projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Remove redundant local zero structure
[deliverable/linux.git] / fs / btrfs / send.c
CommitLineData
31db9f7c
AB		 1/*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/bsearch.h>
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/sort.h>
23#include <linux/mount.h>
24#include <linux/xattr.h>
25#include <linux/posix_acl_xattr.h>
26#include <linux/radix-tree.h>
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;
114
115 u64 send_progress;
116
117 struct list_head new_refs;
118 struct list_head deleted_refs;
119
120 struct radix_tree_root name_cache;
121 struct list_head name_cache_list;
122 int name_cache_size;
123
31db9f7c
AB		 124 char *read_buf;
125};
126
127struct name_cache_entry {
128 struct list_head list;
7e0926fe
AB		 129 /*
130 * radix_tree has only 32bit entries but we need to handle 64bit inums.
131 * We use the lower 32bit of the 64bit inum to store it in the tree. If
132 * more then one inum would fall into the same entry, we use radix_list
133 * to store the additional entries. radix_list is also used to store
134 * entries where two entries have the same inum but different
135 * generations.
136 */
137 struct list_head radix_list;
31db9f7c
AB		 138 u64 ino;
139 u64 gen;
140 u64 parent_ino;
141 u64 parent_gen;
142 int ret;
143 int need_later_update;
144 int name_len;
145 char name[];
146};
147
148static void fs_path_reset(struct fs_path *p)
149{
150 if (p->reversed) {
151 p->start = p->buf + p->buf_len - 1;
152 p->end = p->start;
153 *p->start = 0;
154 } else {
155 p->start = p->buf;
156 p->end = p->start;
157 *p->start = 0;
158 }
159}
160
924794c9 161static struct fs_path *fs_path_alloc(void)
31db9f7c
AB		 162{
163 struct fs_path *p;
164
165 p = kmalloc(sizeof(*p), GFP_NOFS);
166 if (!p)
167 return NULL;
168 p->reversed = 0;
169 p->virtual_mem = 0;
170 p->buf = p->inline_buf;
171 p->buf_len = FS_PATH_INLINE_SIZE;
172 fs_path_reset(p);
173 return p;
174}
175
924794c9 176static struct fs_path *fs_path_alloc_reversed(void)
31db9f7c
AB		 177{
178 struct fs_path *p;
179
924794c9 180 p = fs_path_alloc();
31db9f7c
AB		 181 if (!p)
182 return NULL;
183 p->reversed = 1;
184 fs_path_reset(p);
185 return p;
186}
187
924794c9 188static void fs_path_free(struct fs_path *p)
31db9f7c
AB		 189{
190 if (!p)
191 return;
192 if (p->buf != p->inline_buf) {
193 if (p->virtual_mem)
194 vfree(p->buf);
195 else
196 kfree(p->buf);
197 }
198 kfree(p);
199}
200
201static int fs_path_len(struct fs_path *p)
202{
203 return p->end - p->start;
204}
205
206static int fs_path_ensure_buf(struct fs_path *p, int len)
207{
208 char *tmp_buf;
209 int path_len;
210 int old_buf_len;
211
212 len++;
213
214 if (p->buf_len >= len)
215 return 0;
216
217 path_len = p->end - p->start;
218 old_buf_len = p->buf_len;
219 len = PAGE_ALIGN(len);
220
221 if (p->buf == p->inline_buf) {
8be04b93 222 tmp_buf = kmalloc(len, GFP_NOFS | __GFP_NOWARN);
31db9f7c
AB		 223 if (!tmp_buf) {
224 tmp_buf = vmalloc(len);
225 if (!tmp_buf)
226 return -ENOMEM;
227 p->virtual_mem = 1;
228 }
229 memcpy(tmp_buf, p->buf, p->buf_len);
230 p->buf = tmp_buf;
231 p->buf_len = len;
232 } else {
233 if (p->virtual_mem) {
234 tmp_buf = vmalloc(len);
235 if (!tmp_buf)
236 return -ENOMEM;
237 memcpy(tmp_buf, p->buf, p->buf_len);
238 vfree(p->buf);
239 } else {
240 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
241 if (!tmp_buf) {
242 tmp_buf = vmalloc(len);
243 if (!tmp_buf)
244 return -ENOMEM;
245 memcpy(tmp_buf, p->buf, p->buf_len);
246 kfree(p->buf);
247 p->virtual_mem = 1;
248 }
249 }
250 p->buf = tmp_buf;
251 p->buf_len = len;
252 }
253 if (p->reversed) {
254 tmp_buf = p->buf + old_buf_len - path_len - 1;
255 p->end = p->buf + p->buf_len - 1;
256 p->start = p->end - path_len;
257 memmove(p->start, tmp_buf, path_len + 1);
258 } else {
259 p->start = p->buf;
260 p->end = p->start + path_len;
261 }
262 return 0;
263}
264
265static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
266{
267 int ret;
268 int new_len;
269
270 new_len = p->end - p->start + name_len;
271 if (p->start != p->end)
272 new_len++;
273 ret = fs_path_ensure_buf(p, new_len);
274 if (ret < 0)
275 goto out;
276
277 if (p->reversed) {
278 if (p->start != p->end)
279 *--p->start = '/';
280 p->start -= name_len;
281 p->prepared = p->start;
282 } else {
283 if (p->start != p->end)
284 *p->end++ = '/';
285 p->prepared = p->end;
286 p->end += name_len;
287 *p->end = 0;
288 }
289
290out:
291 return ret;
292}
293
294static int fs_path_add(struct fs_path *p, const char *name, int name_len)
295{
296 int ret;
297
298 ret = fs_path_prepare_for_add(p, name_len);
299 if (ret < 0)
300 goto out;
301 memcpy(p->prepared, name, name_len);
302 p->prepared = NULL;
303
304out:
305 return ret;
306}
307
308static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
309{
310 int ret;
311
312 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
313 if (ret < 0)
314 goto out;
315 memcpy(p->prepared, p2->start, p2->end - p2->start);
316 p->prepared = NULL;
317
318out:
319 return ret;
320}
321
322static int fs_path_add_from_extent_buffer(struct fs_path *p,
323 struct extent_buffer *eb,
324 unsigned long off, int len)
325{
326 int ret;
327
328 ret = fs_path_prepare_for_add(p, len);
329 if (ret < 0)
330 goto out;
331
332 read_extent_buffer(eb, p->prepared, off, len);
333 p->prepared = NULL;
334
335out:
336 return ret;
337}
338
9ea3ef51 339#if 0
31db9f7c
AB		 340static void fs_path_remove(struct fs_path *p)
341{
342 BUG_ON(p->reversed);
343 while (p->start != p->end && *p->end != '/')
344 p->end--;
345 *p->end = 0;
346}
9ea3ef51 347#endif
31db9f7c
AB		 348
349static int fs_path_copy(struct fs_path *p, struct fs_path *from)
350{
351 int ret;
352
353 p->reversed = from->reversed;
354 fs_path_reset(p);
355
356 ret = fs_path_add_path(p, from);
357
358 return ret;
359}
360
361
362static void fs_path_unreverse(struct fs_path *p)
363{
364 char *tmp;
365 int len;
366
367 if (!p->reversed)
368 return;
369
370 tmp = p->start;
371 len = p->end - p->start;
372 p->start = p->buf;
373 p->end = p->start + len;
374 memmove(p->start, tmp, len + 1);
375 p->reversed = 0;
376}
377
378static struct btrfs_path *alloc_path_for_send(void)
379{
380 struct btrfs_path *path;
381
382 path = btrfs_alloc_path();
383 if (!path)
384 return NULL;
385 path->search_commit_root = 1;
386 path->skip_locking = 1;
387 return path;
388}
389
48a3b636 390static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
31db9f7c
AB		 391{
392 int ret;
393 mm_segment_t old_fs;
394 u32 pos = 0;
395
396 old_fs = get_fs();
397 set_fs(KERNEL_DS);
398
399 while (pos < len) {
1bcea355 400 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
31db9f7c
AB		 401 /* TODO handle that correctly */
402 /*if (ret == -ERESTARTSYS) {
403 continue;
404 }*/
405 if (ret < 0)
406 goto out;
407 if (ret == 0) {
408 ret = -EIO;
409 goto out;
410 }
411 pos += ret;
412 }
413
414 ret = 0;
415
416out:
417 set_fs(old_fs);
418 return ret;
419}
420
421static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
422{
423 struct btrfs_tlv_header *hdr;
424 int total_len = sizeof(*hdr) + len;
425 int left = sctx->send_max_size - sctx->send_size;
426
427 if (unlikely(left < total_len))
428 return -EOVERFLOW;
429
430 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
431 hdr->tlv_type = cpu_to_le16(attr);
432 hdr->tlv_len = cpu_to_le16(len);
433 memcpy(hdr + 1, data, len);
434 sctx->send_size += total_len;
435
436 return 0;
437}
438
439#if 0
440static int tlv_put_u8(struct send_ctx *sctx, u16 attr, u8 value)
441{
442 return tlv_put(sctx, attr, &value, sizeof(value));
443}
444
445static int tlv_put_u16(struct send_ctx *sctx, u16 attr, u16 value)
446{
447 __le16 tmp = cpu_to_le16(value);
448 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
449}
450
451static int tlv_put_u32(struct send_ctx *sctx, u16 attr, u32 value)
452{
453 __le32 tmp = cpu_to_le32(value);
454 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
455}
456#endif
457
458static int tlv_put_u64(struct send_ctx *sctx, u16 attr, u64 value)
459{
460 __le64 tmp = cpu_to_le64(value);
461 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
462}
463
464static int tlv_put_string(struct send_ctx *sctx, u16 attr,
465 const char *str, int len)
466{
467 if (len == -1)
468 len = strlen(str);
469 return tlv_put(sctx, attr, str, len);
470}
471
472static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
473 const u8 *uuid)
474{
475 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
476}
477
478#if 0
479static int tlv_put_timespec(struct send_ctx *sctx, u16 attr,
480 struct timespec *ts)
481{
482 struct btrfs_timespec bts;
483 bts.sec = cpu_to_le64(ts->tv_sec);
484 bts.nsec = cpu_to_le32(ts->tv_nsec);
485 return tlv_put(sctx, attr, &bts, sizeof(bts));
486}
487#endif
488
489static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
490 struct extent_buffer *eb,
491 struct btrfs_timespec *ts)
492{
493 struct btrfs_timespec bts;
494 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
495 return tlv_put(sctx, attr, &bts, sizeof(bts));
496}
497
498
499#define TLV_PUT(sctx, attrtype, attrlen, data) \
500 do { \
501 ret = tlv_put(sctx, attrtype, attrlen, data); \
502 if (ret < 0) \
503 goto tlv_put_failure; \
504 } while (0)
505
506#define TLV_PUT_INT(sctx, attrtype, bits, value) \
507 do { \
508 ret = tlv_put_u##bits(sctx, attrtype, value); \
509 if (ret < 0) \
510 goto tlv_put_failure; \
511 } while (0)
512
513#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
514#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
515#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
516#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
517#define TLV_PUT_STRING(sctx, attrtype, str, len) \
518 do { \
519 ret = tlv_put_string(sctx, attrtype, str, len); \
520 if (ret < 0) \
521 goto tlv_put_failure; \
522 } while (0)
523#define TLV_PUT_PATH(sctx, attrtype, p) \
524 do { \
525 ret = tlv_put_string(sctx, attrtype, p->start, \
526 p->end - p->start); \
527 if (ret < 0) \
528 goto tlv_put_failure; \
529 } while(0)
530#define TLV_PUT_UUID(sctx, attrtype, uuid) \
531 do { \
532 ret = tlv_put_uuid(sctx, attrtype, uuid); \
533 if (ret < 0) \
534 goto tlv_put_failure; \
535 } while (0)
536#define TLV_PUT_TIMESPEC(sctx, attrtype, ts) \
537 do { \
538 ret = tlv_put_timespec(sctx, attrtype, ts); \
539 if (ret < 0) \
540 goto tlv_put_failure; \
541 } while (0)
542#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
543 do { \
544 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
545 if (ret < 0) \
546 goto tlv_put_failure; \
547 } while (0)
548
549static int send_header(struct send_ctx *sctx)
550{
551 struct btrfs_stream_header hdr;
552
553 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
554 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
555
1bcea355
AJ		 556 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
557 &sctx->send_off);
31db9f7c
AB		 558}
559
560/*
561 * For each command/item we want to send to userspace, we call this function.
562 */
563static int begin_cmd(struct send_ctx *sctx, int cmd)
564{
565 struct btrfs_cmd_header *hdr;
566
567 if (!sctx->send_buf) {
568 WARN_ON(1);
569 return -EINVAL;
570 }
571
572 BUG_ON(sctx->send_size);
573
574 sctx->send_size += sizeof(*hdr);
575 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
576 hdr->cmd = cpu_to_le16(cmd);
577
578 return 0;
579}
580
581static int send_cmd(struct send_ctx *sctx)
582{
583 int ret;
584 struct btrfs_cmd_header *hdr;
585 u32 crc;
586
587 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
588 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
589 hdr->crc = 0;
590
591 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
592 hdr->crc = cpu_to_le32(crc);
593
1bcea355
AJ		 594 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
595 &sctx->send_off);
31db9f7c
AB		 596
597 sctx->total_send_size += sctx->send_size;
598 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
599 sctx->send_size = 0;
600
601 return ret;
602}
603
604/*
605 * Sends a move instruction to user space
606 */
607static int send_rename(struct send_ctx *sctx,
608 struct fs_path *from, struct fs_path *to)
609{
610 int ret;
611
612verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
613
614 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
615 if (ret < 0)
616 goto out;
617
618 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
619 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
620
621 ret = send_cmd(sctx);
622
623tlv_put_failure:
624out:
625 return ret;
626}
627
628/*
629 * Sends a link instruction to user space
630 */
631static int send_link(struct send_ctx *sctx,
632 struct fs_path *path, struct fs_path *lnk)
633{
634 int ret;
635
636verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
637
638 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
639 if (ret < 0)
640 goto out;
641
642 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
643 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
644
645 ret = send_cmd(sctx);
646
647tlv_put_failure:
648out:
649 return ret;
650}
651
652/*
653 * Sends an unlink instruction to user space
654 */
655static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
656{
657 int ret;
658
659verbose_printk("btrfs: send_unlink %s\n", path->start);
660
661 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
662 if (ret < 0)
663 goto out;
664
665 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
666
667 ret = send_cmd(sctx);
668
669tlv_put_failure:
670out:
671 return ret;
672}
673
674/*
675 * Sends a rmdir instruction to user space
676 */
677static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
678{
679 int ret;
680
681verbose_printk("btrfs: send_rmdir %s\n", path->start);
682
683 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
684 if (ret < 0)
685 goto out;
686
687 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
688
689 ret = send_cmd(sctx);
690
691tlv_put_failure:
692out:
693 return ret;
694}
695
696/*
697 * Helper function to retrieve some fields from an inode item.
698 */
699static int get_inode_info(struct btrfs_root *root,
700 u64 ino, u64 *size, u64 *gen,
85a7b33b
AB		 701 u64 *mode, u64 *uid, u64 *gid,
702 u64 *rdev)
31db9f7c
AB		 703{
704 int ret;
705 struct btrfs_inode_item *ii;
706 struct btrfs_key key;
707 struct btrfs_path *path;
708
709 path = alloc_path_for_send();
710 if (!path)
711 return -ENOMEM;
712
713 key.objectid = ino;
714 key.type = BTRFS_INODE_ITEM_KEY;
715 key.offset = 0;
716 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
717 if (ret < 0)
718 goto out;
719 if (ret) {
720 ret = -ENOENT;
721 goto out;
722 }
723
724 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
725 struct btrfs_inode_item);
726 if (size)
727 *size = btrfs_inode_size(path->nodes[0], ii);
728 if (gen)
729 *gen = btrfs_inode_generation(path->nodes[0], ii);
730 if (mode)
731 *mode = btrfs_inode_mode(path->nodes[0], ii);
732 if (uid)
733 *uid = btrfs_inode_uid(path->nodes[0], ii);
734 if (gid)
735 *gid = btrfs_inode_gid(path->nodes[0], ii);
85a7b33b
AB		 736 if (rdev)
737 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
31db9f7c
AB		 738
739out:
740 btrfs_free_path(path);
741 return ret;
742}
743
744typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
745 struct fs_path *p,
746 void *ctx);
747
748/*
96b5bd77
JS		 749 * Helper function to iterate the entries in ONE btrfs_inode_ref or
750 * btrfs_inode_extref.
31db9f7c
AB		 751 * The iterate callback may return a non zero value to stop iteration. This can
752 * be a negative value for error codes or 1 to simply stop it.
753 *
96b5bd77 754 * path must point to the INODE_REF or INODE_EXTREF when called.
