Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 */
5
6#include <linux/bsearch.h>
7#include <linux/fs.h>
8#include <linux/file.h>
9#include <linux/sort.h>
10#include <linux/mount.h>
11#include <linux/xattr.h>
12#include <linux/posix_acl_xattr.h>
13#include <linux/radix-tree.h>
14#include <linux/vmalloc.h>
15#include <linux/string.h>
16#include <linux/compat.h>
17#include <linux/crc32c.h>
18
19#include "send.h"
20#include "backref.h"
21#include "locking.h"
22#include "disk-io.h"
23#include "btrfs_inode.h"
24#include "transaction.h"
25#include "compression.h"
26
27/*
28 * A fs_path is a helper to dynamically build path names with unknown size.
29 * It reallocates the internal buffer on demand.
30 * It allows fast adding of path elements on the right side (normal path) and
31 * fast adding to the left side (reversed path). A reversed path can also be
32 * unreversed if needed.
33 */
34struct fs_path {
35 union {
36 struct {
37 char *start;
38 char *end;
39
40 char *buf;
41 unsigned short buf_len:15;
42 unsigned short reversed:1;
43 char inline_buf[];
44 };
45 /*
46 * Average path length does not exceed 200 bytes, we'll have
47 * better packing in the slab and higher chance to satisfy
48 * a allocation later during send.
49 */
50 char pad[256];
51 };
52};
53#define FS_PATH_INLINE_SIZE \
54 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
55
56
57/* reused for each extent */
58struct clone_root {
59 struct btrfs_root *root;
60 u64 ino;
61 u64 offset;
62
63 u64 found_refs;
64};
65
66#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
67#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
68
69struct send_ctx {
70 struct file *send_filp;
71 loff_t send_off;
72 char *send_buf;
73 u32 send_size;
74 u32 send_max_size;
75 u64 total_send_size;
76 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
77 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
78
79 struct btrfs_root *send_root;
80 struct btrfs_root *parent_root;
81 struct clone_root *clone_roots;
82 int clone_roots_cnt;
83
84 /* current state of the compare_tree call */
85 struct btrfs_path *left_path;
86 struct btrfs_path *right_path;
87 struct btrfs_key *cmp_key;
88
89 /*
90 * infos of the currently processed inode. In case of deleted inodes,
91 * these are the values from the deleted inode.
92 */
93 u64 cur_ino;
94 u64 cur_inode_gen;
95 int cur_inode_new;
96 int cur_inode_new_gen;
97 int cur_inode_deleted;
98 u64 cur_inode_size;
99 u64 cur_inode_mode;
100 u64 cur_inode_rdev;
101 u64 cur_inode_last_extent;
102 u64 cur_inode_next_write_offset;
103
104 u64 send_progress;
105
106 struct list_head new_refs;
107 struct list_head deleted_refs;
108
109 struct radix_tree_root name_cache;
110 struct list_head name_cache_list;
111 int name_cache_size;
112
113 struct file_ra_state ra;
114
115 char *read_buf;
116
117 /*
118 * We process inodes by their increasing order, so if before an
119 * incremental send we reverse the parent/child relationship of
120 * directories such that a directory with a lower inode number was
121 * the parent of a directory with a higher inode number, and the one
122 * becoming the new parent got renamed too, we can't rename/move the
123 * directory with lower inode number when we finish processing it - we
124 * must process the directory with higher inode number first, then
125 * rename/move it and then rename/move the directory with lower inode
126 * number. Example follows.
127 *
128 * Tree state when the first send was performed:
129 *
130 * .
131 * |-- a (ino 257)
132 * |-- b (ino 258)
133 * |
134 * |
135 * |-- c (ino 259)
136 * | |-- d (ino 260)
137 * |
138 * |-- c2 (ino 261)
139 *
140 * Tree state when the second (incremental) send is performed:
141 *
142 * .
143 * |-- a (ino 257)
144 * |-- b (ino 258)
145 * |-- c2 (ino 261)
146 * |-- d2 (ino 260)
147 * |-- cc (ino 259)
148 *
149 * The sequence of steps that lead to the second state was:
150 *
151 * mv /a/b/c/d /a/b/c2/d2
152 * mv /a/b/c /a/b/c2/d2/cc
153 *
154 * "c" has lower inode number, but we can't move it (2nd mv operation)
155 * before we move "d", which has higher inode number.
156 *
157 * So we just memorize which move/rename operations must be performed
158 * later when their respective parent is processed and moved/renamed.
159 */
160
161 /* Indexed by parent directory inode number. */
162 struct rb_root pending_dir_moves;
163
164 /*
165 * Reverse index, indexed by the inode number of a directory that
166 * is waiting for the move/rename of its immediate parent before its
167 * own move/rename can be performed.
168 */
169 struct rb_root waiting_dir_moves;
170
171 /*
172 * A directory that is going to be rm'ed might have a child directory
173 * which is in the pending directory moves index above. In this case,
174 * the directory can only be removed after the move/rename of its child
175 * is performed. Example:
176 *
177 * Parent snapshot:
178 *
179 * . (ino 256)
180 * |-- a/ (ino 257)
181 * |-- b/ (ino 258)
182 * |-- c/ (ino 259)
183 * | |-- x/ (ino 260)
184 * |
185 * |-- y/ (ino 261)
186 *
187 * Send snapshot:
188 *
189 * . (ino 256)
190 * |-- a/ (ino 257)
191 * |-- b/ (ino 258)
192 * |-- YY/ (ino 261)
193 * |-- x/ (ino 260)
194 *
195 * Sequence of steps that lead to the send snapshot:
196 * rm -f /a/b/c/foo.txt
197 * mv /a/b/y /a/b/YY
198 * mv /a/b/c/x /a/b/YY
199 * rmdir /a/b/c
200 *
201 * When the child is processed, its move/rename is delayed until its
202 * parent is processed (as explained above), but all other operations
203 * like update utimes, chown, chgrp, etc, are performed and the paths
204 * that it uses for those operations must use the orphanized name of
205 * its parent (the directory we're going to rm later), so we need to
206 * memorize that name.
207 *
208 * Indexed by the inode number of the directory to be deleted.
209 */
210 struct rb_root orphan_dirs;
211};
212
213struct pending_dir_move {
214 struct rb_node node;
215 struct list_head list;
216 u64 parent_ino;
217 u64 ino;
218 u64 gen;
219 struct list_head update_refs;
220};
221
222struct waiting_dir_move {
223 struct rb_node node;
224 u64 ino;
225 /*
226 * There might be some directory that could not be removed because it
227 * was waiting for this directory inode to be moved first. Therefore
228 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
229 */
230 u64 rmdir_ino;
231 bool orphanized;
232};
233
234struct orphan_dir_info {
235 struct rb_node node;
236 u64 ino;
237 u64 gen;
238};
239
240struct name_cache_entry {
241 struct list_head list;
242 /*
243 * radix_tree has only 32bit entries but we need to handle 64bit inums.
244 * We use the lower 32bit of the 64bit inum to store it in the tree. If
245 * more then one inum would fall into the same entry, we use radix_list
246 * to store the additional entries. radix_list is also used to store
247 * entries where two entries have the same inum but different
248 * generations.
249 */
250 struct list_head radix_list;
251 u64 ino;
252 u64 gen;
253 u64 parent_ino;
254 u64 parent_gen;
255 int ret;
256 int need_later_update;
257 int name_len;
258 char name[];
259};
260
261__cold
262static void inconsistent_snapshot_error(struct send_ctx *sctx,
263 enum btrfs_compare_tree_result result,
264 const char *what)
265{
266 const char *result_string;
267
268 switch (result) {
269 case BTRFS_COMPARE_TREE_NEW:
270 result_string = "new";
271 break;
272 case BTRFS_COMPARE_TREE_DELETED:
273 result_string = "deleted";
274 break;
275 case BTRFS_COMPARE_TREE_CHANGED:
276 result_string = "updated";
277 break;
278 case BTRFS_COMPARE_TREE_SAME:
279 ASSERT(0);
280 result_string = "unchanged";
281 break;
282 default:
283 ASSERT(0);
284 result_string = "unexpected";
285 }
286
287 btrfs_err(sctx->send_root->fs_info,
288 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
289 result_string, what, sctx->cmp_key->objectid,
290 sctx->send_root->root_key.objectid,
291 (sctx->parent_root ?
292 sctx->parent_root->root_key.objectid : 0));
293}
294
295static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
296
297static struct waiting_dir_move *
298get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
299
300static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
301
302static int need_send_hole(struct send_ctx *sctx)
303{
304 return (sctx->parent_root && !sctx->cur_inode_new &&
305 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
306 S_ISREG(sctx->cur_inode_mode));
307}
308
309static void fs_path_reset(struct fs_path *p)
310{
311 if (p->reversed) {
312 p->start = p->buf + p->buf_len - 1;
313 p->end = p->start;
314 *p->start = 0;
315 } else {
316 p->start = p->buf;
317 p->end = p->start;
318 *p->start = 0;
319 }
320}
321
322static struct fs_path *fs_path_alloc(void)
323{
324 struct fs_path *p;
325
326 p = kmalloc(sizeof(*p), GFP_KERNEL);
327 if (!p)
328 return NULL;
329 p->reversed = 0;
330 p->buf = p->inline_buf;
331 p->buf_len = FS_PATH_INLINE_SIZE;
332 fs_path_reset(p);
333 return p;
334}
335
336static struct fs_path *fs_path_alloc_reversed(void)
337{
338 struct fs_path *p;
339
340 p = fs_path_alloc();
341 if (!p)
342 return NULL;
343 p->reversed = 1;
344 fs_path_reset(p);
345 return p;
346}
347
348static void fs_path_free(struct fs_path *p)
349{
350 if (!p)
351 return;
352 if (p->buf != p->inline_buf)
353 kfree(p->buf);
354 kfree(p);
355}
356
357static int fs_path_len(struct fs_path *p)
358{
359 return p->end - p->start;
360}
361
362static int fs_path_ensure_buf(struct fs_path *p, int len)
363{
364 char *tmp_buf;
365 int path_len;
366 int old_buf_len;
367
368 len++;
369
370 if (p->buf_len >= len)
371 return 0;
372
373 if (len > PATH_MAX) {
374 WARN_ON(1);
375 return -ENOMEM;
376 }
377
378 path_len = p->end - p->start;
379 old_buf_len = p->buf_len;
380
381 /*
382 * First time the inline_buf does not suffice
383 */
384 if (p->buf == p->inline_buf) {
385 tmp_buf = kmalloc(len, GFP_KERNEL);
386 if (tmp_buf)
387 memcpy(tmp_buf, p->buf, old_buf_len);
388 } else {
389 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
390 }
391 if (!tmp_buf)
392 return -ENOMEM;
393 p->buf = tmp_buf;
394 /*
395 * The real size of the buffer is bigger, this will let the fast path
396 * happen most of the time
397 */
398 p->buf_len = ksize(p->buf);
399
400 if (p->reversed) {
401 tmp_buf = p->buf + old_buf_len - path_len - 1;
402 p->end = p->buf + p->buf_len - 1;
403 p->start = p->end - path_len;
404 memmove(p->start, tmp_buf, path_len + 1);
405 } else {
406 p->start = p->buf;
407 p->end = p->start + path_len;
408 }
409 return 0;
410}
411
412static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
413 char **prepared)
414{
415 int ret;
416 int new_len;
417
418 new_len = p->end - p->start + name_len;
419 if (p->start != p->end)
420 new_len++;
421 ret = fs_path_ensure_buf(p, new_len);
422 if (ret < 0)
423 goto out;
424
425 if (p->reversed) {
426 if (p->start != p->end)
427 *--p->start = '/';
428 p->start -= name_len;
429 *prepared = p->start;
430 } else {
431 if (p->start != p->end)
432 *p->end++ = '/';
433 *prepared = p->end;
434 p->end += name_len;
435 *p->end = 0;
436 }
437
438out:
439 return ret;
440}
441
442static int fs_path_add(struct fs_path *p, const char *name, int name_len)
443{
444 int ret;
445 char *prepared;
446
447 ret = fs_path_prepare_for_add(p, name_len, &prepared);
448 if (ret < 0)
449 goto out;
450 memcpy(prepared, name, name_len);
451
452out:
453 return ret;
454}
455
456static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
457{
458 int ret;
459 char *prepared;
460
461 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
462 if (ret < 0)
463 goto out;
464 memcpy(prepared, p2->start, p2->end - p2->start);
465
466out:
467 return ret;
468}
469
470static int fs_path_add_from_extent_buffer(struct fs_path *p,
471 struct extent_buffer *eb,
472 unsigned long off, int len)
473{
474 int ret;
475 char *prepared;
476
477 ret = fs_path_prepare_for_add(p, len, &prepared);
478 if (ret < 0)
479 goto out;
480
481 read_extent_buffer(eb, prepared, off, len);
482
483out:
484 return ret;
485}
486
487static int fs_path_copy(struct fs_path *p, struct fs_path *from)
488{
489 int ret;
490
491 p->reversed = from->reversed;
492 fs_path_reset(p);
493
494 ret = fs_path_add_path(p, from);
495
496 return ret;
497}
498
499
500static void fs_path_unreverse(struct fs_path *p)
501{
502 char *tmp;
503 int len;
504
505 if (!p->reversed)
506 return;
507
508 tmp = p->start;
509 len = p->end - p->start;
510 p->start = p->buf;
511 p->end = p->start + len;
512 memmove(p->start, tmp, len + 1);
513 p->reversed = 0;
514}
515
516static struct btrfs_path *alloc_path_for_send(void)
517{
518 struct btrfs_path *path;
519
520 path = btrfs_alloc_path();
521 if (!path)
522 return NULL;
523 path->search_commit_root = 1;
524 path->skip_locking = 1;
525 path->need_commit_sem = 1;
526 return path;
527}
528
529static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
530{
531 int ret;
532 u32 pos = 0;
533
534 while (pos < len) {
535 ret = kernel_write(filp, buf + pos, len - pos, off);
536 /* TODO handle that correctly */
537 /*if (ret == -ERESTARTSYS) {
538 continue;
539 }*/
540 if (ret < 0)
541 return ret;
542 if (ret == 0) {
543 return -EIO;
544 }
545 pos += ret;
546 }
547
548 return 0;
549}
550
551static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
552{
553 struct btrfs_tlv_header *hdr;
554 int total_len = sizeof(*hdr) + len;
555 int left = sctx->send_max_size - sctx->send_size;
556
557 if (unlikely(left < total_len))
558 return -EOVERFLOW;
559
560 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
561 hdr->tlv_type = cpu_to_le16(attr);
562 hdr->tlv_len = cpu_to_le16(len);
563 memcpy(hdr + 1, data, len);
564 sctx->send_size += total_len;
565
566 return 0;
567}
568
569#define TLV_PUT_DEFINE_INT(bits) \
570 static int tlv_put_u##bits(struct send_ctx *sctx, \
571 u##bits attr, u##bits value) \
572 { \
573 __le##bits __tmp = cpu_to_le##bits(value); \
574 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
575 }
576
577TLV_PUT_DEFINE_INT(64)
578
579static int tlv_put_string(struct send_ctx *sctx, u16 attr,
580 const char *str, int len)
581{
582 if (len == -1)
583 len = strlen(str);
584 return tlv_put(sctx, attr, str, len);
585}
586
587static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
588 const u8 *uuid)
589{
590 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
591}
592
593static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
594 struct extent_buffer *eb,
595 struct btrfs_timespec *ts)
596{
597 struct btrfs_timespec bts;
598 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
599 return tlv_put(sctx, attr, &bts, sizeof(bts));
600}
601
602
603#define TLV_PUT(sctx, attrtype, data, attrlen) \
604 do { \
605 ret = tlv_put(sctx, attrtype, data, attrlen); \
606 if (ret < 0) \
607 goto tlv_put_failure; \
608 } while (0)
609
610#define TLV_PUT_INT(sctx, attrtype, bits, value) \
611 do { \
612 ret = tlv_put_u##bits(sctx, attrtype, value); \
613 if (ret < 0) \
614 goto tlv_put_failure; \
615 } while (0)
616
617#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
618#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
619#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
620#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
621#define TLV_PUT_STRING(sctx, attrtype, str, len) \
622 do { \
623 ret = tlv_put_string(sctx, attrtype, str, len); \
624 if (ret < 0) \
625 goto tlv_put_failure; \
626 } while (0)
627#define TLV_PUT_PATH(sctx, attrtype, p) \
628 do { \
629 ret = tlv_put_string(sctx, attrtype, p->start, \
630 p->end - p->start); \
631 if (ret < 0) \
632 goto tlv_put_failure; \
633 } while(0)
634#define TLV_PUT_UUID(sctx, attrtype, uuid) \
635 do { \
636 ret = tlv_put_uuid(sctx, attrtype, uuid); \
637 if (ret < 0) \
638 goto tlv_put_failure; \
639 } while (0)
640#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
641 do { \
642 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
643 if (ret < 0) \
644 goto tlv_put_failure; \
645 } while (0)
646
647static int send_header(struct send_ctx *sctx)
648{
649 struct btrfs_stream_header hdr;
650
651 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
652 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
653
654 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
655 &sctx->send_off);
656}
657
658/*
659 * For each command/item we want to send to userspace, we call this function.
660 */
661static int begin_cmd(struct send_ctx *sctx, int cmd)
662{
663 struct btrfs_cmd_header *hdr;
664
665 if (WARN_ON(!sctx->send_buf))
666 return -EINVAL;
667
668 BUG_ON(sctx->send_size);
669
670 sctx->send_size += sizeof(*hdr);
671 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
672 hdr->cmd = cpu_to_le16(cmd);
673
674 return 0;
675}
676
677static int send_cmd(struct send_ctx *sctx)
678{
679 int ret;
680 struct btrfs_cmd_header *hdr;
681 u32 crc;
682
683 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
684 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
685 hdr->crc = 0;
686
687 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
688 hdr->crc = cpu_to_le32(crc);
689
690 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
691 &sctx->send_off);
692
693 sctx->total_send_size += sctx->send_size;
694 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
695 sctx->send_size = 0;
696
697 return ret;
698}
699
700/*
701 * Sends a move instruction to user space
702 */
703static int send_rename(struct send_ctx *sctx,
704 struct fs_path *from, struct fs_path *to)
705{
706 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
707 int ret;
708
709 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
710
711 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
712 if (ret < 0)
713 goto out;
714
715 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
716 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
717
718 ret = send_cmd(sctx);
719
720tlv_put_failure:
721out:
722 return ret;
723}
724
725/*
726 * Sends a link instruction to user space
727 */
728static int send_link(struct send_ctx *sctx,
729 struct fs_path *path, struct fs_path *lnk)
730{
731 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
732 int ret;
733
734 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
735
736 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
737 if (ret < 0)
738 goto out;
739
740 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
741 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
742
743 ret = send_cmd(sctx);
744
745tlv_put_failure:
746out:
747 return ret;
748}
749
750/*
751 * Sends an unlink instruction to user space
752 */
753static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
754{
755 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
756 int ret;
757
758 btrfs_debug(fs_info, "send_unlink %s", path->start);
759
760 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
761 if (ret < 0)
762 goto out;
763
764 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
765
766 ret = send_cmd(sctx);
767
768tlv_put_failure:
769out:
770 return ret;
771}
772
773/*
774 * Sends a rmdir instruction to user space
775 */
776static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
777{
778 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
779 int ret;
780
781 btrfs_debug(fs_info, "send_rmdir %s", path->start);
782
783 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
784 if (ret < 0)
785 goto out;
786
787 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
788
789 ret = send_cmd(sctx);
790
791tlv_put_failure:
792out:
793 return ret;
794}
795
796/*
797 * Helper function to retrieve some fields from an inode item.
798 */
799static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
800 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
801 u64 *gid, u64 *rdev)
802{
803 int ret;
804 struct btrfs_inode_item *ii;
805 struct btrfs_key key;
806
807 key.objectid = ino;
808 key.type = BTRFS_INODE_ITEM_KEY;
809 key.offset = 0;
810 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
811 if (ret) {
812 if (ret > 0)
813 ret = -ENOENT;
814 return ret;
815 }
816
817 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
818 struct btrfs_inode_item);
819 if (size)
820 *size = btrfs_inode_size(path->nodes[0], ii);
821 if (gen)
822 *gen = btrfs_inode_generation(path->nodes[0], ii);
823 if (mode)
824 *mode = btrfs_inode_mode(path->nodes[0], ii);
825 if (uid)
826 *uid = btrfs_inode_uid(path->nodes[0], ii);
827 if (gid)
828 *gid = btrfs_inode_gid(path->nodes[0], ii);
829 if (rdev)
830 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
831
832 return ret;
833}
834
835static int get_inode_info(struct btrfs_root *root,
836 u64 ino, u64 *size, u64 *gen,
837 u64 *mode, u64 *uid, u64 *gid,
838 u64 *rdev)
839{
840 struct btrfs_path *path;
841 int ret;
842
843 path = alloc_path_for_send();
844 if (!path)
845 return -ENOMEM;
846 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
847 rdev);
848 btrfs_free_path(path);
849 return ret;
850}
851
852typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
853 struct fs_path *p,
854 void *ctx);
855
856/*
857 * Helper function to iterate the entries in ONE btrfs_inode_ref or
858 * btrfs_inode_extref.
859 * The iterate callback may return a non zero value to stop iteration. This can
860 * be a negative value for error codes or 1 to simply stop it.
861 *
862 * path must point to the INODE_REF or INODE_EXTREF when called.
