;
2302 struct extent_buffer *eb;
2303 struct btrfs_key key;
2304 int slot;
2305
2306verbose_printk("btrfs: send_utimes %llu\n", ino);
2307
924794c9 2308 p = fs_path_alloc();
31db9f7c
AB		 2309 if (!p)
2310 return -ENOMEM;
2311
2312 path = alloc_path_for_send();
2313 if (!path) {
2314 ret = -ENOMEM;
2315 goto out;
2316 }
2317
2318 key.objectid = ino;
2319 key.type = BTRFS_INODE_ITEM_KEY;
2320 key.offset = 0;
2321 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2322 if (ret < 0)
2323 goto out;
2324
2325 eb = path->nodes[0];
2326 slot = path->slots[0];
2327 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2328
2329 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2330 if (ret < 0)
2331 goto out;
2332
2333 ret = get_cur_path(sctx, ino, gen, p);
2334 if (ret < 0)
2335 goto out;
2336 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2337 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2338 btrfs_inode_atime(ii));
2339 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2340 btrfs_inode_mtime(ii));
2341 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2342 btrfs_inode_ctime(ii));
766702ef 2343 /* TODO Add otime support when the otime patches get into upstream */
31db9f7c
AB		 2344
2345 ret = send_cmd(sctx);
2346
2347tlv_put_failure:
2348out:
924794c9 2349 fs_path_free(p);
31db9f7c
AB		 2350 btrfs_free_path(path);
2351 return ret;
2352}
2353
2354/*
2355 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2356 * a valid path yet because we did not process the refs yet. So, the inode
2357 * is created as orphan.
2358 */
1f4692da 2359static int send_create_inode(struct send_ctx *sctx, u64 ino)
31db9f7c
AB		 2360{
2361 int ret = 0;
31db9f7c 2362 struct fs_path *p;
31db9f7c 2363 int cmd;
1f4692da 2364 u64 gen;
31db9f7c 2365 u64 mode;
1f4692da 2366 u64 rdev;
31db9f7c 2367
1f4692da 2368verbose_printk("btrfs: send_create_inode %llu\n", ino);
31db9f7c 2369
924794c9 2370 p = fs_path_alloc();
31db9f7c
AB		 2371 if (!p)
2372 return -ENOMEM;
2373
1f4692da
AB		 2374 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2375 NULL, &rdev);
2376 if (ret < 0)
2377 goto out;
31db9f7c 2378
e938c8ad 2379 if (S_ISREG(mode)) {
31db9f7c 2380 cmd = BTRFS_SEND_C_MKFILE;
e938c8ad 2381 } else if (S_ISDIR(mode)) {
31db9f7c 2382 cmd = BTRFS_SEND_C_MKDIR;
e938c8ad 2383 } else if (S_ISLNK(mode)) {
31db9f7c 2384 cmd = BTRFS_SEND_C_SYMLINK;
e938c8ad 2385 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
31db9f7c 2386 cmd = BTRFS_SEND_C_MKNOD;
e938c8ad 2387 } else if (S_ISFIFO(mode)) {
31db9f7c 2388 cmd = BTRFS_SEND_C_MKFIFO;
e938c8ad 2389 } else if (S_ISSOCK(mode)) {
31db9f7c 2390 cmd = BTRFS_SEND_C_MKSOCK;
e938c8ad 2391 } else {
31db9f7c
AB		 2392 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2393 (int)(mode & S_IFMT));
2394 ret = -ENOTSUPP;
2395 goto out;
2396 }
2397
2398 ret = begin_cmd(sctx, cmd);
2399 if (ret < 0)
2400 goto out;
2401
1f4692da 2402 ret = gen_unique_name(sctx, ino, gen, p);
31db9f7c
AB		 2403 if (ret < 0)
2404 goto out;
2405
2406 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
1f4692da 2407 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
31db9f7c
AB		 2408
2409 if (S_ISLNK(mode)) {
2410 fs_path_reset(p);
924794c9 2411 ret = read_symlink(sctx->send_root, ino, p);
31db9f7c
AB		 2412 if (ret < 0)
2413 goto out;
2414 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2415 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2416 S_ISFIFO(mode) || S_ISSOCK(mode)) {
d79e5043
AJ		 2417 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2418 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
31db9f7c
AB		 2419 }
2420
2421 ret = send_cmd(sctx);
2422 if (ret < 0)
2423 goto out;
2424
2425
2426tlv_put_failure:
2427out:
924794c9 2428 fs_path_free(p);
31db9f7c
AB		 2429 return ret;
2430}
2431
1f4692da
AB		 2432/*
2433 * We need some special handling for inodes that get processed before the parent
2434 * directory got created. See process_recorded_refs for details.
2435 * This function does the check if we already created the dir out of order.
2436 */
2437static int did_create_dir(struct send_ctx *sctx, u64 dir)
2438{
2439 int ret = 0;
2440 struct btrfs_path *path = NULL;
2441 struct btrfs_key key;
2442 struct btrfs_key found_key;
2443 struct btrfs_key di_key;
2444 struct extent_buffer *eb;
2445 struct btrfs_dir_item *di;
2446 int slot;
2447
2448 path = alloc_path_for_send();
2449 if (!path) {
2450 ret = -ENOMEM;
2451 goto out;
2452 }
2453
2454 key.objectid = dir;
2455 key.type = BTRFS_DIR_INDEX_KEY;
2456 key.offset = 0;
2457 while (1) {
2458 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2459 1, 0);
2460 if (ret < 0)
2461 goto out;
2462 if (!ret) {
2463 eb = path->nodes[0];
2464 slot = path->slots[0];
2465 btrfs_item_key_to_cpu(eb, &found_key, slot);
2466 }
2467 if (ret || found_key.objectid != key.objectid ||
2468 found_key.type != key.type) {
2469 ret = 0;
2470 goto out;
2471 }
2472
2473 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2474 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2475
a0525414
JB		 2476 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2477 di_key.objectid < sctx->send_progress) {
1f4692da
AB		 2478 ret = 1;
2479 goto out;
2480 }
2481
2482 key.offset = found_key.offset + 1;
2483 btrfs_release_path(path);
2484 }
2485
2486out:
2487 btrfs_free_path(path);
2488 return ret;
2489}
2490
2491/*
2492 * Only creates the inode if it is:
2493 * 1. Not a directory
2494 * 2. Or a directory which was not created already due to out of order
2495 * directories. See did_create_dir and process_recorded_refs for details.
2496 */
2497static int send_create_inode_if_needed(struct send_ctx *sctx)
2498{
2499 int ret;
2500
2501 if (S_ISDIR(sctx->cur_inode_mode)) {
2502 ret = did_create_dir(sctx, sctx->cur_ino);
2503 if (ret < 0)
2504 goto out;
2505 if (ret) {
2506 ret = 0;
2507 goto out;
2508 }
2509 }
2510
2511 ret = send_create_inode(sctx, sctx->cur_ino);
2512 if (ret < 0)
2513 goto out;
2514
2515out:
2516 return ret;
2517}
2518
31db9f7c
AB		 2519struct recorded_ref {
2520 struct list_head list;
2521 char *dir_path;
2522 char *name;
2523 struct fs_path *full_path;
2524 u64 dir;
2525 u64 dir_gen;
2526 int dir_path_len;
2527 int name_len;
2528};
2529
2530/*
2531 * We need to process new refs before deleted refs, but compare_tree gives us
2532 * everything mixed. So we first record all refs and later process them.
2533 * This function is a helper to record one ref.
2534 */
2535static int record_ref(struct list_head *head, u64 dir,
2536 u64 dir_gen, struct fs_path *path)
2537{
2538 struct recorded_ref *ref;
31db9f7c
AB		 2539
2540 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2541 if (!ref)
2542 return -ENOMEM;
2543
2544 ref->dir = dir;
2545 ref->dir_gen = dir_gen;
2546 ref->full_path = path;
2547
ed84885d
AS		 2548 ref->name = (char *)kbasename(ref->full_path->start);
2549 ref->name_len = ref->full_path->end - ref->name;
2550 ref->dir_path = ref->full_path->start;
2551 if (ref->name == ref->full_path->start)
31db9f7c 2552 ref->dir_path_len = 0;
ed84885d 2553 else
31db9f7c
AB		 2554 ref->dir_path_len = ref->full_path->end -
2555 ref->full_path->start - 1 - ref->name_len;
31db9f7c
AB		 2556
2557 list_add_tail(&ref->list, head);
2558 return 0;
2559}
2560
ba5e8f2e
JB		 2561static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2562{
2563 struct recorded_ref *new;
2564
2565 new = kmalloc(sizeof(*ref), GFP_NOFS);
2566 if (!new)
2567 return -ENOMEM;
2568
2569 new->dir = ref->dir;
2570 new->dir_gen = ref->dir_gen;
2571 new->full_path = NULL;
2572 INIT_LIST_HEAD(&new->list);
2573 list_add_tail(&new->list, list);
2574 return 0;
2575}
2576
924794c9 2577static void __free_recorded_refs(struct list_head *head)
31db9f7c
AB		 2578{
2579 struct recorded_ref *cur;
31db9f7c 2580
e938c8ad
AB		 2581 while (!list_empty(head)) {
2582 cur = list_entry(head->next, struct recorded_ref, list);
924794c9 2583 fs_path_free(cur->full_path);
e938c8ad 2584 list_del(&cur->list);
31db9f7c
AB		 2585 kfree(cur);
2586 }
31db9f7c
AB		 2587}
2588
2589static void free_recorded_refs(struct send_ctx *sctx)
2590{
924794c9
TI		 2591 __free_recorded_refs(&sctx->new_refs);
2592 __free_recorded_refs(&sctx->deleted_refs);
31db9f7c
AB		 2593}
2594
2595/*
766702ef 2596 * Renames/moves a file/dir to its orphan name. Used when the first
31db9f7c
AB		 2597 * ref of an unprocessed inode gets overwritten and for all non empty
2598 * directories.
2599 */
2600static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2601 struct fs_path *path)
2602{
2603 int ret;
2604 struct fs_path *orphan;
2605
924794c9 2606 orphan = fs_path_alloc();
31db9f7c
AB		 2607 if (!orphan)
2608 return -ENOMEM;
2609
2610 ret = gen_unique_name(sctx, ino, gen, orphan);
2611 if (ret < 0)
2612 goto out;
2613
2614 ret = send_rename(sctx, path, orphan);
2615
2616out:
924794c9 2617 fs_path_free(orphan);
31db9f7c
AB		 2618 return ret;
2619}
2620
2621/*
2622 * Returns 1 if a directory can be removed at this point in time.
