projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix check for changed extent in is_extent_unchanged
[linux-2.6-block.git] / fs / btrfs / send.c
CommitLineData
31db9f7c
AB		 1/*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/bsearch.h>
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/sort.h>
23#include <linux/mount.h>
24#include <linux/xattr.h>
25#include <linux/posix_acl_xattr.h>
26#include <linux/radix-tree.h>
27#include <linux/crc32c.h>
a1857ebe 28#include <linux/vmalloc.h>
31db9f7c
AB		 29
30#include "send.h"
31#include "backref.h"
32#include "locking.h"
33#include "disk-io.h"
34#include "btrfs_inode.h"
35#include "transaction.h"
36
37static int g_verbose = 0;
38
39#define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
40
41/*
42 * A fs_path is a helper to dynamically build path names with unknown size.
43 * It reallocates the internal buffer on demand.
44 * It allows fast adding of path elements on the right side (normal path) and
45 * fast adding to the left side (reversed path). A reversed path can also be
46 * unreversed if needed.
47 */
48struct fs_path {
49 union {
50 struct {
51 char *start;
52 char *end;
53 char *prepared;
54
55 char *buf;
56 int buf_len;
57 int reversed:1;
58 int virtual_mem:1;
59 char inline_buf[];
60 };
61 char pad[PAGE_SIZE];
62 };
63};
64#define FS_PATH_INLINE_SIZE \
65 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
66
67
68/* reused for each extent */
69struct clone_root {
70 struct btrfs_root *root;
71 u64 ino;
72 u64 offset;
73
74 u64 found_refs;
75};
76
77#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
78#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
79
80struct send_ctx {
81 struct file *send_filp;
82 loff_t send_off;
83 char *send_buf;
84 u32 send_size;
85 u32 send_max_size;
86 u64 total_send_size;
87 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
88
89 struct vfsmount *mnt;
90
91 struct btrfs_root *send_root;
92 struct btrfs_root *parent_root;
93 struct clone_root *clone_roots;
94 int clone_roots_cnt;
95
96 /* current state of the compare_tree call */
97 struct btrfs_path *left_path;
98 struct btrfs_path *right_path;
99 struct btrfs_key *cmp_key;
100
101 /*
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
104 */
105 u64 cur_ino;
106 u64 cur_inode_gen;
107 int cur_inode_new;
108 int cur_inode_new_gen;
109 int cur_inode_deleted;
31db9f7c
AB		 110 u64 cur_inode_size;
111 u64 cur_inode_mode;
112
113 u64 send_progress;
114
115 struct list_head new_refs;
116 struct list_head deleted_refs;
117
118 struct radix_tree_root name_cache;
119 struct list_head name_cache_list;
120 int name_cache_size;
121
122 struct file *cur_inode_filp;
123 char *read_buf;
124};
125
126struct name_cache_entry {
127 struct list_head list;
7e0926fe
AB		 128 /*
129 * radix_tree has only 32bit entries but we need to handle 64bit inums.
130 * We use the lower 32bit of the 64bit inum to store it in the tree. If
131 * more then one inum would fall into the same entry, we use radix_list
132 * to store the additional entries. radix_list is also used to store
133 * entries where two entries have the same inum but different
134 * generations.
135 */
136 struct list_head radix_list;
31db9f7c
AB		 137 u64 ino;
138 u64 gen;
139 u64 parent_ino;
140 u64 parent_gen;
141 int ret;
142 int need_later_update;
143 int name_len;
144 char name[];
145};
146
147static void fs_path_reset(struct fs_path *p)
148{
149 if (p->reversed) {
150 p->start = p->buf + p->buf_len - 1;
151 p->end = p->start;
152 *p->start = 0;
153 } else {
154 p->start = p->buf;
155 p->end = p->start;
156 *p->start = 0;
157 }
158}
159
160static struct fs_path *fs_path_alloc(struct send_ctx *sctx)
161{
162 struct fs_path *p;
163
164 p = kmalloc(sizeof(*p), GFP_NOFS);
165 if (!p)
166 return NULL;
167 p->reversed = 0;
168 p->virtual_mem = 0;
169 p->buf = p->inline_buf;
170 p->buf_len = FS_PATH_INLINE_SIZE;
171 fs_path_reset(p);
172 return p;
173}
174
175static struct fs_path *fs_path_alloc_reversed(struct send_ctx *sctx)
176{
177 struct fs_path *p;
178
179 p = fs_path_alloc(sctx);
180 if (!p)
181 return NULL;
182 p->reversed = 1;
183 fs_path_reset(p);
184 return p;
185}
186
187static void fs_path_free(struct send_ctx *sctx, struct fs_path *p)
188{
189 if (!p)
190 return;
191 if (p->buf != p->inline_buf) {
192 if (p->virtual_mem)
193 vfree(p->buf);
194 else
195 kfree(p->buf);
196 }
197 kfree(p);
198}
199
200static int fs_path_len(struct fs_path *p)
201{
202 return p->end - p->start;
203}
204
205static int fs_path_ensure_buf(struct fs_path *p, int len)
206{
207 char *tmp_buf;
208 int path_len;
209 int old_buf_len;
210
211 len++;
212
213 if (p->buf_len >= len)
214 return 0;
215
216 path_len = p->end - p->start;
217 old_buf_len = p->buf_len;
218 len = PAGE_ALIGN(len);
219
220 if (p->buf == p->inline_buf) {
221 tmp_buf = kmalloc(len, GFP_NOFS);
222 if (!tmp_buf) {
223 tmp_buf = vmalloc(len);
224 if (!tmp_buf)
225 return -ENOMEM;
226 p->virtual_mem = 1;
227 }
228 memcpy(tmp_buf, p->buf, p->buf_len);
229 p->buf = tmp_buf;
230 p->buf_len = len;
231 } else {
232 if (p->virtual_mem) {
233 tmp_buf = vmalloc(len);
234 if (!tmp_buf)
235 return -ENOMEM;
236 memcpy(tmp_buf, p->buf, p->buf_len);
237 vfree(p->buf);
238 } else {
239 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
240 if (!tmp_buf) {
241 tmp_buf = vmalloc(len);
242 if (!tmp_buf)
243 return -ENOMEM;
244 memcpy(tmp_buf, p->buf, p->buf_len);
245 kfree(p->buf);
246 p->virtual_mem = 1;
247 }
248 }
249 p->buf = tmp_buf;
250 p->buf_len = len;
251 }
252 if (p->reversed) {
253 tmp_buf = p->buf + old_buf_len - path_len - 1;
254 p->end = p->buf + p->buf_len - 1;
255 p->start = p->end - path_len;
256 memmove(p->start, tmp_buf, path_len + 1);
257 } else {
258 p->start = p->buf;
259 p->end = p->start + path_len;
260 }
261 return 0;
262}
263
264static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
265{
266 int ret;
267 int new_len;
268
269 new_len = p->end - p->start + name_len;
270 if (p->start != p->end)
271 new_len++;
272 ret = fs_path_ensure_buf(p, new_len);
273 if (ret < 0)
274 goto out;
275
276 if (p->reversed) {
277 if (p->start != p->end)
278 *--p->start = '/';
279 p->start -= name_len;
280 p->prepared = p->start;
281 } else {
282 if (p->start != p->end)
283 *p->end++ = '/';
284 p->prepared = p->end;
285 p->end += name_len;
286 *p->end = 0;
287 }
288
289out:
290 return ret;
291}
292
293static int fs_path_add(struct fs_path *p, const char *name, int name_len)
294{
295 int ret;
296
297 ret = fs_path_prepare_for_add(p, name_len);
298 if (ret < 0)
299 goto out;
300 memcpy(p->prepared, name, name_len);
301 p->prepared = NULL;
302
303out:
304 return ret;
305}
306
307static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
308{
309 int ret;
310
311 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
312 if (ret < 0)
313 goto out;
314 memcpy(p->prepared, p2->start, p2->end - p2->start);
315 p->prepared = NULL;
316
317out:
318 return ret;
319}
320
321static int fs_path_add_from_extent_buffer(struct fs_path *p,
322 struct extent_buffer *eb,
323 unsigned long off, int len)
324{
325 int ret;
326
327 ret = fs_path_prepare_for_add(p, len);
328 if (ret < 0)
329 goto out;
330
331 read_extent_buffer(eb, p->prepared, off, len);
332 p->prepared = NULL;
333
334out:
335 return ret;
336}
337
9ea3ef51 338#if 0
31db9f7c
AB		 339static void fs_path_remove(struct fs_path *p)
340{
341 BUG_ON(p->reversed);
342 while (p->start != p->end && *p->end != '/')
343 p->end--;
344 *p->end = 0;
345}
9ea3ef51 346#endif
31db9f7c
AB		 347
348static int fs_path_copy(struct fs_path *p, struct fs_path *from)
349{
350 int ret;
351
352 p->reversed = from->reversed;
353 fs_path_reset(p);
354
355 ret = fs_path_add_path(p, from);
356
357 return ret;
358}
359
360
361static void fs_path_unreverse(struct fs_path *p)
362{
363 char *tmp;
364 int len;
365
366 if (!p->reversed)
367 return;
368
369 tmp = p->start;
370 len = p->end - p->start;
371 p->start = p->buf;
372 p->end = p->start + len;
373 memmove(p->start, tmp, len + 1);
374 p->reversed = 0;
375}
376
377static struct btrfs_path *alloc_path_for_send(void)
378{
379 struct btrfs_path *path;
380
381 path = btrfs_alloc_path();
382 if (!path)
383 return NULL;
384 path->search_commit_root = 1;
385 path->skip_locking = 1;
386 return path;
387}
388
389static int write_buf(struct send_ctx *sctx, const void *buf, u32 len)
390{
391 int ret;
392 mm_segment_t old_fs;
393 u32 pos = 0;
394
395 old_fs = get_fs();
396 set_fs(KERNEL_DS);
397
398 while (pos < len) {
399 ret = vfs_write(sctx->send_filp, (char *)buf + pos, len - pos,
400 &sctx->send_off);
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
556 return write_buf(sctx, &hdr, sizeof(hdr));
557}
558
559/*
560 * For each command/item we want to send to userspace, we call this function.
561 */
562static int begin_cmd(struct send_ctx *sctx, int cmd)
563{
564 struct btrfs_cmd_header *hdr;
565
566 if (!sctx->send_buf) {
567 WARN_ON(1);
568 return -EINVAL;
569 }
570
571 BUG_ON(sctx->send_size);
572
573 sctx->send_size += sizeof(*hdr);
574 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
575 hdr->cmd = cpu_to_le16(cmd);
576
577 return 0;
578}
579
580static int send_cmd(struct send_ctx *sctx)
581{
582 int ret;
583 struct btrfs_cmd_header *hdr;
584 u32 crc;
585
586 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
587 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
588 hdr->crc = 0;
589
590 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
591 hdr->crc = cpu_to_le32(crc);
592
593 ret = write_buf(sctx, sctx->send_buf, sctx->send_size);
594
595 sctx->total_send_size += sctx->send_size;
596 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
597 sctx->send_size = 0;
598
599 return ret;
600}
601
602/*
603 * Sends a move instruction to user space
604 */
605static int send_rename(struct send_ctx *sctx,
606 struct fs_path *from, struct fs_path *to)
607{
608 int ret;
609
610verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
611
612 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
613 if (ret < 0)
614 goto out;
615
616 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
617 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
618
619 ret = send_cmd(sctx);
620
621tlv_put_failure:
622out:
623 return ret;
624}
625
626/*
627 * Sends a link instruction to user space
628 */
629static int send_link(struct send_ctx *sctx,
630 struct fs_path *path, struct fs_path *lnk)
631{
632 int ret;
633
634verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
635
636 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
637 if (ret < 0)
638 goto out;
639
640 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
641 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
642
643 ret = send_cmd(sctx);
644
645tlv_put_failure:
646out:
647 return ret;
648}
649
650/*
651 * Sends an unlink instruction to user space
652 */
653static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
654{
655 int ret;
656
657verbose_printk("btrfs: send_unlink %s\n", path->start);
658
659 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
660 if (ret < 0)
661 goto out;
662
663 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
664
665 ret = send_cmd(sctx);
666
667tlv_put_failure:
668out:
669 return ret;
670}
671
672/*
673 * Sends a rmdir instruction to user space
674 */
675static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
676{
677 int ret;
678
679verbose_printk("btrfs: send_rmdir %s\n", path->start);
680
681 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
682 if (ret < 0)
683 goto out;
684
685 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
686
687 ret = send_cmd(sctx);
688
689tlv_put_failure:
690out:
691 return ret;
692}
693
694/*
695 * Helper function to retrieve some fields from an inode item.
696 */
697static int get_inode_info(struct btrfs_root *root,
698 u64 ino, u64 *size, u64 *gen,
85a7b33b
AB		 699 u64 *mode, u64 *uid, u64 *gid,
700 u64 *rdev)
31db9f7c
AB		 701{
702 int ret;
703 struct btrfs_inode_item *ii;
704 struct btrfs_key key;
705 struct btrfs_path *path;
706
707 path = alloc_path_for_send();
708 if (!path)
709 return -ENOMEM;
710
711 key.objectid = ino;
712 key.type = BTRFS_INODE_ITEM_KEY;
713 key.offset = 0;
714 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
715 if (ret < 0)
716 goto out;
717 if (ret) {
718 ret = -ENOENT;
719 goto out;
720 }
721
722 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
723 struct btrfs_inode_item);
724 if (size)
725 *size = btrfs_inode_size(path->nodes[0], ii);
726 if (gen)
727 *gen = btrfs_inode_generation(path->nodes[0], ii);
728 if (mode)
729 *mode = btrfs_inode_mode(path->nodes[0], ii);
730 if (uid)
731 *uid = btrfs_inode_uid(path->nodes[0], ii);
732 if (gid)
733 *gid = btrfs_inode_gid(path->nodes[0], ii);
85a7b33b
AB		 734 if (rdev)
735 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
31db9f7c
AB		 736
737out:
738 btrfs_free_path(path);
739 return ret;
740}
741
742typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
743 struct fs_path *p,
744 void *ctx);
745
746/*
747 * Helper function to iterate the entries in ONE btrfs_inode_ref.
748 * The iterate callback may return a non zero value to stop iteration. This can
749 * be a negative value for error codes or 1 to simply stop it.
750 *
751 * path must point to the INODE_REF when called.
752 */
753static int iterate_inode_ref(struct send_ctx *sctx,
754 struct btrfs_root *root, struct btrfs_path *path,
755 struct btrfs_key *found_key, int resolve,
756 iterate_inode_ref_t iterate, void *ctx)
757{
758 struct extent_buffer *eb;
759 struct btrfs_item *item;
760 struct btrfs_inode_ref *iref;
761 struct btrfs_path *tmp_path;
762 struct fs_path *p;
763 u32 cur;
764 u32 len;
765 u32 total;
766 int slot;
767 u32 name_len;
768 char *start;
769 int ret = 0;
770 int num;
771 int index;
772
773 p = fs_path_alloc_reversed(sctx);
774 if (!p)
775 return -ENOMEM;
776
777 tmp_path = alloc_path_for_send();
778 if (!tmp_path) {
779 fs_path_free(sctx, p);
780 return -ENOMEM;
781 }
782
783 eb = path->nodes[0];
784 slot = path->slots[0];
785 item = btrfs_item_nr(eb, slot);
786 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
787 cur = 0;
788 len = 0;
789 total = btrfs_item_size(eb, item);
790
791 num = 0;
792 while (cur < total) {
793 fs_path_reset(p);
794
795 name_len = btrfs_inode_ref_name_len(eb, iref);
796 index = btrfs_inode_ref_index(eb, iref);
797 if (resolve) {
798 start = btrfs_iref_to_path(root, tmp_path, iref, eb,
799 found_key->offset, p->buf,
800 p->buf_len);
801 if (IS_ERR(start)) {
802 ret = PTR_ERR(start);
803 goto out;
804 }
805 if (start < p->buf) {
806 /* overflow , try again with larger buffer */
807 ret = fs_path_ensure_buf(p,
808 p->buf_len + p->buf - start);
809 if (ret < 0)
810 goto out;
811 start = btrfs_iref_to_path(root, tmp_path, iref,
812 eb, found_key->offset, p->buf,
813 p->buf_len);
814 if (IS_ERR(start)) {
815 ret = PTR_ERR(start);
816 goto out;
817 }
818 BUG_ON(start < p->buf);
819 }
820 p->start = start;
821 } else {
822 ret = fs_path_add_from_extent_buffer(p, eb,
823 (unsigned long)(iref + 1), name_len);
824 if (ret < 0)
825 goto out;
826 }
827
828
829 len = sizeof(*iref) + name_len;
830 iref = (struct btrfs_inode_ref *)((char *)iref + len);
831 cur += len;
832
833 ret = iterate(num, found_key->offset, index, p, ctx);
834 if (ret)
835 goto out;
836
837 num++;
838 }
839
840out:
841 btrfs_free_path(tmp_path);
842 fs_path_free(sctx, p);
843 return ret;
844}
845
846typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
847 const char *name, int name_len,
848 const char *data, int data_len,
849 u8 type, void *ctx);
850
851/*
852 * Helper function to iterate the entries in ONE btrfs_dir_item.