863 */
864static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
865 struct btrfs_key *found_key, int resolve,
866 iterate_inode_ref_t iterate, void *ctx)
867{
868 struct extent_buffer *eb = path->nodes[0];
869 struct btrfs_item *item;
870 struct btrfs_inode_ref *iref;
871 struct btrfs_inode_extref *extref;
872 struct btrfs_path *tmp_path;
873 struct fs_path *p;
874 u32 cur = 0;
875 u32 total;
876 int slot = path->slots[0];
877 u32 name_len;
878 char *start;
879 int ret = 0;
880 int num = 0;
881 int index;
882 u64 dir;
883 unsigned long name_off;
884 unsigned long elem_size;
885 unsigned long ptr;
886
887 p = fs_path_alloc_reversed();
888 if (!p)
889 return -ENOMEM;
890
891 tmp_path = alloc_path_for_send();
892 if (!tmp_path) {
893 fs_path_free(p);
894 return -ENOMEM;
895 }
896
897
898 if (found_key->type == BTRFS_INODE_REF_KEY) {
899 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
900 struct btrfs_inode_ref);
901 item = btrfs_item_nr(slot);
902 total = btrfs_item_size(eb, item);
903 elem_size = sizeof(*iref);
904 } else {
905 ptr = btrfs_item_ptr_offset(eb, slot);
906 total = btrfs_item_size_nr(eb, slot);
907 elem_size = sizeof(*extref);
908 }
909
910 while (cur < total) {
911 fs_path_reset(p);
912
913 if (found_key->type == BTRFS_INODE_REF_KEY) {
914 iref = (struct btrfs_inode_ref *)(ptr + cur);
915 name_len = btrfs_inode_ref_name_len(eb, iref);
916 name_off = (unsigned long)(iref + 1);
917 index = btrfs_inode_ref_index(eb, iref);
918 dir = found_key->offset;
919 } else {
920 extref = (struct btrfs_inode_extref *)(ptr + cur);
921 name_len = btrfs_inode_extref_name_len(eb, extref);
922 name_off = (unsigned long)&extref->name;
923 index = btrfs_inode_extref_index(eb, extref);
924 dir = btrfs_inode_extref_parent(eb, extref);
925 }
926
927 if (resolve) {
928 start = btrfs_ref_to_path(root, tmp_path, name_len,
929 name_off, eb, dir,
930 p->buf, p->buf_len);
931 if (IS_ERR(start)) {
932 ret = PTR_ERR(start);
933 goto out;
934 }
935 if (start < p->buf) {
936 /* overflow , try again with larger buffer */
937 ret = fs_path_ensure_buf(p,
938 p->buf_len + p->buf - start);
939 if (ret < 0)
940 goto out;
941 start = btrfs_ref_to_path(root, tmp_path,
942 name_len, name_off,
943 eb, dir,
944 p->buf, p->buf_len);
945 if (IS_ERR(start)) {
946 ret = PTR_ERR(start);
947 goto out;
948 }
949 BUG_ON(start < p->buf);
950 }
951 p->start = start;
952 } else {
953 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
954 name_len);
955 if (ret < 0)
956 goto out;
957 }
958
959 cur += elem_size + name_len;
960 ret = iterate(num, dir, index, p, ctx);
961 if (ret)
962 goto out;
963 num++;
964 }
965
966out:
967 btrfs_free_path(tmp_path);
968 fs_path_free(p);
969 return ret;
970}
971
972typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
973 const char *name, int name_len,
974 const char *data, int data_len,
975 u8 type, void *ctx);
976
977/*
978 * Helper function to iterate the entries in ONE btrfs_dir_item.
979 * The iterate callback may return a non zero value to stop iteration. This can
980 * be a negative value for error codes or 1 to simply stop it.
981 *
982 * path must point to the dir item when called.
983 */
984static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
985 iterate_dir_item_t iterate, void *ctx)
986{
987 int ret = 0;
988 struct extent_buffer *eb;
989 struct btrfs_item *item;
990 struct btrfs_dir_item *di;
991 struct btrfs_key di_key;
992 char *buf = NULL;
993 int buf_len;
994 u32 name_len;
995 u32 data_len;
996 u32 cur;
997 u32 len;
998 u32 total;
999 int slot;
1000 int num;
1001 u8 type;
1002
1003 /*
1004 * Start with a small buffer (1 page). If later we end up needing more
1005 * space, which can happen for xattrs on a fs with a leaf size greater
1006 * then the page size, attempt to increase the buffer. Typically xattr
1007 * values are small.
1008 */
1009 buf_len = PATH_MAX;
1010 buf = kmalloc(buf_len, GFP_KERNEL);
1011 if (!buf) {
1012 ret = -ENOMEM;
1013 goto out;
1014 }
1015
1016 eb = path->nodes[0];
1017 slot = path->slots[0];
1018 item = btrfs_item_nr(slot);
1019 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1020 cur = 0;
1021 len = 0;
1022 total = btrfs_item_size(eb, item);
1023
1024 num = 0;
1025 while (cur < total) {
1026 name_len = btrfs_dir_name_len(eb, di);
1027 data_len = btrfs_dir_data_len(eb, di);
1028 type = btrfs_dir_type(eb, di);
1029 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1030
1031 if (type == BTRFS_FT_XATTR) {
1032 if (name_len > XATTR_NAME_MAX) {
1033 ret = -ENAMETOOLONG;
1034 goto out;
1035 }
1036 if (name_len + data_len >
1037 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
1038 ret = -E2BIG;
1039 goto out;
1040 }
1041 } else {
1042 /*
1043 * Path too long
1044 */
1045 if (name_len + data_len > PATH_MAX) {
1046 ret = -ENAMETOOLONG;
1047 goto out;
1048 }
1049 }
1050
1051 if (name_len + data_len > buf_len) {
1052 buf_len = name_len + data_len;
1053 if (is_vmalloc_addr(buf)) {
1054 vfree(buf);
1055 buf = NULL;
1056 } else {
1057 char *tmp = krealloc(buf, buf_len,
1058 GFP_KERNEL | __GFP_NOWARN);
1059
1060 if (!tmp)
1061 kfree(buf);
1062 buf = tmp;
1063 }
1064 if (!buf) {
1065 buf = kvmalloc(buf_len, GFP_KERNEL);
1066 if (!buf) {
1067 ret = -ENOMEM;
1068 goto out;
1069 }
1070 }
1071 }
1072
1073 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1074 name_len + data_len);
1075
1076 len = sizeof(*di) + name_len + data_len;
1077 di = (struct btrfs_dir_item *)((char *)di + len);
1078 cur += len;
1079
1080 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1081 data_len, type, ctx);
1082 if (ret < 0)
1083 goto out;
1084 if (ret) {
1085 ret = 0;
1086 goto out;
1087 }
1088
1089 num++;
1090 }
1091
1092out:
1093 kvfree(buf);
1094 return ret;
1095}
1096
1097static int __copy_first_ref(int num, u64 dir, int index,
1098 struct fs_path *p, void *ctx)
1099{
1100 int ret;
1101 struct fs_path *pt = ctx;
1102
1103 ret = fs_path_copy(pt, p);
1104 if (ret < 0)
1105 return ret;
1106
1107 /* we want the first only */
1108 return 1;
1109}
1110
1111/*
1112 * Retrieve the first path of an inode. If an inode has more then one
1113 * ref/hardlink, this is ignored.
1114 */
1115static int get_inode_path(struct btrfs_root *root,
1116 u64 ino, struct fs_path *path)
1117{
1118 int ret;
1119 struct btrfs_key key, found_key;
1120 struct btrfs_path *p;
1121
1122 p = alloc_path_for_send();
1123 if (!p)
1124 return -ENOMEM;
1125
1126 fs_path_reset(path);
1127
1128 key.objectid = ino;
1129 key.type = BTRFS_INODE_REF_KEY;
1130 key.offset = 0;
1131
1132 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1133 if (ret < 0)
1134 goto out;
1135 if (ret) {
1136 ret = 1;
1137 goto out;
1138 }
1139 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1140 if (found_key.objectid != ino ||
1141 (found_key.type != BTRFS_INODE_REF_KEY &&
1142 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1143 ret = -ENOENT;
1144 goto out;
1145 }
1146
1147 ret = iterate_inode_ref(root, p, &found_key, 1,
1148 __copy_first_ref, path);
1149 if (ret < 0)
1150 goto out;
1151 ret = 0;
1152
1153out:
1154 btrfs_free_path(p);
1155 return ret;
1156}
1157
1158struct backref_ctx {
1159 struct send_ctx *sctx;
1160
1161 struct btrfs_path *path;
1162 /* number of total found references */
1163 u64 found;
1164
1165 /*
1166 * used for clones found in send_root. clones found behind cur_objectid
1167 * and cur_offset are not considered as allowed clones.
1168 */
1169 u64 cur_objectid;
1170 u64 cur_offset;
1171
1172 /* may be truncated in case it's the last extent in a file */
1173 u64 extent_len;
1174
1175 /* data offset in the file extent item */
1176 u64 data_offset;
1177
1178 /* Just to check for bugs in backref resolving */
1179 int found_itself;
1180};
1181
1182static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1183{
1184 u64 root = (u64)(uintptr_t)key;
1185 struct clone_root *cr = (struct clone_root *)elt;
1186
1187 if (root < cr->root->objectid)
1188 return -1;
1189 if (root > cr->root->objectid)
1190 return 1;
1191 return 0;
1192}
1193
1194static int __clone_root_cmp_sort(const void *e1, const void *e2)
1195{
1196 struct clone_root *cr1 = (struct clone_root *)e1;
1197 struct clone_root *cr2 = (struct clone_root *)e2;
1198
1199 if (cr1->root->objectid < cr2->root->objectid)
1200 return -1;
1201 if (cr1->root->objectid > cr2->root->objectid)
1202 return 1;
1203 return 0;
1204}
1205
1206/*
1207 * Called for every backref that is found for the current extent.
1208 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1209 */
1210static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1211{
1212 struct backref_ctx *bctx = ctx_;
1213 struct clone_root *found;
1214 int ret;
1215 u64 i_size;
1216
1217 /* First check if the root is in the list of accepted clone sources */
1218 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1219 bctx->sctx->clone_roots_cnt,
1220 sizeof(struct clone_root),
1221 __clone_root_cmp_bsearch);
1222 if (!found)
1223 return 0;
1224
1225 if (found->root == bctx->sctx->send_root &&
1226 ino == bctx->cur_objectid &&
1227 offset == bctx->cur_offset) {
1228 bctx->found_itself = 1;
1229 }
1230
1231 /*
1232 * There are inodes that have extents that lie behind its i_size. Don't
1233 * accept clones from these extents.
1234 */
1235 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1236 NULL, NULL, NULL);
1237 btrfs_release_path(bctx->path);
1238 if (ret < 0)
1239 return ret;
1240
1241 if (offset + bctx->data_offset + bctx->extent_len > i_size)
1242 return 0;
1243
1244 /*
1245 * Make sure we don't consider clones from send_root that are
1246 * behind the current inode/offset.
1247 */
1248 if (found->root == bctx->sctx->send_root) {
1249 /*
1250 * TODO for the moment we don't accept clones from the inode
1251 * that is currently send. We may change this when
1252 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1253 * file.
1254 */
1255 if (ino >= bctx->cur_objectid)
1256 return 0;
1257 }
1258
1259 bctx->found++;
1260 found->found_refs++;
1261 if (ino < found->ino) {
1262 found->ino = ino;
1263 found->offset = offset;
1264 } else if (found->ino == ino) {
1265 /*
1266 * same extent found more then once in the same file.
1267 */
1268 if (found->offset > offset + bctx->extent_len)
1269 found->offset = offset;
1270 }
1271
1272 return 0;
1273}
1274
1275/*
1276 * Given an inode, offset and extent item, it finds a good clone for a clone
1277 * instruction. Returns -ENOENT when none could be found. The function makes
1278 * sure that the returned clone is usable at the point where sending is at the
1279 * moment. This means, that no clones are accepted which lie behind the current
1280 * inode+offset.
1281 *
1282 * path must point to the extent item when called.
1283 */
1284static int find_extent_clone(struct send_ctx *sctx,
1285 struct btrfs_path *path,
1286 u64 ino, u64 data_offset,
1287 u64 ino_size,
1288 struct clone_root **found)
1289{
1290 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1291 int ret;
1292 int extent_type;
1293 u64 logical;
1294 u64 disk_byte;
1295 u64 num_bytes;
1296 u64 extent_item_pos;
1297 u64 flags = 0;
1298 struct btrfs_file_extent_item *fi;
1299 struct extent_buffer *eb = path->nodes[0];
1300 struct backref_ctx *backref_ctx = NULL;
1301 struct clone_root *cur_clone_root;
1302 struct btrfs_key found_key;
1303 struct btrfs_path *tmp_path;
1304 int compressed;
1305 u32 i;
1306
1307 tmp_path = alloc_path_for_send();
1308 if (!tmp_path)
1309 return -ENOMEM;
1310
1311 /* We only use this path under the commit sem */
1312 tmp_path->need_commit_sem = 0;
1313
1314 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1315 if (!backref_ctx) {
1316 ret = -ENOMEM;
1317 goto out;
1318 }
1319
1320 backref_ctx->path = tmp_path;
1321
1322 if (data_offset >= ino_size) {
1323 /*
1324 * There may be extents that lie behind the file's size.
1325 * I at least had this in combination with snapshotting while
1326 * writing large files.
1327 */
1328 ret = 0;
1329 goto out;
1330 }
1331
1332 fi = btrfs_item_ptr(eb, path->slots[0],
1333 struct btrfs_file_extent_item);
1334 extent_type = btrfs_file_extent_type(eb, fi);
1335 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1336 ret = -ENOENT;
1337 goto out;
1338 }
1339 compressed = btrfs_file_extent_compression(eb, fi);
1340
1341 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1342 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1343 if (disk_byte == 0) {
1344 ret = -ENOENT;
1345 goto out;
1346 }
1347 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1348
1349 down_read(&fs_info->commit_root_sem);
1350 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1351 &found_key, &flags);
1352 up_read(&fs_info->commit_root_sem);
1353 btrfs_release_path(tmp_path);
1354
1355 if (ret < 0)
1356 goto out;
1357 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1358 ret = -EIO;
1359 goto out;
1360 }
1361
1362 /*
1363 * Setup the clone roots.
1364 */
1365 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1366 cur_clone_root = sctx->clone_roots + i;
1367 cur_clone_root->ino = (u64)-1;
1368 cur_clone_root->offset = 0;
1369 cur_clone_root->found_refs = 0;
1370 }
1371
1372 backref_ctx->sctx = sctx;
1373 backref_ctx->found = 0;
1374 backref_ctx->cur_objectid = ino;
1375 backref_ctx->cur_offset = data_offset;
1376 backref_ctx->found_itself = 0;
1377 backref_ctx->extent_len = num_bytes;
1378 /*
1379 * For non-compressed extents iterate_extent_inodes() gives us extent
1380 * offsets that already take into account the data offset, but not for
1381 * compressed extents, since the offset is logical and not relative to
1382 * the physical extent locations. We must take this into account to
1383 * avoid sending clone offsets that go beyond the source file's size,
1384 * which would result in the clone ioctl failing with -EINVAL on the
1385 * receiving end.
1386 */
1387 if (compressed == BTRFS_COMPRESS_NONE)
1388 backref_ctx->data_offset = 0;
1389 else
1390 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1391
1392 /*
1393 * The last extent of a file may be too large due to page alignment.
1394 * We need to adjust extent_len in this case so that the checks in
1395 * __iterate_backrefs work.
1396 */
1397 if (data_offset + num_bytes >= ino_size)
1398 backref_ctx->extent_len = ino_size - data_offset;
1399
1400 /*
1401 * Now collect all backrefs.
1402 */
1403 if (compressed == BTRFS_COMPRESS_NONE)
1404 extent_item_pos = logical - found_key.objectid;
1405 else
1406 extent_item_pos = 0;
1407 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1408 extent_item_pos, 1, __iterate_backrefs,
1409 backref_ctx, false);
1410
1411 if (ret < 0)
1412 goto out;
1413
1414 if (!backref_ctx->found_itself) {
1415 /* found a bug in backref code? */
1416 ret = -EIO;
1417 btrfs_err(fs_info,
1418 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1419 ino, data_offset, disk_byte, found_key.objectid);
1420 goto out;
1421 }
1422
1423 btrfs_debug(fs_info,
1424 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1425 data_offset, ino, num_bytes, logical);
1426
1427 if (!backref_ctx->found)
1428 btrfs_debug(fs_info, "no clones found");
1429
1430 cur_clone_root = NULL;
1431 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1432 if (sctx->clone_roots[i].found_refs) {
1433 if (!cur_clone_root)
1434 cur_clone_root = sctx->clone_roots + i;
1435 else if (sctx->clone_roots[i].root == sctx->send_root)
1436 /* prefer clones from send_root over others */
1437 cur_clone_root = sctx->clone_roots + i;
1438 }
1439
1440 }
1441
1442 if (cur_clone_root) {
1443 *found = cur_clone_root;
1444 ret = 0;
1445 } else {
1446 ret = -ENOENT;
1447 }
1448
1449out:
1450 btrfs_free_path(tmp_path);
1451 kfree(backref_ctx);
1452 return ret;
1453}
1454
1455static int read_symlink(struct btrfs_root *root,
1456 u64 ino,
1457 struct fs_path *dest)
1458{
1459 int ret;
1460 struct btrfs_path *path;
1461 struct btrfs_key key;
1462 struct btrfs_file_extent_item *ei;
1463 u8 type;
1464 u8 compression;
1465 unsigned long off;
1466 int len;
1467
1468 path = alloc_path_for_send();
1469 if (!path)
1470 return -ENOMEM;
1471
1472 key.objectid = ino;
1473 key.type = BTRFS_EXTENT_DATA_KEY;
1474 key.offset = 0;
1475 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1476 if (ret < 0)
1477 goto out;
1478 if (ret) {
1479 /*
1480 * An empty symlink inode. Can happen in rare error paths when
1481 * creating a symlink (transaction committed before the inode
1482 * eviction handler removed the symlink inode items and a crash
1483 * happened in between or the subvol was snapshoted in between).
1484 * Print an informative message to dmesg/syslog so that the user
1485 * can delete the symlink.
1486 */
1487 btrfs_err(root->fs_info,
1488 "Found empty symlink inode %llu at root %llu",
1489 ino, root->root_key.objectid);
1490 ret = -EIO;
1491 goto out;
1492 }
1493
1494 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1495 struct btrfs_file_extent_item);
1496 type = btrfs_file_extent_type(path->nodes[0], ei);
1497 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1498 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1499 BUG_ON(compression);
1500
1501 off = btrfs_file_extent_inline_start(ei);
1502 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1503
1504 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1505
1506out:
1507 btrfs_free_path(path);
1508 return ret;
1509}
1510
1511/*
1512 * Helper function to generate a file name that is unique in the root of
1513 * send_root and parent_root. This is used to generate names for orphan inodes.
1514 */
1515static int gen_unique_name(struct send_ctx *sctx,
1516 u64 ino, u64 gen,
1517 struct fs_path *dest)
1518{
1519 int ret = 0;
1520 struct btrfs_path *path;
1521 struct btrfs_dir_item *di;
1522 char tmp[64];
1523 int len;
1524 u64 idx = 0;
1525
1526 path = alloc_path_for_send();
1527 if (!path)
1528 return -ENOMEM;
1529
1530 while (1) {
1531 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1532 ino, gen, idx);
1533 ASSERT(len < sizeof(tmp));
1534
1535 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1536 path, BTRFS_FIRST_FREE_OBJECTID,
1537 tmp, strlen(tmp), 0);
1538 btrfs_release_path(path);
1539 if (IS_ERR(di)) {
1540 ret = PTR_ERR(di);
1541 goto out;
1542 }
1543 if (di) {
1544 /* not unique, try again */
1545 idx++;
1546 continue;
1547 }
1548
1549 if (!sctx->parent_root) {
1550 /* unique */
1551 ret = 0;
1552 break;
1553 }
1554
1555 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1556 path, BTRFS_FIRST_FREE_OBJECTID,
1557 tmp, strlen(tmp), 0);
1558 btrfs_release_path(path);
1559 if (IS_ERR(di)) {
1560 ret = PTR_ERR(di);
1561 goto out;
1562 }
1563 if (di) {
1564 /* not unique, try again */
1565 idx++;
1566 continue;
1567 }
1568 /* unique */
1569 break;
1570 }
1571
1572 ret = fs_path_add(dest, tmp, strlen(tmp));
1573
1574out:
1575 btrfs_free_path(path);
1576 return ret;
1577}
1578
1579enum inode_state {
1580 inode_state_no_change,
1581 inode_state_will_create,
1582 inode_state_did_create,
1583 inode_state_will_delete,
1584 inode_state_did_delete,
1585};
1586
1587static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1588{
1589 int ret;
1590 int left_ret;
1591 int right_ret;
1592 u64 left_gen;
1593 u64 right_gen;
1594
1595 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1596 NULL, NULL);
1597 if (ret < 0 && ret != -ENOENT)
1598 goto out;
1599 left_ret = ret;
1600
1601 if (!sctx->parent_root) {
1602 right_ret = -ENOENT;
1603 } else {
1604 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1605 NULL, NULL, NULL, NULL);
1606 if (ret < 0 && ret != -ENOENT)
1607 goto out;
1608 right_ret = ret;
1609 }
1610
1611 if (!left_ret && !right_ret) {
1612 if (left_gen == gen && right_gen == gen) {
1613 ret = inode_state_no_change;
1614 } else if (left_gen == gen) {
1615 if (ino < sctx->send_progress)
1616 ret = inode_state_did_create;
1617 else
1618 ret = inode_state_will_create;
1619 } else if (right_gen == gen) {
1620 if (ino < sctx->send_progress)
1621 ret = inode_state_did_delete;
1622 else
1623 ret = inode_state_will_delete;
1624 } else {
1625 ret = -ENOENT;
1626 }
1627 } else if (!left_ret) {
1628 if (left_gen == gen) {
1629 if (ino < sctx->send_progress)
1630 ret = inode_state_did_create;
1631 else
1632 ret = inode_state_will_create;
1633 } else {
1634 ret = -ENOENT;
1635 }
1636 } else if (!right_ret) {
1637 if (right_gen == gen) {
1638 if (ino < sctx->send_progress)
1639 ret = inode_state_did_delete;
1640 else
1641 ret = inode_state_will_delete;
1642 } else {
1643 ret = -ENOENT;
1644 }
1645 } else {
1646 ret = -ENOENT;
1647 }
1648
1649out:
1650 return ret;
1651}
1652
1653static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1654{
1655 int ret;
1656
1657 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1658 return 1;
1659
1660 ret = get_cur_inode_state(sctx, ino, gen);
1661 if (ret < 0)
1662 goto out;
1663
1664 if (ret == inode_state_no_change ||
1665 ret == inode_state_did_create ||
1666 ret == inode_state_will_delete)
1667 ret = 1;
1668 else
1669 ret = 0;
1670
1671out:
1672 return ret;
1673}
1674
1675/*
1676 * Helper function to lookup a dir item in a dir.