31db9f7c 755 */
924794c9 756static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 757 struct btrfs_key *found_key, int resolve,
758 iterate_inode_ref_t iterate, void *ctx)
759{
96b5bd77 760 struct extent_buffer *eb = path->nodes[0];
31db9f7c
AB		 761 struct btrfs_item *item;
762 struct btrfs_inode_ref *iref;
96b5bd77 763 struct btrfs_inode_extref *extref;
31db9f7c
AB		 764 struct btrfs_path *tmp_path;
765 struct fs_path *p;
96b5bd77 766 u32 cur = 0;
31db9f7c 767 u32 total;
96b5bd77 768 int slot = path->slots[0];
31db9f7c
AB		 769 u32 name_len;
770 char *start;
771 int ret = 0;
96b5bd77 772 int num = 0;
31db9f7c 773 int index;
96b5bd77
JS		 774 u64 dir;
775 unsigned long name_off;
776 unsigned long elem_size;
777 unsigned long ptr;
31db9f7c 778
924794c9 779 p = fs_path_alloc_reversed();
31db9f7c
AB		 780 if (!p)
781 return -ENOMEM;
782
783 tmp_path = alloc_path_for_send();
784 if (!tmp_path) {
924794c9 785 fs_path_free(p);
31db9f7c
AB		 786 return -ENOMEM;
787 }
788
31db9f7c 789
96b5bd77
JS		 790 if (found_key->type == BTRFS_INODE_REF_KEY) {
791 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
792 struct btrfs_inode_ref);
dd3cc16b 793 item = btrfs_item_nr(slot);
96b5bd77
JS		 794 total = btrfs_item_size(eb, item);
795 elem_size = sizeof(*iref);
796 } else {
797 ptr = btrfs_item_ptr_offset(eb, slot);
798 total = btrfs_item_size_nr(eb, slot);
799 elem_size = sizeof(*extref);
800 }
801
31db9f7c
AB		 802 while (cur < total) {
803 fs_path_reset(p);
804
96b5bd77
JS		 805 if (found_key->type == BTRFS_INODE_REF_KEY) {
806 iref = (struct btrfs_inode_ref *)(ptr + cur);
807 name_len = btrfs_inode_ref_name_len(eb, iref);
808 name_off = (unsigned long)(iref + 1);
809 index = btrfs_inode_ref_index(eb, iref);
810 dir = found_key->offset;
811 } else {
812 extref = (struct btrfs_inode_extref *)(ptr + cur);
813 name_len = btrfs_inode_extref_name_len(eb, extref);
814 name_off = (unsigned long)&extref->name;
815 index = btrfs_inode_extref_index(eb, extref);
816 dir = btrfs_inode_extref_parent(eb, extref);
817 }
818
31db9f7c 819 if (resolve) {
96b5bd77
JS		 820 start = btrfs_ref_to_path(root, tmp_path, name_len,
821 name_off, eb, dir,
822 p->buf, p->buf_len);
31db9f7c
AB		 823 if (IS_ERR(start)) {
824 ret = PTR_ERR(start);
825 goto out;
826 }
827 if (start < p->buf) {
828 /* overflow , try again with larger buffer */
829 ret = fs_path_ensure_buf(p,
830 p->buf_len + p->buf - start);
831 if (ret < 0)
832 goto out;
96b5bd77
JS		 833 start = btrfs_ref_to_path(root, tmp_path,
834 name_len, name_off,
835 eb, dir,
836 p->buf, p->buf_len);
31db9f7c
AB		 837 if (IS_ERR(start)) {
838 ret = PTR_ERR(start);
839 goto out;
840 }
841 BUG_ON(start < p->buf);
842 }
843 p->start = start;
844 } else {
96b5bd77
JS		 845 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
846 name_len);
31db9f7c
AB		 847 if (ret < 0)
848 goto out;
849 }
850
96b5bd77
JS		 851 cur += elem_size + name_len;
852 ret = iterate(num, dir, index, p, ctx);
31db9f7c
AB		 853 if (ret)
854 goto out;
31db9f7c
AB		 855 num++;
856 }
857
858out:
859 btrfs_free_path(tmp_path);
924794c9 860 fs_path_free(p);
31db9f7c
AB		 861 return ret;
862}
863
864typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
865 const char *name, int name_len,
866 const char *data, int data_len,
867 u8 type, void *ctx);
868
869/*
870 * Helper function to iterate the entries in ONE btrfs_dir_item.
871 * The iterate callback may return a non zero value to stop iteration. This can
872 * be a negative value for error codes or 1 to simply stop it.
873 *
874 * path must point to the dir item when called.
875 */
924794c9 876static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 877 struct btrfs_key *found_key,
878 iterate_dir_item_t iterate, void *ctx)
879{
880 int ret = 0;
881 struct extent_buffer *eb;
882 struct btrfs_item *item;
883 struct btrfs_dir_item *di;
31db9f7c
AB		 884 struct btrfs_key di_key;
885 char *buf = NULL;
886 char *buf2 = NULL;
887 int buf_len;
888 int buf_virtual = 0;
889 u32 name_len;
890 u32 data_len;
891 u32 cur;
892 u32 len;
893 u32 total;
894 int slot;
895 int num;
896 u8 type;
897
898 buf_len = PAGE_SIZE;
899 buf = kmalloc(buf_len, GFP_NOFS);
900 if (!buf) {
901 ret = -ENOMEM;
902 goto out;
903 }
904
31db9f7c
AB		 905 eb = path->nodes[0];
906 slot = path->slots[0];
dd3cc16b 907 item = btrfs_item_nr(slot);
31db9f7c
AB		 908 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
909 cur = 0;
910 len = 0;
911 total = btrfs_item_size(eb, item);
912
913 num = 0;
914 while (cur < total) {
915 name_len = btrfs_dir_name_len(eb, di);
916 data_len = btrfs_dir_data_len(eb, di);
917 type = btrfs_dir_type(eb, di);
918 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
919
920 if (name_len + data_len > buf_len) {
921 buf_len = PAGE_ALIGN(name_len + data_len);
922 if (buf_virtual) {
923 buf2 = vmalloc(buf_len);
924 if (!buf2) {
925 ret = -ENOMEM;
926 goto out;
927 }
928 vfree(buf);
929 } else {
930 buf2 = krealloc(buf, buf_len, GFP_NOFS);
931 if (!buf2) {
932 buf2 = vmalloc(buf_len);
933 if (!buf2) {
934 ret = -ENOMEM;
935 goto out;
936 }
937 kfree(buf);
938 buf_virtual = 1;
939 }
940 }
941
942 buf = buf2;
943 buf2 = NULL;
944 }
945
946 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
947 name_len + data_len);
948
949 len = sizeof(*di) + name_len + data_len;
950 di = (struct btrfs_dir_item *)((char *)di + len);
951 cur += len;
952
953 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
954 data_len, type, ctx);
955 if (ret < 0)
956 goto out;
957 if (ret) {
958 ret = 0;
959 goto out;
960 }
961
962 num++;
963 }
964
965out:
31db9f7c
AB		 966 if (buf_virtual)
967 vfree(buf);
968 else
969 kfree(buf);
970 return ret;
971}
972
973static int __copy_first_ref(int num, u64 dir, int index,
974 struct fs_path *p, void *ctx)
975{
976 int ret;
977 struct fs_path *pt = ctx;
978
979 ret = fs_path_copy(pt, p);
980 if (ret < 0)
981 return ret;
982
983 /* we want the first only */
984 return 1;
985}
986
987/*
988 * Retrieve the first path of an inode. If an inode has more then one
989 * ref/hardlink, this is ignored.
990 */
924794c9 991static int get_inode_path(struct btrfs_root *root,
31db9f7c
AB		 992 u64 ino, struct fs_path *path)
993{
994 int ret;
995 struct btrfs_key key, found_key;
996 struct btrfs_path *p;
997
998 p = alloc_path_for_send();
999 if (!p)
1000 return -ENOMEM;
1001
1002 fs_path_reset(path);
1003
1004 key.objectid = ino;
1005 key.type = BTRFS_INODE_REF_KEY;
1006 key.offset = 0;
1007
1008 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1009 if (ret < 0)
1010 goto out;
1011 if (ret) {
1012 ret = 1;
1013 goto out;
1014 }
1015 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1016 if (found_key.objectid != ino ||
96b5bd77
JS		 1017 (found_key.type != BTRFS_INODE_REF_KEY &&
1018 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1019 ret = -ENOENT;
1020 goto out;
1021 }
1022
924794c9
TI		 1023 ret = iterate_inode_ref(root, p, &found_key, 1,
1024 __copy_first_ref, path);
31db9f7c
AB		 1025 if (ret < 0)
1026 goto out;
1027 ret = 0;
1028
1029out:
1030 btrfs_free_path(p);
1031 return ret;
1032}
1033
1034struct backref_ctx {
1035 struct send_ctx *sctx;
1036
1037 /* number of total found references */
1038 u64 found;
1039
1040 /*
1041 * used for clones found in send_root. clones found behind cur_objectid
1042 * and cur_offset are not considered as allowed clones.
1043 */
1044 u64 cur_objectid;
1045 u64 cur_offset;
1046
1047 /* may be truncated in case it's the last extent in a file */
1048 u64 extent_len;
1049
1050 /* Just to check for bugs in backref resolving */
ee849c04 1051 int found_itself;
31db9f7c
AB		 1052};
1053
1054static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1055{
995e01b7 1056 u64 root = (u64)(uintptr_t)key;
31db9f7c
AB		 1057 struct clone_root *cr = (struct clone_root *)elt;
1058
1059 if (root < cr->root->objectid)
1060 return -1;
1061 if (root > cr->root->objectid)
1062 return 1;
1063 return 0;
1064}
1065
1066static int __clone_root_cmp_sort(const void *e1, const void *e2)
1067{
1068 struct clone_root *cr1 = (struct clone_root *)e1;
1069 struct clone_root *cr2 = (struct clone_root *)e2;
1070
1071 if (cr1->root->objectid < cr2->root->objectid)
1072 return -1;
1073 if (cr1->root->objectid > cr2->root->objectid)
1074 return 1;
1075 return 0;
1076}
1077
1078/*
1079 * Called for every backref that is found for the current extent.
766702ef 1080 * Results are collected in sctx->clone_roots->ino/offset/found_refs
31db9f7c
AB		 1081 */
1082static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1083{
1084 struct backref_ctx *bctx = ctx_;
1085 struct clone_root *found;
1086 int ret;
1087 u64 i_size;
1088
1089 /* First check if the root is in the list of accepted clone sources */
995e01b7 1090 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
31db9f7c
AB		 1091 bctx->sctx->clone_roots_cnt,
1092 sizeof(struct clone_root),
1093 __clone_root_cmp_bsearch);
1094 if (!found)
1095 return 0;
1096
1097 if (found->root == bctx->sctx->send_root &&
1098 ino == bctx->cur_objectid &&
1099 offset == bctx->cur_offset) {
ee849c04 1100 bctx->found_itself = 1;
31db9f7c
AB		 1101 }
1102
1103 /*
766702ef 1104 * There are inodes that have extents that lie behind its i_size. Don't
31db9f7c
AB		 1105 * accept clones from these extents.
1106 */
85a7b33b
AB		 1107 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1108 NULL);
31db9f7c
AB		 1109 if (ret < 0)
1110 return ret;
1111
1112 if (offset + bctx->extent_len > i_size)
1113 return 0;
1114
1115 /*
1116 * Make sure we don't consider clones from send_root that are
1117 * behind the current inode/offset.
1118 */
1119 if (found->root == bctx->sctx->send_root) {
1120 /*
1121 * TODO for the moment we don't accept clones from the inode
1122 * that is currently send. We may change this when
1123 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1124 * file.
1125 */
1126 if (ino >= bctx->cur_objectid)
1127 return 0;
e938c8ad
AB		 1128#if 0
1129 if (ino > bctx->cur_objectid)
1130 return 0;
1131 if (offset + bctx->extent_len > bctx->cur_offset)
31db9f7c 1132 return 0;
e938c8ad 1133#endif
31db9f7c
AB		 1134 }
1135
1136 bctx->found++;
1137 found->found_refs++;
1138 if (ino < found->ino) {
1139 found->ino = ino;
1140 found->offset = offset;
1141 } else if (found->ino == ino) {
1142 /*
1143 * same extent found more then once in the same file.
1144 */
1145 if (found->offset > offset + bctx->extent_len)
1146 found->offset = offset;
1147 }
1148
1149 return 0;
1150}
1151
1152/*
766702ef
AB		 1153 * Given an inode, offset and extent item, it finds a good clone for a clone
1154 * instruction. Returns -ENOENT when none could be found. The function makes
1155 * sure that the returned clone is usable at the point where sending is at the
1156 * moment. This means, that no clones are accepted which lie behind the current
1157 * inode+offset.
1158 *
31db9f7c
AB		 1159 * path must point to the extent item when called.
1160 */
1161static int find_extent_clone(struct send_ctx *sctx,
1162 struct btrfs_path *path,
1163 u64 ino, u64 data_offset,
1164 u64 ino_size,
1165 struct clone_root **found)
1166{
1167 int ret;
1168 int extent_type;
1169 u64 logical;
74dd17fb 1170 u64 disk_byte;
31db9f7c
AB		 1171 u64 num_bytes;
1172 u64 extent_item_pos;
69917e43 1173 u64 flags = 0;
31db9f7c
AB		 1174 struct btrfs_file_extent_item *fi;
1175 struct extent_buffer *eb = path->nodes[0];
35075bb0 1176 struct backref_ctx *backref_ctx = NULL;
31db9f7c
AB		 1177 struct clone_root *cur_clone_root;
1178 struct btrfs_key found_key;
1179 struct btrfs_path *tmp_path;
74dd17fb 1180 int compressed;
31db9f7c
AB		 1181 u32 i;
1182
1183 tmp_path = alloc_path_for_send();
1184 if (!tmp_path)
1185 return -ENOMEM;
1186
35075bb0
AB		 1187 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1188 if (!backref_ctx) {
1189 ret = -ENOMEM;
1190 goto out;
1191 }
1192
31db9f7c
AB		 1193 if (data_offset >= ino_size) {
1194 /*
1195 * There may be extents that lie behind the file's size.
1196 * I at least had this in combination with snapshotting while
1197 * writing large files.
1198 */
1199 ret = 0;
1200 goto out;
1201 }
1202
1203 fi = btrfs_item_ptr(eb, path->slots[0],
1204 struct btrfs_file_extent_item);
1205 extent_type = btrfs_file_extent_type(eb, fi);
1206 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1207 ret = -ENOENT;
1208 goto out;
1209 }
74dd17fb 1210 compressed = btrfs_file_extent_compression(eb, fi);
31db9f7c
AB		 1211
1212 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
74dd17fb
CM		 1213 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1214 if (disk_byte == 0) {
31db9f7c
AB		 1215 ret = -ENOENT;
1216 goto out;
1217 }
74dd17fb 1218 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
31db9f7c 1219
69917e43
LB		 1220 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1221 &found_key, &flags);
31db9f7c
AB		 1222 btrfs_release_path(tmp_path);
1223
1224 if (ret < 0)
1225 goto out;
69917e43 1226 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
31db9f7c
AB		 1227 ret = -EIO;
1228 goto out;
1229 }
1230
1231 /*
1232 * Setup the clone roots.
1233 */
1234 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1235 cur_clone_root = sctx->clone_roots + i;
1236 cur_clone_root->ino = (u64)-1;
1237 cur_clone_root->offset = 0;
1238 cur_clone_root->found_refs = 0;
1239 }
1240
35075bb0
AB		 1241 backref_ctx->sctx = sctx;
1242 backref_ctx->found = 0;
1243 backref_ctx->cur_objectid = ino;
1244 backref_ctx->cur_offset = data_offset;
1245 backref_ctx->found_itself = 0;
1246 backref_ctx->extent_len = num_bytes;
31db9f7c
AB		 1247
1248 /*
1249 * The last extent of a file may be too large due to page alignment.
1250 * We need to adjust extent_len in this case so that the checks in
1251 * __iterate_backrefs work.
1252 */
1253 if (data_offset + num_bytes >= ino_size)
35075bb0 1254 backref_ctx->extent_len = ino_size - data_offset;
31db9f7c
AB		 1255
1256 /*
1257 * Now collect all backrefs.
1258 */
74dd17fb
CM		 1259 if (compressed == BTRFS_COMPRESS_NONE)
1260 extent_item_pos = logical - found_key.objectid;
1261 else
1262 extent_item_pos = 0;
1263
31db9f7c
AB		 1264 extent_item_pos = logical - found_key.objectid;
1265 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1266 found_key.objectid, extent_item_pos, 1,
35075bb0 1267 __iterate_backrefs, backref_ctx);
74dd17fb 1268
31db9f7c
AB		 1269 if (ret < 0)
1270 goto out;
1271
35075bb0 1272 if (!backref_ctx->found_itself) {
31db9f7c
AB		 1273 /* found a bug in backref code? */
1274 ret = -EIO;
1275 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1276 "send_root. inode=%llu, offset=%llu, "
74dd17fb
CM		 1277 "disk_byte=%llu found extent=%llu\n",
1278 ino, data_offset, disk_byte, found_key.objectid);
31db9f7c
AB		 1279 goto out;
1280 }
1281
1282verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1283 "ino=%llu, "
1284 "num_bytes=%llu, logical=%llu\n",
1285 data_offset, ino, num_bytes, logical);
1286
35075bb0 1287 if (!backref_ctx->found)
31db9f7c
AB		 1288 verbose_printk("btrfs: no clones found\n");
1289
1290 cur_clone_root = NULL;
1291 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1292 if (sctx->clone_roots[i].found_refs) {
1293 if (!cur_clone_root)
1294 cur_clone_root = sctx->clone_roots + i;
1295 else if (sctx->clone_roots[i].root == sctx->send_root)
1296 /* prefer clones from send_root over others */
1297 cur_clone_root = sctx->clone_roots + i;
31db9f7c
AB		 1298 }
1299
1300 }
1301
1302 if (cur_clone_root) {
1303 *found = cur_clone_root;
1304 ret = 0;
1305 } else {
1306 ret = -ENOENT;
1307 }
1308
1309out:
1310 btrfs_free_path(tmp_path);
35075bb0 1311 kfree(backref_ctx);
31db9f7c
AB		 1312 return ret;
1313}
1314
924794c9 1315static int read_symlink(struct btrfs_root *root,
31db9f7c
AB		 1316 u64 ino,
1317 struct fs_path *dest)
1318{
1319 int ret;
1320 struct btrfs_path *path;
1321 struct btrfs_key key;
1322 struct btrfs_file_extent_item *ei;
1323 u8 type;
1324 u8 compression;
1325 unsigned long off;
1326 int len;
1327
1328 path = alloc_path_for_send();
1329 if (!path)
1330 return -ENOMEM;
1331
1332 key.objectid = ino;
1333 key.type = BTRFS_EXTENT_DATA_KEY;
1334 key.offset = 0;
1335 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1336 if (ret < 0)
1337 goto out;
1338 BUG_ON(ret);
1339
1340 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1341 struct btrfs_file_extent_item);
1342 type = btrfs_file_extent_type(path->nodes[0], ei);
1343 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1344 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1345 BUG_ON(compression);
1346
1347 off = btrfs_file_extent_inline_start(ei);
1348 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1349
1350 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
31db9f7c
AB		 1351
1352out:
1353 btrfs_free_path(path);
1354 return ret;
1355}
1356
1357/*
1358 * Helper function to generate a file name that is unique in the root of
1359 * send_root and parent_root. This is used to generate names for orphan inodes.