853 * The iterate callback may return a non zero value to stop iteration. This can
854 * be a negative value for error codes or 1 to simply stop it.
855 *
856 * path must point to the dir item when called.
857 */
858static int iterate_dir_item(struct send_ctx *sctx,
859 struct btrfs_root *root, struct btrfs_path *path,
860 struct btrfs_key *found_key,
861 iterate_dir_item_t iterate, void *ctx)
862{
863 int ret = 0;
864 struct extent_buffer *eb;
865 struct btrfs_item *item;
866 struct btrfs_dir_item *di;
31db9f7c
AB		 867 struct btrfs_key di_key;
868 char *buf = NULL;
869 char *buf2 = NULL;
870 int buf_len;
871 int buf_virtual = 0;
872 u32 name_len;
873 u32 data_len;
874 u32 cur;
875 u32 len;
876 u32 total;
877 int slot;
878 int num;
879 u8 type;
880
881 buf_len = PAGE_SIZE;
882 buf = kmalloc(buf_len, GFP_NOFS);
883 if (!buf) {
884 ret = -ENOMEM;
885 goto out;
886 }
887
31db9f7c
AB		 888 eb = path->nodes[0];
889 slot = path->slots[0];
890 item = btrfs_item_nr(eb, slot);
891 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
892 cur = 0;
893 len = 0;
894 total = btrfs_item_size(eb, item);
895
896 num = 0;
897 while (cur < total) {
898 name_len = btrfs_dir_name_len(eb, di);
899 data_len = btrfs_dir_data_len(eb, di);
900 type = btrfs_dir_type(eb, di);
901 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
902
903 if (name_len + data_len > buf_len) {
904 buf_len = PAGE_ALIGN(name_len + data_len);
905 if (buf_virtual) {
906 buf2 = vmalloc(buf_len);
907 if (!buf2) {
908 ret = -ENOMEM;
909 goto out;
910 }
911 vfree(buf);
912 } else {
913 buf2 = krealloc(buf, buf_len, GFP_NOFS);
914 if (!buf2) {
915 buf2 = vmalloc(buf_len);
916 if (!buf2) {
917 ret = -ENOMEM;
918 goto out;
919 }
920 kfree(buf);
921 buf_virtual = 1;
922 }
923 }
924
925 buf = buf2;
926 buf2 = NULL;
927 }
928
929 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
930 name_len + data_len);
931
932 len = sizeof(*di) + name_len + data_len;
933 di = (struct btrfs_dir_item *)((char *)di + len);
934 cur += len;
935
936 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
937 data_len, type, ctx);
938 if (ret < 0)
939 goto out;
940 if (ret) {
941 ret = 0;
942 goto out;
943 }
944
945 num++;
946 }
947
948out:
31db9f7c
AB		 949 if (buf_virtual)
950 vfree(buf);
951 else
952 kfree(buf);
953 return ret;
954}
955
956static int __copy_first_ref(int num, u64 dir, int index,
957 struct fs_path *p, void *ctx)
958{
959 int ret;
960 struct fs_path *pt = ctx;
961
962 ret = fs_path_copy(pt, p);
963 if (ret < 0)
964 return ret;
965
966 /* we want the first only */
967 return 1;
968}
969
970/*
971 * Retrieve the first path of an inode. If an inode has more then one
972 * ref/hardlink, this is ignored.
973 */
974static int get_inode_path(struct send_ctx *sctx, struct btrfs_root *root,
975 u64 ino, struct fs_path *path)
976{
977 int ret;
978 struct btrfs_key key, found_key;
979 struct btrfs_path *p;
980
981 p = alloc_path_for_send();
982 if (!p)
983 return -ENOMEM;
984
985 fs_path_reset(path);
986
987 key.objectid = ino;
988 key.type = BTRFS_INODE_REF_KEY;
989 key.offset = 0;
990
991 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
992 if (ret < 0)
993 goto out;
994 if (ret) {
995 ret = 1;
996 goto out;
997 }
998 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
999 if (found_key.objectid != ino ||
1000 found_key.type != BTRFS_INODE_REF_KEY) {
1001 ret = -ENOENT;
1002 goto out;
1003 }
1004
1005 ret = iterate_inode_ref(sctx, root, p, &found_key, 1,
1006 __copy_first_ref, path);
1007 if (ret < 0)
1008 goto out;
1009 ret = 0;
1010
1011out:
1012 btrfs_free_path(p);
1013 return ret;
1014}
1015
1016struct backref_ctx {
1017 struct send_ctx *sctx;
1018
1019 /* number of total found references */
1020 u64 found;
1021
1022 /*
1023 * used for clones found in send_root. clones found behind cur_objectid
1024 * and cur_offset are not considered as allowed clones.
1025 */
1026 u64 cur_objectid;
1027 u64 cur_offset;
1028
1029 /* may be truncated in case it's the last extent in a file */
1030 u64 extent_len;
1031
1032 /* Just to check for bugs in backref resolving */
ee849c04 1033 int found_itself;
31db9f7c
AB		 1034};
1035
1036static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1037{
1038 u64 root = (u64)key;
1039 struct clone_root *cr = (struct clone_root *)elt;
1040
1041 if (root < cr->root->objectid)
1042 return -1;
1043 if (root > cr->root->objectid)
1044 return 1;
1045 return 0;
1046}
1047
1048static int __clone_root_cmp_sort(const void *e1, const void *e2)
1049{
1050 struct clone_root *cr1 = (struct clone_root *)e1;
1051 struct clone_root *cr2 = (struct clone_root *)e2;
1052
1053 if (cr1->root->objectid < cr2->root->objectid)
1054 return -1;
1055 if (cr1->root->objectid > cr2->root->objectid)
1056 return 1;
1057 return 0;
1058}
1059
1060/*
1061 * Called for every backref that is found for the current extent.
766702ef 1062 * Results are collected in sctx->clone_roots->ino/offset/found_refs
31db9f7c
AB		 1063 */
1064static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1065{
1066 struct backref_ctx *bctx = ctx_;
1067 struct clone_root *found;
1068 int ret;
1069 u64 i_size;
1070
1071 /* First check if the root is in the list of accepted clone sources */
1072 found = bsearch((void *)root, bctx->sctx->clone_roots,
1073 bctx->sctx->clone_roots_cnt,
1074 sizeof(struct clone_root),
1075 __clone_root_cmp_bsearch);
1076 if (!found)
1077 return 0;
1078
1079 if (found->root == bctx->sctx->send_root &&
1080 ino == bctx->cur_objectid &&
1081 offset == bctx->cur_offset) {
ee849c04 1082 bctx->found_itself = 1;
31db9f7c
AB		 1083 }
1084
1085 /*
766702ef 1086 * There are inodes that have extents that lie behind its i_size. Don't
31db9f7c
AB		 1087 * accept clones from these extents.
1088 */
85a7b33b
AB		 1089 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1090 NULL);
31db9f7c
AB		 1091 if (ret < 0)
1092 return ret;
1093
1094 if (offset + bctx->extent_len > i_size)
1095 return 0;
1096
1097 /*
1098 * Make sure we don't consider clones from send_root that are
1099 * behind the current inode/offset.
1100 */
1101 if (found->root == bctx->sctx->send_root) {
1102 /*
1103 * TODO for the moment we don't accept clones from the inode
1104 * that is currently send. We may change this when
1105 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1106 * file.
1107 */
1108 if (ino >= bctx->cur_objectid)
1109 return 0;
e938c8ad
AB		 1110#if 0
1111 if (ino > bctx->cur_objectid)
1112 return 0;
1113 if (offset + bctx->extent_len > bctx->cur_offset)
31db9f7c 1114 return 0;
e938c8ad 1115#endif
31db9f7c
AB		 1116 }
1117
1118 bctx->found++;
1119 found->found_refs++;
1120 if (ino < found->ino) {
1121 found->ino = ino;
1122 found->offset = offset;
1123 } else if (found->ino == ino) {
1124 /*
1125 * same extent found more then once in the same file.
1126 */
1127 if (found->offset > offset + bctx->extent_len)
1128 found->offset = offset;
1129 }
1130
1131 return 0;
1132}
1133
1134/*
766702ef
AB		 1135 * Given an inode, offset and extent item, it finds a good clone for a clone
1136 * instruction. Returns -ENOENT when none could be found. The function makes
1137 * sure that the returned clone is usable at the point where sending is at the
1138 * moment. This means, that no clones are accepted which lie behind the current
1139 * inode+offset.
1140 *
31db9f7c
AB		 1141 * path must point to the extent item when called.
1142 */
1143static int find_extent_clone(struct send_ctx *sctx,
1144 struct btrfs_path *path,
1145 u64 ino, u64 data_offset,
1146 u64 ino_size,
1147 struct clone_root **found)
1148{
1149 int ret;
1150 int extent_type;
1151 u64 logical;
1152 u64 num_bytes;
1153 u64 extent_item_pos;
1154 struct btrfs_file_extent_item *fi;
1155 struct extent_buffer *eb = path->nodes[0];
35075bb0 1156 struct backref_ctx *backref_ctx = NULL;
31db9f7c
AB		 1157 struct clone_root *cur_clone_root;
1158 struct btrfs_key found_key;
1159 struct btrfs_path *tmp_path;
1160 u32 i;
1161
1162 tmp_path = alloc_path_for_send();
1163 if (!tmp_path)
1164 return -ENOMEM;
1165
35075bb0
AB		 1166 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1167 if (!backref_ctx) {
1168 ret = -ENOMEM;
1169 goto out;
1170 }
1171
31db9f7c
AB		 1172 if (data_offset >= ino_size) {
1173 /*
1174 * There may be extents that lie behind the file's size.
1175 * I at least had this in combination with snapshotting while
1176 * writing large files.
1177 */
1178 ret = 0;
1179 goto out;
1180 }
1181
1182 fi = btrfs_item_ptr(eb, path->slots[0],
1183 struct btrfs_file_extent_item);
1184 extent_type = btrfs_file_extent_type(eb, fi);
1185 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1186 ret = -ENOENT;
1187 goto out;
1188 }
1189
1190 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1191 logical = btrfs_file_extent_disk_bytenr(eb, fi);
1192 if (logical == 0) {
1193 ret = -ENOENT;
1194 goto out;
1195 }
1196 logical += btrfs_file_extent_offset(eb, fi);
1197
1198 ret = extent_from_logical(sctx->send_root->fs_info,
1199 logical, tmp_path, &found_key);
1200 btrfs_release_path(tmp_path);
1201
1202 if (ret < 0)
1203 goto out;
1204 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1205 ret = -EIO;
1206 goto out;
1207 }
1208
1209 /*
1210 * Setup the clone roots.
1211 */
1212 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1213 cur_clone_root = sctx->clone_roots + i;
1214 cur_clone_root->ino = (u64)-1;
1215 cur_clone_root->offset = 0;
1216 cur_clone_root->found_refs = 0;
1217 }
1218
35075bb0
AB		 1219 backref_ctx->sctx = sctx;
1220 backref_ctx->found = 0;
1221 backref_ctx->cur_objectid = ino;
1222 backref_ctx->cur_offset = data_offset;
1223 backref_ctx->found_itself = 0;
1224 backref_ctx->extent_len = num_bytes;
31db9f7c
AB		 1225
1226 /*
1227 * The last extent of a file may be too large due to page alignment.
1228 * We need to adjust extent_len in this case so that the checks in
1229 * __iterate_backrefs work.
1230 */
1231 if (data_offset + num_bytes >= ino_size)
35075bb0 1232 backref_ctx->extent_len = ino_size - data_offset;
31db9f7c
AB		 1233
1234 /*
1235 * Now collect all backrefs.
1236 */
1237 extent_item_pos = logical - found_key.objectid;
1238 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1239 found_key.objectid, extent_item_pos, 1,
35075bb0 1240 __iterate_backrefs, backref_ctx);
31db9f7c
AB		 1241 if (ret < 0)
1242 goto out;
1243
35075bb0 1244 if (!backref_ctx->found_itself) {
31db9f7c
AB		 1245 /* found a bug in backref code? */
1246 ret = -EIO;
1247 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1248 "send_root. inode=%llu, offset=%llu, "
1249 "logical=%llu\n",
1250 ino, data_offset, logical);
1251 goto out;
1252 }
1253
1254verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1255 "ino=%llu, "
1256 "num_bytes=%llu, logical=%llu\n",
1257 data_offset, ino, num_bytes, logical);
1258
35075bb0 1259 if (!backref_ctx->found)
31db9f7c
AB		 1260 verbose_printk("btrfs: no clones found\n");
1261
1262 cur_clone_root = NULL;
1263 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1264 if (sctx->clone_roots[i].found_refs) {
1265 if (!cur_clone_root)
1266 cur_clone_root = sctx->clone_roots + i;
1267 else if (sctx->clone_roots[i].root == sctx->send_root)
1268 /* prefer clones from send_root over others */
1269 cur_clone_root = sctx->clone_roots + i;
31db9f7c
AB		 1270 }
1271
1272 }
1273
1274 if (cur_clone_root) {
1275 *found = cur_clone_root;
1276 ret = 0;
1277 } else {
1278 ret = -ENOENT;
1279 }
1280
1281out:
1282 btrfs_free_path(tmp_path);
35075bb0 1283 kfree(backref_ctx);
31db9f7c
AB		 1284 return ret;
1285}
1286
1287static int read_symlink(struct send_ctx *sctx,
1288 struct btrfs_root *root,
1289 u64 ino,
1290 struct fs_path *dest)
1291{
1292 int ret;
1293 struct btrfs_path *path;
1294 struct btrfs_key key;
1295 struct btrfs_file_extent_item *ei;
1296 u8 type;
1297 u8 compression;
1298 unsigned long off;
1299 int len;
1300
1301 path = alloc_path_for_send();
1302 if (!path)
1303 return -ENOMEM;
1304
1305 key.objectid = ino;
1306 key.type = BTRFS_EXTENT_DATA_KEY;
1307 key.offset = 0;
1308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1309 if (ret < 0)
1310 goto out;
1311 BUG_ON(ret);
1312
1313 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1314 struct btrfs_file_extent_item);
1315 type = btrfs_file_extent_type(path->nodes[0], ei);
1316 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1317 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1318 BUG_ON(compression);
1319
1320 off = btrfs_file_extent_inline_start(ei);
1321 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1322
1323 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
31db9f7c
AB		 1324
1325out:
1326 btrfs_free_path(path);
1327 return ret;
1328}
1329
1330/*
1331 * Helper function to generate a file name that is unique in the root of
1332 * send_root and parent_root. This is used to generate names for orphan inodes.