2623 * We check this by iterating all dir items and checking if the inode behind
2624 * the dir item was already processed.
2625 */
2626static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2627{
2628 int ret = 0;
2629 struct btrfs_root *root = sctx->parent_root;
2630 struct btrfs_path *path;
2631 struct btrfs_key key;
2632 struct btrfs_key found_key;
2633 struct btrfs_key loc;
2634 struct btrfs_dir_item *di;
2635
6d85ed05
AB		 2636 /*
2637 * Don't try to rmdir the top/root subvolume dir.
2638 */
2639 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2640 return 0;
2641
31db9f7c
AB		 2642 path = alloc_path_for_send();
2643 if (!path)
2644 return -ENOMEM;
2645
2646 key.objectid = dir;
2647 key.type = BTRFS_DIR_INDEX_KEY;
2648 key.offset = 0;
2649
2650 while (1) {
2651 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2652 if (ret < 0)
2653 goto out;
2654 if (!ret) {
2655 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2656 path->slots[0]);
2657 }
2658 if (ret || found_key.objectid != key.objectid ||
2659 found_key.type != key.type) {
2660 break;
2661 }
2662
2663 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2664 struct btrfs_dir_item);
2665 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2666
2667 if (loc.objectid > send_progress) {
2668 ret = 0;
2669 goto out;
2670 }
2671
2672 btrfs_release_path(path);
2673 key.offset = found_key.offset + 1;
2674 }
2675
2676 ret = 1;
2677
2678out:
2679 btrfs_free_path(path);
2680 return ret;
2681}
2682
31db9f7c
AB		 2683/*
2684 * This does all the move/link/unlink/rmdir magic.
2685 */
2686static int process_recorded_refs(struct send_ctx *sctx)
2687{
2688 int ret = 0;
2689 struct recorded_ref *cur;
1f4692da 2690 struct recorded_ref *cur2;
ba5e8f2e 2691 struct list_head check_dirs;
31db9f7c 2692 struct fs_path *valid_path = NULL;
b24baf69 2693 u64 ow_inode = 0;
31db9f7c
AB		 2694 u64 ow_gen;
2695 int did_overwrite = 0;
2696 int is_orphan = 0;
2697
2698verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2699
6d85ed05
AB		 2700 /*
2701 * This should never happen as the root dir always has the same ref
2702 * which is always '..'
2703 */
2704 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
ba5e8f2e 2705 INIT_LIST_HEAD(&check_dirs);
6d85ed05 2706
924794c9 2707 valid_path = fs_path_alloc();
31db9f7c
AB		 2708 if (!valid_path) {
2709 ret = -ENOMEM;
2710 goto out;
2711 }
2712
31db9f7c
AB		 2713 /*
2714 * First, check if the first ref of the current inode was overwritten
2715 * before. If yes, we know that the current inode was already orphanized
2716 * and thus use the orphan name. If not, we can use get_cur_path to
2717 * get the path of the first ref as it would like while receiving at
2718 * this point in time.
2719 * New inodes are always orphan at the beginning, so force to use the
2720 * orphan name in this case.
2721 * The first ref is stored in valid_path and will be updated if it
2722 * gets moved around.
2723 */
2724 if (!sctx->cur_inode_new) {
2725 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2726 sctx->cur_inode_gen);
2727 if (ret < 0)
2728 goto out;
2729 if (ret)
2730 did_overwrite = 1;
2731 }
2732 if (sctx->cur_inode_new || did_overwrite) {
2733 ret = gen_unique_name(sctx, sctx->cur_ino,
2734 sctx->cur_inode_gen, valid_path);
2735 if (ret < 0)
2736 goto out;
2737 is_orphan = 1;
2738 } else {
2739 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2740 valid_path);
2741 if (ret < 0)
2742 goto out;
2743 }
2744
2745 list_for_each_entry(cur, &sctx->new_refs, list) {
1f4692da
AB		 2746 /*
2747 * We may have refs where the parent directory does not exist
2748 * yet. This happens if the parent directories inum is higher
2749 * the the current inum. To handle this case, we create the
2750 * parent directory out of order. But we need to check if this
2751 * did already happen before due to other refs in the same dir.
2752 */
2753 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
2754 if (ret < 0)
2755 goto out;
2756 if (ret == inode_state_will_create) {
2757 ret = 0;
2758 /*
2759 * First check if any of the current inodes refs did
2760 * already create the dir.
2761 */
2762 list_for_each_entry(cur2, &sctx->new_refs, list) {
2763 if (cur == cur2)
2764 break;
2765 if (cur2->dir == cur->dir) {
2766 ret = 1;
2767 break;
2768 }
2769 }
2770
2771 /*
2772 * If that did not happen, check if a previous inode
2773 * did already create the dir.
2774 */
2775 if (!ret)
2776 ret = did_create_dir(sctx, cur->dir);
2777 if (ret < 0)
2778 goto out;
2779 if (!ret) {
2780 ret = send_create_inode(sctx, cur->dir);
2781 if (ret < 0)
2782 goto out;
2783 }
2784 }
2785
31db9f7c
AB		 2786 /*
2787 * Check if this new ref would overwrite the first ref of
2788 * another unprocessed inode. If yes, orphanize the
2789 * overwritten inode. If we find an overwritten ref that is
2790 * not the first ref, simply unlink it.
2791 */
2792 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2793 cur->name, cur->name_len,
2794 &ow_inode, &ow_gen);
2795 if (ret < 0)
2796 goto out;
2797 if (ret) {
924794c9
TI		 2798 ret = is_first_ref(sctx->parent_root,
2799 ow_inode, cur->dir, cur->name,
2800 cur->name_len);
31db9f7c
AB		 2801 if (ret < 0)
2802 goto out;
2803 if (ret) {
2804 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2805 cur->full_path);
2806 if (ret < 0)
2807 goto out;
2808 } else {
2809 ret = send_unlink(sctx, cur->full_path);
2810 if (ret < 0)
2811 goto out;
2812 }
2813 }
2814
2815 /*
2816 * link/move the ref to the new place. If we have an orphan
2817 * inode, move it and update valid_path. If not, link or move
2818 * it depending on the inode mode.
2819 */
1f4692da 2820 if (is_orphan) {
31db9f7c
AB		 2821 ret = send_rename(sctx, valid_path, cur->full_path);
2822 if (ret < 0)
2823 goto out;
2824 is_orphan = 0;
2825 ret = fs_path_copy(valid_path, cur->full_path);
2826 if (ret < 0)
2827 goto out;
2828 } else {
2829 if (S_ISDIR(sctx->cur_inode_mode)) {
2830 /*
2831 * Dirs can't be linked, so move it. For moved
2832 * dirs, we always have one new and one deleted
2833 * ref. The deleted ref is ignored later.
2834 */
2835 ret = send_rename(sctx, valid_path,
2836 cur->full_path);
2837 if (ret < 0)
2838 goto out;
2839 ret = fs_path_copy(valid_path, cur->full_path);
2840 if (ret < 0)
2841 goto out;
2842 } else {
2843 ret = send_link(sctx, cur->full_path,
2844 valid_path);
2845 if (ret < 0)
2846 goto out;
2847 }
2848 }
ba5e8f2e 2849 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 2850 if (ret < 0)
2851 goto out;
2852 }
2853
2854 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2855 /*
2856 * Check if we can already rmdir the directory. If not,
2857 * orphanize it. For every dir item inside that gets deleted
2858 * later, we do this check again and rmdir it then if possible.
2859 * See the use of check_dirs for more details.
2860 */
2861 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2862 if (ret < 0)
2863 goto out;
2864 if (ret) {
2865 ret = send_rmdir(sctx, valid_path);
2866 if (ret < 0)
2867 goto out;
2868 } else if (!is_orphan) {
2869 ret = orphanize_inode(sctx, sctx->cur_ino,
2870 sctx->cur_inode_gen, valid_path);
2871 if (ret < 0)
2872 goto out;
2873 is_orphan = 1;
2874 }
2875
2876 list_for_each_entry(cur, &sctx->deleted_refs, list) {
ba5e8f2e 2877 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 2878 if (ret < 0)
2879 goto out;
2880 }
ccf1626b
AB		 2881 } else if (S_ISDIR(sctx->cur_inode_mode) &&
2882 !list_empty(&sctx->deleted_refs)) {
2883 /*
2884 * We have a moved dir. Add the old parent to check_dirs
2885 */
2886 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
2887 list);
ba5e8f2e 2888 ret = dup_ref(cur, &check_dirs);
ccf1626b
AB		 2889 if (ret < 0)
2890 goto out;
31db9f7c
AB		 2891 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2892 /*
2893 * We have a non dir inode. Go through all deleted refs and
2894 * unlink them if they were not already overwritten by other
2895 * inodes.
2896 */
2897 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2898 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2899 sctx->cur_ino, sctx->cur_inode_gen,
2900 cur->name, cur->name_len);
2901 if (ret < 0)
2902 goto out;
2903 if (!ret) {
1f4692da
AB		 2904 ret = send_unlink(sctx, cur->full_path);
2905 if (ret < 0)
2906 goto out;
31db9f7c 2907 }
ba5e8f2e 2908 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 2909 if (ret < 0)
2910 goto out;
2911 }
31db9f7c
AB		 2912 /*
2913 * If the inode is still orphan, unlink the orphan. This may
2914 * happen when a previous inode did overwrite the first ref
2915 * of this inode and no new refs were added for the current
766702ef
AB		 2916 * inode. Unlinking does not mean that the inode is deleted in
2917 * all cases. There may still be links to this inode in other
2918 * places.
31db9f7c 2919 */
1f4692da 2920 if (is_orphan) {
31db9f7c
AB		 2921 ret = send_unlink(sctx, valid_path);
2922 if (ret < 0)
2923 goto out;
2924 }
2925 }
2926
2927 /*
2928 * We did collect all parent dirs where cur_inode was once located. We
2929 * now go through all these dirs and check if they are pending for
2930 * deletion and if it's finally possible to perform the rmdir now.
2931 * We also update the inode stats of the parent dirs here.
2932 */
ba5e8f2e 2933 list_for_each_entry(cur, &check_dirs, list) {
766702ef
AB		 2934 /*
2935 * In case we had refs into dirs that were not processed yet,
2936 * we don't need to do the utime and rmdir logic for these dirs.
2937 * The dir will be processed later.