1360 */
1361static int gen_unique_name(struct send_ctx *sctx,
1362 u64 ino, u64 gen,
1363 struct fs_path *dest)
1364{
1365 int ret = 0;
1366 struct btrfs_path *path;
1367 struct btrfs_dir_item *di;
1368 char tmp[64];
1369 int len;
1370 u64 idx = 0;
1371
1372 path = alloc_path_for_send();
1373 if (!path)
1374 return -ENOMEM;
1375
1376 while (1) {
1377 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1378 ino, gen, idx);
1379 if (len >= sizeof(tmp)) {
1380 /* should really not happen */
1381 ret = -EOVERFLOW;
1382 goto out;
1383 }
1384
1385 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1386 path, BTRFS_FIRST_FREE_OBJECTID,
1387 tmp, strlen(tmp), 0);
1388 btrfs_release_path(path);
1389 if (IS_ERR(di)) {
1390 ret = PTR_ERR(di);
1391 goto out;
1392 }
1393 if (di) {
1394 /* not unique, try again */
1395 idx++;
1396 continue;
1397 }
1398
1399 if (!sctx->parent_root) {
1400 /* unique */
1401 ret = 0;
1402 break;
1403 }
1404
1405 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1406 path, BTRFS_FIRST_FREE_OBJECTID,
1407 tmp, strlen(tmp), 0);
1408 btrfs_release_path(path);
1409 if (IS_ERR(di)) {
1410 ret = PTR_ERR(di);
1411 goto out;
1412 }
1413 if (di) {
1414 /* not unique, try again */
1415 idx++;
1416 continue;
1417 }
1418 /* unique */
1419 break;
1420 }
1421
1422 ret = fs_path_add(dest, tmp, strlen(tmp));
1423
1424out:
1425 btrfs_free_path(path);
1426 return ret;
1427}
1428
1429enum inode_state {
1430 inode_state_no_change,
1431 inode_state_will_create,
1432 inode_state_did_create,
1433 inode_state_will_delete,
1434 inode_state_did_delete,
1435};
1436
1437static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1438{
1439 int ret;
1440 int left_ret;
1441 int right_ret;
1442 u64 left_gen;
1443 u64 right_gen;
1444
1445 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
85a7b33b 1446 NULL, NULL);
31db9f7c
AB		 1447 if (ret < 0 && ret != -ENOENT)
1448 goto out;
1449 left_ret = ret;
1450
1451 if (!sctx->parent_root) {
1452 right_ret = -ENOENT;
1453 } else {
1454 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
85a7b33b 1455 NULL, NULL, NULL, NULL);
31db9f7c
AB		 1456 if (ret < 0 && ret != -ENOENT)
1457 goto out;
1458 right_ret = ret;
1459 }
1460
1461 if (!left_ret && !right_ret) {
e938c8ad 1462 if (left_gen == gen && right_gen == gen) {
31db9f7c 1463 ret = inode_state_no_change;
e938c8ad 1464 } else if (left_gen == gen) {
31db9f7c
AB		 1465 if (ino < sctx->send_progress)
1466 ret = inode_state_did_create;
1467 else
1468 ret = inode_state_will_create;
1469 } else if (right_gen == gen) {
1470 if (ino < sctx->send_progress)
1471 ret = inode_state_did_delete;
1472 else
1473 ret = inode_state_will_delete;
1474 } else {
1475 ret = -ENOENT;
1476 }
1477 } else if (!left_ret) {
1478 if (left_gen == gen) {
1479 if (ino < sctx->send_progress)
1480 ret = inode_state_did_create;
1481 else
1482 ret = inode_state_will_create;
1483 } else {
1484 ret = -ENOENT;
1485 }
1486 } else if (!right_ret) {
1487 if (right_gen == gen) {
1488 if (ino < sctx->send_progress)
1489 ret = inode_state_did_delete;
1490 else
1491 ret = inode_state_will_delete;
1492 } else {
1493 ret = -ENOENT;
1494 }
1495 } else {
1496 ret = -ENOENT;
1497 }
1498
1499out:
1500 return ret;
1501}
1502
1503static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1504{
1505 int ret;
1506
1507 ret = get_cur_inode_state(sctx, ino, gen);
1508 if (ret < 0)
1509 goto out;
1510
1511 if (ret == inode_state_no_change ||
1512 ret == inode_state_did_create ||
1513 ret == inode_state_will_delete)
1514 ret = 1;
1515 else
1516 ret = 0;
1517
1518out:
1519 return ret;
1520}
1521
1522/*
1523 * Helper function to lookup a dir item in a dir.
1524 */
1525static int lookup_dir_item_inode(struct btrfs_root *root,
1526 u64 dir, const char *name, int name_len,
1527 u64 *found_inode,
1528 u8 *found_type)
1529{
1530 int ret = 0;
1531 struct btrfs_dir_item *di;
1532 struct btrfs_key key;
1533 struct btrfs_path *path;
1534
1535 path = alloc_path_for_send();
1536 if (!path)
1537 return -ENOMEM;
1538
1539 di = btrfs_lookup_dir_item(NULL, root, path,
1540 dir, name, name_len, 0);
1541 if (!di) {
1542 ret = -ENOENT;
1543 goto out;
1544 }
1545 if (IS_ERR(di)) {
1546 ret = PTR_ERR(di);
1547 goto out;
1548 }
1549 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1550 *found_inode = key.objectid;
1551 *found_type = btrfs_dir_type(path->nodes[0], di);
1552
1553out:
1554 btrfs_free_path(path);
1555 return ret;
1556}
1557
766702ef
AB		 1558/*
1559 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1560 * generation of the parent dir and the name of the dir entry.
1561 */
924794c9 1562static int get_first_ref(struct btrfs_root *root, u64 ino,
31db9f7c
AB		 1563 u64 *dir, u64 *dir_gen, struct fs_path *name)
1564{
1565 int ret;
1566 struct btrfs_key key;
1567 struct btrfs_key found_key;
1568 struct btrfs_path *path;
31db9f7c 1569 int len;
96b5bd77 1570 u64 parent_dir;
31db9f7c
AB		 1571
1572 path = alloc_path_for_send();
1573 if (!path)
1574 return -ENOMEM;
1575
1576 key.objectid = ino;
1577 key.type = BTRFS_INODE_REF_KEY;
1578 key.offset = 0;
1579
1580 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1581 if (ret < 0)
1582 goto out;
1583 if (!ret)
1584 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1585 path->slots[0]);
96b5bd77
JS		 1586 if (ret || found_key.objectid != ino ||
1587 (found_key.type != BTRFS_INODE_REF_KEY &&
1588 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1589 ret = -ENOENT;
1590 goto out;
1591 }
1592
96b5bd77
JS		 1593 if (key.type == BTRFS_INODE_REF_KEY) {
1594 struct btrfs_inode_ref *iref;
1595 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1596 struct btrfs_inode_ref);
1597 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1598 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1599 (unsigned long)(iref + 1),
1600 len);
1601 parent_dir = found_key.offset;
1602 } else {
1603 struct btrfs_inode_extref *extref;
1604 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1605 struct btrfs_inode_extref);
1606 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1607 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1608 (unsigned long)&extref->name, len);
1609 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1610 }
31db9f7c
AB		 1611 if (ret < 0)
1612 goto out;
1613 btrfs_release_path(path);
1614
96b5bd77 1615 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
85a7b33b 1616 NULL, NULL);
31db9f7c
AB		 1617 if (ret < 0)
1618 goto out;
1619
96b5bd77 1620 *dir = parent_dir;
31db9f7c
AB		 1621
1622out:
1623 btrfs_free_path(path);
1624 return ret;
1625}
1626
924794c9 1627static int is_first_ref(struct btrfs_root *root,
31db9f7c
AB		 1628 u64 ino, u64 dir,
1629 const char *name, int name_len)
1630{
1631 int ret;
1632 struct fs_path *tmp_name;
1633 u64 tmp_dir;
1634 u64 tmp_dir_gen;
1635
924794c9 1636 tmp_name = fs_path_alloc();
31db9f7c
AB		 1637 if (!tmp_name)
1638 return -ENOMEM;
1639
924794c9 1640 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
31db9f7c
AB		 1641 if (ret < 0)
1642 goto out;
1643
b9291aff 1644 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
31db9f7c
AB		 1645 ret = 0;
1646 goto out;
1647 }
1648
e938c8ad 1649 ret = !memcmp(tmp_name->start, name, name_len);
31db9f7c
AB		 1650
1651out:
924794c9 1652 fs_path_free(tmp_name);
31db9f7c
AB		 1653 return ret;
1654}
1655
766702ef
AB		 1656/*
1657 * Used by process_recorded_refs to determine if a new ref would overwrite an
1658 * already existing ref. In case it detects an overwrite, it returns the
1659 * inode/gen in who_ino/who_gen.
1660 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1661 * to make sure later references to the overwritten inode are possible.
1662 * Orphanizing is however only required for the first ref of an inode.
1663 * process_recorded_refs does an additional is_first_ref check to see if
1664 * orphanizing is really required.
1665 */
31db9f7c
AB		 1666static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1667 const char *name, int name_len,
1668 u64 *who_ino, u64 *who_gen)
1669{
1670 int ret = 0;
ebdad913 1671 u64 gen;
31db9f7c
AB		 1672 u64 other_inode = 0;
1673 u8 other_type = 0;
1674
1675 if (!sctx->parent_root)
1676 goto out;
1677
1678 ret = is_inode_existent(sctx, dir, dir_gen);
1679 if (ret <= 0)
1680 goto out;
1681
ebdad913
JB		 1682 /*
1683 * If we have a parent root we need to verify that the parent dir was
1684 * not delted and then re-created, if it was then we have no overwrite
1685 * and we can just unlink this entry.
1686 */
1687 if (sctx->parent_root) {
1688 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1689 NULL, NULL, NULL);
1690 if (ret < 0 && ret != -ENOENT)
1691 goto out;
1692 if (ret) {
1693 ret = 0;
1694 goto out;
1695 }
1696 if (gen != dir_gen)
1697 goto out;
1698 }
1699
31db9f7c
AB		 1700 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1701 &other_inode, &other_type);
1702 if (ret < 0 && ret != -ENOENT)
1703 goto out;
1704 if (ret) {
1705 ret = 0;
1706 goto out;
1707 }
1708
766702ef
AB		 1709 /*
1710 * Check if the overwritten ref was already processed. If yes, the ref
1711 * was already unlinked/moved, so we can safely assume that we will not
1712 * overwrite anything at this point in time.
1713 */
31db9f7c
AB		 1714 if (other_inode > sctx->send_progress) {
1715 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
85a7b33b 1716 who_gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 1717 if (ret < 0)
1718 goto out;
1719
1720 ret = 1;
1721 *who_ino = other_inode;
1722 } else {
1723 ret = 0;
1724 }
1725
1726out:
1727 return ret;
1728}
1729
766702ef
AB		 1730/*
1731 * Checks if the ref was overwritten by an already processed inode. This is
1732 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1733 * thus the orphan name needs be used.
1734 * process_recorded_refs also uses it to avoid unlinking of refs that were
1735 * overwritten.
1736 */
31db9f7c
AB		 1737static int did_overwrite_ref(struct send_ctx *sctx,
1738 u64 dir, u64 dir_gen,
1739 u64 ino, u64 ino_gen,
1740 const char *name, int name_len)
1741{
1742 int ret = 0;
1743 u64 gen;
1744 u64 ow_inode;
1745 u8 other_type;
1746
1747 if (!sctx->parent_root)
1748 goto out;
1749
1750 ret = is_inode_existent(sctx, dir, dir_gen);
1751 if (ret <= 0)
1752 goto out;
1753
1754 /* check if the ref was overwritten by another ref */
1755 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1756 &ow_inode, &other_type);
1757 if (ret < 0 && ret != -ENOENT)
1758 goto out;
1759 if (ret) {
1760 /* was never and will never be overwritten */
1761 ret = 0;
1762 goto out;
1763 }
1764
1765 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
85a7b33b 1766 NULL, NULL);
31db9f7c
AB		 1767 if (ret < 0)
1768 goto out;
1769
1770 if (ow_inode == ino && gen == ino_gen) {
1771 ret = 0;
1772 goto out;
1773 }
1774
1775 /* we know that it is or will be overwritten. check this now */
1776 if (ow_inode < sctx->send_progress)
1777 ret = 1;
1778 else
1779 ret = 0;
1780
1781out:
1782 return ret;
1783}
1784
766702ef
AB		 1785/*
1786 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1787 * that got overwritten. This is used by process_recorded_refs to determine
1788 * if it has to use the path as returned by get_cur_path or the orphan name.
1789 */
31db9f7c
AB		 1790static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1791{
1792 int ret = 0;
1793 struct fs_path *name = NULL;
1794 u64 dir;
1795 u64 dir_gen;
1796
1797 if (!sctx->parent_root)
1798 goto out;
1799
924794c9 1800 name = fs_path_alloc();
31db9f7c
AB		 1801 if (!name)
1802 return -ENOMEM;
1803
924794c9 1804 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
31db9f7c
AB		 1805 if (ret < 0)
1806 goto out;
1807
1808 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1809 name->start, fs_path_len(name));
31db9f7c
AB		 1810
1811out:
924794c9 1812 fs_path_free(name);
31db9f7c
AB		 1813 return ret;
1814}
1815
766702ef
AB		 1816/*
1817 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1818 * so we need to do some special handling in case we have clashes. This function
1819 * takes care of this with the help of name_cache_entry::radix_list.
5dc67d0b 1820 * In case of error, nce is kfreed.
766702ef 1821 */
31db9f7c
AB		 1822static int name_cache_insert(struct send_ctx *sctx,
1823 struct name_cache_entry *nce)
1824{
1825 int ret = 0;
7e0926fe
AB		 1826 struct list_head *nce_head;
1827
1828 nce_head = radix_tree_lookup(&sctx->name_cache,
1829 (unsigned long)nce->ino);
1830 if (!nce_head) {
1831 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
cfa7a9cc
TI		 1832 if (!nce_head) {
1833 kfree(nce);
31db9f7c 1834 return -ENOMEM;
cfa7a9cc 1835 }
7e0926fe 1836 INIT_LIST_HEAD(nce_head);
31db9f7c 1837
7e0926fe 1838 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
5dc67d0b
AB		 1839 if (ret < 0) {
1840 kfree(nce_head);
1841 kfree(nce);
31db9f7c 1842 return ret;
5dc67d0b 1843 }
31db9f7c 1844 }
7e0926fe 1845 list_add_tail(&nce->radix_list, nce_head);
31db9f7c
AB		 1846 list_add_tail(&nce->list, &sctx->name_cache_list);
1847 sctx->name_cache_size++;
1848
1849 return ret;
1850}
1851
1852static void name_cache_delete(struct send_ctx *sctx,
1853 struct name_cache_entry *nce)
1854{
7e0926fe 1855 struct list_head *nce_head;
31db9f7c 1856
7e0926fe
AB		 1857 nce_head = radix_tree_lookup(&sctx->name_cache,
1858 (unsigned long)nce->ino);
1859 BUG_ON(!nce_head);
31db9f7c 1860
7e0926fe 1861 list_del(&nce->radix_list);
31db9f7c 1862 list_del(&nce->list);
31db9f7c 1863 sctx->name_cache_size--;
7e0926fe
AB		 1864
1865 if (list_empty(nce_head)) {
1866 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1867 kfree(nce_head);
1868 }
31db9f7c
AB		 1869}
1870
1871static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1872 u64 ino, u64 gen)
1873{
7e0926fe
AB		 1874 struct list_head *nce_head;
1875 struct name_cache_entry *cur;
31db9f7c 1876
7e0926fe
AB		 1877 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1878 if (!nce_head)
31db9f7c
AB		 1879 return NULL;
1880
7e0926fe
AB		 1881 list_for_each_entry(cur, nce_head, radix_list) {
1882 if (cur->ino == ino && cur->gen == gen)
1883 return cur;
1884 }
31db9f7c
AB		 1885 return NULL;
1886}
1887
766702ef
AB		 1888/*
1889 * Removes the entry from the list and adds it back to the end. This marks the
1890 * entry as recently used so that name_cache_clean_unused does not remove it.
1891 */
31db9f7c
AB		 1892static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1893{
1894 list_del(&nce->list);
1895 list_add_tail(&nce->list, &sctx->name_cache_list);
1896}
1897
766702ef
AB		 1898/*
1899 * Remove some entries from the beginning of name_cache_list.
1900 */
31db9f7c
AB		 1901static void name_cache_clean_unused(struct send_ctx *sctx)
1902{
1903 struct name_cache_entry *nce;
1904
1905 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1906 return;
1907
1908 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1909 nce = list_entry(sctx->name_cache_list.next,
1910 struct name_cache_entry, list);
1911 name_cache_delete(sctx, nce);
1912 kfree(nce);
1913 }
1914}
1915
1916static void name_cache_free(struct send_ctx *sctx)
1917{
1918 struct name_cache_entry *nce;
31db9f7c 1919
e938c8ad
AB		 1920 while (!list_empty(&sctx->name_cache_list)) {
1921 nce = list_entry(sctx->name_cache_list.next,
1922 struct name_cache_entry, list);
31db9f7c 1923 name_cache_delete(sctx, nce);
17589bd9 1924 kfree(nce);
31db9f7c
AB		 1925 }
1926}
1927
766702ef
AB		 1928/*
1929 * Used by get_cur_path for each ref up to the root.
1930 * Returns 0 if it succeeded.
1931 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1932 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1933 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1934 * Returns <0 in case of error.