1333 */
1334static int gen_unique_name(struct send_ctx *sctx,
1335 u64 ino, u64 gen,
1336 struct fs_path *dest)
1337{
1338 int ret = 0;
1339 struct btrfs_path *path;
1340 struct btrfs_dir_item *di;
1341 char tmp[64];
1342 int len;
1343 u64 idx = 0;
1344
1345 path = alloc_path_for_send();
1346 if (!path)
1347 return -ENOMEM;
1348
1349 while (1) {
1350 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1351 ino, gen, idx);
1352 if (len >= sizeof(tmp)) {
1353 /* should really not happen */
1354 ret = -EOVERFLOW;
1355 goto out;
1356 }
1357
1358 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1359 path, BTRFS_FIRST_FREE_OBJECTID,
1360 tmp, strlen(tmp), 0);
1361 btrfs_release_path(path);
1362 if (IS_ERR(di)) {
1363 ret = PTR_ERR(di);
1364 goto out;
1365 }
1366 if (di) {
1367 /* not unique, try again */
1368 idx++;
1369 continue;
1370 }
1371
1372 if (!sctx->parent_root) {
1373 /* unique */
1374 ret = 0;
1375 break;
1376 }
1377
1378 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1379 path, BTRFS_FIRST_FREE_OBJECTID,
1380 tmp, strlen(tmp), 0);
1381 btrfs_release_path(path);
1382 if (IS_ERR(di)) {
1383 ret = PTR_ERR(di);
1384 goto out;
1385 }
1386 if (di) {
1387 /* not unique, try again */
1388 idx++;
1389 continue;
1390 }
1391 /* unique */
1392 break;
1393 }
1394
1395 ret = fs_path_add(dest, tmp, strlen(tmp));
1396
1397out:
1398 btrfs_free_path(path);
1399 return ret;
1400}
1401
1402enum inode_state {
1403 inode_state_no_change,
1404 inode_state_will_create,
1405 inode_state_did_create,
1406 inode_state_will_delete,
1407 inode_state_did_delete,
1408};
1409
1410static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1411{
1412 int ret;
1413 int left_ret;
1414 int right_ret;
1415 u64 left_gen;
1416 u64 right_gen;
1417
1418 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
85a7b33b 1419 NULL, NULL);
31db9f7c
AB		 1420 if (ret < 0 && ret != -ENOENT)
1421 goto out;
1422 left_ret = ret;
1423
1424 if (!sctx->parent_root) {
1425 right_ret = -ENOENT;
1426 } else {
1427 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
85a7b33b 1428 NULL, NULL, NULL, NULL);
31db9f7c
AB		 1429 if (ret < 0 && ret != -ENOENT)
1430 goto out;
1431 right_ret = ret;
1432 }
1433
1434 if (!left_ret && !right_ret) {
e938c8ad 1435 if (left_gen == gen && right_gen == gen) {
31db9f7c 1436 ret = inode_state_no_change;
e938c8ad 1437 } else if (left_gen == gen) {
31db9f7c
AB		 1438 if (ino < sctx->send_progress)
1439 ret = inode_state_did_create;
1440 else
1441 ret = inode_state_will_create;
1442 } else if (right_gen == gen) {
1443 if (ino < sctx->send_progress)
1444 ret = inode_state_did_delete;
1445 else
1446 ret = inode_state_will_delete;
1447 } else {
1448 ret = -ENOENT;
1449 }
1450 } else if (!left_ret) {
1451 if (left_gen == gen) {
1452 if (ino < sctx->send_progress)
1453 ret = inode_state_did_create;
1454 else
1455 ret = inode_state_will_create;
1456 } else {
1457 ret = -ENOENT;
1458 }
1459 } else if (!right_ret) {
1460 if (right_gen == gen) {
1461 if (ino < sctx->send_progress)
1462 ret = inode_state_did_delete;
1463 else
1464 ret = inode_state_will_delete;
1465 } else {
1466 ret = -ENOENT;
1467 }
1468 } else {
1469 ret = -ENOENT;
1470 }
1471
1472out:
1473 return ret;
1474}
1475
1476static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1477{
1478 int ret;
1479
1480 ret = get_cur_inode_state(sctx, ino, gen);
1481 if (ret < 0)
1482 goto out;
1483
1484 if (ret == inode_state_no_change ||
1485 ret == inode_state_did_create ||
1486 ret == inode_state_will_delete)
1487 ret = 1;
1488 else
1489 ret = 0;
1490
1491out:
1492 return ret;
1493}
1494
1495/*
1496 * Helper function to lookup a dir item in a dir.
1497 */
1498static int lookup_dir_item_inode(struct btrfs_root *root,
1499 u64 dir, const char *name, int name_len,
1500 u64 *found_inode,
1501 u8 *found_type)
1502{
1503 int ret = 0;
1504 struct btrfs_dir_item *di;
1505 struct btrfs_key key;
1506 struct btrfs_path *path;
1507
1508 path = alloc_path_for_send();
1509 if (!path)
1510 return -ENOMEM;
1511
1512 di = btrfs_lookup_dir_item(NULL, root, path,
1513 dir, name, name_len, 0);
1514 if (!di) {
1515 ret = -ENOENT;
1516 goto out;
1517 }
1518 if (IS_ERR(di)) {
1519 ret = PTR_ERR(di);
1520 goto out;
1521 }
1522 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1523 *found_inode = key.objectid;
1524 *found_type = btrfs_dir_type(path->nodes[0], di);
1525
1526out:
1527 btrfs_free_path(path);
1528 return ret;
1529}
1530
766702ef
AB		 1531/*
1532 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1533 * generation of the parent dir and the name of the dir entry.
1534 */
31db9f7c
AB		 1535static int get_first_ref(struct send_ctx *sctx,
1536 struct btrfs_root *root, u64 ino,
1537 u64 *dir, u64 *dir_gen, struct fs_path *name)
1538{
1539 int ret;
1540 struct btrfs_key key;
1541 struct btrfs_key found_key;
1542 struct btrfs_path *path;
1543 struct btrfs_inode_ref *iref;
1544 int len;
1545
1546 path = alloc_path_for_send();
1547 if (!path)
1548 return -ENOMEM;
1549
1550 key.objectid = ino;
1551 key.type = BTRFS_INODE_REF_KEY;
1552 key.offset = 0;
1553
1554 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1555 if (ret < 0)
1556 goto out;
1557 if (!ret)
1558 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1559 path->slots[0]);
1560 if (ret || found_key.objectid != key.objectid ||
1561 found_key.type != key.type) {
1562 ret = -ENOENT;
1563 goto out;
1564 }
1565
1566 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1567 struct btrfs_inode_ref);
1568 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1569 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1570 (unsigned long)(iref + 1), len);
1571 if (ret < 0)
1572 goto out;
1573 btrfs_release_path(path);
1574
1575 ret = get_inode_info(root, found_key.offset, NULL, dir_gen, NULL, NULL,
85a7b33b 1576 NULL, NULL);
31db9f7c
AB		 1577 if (ret < 0)
1578 goto out;
1579
1580 *dir = found_key.offset;
1581
1582out:
1583 btrfs_free_path(path);
1584 return ret;
1585}
1586
1587static int is_first_ref(struct send_ctx *sctx,
1588 struct btrfs_root *root,
1589 u64 ino, u64 dir,
1590 const char *name, int name_len)
1591{
1592 int ret;
1593 struct fs_path *tmp_name;
1594 u64 tmp_dir;
1595 u64 tmp_dir_gen;
1596
1597 tmp_name = fs_path_alloc(sctx);
1598 if (!tmp_name)
1599 return -ENOMEM;
1600
1601 ret = get_first_ref(sctx, root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1602 if (ret < 0)
1603 goto out;
1604
b9291aff 1605 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
31db9f7c
AB		 1606 ret = 0;
1607 goto out;
1608 }
1609
e938c8ad 1610 ret = !memcmp(tmp_name->start, name, name_len);
31db9f7c
AB		 1611
1612out:
1613 fs_path_free(sctx, tmp_name);
1614 return ret;
1615}
1616
766702ef
AB		 1617/*
1618 * Used by process_recorded_refs to determine if a new ref would overwrite an
1619 * already existing ref. In case it detects an overwrite, it returns the
1620 * inode/gen in who_ino/who_gen.
1621 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1622 * to make sure later references to the overwritten inode are possible.
1623 * Orphanizing is however only required for the first ref of an inode.
1624 * process_recorded_refs does an additional is_first_ref check to see if
1625 * orphanizing is really required.
1626 */
31db9f7c
AB		 1627static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1628 const char *name, int name_len,
1629 u64 *who_ino, u64 *who_gen)
1630{
1631 int ret = 0;
1632 u64 other_inode = 0;
1633 u8 other_type = 0;
1634
1635 if (!sctx->parent_root)
1636 goto out;
1637
1638 ret = is_inode_existent(sctx, dir, dir_gen);
1639 if (ret <= 0)
1640 goto out;
1641
1642 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1643 &other_inode, &other_type);
1644 if (ret < 0 && ret != -ENOENT)
1645 goto out;
1646 if (ret) {
1647 ret = 0;
1648 goto out;
1649 }
1650
766702ef
AB		 1651 /*
1652 * Check if the overwritten ref was already processed. If yes, the ref
1653 * was already unlinked/moved, so we can safely assume that we will not
1654 * overwrite anything at this point in time.
1655 */
31db9f7c
AB		 1656 if (other_inode > sctx->send_progress) {
1657 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
85a7b33b 1658 who_gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 1659 if (ret < 0)
1660 goto out;
1661
1662 ret = 1;
1663 *who_ino = other_inode;
1664 } else {
1665 ret = 0;
1666 }
1667
1668out:
1669 return ret;
1670}
1671
766702ef
AB		 1672/*
1673 * Checks if the ref was overwritten by an already processed inode. This is
1674 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1675 * thus the orphan name needs be used.
1676 * process_recorded_refs also uses it to avoid unlinking of refs that were
1677 * overwritten.
1678 */
31db9f7c
AB		 1679static int did_overwrite_ref(struct send_ctx *sctx,
1680 u64 dir, u64 dir_gen,
1681 u64 ino, u64 ino_gen,
1682 const char *name, int name_len)
1683{
1684 int ret = 0;
1685 u64 gen;
1686 u64 ow_inode;
1687 u8 other_type;
1688
1689 if (!sctx->parent_root)
1690 goto out;
1691
1692 ret = is_inode_existent(sctx, dir, dir_gen);
1693 if (ret <= 0)
1694 goto out;
1695
1696 /* check if the ref was overwritten by another ref */
1697 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1698 &ow_inode, &other_type);
1699 if (ret < 0 && ret != -ENOENT)
1700 goto out;
1701 if (ret) {
1702 /* was never and will never be overwritten */
1703 ret = 0;
1704 goto out;
1705 }
1706
1707 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
85a7b33b 1708 NULL, NULL);
31db9f7c
AB		 1709 if (ret < 0)
1710 goto out;
1711
1712 if (ow_inode == ino && gen == ino_gen) {
1713 ret = 0;
1714 goto out;
1715 }
1716
1717 /* we know that it is or will be overwritten. check this now */
1718 if (ow_inode < sctx->send_progress)
1719 ret = 1;
1720 else
1721 ret = 0;
1722
1723out:
1724 return ret;
1725}
1726
766702ef
AB		 1727/*
1728 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1729 * that got overwritten. This is used by process_recorded_refs to determine
1730 * if it has to use the path as returned by get_cur_path or the orphan name.
1731 */
31db9f7c
AB		 1732static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1733{
1734 int ret = 0;
1735 struct fs_path *name = NULL;
1736 u64 dir;
1737 u64 dir_gen;
1738
1739 if (!sctx->parent_root)
1740 goto out;
1741
1742 name = fs_path_alloc(sctx);
1743 if (!name)
1744 return -ENOMEM;
1745
1746 ret = get_first_ref(sctx, sctx->parent_root, ino, &dir, &dir_gen, name);
1747 if (ret < 0)
1748 goto out;
1749
1750 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1751 name->start, fs_path_len(name));
31db9f7c
AB		 1752
1753out:
1754 fs_path_free(sctx, name);
1755 return ret;
1756}
1757
766702ef
AB		 1758/*
1759 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1760 * so we need to do some special handling in case we have clashes. This function
1761 * takes care of this with the help of name_cache_entry::radix_list.
5dc67d0b 1762 * In case of error, nce is kfreed.
766702ef 1763 */
31db9f7c
AB		 1764static int name_cache_insert(struct send_ctx *sctx,
1765 struct name_cache_entry *nce)
1766{
1767 int ret = 0;
7e0926fe
AB		 1768 struct list_head *nce_head;
1769
1770 nce_head = radix_tree_lookup(&sctx->name_cache,
1771 (unsigned long)nce->ino);
1772 if (!nce_head) {
1773 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1774 if (!nce_head)
31db9f7c 1775 return -ENOMEM;
7e0926fe 1776 INIT_LIST_HEAD(nce_head);
31db9f7c 1777
7e0926fe 1778 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
5dc67d0b
AB		 1779 if (ret < 0) {
1780 kfree(nce_head);
1781 kfree(nce);
31db9f7c 1782 return ret;
5dc67d0b 1783 }
31db9f7c 1784 }
7e0926fe 1785 list_add_tail(&nce->radix_list, nce_head);
31db9f7c
AB		 1786 list_add_tail(&nce->list, &sctx->name_cache_list);
1787 sctx->name_cache_size++;
1788
1789 return ret;
1790}
1791
1792static void name_cache_delete(struct send_ctx *sctx,
1793 struct name_cache_entry *nce)
1794{
7e0926fe 1795 struct list_head *nce_head;
31db9f7c 1796
7e0926fe
AB		 1797 nce_head = radix_tree_lookup(&sctx->name_cache,
1798 (unsigned long)nce->ino);
1799 BUG_ON(!nce_head);
31db9f7c 1800
7e0926fe 1801 list_del(&nce->radix_list);
31db9f7c 1802 list_del(&nce->list);
31db9f7c 1803 sctx->name_cache_size--;
7e0926fe
AB		 1804
1805 if (list_empty(nce_head)) {
1806 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1807 kfree(nce_head);
1808 }
31db9f7c
AB		 1809}
1810
1811static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1812 u64 ino, u64 gen)
1813{
7e0926fe
AB		 1814 struct list_head *nce_head;
1815 struct name_cache_entry *cur;
31db9f7c 1816
7e0926fe
AB		 1817 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1818 if (!nce_head)
31db9f7c
AB		 1819 return NULL;
1820
7e0926fe
AB		 1821 list_for_each_entry(cur, nce_head, radix_list) {
1822 if (cur->ino == ino && cur->gen == gen)
1823 return cur;
1824 }
31db9f7c
AB		 1825 return NULL;
1826}
1827
766702ef
AB		 1828/*
1829 * Removes the entry from the list and adds it back to the end. This marks the
1830 * entry as recently used so that name_cache_clean_unused does not remove it.
1831 */
31db9f7c
AB		 1832static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1833{
1834 list_del(&nce->list);
1835 list_add_tail(&nce->list, &sctx->name_cache_list);
1836}
1837
766702ef
AB		 1838/*
1839 * Remove some entries from the beginning of name_cache_list.
1840 */
31db9f7c
AB		 1841static void name_cache_clean_unused(struct send_ctx *sctx)
1842{
1843 struct name_cache_entry *nce;
1844
1845 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1846 return;
1847
1848 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1849 nce = list_entry(sctx->name_cache_list.next,
1850 struct name_cache_entry, list);
1851 name_cache_delete(sctx, nce);
1852 kfree(nce);
1853 }
1854}
1855
1856static void name_cache_free(struct send_ctx *sctx)
1857{
1858 struct name_cache_entry *nce;
31db9f7c 1859
e938c8ad
AB		 1860 while (!list_empty(&sctx->name_cache_list)) {
1861 nce = list_entry(sctx->name_cache_list.next,
1862 struct name_cache_entry, list);
31db9f7c 1863 name_cache_delete(sctx, nce);
17589bd9 1864 kfree(nce);
31db9f7c
AB		 1865 }
1866}
1867
766702ef
AB		 1868/*
1869 * Used by get_cur_path for each ref up to the root.