2938 */
ba5e8f2e 2939 if (cur->dir > sctx->cur_ino)
31db9f7c
AB		 2940 continue;
2941
ba5e8f2e 2942 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 2943 if (ret < 0)
2944 goto out;
2945
2946 if (ret == inode_state_did_create ||
2947 ret == inode_state_no_change) {
2948 /* TODO delayed utimes */
ba5e8f2e 2949 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 2950 if (ret < 0)
2951 goto out;
2952 } else if (ret == inode_state_did_delete) {
ba5e8f2e 2953 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
31db9f7c
AB		 2954 if (ret < 0)
2955 goto out;
2956 if (ret) {
ba5e8f2e
JB		 2957 ret = get_cur_path(sctx, cur->dir,
2958 cur->dir_gen, valid_path);
31db9f7c
AB		 2959 if (ret < 0)
2960 goto out;
2961 ret = send_rmdir(sctx, valid_path);
2962 if (ret < 0)
2963 goto out;
2964 }
2965 }
2966 }
2967
31db9f7c
AB		 2968 ret = 0;
2969
2970out:
ba5e8f2e 2971 __free_recorded_refs(&check_dirs);
31db9f7c 2972 free_recorded_refs(sctx);
924794c9 2973 fs_path_free(valid_path);
31db9f7c
AB		 2974 return ret;
2975}
2976
2977static int __record_new_ref(int num, u64 dir, int index,
2978 struct fs_path *name,
2979 void *ctx)
2980{
2981 int ret = 0;
2982 struct send_ctx *sctx = ctx;
2983 struct fs_path *p;
2984 u64 gen;
2985
924794c9 2986 p = fs_path_alloc();
31db9f7c
AB		 2987 if (!p)
2988 return -ENOMEM;
2989
2990 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 2991 NULL, NULL);
31db9f7c
AB		 2992 if (ret < 0)
2993 goto out;
2994
31db9f7c
AB		 2995 ret = get_cur_path(sctx, dir, gen, p);
2996 if (ret < 0)
2997 goto out;
2998 ret = fs_path_add_path(p, name);
2999 if (ret < 0)
3000 goto out;
3001
3002 ret = record_ref(&sctx->new_refs, dir, gen, p);
3003
3004out:
3005 if (ret)
924794c9 3006 fs_path_free(p);
31db9f7c
AB		 3007 return ret;
3008}
3009
3010static int __record_deleted_ref(int num, u64 dir, int index,
3011 struct fs_path *name,
3012 void *ctx)
3013{
3014 int ret = 0;
3015 struct send_ctx *sctx = ctx;
3016 struct fs_path *p;
3017 u64 gen;
3018
924794c9 3019 p = fs_path_alloc();
31db9f7c
AB		 3020 if (!p)
3021 return -ENOMEM;
3022
3023 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 3024 NULL, NULL);
31db9f7c
AB		 3025 if (ret < 0)
3026 goto out;
3027
3028 ret = get_cur_path(sctx, dir, gen, p);
3029 if (ret < 0)
3030 goto out;
3031 ret = fs_path_add_path(p, name);
3032 if (ret < 0)
3033 goto out;
3034
3035 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3036
3037out:
3038 if (ret)
924794c9 3039 fs_path_free(p);
31db9f7c
AB		 3040 return ret;
3041}
3042
3043static int record_new_ref(struct send_ctx *sctx)
3044{
3045 int ret;
3046
924794c9
TI		 3047 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3048 sctx->cmp_key, 0, __record_new_ref, sctx);
31db9f7c
AB		 3049 if (ret < 0)
3050 goto out;
3051 ret = 0;
3052
3053out:
3054 return ret;
3055}
3056
3057static int record_deleted_ref(struct send_ctx *sctx)
3058{
3059 int ret;
3060
924794c9
TI		 3061 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3062 sctx->cmp_key, 0, __record_deleted_ref, sctx);
31db9f7c
AB		 3063 if (ret < 0)
3064 goto out;
3065 ret = 0;
3066
3067out:
3068 return ret;
3069}
3070
3071struct find_ref_ctx {
3072 u64 dir;
ba5e8f2e
JB		 3073 u64 dir_gen;
3074 struct btrfs_root *root;
31db9f7c
AB		 3075 struct fs_path *name;
3076 int found_idx;
3077};
3078
3079static int __find_iref(int num, u64 dir, int index,
3080 struct fs_path *name,
3081 void *ctx_)
3082{
3083 struct find_ref_ctx *ctx = ctx_;
ba5e8f2e
JB		 3084 u64 dir_gen;
3085 int ret;
31db9f7c
AB		 3086
3087 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3088 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
ba5e8f2e
JB		 3089 /*
3090 * To avoid doing extra lookups we'll only do this if everything
3091 * else matches.
3092 */
3093 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3094 NULL, NULL, NULL);
3095 if (ret)
3096 return ret;
3097 if (dir_gen != ctx->dir_gen)
3098 return 0;
31db9f7c
AB		 3099 ctx->found_idx = num;
3100 return 1;
3101 }
3102 return 0;
3103}
3104
924794c9 3105static int find_iref(struct btrfs_root *root,
31db9f7c
AB		 3106 struct btrfs_path *path,
3107 struct btrfs_key *key,
ba5e8f2e 3108 u64 dir, u64 dir_gen, struct fs_path *name)
31db9f7c
AB		 3109{
3110 int ret;
3111 struct find_ref_ctx ctx;
3112
3113 ctx.dir = dir;
3114 ctx.name = name;
ba5e8f2e 3115 ctx.dir_gen = dir_gen;
31db9f7c 3116 ctx.found_idx = -1;
ba5e8f2e 3117 ctx.root = root;
31db9f7c 3118
924794c9 3119 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
31db9f7c
AB		 3120 if (ret < 0)
3121 return ret;
3122
3123 if (ctx.found_idx == -1)
3124 return -ENOENT;
3125
3126 return ctx.found_idx;
3127}
3128
3129static int __record_changed_new_ref(int num, u64 dir, int index,
3130 struct fs_path *name,
3131 void *ctx)
3132{
ba5e8f2e 3133 u64 dir_gen;
31db9f7c
AB		 3134 int ret;
3135 struct send_ctx *sctx = ctx;
3136
ba5e8f2e
JB		 3137 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3138 NULL, NULL, NULL);
3139 if (ret)
3140 return ret;
3141
924794c9 3142 ret = find_iref(sctx->parent_root, sctx->right_path,
ba5e8f2e 3143 sctx->cmp_key, dir, dir_gen, name);
31db9f7c
AB		 3144 if (ret == -ENOENT)
3145 ret = __record_new_ref(num, dir, index, name, sctx);
3146 else if (ret > 0)
3147 ret = 0;
3148
3149 return ret;
3150}
3151
3152static int __record_changed_deleted_ref(int num, u64 dir, int index,
3153 struct fs_path *name,
3154 void *ctx)
3155{
ba5e8f2e 3156 u64 dir_gen;
31db9f7c
AB		 3157 int ret;
3158 struct send_ctx *sctx = ctx;
3159
ba5e8f2e
JB		 3160 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3161 NULL, NULL, NULL);
3162 if (ret)
3163 return ret;
3164
924794c9 3165 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
ba5e8f2e 3166 dir, dir_gen, name);
31db9f7c
AB		 3167 if (ret == -ENOENT)
3168 ret = __record_deleted_ref(num, dir, index, name, sctx);
3169 else if (ret > 0)
3170 ret = 0;
3171
3172 return ret;
3173}
3174
3175static int record_changed_ref(struct send_ctx *sctx)
3176{
3177 int ret = 0;
3178
924794c9 3179 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
31db9f7c
AB		 3180 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3181 if (ret < 0)
3182 goto out;
924794c9 3183 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 3184 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3185 if (ret < 0)
3186 goto out;
3187 ret = 0;
3188
3189out:
3190 return ret;
3191}
3192
3193/*
3194 * Record and process all refs at once. Needed when an inode changes the
3195 * generation number, which means that it was deleted and recreated.
3196 */
3197static int process_all_refs(struct send_ctx *sctx,
3198 enum btrfs_compare_tree_result cmd)
3199{
3200 int ret;
3201 struct btrfs_root *root;
3202 struct btrfs_path *path;
3203 struct btrfs_key key;
3204 struct btrfs_key found_key;
3205 struct extent_buffer *eb;
3206 int slot;
3207 iterate_inode_ref_t cb;
3208
3209 path = alloc_path_for_send();
3210 if (!path)
3211 return -ENOMEM;
3212
3213 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3214 root = sctx->send_root;
3215 cb = __record_new_ref;
3216 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3217 root = sctx->parent_root;
3218 cb = __record_deleted_ref;
3219 } else {
3220 BUG();
3221 }
3222
3223 key.objectid = sctx->cmp_key->objectid;
3224 key.type = BTRFS_INODE_REF_KEY;
3225 key.offset = 0;
3226 while (1) {
3227 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
e938c8ad 3228 if (ret < 0)
31db9f7c 3229 goto out;
e938c8ad 3230 if (ret)
31db9f7c 3231 break;
31db9f7c
AB		 3232
3233 eb = path->nodes[0];
3234 slot = path->slots[0];
3235 btrfs_item_key_to_cpu(eb, &found_key, slot);
3236
3237 if (found_key.objectid != key.objectid ||
96b5bd77
JS		 3238 (found_key.type != BTRFS_INODE_REF_KEY &&
3239 found_key.type != BTRFS_INODE_EXTREF_KEY))
31db9f7c 3240 break;
31db9f7c 3241
924794c9 3242 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
31db9f7c
AB		 3243 btrfs_release_path(path);
3244 if (ret < 0)
3245 goto out;
3246
3247 key.offset = found_key.offset + 1;
3248 }
e938c8ad 3249 btrfs_release_path(path);
31db9f7c
AB		 3250
3251 ret = process_recorded_refs(sctx);
3252
3253out:
3254 btrfs_free_path(path);
3255 return ret;
3256}
3257
3258static int send_set_xattr(struct send_ctx *sctx,
3259 struct fs_path *path,
3260 const char *name, int name_len,
3261 const char *data, int data_len)
3262{
3263 int ret = 0;
3264
3265 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3266 if (ret < 0)
3267 goto out;
3268
3269 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3270 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3271 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3272
3273 ret = send_cmd(sctx);
3274
3275tlv_put_failure:
3276out:
3277 return ret;
3278}
3279
3280static int send_remove_xattr(struct send_ctx *sctx,
3281 struct fs_path *path,
3282 const char *name, int name_len)
3283{
3284 int ret = 0;
3285
3286 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3287 if (ret < 0)
3288 goto out;
3289
3290 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3291 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3292
3293 ret = send_cmd(sctx);
3294
3295tlv_put_failure:
3296out:
3297 return ret;
3298}
3299
3300static int __process_new_xattr(int num, struct btrfs_key *di_key,
3301 const char *name, int name_len,
3302 const char *data, int data_len,
3303 u8 type, void *ctx)
3304{
3305 int ret;
3306 struct send_ctx *sctx = ctx;
3307 struct fs_path *p;
3308 posix_acl_xattr_header dummy_acl;
3309
924794c9 3310 p = fs_path_alloc();
31db9f7c
AB		 3311 if (!p)
3312 return -ENOMEM;
3313
3314 /*
3315 * This hack is needed because empty acl's are stored as zero byte
3316 * data in xattrs. Problem with that is, that receiving these zero byte
3317 * acl's will fail later. To fix this, we send a dummy acl list that
3318 * only contains the version number and no entries.