1935 */
31db9f7c
AB		 1936static int __get_cur_name_and_parent(struct send_ctx *sctx,
1937 u64 ino, u64 gen,
1938 u64 *parent_ino,
1939 u64 *parent_gen,
1940 struct fs_path *dest)
1941{
1942 int ret;
1943 int nce_ret;
1944 struct btrfs_path *path = NULL;
1945 struct name_cache_entry *nce = NULL;
1946
766702ef
AB		 1947 /*
1948 * First check if we already did a call to this function with the same
1949 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1950 * return the cached result.
1951 */
31db9f7c
AB		 1952 nce = name_cache_search(sctx, ino, gen);
1953 if (nce) {
1954 if (ino < sctx->send_progress && nce->need_later_update) {
1955 name_cache_delete(sctx, nce);
1956 kfree(nce);
1957 nce = NULL;
1958 } else {
1959 name_cache_used(sctx, nce);
1960 *parent_ino = nce->parent_ino;
1961 *parent_gen = nce->parent_gen;
1962 ret = fs_path_add(dest, nce->name, nce->name_len);
1963 if (ret < 0)
1964 goto out;
1965 ret = nce->ret;
1966 goto out;
1967 }
1968 }
1969
1970 path = alloc_path_for_send();
1971 if (!path)
1972 return -ENOMEM;
1973
766702ef
AB		 1974 /*
1975 * If the inode is not existent yet, add the orphan name and return 1.
1976 * This should only happen for the parent dir that we determine in
1977 * __record_new_ref
1978 */
31db9f7c
AB		 1979 ret = is_inode_existent(sctx, ino, gen);
1980 if (ret < 0)
1981 goto out;
1982
1983 if (!ret) {
1984 ret = gen_unique_name(sctx, ino, gen, dest);
1985 if (ret < 0)
1986 goto out;
1987 ret = 1;
1988 goto out_cache;
1989 }
1990
766702ef
AB		 1991 /*
1992 * Depending on whether the inode was already processed or not, use
1993 * send_root or parent_root for ref lookup.
1994 */
31db9f7c 1995 if (ino < sctx->send_progress)
924794c9
TI		 1996 ret = get_first_ref(sctx->send_root, ino,
1997 parent_ino, parent_gen, dest);
31db9f7c 1998 else
924794c9
TI		 1999 ret = get_first_ref(sctx->parent_root, ino,
2000 parent_ino, parent_gen, dest);
31db9f7c
AB		 2001 if (ret < 0)
2002 goto out;
2003
766702ef
AB		 2004 /*
2005 * Check if the ref was overwritten by an inode's ref that was processed
2006 * earlier. If yes, treat as orphan and return 1.
2007 */
31db9f7c
AB		 2008 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2009 dest->start, dest->end - dest->start);
2010 if (ret < 0)
2011 goto out;
2012 if (ret) {
2013 fs_path_reset(dest);
2014 ret = gen_unique_name(sctx, ino, gen, dest);
2015 if (ret < 0)
2016 goto out;
2017 ret = 1;
2018 }
2019
2020out_cache:
766702ef
AB		 2021 /*
2022 * Store the result of the lookup in the name cache.
2023 */
31db9f7c
AB		 2024 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2025 if (!nce) {
2026 ret = -ENOMEM;
2027 goto out;
2028 }
2029
2030 nce->ino = ino;
2031 nce->gen = gen;
2032 nce->parent_ino = *parent_ino;
2033 nce->parent_gen = *parent_gen;
2034 nce->name_len = fs_path_len(dest);
2035 nce->ret = ret;
2036 strcpy(nce->name, dest->start);
31db9f7c
AB		 2037
2038 if (ino < sctx->send_progress)
2039 nce->need_later_update = 0;
2040 else
2041 nce->need_later_update = 1;
2042
2043 nce_ret = name_cache_insert(sctx, nce);
2044 if (nce_ret < 0)
2045 ret = nce_ret;
2046 name_cache_clean_unused(sctx);
2047
2048out:
2049 btrfs_free_path(path);
2050 return ret;
2051}
2052
2053/*
2054 * Magic happens here. This function returns the first ref to an inode as it
2055 * would look like while receiving the stream at this point in time.
2056 * We walk the path up to the root. For every inode in between, we check if it
2057 * was already processed/sent. If yes, we continue with the parent as found
2058 * in send_root. If not, we continue with the parent as found in parent_root.
2059 * If we encounter an inode that was deleted at this point in time, we use the
2060 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2061 * that were not created yet and overwritten inodes/refs.
2062 *
2063 * When do we have have orphan inodes:
2064 * 1. When an inode is freshly created and thus no valid refs are available yet
2065 * 2. When a directory lost all it's refs (deleted) but still has dir items
2066 * inside which were not processed yet (pending for move/delete). If anyone
2067 * tried to get the path to the dir items, it would get a path inside that
2068 * orphan directory.
2069 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2070 * of an unprocessed inode. If in that case the first ref would be
2071 * overwritten, the overwritten inode gets "orphanized". Later when we
2072 * process this overwritten inode, it is restored at a new place by moving
2073 * the orphan inode.
2074 *
2075 * sctx->send_progress tells this function at which point in time receiving
2076 * would be.
2077 */
2078static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2079 struct fs_path *dest)
2080{
2081 int ret = 0;
2082 struct fs_path *name = NULL;
2083 u64 parent_inode = 0;
2084 u64 parent_gen = 0;
2085 int stop = 0;
2086
924794c9 2087 name = fs_path_alloc();
31db9f7c
AB		 2088 if (!name) {
2089 ret = -ENOMEM;
2090 goto out;
2091 }
2092
2093 dest->reversed = 1;
2094 fs_path_reset(dest);
2095
2096 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2097 fs_path_reset(name);
2098
2099 ret = __get_cur_name_and_parent(sctx, ino, gen,
2100 &parent_inode, &parent_gen, name);
2101 if (ret < 0)
2102 goto out;
2103 if (ret)
2104 stop = 1;
2105
2106 ret = fs_path_add_path(dest, name);
2107 if (ret < 0)
2108 goto out;
2109
2110 ino = parent_inode;
2111 gen = parent_gen;
2112 }
2113
2114out:
924794c9 2115 fs_path_free(name);
31db9f7c
AB		 2116 if (!ret)
2117 fs_path_unreverse(dest);
2118 return ret;
2119}
2120
31db9f7c
AB		 2121/*
2122 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2123 */
2124static int send_subvol_begin(struct send_ctx *sctx)
2125{
2126 int ret;
2127 struct btrfs_root *send_root = sctx->send_root;
2128 struct btrfs_root *parent_root = sctx->parent_root;
2129 struct btrfs_path *path;
2130 struct btrfs_key key;
2131 struct btrfs_root_ref *ref;
2132 struct extent_buffer *leaf;
2133 char *name = NULL;
2134 int namelen;
2135
2136 path = alloc_path_for_send();
2137 if (!path)
2138 return -ENOMEM;
2139
2140 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2141 if (!name) {
2142 btrfs_free_path(path);
2143 return -ENOMEM;
2144 }
2145
2146 key.objectid = send_root->objectid;
2147 key.type = BTRFS_ROOT_BACKREF_KEY;
2148 key.offset = 0;
2149
2150 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2151 &key, path, 1, 0);
2152 if (ret < 0)
2153 goto out;
2154 if (ret) {
2155 ret = -ENOENT;
2156 goto out;
2157 }
2158
2159 leaf = path->nodes[0];
2160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2161 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2162 key.objectid != send_root->objectid) {
2163 ret = -ENOENT;
2164 goto out;
2165 }
2166 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2167 namelen = btrfs_root_ref_name_len(leaf, ref);
2168 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2169 btrfs_release_path(path);
2170
31db9f7c
AB		 2171 if (parent_root) {
2172 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2173 if (ret < 0)
2174 goto out;
2175 } else {
2176 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2177 if (ret < 0)
2178 goto out;
2179 }
2180
2181 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2182 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2183 sctx->send_root->root_item.uuid);
2184 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2185 sctx->send_root->root_item.ctransid);
2186 if (parent_root) {
2187 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2188 sctx->parent_root->root_item.uuid);
2189 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2190 sctx->parent_root->root_item.ctransid);
2191 }
2192
2193 ret = send_cmd(sctx);
2194
2195tlv_put_failure:
2196out:
2197 btrfs_free_path(path);
2198 kfree(name);
2199 return ret;
2200}
2201
2202static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2203{
2204 int ret = 0;
2205 struct fs_path *p;
2206
2207verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2208
924794c9 2209 p = fs_path_alloc();
31db9f7c
AB		 2210 if (!p)
2211 return -ENOMEM;
2212
2213 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2214 if (ret < 0)
2215 goto out;
2216
2217 ret = get_cur_path(sctx, ino, gen, p);
2218 if (ret < 0)
2219 goto out;
2220 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2221 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2222
2223 ret = send_cmd(sctx);
2224
2225tlv_put_failure:
2226out:
924794c9 2227 fs_path_free(p);
31db9f7c
AB		 2228 return ret;
2229}
2230
2231static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2232{
2233 int ret = 0;
2234 struct fs_path *p;
2235
2236verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2237
924794c9 2238 p = fs_path_alloc();
31db9f7c
AB		 2239 if (!p)
2240 return -ENOMEM;
2241
2242 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2243 if (ret < 0)
2244 goto out;
2245
2246 ret = get_cur_path(sctx, ino, gen, p);
2247 if (ret < 0)
2248 goto out;
2249 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2250 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2251
2252 ret = send_cmd(sctx);
2253
2254tlv_put_failure:
2255out:
924794c9 2256 fs_path_free(p);
31db9f7c
AB		 2257 return ret;
2258}
2259
2260static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2261{
2262 int ret = 0;
2263 struct fs_path *p;
2264
2265verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2266
924794c9 2267 p = fs_path_alloc();
31db9f7c
AB		 2268 if (!p)
2269 return -ENOMEM;
2270
2271 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2272 if (ret < 0)
2273 goto out;
2274
2275 ret = get_cur_path(sctx, ino, gen, p);
2276 if (ret < 0)
2277 goto out;
2278 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2279 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2280 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2281
2282 ret = send_cmd(sctx);
2283
2284tlv_put_failure:
2285out:
924794c9 2286 fs_path_free(p);
31db9f7c
AB		 2287 return ret;
2288}
2289
2290static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2291{
2292 int ret = 0;
2293 struct fs_path *p = NULL;
2294 struct btrfs_inode_item *ii;
2295 struct btrfs_path *path = NULL;
2296 struct extent_buffer *eb;
2297 struct btrfs_key key;
2298 int slot;
2299
2300verbose_printk("btrfs: send_utimes %llu\n", ino);
2301
924794c9 2302 p = fs_path_alloc();
31db9f7c
AB		 2303 if (!p)
2304 return -ENOMEM;
2305
2306 path = alloc_path_for_send();
2307 if (!path) {
2308 ret = -ENOMEM;
2309 goto out;
2310 }
2311
2312 key.objectid = ino;
2313 key.type = BTRFS_INODE_ITEM_KEY;
2314 key.offset = 0;
2315 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2316 if (ret < 0)
2317 goto out;
2318
2319 eb = path->nodes[0];
2320 slot = path->slots[0];
2321 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2322
2323 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2324 if (ret < 0)
2325 goto out;
2326
2327 ret = get_cur_path(sctx, ino, gen, p);
2328 if (ret < 0)
2329 goto out;
2330 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2331 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2332 btrfs_inode_atime(ii));
2333 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2334 btrfs_inode_mtime(ii));
2335 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2336 btrfs_inode_ctime(ii));
766702ef 2337 /* TODO Add otime support when the otime patches get into upstream */
31db9f7c
AB		 2338
2339 ret = send_cmd(sctx);
2340
2341tlv_put_failure:
2342out:
924794c9 2343 fs_path_free(p);
31db9f7c
AB		 2344 btrfs_free_path(path);
2345 return ret;
2346}
2347
2348/*
2349 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2350 * a valid path yet because we did not process the refs yet. So, the inode
2351 * is created as orphan.
2352 */
1f4692da 2353static int send_create_inode(struct send_ctx *sctx, u64 ino)
31db9f7c
AB		 2354{
2355 int ret = 0;
31db9f7c 2356 struct fs_path *p;
31db9f7c 2357 int cmd;
1f4692da 2358 u64 gen;
31db9f7c 2359 u64 mode;
1f4692da 2360 u64 rdev;
31db9f7c 2361
1f4692da 2362verbose_printk("btrfs: send_create_inode %llu\n", ino);
31db9f7c 2363
924794c9 2364 p = fs_path_alloc();
31db9f7c
AB		 2365 if (!p)
2366 return -ENOMEM;
2367
1f4692da
AB		 2368 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2369 NULL, &rdev);
2370 if (ret < 0)
2371 goto out;
31db9f7c 2372
e938c8ad 2373 if (S_ISREG(mode)) {
31db9f7c 2374 cmd = BTRFS_SEND_C_MKFILE;
e938c8ad 2375 } else if (S_ISDIR(mode)) {
31db9f7c 2376 cmd = BTRFS_SEND_C_MKDIR;
e938c8ad 2377 } else if (S_ISLNK(mode)) {
31db9f7c 2378 cmd = BTRFS_SEND_C_SYMLINK;
e938c8ad 2379 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
31db9f7c 2380 cmd = BTRFS_SEND_C_MKNOD;
e938c8ad 2381 } else if (S_ISFIFO(mode)) {
31db9f7c 2382 cmd = BTRFS_SEND_C_MKFIFO;
e938c8ad 2383 } else if (S_ISSOCK(mode)) {
31db9f7c 2384 cmd = BTRFS_SEND_C_MKSOCK;
e938c8ad 2385 } else {
31db9f7c
AB		 2386 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2387 (int)(mode & S_IFMT));
2388 ret = -ENOTSUPP;
2389 goto out;
2390 }
2391
2392 ret = begin_cmd(sctx, cmd);
2393 if (ret < 0)
2394 goto out;
2395
1f4692da 2396 ret = gen_unique_name(sctx, ino, gen, p);
31db9f7c
AB		 2397 if (ret < 0)
2398 goto out;
2399
2400 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
1f4692da 2401 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
31db9f7c
AB		 2402
2403 if (S_ISLNK(mode)) {
2404 fs_path_reset(p);
924794c9 2405 ret = read_symlink(sctx->send_root, ino, p);
31db9f7c
AB		 2406 if (ret < 0)
2407 goto out;
2408 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2409 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2410 S_ISFIFO(mode) || S_ISSOCK(mode)) {
d79e5043
AJ		 2411 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2412 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
31db9f7c
AB		 2413 }
2414
2415 ret = send_cmd(sctx);
2416 if (ret < 0)
2417 goto out;
2418
2419
2420tlv_put_failure:
2421out:
924794c9 2422 fs_path_free(p);
31db9f7c
AB		 2423 return ret;
2424}
2425
1f4692da
AB		 2426/*
2427 * We need some special handling for inodes that get processed before the parent
2428 * directory got created. See process_recorded_refs for details.
2429 * This function does the check if we already created the dir out of order.
2430 */
2431static int did_create_dir(struct send_ctx *sctx, u64 dir)
2432{
2433 int ret = 0;
2434 struct btrfs_path *path = NULL;
2435 struct btrfs_key key;
2436 struct btrfs_key found_key;
2437 struct btrfs_key di_key;
2438 struct extent_buffer *eb;
2439 struct btrfs_dir_item *di;
2440 int slot;
2441
2442 path = alloc_path_for_send();
2443 if (!path) {
2444 ret = -ENOMEM;
2445 goto out;
2446 }
2447
2448 key.objectid = dir;
2449 key.type = BTRFS_DIR_INDEX_KEY;
2450 key.offset = 0;
2451 while (1) {
2452 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2453 1, 0);
2454 if (ret < 0)
2455 goto out;
2456 if (!ret) {
2457 eb = path->nodes[0];
2458 slot = path->slots[0];
2459 btrfs_item_key_to_cpu(eb, &found_key, slot);
2460 }
2461 if (ret || found_key.objectid != key.objectid ||
2462 found_key.type != key.type) {
2463 ret = 0;
2464 goto out;
2465 }
2466
2467 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2468 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2469
a0525414
JB		 2470 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2471 di_key.objectid < sctx->send_progress) {
1f4692da
AB		 2472 ret = 1;
2473 goto out;
2474 }
2475
2476 key.offset = found_key.offset + 1;
2477 btrfs_release_path(path);
2478 }
2479
2480out:
2481 btrfs_free_path(path);
2482 return ret;
2483}
2484
2485/*
2486 * Only creates the inode if it is:
2487 * 1. Not a directory
2488 * 2. Or a directory which was not created already due to out of order
2489 * directories. See did_create_dir and process_recorded_refs for details.
2490 */
2491static int send_create_inode_if_needed(struct send_ctx *sctx)
2492{
2493 int ret;
2494
2495 if (S_ISDIR(sctx->cur_inode_mode)) {
2496 ret = did_create_dir(sctx, sctx->cur_ino);
2497 if (ret < 0)
2498 goto out;
2499 if (ret) {
2500 ret = 0;
2501 goto out;
2502 }
2503 }
2504
2505 ret = send_create_inode(sctx, sctx->cur_ino);
2506 if (ret < 0)
2507 goto out;
2508
2509out:
2510 return ret;
2511}
2512
31db9f7c
AB		 2513struct recorded_ref {
2514 struct list_head list;
2515 char *dir_path;
2516 char *name;
2517 struct fs_path *full_path;
2518 u64 dir;
2519 u64 dir_gen;
2520 int dir_path_len;
2521 int name_len;
2522};
2523
2524/*
2525 * We need to process new refs before deleted refs, but compare_tree gives us
2526 * everything mixed. So we first record all refs and later process them.
2527 * This function is a helper to record one ref.