1870 * Returns 0 if it succeeded.
1871 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1872 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1873 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1874 * Returns <0 in case of error.
1875 */
31db9f7c
AB		 1876static int __get_cur_name_and_parent(struct send_ctx *sctx,
1877 u64 ino, u64 gen,
1878 u64 *parent_ino,
1879 u64 *parent_gen,
1880 struct fs_path *dest)
1881{
1882 int ret;
1883 int nce_ret;
1884 struct btrfs_path *path = NULL;
1885 struct name_cache_entry *nce = NULL;
1886
766702ef
AB		 1887 /*
1888 * First check if we already did a call to this function with the same
1889 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1890 * return the cached result.
1891 */
31db9f7c
AB		 1892 nce = name_cache_search(sctx, ino, gen);
1893 if (nce) {
1894 if (ino < sctx->send_progress && nce->need_later_update) {
1895 name_cache_delete(sctx, nce);
1896 kfree(nce);
1897 nce = NULL;
1898 } else {
1899 name_cache_used(sctx, nce);
1900 *parent_ino = nce->parent_ino;
1901 *parent_gen = nce->parent_gen;
1902 ret = fs_path_add(dest, nce->name, nce->name_len);
1903 if (ret < 0)
1904 goto out;
1905 ret = nce->ret;
1906 goto out;
1907 }
1908 }
1909
1910 path = alloc_path_for_send();
1911 if (!path)
1912 return -ENOMEM;
1913
766702ef
AB		 1914 /*
1915 * If the inode is not existent yet, add the orphan name and return 1.
1916 * This should only happen for the parent dir that we determine in
1917 * __record_new_ref
1918 */
31db9f7c
AB		 1919 ret = is_inode_existent(sctx, ino, gen);
1920 if (ret < 0)
1921 goto out;
1922
1923 if (!ret) {
1924 ret = gen_unique_name(sctx, ino, gen, dest);
1925 if (ret < 0)
1926 goto out;
1927 ret = 1;
1928 goto out_cache;
1929 }
1930
766702ef
AB		 1931 /*
1932 * Depending on whether the inode was already processed or not, use
1933 * send_root or parent_root for ref lookup.
1934 */
31db9f7c
AB		 1935 if (ino < sctx->send_progress)
1936 ret = get_first_ref(sctx, sctx->send_root, ino,
1937 parent_ino, parent_gen, dest);
1938 else
1939 ret = get_first_ref(sctx, sctx->parent_root, ino,
1940 parent_ino, parent_gen, dest);
1941 if (ret < 0)
1942 goto out;
1943
766702ef
AB		 1944 /*
1945 * Check if the ref was overwritten by an inode's ref that was processed
1946 * earlier. If yes, treat as orphan and return 1.
1947 */
31db9f7c
AB		 1948 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
1949 dest->start, dest->end - dest->start);
1950 if (ret < 0)
1951 goto out;
1952 if (ret) {
1953 fs_path_reset(dest);
1954 ret = gen_unique_name(sctx, ino, gen, dest);
1955 if (ret < 0)
1956 goto out;
1957 ret = 1;
1958 }
1959
1960out_cache:
766702ef
AB		 1961 /*
1962 * Store the result of the lookup in the name cache.
1963 */
31db9f7c
AB		 1964 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
1965 if (!nce) {
1966 ret = -ENOMEM;
1967 goto out;
1968 }
1969
1970 nce->ino = ino;
1971 nce->gen = gen;
1972 nce->parent_ino = *parent_ino;
1973 nce->parent_gen = *parent_gen;
1974 nce->name_len = fs_path_len(dest);
1975 nce->ret = ret;
1976 strcpy(nce->name, dest->start);
31db9f7c
AB		 1977
1978 if (ino < sctx->send_progress)
1979 nce->need_later_update = 0;
1980 else
1981 nce->need_later_update = 1;
1982
1983 nce_ret = name_cache_insert(sctx, nce);
1984 if (nce_ret < 0)
1985 ret = nce_ret;
1986 name_cache_clean_unused(sctx);
1987
1988out:
1989 btrfs_free_path(path);
1990 return ret;
1991}
1992
1993/*
1994 * Magic happens here. This function returns the first ref to an inode as it
1995 * would look like while receiving the stream at this point in time.
1996 * We walk the path up to the root. For every inode in between, we check if it
1997 * was already processed/sent. If yes, we continue with the parent as found
1998 * in send_root. If not, we continue with the parent as found in parent_root.
1999 * If we encounter an inode that was deleted at this point in time, we use the
2000 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2001 * that were not created yet and overwritten inodes/refs.
2002 *
2003 * When do we have have orphan inodes:
2004 * 1. When an inode is freshly created and thus no valid refs are available yet
2005 * 2. When a directory lost all it's refs (deleted) but still has dir items
2006 * inside which were not processed yet (pending for move/delete). If anyone
2007 * tried to get the path to the dir items, it would get a path inside that
2008 * orphan directory.
2009 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2010 * of an unprocessed inode. If in that case the first ref would be
2011 * overwritten, the overwritten inode gets "orphanized". Later when we
2012 * process this overwritten inode, it is restored at a new place by moving
2013 * the orphan inode.
2014 *
2015 * sctx->send_progress tells this function at which point in time receiving
2016 * would be.
2017 */
2018static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2019 struct fs_path *dest)
2020{
2021 int ret = 0;
2022 struct fs_path *name = NULL;
2023 u64 parent_inode = 0;
2024 u64 parent_gen = 0;
2025 int stop = 0;
2026
2027 name = fs_path_alloc(sctx);
2028 if (!name) {
2029 ret = -ENOMEM;
2030 goto out;
2031 }
2032
2033 dest->reversed = 1;
2034 fs_path_reset(dest);
2035
2036 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2037 fs_path_reset(name);
2038
2039 ret = __get_cur_name_and_parent(sctx, ino, gen,
2040 &parent_inode, &parent_gen, name);
2041 if (ret < 0)
2042 goto out;
2043 if (ret)
2044 stop = 1;
2045
2046 ret = fs_path_add_path(dest, name);
2047 if (ret < 0)
2048 goto out;
2049
2050 ino = parent_inode;
2051 gen = parent_gen;
2052 }
2053
2054out:
2055 fs_path_free(sctx, name);
2056 if (!ret)
2057 fs_path_unreverse(dest);
2058 return ret;
2059}
2060
2061/*
2062 * Called for regular files when sending extents data. Opens a struct file
2063 * to read from the file.
2064 */
2065static int open_cur_inode_file(struct send_ctx *sctx)
2066{
2067 int ret = 0;
2068 struct btrfs_key key;
e2aed8df 2069 struct path path;
31db9f7c
AB		 2070 struct inode *inode;
2071 struct dentry *dentry;
2072 struct file *filp;
2073 int new = 0;
2074
2075 if (sctx->cur_inode_filp)
2076 goto out;
2077
2078 key.objectid = sctx->cur_ino;
2079 key.type = BTRFS_INODE_ITEM_KEY;
2080 key.offset = 0;
2081
2082 inode = btrfs_iget(sctx->send_root->fs_info->sb, &key, sctx->send_root,
2083 &new);
2084 if (IS_ERR(inode)) {
2085 ret = PTR_ERR(inode);
2086 goto out;
2087 }
2088
2089 dentry = d_obtain_alias(inode);
2090 inode = NULL;
2091 if (IS_ERR(dentry)) {
2092 ret = PTR_ERR(dentry);
2093 goto out;
2094 }
2095
e2aed8df
LT		 2096 path.mnt = sctx->mnt;
2097 path.dentry = dentry;
2098 filp = dentry_open(&path, O_RDONLY | O_LARGEFILE, current_cred());
2099 dput(dentry);
31db9f7c 2100 dentry = NULL;
31db9f7c
AB		 2101 if (IS_ERR(filp)) {
2102 ret = PTR_ERR(filp);
2103 goto out;
2104 }
2105 sctx->cur_inode_filp = filp;
2106
2107out:
2108 /*
2109 * no xxxput required here as every vfs op
2110 * does it by itself on failure
2111 */
2112 return ret;
2113}
2114
2115/*
2116 * Closes the struct file that was created in open_cur_inode_file
2117 */
2118static int close_cur_inode_file(struct send_ctx *sctx)
2119{
2120 int ret = 0;
2121
2122 if (!sctx->cur_inode_filp)
2123 goto out;
2124
2125 ret = filp_close(sctx->cur_inode_filp, NULL);
2126 sctx->cur_inode_filp = NULL;
2127
2128out:
2129 return ret;
2130}
2131
2132/*
2133 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2134 */
2135static int send_subvol_begin(struct send_ctx *sctx)
2136{
2137 int ret;
2138 struct btrfs_root *send_root = sctx->send_root;
2139 struct btrfs_root *parent_root = sctx->parent_root;
2140 struct btrfs_path *path;
2141 struct btrfs_key key;
2142 struct btrfs_root_ref *ref;
2143 struct extent_buffer *leaf;
2144 char *name = NULL;
2145 int namelen;
2146
2147 path = alloc_path_for_send();
2148 if (!path)
2149 return -ENOMEM;
2150
2151 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2152 if (!name) {
2153 btrfs_free_path(path);
2154 return -ENOMEM;
2155 }
2156
2157 key.objectid = send_root->objectid;
2158 key.type = BTRFS_ROOT_BACKREF_KEY;
2159 key.offset = 0;
2160
2161 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2162 &key, path, 1, 0);
2163 if (ret < 0)
2164 goto out;
2165 if (ret) {
2166 ret = -ENOENT;
2167 goto out;
2168 }
2169
2170 leaf = path->nodes[0];
2171 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2172 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2173 key.objectid != send_root->objectid) {
2174 ret = -ENOENT;
2175 goto out;
2176 }
2177 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2178 namelen = btrfs_root_ref_name_len(leaf, ref);
2179 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2180 btrfs_release_path(path);
2181
31db9f7c
AB		 2182 if (parent_root) {
2183 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2184 if (ret < 0)
2185 goto out;
2186 } else {
2187 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2188 if (ret < 0)
2189 goto out;
2190 }
2191
2192 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2193 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2194 sctx->send_root->root_item.uuid);
2195 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2196 sctx->send_root->root_item.ctransid);
2197 if (parent_root) {
2198 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2199 sctx->parent_root->root_item.uuid);
2200 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2201 sctx->parent_root->root_item.ctransid);
2202 }
2203
2204 ret = send_cmd(sctx);
2205
2206tlv_put_failure:
2207out:
2208 btrfs_free_path(path);
2209 kfree(name);
2210 return ret;
2211}
2212
2213static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2214{
2215 int ret = 0;
2216 struct fs_path *p;
2217
2218verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2219
2220 p = fs_path_alloc(sctx);
2221 if (!p)
2222 return -ENOMEM;
2223
2224 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2225 if (ret < 0)
2226 goto out;
2227
2228 ret = get_cur_path(sctx, ino, gen, p);
2229 if (ret < 0)
2230 goto out;
2231 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2232 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2233
2234 ret = send_cmd(sctx);
2235
2236tlv_put_failure:
2237out:
2238 fs_path_free(sctx, p);
2239 return ret;
2240}
2241
2242static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2243{
2244 int ret = 0;
2245 struct fs_path *p;
2246
2247verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2248
2249 p = fs_path_alloc(sctx);
2250 if (!p)
2251 return -ENOMEM;
2252
2253 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2254 if (ret < 0)
2255 goto out;
2256
2257 ret = get_cur_path(sctx, ino, gen, p);
2258 if (ret < 0)
2259 goto out;
2260 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2261 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2262
2263 ret = send_cmd(sctx);
2264
2265tlv_put_failure:
2266out:
2267 fs_path_free(sctx, p);
2268 return ret;
2269}
2270
2271static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2272{
2273 int ret = 0;
2274 struct fs_path *p;
2275
2276verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2277
2278 p = fs_path_alloc(sctx);
2279 if (!p)
2280 return -ENOMEM;
2281
2282 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2283 if (ret < 0)
2284 goto out;
2285
2286 ret = get_cur_path(sctx, ino, gen, p);
2287 if (ret < 0)
2288 goto out;
2289 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2290 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2291 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2292
2293 ret = send_cmd(sctx);
2294
2295tlv_put_failure:
2296out:
2297 fs_path_free(sctx, p);
2298 return ret;
2299}
2300
2301static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2302{
2303 int ret = 0;
2304 struct fs_path *p = NULL;
2305 struct btrfs_inode_item *ii;
2306 struct btrfs_path *path = NULL;
2307 struct extent_buffer *eb;
2308 struct btrfs_key key;
2309 int slot;
2310
2311verbose_printk("btrfs: send_utimes %llu\n", ino);
2312
2313 p = fs_path_alloc(sctx);
2314 if (!p)
2315 return -ENOMEM;
2316
2317 path = alloc_path_for_send();
2318 if (!path) {
2319 ret = -ENOMEM;
2320 goto out;
2321 }
2322
2323 key.objectid = ino;
2324 key.type = BTRFS_INODE_ITEM_KEY;
2325 key.offset = 0;
2326 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2327 if (ret < 0)
2328 goto out;
2329
2330 eb = path->nodes[0];
2331 slot = path->slots[0];
2332 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2333
2334 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2335 if (ret < 0)
2336 goto out;
2337
2338 ret = get_cur_path(sctx, ino, gen, p);
2339 if (ret < 0)
2340 goto out;
2341 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2342 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2343 btrfs_inode_atime(ii));
2344 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2345 btrfs_inode_mtime(ii));
2346 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2347 btrfs_inode_ctime(ii));
766702ef 2348 /* TODO Add otime support when the otime patches get into upstream */
31db9f7c
AB		 2349
2350 ret = send_cmd(sctx);
2351
2352tlv_put_failure:
2353out:
2354 fs_path_free(sctx, p);
2355 btrfs_free_path(path);
2356 return ret;
2357}
2358
2359/*
2360 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2361 * a valid path yet because we did not process the refs yet. So, the inode
2362 * is created as orphan.
2363 */
1f4692da 2364static int send_create_inode(struct send_ctx *sctx, u64 ino)
31db9f7c
AB		 2365{
2366 int ret = 0;
31db9f7c 2367 struct fs_path *p;
31db9f7c 2368 int cmd;
1f4692da 2369 u64 gen;
31db9f7c 2370 u64 mode;
1f4692da 2371 u64 rdev;
31db9f7c 2372
1f4692da 2373verbose_printk("btrfs: send_create_inode %llu\n", ino);
31db9f7c
AB		 2374
2375 p = fs_path_alloc(sctx);
2376 if (!p)
2377 return -ENOMEM;
2378
1f4692da
AB		 2379 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2380 NULL, &rdev);
2381 if (ret < 0)
2382 goto out;
31db9f7c 2383
e938c8ad 2384 if (S_ISREG(mode)) {
31db9f7c 2385 cmd = BTRFS_SEND_C_MKFILE;
e938c8ad 2386 } else if (S_ISDIR(mode)) {
31db9f7c 2387 cmd = BTRFS_SEND_C_MKDIR;
e938c8ad 2388 } else if (S_ISLNK(mode)) {
31db9f7c 2389 cmd = BTRFS_SEND_C_SYMLINK;
e938c8ad 2390 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
31db9f7c 2391 cmd = BTRFS_SEND_C_MKNOD;
e938c8ad 2392 } else if (S_ISFIFO(mode)) {
31db9f7c 2393 cmd = BTRFS_SEND_C_MKFIFO;
e938c8ad 2394 } else if (S_ISSOCK(mode)) {
31db9f7c 2395 cmd = BTRFS_SEND_C_MKSOCK;
e938c8ad 2396 } else {
31db9f7c
AB		 2397 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2398 (int)(mode & S_IFMT));
2399 ret = -ENOTSUPP;
2400 goto out;
2401 }
2402
2403 ret = begin_cmd(sctx, cmd);
2404 if (ret < 0)
2405 goto out;
2406
1f4692da 2407 ret = gen_unique_name(sctx, ino, gen, p);
31db9f7c
AB		 2408 if (ret < 0)
2409 goto out;
2410
2411 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
1f4692da 2412 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
31db9f7c
AB		 2413
2414 if (S_ISLNK(mode)) {
2415 fs_path_reset(p);
1f4692da 2416 ret = read_symlink(sctx, sctx->send_root, ino, p);
31db9f7c
AB		 2417 if (ret < 0)
2418 goto out;
2419 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2420 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2421 S_ISFIFO(mode) || S_ISSOCK(mode)) {
1f4692da 2422 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, rdev);
31db9f7c
AB		 2423 }
2424
2425 ret = send_cmd(sctx);
2426 if (ret < 0)
2427 goto out;
2428
2429
2430tlv_put_failure:
2431out:
2432 fs_path_free(sctx, p);
2433 return ret;
2434}
2435
1f4692da
AB		 2436/*
2437 * We need some special handling for inodes that get processed before the parent
2438 * directory got created. See process_recorded_refs for details.