3319 */
3320 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3321 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3322 if (data_len == 0) {
3323 dummy_acl.a_version =
3324 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3325 data = (char *)&dummy_acl;
3326 data_len = sizeof(dummy_acl);
3327 }
3328 }
3329
3330 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3331 if (ret < 0)
3332 goto out;
3333
3334 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3335
3336out:
924794c9 3337 fs_path_free(p);
31db9f7c
AB		 3338 return ret;
3339}
3340
3341static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3342 const char *name, int name_len,
3343 const char *data, int data_len,
3344 u8 type, void *ctx)
3345{
3346 int ret;
3347 struct send_ctx *sctx = ctx;
3348 struct fs_path *p;
3349
924794c9 3350 p = fs_path_alloc();
31db9f7c
AB		 3351 if (!p)
3352 return -ENOMEM;
3353
3354 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3355 if (ret < 0)
3356 goto out;
3357
3358 ret = send_remove_xattr(sctx, p, name, name_len);
3359
3360out:
924794c9 3361 fs_path_free(p);
31db9f7c
AB		 3362 return ret;
3363}
3364
3365static int process_new_xattr(struct send_ctx *sctx)
3366{
3367 int ret = 0;
3368
924794c9
TI		 3369 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3370 sctx->cmp_key, __process_new_xattr, sctx);
31db9f7c
AB		 3371
3372 return ret;
3373}
3374
3375static int process_deleted_xattr(struct send_ctx *sctx)
3376{
3377 int ret;
3378
924794c9
TI		 3379 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3380 sctx->cmp_key, __process_deleted_xattr, sctx);
31db9f7c
AB		 3381
3382 return ret;
3383}
3384
3385struct find_xattr_ctx {
3386 const char *name;
3387 int name_len;
3388 int found_idx;
3389 char *found_data;
3390 int found_data_len;
3391};
3392
3393static int __find_xattr(int num, struct btrfs_key *di_key,
3394 const char *name, int name_len,
3395 const char *data, int data_len,
3396 u8 type, void *vctx)
3397{
3398 struct find_xattr_ctx *ctx = vctx;
3399
3400 if (name_len == ctx->name_len &&
3401 strncmp(name, ctx->name, name_len) == 0) {
3402 ctx->found_idx = num;
3403 ctx->found_data_len = data_len;
a5959bc0 3404 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
31db9f7c
AB		 3405 if (!ctx->found_data)
3406 return -ENOMEM;
31db9f7c
AB		 3407 return 1;
3408 }
3409 return 0;
3410}
3411
924794c9 3412static int find_xattr(struct btrfs_root *root,
31db9f7c
AB		 3413 struct btrfs_path *path,
3414 struct btrfs_key *key,
3415 const char *name, int name_len,
3416 char **data, int *data_len)
3417{
3418 int ret;
3419 struct find_xattr_ctx ctx;
3420
3421 ctx.name = name;
3422 ctx.name_len = name_len;
3423 ctx.found_idx = -1;
3424 ctx.found_data = NULL;
3425 ctx.found_data_len = 0;
3426
924794c9 3427 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
31db9f7c
AB		 3428 if (ret < 0)
3429 return ret;
3430
3431 if (ctx.found_idx == -1)
3432 return -ENOENT;
3433 if (data) {
3434 *data = ctx.found_data;
3435 *data_len = ctx.found_data_len;
3436 } else {
3437 kfree(ctx.found_data);
3438 }
3439 return ctx.found_idx;
3440}
3441
3442
3443static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3444 const char *name, int name_len,
3445 const char *data, int data_len,
3446 u8 type, void *ctx)
3447{
3448 int ret;
3449 struct send_ctx *sctx = ctx;
3450 char *found_data = NULL;
3451 int found_data_len = 0;
31db9f7c 3452
924794c9
TI		 3453 ret = find_xattr(sctx->parent_root, sctx->right_path,
3454 sctx->cmp_key, name, name_len, &found_data,
3455 &found_data_len);
31db9f7c
AB		 3456 if (ret == -ENOENT) {
3457 ret = __process_new_xattr(num, di_key, name, name_len, data,
3458 data_len, type, ctx);
3459 } else if (ret >= 0) {
3460 if (data_len != found_data_len ||
3461 memcmp(data, found_data, data_len)) {
3462 ret = __process_new_xattr(num, di_key, name, name_len,
3463 data, data_len, type, ctx);
3464 } else {
3465 ret = 0;
3466 }
3467 }
3468
3469 kfree(found_data);
31db9f7c
AB		 3470 return ret;
3471}
3472
3473static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3474 const char *name, int name_len,
3475 const char *data, int data_len,
3476 u8 type, void *ctx)
3477{
3478 int ret;
3479 struct send_ctx *sctx = ctx;
3480
924794c9
TI		 3481 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3482 name, name_len, NULL, NULL);
31db9f7c
AB		 3483 if (ret == -ENOENT)
3484 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3485 data_len, type, ctx);
3486 else if (ret >= 0)
3487 ret = 0;
3488
3489 return ret;
3490}
3491
3492static int process_changed_xattr(struct send_ctx *sctx)
3493{
3494 int ret = 0;
3495
924794c9 3496 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
31db9f7c
AB		 3497 sctx->cmp_key, __process_changed_new_xattr, sctx);
3498 if (ret < 0)
3499 goto out;
924794c9 3500 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 3501 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3502
3503out:
3504 return ret;
3505}
3506
3507static int process_all_new_xattrs(struct send_ctx *sctx)
3508{
3509 int ret;
3510 struct btrfs_root *root;
3511 struct btrfs_path *path;
3512 struct btrfs_key key;
3513 struct btrfs_key found_key;
3514 struct extent_buffer *eb;
3515 int slot;
3516
3517 path = alloc_path_for_send();
3518 if (!path)
3519 return -ENOMEM;
3520
3521 root = sctx->send_root;
3522
3523 key.objectid = sctx->cmp_key->objectid;
3524 key.type = BTRFS_XATTR_ITEM_KEY;
3525 key.offset = 0;
3526 while (1) {
3527 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3528 if (ret < 0)
3529 goto out;
3530 if (ret) {
3531 ret = 0;
3532 goto out;
3533 }
3534
3535 eb = path->nodes[0];
3536 slot = path->slots[0];
3537 btrfs_item_key_to_cpu(eb, &found_key, slot);
3538
3539 if (found_key.objectid != key.objectid ||
3540 found_key.type != key.type) {
3541 ret = 0;
3542 goto out;
3543 }
3544
924794c9
TI		 3545 ret = iterate_dir_item(root, path, &found_key,
3546 __process_new_xattr, sctx);
31db9f7c
AB		 3547 if (ret < 0)
3548 goto out;
3549
3550 btrfs_release_path(path);
3551 key.offset = found_key.offset + 1;
3552 }
3553
3554out:
3555 btrfs_free_path(path);
3556 return ret;
3557}
3558
ed259095
JB		 3559static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3560{
3561 struct btrfs_root *root = sctx->send_root;
3562 struct btrfs_fs_info *fs_info = root->fs_info;
3563 struct inode *inode;
3564 struct page *page;
3565 char *addr;
3566 struct btrfs_key key;
3567 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3568 pgoff_t last_index;
3569 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3570 ssize_t ret = 0;
3571
3572 key.objectid = sctx->cur_ino;
3573 key.type = BTRFS_INODE_ITEM_KEY;
3574 key.offset = 0;
3575
3576 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3577 if (IS_ERR(inode))
3578 return PTR_ERR(inode);
3579
3580 if (offset + len > i_size_read(inode)) {
3581 if (offset > i_size_read(inode))
3582 len = 0;
3583 else
3584 len = offset - i_size_read(inode);
3585 }
3586 if (len == 0)
3587 goto out;
3588
3589 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
3590 while (index <= last_index) {
3591 unsigned cur_len = min_t(unsigned, len,
3592 PAGE_CACHE_SIZE - pg_offset);
3593 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3594 if (!page) {
3595 ret = -ENOMEM;
3596 break;
3597 }
3598
3599 if (!PageUptodate(page)) {
3600 btrfs_readpage(NULL, page);
3601 lock_page(page);
3602 if (!PageUptodate(page)) {
3603 unlock_page(page);
3604 page_cache_release(page);
3605 ret = -EIO;
3606 break;
3607 }
3608 }
3609
3610 addr = kmap(page);
3611 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
3612 kunmap(page);
3613 unlock_page(page);
3614 page_cache_release(page);
3615 index++;
3616 pg_offset = 0;
3617 len -= cur_len;
3618 ret += cur_len;
3619 }
3620out:
3621 iput(inode);
3622 return ret;
3623}
3624
31db9f7c
AB		 3625/*
3626 * Read some bytes from the current inode/file and send a write command to
3627 * user space.
3628 */
3629static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3630{
3631 int ret = 0;
3632 struct fs_path *p;
ed259095 3633 ssize_t num_read = 0;
31db9f7c 3634
924794c9 3635 p = fs_path_alloc();
31db9f7c
AB		 3636 if (!p)
3637 return -ENOMEM;
3638
31db9f7c
AB		 3639verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3640
ed259095
JB		 3641 num_read = fill_read_buf(sctx, offset, len);
3642 if (num_read <= 0) {
3643 if (num_read < 0)
3644 ret = num_read;
31db9f7c 3645 goto out;
ed259095 3646 }
31db9f7c
AB		 3647
3648 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3649 if (ret < 0)
3650 goto out;
3651
3652 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3653 if (ret < 0)
3654 goto out;
3655
3656 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3657 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
e938c8ad 3658 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
31db9f7c
AB		 3659
3660 ret = send_cmd(sctx);
3661
3662tlv_put_failure:
3663out:
924794c9 3664 fs_path_free(p);
31db9f7c
AB		 3665 if (ret < 0)
3666 return ret;
e938c8ad 3667 return num_read;
31db9f7c
AB		 3668}
3669
3670/*
3671 * Send a clone command to user space.