2528 */
2529static int record_ref(struct list_head *head, u64 dir,
2530 u64 dir_gen, struct fs_path *path)
2531{
2532 struct recorded_ref *ref;
31db9f7c
AB		 2533
2534 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2535 if (!ref)
2536 return -ENOMEM;
2537
2538 ref->dir = dir;
2539 ref->dir_gen = dir_gen;
2540 ref->full_path = path;
2541
ed84885d
AS		 2542 ref->name = (char *)kbasename(ref->full_path->start);
2543 ref->name_len = ref->full_path->end - ref->name;
2544 ref->dir_path = ref->full_path->start;
2545 if (ref->name == ref->full_path->start)
31db9f7c 2546 ref->dir_path_len = 0;
ed84885d 2547 else
31db9f7c
AB		 2548 ref->dir_path_len = ref->full_path->end -
2549 ref->full_path->start - 1 - ref->name_len;
31db9f7c
AB		 2550
2551 list_add_tail(&ref->list, head);
2552 return 0;
2553}
2554
ba5e8f2e
JB		 2555static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2556{
2557 struct recorded_ref *new;
2558
2559 new = kmalloc(sizeof(*ref), GFP_NOFS);
2560 if (!new)
2561 return -ENOMEM;
2562
2563 new->dir = ref->dir;
2564 new->dir_gen = ref->dir_gen;
2565 new->full_path = NULL;
2566 INIT_LIST_HEAD(&new->list);
2567 list_add_tail(&new->list, list);
2568 return 0;
2569}
2570
924794c9 2571static void __free_recorded_refs(struct list_head *head)
31db9f7c
AB		 2572{
2573 struct recorded_ref *cur;
31db9f7c 2574
e938c8ad
AB		 2575 while (!list_empty(head)) {
2576 cur = list_entry(head->next, struct recorded_ref, list);
924794c9 2577 fs_path_free(cur->full_path);
e938c8ad 2578 list_del(&cur->list);
31db9f7c
AB		 2579 kfree(cur);
2580 }
31db9f7c
AB		 2581}
2582
2583static void free_recorded_refs(struct send_ctx *sctx)
2584{
924794c9
TI		 2585 __free_recorded_refs(&sctx->new_refs);
2586 __free_recorded_refs(&sctx->deleted_refs);
31db9f7c
AB		 2587}
2588
2589/*
766702ef 2590 * Renames/moves a file/dir to its orphan name. Used when the first
31db9f7c
AB		 2591 * ref of an unprocessed inode gets overwritten and for all non empty
2592 * directories.
2593 */
2594static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2595 struct fs_path *path)
2596{
2597 int ret;
2598 struct fs_path *orphan;
2599
924794c9 2600 orphan = fs_path_alloc();
31db9f7c
AB		 2601 if (!orphan)
2602 return -ENOMEM;
2603
2604 ret = gen_unique_name(sctx, ino, gen, orphan);
2605 if (ret < 0)
2606 goto out;
2607
2608 ret = send_rename(sctx, path, orphan);
2609
2610out:
924794c9 2611 fs_path_free(orphan);
31db9f7c
AB		 2612 return ret;
2613}
2614
2615/*
2616 * Returns 1 if a directory can be removed at this point in time.
2617 * We check this by iterating all dir items and checking if the inode behind
2618 * the dir item was already processed.
2619 */
2620static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2621{
2622 int ret = 0;
2623 struct btrfs_root *root = sctx->parent_root;
2624 struct btrfs_path *path;
2625 struct btrfs_key key;
2626 struct btrfs_key found_key;
2627 struct btrfs_key loc;
2628 struct btrfs_dir_item *di;
2629
6d85ed05
AB		 2630 /*
2631 * Don't try to rmdir the top/root subvolume dir.
2632 */
2633 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2634 return 0;
2635
31db9f7c
AB		 2636 path = alloc_path_for_send();
2637 if (!path)
2638 return -ENOMEM;
2639
2640 key.objectid = dir;
2641 key.type = BTRFS_DIR_INDEX_KEY;
2642 key.offset = 0;
2643
2644 while (1) {
2645 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2646 if (ret < 0)
2647 goto out;
2648 if (!ret) {
2649 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2650 path->slots[0]);
2651 }
2652 if (ret || found_key.objectid != key.objectid ||
2653 found_key.type != key.type) {
2654 break;
2655 }
2656
2657 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2658 struct btrfs_dir_item);
2659 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2660
2661 if (loc.objectid > send_progress) {
2662 ret = 0;
2663 goto out;
2664 }
2665
2666 btrfs_release_path(path);
2667 key.offset = found_key.offset + 1;
2668 }
2669
2670 ret = 1;
2671
2672out:
2673 btrfs_free_path(path);
2674 return ret;
2675}
2676
31db9f7c
AB		 2677/*
2678 * This does all the move/link/unlink/rmdir magic.
2679 */
2680static int process_recorded_refs(struct send_ctx *sctx)
2681{
2682 int ret = 0;
2683 struct recorded_ref *cur;
1f4692da 2684 struct recorded_ref *cur2;
ba5e8f2e 2685 struct list_head check_dirs;
31db9f7c 2686 struct fs_path *valid_path = NULL;
b24baf69 2687 u64 ow_inode = 0;
31db9f7c
AB		 2688 u64 ow_gen;
2689 int did_overwrite = 0;
2690 int is_orphan = 0;
2691
2692verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2693
6d85ed05
AB		 2694 /*
2695 * This should never happen as the root dir always has the same ref
2696 * which is always '..'
2697 */
2698 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
ba5e8f2e 2699 INIT_LIST_HEAD(&check_dirs);
6d85ed05 2700
924794c9 2701 valid_path = fs_path_alloc();
31db9f7c
AB		 2702 if (!valid_path) {
2703 ret = -ENOMEM;
2704 goto out;
2705 }
2706
31db9f7c
AB		 2707 /*
2708 * First, check if the first ref of the current inode was overwritten
2709 * before. If yes, we know that the current inode was already orphanized
2710 * and thus use the orphan name. If not, we can use get_cur_path to
2711 * get the path of the first ref as it would like while receiving at
2712 * this point in time.
2713 * New inodes are always orphan at the beginning, so force to use the
2714 * orphan name in this case.
2715 * The first ref is stored in valid_path and will be updated if it
2716 * gets moved around.
2717 */
2718 if (!sctx->cur_inode_new) {
2719 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2720 sctx->cur_inode_gen);
2721 if (ret < 0)
2722 goto out;
2723 if (ret)
2724 did_overwrite = 1;
2725 }
2726 if (sctx->cur_inode_new || did_overwrite) {
2727 ret = gen_unique_name(sctx, sctx->cur_ino,
2728 sctx->cur_inode_gen, valid_path);
2729 if (ret < 0)
2730 goto out;
2731 is_orphan = 1;
2732 } else {
2733 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2734 valid_path);
2735 if (ret < 0)
2736 goto out;
2737 }
2738
2739 list_for_each_entry(cur, &sctx->new_refs, list) {
1f4692da
AB		 2740 /*
2741 * We may have refs where the parent directory does not exist
2742 * yet. This happens if the parent directories inum is higher
2743 * the the current inum. To handle this case, we create the
2744 * parent directory out of order. But we need to check if this
2745 * did already happen before due to other refs in the same dir.
2746 */
2747 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
2748 if (ret < 0)
2749 goto out;
2750 if (ret == inode_state_will_create) {
2751 ret = 0;
2752 /*
2753 * First check if any of the current inodes refs did
2754 * already create the dir.
2755 */
2756 list_for_each_entry(cur2, &sctx->new_refs, list) {
2757 if (cur == cur2)
2758 break;
2759 if (cur2->dir == cur->dir) {
2760 ret = 1;
2761 break;
2762 }
2763 }
2764
2765 /*
2766 * If that did not happen, check if a previous inode
2767 * did already create the dir.
2768 */
2769 if (!ret)
2770 ret = did_create_dir(sctx, cur->dir);
2771 if (ret < 0)
2772 goto out;
2773 if (!ret) {
2774 ret = send_create_inode(sctx, cur->dir);
2775 if (ret < 0)
2776 goto out;
2777 }
2778 }
2779
31db9f7c
AB		 2780 /*
2781 * Check if this new ref would overwrite the first ref of
2782 * another unprocessed inode. If yes, orphanize the
2783 * overwritten inode. If we find an overwritten ref that is
2784 * not the first ref, simply unlink it.
2785 */
2786 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2787 cur->name, cur->name_len,
2788 &ow_inode, &ow_gen);
2789 if (ret < 0)
2790 goto out;
2791 if (ret) {
924794c9
TI		 2792 ret = is_first_ref(sctx->parent_root,
2793 ow_inode, cur->dir, cur->name,
2794 cur->name_len);
31db9f7c
AB		 2795 if (ret < 0)
2796 goto out;
2797 if (ret) {
2798 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2799 cur->full_path);
2800 if (ret < 0)
2801 goto out;
2802 } else {
2803 ret = send_unlink(sctx, cur->full_path);
2804 if (ret < 0)
2805 goto out;
2806 }
2807 }
2808
2809 /*
2810 * link/move the ref to the new place. If we have an orphan
2811 * inode, move it and update valid_path. If not, link or move
2812 * it depending on the inode mode.
2813 */
1f4692da 2814 if (is_orphan) {
31db9f7c
AB		 2815 ret = send_rename(sctx, valid_path, cur->full_path);
2816 if (ret < 0)
2817 goto out;
2818 is_orphan = 0;
2819 ret = fs_path_copy(valid_path, cur->full_path);
2820 if (ret < 0)
2821 goto out;
2822 } else {
2823 if (S_ISDIR(sctx->cur_inode_mode)) {
2824 /*
2825 * Dirs can't be linked, so move it. For moved
2826 * dirs, we always have one new and one deleted
2827 * ref. The deleted ref is ignored later.
2828 */
2829 ret = send_rename(sctx, valid_path,
2830 cur->full_path);
2831 if (ret < 0)
2832 goto out;
2833 ret = fs_path_copy(valid_path, cur->full_path);
2834 if (ret < 0)
2835 goto out;
2836 } else {
2837 ret = send_link(sctx, cur->full_path,
2838 valid_path);
2839 if (ret < 0)
2840 goto out;
2841 }
2842 }
ba5e8f2e 2843 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 2844 if (ret < 0)
2845 goto out;
2846 }
2847
2848 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2849 /*
2850 * Check if we can already rmdir the directory. If not,
2851 * orphanize it. For every dir item inside that gets deleted
2852 * later, we do this check again and rmdir it then if possible.
2853 * See the use of check_dirs for more details.
2854 */
2855 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2856 if (ret < 0)
2857 goto out;
2858 if (ret) {
2859 ret = send_rmdir(sctx, valid_path);
2860 if (ret < 0)
2861 goto out;
2862 } else if (!is_orphan) {
2863 ret = orphanize_inode(sctx, sctx->cur_ino,
2864 sctx->cur_inode_gen, valid_path);
2865 if (ret < 0)
2866 goto out;
2867 is_orphan = 1;
2868 }
2869
2870 list_for_each_entry(cur, &sctx->deleted_refs, list) {
ba5e8f2e 2871 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 2872 if (ret < 0)
2873 goto out;
2874 }
ccf1626b
AB		 2875 } else if (S_ISDIR(sctx->cur_inode_mode) &&
2876 !list_empty(&sctx->deleted_refs)) {
2877 /*
2878 * We have a moved dir. Add the old parent to check_dirs
2879 */
2880 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
2881 list);
ba5e8f2e 2882 ret = dup_ref(cur, &check_dirs);
ccf1626b
AB		 2883 if (ret < 0)
2884 goto out;
31db9f7c
AB		 2885 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2886 /*
2887 * We have a non dir inode. Go through all deleted refs and
2888 * unlink them if they were not already overwritten by other
2889 * inodes.
2890 */
2891 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2892 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2893 sctx->cur_ino, sctx->cur_inode_gen,
2894 cur->name, cur->name_len);
2895 if (ret < 0)
2896 goto out;
2897 if (!ret) {
1f4692da
AB		 2898 ret = send_unlink(sctx, cur->full_path);
2899 if (ret < 0)
2900 goto out;
31db9f7c 2901 }
ba5e8f2e 2902 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 2903 if (ret < 0)
2904 goto out;
2905 }
31db9f7c
AB		 2906 /*
2907 * If the inode is still orphan, unlink the orphan. This may
2908 * happen when a previous inode did overwrite the first ref
2909 * of this inode and no new refs were added for the current
766702ef
AB		 2910 * inode. Unlinking does not mean that the inode is deleted in
2911 * all cases. There may still be links to this inode in other
2912 * places.
31db9f7c 2913 */
1f4692da 2914 if (is_orphan) {
31db9f7c
AB		 2915 ret = send_unlink(sctx, valid_path);
2916 if (ret < 0)
2917 goto out;
2918 }
2919 }
2920
2921 /*
2922 * We did collect all parent dirs where cur_inode was once located. We
2923 * now go through all these dirs and check if they are pending for
2924 * deletion and if it's finally possible to perform the rmdir now.
2925 * We also update the inode stats of the parent dirs here.
2926 */
ba5e8f2e 2927 list_for_each_entry(cur, &check_dirs, list) {
766702ef
AB		 2928 /*
2929 * In case we had refs into dirs that were not processed yet,
2930 * we don't need to do the utime and rmdir logic for these dirs.
2931 * The dir will be processed later.
2932 */
ba5e8f2e 2933 if (cur->dir > sctx->cur_ino)
31db9f7c
AB		 2934 continue;
2935
ba5e8f2e 2936 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 2937 if (ret < 0)
2938 goto out;
2939
2940 if (ret == inode_state_did_create ||
2941 ret == inode_state_no_change) {
2942 /* TODO delayed utimes */
ba5e8f2e 2943 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 2944 if (ret < 0)
2945 goto out;
2946 } else if (ret == inode_state_did_delete) {
ba5e8f2e 2947 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
31db9f7c
AB		 2948 if (ret < 0)
2949 goto out;
2950 if (ret) {
ba5e8f2e
JB		 2951 ret = get_cur_path(sctx, cur->dir,
2952 cur->dir_gen, valid_path);
31db9f7c
AB		 2953 if (ret < 0)
2954 goto out;
2955 ret = send_rmdir(sctx, valid_path);
2956 if (ret < 0)
2957 goto out;
2958 }
2959 }
2960 }
2961
31db9f7c
AB		 2962 ret = 0;
2963
2964out:
ba5e8f2e 2965 __free_recorded_refs(&check_dirs);
31db9f7c 2966 free_recorded_refs(sctx);
924794c9 2967 fs_path_free(valid_path);
31db9f7c
AB		 2968 return ret;
2969}
2970
2971static int __record_new_ref(int num, u64 dir, int index,
2972 struct fs_path *name,
2973 void *ctx)
2974{
2975 int ret = 0;
2976 struct send_ctx *sctx = ctx;
2977 struct fs_path *p;
2978 u64 gen;
2979
924794c9 2980 p = fs_path_alloc();
31db9f7c
AB		 2981 if (!p)
2982 return -ENOMEM;
2983
2984 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 2985 NULL, NULL);
31db9f7c
AB		 2986 if (ret < 0)
2987 goto out;
2988
31db9f7c
AB		 2989 ret = get_cur_path(sctx, dir, gen, p);
2990 if (ret < 0)
2991 goto out;
2992 ret = fs_path_add_path(p, name);
2993 if (ret < 0)
2994 goto out;
2995
2996 ret = record_ref(&sctx->new_refs, dir, gen, p);
2997
2998out:
2999 if (ret)
924794c9 3000 fs_path_free(p);
31db9f7c
AB		 3001 return ret;
3002}
3003
3004static int __record_deleted_ref(int num, u64 dir, int index,
3005 struct fs_path *name,
3006 void *ctx)
3007{
3008 int ret = 0;
3009 struct send_ctx *sctx = ctx;
3010 struct fs_path *p;
3011 u64 gen;
3012
924794c9 3013 p = fs_path_alloc();
31db9f7c
AB		 3014 if (!p)
3015 return -ENOMEM;
3016
3017 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 3018 NULL, NULL);
31db9f7c
AB		 3019 if (ret < 0)
3020 goto out;
3021
3022 ret = get_cur_path(sctx, dir, gen, p);
3023 if (ret < 0)
3024 goto out;
3025 ret = fs_path_add_path(p, name);
3026 if (ret < 0)
3027 goto out;
3028
3029 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3030
3031out:
3032 if (ret)
924794c9 3033 fs_path_free(p);
31db9f7c
AB		 3034 return ret;
3035}
3036
3037static int record_new_ref(struct send_ctx *sctx)
3038{
3039 int ret;
3040
924794c9
TI		 3041 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3042 sctx->cmp_key, 0, __record_new_ref, sctx);
31db9f7c
AB		 3043 if (ret < 0)
3044 goto out;
3045 ret = 0;
3046
3047out:
3048 return ret;
3049}
3050
3051static int record_deleted_ref(struct send_ctx *sctx)
3052{
3053 int ret;
3054
924794c9
TI		 3055 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3056 sctx->cmp_key, 0, __record_deleted_ref, sctx);
31db9f7c
AB		 3057 if (ret < 0)
3058 goto out;
3059 ret = 0;
3060
3061out:
3062 return ret;
3063}
3064
3065struct find_ref_ctx {
3066 u64 dir;
ba5e8f2e
JB		 3067 u64 dir_gen;
3068 struct btrfs_root *root;
31db9f7c
AB		 3069 struct fs_path *name;
3070 int found_idx;
3071};
3072
3073static int __find_iref(int num, u64 dir, int index,
3074 struct fs_path *name,
3075 void *ctx_)
3076{
3077 struct find_ref_ctx *ctx = ctx_;
ba5e8f2e
JB		 3078 u64 dir_gen;
3079 int ret;
31db9f7c
AB		 3080
3081 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3082 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
ba5e8f2e
JB		 3083 /*
3084 * To avoid doing extra lookups we'll only do this if everything
3085 * else matches.