2439 * This function does the check if we already created the dir out of order.
2440 */
2441static int did_create_dir(struct send_ctx *sctx, u64 dir)
2442{
2443 int ret = 0;
2444 struct btrfs_path *path = NULL;
2445 struct btrfs_key key;
2446 struct btrfs_key found_key;
2447 struct btrfs_key di_key;
2448 struct extent_buffer *eb;
2449 struct btrfs_dir_item *di;
2450 int slot;
2451
2452 path = alloc_path_for_send();
2453 if (!path) {
2454 ret = -ENOMEM;
2455 goto out;
2456 }
2457
2458 key.objectid = dir;
2459 key.type = BTRFS_DIR_INDEX_KEY;
2460 key.offset = 0;
2461 while (1) {
2462 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2463 1, 0);
2464 if (ret < 0)
2465 goto out;
2466 if (!ret) {
2467 eb = path->nodes[0];
2468 slot = path->slots[0];
2469 btrfs_item_key_to_cpu(eb, &found_key, slot);
2470 }
2471 if (ret || found_key.objectid != key.objectid ||
2472 found_key.type != key.type) {
2473 ret = 0;
2474 goto out;
2475 }
2476
2477 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2478 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2479
2480 if (di_key.objectid < sctx->send_progress) {
2481 ret = 1;
2482 goto out;
2483 }
2484
2485 key.offset = found_key.offset + 1;
2486 btrfs_release_path(path);
2487 }
2488
2489out:
2490 btrfs_free_path(path);
2491 return ret;
2492}
2493
2494/*
2495 * Only creates the inode if it is:
2496 * 1. Not a directory
2497 * 2. Or a directory which was not created already due to out of order
2498 * directories. See did_create_dir and process_recorded_refs for details.
2499 */
2500static int send_create_inode_if_needed(struct send_ctx *sctx)
2501{
2502 int ret;
2503
2504 if (S_ISDIR(sctx->cur_inode_mode)) {
2505 ret = did_create_dir(sctx, sctx->cur_ino);
2506 if (ret < 0)
2507 goto out;
2508 if (ret) {
2509 ret = 0;
2510 goto out;
2511 }
2512 }
2513
2514 ret = send_create_inode(sctx, sctx->cur_ino);
2515 if (ret < 0)
2516 goto out;
2517
2518out:
2519 return ret;
2520}
2521
31db9f7c
AB		 2522struct recorded_ref {
2523 struct list_head list;
2524 char *dir_path;
2525 char *name;
2526 struct fs_path *full_path;
2527 u64 dir;
2528 u64 dir_gen;
2529 int dir_path_len;
2530 int name_len;
2531};
2532
2533/*
2534 * We need to process new refs before deleted refs, but compare_tree gives us
2535 * everything mixed. So we first record all refs and later process them.
2536 * This function is a helper to record one ref.
2537 */
2538static int record_ref(struct list_head *head, u64 dir,
2539 u64 dir_gen, struct fs_path *path)
2540{
2541 struct recorded_ref *ref;
2542 char *tmp;
2543
2544 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2545 if (!ref)
2546 return -ENOMEM;
2547
2548 ref->dir = dir;
2549 ref->dir_gen = dir_gen;
2550 ref->full_path = path;
2551
2552 tmp = strrchr(ref->full_path->start, '/');
2553 if (!tmp) {
2554 ref->name_len = ref->full_path->end - ref->full_path->start;
2555 ref->name = ref->full_path->start;
2556 ref->dir_path_len = 0;
2557 ref->dir_path = ref->full_path->start;
2558 } else {
2559 tmp++;
2560 ref->name_len = ref->full_path->end - tmp;
2561 ref->name = tmp;
2562 ref->dir_path = ref->full_path->start;
2563 ref->dir_path_len = ref->full_path->end -
2564 ref->full_path->start - 1 - ref->name_len;
2565 }
2566
2567 list_add_tail(&ref->list, head);
2568 return 0;
2569}
2570
2571static void __free_recorded_refs(struct send_ctx *sctx, struct list_head *head)
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);
31db9f7c 2577 fs_path_free(sctx, 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{
2585 __free_recorded_refs(sctx, &sctx->new_refs);
2586 __free_recorded_refs(sctx, &sctx->deleted_refs);
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
2600 orphan = fs_path_alloc(sctx);
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:
2611 fs_path_free(sctx, orphan);
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
2630 path = alloc_path_for_send();
2631 if (!path)
2632 return -ENOMEM;
2633
2634 key.objectid = dir;
2635 key.type = BTRFS_DIR_INDEX_KEY;
2636 key.offset = 0;
2637
2638 while (1) {
2639 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2640 if (ret < 0)
2641 goto out;
2642 if (!ret) {
2643 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2644 path->slots[0]);
2645 }
2646 if (ret || found_key.objectid != key.objectid ||
2647 found_key.type != key.type) {
2648 break;
2649 }
2650
2651 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2652 struct btrfs_dir_item);
2653 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2654
2655 if (loc.objectid > send_progress) {
2656 ret = 0;
2657 goto out;
2658 }
2659
2660 btrfs_release_path(path);
2661 key.offset = found_key.offset + 1;
2662 }
2663
2664 ret = 1;
2665
2666out:
2667 btrfs_free_path(path);
2668 return ret;
2669}
2670
31db9f7c
AB		 2671/*
2672 * This does all the move/link/unlink/rmdir magic.
2673 */
2674static int process_recorded_refs(struct send_ctx *sctx)
2675{
2676 int ret = 0;
2677 struct recorded_ref *cur;
1f4692da 2678 struct recorded_ref *cur2;
31db9f7c
AB		 2679 struct ulist *check_dirs = NULL;
2680 struct ulist_iterator uit;
2681 struct ulist_node *un;
2682 struct fs_path *valid_path = NULL;
b24baf69 2683 u64 ow_inode = 0;
31db9f7c
AB		 2684 u64 ow_gen;
2685 int did_overwrite = 0;
2686 int is_orphan = 0;
2687
2688verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2689
2690 valid_path = fs_path_alloc(sctx);
2691 if (!valid_path) {
2692 ret = -ENOMEM;
2693 goto out;
2694 }
2695
2696 check_dirs = ulist_alloc(GFP_NOFS);
2697 if (!check_dirs) {
2698 ret = -ENOMEM;
2699 goto out;
2700 }
2701
2702 /*
2703 * First, check if the first ref of the current inode was overwritten
2704 * before. If yes, we know that the current inode was already orphanized
2705 * and thus use the orphan name. If not, we can use get_cur_path to
2706 * get the path of the first ref as it would like while receiving at
2707 * this point in time.
2708 * New inodes are always orphan at the beginning, so force to use the
2709 * orphan name in this case.
2710 * The first ref is stored in valid_path and will be updated if it
2711 * gets moved around.
2712 */
2713 if (!sctx->cur_inode_new) {
2714 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2715 sctx->cur_inode_gen);
2716 if (ret < 0)
2717 goto out;
2718 if (ret)
2719 did_overwrite = 1;
2720 }
2721 if (sctx->cur_inode_new || did_overwrite) {
2722 ret = gen_unique_name(sctx, sctx->cur_ino,
2723 sctx->cur_inode_gen, valid_path);
2724 if (ret < 0)
2725 goto out;
2726 is_orphan = 1;
2727 } else {
2728 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2729 valid_path);
2730 if (ret < 0)
2731 goto out;
2732 }
2733
2734 list_for_each_entry(cur, &sctx->new_refs, list) {
1f4692da
AB		 2735 /*
2736 * We may have refs where the parent directory does not exist
2737 * yet. This happens if the parent directories inum is higher
2738 * the the current inum. To handle this case, we create the
2739 * parent directory out of order. But we need to check if this
2740 * did already happen before due to other refs in the same dir.
2741 */
2742 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
2743 if (ret < 0)
2744 goto out;
2745 if (ret == inode_state_will_create) {
2746 ret = 0;
2747 /*
2748 * First check if any of the current inodes refs did
2749 * already create the dir.
2750 */
2751 list_for_each_entry(cur2, &sctx->new_refs, list) {
2752 if (cur == cur2)
2753 break;
2754 if (cur2->dir == cur->dir) {
2755 ret = 1;
2756 break;
2757 }
2758 }
2759
2760 /*
2761 * If that did not happen, check if a previous inode
2762 * did already create the dir.
2763 */
2764 if (!ret)
2765 ret = did_create_dir(sctx, cur->dir);
2766 if (ret < 0)
2767 goto out;
2768 if (!ret) {
2769 ret = send_create_inode(sctx, cur->dir);
2770 if (ret < 0)
2771 goto out;
2772 }
2773 }
2774
31db9f7c
AB		 2775 /*
2776 * Check if this new ref would overwrite the first ref of
2777 * another unprocessed inode. If yes, orphanize the
2778 * overwritten inode. If we find an overwritten ref that is
2779 * not the first ref, simply unlink it.
2780 */
2781 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2782 cur->name, cur->name_len,
2783 &ow_inode, &ow_gen);
2784 if (ret < 0)
2785 goto out;
2786 if (ret) {
2787 ret = is_first_ref(sctx, sctx->parent_root,
2788 ow_inode, cur->dir, cur->name,
2789 cur->name_len);
2790 if (ret < 0)
2791 goto out;
2792 if (ret) {
2793 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2794 cur->full_path);
2795 if (ret < 0)
2796 goto out;
2797 } else {
2798 ret = send_unlink(sctx, cur->full_path);
2799 if (ret < 0)
2800 goto out;
2801 }
2802 }
2803
2804 /*
2805 * link/move the ref to the new place. If we have an orphan
2806 * inode, move it and update valid_path. If not, link or move
2807 * it depending on the inode mode.
2808 */
1f4692da 2809 if (is_orphan) {
31db9f7c
AB		 2810 ret = send_rename(sctx, valid_path, cur->full_path);
2811 if (ret < 0)
2812 goto out;
2813 is_orphan = 0;
2814 ret = fs_path_copy(valid_path, cur->full_path);
2815 if (ret < 0)
2816 goto out;
2817 } else {
2818 if (S_ISDIR(sctx->cur_inode_mode)) {
2819 /*
2820 * Dirs can't be linked, so move it. For moved
2821 * dirs, we always have one new and one deleted
2822 * ref. The deleted ref is ignored later.
2823 */
2824 ret = send_rename(sctx, valid_path,
2825 cur->full_path);
2826 if (ret < 0)
2827 goto out;
2828 ret = fs_path_copy(valid_path, cur->full_path);
2829 if (ret < 0)
2830 goto out;
2831 } else {
2832 ret = send_link(sctx, cur->full_path,
2833 valid_path);
2834 if (ret < 0)
2835 goto out;
2836 }
2837 }
2838 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2839 GFP_NOFS);
2840 if (ret < 0)
2841 goto out;
2842 }
2843
2844 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2845 /*
2846 * Check if we can already rmdir the directory. If not,
2847 * orphanize it. For every dir item inside that gets deleted
2848 * later, we do this check again and rmdir it then if possible.
2849 * See the use of check_dirs for more details.
2850 */
2851 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2852 if (ret < 0)
2853 goto out;
2854 if (ret) {
2855 ret = send_rmdir(sctx, valid_path);
2856 if (ret < 0)
2857 goto out;
2858 } else if (!is_orphan) {
2859 ret = orphanize_inode(sctx, sctx->cur_ino,
2860 sctx->cur_inode_gen, valid_path);
2861 if (ret < 0)
2862 goto out;
2863 is_orphan = 1;
2864 }
2865
2866 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2867 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2868 GFP_NOFS);
2869 if (ret < 0)
2870 goto out;
2871 }
ccf1626b
AB		 2872 } else if (S_ISDIR(sctx->cur_inode_mode) &&
2873 !list_empty(&sctx->deleted_refs)) {
2874 /*
2875 * We have a moved dir. Add the old parent to check_dirs
2876 */
2877 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
2878 list);
2879 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2880 GFP_NOFS);
2881 if (ret < 0)
2882 goto out;
31db9f7c
AB		 2883 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2884 /*
2885 * We have a non dir inode. Go through all deleted refs and
2886 * unlink them if they were not already overwritten by other
2887 * inodes.
2888 */
2889 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2890 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2891 sctx->cur_ino, sctx->cur_inode_gen,
2892 cur->name, cur->name_len);
2893 if (ret < 0)
2894 goto out;
2895 if (!ret) {
1f4692da
AB		 2896 ret = send_unlink(sctx, cur->full_path);
2897 if (ret < 0)
2898 goto out;
31db9f7c
AB		 2899 }
2900 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2901 GFP_NOFS);
2902 if (ret < 0)
2903 goto out;
2904 }
2905
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 */
2927 ULIST_ITER_INIT(&uit);
2928 while ((un = ulist_next(check_dirs, &uit))) {
766702ef
AB		 2929 /*
2930 * In case we had refs into dirs that were not processed yet,
2931 * we don't need to do the utime and rmdir logic for these dirs.
2932 * The dir will be processed later.