3672 */
3673static int send_clone(struct send_ctx *sctx,
3674 u64 offset, u32 len,
3675 struct clone_root *clone_root)
3676{
3677 int ret = 0;
31db9f7c
AB		 3678 struct fs_path *p;
3679 u64 gen;
3680
3681verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
3682 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
3683 clone_root->root->objectid, clone_root->ino,
3684 clone_root->offset);
3685
924794c9 3686 p = fs_path_alloc();
31db9f7c
AB		 3687 if (!p)
3688 return -ENOMEM;
3689
3690 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
3691 if (ret < 0)
3692 goto out;
3693
3694 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3695 if (ret < 0)
3696 goto out;
3697
3698 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3699 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
3700 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3701
e938c8ad 3702 if (clone_root->root == sctx->send_root) {
31db9f7c 3703 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
85a7b33b 3704 &gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 3705 if (ret < 0)
3706 goto out;
3707 ret = get_cur_path(sctx, clone_root->ino, gen, p);
3708 } else {
924794c9 3709 ret = get_inode_path(clone_root->root, clone_root->ino, p);
31db9f7c
AB		 3710 }
3711 if (ret < 0)
3712 goto out;
3713
3714 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
e938c8ad 3715 clone_root->root->root_item.uuid);
31db9f7c 3716 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
e938c8ad 3717 clone_root->root->root_item.ctransid);
31db9f7c
AB		 3718 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3719 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3720 clone_root->offset);
3721
3722 ret = send_cmd(sctx);
3723
3724tlv_put_failure:
3725out:
924794c9 3726 fs_path_free(p);
31db9f7c
AB		 3727 return ret;
3728}
3729
cb95e7bf
MF		 3730/*
3731 * Send an update extent command to user space.
3732 */
3733static int send_update_extent(struct send_ctx *sctx,
3734 u64 offset, u32 len)
3735{
3736 int ret = 0;
3737 struct fs_path *p;
3738
924794c9 3739 p = fs_path_alloc();
cb95e7bf
MF		 3740 if (!p)
3741 return -ENOMEM;
3742
3743 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
3744 if (ret < 0)
3745 goto out;
3746
3747 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3748 if (ret < 0)
3749 goto out;
3750
3751 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3752 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3753 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
3754
3755 ret = send_cmd(sctx);
3756
3757tlv_put_failure:
3758out:
924794c9 3759 fs_path_free(p);
cb95e7bf
MF		 3760 return ret;
3761}
3762
16e7549f
JB		 3763static int send_hole(struct send_ctx *sctx, u64 end)
3764{
3765 struct fs_path *p = NULL;
3766 u64 offset = sctx->cur_inode_last_extent;
3767 u64 len;
3768 int ret = 0;
3769
3770 p = fs_path_alloc();
3771 if (!p)
3772 return -ENOMEM;
3773 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
3774 while (offset < end) {
3775 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
3776
3777 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3778 if (ret < 0)
3779 break;
3780 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3781 if (ret < 0)
3782 break;
3783 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3784 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3785 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
3786 ret = send_cmd(sctx);
3787 if (ret < 0)
3788 break;
3789 offset += len;
3790 }
3791tlv_put_failure:
3792 fs_path_free(p);
3793 return ret;
3794}
3795
31db9f7c
AB		 3796static int send_write_or_clone(struct send_ctx *sctx,
3797 struct btrfs_path *path,
3798 struct btrfs_key *key,
3799 struct clone_root *clone_root)
3800{
3801 int ret = 0;
3802 struct btrfs_file_extent_item *ei;
3803 u64 offset = key->offset;
3804 u64 pos = 0;
3805 u64 len;
3806 u32 l;
3807 u8 type;
3808
3809 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3810 struct btrfs_file_extent_item);
3811 type = btrfs_file_extent_type(path->nodes[0], ei);
74dd17fb 3812 if (type == BTRFS_FILE_EXTENT_INLINE) {
31db9f7c 3813 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
74dd17fb
CM		 3814 /*
3815 * it is possible the inline item won't cover the whole page,
3816 * but there may be items after this page. Make
3817 * sure to send the whole thing
3818 */
3819 len = PAGE_CACHE_ALIGN(len);
3820 } else {
31db9f7c 3821 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
74dd17fb 3822 }
31db9f7c
AB		 3823
3824 if (offset + len > sctx->cur_inode_size)
3825 len = sctx->cur_inode_size - offset;
3826 if (len == 0) {
3827 ret = 0;
3828 goto out;
3829 }
3830
cb95e7bf
MF		 3831 if (clone_root) {
3832 ret = send_clone(sctx, offset, len, clone_root);
3833 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
3834 ret = send_update_extent(sctx, offset, len);
3835 } else {
31db9f7c
AB		 3836 while (pos < len) {
3837 l = len - pos;
3838 if (l > BTRFS_SEND_READ_SIZE)
3839 l = BTRFS_SEND_READ_SIZE;
3840 ret = send_write(sctx, pos + offset, l);
3841 if (ret < 0)
3842 goto out;
3843 if (!ret)
3844 break;
3845 pos += ret;
3846 }
3847 ret = 0;
31db9f7c 3848 }
31db9f7c
AB		 3849out:
3850 return ret;
3851}
3852
3853static int is_extent_unchanged(struct send_ctx *sctx,
3854 struct btrfs_path *left_path,
3855 struct btrfs_key *ekey)
3856{
3857 int ret = 0;
3858 struct btrfs_key key;
3859 struct btrfs_path *path = NULL;
3860 struct extent_buffer *eb;
3861 int slot;
3862 struct btrfs_key found_key;
3863 struct btrfs_file_extent_item *ei;
3864 u64 left_disknr;
3865 u64 right_disknr;
3866 u64 left_offset;
3867 u64 right_offset;
3868 u64 left_offset_fixed;
3869 u64 left_len;
3870 u64 right_len;
74dd17fb
CM		 3871 u64 left_gen;
3872 u64 right_gen;
31db9f7c
AB		 3873 u8 left_type;
3874 u8 right_type;
3875
3876 path = alloc_path_for_send();
3877 if (!path)
3878 return -ENOMEM;
3879
3880 eb = left_path->nodes[0];
3881 slot = left_path->slots[0];
31db9f7c
AB		 3882 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3883 left_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 3884
3885 if (left_type != BTRFS_FILE_EXTENT_REG) {
3886 ret = 0;
3887 goto out;
3888 }
74dd17fb
CM		 3889 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3890 left_len = btrfs_file_extent_num_bytes(eb, ei);
3891 left_offset = btrfs_file_extent_offset(eb, ei);
3892 left_gen = btrfs_file_extent_generation(eb, ei);
31db9f7c
AB		 3893
3894 /*
3895 * Following comments will refer to these graphics. L is the left
3896 * extents which we are checking at the moment. 1-8 are the right
3897 * extents that we iterate.
3898 *
3899 * |-----L-----|
3900 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3901 *
3902 * |-----L-----|
3903 * |--1--|-2b-|...(same as above)
3904 *
3905 * Alternative situation. Happens on files where extents got split.
3906 * |-----L-----|
3907 * |-----------7-----------|-6-|
3908 *
3909 * Alternative situation. Happens on files which got larger.
3910 * |-----L-----|
3911 * |-8-|
3912 * Nothing follows after 8.
3913 */
3914
3915 key.objectid = ekey->objectid;
3916 key.type = BTRFS_EXTENT_DATA_KEY;
3917 key.offset = ekey->offset;
3918 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3919 if (ret < 0)
3920 goto out;
3921 if (ret) {
3922 ret = 0;
3923 goto out;
3924 }
3925
3926 /*
3927 * Handle special case where the right side has no extents at all.
3928 */
3929 eb = path->nodes[0];
3930 slot = path->slots[0];
3931 btrfs_item_key_to_cpu(eb, &found_key, slot);
3932 if (found_key.objectid != key.objectid ||
3933 found_key.type != key.type) {
57cfd462
JB		 3934 /* If we're a hole then just pretend nothing changed */
3935 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 3936 goto out;
3937 }
3938
3939 /*
3940 * We're now on 2a, 2b or 7.
3941 */
3942 key = found_key;
3943 while (key.offset < ekey->offset + left_len) {
3944 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3945 right_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 3946 if (right_type != BTRFS_FILE_EXTENT_REG) {
3947 ret = 0;
3948 goto out;
3949 }
3950
007d31f7
JB		 3951 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3952 right_len = btrfs_file_extent_num_bytes(eb, ei);
3953 right_offset = btrfs_file_extent_offset(eb, ei);
3954 right_gen = btrfs_file_extent_generation(eb, ei);
3955
31db9f7c
AB		 3956 /*
3957 * Are we at extent 8? If yes, we know the extent is changed.
3958 * This may only happen on the first iteration.
3959 */
d8347fa4 3960 if (found_key.offset + right_len <= ekey->offset) {
57cfd462
JB		 3961 /* If we're a hole just pretend nothing changed */
3962 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 3963 goto out;
3964 }
3965
3966 left_offset_fixed = left_offset;
3967 if (key.offset < ekey->offset) {
3968 /* Fix the right offset for 2a and 7. */
3969 right_offset += ekey->offset - key.offset;
3970 } else {
3971 /* Fix the left offset for all behind 2a and 2b */
3972 left_offset_fixed += key.offset - ekey->offset;
3973 }
3974
3975 /*
3976 * Check if we have the same extent.
3977 */
3954096d 3978 if (left_disknr != right_disknr ||
74dd17fb
CM		 3979 left_offset_fixed != right_offset ||
3980 left_gen != right_gen) {
31db9f7c
AB		 3981 ret = 0;
3982 goto out;
3983 }
3984
3985 /*
3986 * Go to the next extent.
3987 */
3988 ret = btrfs_next_item(sctx->parent_root, path);
3989 if (ret < 0)
3990 goto out;
3991 if (!ret) {
3992 eb = path->nodes[0];
3993 slot = path->slots[0];
3994 btrfs_item_key_to_cpu(eb, &found_key, slot);
3995 }
3996 if (ret || found_key.objectid != key.objectid ||
3997 found_key.type != key.type) {
3998 key.offset += right_len;
3999 break;
adaa4b8e
JS		 4000 }
4001 if (found_key.offset != key.offset + right_len) {
4002 ret = 0;
4003 goto out;
31db9f7c
AB		 4004 }
4005 key = found_key;
4006 }
4007
4008 /*
4009 * We're now behind the left extent (treat as unchanged) or at the end
4010 * of the right side (treat as changed).