1677 */
1678static int lookup_dir_item_inode(struct btrfs_root *root,
1679 u64 dir, const char *name, int name_len,
1680 u64 *found_inode,
1681 u8 *found_type)
1682{
1683 int ret = 0;
1684 struct btrfs_dir_item *di;
1685 struct btrfs_key key;
1686 struct btrfs_path *path;
1687
1688 path = alloc_path_for_send();
1689 if (!path)
1690 return -ENOMEM;
1691
1692 di = btrfs_lookup_dir_item(NULL, root, path,
1693 dir, name, name_len, 0);
1694 if (!di) {
1695 ret = -ENOENT;
1696 goto out;
1697 }
1698 if (IS_ERR(di)) {
1699 ret = PTR_ERR(di);
1700 goto out;
1701 }
1702 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1703 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1704 ret = -ENOENT;
1705 goto out;
1706 }
1707 *found_inode = key.objectid;
1708 *found_type = btrfs_dir_type(path->nodes[0], di);
1709
1710out:
1711 btrfs_free_path(path);
1712 return ret;
1713}
1714
1715/*
1716 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1717 * generation of the parent dir and the name of the dir entry.
1718 */
1719static int get_first_ref(struct btrfs_root *root, u64 ino,
1720 u64 *dir, u64 *dir_gen, struct fs_path *name)
1721{
1722 int ret;
1723 struct btrfs_key key;
1724 struct btrfs_key found_key;
1725 struct btrfs_path *path;
1726 int len;
1727 u64 parent_dir;
1728
1729 path = alloc_path_for_send();
1730 if (!path)
1731 return -ENOMEM;
1732
1733 key.objectid = ino;
1734 key.type = BTRFS_INODE_REF_KEY;
1735 key.offset = 0;
1736
1737 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1738 if (ret < 0)
1739 goto out;
1740 if (!ret)
1741 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1742 path->slots[0]);
1743 if (ret || found_key.objectid != ino ||
1744 (found_key.type != BTRFS_INODE_REF_KEY &&
1745 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1746 ret = -ENOENT;
1747 goto out;
1748 }
1749
1750 if (found_key.type == BTRFS_INODE_REF_KEY) {
1751 struct btrfs_inode_ref *iref;
1752 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1753 struct btrfs_inode_ref);
1754 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1755 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1756 (unsigned long)(iref + 1),
1757 len);
1758 parent_dir = found_key.offset;
1759 } else {
1760 struct btrfs_inode_extref *extref;
1761 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1762 struct btrfs_inode_extref);
1763 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1764 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1765 (unsigned long)&extref->name, len);
1766 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1767 }
1768 if (ret < 0)
1769 goto out;
1770 btrfs_release_path(path);
1771
1772 if (dir_gen) {
1773 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1774 NULL, NULL, NULL);
1775 if (ret < 0)
1776 goto out;
1777 }
1778
1779 *dir = parent_dir;
1780
1781out:
1782 btrfs_free_path(path);
1783 return ret;
1784}
1785
1786static int is_first_ref(struct btrfs_root *root,
1787 u64 ino, u64 dir,
1788 const char *name, int name_len)
1789{
1790 int ret;
1791 struct fs_path *tmp_name;
1792 u64 tmp_dir;
1793
1794 tmp_name = fs_path_alloc();
1795 if (!tmp_name)
1796 return -ENOMEM;
1797
1798 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1799 if (ret < 0)
1800 goto out;
1801
1802 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1803 ret = 0;
1804 goto out;
1805 }
1806
1807 ret = !memcmp(tmp_name->start, name, name_len);
1808
1809out:
1810 fs_path_free(tmp_name);
1811 return ret;
1812}
1813
1814/*
1815 * Used by process_recorded_refs to determine if a new ref would overwrite an
1816 * already existing ref. In case it detects an overwrite, it returns the
1817 * inode/gen in who_ino/who_gen.
1818 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1819 * to make sure later references to the overwritten inode are possible.
1820 * Orphanizing is however only required for the first ref of an inode.
1821 * process_recorded_refs does an additional is_first_ref check to see if
1822 * orphanizing is really required.
1823 */
1824static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1825 const char *name, int name_len,
1826 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1827{
1828 int ret = 0;
1829 u64 gen;
1830 u64 other_inode = 0;
1831 u8 other_type = 0;
1832
1833 if (!sctx->parent_root)
1834 goto out;
1835
1836 ret = is_inode_existent(sctx, dir, dir_gen);
1837 if (ret <= 0)
1838 goto out;
1839
1840 /*
1841 * If we have a parent root we need to verify that the parent dir was
1842 * not deleted and then re-created, if it was then we have no overwrite
1843 * and we can just unlink this entry.
1844 */
1845 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1846 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1847 NULL, NULL, NULL);
1848 if (ret < 0 && ret != -ENOENT)
1849 goto out;
1850 if (ret) {
1851 ret = 0;
1852 goto out;
1853 }
1854 if (gen != dir_gen)
1855 goto out;
1856 }
1857
1858 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1859 &other_inode, &other_type);
1860 if (ret < 0 && ret != -ENOENT)
1861 goto out;
1862 if (ret) {
1863 ret = 0;
1864 goto out;
1865 }
1866
1867 /*
1868 * Check if the overwritten ref was already processed. If yes, the ref
1869 * was already unlinked/moved, so we can safely assume that we will not
1870 * overwrite anything at this point in time.
1871 */
1872 if (other_inode > sctx->send_progress ||
1873 is_waiting_for_move(sctx, other_inode)) {
1874 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1875 who_gen, who_mode, NULL, NULL, NULL);
1876 if (ret < 0)
1877 goto out;
1878
1879 ret = 1;
1880 *who_ino = other_inode;
1881 } else {
1882 ret = 0;
1883 }
1884
1885out:
1886 return ret;
1887}
1888
1889/*
1890 * Checks if the ref was overwritten by an already processed inode. This is
1891 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1892 * thus the orphan name needs be used.
1893 * process_recorded_refs also uses it to avoid unlinking of refs that were
1894 * overwritten.
1895 */
1896static int did_overwrite_ref(struct send_ctx *sctx,
1897 u64 dir, u64 dir_gen,
1898 u64 ino, u64 ino_gen,
1899 const char *name, int name_len)
1900{
1901 int ret = 0;
1902 u64 gen;
1903 u64 ow_inode;
1904 u8 other_type;
1905
1906 if (!sctx->parent_root)
1907 goto out;
1908
1909 ret = is_inode_existent(sctx, dir, dir_gen);
1910 if (ret <= 0)
1911 goto out;
1912
1913 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1914 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1915 NULL, NULL, NULL);
1916 if (ret < 0 && ret != -ENOENT)
1917 goto out;
1918 if (ret) {
1919 ret = 0;
1920 goto out;
1921 }
1922 if (gen != dir_gen)
1923 goto out;
1924 }
1925
1926 /* check if the ref was overwritten by another ref */
1927 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1928 &ow_inode, &other_type);
1929 if (ret < 0 && ret != -ENOENT)
1930 goto out;
1931 if (ret) {
1932 /* was never and will never be overwritten */
1933 ret = 0;
1934 goto out;
1935 }
1936
1937 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1938 NULL, NULL);
1939 if (ret < 0)
1940 goto out;
1941
1942 if (ow_inode == ino && gen == ino_gen) {
1943 ret = 0;
1944 goto out;
1945 }
1946
1947 /*
1948 * We know that it is or will be overwritten. Check this now.
1949 * The current inode being processed might have been the one that caused
1950 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1951 * the current inode being processed.
1952 */
1953 if ((ow_inode < sctx->send_progress) ||
1954 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1955 gen == sctx->cur_inode_gen))
1956 ret = 1;
1957 else
1958 ret = 0;
1959
1960out:
1961 return ret;
1962}
1963
1964/*
1965 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1966 * that got overwritten. This is used by process_recorded_refs to determine
1967 * if it has to use the path as returned by get_cur_path or the orphan name.
1968 */
1969static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1970{
1971 int ret = 0;
1972 struct fs_path *name = NULL;
1973 u64 dir;
1974 u64 dir_gen;
1975
1976 if (!sctx->parent_root)
1977 goto out;
1978
1979 name = fs_path_alloc();
1980 if (!name)
1981 return -ENOMEM;
1982
1983 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1984 if (ret < 0)
1985 goto out;
1986
1987 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1988 name->start, fs_path_len(name));
1989
1990out:
1991 fs_path_free(name);
1992 return ret;
1993}
1994
1995/*
1996 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1997 * so we need to do some special handling in case we have clashes. This function
1998 * takes care of this with the help of name_cache_entry::radix_list.
1999 * In case of error, nce is kfreed.
2000 */
2001static int name_cache_insert(struct send_ctx *sctx,
2002 struct name_cache_entry *nce)
2003{
2004 int ret = 0;
2005 struct list_head *nce_head;
2006
2007 nce_head = radix_tree_lookup(&sctx->name_cache,
2008 (unsigned long)nce->ino);
2009 if (!nce_head) {
2010 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2011 if (!nce_head) {
2012 kfree(nce);
2013 return -ENOMEM;
2014 }
2015 INIT_LIST_HEAD(nce_head);
2016
2017 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2018 if (ret < 0) {
2019 kfree(nce_head);
2020 kfree(nce);
2021 return ret;
2022 }
2023 }
2024 list_add_tail(&nce->radix_list, nce_head);
2025 list_add_tail(&nce->list, &sctx->name_cache_list);
2026 sctx->name_cache_size++;
2027
2028 return ret;
2029}
2030
2031static void name_cache_delete(struct send_ctx *sctx,
2032 struct name_cache_entry *nce)
2033{
2034 struct list_head *nce_head;
2035
2036 nce_head = radix_tree_lookup(&sctx->name_cache,
2037 (unsigned long)nce->ino);
2038 if (!nce_head) {
2039 btrfs_err(sctx->send_root->fs_info,
2040 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2041 nce->ino, sctx->name_cache_size);
2042 }
2043
2044 list_del(&nce->radix_list);
2045 list_del(&nce->list);
2046 sctx->name_cache_size--;
2047
2048 /*
2049 * We may not get to the final release of nce_head if the lookup fails
2050 */
2051 if (nce_head && list_empty(nce_head)) {
2052 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2053 kfree(nce_head);
2054 }
2055}
2056
2057static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2058 u64 ino, u64 gen)
2059{
2060 struct list_head *nce_head;
2061 struct name_cache_entry *cur;
2062
2063 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2064 if (!nce_head)
2065 return NULL;
2066
2067 list_for_each_entry(cur, nce_head, radix_list) {
2068 if (cur->ino == ino && cur->gen == gen)
2069 return cur;
2070 }
2071 return NULL;
2072}
2073
2074/*
2075 * Removes the entry from the list and adds it back to the end. This marks the
2076 * entry as recently used so that name_cache_clean_unused does not remove it.
2077 */
2078static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2079{
2080 list_del(&nce->list);
2081 list_add_tail(&nce->list, &sctx->name_cache_list);
2082}
2083
2084/*
2085 * Remove some entries from the beginning of name_cache_list.
2086 */
2087static void name_cache_clean_unused(struct send_ctx *sctx)
2088{
2089 struct name_cache_entry *nce;
2090
2091 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2092 return;
2093
2094 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2095 nce = list_entry(sctx->name_cache_list.next,
2096 struct name_cache_entry, list);
2097 name_cache_delete(sctx, nce);
2098 kfree(nce);
2099 }
2100}
2101
2102static void name_cache_free(struct send_ctx *sctx)
2103{
2104 struct name_cache_entry *nce;
2105
2106 while (!list_empty(&sctx->name_cache_list)) {
2107 nce = list_entry(sctx->name_cache_list.next,
2108 struct name_cache_entry, list);
2109 name_cache_delete(sctx, nce);
2110 kfree(nce);
2111 }
2112}
2113
2114/*
2115 * Used by get_cur_path for each ref up to the root.
2116 * Returns 0 if it succeeded.
2117 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2118 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2119 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2120 * Returns <0 in case of error.
2121 */
2122static int __get_cur_name_and_parent(struct send_ctx *sctx,
2123 u64 ino, u64 gen,
2124 u64 *parent_ino,
2125 u64 *parent_gen,
2126 struct fs_path *dest)
2127{
2128 int ret;
2129 int nce_ret;
2130 struct name_cache_entry *nce = NULL;
2131
2132 /*
2133 * First check if we already did a call to this function with the same
2134 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2135 * return the cached result.
2136 */
2137 nce = name_cache_search(sctx, ino, gen);
2138 if (nce) {
2139 if (ino < sctx->send_progress && nce->need_later_update) {
2140 name_cache_delete(sctx, nce);
2141 kfree(nce);
2142 nce = NULL;
2143 } else {
2144 name_cache_used(sctx, nce);
2145 *parent_ino = nce->parent_ino;
2146 *parent_gen = nce->parent_gen;
2147 ret = fs_path_add(dest, nce->name, nce->name_len);
2148 if (ret < 0)
2149 goto out;
2150 ret = nce->ret;
2151 goto out;
2152 }
2153 }
2154
2155 /*
2156 * If the inode is not existent yet, add the orphan name and return 1.
2157 * This should only happen for the parent dir that we determine in
2158 * __record_new_ref
2159 */
2160 ret = is_inode_existent(sctx, ino, gen);
2161 if (ret < 0)
2162 goto out;
2163
2164 if (!ret) {
2165 ret = gen_unique_name(sctx, ino, gen, dest);
2166 if (ret < 0)
2167 goto out;
2168 ret = 1;
2169 goto out_cache;
2170 }
2171
2172 /*
2173 * Depending on whether the inode was already processed or not, use
2174 * send_root or parent_root for ref lookup.
2175 */
2176 if (ino < sctx->send_progress)
2177 ret = get_first_ref(sctx->send_root, ino,
2178 parent_ino, parent_gen, dest);
2179 else
2180 ret = get_first_ref(sctx->parent_root, ino,
2181 parent_ino, parent_gen, dest);
2182 if (ret < 0)
2183 goto out;
2184
2185 /*
2186 * Check if the ref was overwritten by an inode's ref that was processed
2187 * earlier. If yes, treat as orphan and return 1.
2188 */
2189 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2190 dest->start, dest->end - dest->start);
2191 if (ret < 0)
2192 goto out;
2193 if (ret) {
2194 fs_path_reset(dest);
2195 ret = gen_unique_name(sctx, ino, gen, dest);
2196 if (ret < 0)
2197 goto out;
2198 ret = 1;
2199 }
2200
2201out_cache:
2202 /*
2203 * Store the result of the lookup in the name cache.
2204 */
2205 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2206 if (!nce) {
2207 ret = -ENOMEM;
2208 goto out;
2209 }
2210
2211 nce->ino = ino;
2212 nce->gen = gen;
2213 nce->parent_ino = *parent_ino;
2214 nce->parent_gen = *parent_gen;
2215 nce->name_len = fs_path_len(dest);
2216 nce->ret = ret;
2217 strcpy(nce->name, dest->start);
2218
2219 if (ino < sctx->send_progress)
2220 nce->need_later_update = 0;
2221 else
2222 nce->need_later_update = 1;
2223
2224 nce_ret = name_cache_insert(sctx, nce);
2225 if (nce_ret < 0)
2226 ret = nce_ret;
2227 name_cache_clean_unused(sctx);
2228
2229out:
2230 return ret;
2231}
2232
2233/*
2234 * Magic happens here. This function returns the first ref to an inode as it
2235 * would look like while receiving the stream at this point in time.
2236 * We walk the path up to the root. For every inode in between, we check if it
2237 * was already processed/sent. If yes, we continue with the parent as found
2238 * in send_root. If not, we continue with the parent as found in parent_root.
2239 * If we encounter an inode that was deleted at this point in time, we use the
2240 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2241 * that were not created yet and overwritten inodes/refs.
2242 *
2243 * When do we have have orphan inodes:
2244 * 1. When an inode is freshly created and thus no valid refs are available yet
2245 * 2. When a directory lost all it's refs (deleted) but still has dir items
2246 * inside which were not processed yet (pending for move/delete). If anyone
2247 * tried to get the path to the dir items, it would get a path inside that
2248 * orphan directory.
2249 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2250 * of an unprocessed inode. If in that case the first ref would be
2251 * overwritten, the overwritten inode gets "orphanized". Later when we
2252 * process this overwritten inode, it is restored at a new place by moving
2253 * the orphan inode.
2254 *
2255 * sctx->send_progress tells this function at which point in time receiving
2256 * would be.
2257 */
2258static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2259 struct fs_path *dest)
2260{
2261 int ret = 0;
2262 struct fs_path *name = NULL;
2263 u64 parent_inode = 0;
2264 u64 parent_gen = 0;
2265 int stop = 0;
2266
2267 name = fs_path_alloc();
2268 if (!name) {
2269 ret = -ENOMEM;
2270 goto out;
2271 }
2272
2273 dest->reversed = 1;
2274 fs_path_reset(dest);
2275
2276 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2277 struct waiting_dir_move *wdm;
2278
2279 fs_path_reset(name);
2280
2281 if (is_waiting_for_rm(sctx, ino)) {
2282 ret = gen_unique_name(sctx, ino, gen, name);
2283 if (ret < 0)
2284 goto out;
2285 ret = fs_path_add_path(dest, name);
2286 break;
2287 }
2288
2289 wdm = get_waiting_dir_move(sctx, ino);
2290 if (wdm && wdm->orphanized) {
2291 ret = gen_unique_name(sctx, ino, gen, name);
2292 stop = 1;
2293 } else if (wdm) {
2294 ret = get_first_ref(sctx->parent_root, ino,
2295 &parent_inode, &parent_gen, name);
2296 } else {
2297 ret = __get_cur_name_and_parent(sctx, ino, gen,
2298 &parent_inode,
2299 &parent_gen, name);
2300 if (ret)
2301 stop = 1;
2302 }
2303
2304 if (ret < 0)
2305 goto out;
2306
2307 ret = fs_path_add_path(dest, name);
2308 if (ret < 0)
2309 goto out;
2310
2311 ino = parent_inode;
2312 gen = parent_gen;
2313 }
2314
2315out:
2316 fs_path_free(name);
2317 if (!ret)
2318 fs_path_unreverse(dest);
2319 return ret;
2320}
2321
2322/*
2323 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2324 */
2325static int send_subvol_begin(struct send_ctx *sctx)
2326{
2327 int ret;
2328 struct btrfs_root *send_root = sctx->send_root;
2329 struct btrfs_root *parent_root = sctx->parent_root;
2330 struct btrfs_path *path;
2331 struct btrfs_key key;
2332 struct btrfs_root_ref *ref;
2333 struct extent_buffer *leaf;
2334 char *name = NULL;
2335 int namelen;
2336
2337 path = btrfs_alloc_path();
2338 if (!path)
2339 return -ENOMEM;
2340
2341 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2342 if (!name) {
2343 btrfs_free_path(path);
2344 return -ENOMEM;
2345 }
2346
2347 key.objectid = send_root->objectid;
2348 key.type = BTRFS_ROOT_BACKREF_KEY;
2349 key.offset = 0;
2350
2351 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2352 &key, path, 1, 0);
2353 if (ret < 0)
2354 goto out;
2355 if (ret) {
2356 ret = -ENOENT;
2357 goto out;
2358 }
2359
2360 leaf = path->nodes[0];
2361 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2362 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2363 key.objectid != send_root->objectid) {
2364 ret = -ENOENT;
2365 goto out;
2366 }
2367 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2368 namelen = btrfs_root_ref_name_len(leaf, ref);
2369 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2370 btrfs_release_path(path);
2371
2372 if (parent_root) {
2373 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2374 if (ret < 0)
2375 goto out;
2376 } else {
2377 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2378 if (ret < 0)
2379 goto out;
2380 }
2381
2382 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2383
2384 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2385 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2386 sctx->send_root->root_item.received_uuid);
2387 else
2388 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2389 sctx->send_root->root_item.uuid);
2390
2391 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2392 le64_to_cpu(sctx->send_root->root_item.ctransid));
2393 if (parent_root) {
2394 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2395 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2396 parent_root->root_item.received_uuid);
2397 else
2398 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2399 parent_root->root_item.uuid);
2400 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2401 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2402 }
2403
2404 ret = send_cmd(sctx);
2405
2406tlv_put_failure:
2407out:
2408 btrfs_free_path(path);
2409 kfree(name);
2410 return ret;
2411}
2412
2413static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2414{
2415 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2416 int ret = 0;
2417 struct fs_path *p;
2418
2419 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2420
2421 p = fs_path_alloc();
2422 if (!p)
2423 return -ENOMEM;
2424
2425 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2426 if (ret < 0)
2427 goto out;
2428
2429 ret = get_cur_path(sctx, ino, gen, p);
2430 if (ret < 0)
2431 goto out;
2432 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2433 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2434
2435 ret = send_cmd(sctx);
2436
2437tlv_put_failure:
2438out:
2439 fs_path_free(p);
2440 return ret;
2441}
2442
2443static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2444{
2445 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2446 int ret = 0;
2447 struct fs_path *p;
2448
2449 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2450
2451 p = fs_path_alloc();
2452 if (!p)
2453 return -ENOMEM;
2454
2455 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2456 if (ret < 0)
2457 goto out;
2458
2459 ret = get_cur_path(sctx, ino, gen, p);
2460 if (ret < 0)
2461 goto out;
2462 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2463 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2464
2465 ret = send_cmd(sctx);
2466
2467tlv_put_failure:
2468out:
2469 fs_path_free(p);
2470 return ret;
2471}
2472
2473static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2474{
2475 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2476 int ret = 0;
2477 struct fs_path *p;
2478
2479 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2480 ino, uid, gid);
2481
2482 p = fs_path_alloc();
2483 if (!p)
2484 return -ENOMEM;
2485
2486 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2487 if (ret < 0)
2488 goto out;
2489
2490 ret = get_cur_path(sctx, ino, gen, p);
2491 if (ret < 0)
2492 goto out;
2493 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2494 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2495 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2496
2497 ret = send_cmd(sctx);
2498
2499tlv_put_failure:
2500out:
2501 fs_path_free(p);
2502 return ret;
2503}
2504
2505static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2506{
2507 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2508 int ret = 0;
2509 struct fs_path *p = NULL;
2510 struct btrfs_inode_item *ii;
2511 struct btrfs_path *path = NULL;
2512 struct extent_buffer *eb;
2513 struct btrfs_key key;
2514 int slot;
2515
2516 btrfs_debug(fs_info, "send_utimes %llu", ino);
2517
2518 p = fs_path_alloc();
2519 if (!p)
2520 return -ENOMEM;
2521
2522 path = alloc_path_for_send();
2523 if (!path) {
2524 ret = -ENOMEM;
2525 goto out;
2526 }
2527
2528 key.objectid = ino;
2529 key.type = BTRFS_INODE_ITEM_KEY;
2530 key.offset = 0;
2531 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2532 if (ret > 0)
2533 ret = -ENOENT;
2534 if (ret < 0)
2535 goto out;
2536
2537 eb = path->nodes[0];
2538 slot = path->slots[0];
2539 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2540
2541 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2542 if (ret < 0)
2543 goto out;
2544
2545 ret = get_cur_path(sctx, ino, gen, p);
2546 if (ret < 0)
2547 goto out;
2548 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2549 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2550 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2551 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2552 /* TODO Add otime support when the otime patches get into upstream */
2553
2554 ret = send_cmd(sctx);
2555
2556tlv_put_failure:
2557out:
2558 fs_path_free(p);
2559 btrfs_free_path(path);
2560 return ret;
2561}
2562
2563/*
2564 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2565 * a valid path yet because we did not process the refs yet. So, the inode
2566 * is created as orphan.