2933 */
31db9f7c
AB		 2934 if (un->val > sctx->cur_ino)
2935 continue;
2936
2937 ret = get_cur_inode_state(sctx, un->val, un->aux);
2938 if (ret < 0)
2939 goto out;
2940
2941 if (ret == inode_state_did_create ||
2942 ret == inode_state_no_change) {
2943 /* TODO delayed utimes */
2944 ret = send_utimes(sctx, un->val, un->aux);
2945 if (ret < 0)
2946 goto out;
2947 } else if (ret == inode_state_did_delete) {
2948 ret = can_rmdir(sctx, un->val, sctx->cur_ino);
2949 if (ret < 0)
2950 goto out;
2951 if (ret) {
2952 ret = get_cur_path(sctx, un->val, un->aux,
2953 valid_path);
2954 if (ret < 0)
2955 goto out;
2956 ret = send_rmdir(sctx, valid_path);
2957 if (ret < 0)
2958 goto out;
2959 }
2960 }
2961 }
2962
31db9f7c
AB		 2963 ret = 0;
2964
2965out:
2966 free_recorded_refs(sctx);
2967 ulist_free(check_dirs);
2968 fs_path_free(sctx, valid_path);
2969 return ret;
2970}
2971
2972static int __record_new_ref(int num, u64 dir, int index,
2973 struct fs_path *name,
2974 void *ctx)
2975{
2976 int ret = 0;
2977 struct send_ctx *sctx = ctx;
2978 struct fs_path *p;
2979 u64 gen;
2980
2981 p = fs_path_alloc(sctx);
2982 if (!p)
2983 return -ENOMEM;
2984
2985 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 2986 NULL, NULL);
31db9f7c
AB		 2987 if (ret < 0)
2988 goto out;
2989
31db9f7c
AB		 2990 ret = get_cur_path(sctx, dir, gen, p);
2991 if (ret < 0)
2992 goto out;
2993 ret = fs_path_add_path(p, name);
2994 if (ret < 0)
2995 goto out;
2996
2997 ret = record_ref(&sctx->new_refs, dir, gen, p);
2998
2999out:
3000 if (ret)
3001 fs_path_free(sctx, p);
3002 return ret;
3003}
3004
3005static int __record_deleted_ref(int num, u64 dir, int index,
3006 struct fs_path *name,
3007 void *ctx)
3008{
3009 int ret = 0;
3010 struct send_ctx *sctx = ctx;
3011 struct fs_path *p;
3012 u64 gen;
3013
3014 p = fs_path_alloc(sctx);
3015 if (!p)
3016 return -ENOMEM;
3017
3018 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
85a7b33b 3019 NULL, NULL);
31db9f7c
AB		 3020 if (ret < 0)
3021 goto out;
3022
3023 ret = get_cur_path(sctx, dir, gen, p);
3024 if (ret < 0)
3025 goto out;
3026 ret = fs_path_add_path(p, name);
3027 if (ret < 0)
3028 goto out;
3029
3030 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3031
3032out:
3033 if (ret)
3034 fs_path_free(sctx, p);
3035 return ret;
3036}
3037
3038static int record_new_ref(struct send_ctx *sctx)
3039{
3040 int ret;
3041
3042 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3043 sctx->cmp_key, 0, __record_new_ref, sctx);
3044 if (ret < 0)
3045 goto out;
3046 ret = 0;
3047
3048out:
3049 return ret;
3050}
3051
3052static int record_deleted_ref(struct send_ctx *sctx)
3053{
3054 int ret;
3055
3056 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3057 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3058 if (ret < 0)
3059 goto out;
3060 ret = 0;
3061
3062out:
3063 return ret;
3064}
3065
3066struct find_ref_ctx {
3067 u64 dir;
3068 struct fs_path *name;
3069 int found_idx;
3070};
3071
3072static int __find_iref(int num, u64 dir, int index,
3073 struct fs_path *name,
3074 void *ctx_)
3075{
3076 struct find_ref_ctx *ctx = ctx_;
3077
3078 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3079 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3080 ctx->found_idx = num;
3081 return 1;
3082 }
3083 return 0;
3084}
3085
3086static int find_iref(struct send_ctx *sctx,
3087 struct btrfs_root *root,
3088 struct btrfs_path *path,
3089 struct btrfs_key *key,
3090 u64 dir, struct fs_path *name)
3091{
3092 int ret;
3093 struct find_ref_ctx ctx;
3094
3095 ctx.dir = dir;
3096 ctx.name = name;
3097 ctx.found_idx = -1;
3098
3099 ret = iterate_inode_ref(sctx, root, path, key, 0, __find_iref, &ctx);
3100 if (ret < 0)
3101 return ret;
3102
3103 if (ctx.found_idx == -1)
3104 return -ENOENT;
3105
3106 return ctx.found_idx;
3107}
3108
3109static int __record_changed_new_ref(int num, u64 dir, int index,
3110 struct fs_path *name,
3111 void *ctx)
3112{
3113 int ret;
3114 struct send_ctx *sctx = ctx;
3115
3116 ret = find_iref(sctx, sctx->parent_root, sctx->right_path,
3117 sctx->cmp_key, dir, name);
3118 if (ret == -ENOENT)
3119 ret = __record_new_ref(num, dir, index, name, sctx);
3120 else if (ret > 0)
3121 ret = 0;
3122
3123 return ret;
3124}
3125
3126static int __record_changed_deleted_ref(int num, u64 dir, int index,
3127 struct fs_path *name,
3128 void *ctx)
3129{
3130 int ret;
3131 struct send_ctx *sctx = ctx;
3132
3133 ret = find_iref(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3134 dir, name);
3135 if (ret == -ENOENT)
3136 ret = __record_deleted_ref(num, dir, index, name, sctx);
3137 else if (ret > 0)
3138 ret = 0;
3139
3140 return ret;
3141}
3142
3143static int record_changed_ref(struct send_ctx *sctx)
3144{
3145 int ret = 0;
3146
3147 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3148 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3149 if (ret < 0)
3150 goto out;
3151 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3152 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3153 if (ret < 0)
3154 goto out;
3155 ret = 0;
3156
3157out:
3158 return ret;
3159}
3160
3161/*
3162 * Record and process all refs at once. Needed when an inode changes the
3163 * generation number, which means that it was deleted and recreated.
3164 */
3165static int process_all_refs(struct send_ctx *sctx,
3166 enum btrfs_compare_tree_result cmd)
3167{
3168 int ret;
3169 struct btrfs_root *root;
3170 struct btrfs_path *path;
3171 struct btrfs_key key;
3172 struct btrfs_key found_key;
3173 struct extent_buffer *eb;
3174 int slot;
3175 iterate_inode_ref_t cb;
3176
3177 path = alloc_path_for_send();
3178 if (!path)
3179 return -ENOMEM;
3180
3181 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3182 root = sctx->send_root;
3183 cb = __record_new_ref;
3184 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3185 root = sctx->parent_root;
3186 cb = __record_deleted_ref;
3187 } else {
3188 BUG();
3189 }
3190
3191 key.objectid = sctx->cmp_key->objectid;
3192 key.type = BTRFS_INODE_REF_KEY;
3193 key.offset = 0;
3194 while (1) {
3195 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
e938c8ad 3196 if (ret < 0)
31db9f7c 3197 goto out;
e938c8ad 3198 if (ret)
31db9f7c 3199 break;
31db9f7c
AB		 3200
3201 eb = path->nodes[0];
3202 slot = path->slots[0];
3203 btrfs_item_key_to_cpu(eb, &found_key, slot);
3204
3205 if (found_key.objectid != key.objectid ||
e938c8ad 3206 found_key.type != key.type)
31db9f7c 3207 break;
31db9f7c
AB		 3208
3209 ret = iterate_inode_ref(sctx, sctx->parent_root, path,
3210 &found_key, 0, cb, sctx);
3211 btrfs_release_path(path);
3212 if (ret < 0)
3213 goto out;
3214
3215 key.offset = found_key.offset + 1;
3216 }
e938c8ad 3217 btrfs_release_path(path);
31db9f7c
AB		 3218
3219 ret = process_recorded_refs(sctx);
3220
3221out:
3222 btrfs_free_path(path);
3223 return ret;
3224}
3225
3226static int send_set_xattr(struct send_ctx *sctx,
3227 struct fs_path *path,
3228 const char *name, int name_len,
3229 const char *data, int data_len)
3230{
3231 int ret = 0;
3232
3233 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3234 if (ret < 0)
3235 goto out;
3236
3237 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3238 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3239 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3240
3241 ret = send_cmd(sctx);
3242
3243tlv_put_failure:
3244out:
3245 return ret;
3246}
3247
3248static int send_remove_xattr(struct send_ctx *sctx,
3249 struct fs_path *path,
3250 const char *name, int name_len)
3251{
3252 int ret = 0;
3253
3254 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3255 if (ret < 0)
3256 goto out;
3257
3258 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3259 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3260
3261 ret = send_cmd(sctx);
3262
3263tlv_put_failure:
3264out:
3265 return ret;
3266}
3267
3268static int __process_new_xattr(int num, struct btrfs_key *di_key,
3269 const char *name, int name_len,
3270 const char *data, int data_len,
3271 u8 type, void *ctx)
3272{
3273 int ret;
3274 struct send_ctx *sctx = ctx;
3275 struct fs_path *p;
3276 posix_acl_xattr_header dummy_acl;
3277
3278 p = fs_path_alloc(sctx);
3279 if (!p)
3280 return -ENOMEM;
3281
3282 /*
3283 * This hack is needed because empty acl's are stored as zero byte
3284 * data in xattrs. Problem with that is, that receiving these zero byte
3285 * acl's will fail later. To fix this, we send a dummy acl list that
3286 * only contains the version number and no entries.
3287 */
3288 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3289 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3290 if (data_len == 0) {
3291 dummy_acl.a_version =
3292 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3293 data = (char *)&dummy_acl;
3294 data_len = sizeof(dummy_acl);
3295 }
3296 }
3297
3298 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3299 if (ret < 0)
3300 goto out;
3301
3302 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3303
3304out:
3305 fs_path_free(sctx, p);
3306 return ret;
3307}
3308
3309static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3310 const char *name, int name_len,
3311 const char *data, int data_len,
3312 u8 type, void *ctx)
3313{
3314 int ret;
3315 struct send_ctx *sctx = ctx;
3316 struct fs_path *p;
3317
3318 p = fs_path_alloc(sctx);
3319 if (!p)
3320 return -ENOMEM;
3321
3322 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3323 if (ret < 0)
3324 goto out;
3325
3326 ret = send_remove_xattr(sctx, p, name, name_len);
3327
3328out:
3329 fs_path_free(sctx, p);
3330 return ret;
3331}
3332
3333static int process_new_xattr(struct send_ctx *sctx)
3334{
3335 int ret = 0;
3336
3337 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3338 sctx->cmp_key, __process_new_xattr, sctx);
3339
3340 return ret;
3341}
3342
3343static int process_deleted_xattr(struct send_ctx *sctx)
3344{
3345 int ret;
3346
3347 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3348 sctx->cmp_key, __process_deleted_xattr, sctx);
3349
3350 return ret;
3351}
3352
3353struct find_xattr_ctx {
3354 const char *name;
3355 int name_len;
3356 int found_idx;
3357 char *found_data;
3358 int found_data_len;
3359};
3360
3361static int __find_xattr(int num, struct btrfs_key *di_key,
3362 const char *name, int name_len,
3363 const char *data, int data_len,
3364 u8 type, void *vctx)
3365{
3366 struct find_xattr_ctx *ctx = vctx;
3367
3368 if (name_len == ctx->name_len &&
3369 strncmp(name, ctx->name, name_len) == 0) {
3370 ctx->found_idx = num;
3371 ctx->found_data_len = data_len;
3372 ctx->found_data = kmalloc(data_len, GFP_NOFS);
3373 if (!ctx->found_data)
3374 return -ENOMEM;
3375 memcpy(ctx->found_data, data, data_len);
3376 return 1;
3377 }
3378 return 0;
3379}
3380
3381static int find_xattr(struct send_ctx *sctx,
3382 struct btrfs_root *root,
3383 struct btrfs_path *path,
3384 struct btrfs_key *key,
3385 const char *name, int name_len,
3386 char **data, int *data_len)
3387{
3388 int ret;
3389 struct find_xattr_ctx ctx;
3390
3391 ctx.name = name;
3392 ctx.name_len = name_len;
3393 ctx.found_idx = -1;
3394 ctx.found_data = NULL;
3395 ctx.found_data_len = 0;
3396
3397 ret = iterate_dir_item(sctx, root, path, key, __find_xattr, &ctx);
3398 if (ret < 0)
3399 return ret;
3400
3401 if (ctx.found_idx == -1)
3402 return -ENOENT;
3403 if (data) {
3404 *data = ctx.found_data;
3405 *data_len = ctx.found_data_len;
3406 } else {
3407 kfree(ctx.found_data);
3408 }
3409 return ctx.found_idx;
3410}
3411
3412
3413static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3414 const char *name, int name_len,
3415 const char *data, int data_len,
3416 u8 type, void *ctx)
3417{
3418 int ret;
3419 struct send_ctx *sctx = ctx;
3420 char *found_data = NULL;
3421 int found_data_len = 0;
3422 struct fs_path *p = NULL;
3423
3424 ret = find_xattr(sctx, sctx->parent_root, sctx->right_path,
3425 sctx->cmp_key, name, name_len, &found_data,
3426 &found_data_len);
3427 if (ret == -ENOENT) {
3428 ret = __process_new_xattr(num, di_key, name, name_len, data,
3429 data_len, type, ctx);
3430 } else if (ret >= 0) {
3431 if (data_len != found_data_len ||
3432 memcmp(data, found_data, data_len)) {
3433 ret = __process_new_xattr(num, di_key, name, name_len,
3434 data, data_len, type, ctx);
3435 } else {
3436 ret = 0;
3437 }
3438 }
3439
3440 kfree(found_data);
3441 fs_path_free(sctx, p);
3442 return ret;
3443}
3444
3445static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3446 const char *name, int name_len,
3447 const char *data, int data_len,
3448 u8 type, void *ctx)
3449{
3450 int ret;
3451 struct send_ctx *sctx = ctx;
3452
3453 ret = find_xattr(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3454 name, name_len, NULL, NULL);
3455 if (ret == -ENOENT)
3456 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3457 data_len, type, ctx);
3458 else if (ret >= 0)
3459 ret = 0;
3460
3461 return ret;
3462}
3463
3464static int process_changed_xattr(struct send_ctx *sctx)
3465{
3466 int ret = 0;
3467
3468 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3469 sctx->cmp_key, __process_changed_new_xattr, sctx);
3470 if (ret < 0)
3471 goto out;
3472 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3473 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3474
3475out:
3476 return ret;
3477}
3478
3479static int process_all_new_xattrs(struct send_ctx *sctx)
3480{
3481 int ret;
3482 struct btrfs_root *root;
3483 struct btrfs_path *path;
3484 struct btrfs_key key;
3485 struct btrfs_key found_key;
3486 struct extent_buffer *eb;
3487 int slot;
3488
3489 path = alloc_path_for_send();
3490 if (!path)
3491 return -ENOMEM;
3492
3493 root = sctx->send_root;
3494
3495 key.objectid = sctx->cmp_key->objectid;
3496 key.type = BTRFS_XATTR_ITEM_KEY;
3497 key.offset = 0;
3498 while (1) {
3499 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3500 if (ret < 0)
3501 goto out;
3502 if (ret) {
3503 ret = 0;
3504 goto out;
3505 }
3506
3507 eb = path->nodes[0];
3508 slot = path->slots[0];
3509 btrfs_item_key_to_cpu(eb, &found_key, slot);
3510
3511 if (found_key.objectid != key.objectid ||
3512 found_key.type != key.type) {
3513 ret = 0;
3514 goto out;
3515 }
3516
3517 ret = iterate_dir_item(sctx, root, path, &found_key,
3518 __process_new_xattr, sctx);
3519 if (ret < 0)
3520 goto out;
3521
3522 btrfs_release_path(path);
3523 key.offset = found_key.offset + 1;
3524 }
3525
3526out:
3527 btrfs_free_path(path);
3528 return ret;
3529}
3530
3531/*
3532 * Read some bytes from the current inode/file and send a write command to
3533 * user space.
3534 */
3535static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3536{
3537 int ret = 0;
3538 struct fs_path *p;
3539 loff_t pos = offset;
e938c8ad 3540 int num_read = 0;
31db9f7c
AB		 3541 mm_segment_t old_fs;
3542
3543 p = fs_path_alloc(sctx);
3544 if (!p)
3545 return -ENOMEM;
3546
3547 /*
3548 * vfs normally only accepts user space buffers for security reasons.
3549 * we only read from the file and also only provide the read_buf buffer
3550 * to vfs. As this buffer does not come from a user space call, it's
3551 * ok to temporary allow kernel space buffers.