4011 */
4012 if (key.offset >= ekey->offset + left_len)
4013 ret = 1;
4014 else
4015 ret = 0;
4016
4017
4018out:
4019 btrfs_free_path(path);
4020 return ret;
4021}
4022
16e7549f
JB		 4023static int get_last_extent(struct send_ctx *sctx, u64 offset)
4024{
4025 struct btrfs_path *path;
4026 struct btrfs_root *root = sctx->send_root;
4027 struct btrfs_file_extent_item *fi;
4028 struct btrfs_key key;
4029 u64 extent_end;
4030 u8 type;
4031 int ret;
4032
4033 path = alloc_path_for_send();
4034 if (!path)
4035 return -ENOMEM;
4036
4037 sctx->cur_inode_last_extent = 0;
4038
4039 key.objectid = sctx->cur_ino;
4040 key.type = BTRFS_EXTENT_DATA_KEY;
4041 key.offset = offset;
4042 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4043 if (ret < 0)
4044 goto out;
4045 ret = 0;
4046 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4047 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4048 goto out;
4049
4050 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4051 struct btrfs_file_extent_item);
4052 type = btrfs_file_extent_type(path->nodes[0], fi);
4053 if (type == BTRFS_FILE_EXTENT_INLINE) {
4054 u64 size = btrfs_file_extent_inline_len(path->nodes[0], fi);
4055 extent_end = ALIGN(key.offset + size,
4056 sctx->send_root->sectorsize);
4057 } else {
4058 extent_end = key.offset +
4059 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4060 }
4061 sctx->cur_inode_last_extent = extent_end;
4062out:
4063 btrfs_free_path(path);
4064 return ret;
4065}
4066
4067static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4068 struct btrfs_key *key)
4069{
4070 struct btrfs_file_extent_item *fi;
4071 u64 extent_end;
4072 u8 type;
4073 int ret = 0;
4074
4075 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4076 return 0;
4077
4078 if (sctx->cur_inode_last_extent == (u64)-1) {
4079 ret = get_last_extent(sctx, key->offset - 1);
4080 if (ret)
4081 return ret;
4082 }
4083
4084 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4085 struct btrfs_file_extent_item);
4086 type = btrfs_file_extent_type(path->nodes[0], fi);
4087 if (type == BTRFS_FILE_EXTENT_INLINE) {
4088 u64 size = btrfs_file_extent_inline_len(path->nodes[0], fi);
4089 extent_end = ALIGN(key->offset + size,
4090 sctx->send_root->sectorsize);
4091 } else {
4092 extent_end = key->offset +
4093 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4094 }
4095 if (sctx->cur_inode_last_extent < key->offset)
4096 ret = send_hole(sctx, key->offset);
4097 sctx->cur_inode_last_extent = extent_end;
4098 return ret;
4099}
4100
31db9f7c
AB		 4101static int process_extent(struct send_ctx *sctx,
4102 struct btrfs_path *path,
4103 struct btrfs_key *key)
4104{
31db9f7c 4105 struct clone_root *found_clone = NULL;
57cfd462 4106 int ret = 0;
31db9f7c
AB		 4107
4108 if (S_ISLNK(sctx->cur_inode_mode))
4109 return 0;
4110
4111 if (sctx->parent_root && !sctx->cur_inode_new) {
4112 ret = is_extent_unchanged(sctx, path, key);
4113 if (ret < 0)
4114 goto out;
4115 if (ret) {
4116 ret = 0;
16e7549f 4117 goto out_hole;
31db9f7c 4118 }
57cfd462
JB		 4119 } else {
4120 struct btrfs_file_extent_item *ei;
4121 u8 type;
4122
4123 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4124 struct btrfs_file_extent_item);
4125 type = btrfs_file_extent_type(path->nodes[0], ei);
4126 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4127 type == BTRFS_FILE_EXTENT_REG) {
4128 /*
4129 * The send spec does not have a prealloc command yet,
4130 * so just leave a hole for prealloc'ed extents until
4131 * we have enough commands queued up to justify rev'ing
4132 * the send spec.
4133 */
4134 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4135 ret = 0;
4136 goto out;
4137 }
4138
4139 /* Have a hole, just skip it. */
4140 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4141 ret = 0;
4142 goto out;
4143 }
4144 }
31db9f7c
AB		 4145 }
4146
4147 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4148 sctx->cur_inode_size, &found_clone);
4149 if (ret != -ENOENT && ret < 0)
4150 goto out;
4151
4152 ret = send_write_or_clone(sctx, path, key, found_clone);
16e7549f
JB		 4153 if (ret)
4154 goto out;
4155out_hole:
4156 ret = maybe_send_hole(sctx, path, key);
31db9f7c
AB		 4157out:
4158 return ret;
4159}
4160
4161static int process_all_extents(struct send_ctx *sctx)
4162{
4163 int ret;
4164 struct btrfs_root *root;
4165 struct btrfs_path *path;
4166 struct btrfs_key key;
4167 struct btrfs_key found_key;
4168 struct extent_buffer *eb;
4169 int slot;
4170
4171 root = sctx->send_root;
4172 path = alloc_path_for_send();
4173 if (!path)
4174 return -ENOMEM;
4175
4176 key.objectid = sctx->cmp_key->objectid;
4177 key.type = BTRFS_EXTENT_DATA_KEY;
4178 key.offset = 0;
4179 while (1) {
4180 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
4181 if (ret < 0)
4182 goto out;
4183 if (ret) {
4184 ret = 0;
4185 goto out;
4186 }
4187
4188 eb = path->nodes[0];
4189 slot = path->slots[0];
4190 btrfs_item_key_to_cpu(eb, &found_key, slot);
4191
4192 if (found_key.objectid != key.objectid ||
4193 found_key.type != key.type) {
4194 ret = 0;
4195 goto out;
4196 }
4197
4198 ret = process_extent(sctx, path, &found_key);
4199 if (ret < 0)
4200 goto out;
4201
4202 btrfs_release_path(path);
4203 key.offset = found_key.offset + 1;
4204 }
4205
4206out:
4207 btrfs_free_path(path);
4208 return ret;
4209}
4210
4211static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
4212{
4213 int ret = 0;
4214
4215 if (sctx->cur_ino == 0)
4216 goto out;
4217 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
96b5bd77 4218 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 4219 goto out;
4220 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4221 goto out;
4222
4223 ret = process_recorded_refs(sctx);
e479d9bb
AB		 4224 if (ret < 0)
4225 goto out;
4226
4227 /*
4228 * We have processed the refs and thus need to advance send_progress.
4229 * Now, calls to get_cur_xxx will take the updated refs of the current
4230 * inode into account.
4231 */
4232 sctx->send_progress = sctx->cur_ino + 1;
31db9f7c
AB		 4233
4234out:
4235 return ret;
4236}
4237
4238static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4239{
4240 int ret = 0;
4241 u64 left_mode;
4242 u64 left_uid;
4243 u64 left_gid;
4244 u64 right_mode;
4245 u64 right_uid;
4246 u64 right_gid;
4247 int need_chmod = 0;
4248 int need_chown = 0;
4249
4250 ret = process_recorded_refs_if_needed(sctx, at_end);
4251 if (ret < 0)
4252 goto out;
4253
4254 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4255 goto out;
4256 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4257 goto out;
4258
4259 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
85a7b33b 4260 &left_mode, &left_uid, &left_gid, NULL);
31db9f7c
AB		 4261 if (ret < 0)
4262 goto out;
4263
e2d044fe
AL		 4264 if (!sctx->parent_root || sctx->cur_inode_new) {
4265 need_chown = 1;
4266 if (!S_ISLNK(sctx->cur_inode_mode))
31db9f7c 4267 need_chmod = 1;
e2d044fe
AL		 4268 } else {
4269 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4270 NULL, NULL, &right_mode, &right_uid,
4271 &right_gid, NULL);
4272 if (ret < 0)
4273 goto out;
31db9f7c 4274
e2d044fe
AL		 4275 if (left_uid != right_uid || left_gid != right_gid)
4276 need_chown = 1;
4277 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4278 need_chmod = 1;
31db9f7c
AB		 4279 }
4280
4281 if (S_ISREG(sctx->cur_inode_mode)) {
16e7549f
JB		 4282 if (need_send_hole(sctx)) {
4283 if (sctx->cur_inode_last_extent == (u64)-1) {
4284 ret = get_last_extent(sctx, (u64)-1);
4285 if (ret)
4286 goto out;
4287 }
4288 if (sctx->cur_inode_last_extent <
4289 sctx->cur_inode_size) {
4290 ret = send_hole(sctx, sctx->cur_inode_size);
4291 if (ret)
4292 goto out;
4293 }
4294 }
31db9f7c
AB		 4295 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4296 sctx->cur_inode_size);
4297 if (ret < 0)
4298 goto out;
4299 }
4300
4301 if (need_chown) {
4302 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4303 left_uid, left_gid);
4304 if (ret < 0)
4305 goto out;
4306 }
4307 if (need_chmod) {
4308 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4309 left_mode);
4310 if (ret < 0)
4311 goto out;
4312 }
4313
4314 /*
4315 * Need to send that every time, no matter if it actually changed
4316 * between the two trees as we have done changes to the inode before.
4317 */
4318 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4319 if (ret < 0)
4320 goto out;
4321
4322out:
4323 return ret;
4324}
4325
4326static int changed_inode(struct send_ctx *sctx,
4327 enum btrfs_compare_tree_result result)
4328{
4329 int ret = 0;
4330 struct btrfs_key *key = sctx->cmp_key;
4331 struct btrfs_inode_item *left_ii = NULL;
4332 struct btrfs_inode_item *right_ii = NULL;
4333 u64 left_gen = 0;
4334 u64 right_gen = 0;
4335
31db9f7c
AB		 4336 sctx->cur_ino = key->objectid;
4337 sctx->cur_inode_new_gen = 0;
16e7549f 4338 sctx->cur_inode_last_extent = (u64)-1;
e479d9bb
AB		 4339
4340 /*
4341 * Set send_progress to current inode. This will tell all get_cur_xxx
4342 * functions that the current inode's refs are not updated yet. Later,
4343 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4344 */
31db9f7c
AB		 4345 sctx->send_progress = sctx->cur_ino;
4346
4347 if (result == BTRFS_COMPARE_TREE_NEW ||
4348 result == BTRFS_COMPARE_TREE_CHANGED) {
4349 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4350 sctx->left_path->slots[0],
4351 struct btrfs_inode_item);
4352 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4353 left_ii);
4354 } else {
4355 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4356 sctx->right_path->slots[0],
4357 struct btrfs_inode_item);
4358 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4359 right_ii);
4360 }
4361 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4362 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4363 sctx->right_path->slots[0],
4364 struct btrfs_inode_item);
4365
4366 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4367 right_ii);
6d85ed05
AB		 4368
4369 /*
4370 * The cur_ino = root dir case is special here. We can't treat
4371 * the inode as deleted+reused because it would generate a
4372 * stream that tries to delete/mkdir the root dir.