2567 */
2568static int send_create_inode(struct send_ctx *sctx, u64 ino)
2569{
2570 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2571 int ret = 0;
2572 struct fs_path *p;
2573 int cmd;
2574 u64 gen;
2575 u64 mode;
2576 u64 rdev;
2577
2578 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2579
2580 p = fs_path_alloc();
2581 if (!p)
2582 return -ENOMEM;
2583
2584 if (ino != sctx->cur_ino) {
2585 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2586 NULL, NULL, &rdev);
2587 if (ret < 0)
2588 goto out;
2589 } else {
2590 gen = sctx->cur_inode_gen;
2591 mode = sctx->cur_inode_mode;
2592 rdev = sctx->cur_inode_rdev;
2593 }
2594
2595 if (S_ISREG(mode)) {
2596 cmd = BTRFS_SEND_C_MKFILE;
2597 } else if (S_ISDIR(mode)) {
2598 cmd = BTRFS_SEND_C_MKDIR;
2599 } else if (S_ISLNK(mode)) {
2600 cmd = BTRFS_SEND_C_SYMLINK;
2601 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2602 cmd = BTRFS_SEND_C_MKNOD;
2603 } else if (S_ISFIFO(mode)) {
2604 cmd = BTRFS_SEND_C_MKFIFO;
2605 } else if (S_ISSOCK(mode)) {
2606 cmd = BTRFS_SEND_C_MKSOCK;
2607 } else {
2608 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2609 (int)(mode & S_IFMT));
2610 ret = -EOPNOTSUPP;
2611 goto out;
2612 }
2613
2614 ret = begin_cmd(sctx, cmd);
2615 if (ret < 0)
2616 goto out;
2617
2618 ret = gen_unique_name(sctx, ino, gen, p);
2619 if (ret < 0)
2620 goto out;
2621
2622 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2623 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2624
2625 if (S_ISLNK(mode)) {
2626 fs_path_reset(p);
2627 ret = read_symlink(sctx->send_root, ino, p);
2628 if (ret < 0)
2629 goto out;
2630 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2631 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2632 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2633 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2634 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2635 }
2636
2637 ret = send_cmd(sctx);
2638 if (ret < 0)
2639 goto out;
2640
2641
2642tlv_put_failure:
2643out:
2644 fs_path_free(p);
2645 return ret;
2646}
2647
2648/*
2649 * We need some special handling for inodes that get processed before the parent
2650 * directory got created. See process_recorded_refs for details.
2651 * This function does the check if we already created the dir out of order.
2652 */
2653static int did_create_dir(struct send_ctx *sctx, u64 dir)
2654{
2655 int ret = 0;
2656 struct btrfs_path *path = NULL;
2657 struct btrfs_key key;
2658 struct btrfs_key found_key;
2659 struct btrfs_key di_key;
2660 struct extent_buffer *eb;
2661 struct btrfs_dir_item *di;
2662 int slot;
2663
2664 path = alloc_path_for_send();
2665 if (!path) {
2666 ret = -ENOMEM;
2667 goto out;
2668 }
2669
2670 key.objectid = dir;
2671 key.type = BTRFS_DIR_INDEX_KEY;
2672 key.offset = 0;
2673 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2674 if (ret < 0)
2675 goto out;
2676
2677 while (1) {
2678 eb = path->nodes[0];
2679 slot = path->slots[0];
2680 if (slot >= btrfs_header_nritems(eb)) {
2681 ret = btrfs_next_leaf(sctx->send_root, path);
2682 if (ret < 0) {
2683 goto out;
2684 } else if (ret > 0) {
2685 ret = 0;
2686 break;
2687 }
2688 continue;
2689 }
2690
2691 btrfs_item_key_to_cpu(eb, &found_key, slot);
2692 if (found_key.objectid != key.objectid ||
2693 found_key.type != key.type) {
2694 ret = 0;
2695 goto out;
2696 }
2697
2698 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2699 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2700
2701 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2702 di_key.objectid < sctx->send_progress) {
2703 ret = 1;
2704 goto out;
2705 }
2706
2707 path->slots[0]++;
2708 }
2709
2710out:
2711 btrfs_free_path(path);
2712 return ret;
2713}
2714
2715/*
2716 * Only creates the inode if it is:
2717 * 1. Not a directory
2718 * 2. Or a directory which was not created already due to out of order
2719 * directories. See did_create_dir and process_recorded_refs for details.
2720 */
2721static int send_create_inode_if_needed(struct send_ctx *sctx)
2722{
2723 int ret;
2724
2725 if (S_ISDIR(sctx->cur_inode_mode)) {
2726 ret = did_create_dir(sctx, sctx->cur_ino);
2727 if (ret < 0)
2728 goto out;
2729 if (ret) {
2730 ret = 0;
2731 goto out;
2732 }
2733 }
2734
2735 ret = send_create_inode(sctx, sctx->cur_ino);
2736 if (ret < 0)
2737 goto out;
2738
2739out:
2740 return ret;
2741}
2742
2743struct recorded_ref {
2744 struct list_head list;
2745 char *name;
2746 struct fs_path *full_path;
2747 u64 dir;
2748 u64 dir_gen;
2749 int name_len;
2750};
2751
2752static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2753{
2754 ref->full_path = path;
2755 ref->name = (char *)kbasename(ref->full_path->start);
2756 ref->name_len = ref->full_path->end - ref->name;
2757}
2758
2759/*
2760 * We need to process new refs before deleted refs, but compare_tree gives us
2761 * everything mixed. So we first record all refs and later process them.
2762 * This function is a helper to record one ref.
2763 */
2764static int __record_ref(struct list_head *head, u64 dir,
2765 u64 dir_gen, struct fs_path *path)
2766{
2767 struct recorded_ref *ref;
2768
2769 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2770 if (!ref)
2771 return -ENOMEM;
2772
2773 ref->dir = dir;
2774 ref->dir_gen = dir_gen;
2775 set_ref_path(ref, path);
2776 list_add_tail(&ref->list, head);
2777 return 0;
2778}
2779
2780static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2781{
2782 struct recorded_ref *new;
2783
2784 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2785 if (!new)
2786 return -ENOMEM;
2787
2788 new->dir = ref->dir;
2789 new->dir_gen = ref->dir_gen;
2790 new->full_path = NULL;
2791 INIT_LIST_HEAD(&new->list);
2792 list_add_tail(&new->list, list);
2793 return 0;
2794}
2795
2796static void __free_recorded_refs(struct list_head *head)
2797{
2798 struct recorded_ref *cur;
2799
2800 while (!list_empty(head)) {
2801 cur = list_entry(head->next, struct recorded_ref, list);
2802 fs_path_free(cur->full_path);
2803 list_del(&cur->list);
2804 kfree(cur);
2805 }
2806}
2807
2808static void free_recorded_refs(struct send_ctx *sctx)
2809{
2810 __free_recorded_refs(&sctx->new_refs);
2811 __free_recorded_refs(&sctx->deleted_refs);
2812}
2813
2814/*
2815 * Renames/moves a file/dir to its orphan name. Used when the first
2816 * ref of an unprocessed inode gets overwritten and for all non empty
2817 * directories.
2818 */
2819static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2820 struct fs_path *path)
2821{
2822 int ret;
2823 struct fs_path *orphan;
2824
2825 orphan = fs_path_alloc();
2826 if (!orphan)
2827 return -ENOMEM;
2828
2829 ret = gen_unique_name(sctx, ino, gen, orphan);
2830 if (ret < 0)
2831 goto out;
2832
2833 ret = send_rename(sctx, path, orphan);
2834
2835out:
2836 fs_path_free(orphan);
2837 return ret;
2838}
2839
2840static struct orphan_dir_info *
2841add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2842{
2843 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2844 struct rb_node *parent = NULL;
2845 struct orphan_dir_info *entry, *odi;
2846
2847 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2848 if (!odi)
2849 return ERR_PTR(-ENOMEM);
2850 odi->ino = dir_ino;
2851 odi->gen = 0;
2852
2853 while (*p) {
2854 parent = *p;
2855 entry = rb_entry(parent, struct orphan_dir_info, node);
2856 if (dir_ino < entry->ino) {
2857 p = &(*p)->rb_left;
2858 } else if (dir_ino > entry->ino) {
2859 p = &(*p)->rb_right;
2860 } else {
2861 kfree(odi);
2862 return entry;
2863 }
2864 }
2865
2866 rb_link_node(&odi->node, parent, p);
2867 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2868 return odi;
2869}
2870
2871static struct orphan_dir_info *
2872get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2873{
2874 struct rb_node *n = sctx->orphan_dirs.rb_node;
2875 struct orphan_dir_info *entry;
2876
2877 while (n) {
2878 entry = rb_entry(n, struct orphan_dir_info, node);
2879 if (dir_ino < entry->ino)
2880 n = n->rb_left;
2881 else if (dir_ino > entry->ino)
2882 n = n->rb_right;
2883 else
2884 return entry;
2885 }
2886 return NULL;
2887}
2888
2889static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2890{
2891 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2892
2893 return odi != NULL;
2894}
2895
2896static void free_orphan_dir_info(struct send_ctx *sctx,
2897 struct orphan_dir_info *odi)
2898{
2899 if (!odi)
2900 return;
2901 rb_erase(&odi->node, &sctx->orphan_dirs);
2902 kfree(odi);
2903}
2904
2905/*
2906 * Returns 1 if a directory can be removed at this point in time.
2907 * We check this by iterating all dir items and checking if the inode behind
2908 * the dir item was already processed.
2909 */
2910static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2911 u64 send_progress)
2912{
2913 int ret = 0;
2914 struct btrfs_root *root = sctx->parent_root;
2915 struct btrfs_path *path;
2916 struct btrfs_key key;
2917 struct btrfs_key found_key;
2918 struct btrfs_key loc;
2919 struct btrfs_dir_item *di;
2920
2921 /*
2922 * Don't try to rmdir the top/root subvolume dir.
2923 */
2924 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2925 return 0;
2926
2927 path = alloc_path_for_send();
2928 if (!path)
2929 return -ENOMEM;
2930
2931 key.objectid = dir;
2932 key.type = BTRFS_DIR_INDEX_KEY;
2933 key.offset = 0;
2934 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2935 if (ret < 0)
2936 goto out;
2937
2938 while (1) {
2939 struct waiting_dir_move *dm;
2940
2941 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2942 ret = btrfs_next_leaf(root, path);
2943 if (ret < 0)
2944 goto out;
2945 else if (ret > 0)
2946 break;
2947 continue;
2948 }
2949 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2950 path->slots[0]);
2951 if (found_key.objectid != key.objectid ||
2952 found_key.type != key.type)
2953 break;
2954
2955 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2956 struct btrfs_dir_item);
2957 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2958
2959 dm = get_waiting_dir_move(sctx, loc.objectid);
2960 if (dm) {
2961 struct orphan_dir_info *odi;
2962
2963 odi = add_orphan_dir_info(sctx, dir);
2964 if (IS_ERR(odi)) {
2965 ret = PTR_ERR(odi);
2966 goto out;
2967 }
2968 odi->gen = dir_gen;
2969 dm->rmdir_ino = dir;
2970 ret = 0;
2971 goto out;
2972 }
2973
2974 if (loc.objectid > send_progress) {
2975 struct orphan_dir_info *odi;
2976
2977 odi = get_orphan_dir_info(sctx, dir);
2978 free_orphan_dir_info(sctx, odi);
2979 ret = 0;
2980 goto out;
2981 }
2982
2983 path->slots[0]++;
2984 }
2985
2986 ret = 1;
2987
2988out:
2989 btrfs_free_path(path);
2990 return ret;
2991}
2992
2993static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2994{
2995 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2996
2997 return entry != NULL;
2998}
2999
3000static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3001{
3002 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3003 struct rb_node *parent = NULL;
3004 struct waiting_dir_move *entry, *dm;
3005
3006 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3007 if (!dm)
3008 return -ENOMEM;
3009 dm->ino = ino;
3010 dm->rmdir_ino = 0;
3011 dm->orphanized = orphanized;
3012
3013 while (*p) {
3014 parent = *p;
3015 entry = rb_entry(parent, struct waiting_dir_move, node);
3016 if (ino < entry->ino) {
3017 p = &(*p)->rb_left;
3018 } else if (ino > entry->ino) {
3019 p = &(*p)->rb_right;
3020 } else {
3021 kfree(dm);
3022 return -EEXIST;
3023 }
3024 }
3025
3026 rb_link_node(&dm->node, parent, p);
3027 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3028 return 0;
3029}
3030
3031static struct waiting_dir_move *
3032get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3033{
3034 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3035 struct waiting_dir_move *entry;
3036
3037 while (n) {
3038 entry = rb_entry(n, struct waiting_dir_move, node);
3039 if (ino < entry->ino)
3040 n = n->rb_left;
3041 else if (ino > entry->ino)
3042 n = n->rb_right;
3043 else
3044 return entry;
3045 }
3046 return NULL;
3047}
3048
3049static void free_waiting_dir_move(struct send_ctx *sctx,
3050 struct waiting_dir_move *dm)
3051{
3052 if (!dm)
3053 return;
3054 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3055 kfree(dm);
3056}
3057
3058static int add_pending_dir_move(struct send_ctx *sctx,
3059 u64 ino,
3060 u64 ino_gen,
3061 u64 parent_ino,
3062 struct list_head *new_refs,
3063 struct list_head *deleted_refs,
3064 const bool is_orphan)
3065{
3066 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3067 struct rb_node *parent = NULL;
3068 struct pending_dir_move *entry = NULL, *pm;
3069 struct recorded_ref *cur;
3070 int exists = 0;
3071 int ret;
3072
3073 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3074 if (!pm)
3075 return -ENOMEM;
3076 pm->parent_ino = parent_ino;
3077 pm->ino = ino;
3078 pm->gen = ino_gen;
3079 INIT_LIST_HEAD(&pm->list);
3080 INIT_LIST_HEAD(&pm->update_refs);
3081 RB_CLEAR_NODE(&pm->node);
3082
3083 while (*p) {
3084 parent = *p;
3085 entry = rb_entry(parent, struct pending_dir_move, node);
3086 if (parent_ino < entry->parent_ino) {
3087 p = &(*p)->rb_left;
3088 } else if (parent_ino > entry->parent_ino) {
3089 p = &(*p)->rb_right;
3090 } else {
3091 exists = 1;
3092 break;
3093 }
3094 }
3095
3096 list_for_each_entry(cur, deleted_refs, list) {
3097 ret = dup_ref(cur, &pm->update_refs);
3098 if (ret < 0)
3099 goto out;
3100 }
3101 list_for_each_entry(cur, new_refs, list) {
3102 ret = dup_ref(cur, &pm->update_refs);
3103 if (ret < 0)
3104 goto out;
3105 }
3106
3107 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3108 if (ret)
3109 goto out;
3110
3111 if (exists) {
3112 list_add_tail(&pm->list, &entry->list);
3113 } else {
3114 rb_link_node(&pm->node, parent, p);
3115 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3116 }
3117 ret = 0;
3118out:
3119 if (ret) {
3120 __free_recorded_refs(&pm->update_refs);
3121 kfree(pm);
3122 }
3123 return ret;
3124}
3125
3126static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3127 u64 parent_ino)
3128{
3129 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3130 struct pending_dir_move *entry;
3131
3132 while (n) {
3133 entry = rb_entry(n, struct pending_dir_move, node);
3134 if (parent_ino < entry->parent_ino)
3135 n = n->rb_left;
3136 else if (parent_ino > entry->parent_ino)
3137 n = n->rb_right;
3138 else
3139 return entry;
3140 }
3141 return NULL;
3142}
3143
3144static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3145 u64 ino, u64 gen, u64 *ancestor_ino)
3146{
3147 int ret = 0;
3148 u64 parent_inode = 0;
3149 u64 parent_gen = 0;
3150 u64 start_ino = ino;
3151
3152 *ancestor_ino = 0;
3153 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3154 fs_path_reset(name);
3155
3156 if (is_waiting_for_rm(sctx, ino))
3157 break;
3158 if (is_waiting_for_move(sctx, ino)) {
3159 if (*ancestor_ino == 0)
3160 *ancestor_ino = ino;
3161 ret = get_first_ref(sctx->parent_root, ino,
3162 &parent_inode, &parent_gen, name);
3163 } else {
3164 ret = __get_cur_name_and_parent(sctx, ino, gen,
3165 &parent_inode,
3166 &parent_gen, name);
3167 if (ret > 0) {
3168 ret = 0;
3169 break;
3170 }
3171 }
3172 if (ret < 0)
3173 break;
3174 if (parent_inode == start_ino) {
3175 ret = 1;
3176 if (*ancestor_ino == 0)
3177 *ancestor_ino = ino;
3178 break;
3179 }
3180 ino = parent_inode;
3181 gen = parent_gen;
3182 }
3183 return ret;
3184}
3185
3186static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3187{
3188 struct fs_path *from_path = NULL;
3189 struct fs_path *to_path = NULL;
3190 struct fs_path *name = NULL;
3191 u64 orig_progress = sctx->send_progress;
3192 struct recorded_ref *cur;
3193 u64 parent_ino, parent_gen;
3194 struct waiting_dir_move *dm = NULL;
3195 u64 rmdir_ino = 0;
3196 u64 ancestor;
3197 bool is_orphan;
3198 int ret;
3199
3200 name = fs_path_alloc();
3201 from_path = fs_path_alloc();
3202 if (!name || !from_path) {
3203 ret = -ENOMEM;
3204 goto out;
3205 }
3206
3207 dm = get_waiting_dir_move(sctx, pm->ino);
3208 ASSERT(dm);
3209 rmdir_ino = dm->rmdir_ino;
3210 is_orphan = dm->orphanized;
3211 free_waiting_dir_move(sctx, dm);
3212
3213 if (is_orphan) {
3214 ret = gen_unique_name(sctx, pm->ino,
3215 pm->gen, from_path);
3216 } else {
3217 ret = get_first_ref(sctx->parent_root, pm->ino,
3218 &parent_ino, &parent_gen, name);
3219 if (ret < 0)
3220 goto out;
3221 ret = get_cur_path(sctx, parent_ino, parent_gen,
3222 from_path);
3223 if (ret < 0)
3224 goto out;
3225 ret = fs_path_add_path(from_path, name);
3226 }
3227 if (ret < 0)
3228 goto out;
3229
3230 sctx->send_progress = sctx->cur_ino + 1;
3231 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3232 if (ret < 0)
3233 goto out;
3234 if (ret) {
3235 LIST_HEAD(deleted_refs);
3236 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3237 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3238 &pm->update_refs, &deleted_refs,
3239 is_orphan);
3240 if (ret < 0)
3241 goto out;
3242 if (rmdir_ino) {
3243 dm = get_waiting_dir_move(sctx, pm->ino);
3244 ASSERT(dm);
3245 dm->rmdir_ino = rmdir_ino;
3246 }
3247 goto out;
3248 }
3249 fs_path_reset(name);
3250 to_path = name;
3251 name = NULL;
3252 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3253 if (ret < 0)
3254 goto out;
3255
3256 ret = send_rename(sctx, from_path, to_path);
3257 if (ret < 0)
3258 goto out;
3259
3260 if (rmdir_ino) {
3261 struct orphan_dir_info *odi;
3262
3263 odi = get_orphan_dir_info(sctx, rmdir_ino);
3264 if (!odi) {
3265 /* already deleted */
3266 goto finish;
3267 }
3268 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino);
3269 if (ret < 0)
3270 goto out;
3271 if (!ret)
3272 goto finish;
3273
3274 name = fs_path_alloc();
3275 if (!name) {
3276 ret = -ENOMEM;
3277 goto out;
3278 }
3279 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3280 if (ret < 0)
3281 goto out;
3282 ret = send_rmdir(sctx, name);
3283 if (ret < 0)
3284 goto out;
3285 free_orphan_dir_info(sctx, odi);
3286 }
3287
3288finish:
3289 ret = send_utimes(sctx, pm->ino, pm->gen);
3290 if (ret < 0)
3291 goto out;
3292
3293 /*
3294 * After rename/move, need to update the utimes of both new parent(s)
3295 * and old parent(s).