3552 */
3553 old_fs = get_fs();
3554 set_fs(KERNEL_DS);
3555
3556verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3557
3558 ret = open_cur_inode_file(sctx);
3559 if (ret < 0)
3560 goto out;
3561
3562 ret = vfs_read(sctx->cur_inode_filp, sctx->read_buf, len, &pos);
3563 if (ret < 0)
3564 goto out;
e938c8ad
AB		 3565 num_read = ret;
3566 if (!num_read)
31db9f7c
AB		 3567 goto out;
3568
3569 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3570 if (ret < 0)
3571 goto out;
3572
3573 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3574 if (ret < 0)
3575 goto out;
3576
3577 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3578 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
e938c8ad 3579 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
31db9f7c
AB		 3580
3581 ret = send_cmd(sctx);
3582
3583tlv_put_failure:
3584out:
3585 fs_path_free(sctx, p);
3586 set_fs(old_fs);
3587 if (ret < 0)
3588 return ret;
e938c8ad 3589 return num_read;
31db9f7c
AB		 3590}
3591
3592/*
3593 * Send a clone command to user space.
3594 */
3595static int send_clone(struct send_ctx *sctx,
3596 u64 offset, u32 len,
3597 struct clone_root *clone_root)
3598{
3599 int ret = 0;
31db9f7c
AB		 3600 struct fs_path *p;
3601 u64 gen;
3602
3603verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
3604 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
3605 clone_root->root->objectid, clone_root->ino,
3606 clone_root->offset);
3607
3608 p = fs_path_alloc(sctx);
3609 if (!p)
3610 return -ENOMEM;
3611
3612 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
3613 if (ret < 0)
3614 goto out;
3615
3616 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3617 if (ret < 0)
3618 goto out;
3619
3620 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3621 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
3622 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3623
e938c8ad 3624 if (clone_root->root == sctx->send_root) {
31db9f7c 3625 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
85a7b33b 3626 &gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 3627 if (ret < 0)
3628 goto out;
3629 ret = get_cur_path(sctx, clone_root->ino, gen, p);
3630 } else {
e938c8ad
AB		 3631 ret = get_inode_path(sctx, clone_root->root,
3632 clone_root->ino, p);
31db9f7c
AB		 3633 }
3634 if (ret < 0)
3635 goto out;
3636
3637 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
e938c8ad 3638 clone_root->root->root_item.uuid);
31db9f7c 3639 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
e938c8ad 3640 clone_root->root->root_item.ctransid);
31db9f7c
AB		 3641 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3642 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3643 clone_root->offset);
3644
3645 ret = send_cmd(sctx);
3646
3647tlv_put_failure:
3648out:
3649 fs_path_free(sctx, p);
3650 return ret;
3651}
3652
3653static int send_write_or_clone(struct send_ctx *sctx,
3654 struct btrfs_path *path,
3655 struct btrfs_key *key,
3656 struct clone_root *clone_root)
3657{
3658 int ret = 0;
3659 struct btrfs_file_extent_item *ei;
3660 u64 offset = key->offset;
3661 u64 pos = 0;
3662 u64 len;
3663 u32 l;
3664 u8 type;
3665
3666 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3667 struct btrfs_file_extent_item);
3668 type = btrfs_file_extent_type(path->nodes[0], ei);
3669 if (type == BTRFS_FILE_EXTENT_INLINE)
3670 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
3671 else
3672 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3673
3674 if (offset + len > sctx->cur_inode_size)
3675 len = sctx->cur_inode_size - offset;
3676 if (len == 0) {
3677 ret = 0;
3678 goto out;
3679 }
3680
3681 if (!clone_root) {
3682 while (pos < len) {
3683 l = len - pos;
3684 if (l > BTRFS_SEND_READ_SIZE)
3685 l = BTRFS_SEND_READ_SIZE;
3686 ret = send_write(sctx, pos + offset, l);
3687 if (ret < 0)
3688 goto out;
3689 if (!ret)
3690 break;
3691 pos += ret;
3692 }
3693 ret = 0;
3694 } else {
3695 ret = send_clone(sctx, offset, len, clone_root);
3696 }
3697
3698out:
3699 return ret;
3700}
3701
3702static int is_extent_unchanged(struct send_ctx *sctx,
3703 struct btrfs_path *left_path,
3704 struct btrfs_key *ekey)
3705{
3706 int ret = 0;
3707 struct btrfs_key key;
3708 struct btrfs_path *path = NULL;
3709 struct extent_buffer *eb;
3710 int slot;
3711 struct btrfs_key found_key;
3712 struct btrfs_file_extent_item *ei;
3713 u64 left_disknr;
3714 u64 right_disknr;
3715 u64 left_offset;
3716 u64 right_offset;
3717 u64 left_offset_fixed;
3718 u64 left_len;
3719 u64 right_len;
3720 u8 left_type;
3721 u8 right_type;
3722
3723 path = alloc_path_for_send();
3724 if (!path)
3725 return -ENOMEM;
3726
3727 eb = left_path->nodes[0];
3728 slot = left_path->slots[0];
3729
3730 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3731 left_type = btrfs_file_extent_type(eb, ei);
3732 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3733 left_len = btrfs_file_extent_num_bytes(eb, ei);
3734 left_offset = btrfs_file_extent_offset(eb, ei);
3735
3736 if (left_type != BTRFS_FILE_EXTENT_REG) {
3737 ret = 0;
3738 goto out;
3739 }
3740
3741 /*
3742 * Following comments will refer to these graphics. L is the left
3743 * extents which we are checking at the moment. 1-8 are the right
3744 * extents that we iterate.
3745 *
3746 * |-----L-----|
3747 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3748 *
3749 * |-----L-----|
3750 * |--1--|-2b-|...(same as above)
3751 *
3752 * Alternative situation. Happens on files where extents got split.
3753 * |-----L-----|
3754 * |-----------7-----------|-6-|
3755 *
3756 * Alternative situation. Happens on files which got larger.
3757 * |-----L-----|
3758 * |-8-|
3759 * Nothing follows after 8.
3760 */
3761
3762 key.objectid = ekey->objectid;
3763 key.type = BTRFS_EXTENT_DATA_KEY;
3764 key.offset = ekey->offset;
3765 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3766 if (ret < 0)
3767 goto out;
3768 if (ret) {
3769 ret = 0;
3770 goto out;
3771 }
3772
3773 /*
3774 * Handle special case where the right side has no extents at all.
3775 */
3776 eb = path->nodes[0];
3777 slot = path->slots[0];
3778 btrfs_item_key_to_cpu(eb, &found_key, slot);
3779 if (found_key.objectid != key.objectid ||
3780 found_key.type != key.type) {
3781 ret = 0;
3782 goto out;
3783 }
3784
3785 /*
3786 * We're now on 2a, 2b or 7.
3787 */
3788 key = found_key;
3789 while (key.offset < ekey->offset + left_len) {
3790 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3791 right_type = btrfs_file_extent_type(eb, ei);
3792 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3793 right_len = btrfs_file_extent_num_bytes(eb, ei);
3794 right_offset = btrfs_file_extent_offset(eb, ei);
3795
3796 if (right_type != BTRFS_FILE_EXTENT_REG) {
3797 ret = 0;
3798 goto out;
3799 }
3800
3801 /*
3802 * Are we at extent 8? If yes, we know the extent is changed.
3803 * This may only happen on the first iteration.
3804 */
3805 if (found_key.offset + right_len < ekey->offset) {
3806 ret = 0;
3807 goto out;
3808 }
3809
3810 left_offset_fixed = left_offset;
3811 if (key.offset < ekey->offset) {
3812 /* Fix the right offset for 2a and 7. */
3813 right_offset += ekey->offset - key.offset;
3814 } else {
3815 /* Fix the left offset for all behind 2a and 2b */
3816 left_offset_fixed += key.offset - ekey->offset;
3817 }
3818
3819 /*
3820 * Check if we have the same extent.
3821 */
3954096d
AB		 3822 if (left_disknr != right_disknr ||
3823 left_offset_fixed != right_offset) {
31db9f7c
AB		 3824 ret = 0;
3825 goto out;
3826 }
3827
3828 /*
3829 * Go to the next extent.
3830 */
3831 ret = btrfs_next_item(sctx->parent_root, path);
3832 if (ret < 0)
3833 goto out;
3834 if (!ret) {
3835 eb = path->nodes[0];
3836 slot = path->slots[0];
3837 btrfs_item_key_to_cpu(eb, &found_key, slot);
3838 }
3839 if (ret || found_key.objectid != key.objectid ||
3840 found_key.type != key.type) {
3841 key.offset += right_len;
3842 break;
3843 } else {
3844 if (found_key.offset != key.offset + right_len) {
3845 /* Should really not happen */
3846 ret = -EIO;
3847 goto out;
3848 }
3849 }
3850 key = found_key;
3851 }
3852
3853 /*
3854 * We're now behind the left extent (treat as unchanged) or at the end
3855 * of the right side (treat as changed).
3856 */
3857 if (key.offset >= ekey->offset + left_len)
3858 ret = 1;
3859 else
3860 ret = 0;
3861
3862
3863out:
3864 btrfs_free_path(path);
3865 return ret;
3866}
3867
3868static int process_extent(struct send_ctx *sctx,
3869 struct btrfs_path *path,
3870 struct btrfs_key *key)
3871{
3872 int ret = 0;
3873 struct clone_root *found_clone = NULL;
3874
3875 if (S_ISLNK(sctx->cur_inode_mode))
3876 return 0;
3877
3878 if (sctx->parent_root && !sctx->cur_inode_new) {
3879 ret = is_extent_unchanged(sctx, path, key);
3880 if (ret < 0)
3881 goto out;
3882 if (ret) {
3883 ret = 0;
3884 goto out;
3885 }
3886 }
3887
3888 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
3889 sctx->cur_inode_size, &found_clone);
3890 if (ret != -ENOENT && ret < 0)
3891 goto out;
3892
3893 ret = send_write_or_clone(sctx, path, key, found_clone);
3894
3895out:
3896 return ret;
3897}
3898
3899static int process_all_extents(struct send_ctx *sctx)
3900{
3901 int ret;
3902 struct btrfs_root *root;
3903 struct btrfs_path *path;
3904 struct btrfs_key key;
3905 struct btrfs_key found_key;
3906 struct extent_buffer *eb;
3907 int slot;
3908
3909 root = sctx->send_root;
3910 path = alloc_path_for_send();
3911 if (!path)
3912 return -ENOMEM;
3913
3914 key.objectid = sctx->cmp_key->objectid;
3915 key.type = BTRFS_EXTENT_DATA_KEY;
3916 key.offset = 0;
3917 while (1) {
3918 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3919 if (ret < 0)
3920 goto out;
3921 if (ret) {
3922 ret = 0;
3923 goto out;
3924 }
3925
3926 eb = path->nodes[0];
3927 slot = path->slots[0];
3928 btrfs_item_key_to_cpu(eb, &found_key, slot);
3929
3930 if (found_key.objectid != key.objectid ||
3931 found_key.type != key.type) {
3932 ret = 0;
3933 goto out;
3934 }
3935
3936 ret = process_extent(sctx, path, &found_key);
3937 if (ret < 0)
3938 goto out;
3939
3940 btrfs_release_path(path);
3941 key.offset = found_key.offset + 1;
3942 }
3943
3944out:
3945 btrfs_free_path(path);
3946 return ret;
3947}
3948
3949static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
3950{
3951 int ret = 0;
3952
3953 if (sctx->cur_ino == 0)
3954 goto out;
3955 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
3956 sctx->cmp_key->type <= BTRFS_INODE_REF_KEY)
3957 goto out;
3958 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
3959 goto out;
3960
3961 ret = process_recorded_refs(sctx);
e479d9bb
AB		 3962 if (ret < 0)
3963 goto out;
3964
3965 /*
3966 * We have processed the refs and thus need to advance send_progress.
3967 * Now, calls to get_cur_xxx will take the updated refs of the current
3968 * inode into account.
3969 */
3970 sctx->send_progress = sctx->cur_ino + 1;
31db9f7c
AB		 3971
3972out:
3973 return ret;
3974}
3975
3976static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
3977{
3978 int ret = 0;
3979 u64 left_mode;
3980 u64 left_uid;
3981 u64 left_gid;
3982 u64 right_mode;
3983 u64 right_uid;
3984 u64 right_gid;
3985 int need_chmod = 0;
3986 int need_chown = 0;
3987
3988 ret = process_recorded_refs_if_needed(sctx, at_end);
3989 if (ret < 0)
3990 goto out;
3991
3992 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
3993 goto out;
3994 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
3995 goto out;
3996
3997 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
85a7b33b 3998 &left_mode, &left_uid, &left_gid, NULL);
31db9f7c
AB		 3999 if (ret < 0)
4000 goto out;
4001
4002 if (!S_ISLNK(sctx->cur_inode_mode)) {
4003 if (!sctx->parent_root || sctx->cur_inode_new) {
4004 need_chmod = 1;
4005 need_chown = 1;
4006 } else {
4007 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4008 NULL, NULL, &right_mode, &right_uid,
85a7b33b 4009 &right_gid, NULL);
31db9f7c
AB		 4010 if (ret < 0)
4011 goto out;
4012
4013 if (left_uid != right_uid || left_gid != right_gid)
4014 need_chown = 1;
4015 if (left_mode != right_mode)
4016 need_chmod = 1;
4017 }
4018 }
4019
4020 if (S_ISREG(sctx->cur_inode_mode)) {
4021 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4022 sctx->cur_inode_size);
4023 if (ret < 0)
4024 goto out;
4025 }
4026
4027 if (need_chown) {
4028 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4029 left_uid, left_gid);
4030 if (ret < 0)
4031 goto out;
4032 }
4033 if (need_chmod) {
4034 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4035 left_mode);
4036 if (ret < 0)
4037 goto out;
4038 }
4039
4040 /*
4041 * Need to send that every time, no matter if it actually changed
4042 * between the two trees as we have done changes to the inode before.
4043 */
4044 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4045 if (ret < 0)
4046 goto out;
4047
4048out:
4049 return ret;
4050}
4051
4052static int changed_inode(struct send_ctx *sctx,
4053 enum btrfs_compare_tree_result result)
4054{
4055 int ret = 0;
4056 struct btrfs_key *key = sctx->cmp_key;
4057 struct btrfs_inode_item *left_ii = NULL;
4058 struct btrfs_inode_item *right_ii = NULL;
4059 u64 left_gen = 0;
4060 u64 right_gen = 0;
4061
4062 ret = close_cur_inode_file(sctx);
4063 if (ret < 0)
4064 goto out;
4065
4066 sctx->cur_ino = key->objectid;
4067 sctx->cur_inode_new_gen = 0;
e479d9bb
AB		 4068
4069 /*
4070 * Set send_progress to current inode. This will tell all get_cur_xxx
4071 * functions that the current inode's refs are not updated yet. Later,
4072 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4073 */
31db9f7c
AB		 4074 sctx->send_progress = sctx->cur_ino;
4075
4076 if (result == BTRFS_COMPARE_TREE_NEW ||
4077 result == BTRFS_COMPARE_TREE_CHANGED) {
4078 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4079 sctx->left_path->slots[0],
4080 struct btrfs_inode_item);
4081 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4082 left_ii);
4083 } else {
4084 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4085 sctx->right_path->slots[0],
4086 struct btrfs_inode_item);
4087 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4088 right_ii);
4089 }
4090 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4091 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4092 sctx->right_path->slots[0],
4093 struct btrfs_inode_item);
4094
4095 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4096 right_ii);
4097 if (left_gen != right_gen)
4098 sctx->cur_inode_new_gen = 1;
4099 }
4100
4101 if (result == BTRFS_COMPARE_TREE_NEW) {
4102 sctx->cur_inode_gen = left_gen;
4103 sctx->cur_inode_new = 1;
4104 sctx->cur_inode_deleted = 0;
4105 sctx->cur_inode_size = btrfs_inode_size(
4106 sctx->left_path->nodes[0], left_ii);
4107 sctx->cur_inode_mode = btrfs_inode_mode(
4108 sctx->left_path->nodes[0], left_ii);
4109 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
1f4692da 4110 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4111 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4112 sctx->cur_inode_gen = right_gen;
4113 sctx->cur_inode_new = 0;
4114 sctx->cur_inode_deleted = 1;
4115 sctx->cur_inode_size = btrfs_inode_size(
4116 sctx->right_path->nodes[0], right_ii);
4117 sctx->cur_inode_mode = btrfs_inode_mode(
4118 sctx->right_path->nodes[0], right_ii);
4119 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
766702ef
AB		 4120 /*
4121 * We need to do some special handling in case the inode was
4122 * reported as changed with a changed generation number. This
4123 * means that the original inode was deleted and new inode
4124 * reused the same inum. So we have to treat the old inode as
4125 * deleted and the new one as new.