3086 */
3087 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3088 NULL, NULL, NULL);
3089 if (ret)
3090 return ret;
3091 if (dir_gen != ctx->dir_gen)
3092 return 0;
31db9f7c
AB		 3093 ctx->found_idx = num;
3094 return 1;
3095 }
3096 return 0;
3097}
3098
924794c9 3099static int find_iref(struct btrfs_root *root,
31db9f7c
AB		 3100 struct btrfs_path *path,
3101 struct btrfs_key *key,
ba5e8f2e 3102 u64 dir, u64 dir_gen, struct fs_path *name)
31db9f7c
AB		 3103{
3104 int ret;
3105 struct find_ref_ctx ctx;
3106
3107 ctx.dir = dir;
3108 ctx.name = name;
ba5e8f2e 3109 ctx.dir_gen = dir_gen;
31db9f7c 3110 ctx.found_idx = -1;
ba5e8f2e 3111 ctx.root = root;
31db9f7c 3112
924794c9 3113 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
31db9f7c
AB		 3114 if (ret < 0)
3115 return ret;
3116
3117 if (ctx.found_idx == -1)
3118 return -ENOENT;
3119
3120 return ctx.found_idx;
3121}
3122
3123static int __record_changed_new_ref(int num, u64 dir, int index,
3124 struct fs_path *name,
3125 void *ctx)
3126{
ba5e8f2e 3127 u64 dir_gen;
31db9f7c
AB		 3128 int ret;
3129 struct send_ctx *sctx = ctx;
3130
ba5e8f2e
JB		 3131 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3132 NULL, NULL, NULL);
3133 if (ret)
3134 return ret;
3135
924794c9 3136 ret = find_iref(sctx->parent_root, sctx->right_path,
ba5e8f2e 3137 sctx->cmp_key, dir, dir_gen, name);
31db9f7c
AB		 3138 if (ret == -ENOENT)
3139 ret = __record_new_ref(num, dir, index, name, sctx);
3140 else if (ret > 0)
3141 ret = 0;
3142
3143 return ret;
3144}
3145
3146static int __record_changed_deleted_ref(int num, u64 dir, int index,
3147 struct fs_path *name,
3148 void *ctx)
3149{
ba5e8f2e 3150 u64 dir_gen;
31db9f7c
AB		 3151 int ret;
3152 struct send_ctx *sctx = ctx;
3153
ba5e8f2e
JB		 3154 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3155 NULL, NULL, NULL);
3156 if (ret)
3157 return ret;
3158
924794c9 3159 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
ba5e8f2e 3160 dir, dir_gen, name);
31db9f7c
AB		 3161 if (ret == -ENOENT)
3162 ret = __record_deleted_ref(num, dir, index, name, sctx);
3163 else if (ret > 0)
3164 ret = 0;
3165
3166 return ret;
3167}
3168
3169static int record_changed_ref(struct send_ctx *sctx)
3170{
3171 int ret = 0;
3172
924794c9 3173 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
31db9f7c
AB		 3174 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3175 if (ret < 0)
3176 goto out;
924794c9 3177 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 3178 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3179 if (ret < 0)
3180 goto out;
3181 ret = 0;
3182
3183out:
3184 return ret;
3185}
3186
3187/*
3188 * Record and process all refs at once. Needed when an inode changes the
3189 * generation number, which means that it was deleted and recreated.
3190 */
3191static int process_all_refs(struct send_ctx *sctx,
3192 enum btrfs_compare_tree_result cmd)
3193{
3194 int ret;
3195 struct btrfs_root *root;
3196 struct btrfs_path *path;
3197 struct btrfs_key key;
3198 struct btrfs_key found_key;
3199 struct extent_buffer *eb;
3200 int slot;
3201 iterate_inode_ref_t cb;
3202
3203 path = alloc_path_for_send();
3204 if (!path)
3205 return -ENOMEM;
3206
3207 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3208 root = sctx->send_root;
3209 cb = __record_new_ref;
3210 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3211 root = sctx->parent_root;
3212 cb = __record_deleted_ref;
3213 } else {
3214 BUG();
3215 }
3216
3217 key.objectid = sctx->cmp_key->objectid;
3218 key.type = BTRFS_INODE_REF_KEY;
3219 key.offset = 0;
3220 while (1) {
3221 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
e938c8ad 3222 if (ret < 0)
31db9f7c 3223 goto out;
e938c8ad 3224 if (ret)
31db9f7c 3225 break;
31db9f7c
AB		 3226
3227 eb = path->nodes[0];
3228 slot = path->slots[0];
3229 btrfs_item_key_to_cpu(eb, &found_key, slot);
3230
3231 if (found_key.objectid != key.objectid ||
96b5bd77
JS		 3232 (found_key.type != BTRFS_INODE_REF_KEY &&
3233 found_key.type != BTRFS_INODE_EXTREF_KEY))
31db9f7c 3234 break;
31db9f7c 3235
924794c9 3236 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
31db9f7c
AB		 3237 btrfs_release_path(path);
3238 if (ret < 0)
3239 goto out;
3240
3241 key.offset = found_key.offset + 1;
3242 }
e938c8ad 3243 btrfs_release_path(path);
31db9f7c
AB		 3244
3245 ret = process_recorded_refs(sctx);
3246
3247out:
3248 btrfs_free_path(path);
3249 return ret;
3250}
3251
3252static int send_set_xattr(struct send_ctx *sctx,
3253 struct fs_path *path,
3254 const char *name, int name_len,
3255 const char *data, int data_len)
3256{
3257 int ret = 0;
3258
3259 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3260 if (ret < 0)
3261 goto out;
3262
3263 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3264 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3265 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3266
3267 ret = send_cmd(sctx);
3268
3269tlv_put_failure:
3270out:
3271 return ret;
3272}
3273
3274static int send_remove_xattr(struct send_ctx *sctx,
3275 struct fs_path *path,
3276 const char *name, int name_len)
3277{
3278 int ret = 0;
3279
3280 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3281 if (ret < 0)
3282 goto out;
3283
3284 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3285 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3286
3287 ret = send_cmd(sctx);
3288
3289tlv_put_failure:
3290out:
3291 return ret;
3292}
3293
3294static int __process_new_xattr(int num, struct btrfs_key *di_key,
3295 const char *name, int name_len,
3296 const char *data, int data_len,
3297 u8 type, void *ctx)
3298{
3299 int ret;
3300 struct send_ctx *sctx = ctx;
3301 struct fs_path *p;
3302 posix_acl_xattr_header dummy_acl;
3303
924794c9 3304 p = fs_path_alloc();
31db9f7c
AB		 3305 if (!p)
3306 return -ENOMEM;
3307
3308 /*
3309 * This hack is needed because empty acl's are stored as zero byte
3310 * data in xattrs. Problem with that is, that receiving these zero byte
3311 * acl's will fail later. To fix this, we send a dummy acl list that
3312 * only contains the version number and no entries.
3313 */
3314 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3315 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3316 if (data_len == 0) {
3317 dummy_acl.a_version =
3318 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3319 data = (char *)&dummy_acl;
3320 data_len = sizeof(dummy_acl);
3321 }
3322 }
3323
3324 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3325 if (ret < 0)
3326 goto out;
3327
3328 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3329
3330out:
924794c9 3331 fs_path_free(p);
31db9f7c
AB		 3332 return ret;
3333}
3334
3335static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3336 const char *name, int name_len,
3337 const char *data, int data_len,
3338 u8 type, void *ctx)
3339{
3340 int ret;
3341 struct send_ctx *sctx = ctx;
3342 struct fs_path *p;
3343
924794c9 3344 p = fs_path_alloc();
31db9f7c
AB		 3345 if (!p)
3346 return -ENOMEM;
3347
3348 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3349 if (ret < 0)
3350 goto out;
3351
3352 ret = send_remove_xattr(sctx, p, name, name_len);
3353
3354out:
924794c9 3355 fs_path_free(p);
31db9f7c
AB		 3356 return ret;
3357}
3358
3359static int process_new_xattr(struct send_ctx *sctx)
3360{
3361 int ret = 0;
3362
924794c9
TI		 3363 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3364 sctx->cmp_key, __process_new_xattr, sctx);
31db9f7c
AB		 3365
3366 return ret;
3367}
3368
3369static int process_deleted_xattr(struct send_ctx *sctx)
3370{
3371 int ret;
3372
924794c9
TI		 3373 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3374 sctx->cmp_key, __process_deleted_xattr, sctx);
31db9f7c
AB		 3375
3376 return ret;
3377}
3378
3379struct find_xattr_ctx {
3380 const char *name;
3381 int name_len;
3382 int found_idx;
3383 char *found_data;
3384 int found_data_len;
3385};
3386
3387static int __find_xattr(int num, struct btrfs_key *di_key,
3388 const char *name, int name_len,
3389 const char *data, int data_len,
3390 u8 type, void *vctx)
3391{
3392 struct find_xattr_ctx *ctx = vctx;
3393
3394 if (name_len == ctx->name_len &&
3395 strncmp(name, ctx->name, name_len) == 0) {
3396 ctx->found_idx = num;
3397 ctx->found_data_len = data_len;
a5959bc0 3398 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
31db9f7c
AB		 3399 if (!ctx->found_data)
3400 return -ENOMEM;
31db9f7c
AB		 3401 return 1;
3402 }
3403 return 0;
3404}
3405
924794c9 3406static int find_xattr(struct btrfs_root *root,
31db9f7c
AB		 3407 struct btrfs_path *path,
3408 struct btrfs_key *key,
3409 const char *name, int name_len,
3410 char **data, int *data_len)
3411{
3412 int ret;
3413 struct find_xattr_ctx ctx;
3414
3415 ctx.name = name;
3416 ctx.name_len = name_len;
3417 ctx.found_idx = -1;
3418 ctx.found_data = NULL;
3419 ctx.found_data_len = 0;
3420
924794c9 3421 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
31db9f7c
AB		 3422 if (ret < 0)
3423 return ret;
3424
3425 if (ctx.found_idx == -1)
3426 return -ENOENT;
3427 if (data) {
3428 *data = ctx.found_data;
3429 *data_len = ctx.found_data_len;
3430 } else {
3431 kfree(ctx.found_data);
3432 }
3433 return ctx.found_idx;
3434}
3435
3436
3437static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3438 const char *name, int name_len,
3439 const char *data, int data_len,
3440 u8 type, void *ctx)
3441{
3442 int ret;
3443 struct send_ctx *sctx = ctx;
3444 char *found_data = NULL;
3445 int found_data_len = 0;
31db9f7c 3446
924794c9
TI		 3447 ret = find_xattr(sctx->parent_root, sctx->right_path,
3448 sctx->cmp_key, name, name_len, &found_data,
3449 &found_data_len);
31db9f7c
AB		 3450 if (ret == -ENOENT) {
3451 ret = __process_new_xattr(num, di_key, name, name_len, data,
3452 data_len, type, ctx);
3453 } else if (ret >= 0) {
3454 if (data_len != found_data_len ||
3455 memcmp(data, found_data, data_len)) {
3456 ret = __process_new_xattr(num, di_key, name, name_len,
3457 data, data_len, type, ctx);
3458 } else {
3459 ret = 0;
3460 }
3461 }
3462
3463 kfree(found_data);
31db9f7c
AB		 3464 return ret;
3465}
3466
3467static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3468 const char *name, int name_len,
3469 const char *data, int data_len,
3470 u8 type, void *ctx)
3471{
3472 int ret;
3473 struct send_ctx *sctx = ctx;
3474
924794c9
TI		 3475 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3476 name, name_len, NULL, NULL);
31db9f7c
AB		 3477 if (ret == -ENOENT)
3478 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3479 data_len, type, ctx);
3480 else if (ret >= 0)
3481 ret = 0;
3482
3483 return ret;
3484}
3485
3486static int process_changed_xattr(struct send_ctx *sctx)
3487{
3488 int ret = 0;
3489
924794c9 3490 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
31db9f7c
AB		 3491 sctx->cmp_key, __process_changed_new_xattr, sctx);
3492 if (ret < 0)
3493 goto out;
924794c9 3494 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 3495 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3496
3497out:
3498 return ret;
3499}
3500
3501static int process_all_new_xattrs(struct send_ctx *sctx)
3502{
3503 int ret;
3504 struct btrfs_root *root;
3505 struct btrfs_path *path;
3506 struct btrfs_key key;
3507 struct btrfs_key found_key;
3508 struct extent_buffer *eb;
3509 int slot;
3510
3511 path = alloc_path_for_send();
3512 if (!path)
3513 return -ENOMEM;
3514
3515 root = sctx->send_root;
3516
3517 key.objectid = sctx->cmp_key->objectid;
3518 key.type = BTRFS_XATTR_ITEM_KEY;
3519 key.offset = 0;
3520 while (1) {
3521 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3522 if (ret < 0)
3523 goto out;
3524 if (ret) {
3525 ret = 0;
3526 goto out;
3527 }
3528
3529 eb = path->nodes[0];
3530 slot = path->slots[0];
3531 btrfs_item_key_to_cpu(eb, &found_key, slot);
3532
3533 if (found_key.objectid != key.objectid ||
3534 found_key.type != key.type) {
3535 ret = 0;
3536 goto out;
3537 }
3538
924794c9
TI		 3539 ret = iterate_dir_item(root, path, &found_key,
3540 __process_new_xattr, sctx);
31db9f7c
AB		 3541 if (ret < 0)
3542 goto out;
3543
3544 btrfs_release_path(path);
3545 key.offset = found_key.offset + 1;
3546 }
3547
3548out:
3549 btrfs_free_path(path);
3550 return ret;
3551}
3552
ed259095
JB		 3553static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3554{
3555 struct btrfs_root *root = sctx->send_root;
3556 struct btrfs_fs_info *fs_info = root->fs_info;
3557 struct inode *inode;
3558 struct page *page;
3559 char *addr;
3560 struct btrfs_key key;
3561 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3562 pgoff_t last_index;
3563 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3564 ssize_t ret = 0;
3565
3566 key.objectid = sctx->cur_ino;
3567 key.type = BTRFS_INODE_ITEM_KEY;
3568 key.offset = 0;
3569
3570 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3571 if (IS_ERR(inode))
3572 return PTR_ERR(inode);
3573
3574 if (offset + len > i_size_read(inode)) {
3575 if (offset > i_size_read(inode))
3576 len = 0;
3577 else
3578 len = offset - i_size_read(inode);
3579 }
3580 if (len == 0)
3581 goto out;
3582
3583 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
3584 while (index <= last_index) {
3585 unsigned cur_len = min_t(unsigned, len,
3586 PAGE_CACHE_SIZE - pg_offset);
3587 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3588 if (!page) {
3589 ret = -ENOMEM;
3590 break;
3591 }
3592
3593 if (!PageUptodate(page)) {
3594 btrfs_readpage(NULL, page);
3595 lock_page(page);
3596 if (!PageUptodate(page)) {
3597 unlock_page(page);
3598 page_cache_release(page);
3599 ret = -EIO;
3600 break;
3601 }
3602 }
3603
3604 addr = kmap(page);
3605 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
3606 kunmap(page);
3607 unlock_page(page);
3608 page_cache_release(page);
3609 index++;
3610 pg_offset = 0;
3611 len -= cur_len;
3612 ret += cur_len;
3613 }
3614out:
3615 iput(inode);
3616 return ret;
3617}
3618
31db9f7c
AB		 3619/*
3620 * Read some bytes from the current inode/file and send a write command to
3621 * user space.
3622 */
3623static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3624{
3625 int ret = 0;
3626 struct fs_path *p;
ed259095 3627 ssize_t num_read = 0;
31db9f7c 3628
924794c9 3629 p = fs_path_alloc();
31db9f7c
AB		 3630 if (!p)
3631 return -ENOMEM;
3632
31db9f7c
AB		 3633verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3634
ed259095
JB		 3635 num_read = fill_read_buf(sctx, offset, len);
3636 if (num_read <= 0) {
3637 if (num_read < 0)
3638 ret = num_read;
31db9f7c 3639 goto out;
ed259095 3640 }
31db9f7c
AB		 3641
3642 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3643 if (ret < 0)
3644 goto out;
3645
3646 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3647 if (ret < 0)
3648 goto out;
3649
3650 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3651 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
e938c8ad 3652 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
31db9f7c
AB		 3653
3654 ret = send_cmd(sctx);
3655
3656tlv_put_failure:
3657out:
924794c9 3658 fs_path_free(p);
31db9f7c
AB		 3659 if (ret < 0)
3660 return ret;
e938c8ad 3661 return num_read;
31db9f7c
AB		 3662}
3663
3664/*
3665 * Send a clone command to user space.
3666 */
3667static int send_clone(struct send_ctx *sctx,
3668 u64 offset, u32 len,
3669 struct clone_root *clone_root)
3670{
3671 int ret = 0;
31db9f7c
AB		 3672 struct fs_path *p;
3673 u64 gen;
3674
3675verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
3676 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
3677 clone_root->root->objectid, clone_root->ino,
3678 clone_root->offset);
3679
924794c9 3680 p = fs_path_alloc();
31db9f7c
AB		 3681 if (!p)
3682 return -ENOMEM;
3683
3684 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
3685 if (ret < 0)
3686 goto out;
3687
3688 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3689 if (ret < 0)
3690 goto out;
3691
3692 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3693 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
3694 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3695
e938c8ad 3696 if (clone_root->root == sctx->send_root) {
31db9f7c 3697 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
85a7b33b 3698 &gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 3699 if (ret < 0)
3700 goto out;
3701 ret = get_cur_path(sctx, clone_root->ino, gen, p);
3702 } else {
924794c9 3703 ret = get_inode_path(clone_root->root, clone_root->ino, p);
31db9f7c
AB		 3704 }
3705 if (ret < 0)
3706 goto out;
3707
3708 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
e938c8ad 3709 clone_root->root->root_item.uuid);
31db9f7c 3710 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
e938c8ad 3711 clone_root->root->root_item.ctransid);
31db9f7c
AB		 3712 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3713 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3714 clone_root->offset);
3715
3716 ret = send_cmd(sctx);
3717
3718tlv_put_failure:
3719out:
924794c9 3720 fs_path_free(p);
31db9f7c
AB		 3721 return ret;
3722}
3723
cb95e7bf
MF		 3724/*
3725 * Send an update extent command to user space.