3296 */
3297 list_for_each_entry(cur, &pm->update_refs, list) {
3298 /*
3299 * The parent inode might have been deleted in the send snapshot
3300 */
3301 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3302 NULL, NULL, NULL, NULL, NULL);
3303 if (ret == -ENOENT) {
3304 ret = 0;
3305 continue;
3306 }
3307 if (ret < 0)
3308 goto out;
3309
3310 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3311 if (ret < 0)
3312 goto out;
3313 }
3314
3315out:
3316 fs_path_free(name);
3317 fs_path_free(from_path);
3318 fs_path_free(to_path);
3319 sctx->send_progress = orig_progress;
3320
3321 return ret;
3322}
3323
3324static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3325{
3326 if (!list_empty(&m->list))
3327 list_del(&m->list);
3328 if (!RB_EMPTY_NODE(&m->node))
3329 rb_erase(&m->node, &sctx->pending_dir_moves);
3330 __free_recorded_refs(&m->update_refs);
3331 kfree(m);
3332}
3333
3334static void tail_append_pending_moves(struct pending_dir_move *moves,
3335 struct list_head *stack)
3336{
3337 if (list_empty(&moves->list)) {
3338 list_add_tail(&moves->list, stack);
3339 } else {
3340 LIST_HEAD(list);
3341 list_splice_init(&moves->list, &list);
3342 list_add_tail(&moves->list, stack);
3343 list_splice_tail(&list, stack);
3344 }
3345}
3346
3347static int apply_children_dir_moves(struct send_ctx *sctx)
3348{
3349 struct pending_dir_move *pm;
3350 struct list_head stack;
3351 u64 parent_ino = sctx->cur_ino;
3352 int ret = 0;
3353
3354 pm = get_pending_dir_moves(sctx, parent_ino);
3355 if (!pm)
3356 return 0;
3357
3358 INIT_LIST_HEAD(&stack);
3359 tail_append_pending_moves(pm, &stack);
3360
3361 while (!list_empty(&stack)) {
3362 pm = list_first_entry(&stack, struct pending_dir_move, list);
3363 parent_ino = pm->ino;
3364 ret = apply_dir_move(sctx, pm);
3365 free_pending_move(sctx, pm);
3366 if (ret)
3367 goto out;
3368 pm = get_pending_dir_moves(sctx, parent_ino);
3369 if (pm)
3370 tail_append_pending_moves(pm, &stack);
3371 }
3372 return 0;
3373
3374out:
3375 while (!list_empty(&stack)) {
3376 pm = list_first_entry(&stack, struct pending_dir_move, list);
3377 free_pending_move(sctx, pm);
3378 }
3379 return ret;
3380}
3381
3382/*
3383 * We might need to delay a directory rename even when no ancestor directory
3384 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3385 * renamed. This happens when we rename a directory to the old name (the name
3386 * in the parent root) of some other unrelated directory that got its rename
3387 * delayed due to some ancestor with higher number that got renamed.
3388 *
3389 * Example:
3390 *
3391 * Parent snapshot:
3392 * . (ino 256)
3393 * |---- a/ (ino 257)
3394 * | |---- file (ino 260)
3395 * |
3396 * |---- b/ (ino 258)
3397 * |---- c/ (ino 259)
3398 *
3399 * Send snapshot:
3400 * . (ino 256)
3401 * |---- a/ (ino 258)
3402 * |---- x/ (ino 259)
3403 * |---- y/ (ino 257)
3404 * |----- file (ino 260)
3405 *
3406 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3407 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3408 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3409 * must issue is:
3410 *
3411 * 1 - rename 259 from 'c' to 'x'
3412 * 2 - rename 257 from 'a' to 'x/y'
3413 * 3 - rename 258 from 'b' to 'a'
3414 *
3415 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3416 * be done right away and < 0 on error.
3417 */
3418static int wait_for_dest_dir_move(struct send_ctx *sctx,
3419 struct recorded_ref *parent_ref,
3420 const bool is_orphan)
3421{
3422 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3423 struct btrfs_path *path;
3424 struct btrfs_key key;
3425 struct btrfs_key di_key;
3426 struct btrfs_dir_item *di;
3427 u64 left_gen;
3428 u64 right_gen;
3429 int ret = 0;
3430 struct waiting_dir_move *wdm;
3431
3432 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3433 return 0;
3434
3435 path = alloc_path_for_send();
3436 if (!path)
3437 return -ENOMEM;
3438
3439 key.objectid = parent_ref->dir;
3440 key.type = BTRFS_DIR_ITEM_KEY;
3441 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3442
3443 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3444 if (ret < 0) {
3445 goto out;
3446 } else if (ret > 0) {
3447 ret = 0;
3448 goto out;
3449 }
3450
3451 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3452 parent_ref->name_len);
3453 if (!di) {
3454 ret = 0;
3455 goto out;
3456 }
3457 /*
3458 * di_key.objectid has the number of the inode that has a dentry in the
3459 * parent directory with the same name that sctx->cur_ino is being
3460 * renamed to. We need to check if that inode is in the send root as
3461 * well and if it is currently marked as an inode with a pending rename,
3462 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3463 * that it happens after that other inode is renamed.
3464 */
3465 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3466 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3467 ret = 0;
3468 goto out;
3469 }
3470
3471 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3472 &left_gen, NULL, NULL, NULL, NULL);
3473 if (ret < 0)
3474 goto out;
3475 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3476 &right_gen, NULL, NULL, NULL, NULL);
3477 if (ret < 0) {
3478 if (ret == -ENOENT)
3479 ret = 0;
3480 goto out;
3481 }
3482
3483 /* Different inode, no need to delay the rename of sctx->cur_ino */
3484 if (right_gen != left_gen) {
3485 ret = 0;
3486 goto out;
3487 }
3488
3489 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3490 if (wdm && !wdm->orphanized) {
3491 ret = add_pending_dir_move(sctx,
3492 sctx->cur_ino,
3493 sctx->cur_inode_gen,
3494 di_key.objectid,
3495 &sctx->new_refs,
3496 &sctx->deleted_refs,
3497 is_orphan);
3498 if (!ret)
3499 ret = 1;
3500 }
3501out:
3502 btrfs_free_path(path);
3503 return ret;
3504}
3505
3506/*
3507 * Check if inode ino2, or any of its ancestors, is inode ino1.
3508 * Return 1 if true, 0 if false and < 0 on error.
3509 */
3510static int check_ino_in_path(struct btrfs_root *root,
3511 const u64 ino1,
3512 const u64 ino1_gen,
3513 const u64 ino2,
3514 const u64 ino2_gen,
3515 struct fs_path *fs_path)
3516{
3517 u64 ino = ino2;
3518
3519 if (ino1 == ino2)
3520 return ino1_gen == ino2_gen;
3521
3522 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3523 u64 parent;
3524 u64 parent_gen;
3525 int ret;
3526
3527 fs_path_reset(fs_path);
3528 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3529 if (ret < 0)
3530 return ret;
3531 if (parent == ino1)
3532 return parent_gen == ino1_gen;
3533 ino = parent;
3534 }
3535 return 0;
3536}
3537
3538/*
3539 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3540 * possible path (in case ino2 is not a directory and has multiple hard links).
3541 * Return 1 if true, 0 if false and < 0 on error.
3542 */
3543static int is_ancestor(struct btrfs_root *root,
3544 const u64 ino1,
3545 const u64 ino1_gen,
3546 const u64 ino2,
3547 struct fs_path *fs_path)
3548{
3549 bool free_fs_path = false;
3550 int ret = 0;
3551 struct btrfs_path *path = NULL;
3552 struct btrfs_key key;
3553
3554 if (!fs_path) {
3555 fs_path = fs_path_alloc();
3556 if (!fs_path)
3557 return -ENOMEM;
3558 free_fs_path = true;
3559 }
3560
3561 path = alloc_path_for_send();
3562 if (!path) {
3563 ret = -ENOMEM;
3564 goto out;
3565 }
3566
3567 key.objectid = ino2;
3568 key.type = BTRFS_INODE_REF_KEY;
3569 key.offset = 0;
3570
3571 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3572 if (ret < 0)
3573 goto out;
3574
3575 while (true) {
3576 struct extent_buffer *leaf = path->nodes[0];
3577 int slot = path->slots[0];
3578 u32 cur_offset = 0;
3579 u32 item_size;
3580
3581 if (slot >= btrfs_header_nritems(leaf)) {
3582 ret = btrfs_next_leaf(root, path);
3583 if (ret < 0)
3584 goto out;
3585 if (ret > 0)
3586 break;
3587 continue;
3588 }
3589
3590 btrfs_item_key_to_cpu(leaf, &key, slot);
3591 if (key.objectid != ino2)
3592 break;
3593 if (key.type != BTRFS_INODE_REF_KEY &&
3594 key.type != BTRFS_INODE_EXTREF_KEY)
3595 break;
3596
3597 item_size = btrfs_item_size_nr(leaf, slot);
3598 while (cur_offset < item_size) {
3599 u64 parent;
3600 u64 parent_gen;
3601
3602 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3603 unsigned long ptr;
3604 struct btrfs_inode_extref *extref;
3605
3606 ptr = btrfs_item_ptr_offset(leaf, slot);
3607 extref = (struct btrfs_inode_extref *)
3608 (ptr + cur_offset);
3609 parent = btrfs_inode_extref_parent(leaf,
3610 extref);
3611 cur_offset += sizeof(*extref);
3612 cur_offset += btrfs_inode_extref_name_len(leaf,
3613 extref);
3614 } else {
3615 parent = key.offset;
3616 cur_offset = item_size;
3617 }
3618
3619 ret = get_inode_info(root, parent, NULL, &parent_gen,
3620 NULL, NULL, NULL, NULL);
3621 if (ret < 0)
3622 goto out;
3623 ret = check_ino_in_path(root, ino1, ino1_gen,
3624 parent, parent_gen, fs_path);
3625 if (ret)
3626 goto out;
3627 }
3628 path->slots[0]++;
3629 }
3630 ret = 0;
3631 out:
3632 btrfs_free_path(path);
3633 if (free_fs_path)
3634 fs_path_free(fs_path);
3635 return ret;
3636}
3637
3638static int wait_for_parent_move(struct send_ctx *sctx,
3639 struct recorded_ref *parent_ref,
3640 const bool is_orphan)
3641{
3642 int ret = 0;
3643 u64 ino = parent_ref->dir;
3644 u64 ino_gen = parent_ref->dir_gen;
3645 u64 parent_ino_before, parent_ino_after;
3646 struct fs_path *path_before = NULL;
3647 struct fs_path *path_after = NULL;
3648 int len1, len2;
3649
3650 path_after = fs_path_alloc();
3651 path_before = fs_path_alloc();
3652 if (!path_after || !path_before) {
3653 ret = -ENOMEM;
3654 goto out;
3655 }
3656
3657 /*
3658 * Our current directory inode may not yet be renamed/moved because some
3659 * ancestor (immediate or not) has to be renamed/moved first. So find if
3660 * such ancestor exists and make sure our own rename/move happens after
3661 * that ancestor is processed to avoid path build infinite loops (done
3662 * at get_cur_path()).
3663 */
3664 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3665 u64 parent_ino_after_gen;
3666
3667 if (is_waiting_for_move(sctx, ino)) {
3668 /*
3669 * If the current inode is an ancestor of ino in the
3670 * parent root, we need to delay the rename of the
3671 * current inode, otherwise don't delayed the rename
3672 * because we can end up with a circular dependency
3673 * of renames, resulting in some directories never
3674 * getting the respective rename operations issued in
3675 * the send stream or getting into infinite path build
3676 * loops.
3677 */
3678 ret = is_ancestor(sctx->parent_root,
3679 sctx->cur_ino, sctx->cur_inode_gen,
3680 ino, path_before);
3681 if (ret)
3682 break;
3683 }
3684
3685 fs_path_reset(path_before);
3686 fs_path_reset(path_after);
3687
3688 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3689 &parent_ino_after_gen, path_after);
3690 if (ret < 0)
3691 goto out;
3692 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3693 NULL, path_before);
3694 if (ret < 0 && ret != -ENOENT) {
3695 goto out;
3696 } else if (ret == -ENOENT) {
3697 ret = 0;
3698 break;
3699 }
3700
3701 len1 = fs_path_len(path_before);
3702 len2 = fs_path_len(path_after);
3703 if (ino > sctx->cur_ino &&
3704 (parent_ino_before != parent_ino_after || len1 != len2 ||
3705 memcmp(path_before->start, path_after->start, len1))) {
3706 u64 parent_ino_gen;
3707
3708 ret = get_inode_info(sctx->parent_root, ino, NULL,
3709 &parent_ino_gen, NULL, NULL, NULL,
3710 NULL);
3711 if (ret < 0)
3712 goto out;
3713 if (ino_gen == parent_ino_gen) {
3714 ret = 1;
3715 break;
3716 }
3717 }
3718 ino = parent_ino_after;
3719 ino_gen = parent_ino_after_gen;
3720 }
3721
3722out:
3723 fs_path_free(path_before);
3724 fs_path_free(path_after);
3725
3726 if (ret == 1) {
3727 ret = add_pending_dir_move(sctx,
3728 sctx->cur_ino,
3729 sctx->cur_inode_gen,
3730 ino,
3731 &sctx->new_refs,
3732 &sctx->deleted_refs,
3733 is_orphan);
3734 if (!ret)
3735 ret = 1;
3736 }
3737
3738 return ret;
3739}
3740
3741static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3742{
3743 int ret;
3744 struct fs_path *new_path;
3745
3746 /*
3747 * Our reference's name member points to its full_path member string, so
3748 * we use here a new path.
3749 */
3750 new_path = fs_path_alloc();
3751 if (!new_path)
3752 return -ENOMEM;
3753
3754 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3755 if (ret < 0) {
3756 fs_path_free(new_path);
3757 return ret;
3758 }
3759 ret = fs_path_add(new_path, ref->name, ref->name_len);
3760 if (ret < 0) {
3761 fs_path_free(new_path);
3762 return ret;
3763 }
3764
3765 fs_path_free(ref->full_path);
3766 set_ref_path(ref, new_path);
3767
3768 return 0;
3769}
3770
3771/*
3772 * This does all the move/link/unlink/rmdir magic.
3773 */
3774static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3775{
3776 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3777 int ret = 0;
3778 struct recorded_ref *cur;
3779 struct recorded_ref *cur2;
3780 struct list_head check_dirs;
3781 struct fs_path *valid_path = NULL;
3782 u64 ow_inode = 0;
3783 u64 ow_gen;
3784 u64 ow_mode;
3785 int did_overwrite = 0;
3786 int is_orphan = 0;
3787 u64 last_dir_ino_rm = 0;
3788 bool can_rename = true;
3789 bool orphanized_dir = false;
3790 bool orphanized_ancestor = false;
3791
3792 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3793
3794 /*
3795 * This should never happen as the root dir always has the same ref
3796 * which is always '..'
3797 */
3798 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3799 INIT_LIST_HEAD(&check_dirs);
3800
3801 valid_path = fs_path_alloc();
3802 if (!valid_path) {
3803 ret = -ENOMEM;
3804 goto out;
3805 }
3806
3807 /*
3808 * First, check if the first ref of the current inode was overwritten
3809 * before. If yes, we know that the current inode was already orphanized
3810 * and thus use the orphan name. If not, we can use get_cur_path to
3811 * get the path of the first ref as it would like while receiving at
3812 * this point in time.
3813 * New inodes are always orphan at the beginning, so force to use the
3814 * orphan name in this case.
3815 * The first ref is stored in valid_path and will be updated if it
3816 * gets moved around.
3817 */
3818 if (!sctx->cur_inode_new) {
3819 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3820 sctx->cur_inode_gen);
3821 if (ret < 0)
3822 goto out;
3823 if (ret)
3824 did_overwrite = 1;
3825 }
3826 if (sctx->cur_inode_new || did_overwrite) {
3827 ret = gen_unique_name(sctx, sctx->cur_ino,
3828 sctx->cur_inode_gen, valid_path);
3829 if (ret < 0)
3830 goto out;
3831 is_orphan = 1;
3832 } else {
3833 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3834 valid_path);
3835 if (ret < 0)
3836 goto out;
3837 }
3838
3839 list_for_each_entry(cur, &sctx->new_refs, list) {
3840 /*
3841 * We may have refs where the parent directory does not exist
3842 * yet. This happens if the parent directories inum is higher
3843 * the the current inum. To handle this case, we create the
3844 * parent directory out of order. But we need to check if this
3845 * did already happen before due to other refs in the same dir.
3846 */
3847 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3848 if (ret < 0)
3849 goto out;
3850 if (ret == inode_state_will_create) {
3851 ret = 0;
3852 /*
3853 * First check if any of the current inodes refs did
3854 * already create the dir.
3855 */
3856 list_for_each_entry(cur2, &sctx->new_refs, list) {
3857 if (cur == cur2)
3858 break;
3859 if (cur2->dir == cur->dir) {
3860 ret = 1;
3861 break;
3862 }
3863 }
3864
3865 /*
3866 * If that did not happen, check if a previous inode
3867 * did already create the dir.
3868 */
3869 if (!ret)
3870 ret = did_create_dir(sctx, cur->dir);
3871 if (ret < 0)
3872 goto out;
3873 if (!ret) {
3874 ret = send_create_inode(sctx, cur->dir);
3875 if (ret < 0)
3876 goto out;
3877 }
3878 }
3879
3880 /*
3881 * Check if this new ref would overwrite the first ref of
3882 * another unprocessed inode. If yes, orphanize the
3883 * overwritten inode. If we find an overwritten ref that is
3884 * not the first ref, simply unlink it.
3885 */
3886 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3887 cur->name, cur->name_len,
3888 &ow_inode, &ow_gen, &ow_mode);
3889 if (ret < 0)
3890 goto out;
3891 if (ret) {
3892 ret = is_first_ref(sctx->parent_root,
3893 ow_inode, cur->dir, cur->name,
3894 cur->name_len);
3895 if (ret < 0)
3896 goto out;
3897 if (ret) {
3898 struct name_cache_entry *nce;
3899 struct waiting_dir_move *wdm;
3900
3901 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3902 cur->full_path);
3903 if (ret < 0)
3904 goto out;
3905 if (S_ISDIR(ow_mode))
3906 orphanized_dir = true;
3907
3908 /*
3909 * If ow_inode has its rename operation delayed
3910 * make sure that its orphanized name is used in
3911 * the source path when performing its rename
3912 * operation.
3913 */
3914 if (is_waiting_for_move(sctx, ow_inode)) {
3915 wdm = get_waiting_dir_move(sctx,
3916 ow_inode);
3917 ASSERT(wdm);
3918 wdm->orphanized = true;
3919 }
3920
3921 /*
3922 * Make sure we clear our orphanized inode's
3923 * name from the name cache. This is because the
3924 * inode ow_inode might be an ancestor of some
3925 * other inode that will be orphanized as well
3926 * later and has an inode number greater than
3927 * sctx->send_progress. We need to prevent
3928 * future name lookups from using the old name
3929 * and get instead the orphan name.
3930 */
3931 nce = name_cache_search(sctx, ow_inode, ow_gen);
3932 if (nce) {
3933 name_cache_delete(sctx, nce);
3934 kfree(nce);
3935 }
3936
3937 /*
3938 * ow_inode might currently be an ancestor of
3939 * cur_ino, therefore compute valid_path (the
3940 * current path of cur_ino) again because it
3941 * might contain the pre-orphanization name of
3942 * ow_inode, which is no longer valid.
3943 */
3944 ret = is_ancestor(sctx->parent_root,
3945 ow_inode, ow_gen,
3946 sctx->cur_ino, NULL);
3947 if (ret > 0) {
3948 orphanized_ancestor = true;
3949 fs_path_reset(valid_path);
3950 ret = get_cur_path(sctx, sctx->cur_ino,
3951 sctx->cur_inode_gen,
3952 valid_path);
3953 }
3954 if (ret < 0)
3955 goto out;
3956 } else {
3957 ret = send_unlink(sctx, cur->full_path);
3958 if (ret < 0)
3959 goto out;
3960 }
3961 }
3962
3963 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3964 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3965 if (ret < 0)
3966 goto out;
3967 if (ret == 1) {
3968 can_rename = false;
3969 *pending_move = 1;
3970 }
3971 }
3972
3973 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3974 can_rename) {
3975 ret = wait_for_parent_move(sctx, cur, is_orphan);
3976 if (ret < 0)
3977 goto out;
3978 if (ret == 1) {
3979 can_rename = false;
3980 *pending_move = 1;
3981 }
3982 }
3983
3984 /*
3985 * link/move the ref to the new place. If we have an orphan
3986 * inode, move it and update valid_path. If not, link or move
3987 * it depending on the inode mode.
3988 */
3989 if (is_orphan && can_rename) {
3990 ret = send_rename(sctx, valid_path, cur->full_path);
3991 if (ret < 0)
3992 goto out;
3993 is_orphan = 0;
3994 ret = fs_path_copy(valid_path, cur->full_path);
3995 if (ret < 0)
3996 goto out;
3997 } else if (can_rename) {
3998 if (S_ISDIR(sctx->cur_inode_mode)) {
3999 /*
4000 * Dirs can't be linked, so move it. For moved
4001 * dirs, we always have one new and one deleted
4002 * ref. The deleted ref is ignored later.
4003 */
4004 ret = send_rename(sctx, valid_path,
4005 cur->full_path);
4006 if (!ret)
4007 ret = fs_path_copy(valid_path,
4008 cur->full_path);
4009 if (ret < 0)
4010 goto out;
4011 } else {
4012 /*
4013 * We might have previously orphanized an inode
4014 * which is an ancestor of our current inode,
4015 * so our reference's full path, which was
4016 * computed before any such orphanizations, must
4017 * be updated.
4018 */
4019 if (orphanized_dir) {
4020 ret = update_ref_path(sctx, cur);
4021 if (ret < 0)
4022 goto out;
4023 }
4024 ret = send_link(sctx, cur->full_path,
4025 valid_path);
4026 if (ret < 0)
4027 goto out;
4028 }
4029 }
4030 ret = dup_ref(cur, &check_dirs);
4031 if (ret < 0)
4032 goto out;
4033 }
4034
4035 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4036 /*
4037 * Check if we can already rmdir the directory. If not,
4038 * orphanize it. For every dir item inside that gets deleted
4039 * later, we do this check again and rmdir it then if possible.
4040 * See the use of check_dirs for more details.