4373 */
4374 if (left_gen != right_gen &&
4375 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
31db9f7c
AB		 4376 sctx->cur_inode_new_gen = 1;
4377 }
4378
4379 if (result == BTRFS_COMPARE_TREE_NEW) {
4380 sctx->cur_inode_gen = left_gen;
4381 sctx->cur_inode_new = 1;
4382 sctx->cur_inode_deleted = 0;
4383 sctx->cur_inode_size = btrfs_inode_size(
4384 sctx->left_path->nodes[0], left_ii);
4385 sctx->cur_inode_mode = btrfs_inode_mode(
4386 sctx->left_path->nodes[0], left_ii);
4387 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
1f4692da 4388 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4389 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4390 sctx->cur_inode_gen = right_gen;
4391 sctx->cur_inode_new = 0;
4392 sctx->cur_inode_deleted = 1;
4393 sctx->cur_inode_size = btrfs_inode_size(
4394 sctx->right_path->nodes[0], right_ii);
4395 sctx->cur_inode_mode = btrfs_inode_mode(
4396 sctx->right_path->nodes[0], right_ii);
4397 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
766702ef
AB		 4398 /*
4399 * We need to do some special handling in case the inode was
4400 * reported as changed with a changed generation number. This
4401 * means that the original inode was deleted and new inode
4402 * reused the same inum. So we have to treat the old inode as
4403 * deleted and the new one as new.
4404 */
31db9f7c 4405 if (sctx->cur_inode_new_gen) {
766702ef
AB		 4406 /*
4407 * First, process the inode as if it was deleted.
4408 */
31db9f7c
AB		 4409 sctx->cur_inode_gen = right_gen;
4410 sctx->cur_inode_new = 0;
4411 sctx->cur_inode_deleted = 1;
4412 sctx->cur_inode_size = btrfs_inode_size(
4413 sctx->right_path->nodes[0], right_ii);
4414 sctx->cur_inode_mode = btrfs_inode_mode(
4415 sctx->right_path->nodes[0], right_ii);
4416 ret = process_all_refs(sctx,
4417 BTRFS_COMPARE_TREE_DELETED);
4418 if (ret < 0)
4419 goto out;
4420
766702ef
AB		 4421 /*
4422 * Now process the inode as if it was new.
4423 */
31db9f7c
AB		 4424 sctx->cur_inode_gen = left_gen;
4425 sctx->cur_inode_new = 1;
4426 sctx->cur_inode_deleted = 0;
4427 sctx->cur_inode_size = btrfs_inode_size(
4428 sctx->left_path->nodes[0], left_ii);
4429 sctx->cur_inode_mode = btrfs_inode_mode(
4430 sctx->left_path->nodes[0], left_ii);
1f4692da 4431 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4432 if (ret < 0)
4433 goto out;
4434
4435 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4436 if (ret < 0)
4437 goto out;
e479d9bb
AB		 4438 /*
4439 * Advance send_progress now as we did not get into
4440 * process_recorded_refs_if_needed in the new_gen case.
4441 */
4442 sctx->send_progress = sctx->cur_ino + 1;
766702ef
AB		 4443
4444 /*
4445 * Now process all extents and xattrs of the inode as if
4446 * they were all new.
4447 */
31db9f7c
AB		 4448 ret = process_all_extents(sctx);
4449 if (ret < 0)
4450 goto out;
4451 ret = process_all_new_xattrs(sctx);
4452 if (ret < 0)
4453 goto out;
4454 } else {
4455 sctx->cur_inode_gen = left_gen;
4456 sctx->cur_inode_new = 0;
4457 sctx->cur_inode_new_gen = 0;
4458 sctx->cur_inode_deleted = 0;
4459 sctx->cur_inode_size = btrfs_inode_size(
4460 sctx->left_path->nodes[0], left_ii);
4461 sctx->cur_inode_mode = btrfs_inode_mode(
4462 sctx->left_path->nodes[0], left_ii);
4463 }
4464 }
4465
4466out:
4467 return ret;
4468}
4469
766702ef
AB		 4470/*
4471 * We have to process new refs before deleted refs, but compare_trees gives us
4472 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4473 * first and later process them in process_recorded_refs.
4474 * For the cur_inode_new_gen case, we skip recording completely because
4475 * changed_inode did already initiate processing of refs. The reason for this is
4476 * that in this case, compare_tree actually compares the refs of 2 different
4477 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4478 * refs of the right tree as deleted and all refs of the left tree as new.
4479 */
31db9f7c
AB		 4480static int changed_ref(struct send_ctx *sctx,
4481 enum btrfs_compare_tree_result result)
4482{
4483 int ret = 0;
4484
4485 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4486
4487 if (!sctx->cur_inode_new_gen &&
4488 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4489 if (result == BTRFS_COMPARE_TREE_NEW)
4490 ret = record_new_ref(sctx);
4491 else if (result == BTRFS_COMPARE_TREE_DELETED)
4492 ret = record_deleted_ref(sctx);
4493 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4494 ret = record_changed_ref(sctx);
4495 }
4496
4497 return ret;
4498}
4499
766702ef
AB		 4500/*
4501 * Process new/deleted/changed xattrs. We skip processing in the
4502 * cur_inode_new_gen case because changed_inode did already initiate processing
4503 * of xattrs. The reason is the same as in changed_ref
4504 */
31db9f7c
AB		 4505static int changed_xattr(struct send_ctx *sctx,
4506 enum btrfs_compare_tree_result result)
4507{
4508 int ret = 0;
4509
4510 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4511
4512 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4513 if (result == BTRFS_COMPARE_TREE_NEW)
4514 ret = process_new_xattr(sctx);
4515 else if (result == BTRFS_COMPARE_TREE_DELETED)
4516 ret = process_deleted_xattr(sctx);
4517 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4518 ret = process_changed_xattr(sctx);
4519 }
4520
4521 return ret;
4522}
4523
766702ef
AB		 4524/*
4525 * Process new/deleted/changed extents. We skip processing in the
4526 * cur_inode_new_gen case because changed_inode did already initiate processing
4527 * of extents. The reason is the same as in changed_ref
4528 */
31db9f7c
AB		 4529static int changed_extent(struct send_ctx *sctx,
4530 enum btrfs_compare_tree_result result)
4531{
4532 int ret = 0;
4533
4534 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4535
4536 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4537 if (result != BTRFS_COMPARE_TREE_DELETED)
4538 ret = process_extent(sctx, sctx->left_path,
4539 sctx->cmp_key);
4540 }
4541
4542 return ret;
4543}
4544
ba5e8f2e
JB		 4545static int dir_changed(struct send_ctx *sctx, u64 dir)
4546{
4547 u64 orig_gen, new_gen;
4548 int ret;
4549
4550 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
4551 NULL, NULL);
4552 if (ret)
4553 return ret;
4554
4555 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
4556 NULL, NULL, NULL);
4557 if (ret)
4558 return ret;
4559
4560 return (orig_gen != new_gen) ? 1 : 0;
4561}
4562
4563static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
4564 struct btrfs_key *key)
4565{
4566 struct btrfs_inode_extref *extref;
4567 struct extent_buffer *leaf;
4568 u64 dirid = 0, last_dirid = 0;
4569 unsigned long ptr;
4570 u32 item_size;
4571 u32 cur_offset = 0;
4572 int ref_name_len;
4573 int ret = 0;
4574
4575 /* Easy case, just check this one dirid */
4576 if (key->type == BTRFS_INODE_REF_KEY) {
4577 dirid = key->offset;
4578
4579 ret = dir_changed(sctx, dirid);
4580 goto out;
4581 }
4582
4583 leaf = path->nodes[0];
4584 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4585 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4586 while (cur_offset < item_size) {
4587 extref = (struct btrfs_inode_extref *)(ptr +
4588 cur_offset);
4589 dirid = btrfs_inode_extref_parent(leaf, extref);
4590 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
4591 cur_offset += ref_name_len + sizeof(*extref);
4592 if (dirid == last_dirid)
4593 continue;
4594 ret = dir_changed(sctx, dirid);
4595 if (ret)
4596 break;
4597 last_dirid = dirid;
4598 }
4599out:
4600 return ret;
4601}
4602
766702ef
AB		 4603/*
4604 * Updates compare related fields in sctx and simply forwards to the actual
4605 * changed_xxx functions.
4606 */
31db9f7c
AB		 4607static int changed_cb(struct btrfs_root *left_root,
4608 struct btrfs_root *right_root,
4609 struct btrfs_path *left_path,
4610 struct btrfs_path *right_path,
4611 struct btrfs_key *key,
4612 enum btrfs_compare_tree_result result,
4613 void *ctx)
4614{
4615 int ret = 0;
4616 struct send_ctx *sctx = ctx;
4617
ba5e8f2e 4618 if (result == BTRFS_COMPARE_TREE_SAME) {
16e7549f
JB		 4619 if (key->type == BTRFS_INODE_REF_KEY ||
4620 key->type == BTRFS_INODE_EXTREF_KEY) {
4621 ret = compare_refs(sctx, left_path, key);
4622 if (!ret)
4623 return 0;
4624 if (ret < 0)
4625 return ret;
4626 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
4627 return maybe_send_hole(sctx, left_path, key);
4628 } else {
ba5e8f2e 4629 return 0;
16e7549f 4630 }
ba5e8f2e
JB		 4631 result = BTRFS_COMPARE_TREE_CHANGED;
4632 ret = 0;
4633 }
4634
31db9f7c
AB		 4635 sctx->left_path = left_path;
4636 sctx->right_path = right_path;
4637 sctx->cmp_key = key;
4638
4639 ret = finish_inode_if_needed(sctx, 0);
4640 if (ret < 0)
4641 goto out;
4642
2981e225
AB		 4643 /* Ignore non-FS objects */
4644 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
4645 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
4646 goto out;
4647
31db9f7c
AB		 4648 if (key->type == BTRFS_INODE_ITEM_KEY)
4649 ret = changed_inode(sctx, result);
96b5bd77
JS		 4650 else if (key->type == BTRFS_INODE_REF_KEY ||
4651 key->type == BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 4652 ret = changed_ref(sctx, result);
4653 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4654 ret = changed_xattr(sctx, result);
4655 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4656 ret = changed_extent(sctx, result);
4657
4658out:
4659 return ret;
4660}
4661
4662static int full_send_tree(struct send_ctx *sctx)
4663{
4664 int ret;
4665 struct btrfs_trans_handle *trans = NULL;
4666 struct btrfs_root *send_root = sctx->send_root;
4667 struct btrfs_key key;
4668 struct btrfs_key found_key;
4669 struct btrfs_path *path;
4670 struct extent_buffer *eb;
4671 int slot;
4672 u64 start_ctransid;
4673 u64 ctransid;
4674
4675 path = alloc_path_for_send();
4676 if (!path)
4677 return -ENOMEM;
4678
5f3ab90a 4679 spin_lock(&send_root->root_item_lock);
31db9f7c 4680 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5f3ab90a 4681 spin_unlock(&send_root->root_item_lock);
31db9f7c
AB		 4682
4683 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4684 key.type = BTRFS_INODE_ITEM_KEY;
4685 key.offset = 0;
4686
4687join_trans:
4688 /*
4689 * We need to make sure the transaction does not get committed
4690 * while we do anything on commit roots. Join a transaction to prevent
4691 * this.