4126 */
31db9f7c 4127 if (sctx->cur_inode_new_gen) {
766702ef
AB		 4128 /*
4129 * First, process the inode as if it was deleted.
4130 */
31db9f7c
AB		 4131 sctx->cur_inode_gen = right_gen;
4132 sctx->cur_inode_new = 0;
4133 sctx->cur_inode_deleted = 1;
4134 sctx->cur_inode_size = btrfs_inode_size(
4135 sctx->right_path->nodes[0], right_ii);
4136 sctx->cur_inode_mode = btrfs_inode_mode(
4137 sctx->right_path->nodes[0], right_ii);
4138 ret = process_all_refs(sctx,
4139 BTRFS_COMPARE_TREE_DELETED);
4140 if (ret < 0)
4141 goto out;
4142
766702ef
AB		 4143 /*
4144 * Now process the inode as if it was new.
4145 */
31db9f7c
AB		 4146 sctx->cur_inode_gen = left_gen;
4147 sctx->cur_inode_new = 1;
4148 sctx->cur_inode_deleted = 0;
4149 sctx->cur_inode_size = btrfs_inode_size(
4150 sctx->left_path->nodes[0], left_ii);
4151 sctx->cur_inode_mode = btrfs_inode_mode(
4152 sctx->left_path->nodes[0], left_ii);
1f4692da 4153 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 4154 if (ret < 0)
4155 goto out;
4156
4157 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4158 if (ret < 0)
4159 goto out;
e479d9bb
AB		 4160 /*
4161 * Advance send_progress now as we did not get into
4162 * process_recorded_refs_if_needed in the new_gen case.
4163 */
4164 sctx->send_progress = sctx->cur_ino + 1;
766702ef
AB		 4165
4166 /*
4167 * Now process all extents and xattrs of the inode as if
4168 * they were all new.
4169 */
31db9f7c
AB		 4170 ret = process_all_extents(sctx);
4171 if (ret < 0)
4172 goto out;
4173 ret = process_all_new_xattrs(sctx);
4174 if (ret < 0)
4175 goto out;
4176 } else {
4177 sctx->cur_inode_gen = left_gen;
4178 sctx->cur_inode_new = 0;
4179 sctx->cur_inode_new_gen = 0;
4180 sctx->cur_inode_deleted = 0;
4181 sctx->cur_inode_size = btrfs_inode_size(
4182 sctx->left_path->nodes[0], left_ii);
4183 sctx->cur_inode_mode = btrfs_inode_mode(
4184 sctx->left_path->nodes[0], left_ii);
4185 }
4186 }
4187
4188out:
4189 return ret;
4190}
4191
766702ef
AB		 4192/*
4193 * We have to process new refs before deleted refs, but compare_trees gives us
4194 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4195 * first and later process them in process_recorded_refs.
4196 * For the cur_inode_new_gen case, we skip recording completely because
4197 * changed_inode did already initiate processing of refs. The reason for this is
4198 * that in this case, compare_tree actually compares the refs of 2 different
4199 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4200 * refs of the right tree as deleted and all refs of the left tree as new.
4201 */
31db9f7c
AB		 4202static int changed_ref(struct send_ctx *sctx,
4203 enum btrfs_compare_tree_result result)
4204{
4205 int ret = 0;
4206
4207 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4208
4209 if (!sctx->cur_inode_new_gen &&
4210 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4211 if (result == BTRFS_COMPARE_TREE_NEW)
4212 ret = record_new_ref(sctx);
4213 else if (result == BTRFS_COMPARE_TREE_DELETED)
4214 ret = record_deleted_ref(sctx);
4215 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4216 ret = record_changed_ref(sctx);
4217 }
4218
4219 return ret;
4220}
4221
766702ef
AB		 4222/*
4223 * Process new/deleted/changed xattrs. We skip processing in the
4224 * cur_inode_new_gen case because changed_inode did already initiate processing
4225 * of xattrs. The reason is the same as in changed_ref
4226 */
31db9f7c
AB		 4227static int changed_xattr(struct send_ctx *sctx,
4228 enum btrfs_compare_tree_result result)
4229{
4230 int ret = 0;
4231
4232 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4233
4234 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4235 if (result == BTRFS_COMPARE_TREE_NEW)
4236 ret = process_new_xattr(sctx);
4237 else if (result == BTRFS_COMPARE_TREE_DELETED)
4238 ret = process_deleted_xattr(sctx);
4239 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4240 ret = process_changed_xattr(sctx);
4241 }
4242
4243 return ret;
4244}
4245
766702ef
AB		 4246/*
4247 * Process new/deleted/changed extents. We skip processing in the
4248 * cur_inode_new_gen case because changed_inode did already initiate processing
4249 * of extents. The reason is the same as in changed_ref
4250 */
31db9f7c
AB		 4251static int changed_extent(struct send_ctx *sctx,
4252 enum btrfs_compare_tree_result result)
4253{
4254 int ret = 0;
4255
4256 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4257
4258 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4259 if (result != BTRFS_COMPARE_TREE_DELETED)
4260 ret = process_extent(sctx, sctx->left_path,
4261 sctx->cmp_key);
4262 }
4263
4264 return ret;
4265}
4266
766702ef
AB		 4267/*
4268 * Updates compare related fields in sctx and simply forwards to the actual
4269 * changed_xxx functions.
4270 */
31db9f7c
AB		 4271static int changed_cb(struct btrfs_root *left_root,
4272 struct btrfs_root *right_root,
4273 struct btrfs_path *left_path,
4274 struct btrfs_path *right_path,
4275 struct btrfs_key *key,
4276 enum btrfs_compare_tree_result result,
4277 void *ctx)
4278{
4279 int ret = 0;
4280 struct send_ctx *sctx = ctx;
4281
4282 sctx->left_path = left_path;
4283 sctx->right_path = right_path;
4284 sctx->cmp_key = key;
4285
4286 ret = finish_inode_if_needed(sctx, 0);
4287 if (ret < 0)
4288 goto out;
4289
4290 if (key->type == BTRFS_INODE_ITEM_KEY)
4291 ret = changed_inode(sctx, result);
4292 else if (key->type == BTRFS_INODE_REF_KEY)
4293 ret = changed_ref(sctx, result);
4294 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4295 ret = changed_xattr(sctx, result);
4296 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4297 ret = changed_extent(sctx, result);
4298
4299out:
4300 return ret;
4301}
4302
4303static int full_send_tree(struct send_ctx *sctx)
4304{
4305 int ret;
4306 struct btrfs_trans_handle *trans = NULL;
4307 struct btrfs_root *send_root = sctx->send_root;
4308 struct btrfs_key key;
4309 struct btrfs_key found_key;
4310 struct btrfs_path *path;
4311 struct extent_buffer *eb;
4312 int slot;
4313 u64 start_ctransid;
4314 u64 ctransid;
4315
4316 path = alloc_path_for_send();
4317 if (!path)
4318 return -ENOMEM;
4319
4320 spin_lock(&send_root->root_times_lock);
4321 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
4322 spin_unlock(&send_root->root_times_lock);
4323
4324 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4325 key.type = BTRFS_INODE_ITEM_KEY;
4326 key.offset = 0;
4327
4328join_trans:
4329 /*
4330 * We need to make sure the transaction does not get committed
4331 * while we do anything on commit roots. Join a transaction to prevent
4332 * this.
4333 */
4334 trans = btrfs_join_transaction(send_root);
4335 if (IS_ERR(trans)) {
4336 ret = PTR_ERR(trans);
4337 trans = NULL;
4338 goto out;
4339 }
4340
4341 /*
766702ef
AB		 4342 * Make sure the tree has not changed after re-joining. We detect this
4343 * by comparing start_ctransid and ctransid. They should always match.
31db9f7c
AB		 4344 */
4345 spin_lock(&send_root->root_times_lock);
4346 ctransid = btrfs_root_ctransid(&send_root->root_item);
4347 spin_unlock(&send_root->root_times_lock);
4348
4349 if (ctransid != start_ctransid) {
4350 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4351 "send was modified in between. This is "
4352 "probably a bug.\n");
4353 ret = -EIO;
4354 goto out;
4355 }
4356
4357 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4358 if (ret < 0)
4359 goto out;
4360 if (ret)
4361 goto out_finish;
4362
4363 while (1) {
4364 /*
4365 * When someone want to commit while we iterate, end the
4366 * joined transaction and rejoin.
4367 */
4368 if (btrfs_should_end_transaction(trans, send_root)) {
4369 ret = btrfs_end_transaction(trans, send_root);
4370 trans = NULL;
4371 if (ret < 0)
4372 goto out;
4373 btrfs_release_path(path);
4374 goto join_trans;
4375 }
4376
4377 eb = path->nodes[0];
4378 slot = path->slots[0];
4379 btrfs_item_key_to_cpu(eb, &found_key, slot);
4380
4381 ret = changed_cb(send_root, NULL, path, NULL,
4382 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4383 if (ret < 0)
4384 goto out;
4385
4386 key.objectid = found_key.objectid;
4387 key.type = found_key.type;
4388 key.offset = found_key.offset + 1;
4389
4390 ret = btrfs_next_item(send_root, path);
4391 if (ret < 0)
4392 goto out;
4393 if (ret) {
4394 ret = 0;
4395 break;
4396 }
4397 }
4398
4399out_finish:
4400 ret = finish_inode_if_needed(sctx, 1);
4401
4402out:
4403 btrfs_free_path(path);
4404 if (trans) {
4405 if (!ret)
4406 ret = btrfs_end_transaction(trans, send_root);
4407 else
4408 btrfs_end_transaction(trans, send_root);
4409 }
4410 return ret;
4411}
4412
4413static int send_subvol(struct send_ctx *sctx)
4414{
4415 int ret;
4416
4417 ret = send_header(sctx);
4418 if (ret < 0)
4419 goto out;
4420
4421 ret = send_subvol_begin(sctx);
4422 if (ret < 0)
4423 goto out;
4424
4425 if (sctx->parent_root) {
4426 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4427 changed_cb, sctx);
4428 if (ret < 0)
4429 goto out;
4430 ret = finish_inode_if_needed(sctx, 1);
4431 if (ret < 0)
4432 goto out;
4433 } else {
4434 ret = full_send_tree(sctx);
4435 if (ret < 0)
4436 goto out;
4437 }
4438
4439out:
4440 if (!ret)
4441 ret = close_cur_inode_file(sctx);
4442 else
4443 close_cur_inode_file(sctx);
4444
4445 free_recorded_refs(sctx);
4446 return ret;
4447}
4448
4449long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4450{
4451 int ret = 0;
4452 struct btrfs_root *send_root;
4453 struct btrfs_root *clone_root;
4454 struct btrfs_fs_info *fs_info;
4455 struct btrfs_ioctl_send_args *arg = NULL;
4456 struct btrfs_key key;
4457 struct file *filp = NULL;
4458 struct send_ctx *sctx = NULL;
4459 u32 i;
4460 u64 *clone_sources_tmp = NULL;
4461
4462 if (!capable(CAP_SYS_ADMIN))
4463 return -EPERM;
4464
4465 send_root = BTRFS_I(fdentry(mnt_file)->d_inode)->root;
4466 fs_info = send_root->fs_info;
4467
4468 arg = memdup_user(arg_, sizeof(*arg));
4469 if (IS_ERR(arg)) {
4470 ret = PTR_ERR(arg);
4471 arg = NULL;
4472 goto out;
4473 }
4474
4475 if (!access_ok(VERIFY_READ, arg->clone_sources,
4476 sizeof(*arg->clone_sources *
4477 arg->clone_sources_count))) {
4478 ret = -EFAULT;
4479 goto out;
4480 }
4481
4482 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4483 if (!sctx) {
4484 ret = -ENOMEM;
4485 goto out;
4486 }
4487
4488 INIT_LIST_HEAD(&sctx->new_refs);
4489 INIT_LIST_HEAD(&sctx->deleted_refs);
4490 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4491 INIT_LIST_HEAD(&sctx->name_cache_list);
4492
4493 sctx->send_filp = fget(arg->send_fd);
4494 if (IS_ERR(sctx->send_filp)) {
4495 ret = PTR_ERR(sctx->send_filp);
4496 goto out;
4497 }
4498
4499 sctx->mnt = mnt_file->f_path.mnt;
4500
4501 sctx->send_root = send_root;
4502 sctx->clone_roots_cnt = arg->clone_sources_count;
4503
4504 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4505 sctx->send_buf = vmalloc(sctx->send_max_size);
4506 if (!sctx->send_buf) {
4507 ret = -ENOMEM;
4508 goto out;
4509 }
4510
4511 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4512 if (!sctx->read_buf) {
4513 ret = -ENOMEM;
4514 goto out;
4515 }
4516
4517 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4518 (arg->clone_sources_count + 1));
4519 if (!sctx->clone_roots) {
4520 ret = -ENOMEM;
4521 goto out;
4522 }
4523
4524 if (arg->clone_sources_count) {
4525 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4526 sizeof(*arg->clone_sources));
4527 if (!clone_sources_tmp) {
4528 ret = -ENOMEM;
4529 goto out;
4530 }
4531
4532 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4533 arg->clone_sources_count *
4534 sizeof(*arg->clone_sources));
4535 if (ret) {
4536 ret = -EFAULT;
4537 goto out;
4538 }
4539
4540 for (i = 0; i < arg->clone_sources_count; i++) {
4541 key.objectid = clone_sources_tmp[i];
4542 key.type = BTRFS_ROOT_ITEM_KEY;
4543 key.offset = (u64)-1;
4544 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
4545 if (!clone_root) {
4546 ret = -EINVAL;
4547 goto out;
4548 }
4549 if (IS_ERR(clone_root)) {
4550 ret = PTR_ERR(clone_root);
4551 goto out;
4552 }
4553 sctx->clone_roots[i].root = clone_root;
4554 }
4555 vfree(clone_sources_tmp);
4556 clone_sources_tmp = NULL;
4557 }
4558
4559 if (arg->parent_root) {
4560 key.objectid = arg->parent_root;
4561 key.type = BTRFS_ROOT_ITEM_KEY;
4562 key.offset = (u64)-1;
4563 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
4564 if (!sctx->parent_root) {
4565 ret = -EINVAL;
4566 goto out;
4567 }
4568 }
4569
4570 /*
4571 * Clones from send_root are allowed, but only if the clone source
4572 * is behind the current send position. This is checked while searching
4573 * for possible clone sources.
4574 */
4575 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4576
4577 /* We do a bsearch later */
4578 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4579 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4580 NULL);
4581
4582 ret = send_subvol(sctx);
4583 if (ret < 0)
4584 goto out;
4585
4586 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4587 if (ret < 0)
4588 goto out;
4589 ret = send_cmd(sctx);
4590 if (ret < 0)
4591 goto out;
4592
4593out:
4594 if (filp)
4595 fput(filp);
4596 kfree(arg);
4597 vfree(clone_sources_tmp);
4598
4599 if (sctx) {
4600 if (sctx->send_filp)
4601 fput(sctx->send_filp);
4602
4603 vfree(sctx->clone_roots);
4604 vfree(sctx->send_buf);
4605 vfree(sctx->read_buf);
4606
4607 name_cache_free(sctx);
4608
4609 kfree(sctx);
4610 }
4611
4612 return ret;
4613}
Linux 6.x block layer and io_uring tree(s)