3726 */
3727static int send_update_extent(struct send_ctx *sctx,
3728 u64 offset, u32 len)
3729{
3730 int ret = 0;
3731 struct fs_path *p;
3732
924794c9 3733 p = fs_path_alloc();
cb95e7bf
MF		 3734 if (!p)
3735 return -ENOMEM;
3736
3737 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
3738 if (ret < 0)
3739 goto out;
3740
3741 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3742 if (ret < 0)
3743 goto out;
3744
3745 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3746 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3747 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
3748
3749 ret = send_cmd(sctx);
3750
3751tlv_put_failure:
3752out:
924794c9 3753 fs_path_free(p);
cb95e7bf
MF		 3754 return ret;
3755}
3756
31db9f7c
AB		 3757static int send_write_or_clone(struct send_ctx *sctx,
3758 struct btrfs_path *path,
3759 struct btrfs_key *key,
3760 struct clone_root *clone_root)
3761{
3762 int ret = 0;
3763 struct btrfs_file_extent_item *ei;
3764 u64 offset = key->offset;
3765 u64 pos = 0;
3766 u64 len;
3767 u32 l;
3768 u8 type;
3769
3770 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3771 struct btrfs_file_extent_item);
3772 type = btrfs_file_extent_type(path->nodes[0], ei);
74dd17fb 3773 if (type == BTRFS_FILE_EXTENT_INLINE) {
31db9f7c 3774 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
74dd17fb
CM		 3775 /*
3776 * it is possible the inline item won't cover the whole page,
3777 * but there may be items after this page. Make
3778 * sure to send the whole thing
3779 */
3780 len = PAGE_CACHE_ALIGN(len);
3781 } else {
31db9f7c 3782 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
74dd17fb 3783 }
31db9f7c
AB		 3784
3785 if (offset + len > sctx->cur_inode_size)
3786 len = sctx->cur_inode_size - offset;
3787 if (len == 0) {
3788 ret = 0;
3789 goto out;
3790 }
3791
cb95e7bf
MF		 3792 if (clone_root) {
3793 ret = send_clone(sctx, offset, len, clone_root);
3794 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
3795 ret = send_update_extent(sctx, offset, len);
3796 } else {
31db9f7c
AB		 3797 while (pos < len) {
3798 l = len - pos;
3799 if (l > BTRFS_SEND_READ_SIZE)
3800 l = BTRFS_SEND_READ_SIZE;
3801 ret = send_write(sctx, pos + offset, l);
3802 if (ret < 0)
3803 goto out;
3804 if (!ret)
3805 break;
3806 pos += ret;
3807 }
3808 ret = 0;
31db9f7c 3809 }
31db9f7c
AB		 3810out:
3811 return ret;
3812}
3813
3814static int is_extent_unchanged(struct send_ctx *sctx,
3815 struct btrfs_path *left_path,
3816 struct btrfs_key *ekey)
3817{
3818 int ret = 0;
3819 struct btrfs_key key;
3820 struct btrfs_path *path = NULL;
3821 struct extent_buffer *eb;
3822 int slot;
3823 struct btrfs_key found_key;
3824 struct btrfs_file_extent_item *ei;
3825 u64 left_disknr;
3826 u64 right_disknr;
3827 u64 left_offset;
3828 u64 right_offset;
3829 u64 left_offset_fixed;
3830 u64 left_len;
3831 u64 right_len;
74dd17fb
CM		 3832 u64 left_gen;
3833 u64 right_gen;
31db9f7c
AB		 3834 u8 left_type;
3835 u8 right_type;
3836
3837 path = alloc_path_for_send();
3838 if (!path)
3839 return -ENOMEM;
3840
3841 eb = left_path->nodes[0];
3842 slot = left_path->slots[0];
31db9f7c
AB		 3843 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3844 left_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 3845
3846 if (left_type != BTRFS_FILE_EXTENT_REG) {
3847 ret = 0;
3848 goto out;
3849 }
74dd17fb
CM		 3850 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3851 left_len = btrfs_file_extent_num_bytes(eb, ei);
3852 left_offset = btrfs_file_extent_offset(eb, ei);
3853 left_gen = btrfs_file_extent_generation(eb, ei);
31db9f7c
AB		 3854
3855 /*
3856 * Following comments will refer to these graphics. L is the left
3857 * extents which we are checking at the moment. 1-8 are the right
3858 * extents that we iterate.
3859 *
3860 * |-----L-----|
3861 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3862 *
3863 * |-----L-----|
3864 * |--1--|-2b-|...(same as above)
3865 *
3866 * Alternative situation. Happens on files where extents got split.
3867 * |-----L-----|
3868 * |-----------7-----------|-6-|
3869 *
3870 * Alternative situation. Happens on files which got larger.
3871 * |-----L-----|
3872 * |-8-|
3873 * Nothing follows after 8.
3874 */
3875
3876 key.objectid = ekey->objectid;
3877 key.type = BTRFS_EXTENT_DATA_KEY;
3878 key.offset = ekey->offset;
3879 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3880 if (ret < 0)
3881 goto out;
3882 if (ret) {
3883 ret = 0;
3884 goto out;
3885 }
3886
3887 /*
3888 * Handle special case where the right side has no extents at all.
3889 */
3890 eb = path->nodes[0];
3891 slot = path->slots[0];
3892 btrfs_item_key_to_cpu(eb, &found_key, slot);
3893 if (found_key.objectid != key.objectid ||
3894 found_key.type != key.type) {
57cfd462
JB		 3895 /* If we're a hole then just pretend nothing changed */
3896 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 3897 goto out;
3898 }
3899
3900 /*
3901 * We're now on 2a, 2b or 7.
3902 */
3903 key = found_key;
3904 while (key.offset < ekey->offset + left_len) {
3905 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3906 right_type = btrfs_file_extent_type(eb, ei);
3907 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3908 right_len = btrfs_file_extent_num_bytes(eb, ei);
3909 right_offset = btrfs_file_extent_offset(eb, ei);
74dd17fb 3910 right_gen = btrfs_file_extent_generation(eb, ei);
31db9f7c
AB		 3911
3912 if (right_type != BTRFS_FILE_EXTENT_REG) {
3913 ret = 0;
3914 goto out;
3915 }
3916
3917 /*
3918 * Are we at extent 8? If yes, we know the extent is changed.
3919 * This may only happen on the first iteration.
3920 */
d8347fa4 3921 if (found_key.offset + right_len <= ekey->offset) {
57cfd462
JB		 3922 /* If we're a hole just pretend nothing changed */
3923 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 3924 goto out;
3925 }
3926
3927 left_offset_fixed = left_offset;
3928 if (key.offset < ekey->offset) {
3929 /* Fix the right offset for 2a and 7. */
3930 right_offset += ekey->offset - key.offset;
3931 } else {
3932 /* Fix the left offset for all behind 2a and 2b */
3933 left_offset_fixed += key.offset - ekey->offset;
3934 }
3935
3936 /*
3937 * Check if we have the same extent.
3938 */
3954096d 3939 if (left_disknr != right_disknr ||
74dd17fb
CM		 3940 left_offset_fixed != right_offset ||
3941 left_gen != right_gen) {
31db9f7c
AB		 3942 ret = 0;
3943 goto out;
3944 }
3945
3946 /*
3947 * Go to the next extent.
3948 */
3949 ret = btrfs_next_item(sctx->parent_root, path);
3950 if (ret < 0)
3951 goto out;
3952 if (!ret) {
3953 eb = path->nodes[0];
3954 slot = path->slots[0];
3955 btrfs_item_key_to_cpu(eb, &found_key, slot);
3956 }
3957 if (ret || found_key.objectid != key.objectid ||
3958 found_key.type != key.type) {
3959 key.offset += right_len;
3960 break;
adaa4b8e
JS		 3961 }
3962 if (found_key.offset != key.offset + right_len) {
3963 ret = 0;
3964 goto out;
31db9f7c
AB		 3965 }
3966 key = found_key;
3967 }
3968
3969 /*
3970 * We're now behind the left extent (treat as unchanged) or at the end
3971 * of the right side (treat as changed).
3972 */
3973 if (key.offset >= ekey->offset + left_len)
3974 ret = 1;
3975 else
3976 ret = 0;
3977
3978
3979out:
3980 btrfs_free_path(path);
3981 return ret;
3982}
3983
3984static int process_extent(struct send_ctx *sctx,
3985 struct btrfs_path *path,
3986 struct btrfs_key *key)
3987{
31db9f7c 3988 struct clone_root *found_clone = NULL;
57cfd462 3989 int ret = 0;
31db9f7c
AB		 3990
3991 if (S_ISLNK(sctx->cur_inode_mode))
3992 return 0;
3993
3994 if (sctx->parent_root && !sctx->cur_inode_new) {
3995 ret = is_extent_unchanged(sctx, path, key);
3996 if (ret < 0)
3997 goto out;
3998 if (ret) {
3999 ret = 0;
4000 goto out;
4001 }
57cfd462
JB		 4002 } else {
4003 struct btrfs_file_extent_item *ei;
4004 u8 type;
4005
4006 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4007 struct btrfs_file_extent_item);
4008 type = btrfs_file_extent_type(path->nodes[0], ei);
4009 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4010 type == BTRFS_FILE_EXTENT_REG) {
4011 /*
4012 * The send spec does not have a prealloc command yet,
4013 * so just leave a hole for prealloc'ed extents until
4014 * we have enough commands queued up to justify rev'ing
4015 * the send spec.
4016 */
4017 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4018 ret = 0;
4019 goto out;
4020 }
4021
4022 /* Have a hole, just skip it. */
4023 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4024 ret = 0;
4025 goto out;
4026 }
4027 }
31db9f7c
AB		 4028 }
4029
4030 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4031 sctx->cur_inode_size, &found_clone);
4032 if (ret != -ENOENT && ret < 0)
4033 goto out;
4034
4035 ret = send_write_or_clone(sctx, path, key, found_clone);
4036
4037out:
4038 return ret;
4039}
4040
4041static int process_all_extents(struct send_ctx *sctx)
4042{
4043 int ret;
4044 struct btrfs_root *root;
4045 struct btrfs_path *path;
4046 struct btrfs_key key;
4047 struct btrfs_key found_key;
4048 struct extent_buffer *eb;
4049 int slot;
4050
4051 root = sctx->send_root;
4052 path = alloc_path_for_send();
4053 if (!path)
4054 return -ENOMEM;
4055
4056 key.objectid = sctx->cmp_key->objectid;
4057 key.type = BTRFS_EXTENT_DATA_KEY;
4058 key.offset = 0;
4059 while (1) {
4060 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
4061 if (ret < 0)
4062 goto out;
4063 if (ret) {
4064 ret = 0;
4065 goto out;
4066 }
4067
4068 eb = path->nodes[0];
4069 slot = path->slots[0];
4070 btrfs_item_key_to_cpu(eb, &found_key, slot);
4071
4072 if (found_key.objectid != key.objectid ||
4073 found_key.type != key.type) {
4074 ret = 0;
4075 goto out;
4076 }
4077
4078 ret = process_extent(sctx, path, &found_key);
4079 if (ret < 0)
4080 goto out;
4081
4082 btrfs_release_path(path);
4083 key.offset = found_key.offset + 1;
4084 }
4085
4086out:
4087 btrfs_free_path(path);
4088 return ret;
4089}
4090
4091static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
4092{
4093 int ret = 0;
4094
4095 if (sctx->cur_ino == 0)
4096 goto out;
4097 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
96b5bd77 4098 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 4099 goto out;
4100 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4101 goto out;
4102
4103 ret = process_recorded_refs(sctx);
e479d9bb
AB		 4104 if (ret < 0)
4105 goto out;
4106
4107 /*
4108 * We have processed the refs and thus need to advance send_progress.
4109 * Now, calls to get_cur_xxx will take the updated refs of the current
4110 * inode into account.
4111 */
4112 sctx->send_progress = sctx->cur_ino + 1;
31db9f7c
AB		 4113
4114out:
4115 return ret;
4116}
4117
4118static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4119{
4120 int ret = 0;
4121 u64 left_mode;
4122 u64 left_uid;
4123 u64 left_gid;
4124 u64 right_mode;
4125 u64 right_uid;
4126 u64 right_gid;
4127 int need_chmod = 0;
4128 int need_chown = 0;
4129
4130 ret = process_recorded_refs_if_needed(sctx, at_end);
4131 if (ret < 0)
4132 goto out;
4133
4134 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4135 goto out;
4136 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4137 goto out;
4138
4139 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
85a7b33b 4140 &left_mode, &left_uid, &left_gid, NULL);
31db9f7c
AB		 4141 if (ret < 0)
4142 goto out;
4143
e2d044fe
AL		 4144 if (!sctx->parent_root || sctx->cur_inode_new) {
4145 need_chown = 1;
4146 if (!S_ISLNK(sctx->cur_inode_mode))
31db9f7c 4147 need_chmod = 1;
e2d044fe
AL		 4148 } else {
4149 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4150 NULL, NULL, &right_mode, &right_uid,
4151 &right_gid, NULL);
4152 if (ret < 0)
4153 goto out;
31db9f7c 4154
e2d044fe
AL		 4155 if (left_uid != right_uid || left_gid != right_gid)
4156 need_chown = 1;
4157 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4158 need_chmod = 1;
31db9f7c
AB		 4159 }
4160
4161 if (S_ISREG(sctx->cur_inode_mode)) {
4162 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4163 sctx->cur_inode_size);
4164 if (ret < 0)
4165 goto out;
4166 }
4167
4168 if (need_chown) {
4169 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4170 left_uid, left_gid);
4171 if (ret < 0)
4172 goto out;
4173 }
4174 if (need_chmod) {
4175 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4176 left_mode);
4177 if (ret < 0)
4178 goto out;
4179 }
4180
4181 /*
4182 * Need to send that every time, no matter if it actually changed
4183 * between the two trees as we have done changes to the inode before.
4184 */
4185 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4186 if (ret < 0)
4187 goto out;
4188
4189out:
4190 return ret;
4191}
4192
4193static int changed_inode(struct send_ctx *sctx,
4194 enum btrfs_compare_tree_result result)
4195{
4196 int ret = 0;
4197 struct btrfs_key *key = sctx->cmp_key;
4198 struct btrfs_inode_item *left_ii = NULL;
4199 struct btrfs_inode_item *right_ii = NULL;
4200 u64 left_gen = 0;
4201 u64 right_gen = 0;
4202
31db9f7c
AB		 4203 sctx->cur_ino = key->objectid;
4204 sctx->cur_inode_new_gen = 0;
e479d9bb
AB		 4205
4206 /*
4207 * Set send_progress to current inode. This will tell all get_cur_xxx
4208 * functions that the current inode's refs are not updated yet. Later,
4209 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4210 */
31db9f7c
AB		 4211 sctx->send_progress = sctx->cur_ino;
4212
4213 if (result == BTRFS_COMPARE_TREE_NEW ||
4214 result == BTRFS_COMPARE_TREE_CHANGED) {
4215 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4216 sctx->left_path->slots[0],
4217 struct btrfs_inode_item);
4218 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4219 left_ii);
4220 } else {
4221 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4222 sctx->right_path->slots[0],
4223 struct btrfs_inode_item);
4224 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4225 right_ii);
4226 }
4227 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4228 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4229 sctx->right_path->slots[0],
4230 struct btrfs_inode_item);
4231
4232 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4233 right_ii);
6d85ed05
AB		 4234
4235 /*
4236 * The cur_ino = root dir case is special here. We can't treat
4237 * the inode as deleted+reused because it would generate a
4238 * stream that tries to delete/mkdir the root dir.
4239 */
4240 if (left_gen != right_gen &&
4241 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
31db9f7c
AB		 4242 sctx->cur_inode_new_gen = 1;
4243 }
4244
4245 if (result == BTRFS_COMPARE_TREE_NEW) {
4246 sctx->cur_inode_gen = left_gen;
4247 sctx->cur_inode_new = 1;
4248 sctx->cur_inode_deleted = 0;
4249 sctx->cur_inode_size = btrfs_inode_size(
4250 sctx->left_path->nodes[0], left_ii);
4251 sctx->cur_inode_mode = btrfs_inode_mode(
4252 sctx->left_path->nodes[0], left_ii);
4253 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
1f4692da 4254 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4255 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4256 sctx->cur_inode_gen = right_gen;
4257 sctx->cur_inode_new = 0;
4258 sctx->cur_inode_deleted = 1;
4259 sctx->cur_inode_size = btrfs_inode_size(
4260 sctx->right_path->nodes[0], right_ii);
4261 sctx->cur_inode_mode = btrfs_inode_mode(
4262 sctx->right_path->nodes[0], right_ii);
4263 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
766702ef
AB		 4264 /*
4265 * We need to do some special handling in case the inode was
4266 * reported as changed with a changed generation number. This
4267 * means that the original inode was deleted and new inode
4268 * reused the same inum. So we have to treat the old inode as
4269 * deleted and the new one as new.
4270 */
31db9f7c 4271 if (sctx->cur_inode_new_gen) {
766702ef
AB		 4272 /*
4273 * First, process the inode as if it was deleted.
4274 */
31db9f7c
AB		 4275 sctx->cur_inode_gen = right_gen;
4276 sctx->cur_inode_new = 0;
4277 sctx->cur_inode_deleted = 1;
4278 sctx->cur_inode_size = btrfs_inode_size(
4279 sctx->right_path->nodes[0], right_ii);
4280 sctx->cur_inode_mode = btrfs_inode_mode(
4281 sctx->right_path->nodes[0], right_ii);
4282 ret = process_all_refs(sctx,
4283 BTRFS_COMPARE_TREE_DELETED);
4284 if (ret < 0)
4285 goto out;
4286
766702ef
AB		 4287 /*
4288 * Now process the inode as if it was new.
4289 */
31db9f7c
AB		 4290 sctx->cur_inode_gen = left_gen;
4291 sctx->cur_inode_new = 1;
4292 sctx->cur_inode_deleted = 0;
4293 sctx->cur_inode_size = btrfs_inode_size(
4294 sctx->left_path->nodes[0], left_ii);
4295 sctx->cur_inode_mode = btrfs_inode_mode(
4296 sctx->left_path->nodes[0], left_ii);
1f4692da 4297 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4298 if (ret < 0)
4299 goto out;
4300
4301 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4302 if (ret < 0)
4303 goto out;
e479d9bb
AB		 4304 /*
4305 * Advance send_progress now as we did not get into
4306 * process_recorded_refs_if_needed in the new_gen case.