4692 */
4693 trans = btrfs_join_transaction(send_root);
4694 if (IS_ERR(trans)) {
4695 ret = PTR_ERR(trans);
4696 trans = NULL;
4697 goto out;
4698 }
4699
4700 /*
766702ef
AB		 4701 * Make sure the tree has not changed after re-joining. We detect this
4702 * by comparing start_ctransid and ctransid. They should always match.
31db9f7c 4703 */
5f3ab90a 4704 spin_lock(&send_root->root_item_lock);
31db9f7c 4705 ctransid = btrfs_root_ctransid(&send_root->root_item);
5f3ab90a 4706 spin_unlock(&send_root->root_item_lock);
31db9f7c
AB		 4707
4708 if (ctransid != start_ctransid) {
4709 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4710 "send was modified in between. This is "
4711 "probably a bug.\n");
4712 ret = -EIO;
4713 goto out;
4714 }
4715
4716 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4717 if (ret < 0)
4718 goto out;
4719 if (ret)
4720 goto out_finish;
4721
4722 while (1) {
4723 /*
4724 * When someone want to commit while we iterate, end the
4725 * joined transaction and rejoin.
4726 */
4727 if (btrfs_should_end_transaction(trans, send_root)) {
4728 ret = btrfs_end_transaction(trans, send_root);
4729 trans = NULL;
4730 if (ret < 0)
4731 goto out;
4732 btrfs_release_path(path);
4733 goto join_trans;
4734 }
4735
4736 eb = path->nodes[0];
4737 slot = path->slots[0];
4738 btrfs_item_key_to_cpu(eb, &found_key, slot);
4739
4740 ret = changed_cb(send_root, NULL, path, NULL,
4741 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4742 if (ret < 0)
4743 goto out;
4744
4745 key.objectid = found_key.objectid;
4746 key.type = found_key.type;
4747 key.offset = found_key.offset + 1;
4748
4749 ret = btrfs_next_item(send_root, path);
4750 if (ret < 0)
4751 goto out;
4752 if (ret) {
4753 ret = 0;
4754 break;
4755 }
4756 }
4757
4758out_finish:
4759 ret = finish_inode_if_needed(sctx, 1);
4760
4761out:
4762 btrfs_free_path(path);
4763 if (trans) {
4764 if (!ret)
4765 ret = btrfs_end_transaction(trans, send_root);
4766 else
4767 btrfs_end_transaction(trans, send_root);
4768 }
4769 return ret;
4770}
4771
4772static int send_subvol(struct send_ctx *sctx)
4773{
4774 int ret;
4775
c2c71324
SB		 4776 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
4777 ret = send_header(sctx);
4778 if (ret < 0)
4779 goto out;
4780 }
31db9f7c
AB		 4781
4782 ret = send_subvol_begin(sctx);
4783 if (ret < 0)
4784 goto out;
4785
4786 if (sctx->parent_root) {
4787 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4788 changed_cb, sctx);
4789 if (ret < 0)
4790 goto out;
4791 ret = finish_inode_if_needed(sctx, 1);
4792 if (ret < 0)
4793 goto out;
4794 } else {
4795 ret = full_send_tree(sctx);
4796 if (ret < 0)
4797 goto out;
4798 }
4799
4800out:
31db9f7c
AB		 4801 free_recorded_refs(sctx);
4802 return ret;
4803}
4804
4805long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4806{
4807 int ret = 0;
4808 struct btrfs_root *send_root;
4809 struct btrfs_root *clone_root;
4810 struct btrfs_fs_info *fs_info;
4811 struct btrfs_ioctl_send_args *arg = NULL;
4812 struct btrfs_key key;
31db9f7c
AB		 4813 struct send_ctx *sctx = NULL;
4814 u32 i;
4815 u64 *clone_sources_tmp = NULL;
4816
4817 if (!capable(CAP_SYS_ADMIN))
4818 return -EPERM;
4819
496ad9aa 4820 send_root = BTRFS_I(file_inode(mnt_file))->root;
31db9f7c
AB		 4821 fs_info = send_root->fs_info;
4822
139f807a
JB		 4823 /*
4824 * This is done when we lookup the root, it should already be complete
4825 * by the time we get here.
4826 */
4827 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
4828
4829 /*
4830 * If we just created this root we need to make sure that the orphan
4831 * cleanup has been done and committed since we search the commit root,
4832 * so check its commit root transid with our otransid and if they match
4833 * commit the transaction to make sure everything is updated.
4834 */
4835 down_read(&send_root->fs_info->extent_commit_sem);
4836 if (btrfs_header_generation(send_root->commit_root) ==
4837 btrfs_root_otransid(&send_root->root_item)) {
4838 struct btrfs_trans_handle *trans;
4839
4840 up_read(&send_root->fs_info->extent_commit_sem);
4841
4842 trans = btrfs_attach_transaction_barrier(send_root);
4843 if (IS_ERR(trans)) {
4844 if (PTR_ERR(trans) != -ENOENT) {
4845 ret = PTR_ERR(trans);
4846 goto out;
4847 }
4848 /* ENOENT means theres no transaction */
4849 } else {
4850 ret = btrfs_commit_transaction(trans, send_root);
4851 if (ret)
4852 goto out;
4853 }
4854 } else {
4855 up_read(&send_root->fs_info->extent_commit_sem);
4856 }
4857
31db9f7c
AB		 4858 arg = memdup_user(arg_, sizeof(*arg));
4859 if (IS_ERR(arg)) {
4860 ret = PTR_ERR(arg);
4861 arg = NULL;
4862 goto out;
4863 }
4864
4865 if (!access_ok(VERIFY_READ, arg->clone_sources,
700ff4f0
DC		 4866 sizeof(*arg->clone_sources) *
4867 arg->clone_sources_count)) {
31db9f7c
AB		 4868 ret = -EFAULT;
4869 goto out;
4870 }
4871
c2c71324 4872 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
cb95e7bf
MF		 4873 ret = -EINVAL;
4874 goto out;
4875 }
4876
31db9f7c
AB		 4877 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4878 if (!sctx) {
4879 ret = -ENOMEM;
4880 goto out;
4881 }
4882
4883 INIT_LIST_HEAD(&sctx->new_refs);
4884 INIT_LIST_HEAD(&sctx->deleted_refs);
4885 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4886 INIT_LIST_HEAD(&sctx->name_cache_list);
4887
cb95e7bf
MF		 4888 sctx->flags = arg->flags;
4889
31db9f7c 4890 sctx->send_filp = fget(arg->send_fd);
ecc7ada7
TI		 4891 if (!sctx->send_filp) {
4892 ret = -EBADF;
31db9f7c
AB		 4893 goto out;
4894 }
4895
4896 sctx->mnt = mnt_file->f_path.mnt;
4897
4898 sctx->send_root = send_root;
4899 sctx->clone_roots_cnt = arg->clone_sources_count;
4900
4901 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4902 sctx->send_buf = vmalloc(sctx->send_max_size);
4903 if (!sctx->send_buf) {
4904 ret = -ENOMEM;
4905 goto out;
4906 }
4907
4908 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4909 if (!sctx->read_buf) {
4910 ret = -ENOMEM;
4911 goto out;
4912 }
4913
4914 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4915 (arg->clone_sources_count + 1));
4916 if (!sctx->clone_roots) {
4917 ret = -ENOMEM;
4918 goto out;
4919 }
4920
4921 if (arg->clone_sources_count) {
4922 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4923 sizeof(*arg->clone_sources));
4924 if (!clone_sources_tmp) {
4925 ret = -ENOMEM;
4926 goto out;
4927 }
4928
4929 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4930 arg->clone_sources_count *
4931 sizeof(*arg->clone_sources));
4932 if (ret) {
4933 ret = -EFAULT;
4934 goto out;
4935 }
4936
4937 for (i = 0; i < arg->clone_sources_count; i++) {
4938 key.objectid = clone_sources_tmp[i];
4939 key.type = BTRFS_ROOT_ITEM_KEY;
4940 key.offset = (u64)-1;
4941 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
31db9f7c
AB		 4942 if (IS_ERR(clone_root)) {
4943 ret = PTR_ERR(clone_root);
4944 goto out;
4945 }
4946 sctx->clone_roots[i].root = clone_root;
4947 }
4948 vfree(clone_sources_tmp);
4949 clone_sources_tmp = NULL;
4950 }
4951
4952 if (arg->parent_root) {
4953 key.objectid = arg->parent_root;
4954 key.type = BTRFS_ROOT_ITEM_KEY;
4955 key.offset = (u64)-1;
4956 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
b1b19596
SB		 4957 if (IS_ERR(sctx->parent_root)) {
4958 ret = PTR_ERR(sctx->parent_root);
31db9f7c
AB		 4959 goto out;
4960 }
4961 }
4962
4963 /*
4964 * Clones from send_root are allowed, but only if the clone source
4965 * is behind the current send position. This is checked while searching
4966 * for possible clone sources.
4967 */
4968 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4969
4970 /* We do a bsearch later */
4971 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4972 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4973 NULL);
4974
4975 ret = send_subvol(sctx);
4976 if (ret < 0)
4977 goto out;
4978
c2c71324
SB		 4979 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
4980 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4981 if (ret < 0)
4982 goto out;
4983 ret = send_cmd(sctx);
4984 if (ret < 0)
4985 goto out;
4986 }
31db9f7c
AB		 4987
4988out:
31db9f7c
AB		 4989 kfree(arg);
4990 vfree(clone_sources_tmp);
4991
4992 if (sctx) {
4993 if (sctx->send_filp)
4994 fput(sctx->send_filp);
4995
4996 vfree(sctx->clone_roots);
4997 vfree(sctx->send_buf);
4998 vfree(sctx->read_buf);
4999
5000 name_cache_free(sctx);
5001
5002 kfree(sctx);
5003 }
5004
5005 return ret;
5006}
Linux 6.x block layer and io_uring tree(s)