4041 */
4042 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4043 sctx->cur_ino);
4044 if (ret < 0)
4045 goto out;
4046 if (ret) {
4047 ret = send_rmdir(sctx, valid_path);
4048 if (ret < 0)
4049 goto out;
4050 } else if (!is_orphan) {
4051 ret = orphanize_inode(sctx, sctx->cur_ino,
4052 sctx->cur_inode_gen, valid_path);
4053 if (ret < 0)
4054 goto out;
4055 is_orphan = 1;
4056 }
4057
4058 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4059 ret = dup_ref(cur, &check_dirs);
4060 if (ret < 0)
4061 goto out;
4062 }
4063 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4064 !list_empty(&sctx->deleted_refs)) {
4065 /*
4066 * We have a moved dir. Add the old parent to check_dirs
4067 */
4068 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4069 list);
4070 ret = dup_ref(cur, &check_dirs);
4071 if (ret < 0)
4072 goto out;
4073 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4074 /*
4075 * We have a non dir inode. Go through all deleted refs and
4076 * unlink them if they were not already overwritten by other
4077 * inodes.
4078 */
4079 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4080 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4081 sctx->cur_ino, sctx->cur_inode_gen,
4082 cur->name, cur->name_len);
4083 if (ret < 0)
4084 goto out;
4085 if (!ret) {
4086 /*
4087 * If we orphanized any ancestor before, we need
4088 * to recompute the full path for deleted names,
4089 * since any such path was computed before we
4090 * processed any references and orphanized any
4091 * ancestor inode.
4092 */
4093 if (orphanized_ancestor) {
4094 ret = update_ref_path(sctx, cur);
4095 if (ret < 0)
4096 goto out;
4097 }
4098 ret = send_unlink(sctx, cur->full_path);
4099 if (ret < 0)
4100 goto out;
4101 }
4102 ret = dup_ref(cur, &check_dirs);
4103 if (ret < 0)
4104 goto out;
4105 }
4106 /*
4107 * If the inode is still orphan, unlink the orphan. This may
4108 * happen when a previous inode did overwrite the first ref
4109 * of this inode and no new refs were added for the current
4110 * inode. Unlinking does not mean that the inode is deleted in
4111 * all cases. There may still be links to this inode in other
4112 * places.
4113 */
4114 if (is_orphan) {
4115 ret = send_unlink(sctx, valid_path);
4116 if (ret < 0)
4117 goto out;
4118 }
4119 }
4120
4121 /*
4122 * We did collect all parent dirs where cur_inode was once located. We
4123 * now go through all these dirs and check if they are pending for
4124 * deletion and if it's finally possible to perform the rmdir now.
4125 * We also update the inode stats of the parent dirs here.
4126 */
4127 list_for_each_entry(cur, &check_dirs, list) {
4128 /*
4129 * In case we had refs into dirs that were not processed yet,
4130 * we don't need to do the utime and rmdir logic for these dirs.
4131 * The dir will be processed later.
4132 */
4133 if (cur->dir > sctx->cur_ino)
4134 continue;
4135
4136 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4137 if (ret < 0)
4138 goto out;
4139
4140 if (ret == inode_state_did_create ||
4141 ret == inode_state_no_change) {
4142 /* TODO delayed utimes */
4143 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4144 if (ret < 0)
4145 goto out;
4146 } else if (ret == inode_state_did_delete &&
4147 cur->dir != last_dir_ino_rm) {
4148 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4149 sctx->cur_ino);
4150 if (ret < 0)
4151 goto out;
4152 if (ret) {
4153 ret = get_cur_path(sctx, cur->dir,
4154 cur->dir_gen, valid_path);
4155 if (ret < 0)
4156 goto out;
4157 ret = send_rmdir(sctx, valid_path);
4158 if (ret < 0)
4159 goto out;
4160 last_dir_ino_rm = cur->dir;
4161 }
4162 }
4163 }
4164
4165 ret = 0;
4166
4167out:
4168 __free_recorded_refs(&check_dirs);
4169 free_recorded_refs(sctx);
4170 fs_path_free(valid_path);
4171 return ret;
4172}
4173
4174static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4175 void *ctx, struct list_head *refs)
4176{
4177 int ret = 0;
4178 struct send_ctx *sctx = ctx;
4179 struct fs_path *p;
4180 u64 gen;
4181
4182 p = fs_path_alloc();
4183 if (!p)
4184 return -ENOMEM;
4185
4186 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4187 NULL, NULL);
4188 if (ret < 0)
4189 goto out;
4190
4191 ret = get_cur_path(sctx, dir, gen, p);
4192 if (ret < 0)
4193 goto out;
4194 ret = fs_path_add_path(p, name);
4195 if (ret < 0)
4196 goto out;
4197
4198 ret = __record_ref(refs, dir, gen, p);
4199
4200out:
4201 if (ret)
4202 fs_path_free(p);
4203 return ret;
4204}
4205
4206static int __record_new_ref(int num, u64 dir, int index,
4207 struct fs_path *name,
4208 void *ctx)
4209{
4210 struct send_ctx *sctx = ctx;
4211 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4212}
4213
4214
4215static int __record_deleted_ref(int num, u64 dir, int index,
4216 struct fs_path *name,
4217 void *ctx)
4218{
4219 struct send_ctx *sctx = ctx;
4220 return record_ref(sctx->parent_root, dir, name, ctx,
4221 &sctx->deleted_refs);
4222}
4223
4224static int record_new_ref(struct send_ctx *sctx)
4225{
4226 int ret;
4227
4228 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4229 sctx->cmp_key, 0, __record_new_ref, sctx);
4230 if (ret < 0)
4231 goto out;
4232 ret = 0;
4233
4234out:
4235 return ret;
4236}
4237
4238static int record_deleted_ref(struct send_ctx *sctx)
4239{
4240 int ret;
4241
4242 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4243 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4244 if (ret < 0)
4245 goto out;
4246 ret = 0;
4247
4248out:
4249 return ret;
4250}
4251
4252struct find_ref_ctx {
4253 u64 dir;
4254 u64 dir_gen;
4255 struct btrfs_root *root;
4256 struct fs_path *name;
4257 int found_idx;
4258};
4259
4260static int __find_iref(int num, u64 dir, int index,
4261 struct fs_path *name,
4262 void *ctx_)
4263{
4264 struct find_ref_ctx *ctx = ctx_;
4265 u64 dir_gen;
4266 int ret;
4267
4268 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4269 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4270 /*
4271 * To avoid doing extra lookups we'll only do this if everything
4272 * else matches.
4273 */
4274 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4275 NULL, NULL, NULL);
4276 if (ret)
4277 return ret;
4278 if (dir_gen != ctx->dir_gen)
4279 return 0;
4280 ctx->found_idx = num;
4281 return 1;
4282 }
4283 return 0;
4284}
4285
4286static int find_iref(struct btrfs_root *root,
4287 struct btrfs_path *path,
4288 struct btrfs_key *key,
4289 u64 dir, u64 dir_gen, struct fs_path *name)
4290{
4291 int ret;
4292 struct find_ref_ctx ctx;
4293
4294 ctx.dir = dir;
4295 ctx.name = name;
4296 ctx.dir_gen = dir_gen;
4297 ctx.found_idx = -1;
4298 ctx.root = root;
4299
4300 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4301 if (ret < 0)
4302 return ret;
4303
4304 if (ctx.found_idx == -1)
4305 return -ENOENT;
4306
4307 return ctx.found_idx;
4308}
4309
4310static int __record_changed_new_ref(int num, u64 dir, int index,
4311 struct fs_path *name,
4312 void *ctx)
4313{
4314 u64 dir_gen;
4315 int ret;
4316 struct send_ctx *sctx = ctx;
4317
4318 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4319 NULL, NULL, NULL);
4320 if (ret)
4321 return ret;
4322
4323 ret = find_iref(sctx->parent_root, sctx->right_path,
4324 sctx->cmp_key, dir, dir_gen, name);
4325 if (ret == -ENOENT)
4326 ret = __record_new_ref(num, dir, index, name, sctx);
4327 else if (ret > 0)
4328 ret = 0;
4329
4330 return ret;
4331}
4332
4333static int __record_changed_deleted_ref(int num, u64 dir, int index,
4334 struct fs_path *name,
4335 void *ctx)
4336{
4337 u64 dir_gen;
4338 int ret;
4339 struct send_ctx *sctx = ctx;
4340
4341 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4342 NULL, NULL, NULL);
4343 if (ret)
4344 return ret;
4345
4346 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4347 dir, dir_gen, name);
4348 if (ret == -ENOENT)
4349 ret = __record_deleted_ref(num, dir, index, name, sctx);
4350 else if (ret > 0)
4351 ret = 0;
4352
4353 return ret;
4354}
4355
4356static int record_changed_ref(struct send_ctx *sctx)
4357{
4358 int ret = 0;
4359
4360 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4361 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4362 if (ret < 0)
4363 goto out;
4364 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4365 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4366 if (ret < 0)
4367 goto out;
4368 ret = 0;
4369
4370out:
4371 return ret;
4372}
4373
4374/*
4375 * Record and process all refs at once. Needed when an inode changes the
4376 * generation number, which means that it was deleted and recreated.
4377 */
4378static int process_all_refs(struct send_ctx *sctx,
4379 enum btrfs_compare_tree_result cmd)
4380{
4381 int ret;
4382 struct btrfs_root *root;
4383 struct btrfs_path *path;
4384 struct btrfs_key key;
4385 struct btrfs_key found_key;
4386 struct extent_buffer *eb;
4387 int slot;
4388 iterate_inode_ref_t cb;
4389 int pending_move = 0;
4390
4391 path = alloc_path_for_send();
4392 if (!path)
4393 return -ENOMEM;
4394
4395 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4396 root = sctx->send_root;
4397 cb = __record_new_ref;
4398 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4399 root = sctx->parent_root;
4400 cb = __record_deleted_ref;
4401 } else {
4402 btrfs_err(sctx->send_root->fs_info,
4403 "Wrong command %d in process_all_refs", cmd);
4404 ret = -EINVAL;
4405 goto out;
4406 }
4407
4408 key.objectid = sctx->cmp_key->objectid;
4409 key.type = BTRFS_INODE_REF_KEY;
4410 key.offset = 0;
4411 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4412 if (ret < 0)
4413 goto out;
4414
4415 while (1) {
4416 eb = path->nodes[0];
4417 slot = path->slots[0];
4418 if (slot >= btrfs_header_nritems(eb)) {
4419 ret = btrfs_next_leaf(root, path);
4420 if (ret < 0)
4421 goto out;
4422 else if (ret > 0)
4423 break;
4424 continue;
4425 }
4426
4427 btrfs_item_key_to_cpu(eb, &found_key, slot);
4428
4429 if (found_key.objectid != key.objectid ||
4430 (found_key.type != BTRFS_INODE_REF_KEY &&
4431 found_key.type != BTRFS_INODE_EXTREF_KEY))
4432 break;
4433
4434 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4435 if (ret < 0)
4436 goto out;
4437
4438 path->slots[0]++;
4439 }
4440 btrfs_release_path(path);
4441
4442 /*
4443 * We don't actually care about pending_move as we are simply
4444 * re-creating this inode and will be rename'ing it into place once we
4445 * rename the parent directory.
4446 */
4447 ret = process_recorded_refs(sctx, &pending_move);
4448out:
4449 btrfs_free_path(path);
4450 return ret;
4451}
4452
4453static int send_set_xattr(struct send_ctx *sctx,
4454 struct fs_path *path,
4455 const char *name, int name_len,
4456 const char *data, int data_len)
4457{
4458 int ret = 0;
4459
4460 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4461 if (ret < 0)
4462 goto out;
4463
4464 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4465 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4466 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4467
4468 ret = send_cmd(sctx);
4469
4470tlv_put_failure:
4471out:
4472 return ret;
4473}
4474
4475static int send_remove_xattr(struct send_ctx *sctx,
4476 struct fs_path *path,
4477 const char *name, int name_len)
4478{
4479 int ret = 0;
4480
4481 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4482 if (ret < 0)
4483 goto out;
4484
4485 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4486 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4487
4488 ret = send_cmd(sctx);
4489
4490tlv_put_failure:
4491out:
4492 return ret;
4493}
4494
4495static int __process_new_xattr(int num, struct btrfs_key *di_key,
4496 const char *name, int name_len,
4497 const char *data, int data_len,
4498 u8 type, void *ctx)
4499{
4500 int ret;
4501 struct send_ctx *sctx = ctx;
4502 struct fs_path *p;
4503 struct posix_acl_xattr_header dummy_acl;
4504
4505 p = fs_path_alloc();
4506 if (!p)
4507 return -ENOMEM;
4508
4509 /*
4510 * This hack is needed because empty acls are stored as zero byte
4511 * data in xattrs. Problem with that is, that receiving these zero byte
4512 * acls will fail later. To fix this, we send a dummy acl list that
4513 * only contains the version number and no entries.
4514 */
4515 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4516 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4517 if (data_len == 0) {
4518 dummy_acl.a_version =
4519 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4520 data = (char *)&dummy_acl;
4521 data_len = sizeof(dummy_acl);
4522 }
4523 }
4524
4525 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4526 if (ret < 0)
4527 goto out;
4528
4529 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4530
4531out:
4532 fs_path_free(p);
4533 return ret;
4534}
4535
4536static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4537 const char *name, int name_len,
4538 const char *data, int data_len,
4539 u8 type, void *ctx)
4540{
4541 int ret;
4542 struct send_ctx *sctx = ctx;
4543 struct fs_path *p;
4544
4545 p = fs_path_alloc();
4546 if (!p)
4547 return -ENOMEM;
4548
4549 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4550 if (ret < 0)
4551 goto out;
4552
4553 ret = send_remove_xattr(sctx, p, name, name_len);
4554
4555out:
4556 fs_path_free(p);
4557 return ret;
4558}
4559
4560static int process_new_xattr(struct send_ctx *sctx)
4561{
4562 int ret = 0;
4563
4564 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4565 __process_new_xattr, sctx);
4566
4567 return ret;
4568}
4569
4570static int process_deleted_xattr(struct send_ctx *sctx)
4571{
4572 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4573 __process_deleted_xattr, sctx);
4574}
4575
4576struct find_xattr_ctx {
4577 const char *name;
4578 int name_len;
4579 int found_idx;
4580 char *found_data;
4581 int found_data_len;
4582};
4583
4584static int __find_xattr(int num, struct btrfs_key *di_key,
4585 const char *name, int name_len,
4586 const char *data, int data_len,
4587 u8 type, void *vctx)
4588{
4589 struct find_xattr_ctx *ctx = vctx;
4590
4591 if (name_len == ctx->name_len &&
4592 strncmp(name, ctx->name, name_len) == 0) {
4593 ctx->found_idx = num;
4594 ctx->found_data_len = data_len;
4595 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4596 if (!ctx->found_data)
4597 return -ENOMEM;
4598 return 1;
4599 }
4600 return 0;
4601}
4602
4603static int find_xattr(struct btrfs_root *root,
4604 struct btrfs_path *path,
4605 struct btrfs_key *key,
4606 const char *name, int name_len,
4607 char **data, int *data_len)
4608{
4609 int ret;
4610 struct find_xattr_ctx ctx;
4611
4612 ctx.name = name;
4613 ctx.name_len = name_len;
4614 ctx.found_idx = -1;
4615 ctx.found_data = NULL;
4616 ctx.found_data_len = 0;
4617
4618 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4619 if (ret < 0)
4620 return ret;
4621
4622 if (ctx.found_idx == -1)
4623 return -ENOENT;
4624 if (data) {
4625 *data = ctx.found_data;
4626 *data_len = ctx.found_data_len;
4627 } else {
4628 kfree(ctx.found_data);
4629 }
4630 return ctx.found_idx;
4631}
4632
4633
4634static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4635 const char *name, int name_len,
4636 const char *data, int data_len,
4637 u8 type, void *ctx)
4638{
4639 int ret;
4640 struct send_ctx *sctx = ctx;
4641 char *found_data = NULL;
4642 int found_data_len = 0;
4643
4644 ret = find_xattr(sctx->parent_root, sctx->right_path,
4645 sctx->cmp_key, name, name_len, &found_data,
4646 &found_data_len);
4647 if (ret == -ENOENT) {
4648 ret = __process_new_xattr(num, di_key, name, name_len, data,
4649 data_len, type, ctx);
4650 } else if (ret >= 0) {
4651 if (data_len != found_data_len ||
4652 memcmp(data, found_data, data_len)) {
4653 ret = __process_new_xattr(num, di_key, name, name_len,
4654 data, data_len, type, ctx);
4655 } else {
4656 ret = 0;
4657 }
4658 }
4659
4660 kfree(found_data);
4661 return ret;
4662}
4663
4664static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4665 const char *name, int name_len,
4666 const char *data, int data_len,
4667 u8 type, void *ctx)
4668{
4669 int ret;
4670 struct send_ctx *sctx = ctx;
4671
4672 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4673 name, name_len, NULL, NULL);
4674 if (ret == -ENOENT)
4675 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4676 data_len, type, ctx);
4677 else if (ret >= 0)
4678 ret = 0;
4679
4680 return ret;
4681}
4682
4683static int process_changed_xattr(struct send_ctx *sctx)
4684{
4685 int ret = 0;
4686
4687 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4688 __process_changed_new_xattr, sctx);
4689 if (ret < 0)
4690 goto out;
4691 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4692 __process_changed_deleted_xattr, sctx);
4693
4694out:
4695 return ret;
4696}
4697
4698static int process_all_new_xattrs(struct send_ctx *sctx)
4699{
4700 int ret;
4701 struct btrfs_root *root;
4702 struct btrfs_path *path;
4703 struct btrfs_key key;
4704 struct btrfs_key found_key;
4705 struct extent_buffer *eb;
4706 int slot;
4707
4708 path = alloc_path_for_send();
4709 if (!path)
4710 return -ENOMEM;
4711
4712 root = sctx->send_root;
4713
4714 key.objectid = sctx->cmp_key->objectid;
4715 key.type = BTRFS_XATTR_ITEM_KEY;
4716 key.offset = 0;
4717 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4718 if (ret < 0)
4719 goto out;
4720
4721 while (1) {
4722 eb = path->nodes[0];
4723 slot = path->slots[0];
4724 if (slot >= btrfs_header_nritems(eb)) {
4725 ret = btrfs_next_leaf(root, path);
4726 if (ret < 0) {
4727 goto out;
4728 } else if (ret > 0) {
4729 ret = 0;
4730 break;
4731 }
4732 continue;
4733 }
4734
4735 btrfs_item_key_to_cpu(eb, &found_key, slot);
4736 if (found_key.objectid != key.objectid ||
4737 found_key.type != key.type) {
4738 ret = 0;
4739 goto out;
4740 }
4741
4742 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4743 if (ret < 0)
4744 goto out;
4745
4746 path->slots[0]++;
4747 }
4748
4749out:
4750 btrfs_free_path(path);
4751 return ret;
4752}
4753
4754static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4755{
4756 struct btrfs_root *root = sctx->send_root;
4757 struct btrfs_fs_info *fs_info = root->fs_info;
4758 struct inode *inode;
4759 struct page *page;
4760 char *addr;
4761 struct btrfs_key key;
4762 pgoff_t index = offset >> PAGE_SHIFT;
4763 pgoff_t last_index;
4764 unsigned pg_offset = offset & ~PAGE_MASK;
4765 ssize_t ret = 0;
4766
4767 key.objectid = sctx->cur_ino;
4768 key.type = BTRFS_INODE_ITEM_KEY;
4769 key.offset = 0;
4770
4771 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4772 if (IS_ERR(inode))
4773 return PTR_ERR(inode);
4774
4775 if (offset + len > i_size_read(inode)) {
4776 if (offset > i_size_read(inode))
4777 len = 0;
4778 else
4779 len = offset - i_size_read(inode);
4780 }
4781 if (len == 0)
4782 goto out;
4783
4784 last_index = (offset + len - 1) >> PAGE_SHIFT;
4785
4786 /* initial readahead */
4787 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4788 file_ra_state_init(&sctx->ra, inode->i_mapping);
4789
4790 while (index <= last_index) {
4791 unsigned cur_len = min_t(unsigned, len,
4792 PAGE_SIZE - pg_offset);
4793
4794 page = find_lock_page(inode->i_mapping, index);
4795 if (!page) {
4796 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4797 NULL, index, last_index + 1 - index);
4798
4799 page = find_or_create_page(inode->i_mapping, index,
4800 GFP_KERNEL);
4801 if (!page) {
4802 ret = -ENOMEM;
4803 break;
4804 }
4805 }
4806
4807 if (PageReadahead(page)) {
4808 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4809 NULL, page, index, last_index + 1 - index);
4810 }
4811
4812 if (!PageUptodate(page)) {
4813 btrfs_readpage(NULL, page);
4814 lock_page(page);
4815 if (!PageUptodate(page)) {
4816 unlock_page(page);
4817 put_page(page);
4818 ret = -EIO;
4819 break;
4820 }
4821 }
4822
4823 addr = kmap(page);
4824 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4825 kunmap(page);
4826 unlock_page(page);
4827 put_page(page);
4828 index++;
4829 pg_offset = 0;
4830 len -= cur_len;
4831 ret += cur_len;
4832 }
4833out:
4834 iput(inode);
4835 return ret;
4836}
4837
4838/*
4839 * Read some bytes from the current inode/file and send a write command to
4840 * user space.
4841 */
4842static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4843{
4844 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4845 int ret = 0;
4846 struct fs_path *p;
4847 ssize_t num_read = 0;
4848
4849 p = fs_path_alloc();
4850 if (!p)
4851 return -ENOMEM;
4852
4853 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4854
4855 num_read = fill_read_buf(sctx, offset, len);
4856 if (num_read <= 0) {
4857 if (num_read < 0)
4858 ret = num_read;
4859 goto out;
4860 }
4861
4862 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4863 if (ret < 0)
4864 goto out;
4865
4866 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4867 if (ret < 0)
4868 goto out;
4869
4870 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4871 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4872 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4873
4874 ret = send_cmd(sctx);
4875
4876tlv_put_failure:
4877out:
4878 fs_path_free(p);
4879 if (ret < 0)
4880 return ret;
4881 return num_read;
4882}
4883
4884/*
4885 * Send a clone command to user space.