4307 */
4308 sctx->send_progress = sctx->cur_ino + 1;
766702ef
AB		 4309
4310 /*
4311 * Now process all extents and xattrs of the inode as if
4312 * they were all new.
4313 */
31db9f7c
AB		 4314 ret = process_all_extents(sctx);
4315 if (ret < 0)
4316 goto out;
4317 ret = process_all_new_xattrs(sctx);
4318 if (ret < 0)
4319 goto out;
4320 } else {
4321 sctx->cur_inode_gen = left_gen;
4322 sctx->cur_inode_new = 0;
4323 sctx->cur_inode_new_gen = 0;
4324 sctx->cur_inode_deleted = 0;
4325 sctx->cur_inode_size = btrfs_inode_size(
4326 sctx->left_path->nodes[0], left_ii);
4327 sctx->cur_inode_mode = btrfs_inode_mode(
4328 sctx->left_path->nodes[0], left_ii);
4329 }
4330 }
4331
4332out:
4333 return ret;
4334}
4335
766702ef
AB		 4336/*
4337 * We have to process new refs before deleted refs, but compare_trees gives us
4338 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4339 * first and later process them in process_recorded_refs.
4340 * For the cur_inode_new_gen case, we skip recording completely because
4341 * changed_inode did already initiate processing of refs. The reason for this is
4342 * that in this case, compare_tree actually compares the refs of 2 different
4343 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4344 * refs of the right tree as deleted and all refs of the left tree as new.
4345 */
31db9f7c
AB		 4346static int changed_ref(struct send_ctx *sctx,
4347 enum btrfs_compare_tree_result result)
4348{
4349 int ret = 0;
4350
4351 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4352
4353 if (!sctx->cur_inode_new_gen &&
4354 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4355 if (result == BTRFS_COMPARE_TREE_NEW)
4356 ret = record_new_ref(sctx);
4357 else if (result == BTRFS_COMPARE_TREE_DELETED)
4358 ret = record_deleted_ref(sctx);
4359 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4360 ret = record_changed_ref(sctx);
4361 }
4362
4363 return ret;
4364}
4365
766702ef
AB		 4366/*
4367 * Process new/deleted/changed xattrs. We skip processing in the
4368 * cur_inode_new_gen case because changed_inode did already initiate processing
4369 * of xattrs. The reason is the same as in changed_ref
4370 */
31db9f7c
AB		 4371static int changed_xattr(struct send_ctx *sctx,
4372 enum btrfs_compare_tree_result result)
4373{
4374 int ret = 0;
4375
4376 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4377
4378 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4379 if (result == BTRFS_COMPARE_TREE_NEW)
4380 ret = process_new_xattr(sctx);
4381 else if (result == BTRFS_COMPARE_TREE_DELETED)
4382 ret = process_deleted_xattr(sctx);
4383 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4384 ret = process_changed_xattr(sctx);
4385 }
4386
4387 return ret;
4388}
4389
766702ef
AB		 4390/*
4391 * Process new/deleted/changed extents. We skip processing in the
4392 * cur_inode_new_gen case because changed_inode did already initiate processing
4393 * of extents. The reason is the same as in changed_ref
4394 */
31db9f7c
AB		 4395static int changed_extent(struct send_ctx *sctx,
4396 enum btrfs_compare_tree_result result)
4397{
4398 int ret = 0;
4399
4400 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4401
4402 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4403 if (result != BTRFS_COMPARE_TREE_DELETED)
4404 ret = process_extent(sctx, sctx->left_path,
4405 sctx->cmp_key);
4406 }
4407
4408 return ret;
4409}
4410
ba5e8f2e
JB		 4411static int dir_changed(struct send_ctx *sctx, u64 dir)
4412{
4413 u64 orig_gen, new_gen;
4414 int ret;
4415
4416 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
4417 NULL, NULL);
4418 if (ret)
4419 return ret;
4420
4421 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
4422 NULL, NULL, NULL);
4423 if (ret)
4424 return ret;
4425
4426 return (orig_gen != new_gen) ? 1 : 0;
4427}
4428
4429static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
4430 struct btrfs_key *key)
4431{
4432 struct btrfs_inode_extref *extref;
4433 struct extent_buffer *leaf;
4434 u64 dirid = 0, last_dirid = 0;
4435 unsigned long ptr;
4436 u32 item_size;
4437 u32 cur_offset = 0;
4438 int ref_name_len;
4439 int ret = 0;
4440
4441 /* Easy case, just check this one dirid */
4442 if (key->type == BTRFS_INODE_REF_KEY) {
4443 dirid = key->offset;
4444
4445 ret = dir_changed(sctx, dirid);
4446 goto out;
4447 }
4448
4449 leaf = path->nodes[0];
4450 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4451 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4452 while (cur_offset < item_size) {
4453 extref = (struct btrfs_inode_extref *)(ptr +
4454 cur_offset);
4455 dirid = btrfs_inode_extref_parent(leaf, extref);
4456 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
4457 cur_offset += ref_name_len + sizeof(*extref);
4458 if (dirid == last_dirid)
4459 continue;
4460 ret = dir_changed(sctx, dirid);
4461 if (ret)
4462 break;
4463 last_dirid = dirid;
4464 }
4465out:
4466 return ret;
4467}
4468
766702ef
AB		 4469/*
4470 * Updates compare related fields in sctx and simply forwards to the actual
4471 * changed_xxx functions.
4472 */
31db9f7c
AB		 4473static int changed_cb(struct btrfs_root *left_root,
4474 struct btrfs_root *right_root,
4475 struct btrfs_path *left_path,
4476 struct btrfs_path *right_path,
4477 struct btrfs_key *key,
4478 enum btrfs_compare_tree_result result,
4479 void *ctx)
4480{
4481 int ret = 0;
4482 struct send_ctx *sctx = ctx;
4483
ba5e8f2e
JB		 4484 if (result == BTRFS_COMPARE_TREE_SAME) {
4485 if (key->type != BTRFS_INODE_REF_KEY &&
4486 key->type != BTRFS_INODE_EXTREF_KEY)
4487 return 0;
4488 ret = compare_refs(sctx, left_path, key);
4489 if (!ret)
4490 return 0;
4491 if (ret < 0)
4492 return ret;
4493 result = BTRFS_COMPARE_TREE_CHANGED;
4494 ret = 0;
4495 }
4496
31db9f7c
AB		 4497 sctx->left_path = left_path;
4498 sctx->right_path = right_path;
4499 sctx->cmp_key = key;
4500
4501 ret = finish_inode_if_needed(sctx, 0);
4502 if (ret < 0)
4503 goto out;
4504
2981e225
AB		 4505 /* Ignore non-FS objects */
4506 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
4507 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
4508 goto out;
4509
31db9f7c
AB		 4510 if (key->type == BTRFS_INODE_ITEM_KEY)
4511 ret = changed_inode(sctx, result);
96b5bd77
JS		 4512 else if (key->type == BTRFS_INODE_REF_KEY ||
4513 key->type == BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 4514 ret = changed_ref(sctx, result);
4515 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4516 ret = changed_xattr(sctx, result);
4517 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4518 ret = changed_extent(sctx, result);
4519
4520out:
4521 return ret;
4522}
4523
4524static int full_send_tree(struct send_ctx *sctx)
4525{
4526 int ret;
4527 struct btrfs_trans_handle *trans = NULL;
4528 struct btrfs_root *send_root = sctx->send_root;
4529 struct btrfs_key key;
4530 struct btrfs_key found_key;
4531 struct btrfs_path *path;
4532 struct extent_buffer *eb;
4533 int slot;
4534 u64 start_ctransid;
4535 u64 ctransid;
4536
4537 path = alloc_path_for_send();
4538 if (!path)
4539 return -ENOMEM;
4540
5f3ab90a 4541 spin_lock(&send_root->root_item_lock);
31db9f7c 4542 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5f3ab90a 4543 spin_unlock(&send_root->root_item_lock);
31db9f7c
AB		 4544
4545 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4546 key.type = BTRFS_INODE_ITEM_KEY;
4547 key.offset = 0;
4548
4549join_trans:
4550 /*
4551 * We need to make sure the transaction does not get committed
4552 * while we do anything on commit roots. Join a transaction to prevent
4553 * this.
4554 */
4555 trans = btrfs_join_transaction(send_root);
4556 if (IS_ERR(trans)) {
4557 ret = PTR_ERR(trans);
4558 trans = NULL;
4559 goto out;
4560 }
4561
4562 /*
766702ef
AB		 4563 * Make sure the tree has not changed after re-joining. We detect this
4564 * by comparing start_ctransid and ctransid. They should always match.
31db9f7c 4565 */
5f3ab90a 4566 spin_lock(&send_root->root_item_lock);
31db9f7c 4567 ctransid = btrfs_root_ctransid(&send_root->root_item);
5f3ab90a 4568 spin_unlock(&send_root->root_item_lock);
31db9f7c
AB		 4569
4570 if (ctransid != start_ctransid) {
4571 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4572 "send was modified in between. This is "
4573 "probably a bug.\n");
4574 ret = -EIO;
4575 goto out;
4576 }
4577
4578 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4579 if (ret < 0)
4580 goto out;
4581 if (ret)
4582 goto out_finish;
4583
4584 while (1) {
4585 /*
4586 * When someone want to commit while we iterate, end the
4587 * joined transaction and rejoin.
4588 */
4589 if (btrfs_should_end_transaction(trans, send_root)) {
4590 ret = btrfs_end_transaction(trans, send_root);
4591 trans = NULL;
4592 if (ret < 0)
4593 goto out;
4594 btrfs_release_path(path);
4595 goto join_trans;
4596 }
4597
4598 eb = path->nodes[0];
4599 slot = path->slots[0];
4600 btrfs_item_key_to_cpu(eb, &found_key, slot);
4601
4602 ret = changed_cb(send_root, NULL, path, NULL,
4603 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4604 if (ret < 0)
4605 goto out;
4606
4607 key.objectid = found_key.objectid;
4608 key.type = found_key.type;
4609 key.offset = found_key.offset + 1;
4610
4611 ret = btrfs_next_item(send_root, path);
4612 if (ret < 0)
4613 goto out;
4614 if (ret) {
4615 ret = 0;
4616 break;
4617 }
4618 }
4619
4620out_finish:
4621 ret = finish_inode_if_needed(sctx, 1);
4622
4623out:
4624 btrfs_free_path(path);
4625 if (trans) {
4626 if (!ret)
4627 ret = btrfs_end_transaction(trans, send_root);
4628 else
4629 btrfs_end_transaction(trans, send_root);
4630 }
4631 return ret;
4632}
4633
4634static int send_subvol(struct send_ctx *sctx)
4635{
4636 int ret;
4637
c2c71324
SB		 4638 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
4639 ret = send_header(sctx);
4640 if (ret < 0)
4641 goto out;
4642 }
31db9f7c
AB		 4643
4644 ret = send_subvol_begin(sctx);
4645 if (ret < 0)
4646 goto out;
4647
4648 if (sctx->parent_root) {
4649 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4650 changed_cb, sctx);
4651 if (ret < 0)
4652 goto out;
4653 ret = finish_inode_if_needed(sctx, 1);
4654 if (ret < 0)
4655 goto out;
4656 } else {
4657 ret = full_send_tree(sctx);
4658 if (ret < 0)
4659 goto out;
4660 }
4661
4662out:
31db9f7c
AB		 4663 free_recorded_refs(sctx);
4664 return ret;
4665}
4666
4667long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4668{
4669 int ret = 0;
4670 struct btrfs_root *send_root;
4671 struct btrfs_root *clone_root;
4672 struct btrfs_fs_info *fs_info;
4673 struct btrfs_ioctl_send_args *arg = NULL;
4674 struct btrfs_key key;
31db9f7c
AB		 4675 struct send_ctx *sctx = NULL;
4676 u32 i;
4677 u64 *clone_sources_tmp = NULL;
4678
4679 if (!capable(CAP_SYS_ADMIN))
4680 return -EPERM;
4681
496ad9aa 4682 send_root = BTRFS_I(file_inode(mnt_file))->root;
31db9f7c
AB		 4683 fs_info = send_root->fs_info;
4684
139f807a
JB		 4685 /*
4686 * This is done when we lookup the root, it should already be complete
4687 * by the time we get here.
4688 */
4689 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
4690
4691 /*
4692 * If we just created this root we need to make sure that the orphan
4693 * cleanup has been done and committed since we search the commit root,
4694 * so check its commit root transid with our otransid and if they match
4695 * commit the transaction to make sure everything is updated.
4696 */
4697 down_read(&send_root->fs_info->extent_commit_sem);
4698 if (btrfs_header_generation(send_root->commit_root) ==
4699 btrfs_root_otransid(&send_root->root_item)) {
4700 struct btrfs_trans_handle *trans;
4701
4702 up_read(&send_root->fs_info->extent_commit_sem);
4703
4704 trans = btrfs_attach_transaction_barrier(send_root);
4705 if (IS_ERR(trans)) {
4706 if (PTR_ERR(trans) != -ENOENT) {
4707 ret = PTR_ERR(trans);
4708 goto out;
4709 }
4710 /* ENOENT means theres no transaction */
4711 } else {
4712 ret = btrfs_commit_transaction(trans, send_root);
4713 if (ret)
4714 goto out;
4715 }
4716 } else {
4717 up_read(&send_root->fs_info->extent_commit_sem);
4718 }
4719
31db9f7c
AB		 4720 arg = memdup_user(arg_, sizeof(*arg));
4721 if (IS_ERR(arg)) {
4722 ret = PTR_ERR(arg);
4723 arg = NULL;
4724 goto out;
4725 }
4726
4727 if (!access_ok(VERIFY_READ, arg->clone_sources,
4728 sizeof(*arg->clone_sources *
4729 arg->clone_sources_count))) {
4730 ret = -EFAULT;
4731 goto out;
4732 }
4733
c2c71324 4734 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
cb95e7bf
MF		 4735 ret = -EINVAL;
4736 goto out;
4737 }
4738
31db9f7c
AB		 4739 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4740 if (!sctx) {
4741 ret = -ENOMEM;
4742 goto out;
4743 }
4744
4745 INIT_LIST_HEAD(&sctx->new_refs);
4746 INIT_LIST_HEAD(&sctx->deleted_refs);
4747 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4748 INIT_LIST_HEAD(&sctx->name_cache_list);
4749
cb95e7bf
MF		 4750 sctx->flags = arg->flags;
4751
31db9f7c 4752 sctx->send_filp = fget(arg->send_fd);
ecc7ada7
TI		 4753 if (!sctx->send_filp) {
4754 ret = -EBADF;
31db9f7c
AB		 4755 goto out;
4756 }
4757
4758 sctx->mnt = mnt_file->f_path.mnt;
4759
4760 sctx->send_root = send_root;
4761 sctx->clone_roots_cnt = arg->clone_sources_count;
4762
4763 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4764 sctx->send_buf = vmalloc(sctx->send_max_size);
4765 if (!sctx->send_buf) {
4766 ret = -ENOMEM;
4767 goto out;
4768 }
4769
4770 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4771 if (!sctx->read_buf) {
4772 ret = -ENOMEM;
4773 goto out;
4774 }
4775
4776 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4777 (arg->clone_sources_count + 1));
4778 if (!sctx->clone_roots) {
4779 ret = -ENOMEM;
4780 goto out;
4781 }
4782
4783 if (arg->clone_sources_count) {
4784 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4785 sizeof(*arg->clone_sources));
4786 if (!clone_sources_tmp) {
4787 ret = -ENOMEM;
4788 goto out;
4789 }
4790
4791 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4792 arg->clone_sources_count *
4793 sizeof(*arg->clone_sources));
4794 if (ret) {
4795 ret = -EFAULT;
4796 goto out;
4797 }
4798
4799 for (i = 0; i < arg->clone_sources_count; i++) {
4800 key.objectid = clone_sources_tmp[i];
4801 key.type = BTRFS_ROOT_ITEM_KEY;
4802 key.offset = (u64)-1;
4803 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
31db9f7c
AB		 4804 if (IS_ERR(clone_root)) {
4805 ret = PTR_ERR(clone_root);
4806 goto out;
4807 }
4808 sctx->clone_roots[i].root = clone_root;
4809 }
4810 vfree(clone_sources_tmp);
4811 clone_sources_tmp = NULL;
4812 }
4813
4814 if (arg->parent_root) {
4815 key.objectid = arg->parent_root;
4816 key.type = BTRFS_ROOT_ITEM_KEY;
4817 key.offset = (u64)-1;
4818 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
b1b19596
SB		 4819 if (IS_ERR(sctx->parent_root)) {
4820 ret = PTR_ERR(sctx->parent_root);
31db9f7c
AB		 4821 goto out;
4822 }
4823 }
4824
4825 /*
4826 * Clones from send_root are allowed, but only if the clone source
4827 * is behind the current send position. This is checked while searching
4828 * for possible clone sources.
4829 */
4830 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4831
4832 /* We do a bsearch later */
4833 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4834 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4835 NULL);
4836
4837 ret = send_subvol(sctx);
4838 if (ret < 0)
4839 goto out;
4840
c2c71324
SB		 4841 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
4842 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4843 if (ret < 0)
4844 goto out;
4845 ret = send_cmd(sctx);
4846 if (ret < 0)
4847 goto out;
4848 }
31db9f7c
AB		 4849
4850out:
31db9f7c
AB		 4851 kfree(arg);
4852 vfree(clone_sources_tmp);
4853
4854 if (sctx) {
4855 if (sctx->send_filp)
4856 fput(sctx->send_filp);
4857
4858 vfree(sctx->clone_roots);
4859 vfree(sctx->send_buf);
4860 vfree(sctx->read_buf);
4861
4862 name_cache_free(sctx);
4863
4864 kfree(sctx);
4865 }
4866
4867 return ret;
4868}
Linux kernel - Deliverables
This page took 0.427116 seconds and 5 git commands to generate.