4886 */
4887static int send_clone(struct send_ctx *sctx,
4888 u64 offset, u32 len,
4889 struct clone_root *clone_root)
4890{
4891 int ret = 0;
4892 struct fs_path *p;
4893 u64 gen;
4894
4895 btrfs_debug(sctx->send_root->fs_info,
4896 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4897 offset, len, clone_root->root->objectid, clone_root->ino,
4898 clone_root->offset);
4899
4900 p = fs_path_alloc();
4901 if (!p)
4902 return -ENOMEM;
4903
4904 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4905 if (ret < 0)
4906 goto out;
4907
4908 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4909 if (ret < 0)
4910 goto out;
4911
4912 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4913 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4914 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4915
4916 if (clone_root->root == sctx->send_root) {
4917 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4918 &gen, NULL, NULL, NULL, NULL);
4919 if (ret < 0)
4920 goto out;
4921 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4922 } else {
4923 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4924 }
4925 if (ret < 0)
4926 goto out;
4927
4928 /*
4929 * If the parent we're using has a received_uuid set then use that as
4930 * our clone source as that is what we will look for when doing a
4931 * receive.
4932 *
4933 * This covers the case that we create a snapshot off of a received
4934 * subvolume and then use that as the parent and try to receive on a
4935 * different host.
4936 */
4937 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4938 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4939 clone_root->root->root_item.received_uuid);
4940 else
4941 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4942 clone_root->root->root_item.uuid);
4943 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4944 le64_to_cpu(clone_root->root->root_item.ctransid));
4945 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4946 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4947 clone_root->offset);
4948
4949 ret = send_cmd(sctx);
4950
4951tlv_put_failure:
4952out:
4953 fs_path_free(p);
4954 return ret;
4955}
4956
4957/*
4958 * Send an update extent command to user space.
4959 */
4960static int send_update_extent(struct send_ctx *sctx,
4961 u64 offset, u32 len)
4962{
4963 int ret = 0;
4964 struct fs_path *p;
4965
4966 p = fs_path_alloc();
4967 if (!p)
4968 return -ENOMEM;
4969
4970 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4971 if (ret < 0)
4972 goto out;
4973
4974 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4975 if (ret < 0)
4976 goto out;
4977
4978 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4979 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4980 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4981
4982 ret = send_cmd(sctx);
4983
4984tlv_put_failure:
4985out:
4986 fs_path_free(p);
4987 return ret;
4988}
4989
4990static int send_hole(struct send_ctx *sctx, u64 end)
4991{
4992 struct fs_path *p = NULL;
4993 u64 offset = sctx->cur_inode_last_extent;
4994 u64 len;
4995 int ret = 0;
4996
4997 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4998 return send_update_extent(sctx, offset, end - offset);
4999
5000 p = fs_path_alloc();
5001 if (!p)
5002 return -ENOMEM;
5003 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5004 if (ret < 0)
5005 goto tlv_put_failure;
5006 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5007 while (offset < end) {
5008 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5009
5010 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5011 if (ret < 0)
5012 break;
5013 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5014 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5015 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5016 ret = send_cmd(sctx);
5017 if (ret < 0)
5018 break;
5019 offset += len;
5020 }
5021 sctx->cur_inode_next_write_offset = offset;
5022tlv_put_failure:
5023 fs_path_free(p);
5024 return ret;
5025}
5026
5027static int send_extent_data(struct send_ctx *sctx,
5028 const u64 offset,
5029 const u64 len)
5030{
5031 u64 sent = 0;
5032
5033 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5034 return send_update_extent(sctx, offset, len);
5035
5036 while (sent < len) {
5037 u64 size = len - sent;
5038 int ret;
5039
5040 if (size > BTRFS_SEND_READ_SIZE)
5041 size = BTRFS_SEND_READ_SIZE;
5042 ret = send_write(sctx, offset + sent, size);
5043 if (ret < 0)
5044 return ret;
5045 if (!ret)
5046 break;
5047 sent += ret;
5048 }
5049 return 0;
5050}
5051
5052static int clone_range(struct send_ctx *sctx,
5053 struct clone_root *clone_root,
5054 const u64 disk_byte,
5055 u64 data_offset,
5056 u64 offset,
5057 u64 len)
5058{
5059 struct btrfs_path *path;
5060 struct btrfs_key key;
5061 int ret;
5062
5063 /*
5064 * Prevent cloning from a zero offset with a length matching the sector
5065 * size because in some scenarios this will make the receiver fail.
5066 *
5067 * For example, if in the source filesystem the extent at offset 0
5068 * has a length of sectorsize and it was written using direct IO, then
5069 * it can never be an inline extent (even if compression is enabled).
5070 * Then this extent can be cloned in the original filesystem to a non
5071 * zero file offset, but it may not be possible to clone in the
5072 * destination filesystem because it can be inlined due to compression
5073 * on the destination filesystem (as the receiver's write operations are
5074 * always done using buffered IO). The same happens when the original
5075 * filesystem does not have compression enabled but the destination
5076 * filesystem has.
5077 */
5078 if (clone_root->offset == 0 &&
5079 len == sctx->send_root->fs_info->sectorsize)
5080 return send_extent_data(sctx, offset, len);
5081
5082 path = alloc_path_for_send();
5083 if (!path)
5084 return -ENOMEM;
5085
5086 /*
5087 * We can't send a clone operation for the entire range if we find
5088 * extent items in the respective range in the source file that
5089 * refer to different extents or if we find holes.
5090 * So check for that and do a mix of clone and regular write/copy
5091 * operations if needed.
5092 *
5093 * Example:
5094 *
5095 * mkfs.btrfs -f /dev/sda
5096 * mount /dev/sda /mnt
5097 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5098 * cp --reflink=always /mnt/foo /mnt/bar
5099 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5100 * btrfs subvolume snapshot -r /mnt /mnt/snap
5101 *
5102 * If when we send the snapshot and we are processing file bar (which
5103 * has a higher inode number than foo) we blindly send a clone operation
5104 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5105 * a file bar that matches the content of file foo - iow, doesn't match
5106 * the content from bar in the original filesystem.
5107 */
5108 key.objectid = clone_root->ino;
5109 key.type = BTRFS_EXTENT_DATA_KEY;
5110 key.offset = clone_root->offset;
5111 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5112 if (ret < 0)
5113 goto out;
5114 if (ret > 0 && path->slots[0] > 0) {
5115 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5116 if (key.objectid == clone_root->ino &&
5117 key.type == BTRFS_EXTENT_DATA_KEY)
5118 path->slots[0]--;
5119 }
5120
5121 while (true) {
5122 struct extent_buffer *leaf = path->nodes[0];
5123 int slot = path->slots[0];
5124 struct btrfs_file_extent_item *ei;
5125 u8 type;
5126 u64 ext_len;
5127 u64 clone_len;
5128
5129 if (slot >= btrfs_header_nritems(leaf)) {
5130 ret = btrfs_next_leaf(clone_root->root, path);
5131 if (ret < 0)
5132 goto out;
5133 else if (ret > 0)
5134 break;
5135 continue;
5136 }
5137
5138 btrfs_item_key_to_cpu(leaf, &key, slot);
5139
5140 /*
5141 * We might have an implicit trailing hole (NO_HOLES feature
5142 * enabled). We deal with it after leaving this loop.
5143 */
5144 if (key.objectid != clone_root->ino ||
5145 key.type != BTRFS_EXTENT_DATA_KEY)
5146 break;
5147
5148 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5149 type = btrfs_file_extent_type(leaf, ei);
5150 if (type == BTRFS_FILE_EXTENT_INLINE) {
5151 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
5152 ext_len = PAGE_ALIGN(ext_len);
5153 } else {
5154 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5155 }
5156
5157 if (key.offset + ext_len <= clone_root->offset)
5158 goto next;
5159
5160 if (key.offset > clone_root->offset) {
5161 /* Implicit hole, NO_HOLES feature enabled. */
5162 u64 hole_len = key.offset - clone_root->offset;
5163
5164 if (hole_len > len)
5165 hole_len = len;
5166 ret = send_extent_data(sctx, offset, hole_len);
5167 if (ret < 0)
5168 goto out;
5169
5170 len -= hole_len;
5171 if (len == 0)
5172 break;
5173 offset += hole_len;
5174 clone_root->offset += hole_len;
5175 data_offset += hole_len;
5176 }
5177
5178 if (key.offset >= clone_root->offset + len)
5179 break;
5180
5181 clone_len = min_t(u64, ext_len, len);
5182
5183 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5184 btrfs_file_extent_offset(leaf, ei) == data_offset)
5185 ret = send_clone(sctx, offset, clone_len, clone_root);
5186 else
5187 ret = send_extent_data(sctx, offset, clone_len);
5188
5189 if (ret < 0)
5190 goto out;
5191
5192 len -= clone_len;
5193 if (len == 0)
5194 break;
5195 offset += clone_len;
5196 clone_root->offset += clone_len;
5197 data_offset += clone_len;
5198next:
5199 path->slots[0]++;
5200 }
5201
5202 if (len > 0)
5203 ret = send_extent_data(sctx, offset, len);
5204 else
5205 ret = 0;
5206out:
5207 btrfs_free_path(path);
5208 return ret;
5209}
5210
5211static int send_write_or_clone(struct send_ctx *sctx,
5212 struct btrfs_path *path,
5213 struct btrfs_key *key,
5214 struct clone_root *clone_root)
5215{
5216 int ret = 0;
5217 struct btrfs_file_extent_item *ei;
5218 u64 offset = key->offset;
5219 u64 len;
5220 u8 type;
5221 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5222
5223 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5224 struct btrfs_file_extent_item);
5225 type = btrfs_file_extent_type(path->nodes[0], ei);
5226 if (type == BTRFS_FILE_EXTENT_INLINE) {
5227 len = btrfs_file_extent_inline_len(path->nodes[0],
5228 path->slots[0], ei);
5229 /*
5230 * it is possible the inline item won't cover the whole page,
5231 * but there may be items after this page. Make
5232 * sure to send the whole thing
5233 */
5234 len = PAGE_ALIGN(len);
5235 } else {
5236 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5237 }
5238
5239 if (offset >= sctx->cur_inode_size) {
5240 ret = 0;
5241 goto out;
5242 }
5243 if (offset + len > sctx->cur_inode_size)
5244 len = sctx->cur_inode_size - offset;
5245 if (len == 0) {
5246 ret = 0;
5247 goto out;
5248 }
5249
5250 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5251 u64 disk_byte;
5252 u64 data_offset;
5253
5254 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5255 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5256 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5257 offset, len);
5258 } else {
5259 ret = send_extent_data(sctx, offset, len);
5260 }
5261 sctx->cur_inode_next_write_offset = offset + len;
5262out:
5263 return ret;
5264}
5265
5266static int is_extent_unchanged(struct send_ctx *sctx,
5267 struct btrfs_path *left_path,
5268 struct btrfs_key *ekey)
5269{
5270 int ret = 0;
5271 struct btrfs_key key;
5272 struct btrfs_path *path = NULL;
5273 struct extent_buffer *eb;
5274 int slot;
5275 struct btrfs_key found_key;
5276 struct btrfs_file_extent_item *ei;
5277 u64 left_disknr;
5278 u64 right_disknr;
5279 u64 left_offset;
5280 u64 right_offset;
5281 u64 left_offset_fixed;
5282 u64 left_len;
5283 u64 right_len;
5284 u64 left_gen;
5285 u64 right_gen;
5286 u8 left_type;
5287 u8 right_type;
5288
5289 path = alloc_path_for_send();
5290 if (!path)
5291 return -ENOMEM;
5292
5293 eb = left_path->nodes[0];
5294 slot = left_path->slots[0];
5295 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5296 left_type = btrfs_file_extent_type(eb, ei);
5297
5298 if (left_type != BTRFS_FILE_EXTENT_REG) {
5299 ret = 0;
5300 goto out;
5301 }
5302 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5303 left_len = btrfs_file_extent_num_bytes(eb, ei);
5304 left_offset = btrfs_file_extent_offset(eb, ei);
5305 left_gen = btrfs_file_extent_generation(eb, ei);
5306
5307 /*
5308 * Following comments will refer to these graphics. L is the left
5309 * extents which we are checking at the moment. 1-8 are the right
5310 * extents that we iterate.
5311 *
5312 * |-----L-----|
5313 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5314 *
5315 * |-----L-----|
5316 * |--1--|-2b-|...(same as above)
5317 *
5318 * Alternative situation. Happens on files where extents got split.
5319 * |-----L-----|
5320 * |-----------7-----------|-6-|
5321 *
5322 * Alternative situation. Happens on files which got larger.
5323 * |-----L-----|
5324 * |-8-|
5325 * Nothing follows after 8.
5326 */
5327
5328 key.objectid = ekey->objectid;
5329 key.type = BTRFS_EXTENT_DATA_KEY;
5330 key.offset = ekey->offset;
5331 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5332 if (ret < 0)
5333 goto out;
5334 if (ret) {
5335 ret = 0;
5336 goto out;
5337 }
5338
5339 /*
5340 * Handle special case where the right side has no extents at all.
5341 */
5342 eb = path->nodes[0];
5343 slot = path->slots[0];
5344 btrfs_item_key_to_cpu(eb, &found_key, slot);
5345 if (found_key.objectid != key.objectid ||
5346 found_key.type != key.type) {
5347 /* If we're a hole then just pretend nothing changed */
5348 ret = (left_disknr) ? 0 : 1;
5349 goto out;
5350 }
5351
5352 /*
5353 * We're now on 2a, 2b or 7.
5354 */
5355 key = found_key;
5356 while (key.offset < ekey->offset + left_len) {
5357 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5358 right_type = btrfs_file_extent_type(eb, ei);
5359 if (right_type != BTRFS_FILE_EXTENT_REG &&
5360 right_type != BTRFS_FILE_EXTENT_INLINE) {
5361 ret = 0;
5362 goto out;
5363 }
5364
5365 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5366 right_len = btrfs_file_extent_inline_len(eb, slot, ei);
5367 right_len = PAGE_ALIGN(right_len);
5368 } else {
5369 right_len = btrfs_file_extent_num_bytes(eb, ei);
5370 }
5371
5372 /*
5373 * Are we at extent 8? If yes, we know the extent is changed.
5374 * This may only happen on the first iteration.
5375 */
5376 if (found_key.offset + right_len <= ekey->offset) {
5377 /* If we're a hole just pretend nothing changed */
5378 ret = (left_disknr) ? 0 : 1;
5379 goto out;
5380 }
5381
5382 /*
5383 * We just wanted to see if when we have an inline extent, what
5384 * follows it is a regular extent (wanted to check the above
5385 * condition for inline extents too). This should normally not
5386 * happen but it's possible for example when we have an inline
5387 * compressed extent representing data with a size matching
5388 * the page size (currently the same as sector size).
5389 */
5390 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5391 ret = 0;
5392 goto out;
5393 }
5394
5395 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5396 right_offset = btrfs_file_extent_offset(eb, ei);
5397 right_gen = btrfs_file_extent_generation(eb, ei);
5398
5399 left_offset_fixed = left_offset;
5400 if (key.offset < ekey->offset) {
5401 /* Fix the right offset for 2a and 7. */
5402 right_offset += ekey->offset - key.offset;
5403 } else {
5404 /* Fix the left offset for all behind 2a and 2b */
5405 left_offset_fixed += key.offset - ekey->offset;
5406 }
5407
5408 /*
5409 * Check if we have the same extent.
5410 */
5411 if (left_disknr != right_disknr ||
5412 left_offset_fixed != right_offset ||
5413 left_gen != right_gen) {
5414 ret = 0;
5415 goto out;
5416 }
5417
5418 /*
5419 * Go to the next extent.
5420 */
5421 ret = btrfs_next_item(sctx->parent_root, path);
5422 if (ret < 0)
5423 goto out;
5424 if (!ret) {
5425 eb = path->nodes[0];
5426 slot = path->slots[0];
5427 btrfs_item_key_to_cpu(eb, &found_key, slot);
5428 }
5429 if (ret || found_key.objectid != key.objectid ||
5430 found_key.type != key.type) {
5431 key.offset += right_len;
5432 break;
5433 }
5434 if (found_key.offset != key.offset + right_len) {
5435 ret = 0;
5436 goto out;
5437 }
5438 key = found_key;
5439 }
5440
5441 /*
5442 * We're now behind the left extent (treat as unchanged) or at the end
5443 * of the right side (treat as changed).
5444 */
5445 if (key.offset >= ekey->offset + left_len)
5446 ret = 1;
5447 else
5448 ret = 0;
5449
5450
5451out:
5452 btrfs_free_path(path);
5453 return ret;
5454}
5455
5456static int get_last_extent(struct send_ctx *sctx, u64 offset)
5457{
5458 struct btrfs_path *path;
5459 struct btrfs_root *root = sctx->send_root;
5460 struct btrfs_file_extent_item *fi;
5461 struct btrfs_key key;
5462 u64 extent_end;
5463 u8 type;
5464 int ret;
5465
5466 path = alloc_path_for_send();
5467 if (!path)
5468 return -ENOMEM;
5469
5470 sctx->cur_inode_last_extent = 0;
5471
5472 key.objectid = sctx->cur_ino;
5473 key.type = BTRFS_EXTENT_DATA_KEY;
5474 key.offset = offset;
5475 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5476 if (ret < 0)
5477 goto out;
5478 ret = 0;
5479 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5480 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5481 goto out;
5482
5483 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5484 struct btrfs_file_extent_item);
5485 type = btrfs_file_extent_type(path->nodes[0], fi);
5486 if (type == BTRFS_FILE_EXTENT_INLINE) {
5487 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5488 path->slots[0], fi);
5489 extent_end = ALIGN(key.offset + size,
5490 sctx->send_root->fs_info->sectorsize);
5491 } else {
5492 extent_end = key.offset +
5493 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5494 }
5495 sctx->cur_inode_last_extent = extent_end;
5496out:
5497 btrfs_free_path(path);
5498 return ret;
5499}
5500
5501static int range_is_hole_in_parent(struct send_ctx *sctx,
5502 const u64 start,
5503 const u64 end)
5504{
5505 struct btrfs_path *path;
5506 struct btrfs_key key;
5507 struct btrfs_root *root = sctx->parent_root;
5508 u64 search_start = start;
5509 int ret;
5510
5511 path = alloc_path_for_send();
5512 if (!path)
5513 return -ENOMEM;
5514
5515 key.objectid = sctx->cur_ino;
5516 key.type = BTRFS_EXTENT_DATA_KEY;
5517 key.offset = search_start;
5518 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5519 if (ret < 0)
5520 goto out;
5521 if (ret > 0 && path->slots[0] > 0)
5522 path->slots[0]--;
5523
5524 while (search_start < end) {
5525 struct extent_buffer *leaf = path->nodes[0];
5526 int slot = path->slots[0];
5527 struct btrfs_file_extent_item *fi;
5528 u64 extent_end;
5529
5530 if (slot >= btrfs_header_nritems(leaf)) {
5531 ret = btrfs_next_leaf(root, path);
5532 if (ret < 0)
5533 goto out;
5534 else if (ret > 0)
5535 break;
5536 continue;
5537 }
5538
5539 btrfs_item_key_to_cpu(leaf, &key, slot);
5540 if (key.objectid < sctx->cur_ino ||
5541 key.type < BTRFS_EXTENT_DATA_KEY)
5542 goto next;
5543 if (key.objectid > sctx->cur_ino ||
5544 key.type > BTRFS_EXTENT_DATA_KEY ||
5545 key.offset >= end)
5546 break;
5547
5548 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5549 if (btrfs_file_extent_type(leaf, fi) ==
5550 BTRFS_FILE_EXTENT_INLINE) {
5551 u64 size = btrfs_file_extent_inline_len(leaf, slot, fi);
5552
5553 extent_end = ALIGN(key.offset + size,
5554 root->fs_info->sectorsize);
5555 } else {
5556 extent_end = key.offset +
5557 btrfs_file_extent_num_bytes(leaf, fi);
5558 }
5559 if (extent_end <= start)
5560 goto next;
5561 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5562 search_start = extent_end;
5563 goto next;
5564 }
5565 ret = 0;
5566 goto out;
5567next:
5568 path->slots[0]++;
5569 }
5570 ret = 1;
5571out:
5572 btrfs_free_path(path);
5573 return ret;
5574}
5575
5576static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5577 struct btrfs_key *key)
5578{
5579 struct btrfs_file_extent_item *fi;
5580 u64 extent_end;
5581 u8 type;
5582 int ret = 0;
5583
5584 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5585 return 0;
5586
5587 if (sctx->cur_inode_last_extent == (u64)-1) {
5588 ret = get_last_extent(sctx, key->offset - 1);
5589 if (ret)
5590 return ret;
5591 }
5592
5593 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5594 struct btrfs_file_extent_item);
5595 type = btrfs_file_extent_type(path->nodes[0], fi);
5596 if (type == BTRFS_FILE_EXTENT_INLINE) {
5597 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5598 path->slots[0], fi);
5599 extent_end = ALIGN(key->offset + size,
5600 sctx->send_root->fs_info->sectorsize);
5601 } else {
5602 extent_end = key->offset +
5603 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5604 }
5605
5606 if (path->slots[0] == 0 &&
5607 sctx->cur_inode_last_extent < key->offset) {
5608 /*
5609 * We might have skipped entire leafs that contained only
5610 * file extent items for our current inode. These leafs have
5611 * a generation number smaller (older) than the one in the
5612 * current leaf and the leaf our last extent came from, and
5613 * are located between these 2 leafs.
5614 */
5615 ret = get_last_extent(sctx, key->offset - 1);
5616 if (ret)
5617 return ret;
5618 }
5619
5620 if (sctx->cur_inode_last_extent < key->offset) {
5621 ret = range_is_hole_in_parent(sctx,
5622 sctx->cur_inode_last_extent,
5623 key->offset);
5624 if (ret < 0)
5625 return ret;
5626 else if (ret == 0)
5627 ret = send_hole(sctx, key->offset);
5628 else
5629 ret = 0;
5630 }
5631 sctx->cur_inode_last_extent = extent_end;
5632 return ret;
5633}
5634
5635static int process_extent(struct send_ctx *sctx,
5636 struct btrfs_path *path,
5637 struct btrfs_key *key)
5638{
5639 struct clone_root *found_clone = NULL;
5640 int ret = 0;
5641
5642 if (S_ISLNK(sctx->cur_inode_mode))
5643 return 0;
5644
5645 if (sctx->parent_root && !sctx->cur_inode_new) {
5646 ret = is_extent_unchanged(sctx, path, key);
5647 if (ret < 0)
5648 goto out;
5649 if (ret) {
5650 ret = 0;
5651 goto out_hole;
5652 }
5653 } else {
5654 struct btrfs_file_extent_item *ei;
5655 u8 type;
5656
5657 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5658 struct btrfs_file_extent_item);
5659 type = btrfs_file_extent_type(path->nodes[0], ei);
5660 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5661 type == BTRFS_FILE_EXTENT_REG) {
5662 /*
5663 * The send spec does not have a prealloc command yet,
5664 * so just leave a hole for prealloc'ed extents until
5665 * we have enough commands queued up to justify rev'ing
5666 * the send spec.
5667 */
5668 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5669 ret = 0;
5670 goto out;
5671 }
5672
5673 /* Have a hole, just skip it. */
5674 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5675 ret = 0;
5676 goto out;
5677 }
5678 }
5679 }
5680
5681 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5682 sctx->cur_inode_size, &found_clone);
5683 if (ret != -ENOENT && ret < 0)
5684 goto out;
5685
5686 ret = send_write_or_clone(sctx, path, key, found_clone);
5687 if (ret)
5688 goto out;
5689out_hole:
5690 ret = maybe_send_hole(sctx, path, key);
5691out:
5692 return ret;
5693}
5694
5695static int process_all_extents(struct send_ctx *sctx)
5696{
5697 int ret;
5698 struct btrfs_root *root;
5699 struct btrfs_path *path;
5700 struct btrfs_key key;
5701 struct btrfs_key found_key;
5702 struct extent_buffer *eb;
5703 int slot;
5704
5705 root = sctx->send_root;
5706 path = alloc_path_for_send();
5707 if (!path)
5708 return -ENOMEM;
5709
5710 key.objectid = sctx->cmp_key->objectid;
5711 key.type = BTRFS_EXTENT_DATA_KEY;
5712 key.offset = 0;
5713 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5714 if (ret < 0)
5715 goto out;
5716
5717 while (1) {
5718 eb = path->nodes[0];
5719 slot = path->slots[0];
5720
5721 if (slot >= btrfs_header_nritems(eb)) {
5722 ret = btrfs_next_leaf(root, path);
5723 if (ret < 0) {
5724 goto out;
5725 } else if (ret > 0) {
5726 ret = 0;
5727 break;
5728 }
5729 continue;
5730 }
5731
5732 btrfs_item_key_to_cpu(eb, &found_key, slot);
5733
5734 if (found_key.objectid != key.objectid ||
5735 found_key.type != key.type) {
5736 ret = 0;
5737 goto out;
5738 }
5739
5740 ret = process_extent(sctx, path, &found_key);
5741 if (ret < 0)
5742 goto out;
5743
5744 path->slots[0]++;
5745 }
5746
5747out:
5748 btrfs_free_path(path);
5749 return ret;
5750}
5751
5752static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5753 int *pending_move,
5754 int *refs_processed)
5755{
5756 int ret = 0;
5757
5758 if (sctx->cur_ino == 0)
5759 goto out;
5760 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5761 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5762 goto out;
5763 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5764 goto out;
5765
5766 ret = process_recorded_refs(sctx, pending_move);
5767 if (ret < 0)
5768 goto out;
5769
5770 *refs_processed = 1;
5771out:
5772 return ret;
5773}
5774
5775static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5776{
5777 int ret = 0;
5778 u64 left_mode;
5779 u64 left_uid;
5780 u64 left_gid;
5781 u64 right_mode;
5782 u64 right_uid;
5783 u64 right_gid;
5784 int need_chmod = 0;
5785 int need_chown = 0;
5786 int need_truncate = 1;
5787 int pending_move = 0;
5788 int refs_processed = 0;
5789
5790 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5791 &refs_processed);
5792 if (ret < 0)
5793 goto out;
5794
5795 /*
5796 * We have processed the refs and thus need to advance send_progress.
5797 * Now, calls to get_cur_xxx will take the updated refs of the current
5798 * inode into account.
5799 *
5800 * On the other hand, if our current inode is a directory and couldn't
5801 * be moved/renamed because its parent was renamed/moved too and it has
5802 * a higher inode number, we can only move/rename our current inode
5803 * after we moved/renamed its parent. Therefore in this case operate on
5804 * the old path (pre move/rename) of our current inode, and the
5805 * move/rename will be performed later.
5806 */
5807 if (refs_processed && !pending_move)
5808 sctx->send_progress = sctx->cur_ino + 1;
5809
5810 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5811 goto out;
5812 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5813 goto out;
5814
5815 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5816 &left_mode, &left_uid, &left_gid, NULL);
5817 if (ret < 0)
5818 goto out;
5819
5820 if (!sctx->parent_root || sctx->cur_inode_new) {
5821 need_chown = 1;
5822 if (!S_ISLNK(sctx->cur_inode_mode))
5823 need_chmod = 1;
5824 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5825 need_truncate = 0;
5826 } else {
5827 u64 old_size;
5828
5829 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5830 &old_size, NULL, &right_mode, &right_uid,
5831 &right_gid, NULL);
5832 if (ret < 0)
5833 goto out;
5834
5835 if (left_uid != right_uid || left_gid != right_gid)
5836 need_chown = 1;
5837 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5838 need_chmod = 1;
5839 if ((old_size == sctx->cur_inode_size) ||
5840 (sctx->cur_inode_size > old_size &&
5841 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5842 need_truncate = 0;
5843 }
5844
5845 if (S_ISREG(sctx->cur_inode_mode)) {
5846 if (need_send_hole(sctx)) {
5847 if (sctx->cur_inode_last_extent == (u64)-1 ||
5848 sctx->cur_inode_last_extent <
5849 sctx->cur_inode_size) {
5850 ret = get_last_extent(sctx, (u64)-1);
5851 if (ret)
5852 goto out;
5853 }
5854 if (sctx->cur_inode_last_extent <
5855 sctx->cur_inode_size) {
5856 ret = send_hole(sctx, sctx->cur_inode_size);
5857 if (ret)
5858 goto out;
5859 }
5860 }
5861 if (need_truncate) {
5862 ret = send_truncate(sctx, sctx->cur_ino,
5863 sctx->cur_inode_gen,
5864 sctx->cur_inode_size);
5865 if (ret < 0)
5866 goto out;
5867 }
5868 }
5869
5870 if (need_chown) {
5871 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5872 left_uid, left_gid);
5873 if (ret < 0)
5874 goto out;
5875 }
5876 if (need_chmod) {
5877 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5878 left_mode);
5879 if (ret < 0)
5880 goto out;
5881 }
5882
5883 /*
5884 * If other directory inodes depended on our current directory
5885 * inode's move/rename, now do their move/rename operations.
5886 */
5887 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5888 ret = apply_children_dir_moves(sctx);
5889 if (ret)
5890 goto out;
5891 /*
5892 * Need to send that every time, no matter if it actually
5893 * changed between the two trees as we have done changes to
5894 * the inode before. If our inode is a directory and it's
5895 * waiting to be moved/renamed, we will send its utimes when
5896 * it's moved/renamed, therefore we don't need to do it here.
5897 */
5898 sctx->send_progress = sctx->cur_ino + 1;
5899 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5900 if (ret < 0)
5901 goto out;
5902 }
5903
5904out:
5905 return ret;
5906}
5907
5908static int changed_inode(struct send_ctx *sctx,
5909 enum btrfs_compare_tree_result result)
5910{
5911 int ret = 0;
5912 struct btrfs_key *key = sctx->cmp_key;
5913 struct btrfs_inode_item *left_ii = NULL;
5914 struct btrfs_inode_item *right_ii = NULL;
5915 u64 left_gen = 0;
5916 u64 right_gen = 0;
5917
5918 sctx->cur_ino = key->objectid;
5919 sctx->cur_inode_new_gen = 0;
5920 sctx->cur_inode_last_extent = (u64)-1;
5921 sctx->cur_inode_next_write_offset = 0;
5922
5923 /*
5924 * Set send_progress to current inode. This will tell all get_cur_xxx
5925 * functions that the current inode's refs are not updated yet. Later,
5926 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5927 */
5928 sctx->send_progress = sctx->cur_ino;
5929
5930 if (result == BTRFS_COMPARE_TREE_NEW ||
5931 result == BTRFS_COMPARE_TREE_CHANGED) {
5932 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5933 sctx->left_path->slots[0],
5934 struct btrfs_inode_item);
5935 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5936 left_ii);
5937 } else {
5938 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5939 sctx->right_path->slots[0],
5940 struct btrfs_inode_item);
5941 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5942 right_ii);
5943 }
5944 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5945 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5946 sctx->right_path->slots[0],
5947 struct btrfs_inode_item);
5948
5949 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5950 right_ii);
5951
5952 /*
5953 * The cur_ino = root dir case is special here. We can't treat
5954 * the inode as deleted+reused because it would generate a
5955 * stream that tries to delete/mkdir the root dir.
5956 */
5957 if (left_gen != right_gen &&
5958 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5959 sctx->cur_inode_new_gen = 1;
5960 }
5961
5962 if (result == BTRFS_COMPARE_TREE_NEW) {
5963 sctx->cur_inode_gen = left_gen;
5964 sctx->cur_inode_new = 1;
5965 sctx->cur_inode_deleted = 0;
5966 sctx->cur_inode_size = btrfs_inode_size(
5967 sctx->left_path->nodes[0], left_ii);
5968 sctx->cur_inode_mode = btrfs_inode_mode(
5969 sctx->left_path->nodes[0], left_ii);
5970 sctx->cur_inode_rdev = btrfs_inode_rdev(
5971 sctx->left_path->nodes[0], left_ii);
5972 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5973 ret = send_create_inode_if_needed(sctx);
5974 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5975 sctx->cur_inode_gen = right_gen;
5976 sctx->cur_inode_new = 0;
5977 sctx->cur_inode_deleted = 1;
5978 sctx->cur_inode_size = btrfs_inode_size(
5979 sctx->right_path->nodes[0], right_ii);
5980 sctx->cur_inode_mode = btrfs_inode_mode(
5981 sctx->right_path->nodes[0], right_ii);
5982 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5983 /*
5984 * We need to do some special handling in case the inode was
5985 * reported as changed with a changed generation number. This
5986 * means that the original inode was deleted and new inode
5987 * reused the same inum. So we have to treat the old inode as
5988 * deleted and the new one as new.
5989 */
5990 if (sctx->cur_inode_new_gen) {
5991 /*
5992 * First, process the inode as if it was deleted.
5993 */
5994 sctx->cur_inode_gen = right_gen;
5995 sctx->cur_inode_new = 0;
5996 sctx->cur_inode_deleted = 1;
5997 sctx->cur_inode_size = btrfs_inode_size(
5998 sctx->right_path->nodes[0], right_ii);
5999 sctx->cur_inode_mode = btrfs_inode_mode(
6000 sctx->right_path->nodes[0], right_ii);
6001 ret = process_all_refs(sctx,
6002 BTRFS_COMPARE_TREE_DELETED);
6003 if (ret < 0)
6004 goto out;
6005
6006 /*
6007 * Now process the inode as if it was new.
6008 */
6009 sctx->cur_inode_gen = left_gen;
6010 sctx->cur_inode_new = 1;
6011 sctx->cur_inode_deleted = 0;
6012 sctx->cur_inode_size = btrfs_inode_size(
6013 sctx->left_path->nodes[0], left_ii);
6014 sctx->cur_inode_mode = btrfs_inode_mode(
6015 sctx->left_path->nodes[0], left_ii);
6016 sctx->cur_inode_rdev = btrfs_inode_rdev(
6017 sctx->left_path->nodes[0], left_ii);
6018 ret = send_create_inode_if_needed(sctx);
6019 if (ret < 0)
6020 goto out;
6021
6022 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6023 if (ret < 0)
6024 goto out;
6025 /*
6026 * Advance send_progress now as we did not get into
6027 * process_recorded_refs_if_needed in the new_gen case.
6028 */
6029 sctx->send_progress = sctx->cur_ino + 1;
6030
6031 /*
6032 * Now process all extents and xattrs of the inode as if
6033 * they were all new.
6034 */
6035 ret = process_all_extents(sctx);
6036 if (ret < 0)
6037 goto out;
6038 ret = process_all_new_xattrs(sctx);
6039 if (ret < 0)
6040 goto out;
6041 } else {
6042 sctx->cur_inode_gen = left_gen;
6043 sctx->cur_inode_new = 0;
6044 sctx->cur_inode_new_gen = 0;
6045 sctx->cur_inode_deleted = 0;
6046 sctx->cur_inode_size = btrfs_inode_size(
6047 sctx->left_path->nodes[0], left_ii);
6048 sctx->cur_inode_mode = btrfs_inode_mode(
6049 sctx->left_path->nodes[0], left_ii);
6050 }
6051 }
6052
6053out:
6054 return ret;
6055}
6056
6057/*
6058 * We have to process new refs before deleted refs, but compare_trees gives us
6059 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6060 * first and later process them in process_recorded_refs.
6061 * For the cur_inode_new_gen case, we skip recording completely because
6062 * changed_inode did already initiate processing of refs. The reason for this is
6063 * that in this case, compare_tree actually compares the refs of 2 different
6064 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6065 * refs of the right tree as deleted and all refs of the left tree as new.
6066 */
6067static int changed_ref(struct send_ctx *sctx,
6068 enum btrfs_compare_tree_result result)
6069{
6070 int ret = 0;
6071
6072 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6073 inconsistent_snapshot_error(sctx, result, "reference");
6074 return -EIO;
6075 }
6076
6077 if (!sctx->cur_inode_new_gen &&
6078 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6079 if (result == BTRFS_COMPARE_TREE_NEW)
6080 ret = record_new_ref(sctx);
6081 else if (result == BTRFS_COMPARE_TREE_DELETED)
6082 ret = record_deleted_ref(sctx);
6083 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6084 ret = record_changed_ref(sctx);
6085 }
6086
6087 return ret;
6088}
6089
6090/*
6091 * Process new/deleted/changed xattrs. We skip processing in the
6092 * cur_inode_new_gen case because changed_inode did already initiate processing
6093 * of xattrs. The reason is the same as in changed_ref
6094 */
6095static int changed_xattr(struct send_ctx *sctx,
6096 enum btrfs_compare_tree_result result)
6097{
6098 int ret = 0;
6099
6100 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6101 inconsistent_snapshot_error(sctx, result, "xattr");
6102 return -EIO;
6103 }
6104
6105 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6106 if (result == BTRFS_COMPARE_TREE_NEW)
6107 ret = process_new_xattr(sctx);
6108 else if (result == BTRFS_COMPARE_TREE_DELETED)
6109 ret = process_deleted_xattr(sctx);
6110 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6111 ret = process_changed_xattr(sctx);
6112 }
6113
6114 return ret;
6115}
6116
6117/*
6118 * Process new/deleted/changed extents. We skip processing in the
6119 * cur_inode_new_gen case because changed_inode did already initiate processing
6120 * of extents. The reason is the same as in changed_ref
6121 */
6122static int changed_extent(struct send_ctx *sctx,
6123 enum btrfs_compare_tree_result result)
6124{
6125 int ret = 0;
6126
6127 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6128
6129 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6130 struct extent_buffer *leaf_l;
6131 struct extent_buffer *leaf_r;
6132 struct btrfs_file_extent_item *ei_l;
6133 struct btrfs_file_extent_item *ei_r;
6134
6135 leaf_l = sctx->left_path->nodes[0];
6136 leaf_r = sctx->right_path->nodes[0];
6137 ei_l = btrfs_item_ptr(leaf_l,
6138 sctx->left_path->slots[0],
6139 struct btrfs_file_extent_item);
6140 ei_r = btrfs_item_ptr(leaf_r,
6141 sctx->right_path->slots[0],
6142 struct btrfs_file_extent_item);
6143
6144 /*
6145 * We may have found an extent item that has changed
6146 * only its disk_bytenr field and the corresponding
6147 * inode item was not updated. This case happens due to
6148 * very specific timings during relocation when a leaf
6149 * that contains file extent items is COWed while
6150 * relocation is ongoing and its in the stage where it
6151 * updates data pointers. So when this happens we can
6152 * safely ignore it since we know it's the same extent,
6153 * but just at different logical and physical locations
6154 * (when an extent is fully replaced with a new one, we
6155 * know the generation number must have changed too,
6156 * since snapshot creation implies committing the current
6157 * transaction, and the inode item must have been updated
6158 * as well).
6159 * This replacement of the disk_bytenr happens at
6160 * relocation.c:replace_file_extents() through
6161 * relocation.c:btrfs_reloc_cow_block().
6162 */
6163 if (btrfs_file_extent_generation(leaf_l, ei_l) ==
6164 btrfs_file_extent_generation(leaf_r, ei_r) &&
6165 btrfs_file_extent_ram_bytes(leaf_l, ei_l) ==
6166 btrfs_file_extent_ram_bytes(leaf_r, ei_r) &&
6167 btrfs_file_extent_compression(leaf_l, ei_l) ==
6168 btrfs_file_extent_compression(leaf_r, ei_r) &&
6169 btrfs_file_extent_encryption(leaf_l, ei_l) ==
6170 btrfs_file_extent_encryption(leaf_r, ei_r) &&
6171 btrfs_file_extent_other_encoding(leaf_l, ei_l) ==
6172 btrfs_file_extent_other_encoding(leaf_r, ei_r) &&
6173 btrfs_file_extent_type(leaf_l, ei_l) ==
6174 btrfs_file_extent_type(leaf_r, ei_r) &&
6175 btrfs_file_extent_disk_bytenr(leaf_l, ei_l) !=
6176 btrfs_file_extent_disk_bytenr(leaf_r, ei_r) &&
6177 btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) ==
6178 btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) &&
6179 btrfs_file_extent_offset(leaf_l, ei_l) ==
6180 btrfs_file_extent_offset(leaf_r, ei_r) &&
6181 btrfs_file_extent_num_bytes(leaf_l, ei_l) ==
6182 btrfs_file_extent_num_bytes(leaf_r, ei_r))
6183 return 0;
6184 }
6185
6186 inconsistent_snapshot_error(sctx, result, "extent");
6187 return -EIO;
6188 }
6189
6190 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6191 if (result != BTRFS_COMPARE_TREE_DELETED)
6192 ret = process_extent(sctx, sctx->left_path,
6193 sctx->cmp_key);
6194 }
6195
6196 return ret;
6197}
6198
6199static int dir_changed(struct send_ctx *sctx, u64 dir)
6200{
6201 u64 orig_gen, new_gen;
6202 int ret;
6203
6204 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6205 NULL, NULL);
6206 if (ret)
6207 return ret;
6208
6209 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6210 NULL, NULL, NULL);
6211 if (ret)
6212 return ret;
6213
6214 return (orig_gen != new_gen) ? 1 : 0;
6215}
6216
6217static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6218 struct btrfs_key *key)
6219{
6220 struct btrfs_inode_extref *extref;
6221 struct extent_buffer *leaf;
6222 u64 dirid = 0, last_dirid = 0;
6223 unsigned long ptr;
6224 u32 item_size;
6225 u32 cur_offset = 0;
6226 int ref_name_len;
6227 int ret = 0;
6228
6229 /* Easy case, just check this one dirid */
6230 if (key->type == BTRFS_INODE_REF_KEY) {
6231 dirid = key->offset;
6232
6233 ret = dir_changed(sctx, dirid);
6234 goto out;
6235 }
6236
6237 leaf = path->nodes[0];
6238 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6239 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6240 while (cur_offset < item_size) {
6241 extref = (struct btrfs_inode_extref *)(ptr +
6242 cur_offset);
6243 dirid = btrfs_inode_extref_parent(leaf, extref);
6244 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6245 cur_offset += ref_name_len + sizeof(*extref);
6246 if (dirid == last_dirid)
6247 continue;
6248 ret = dir_changed(sctx, dirid);
6249 if (ret)
6250 break;
6251 last_dirid = dirid;
6252 }
6253out:
6254 return ret;
6255}
6256
6257/*
6258 * Updates compare related fields in sctx and simply forwards to the actual
6259 * changed_xxx functions.
6260 */
6261static int changed_cb(struct btrfs_path *left_path,
6262 struct btrfs_path *right_path,
6263 struct btrfs_key *key,
6264 enum btrfs_compare_tree_result result,
6265 void *ctx)
6266{
6267 int ret = 0;
6268 struct send_ctx *sctx = ctx;
6269
6270 if (result == BTRFS_COMPARE_TREE_SAME) {
6271 if (key->type == BTRFS_INODE_REF_KEY ||
6272 key->type == BTRFS_INODE_EXTREF_KEY) {
6273 ret = compare_refs(sctx, left_path, key);
6274 if (!ret)
6275 return 0;
6276 if (ret < 0)
6277 return ret;
6278 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6279 return maybe_send_hole(sctx, left_path, key);
6280 } else {
6281 return 0;
6282 }
6283 result = BTRFS_COMPARE_TREE_CHANGED;
6284 ret = 0;
6285 }
6286
6287 sctx->left_path = left_path;
6288 sctx->right_path = right_path;
6289 sctx->cmp_key = key;
6290
6291 ret = finish_inode_if_needed(sctx, 0);
6292 if (ret < 0)
6293 goto out;
6294
6295 /* Ignore non-FS objects */
6296 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6297 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6298 goto out;
6299
6300 if (key->type == BTRFS_INODE_ITEM_KEY)
6301 ret = changed_inode(sctx, result);
6302 else if (key->type == BTRFS_INODE_REF_KEY ||
6303 key->type == BTRFS_INODE_EXTREF_KEY)
6304 ret = changed_ref(sctx, result);
6305 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6306 ret = changed_xattr(sctx, result);
6307 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6308 ret = changed_extent(sctx, result);
6309
6310out:
6311 return ret;
6312}
6313
6314static int full_send_tree(struct send_ctx *sctx)
6315{
6316 int ret;
6317 struct btrfs_root *send_root = sctx->send_root;
6318 struct btrfs_key key;
6319 struct btrfs_key found_key;
6320 struct btrfs_path *path;
6321 struct extent_buffer *eb;
6322 int slot;
6323
6324 path = alloc_path_for_send();
6325 if (!path)
6326 return -ENOMEM;
6327
6328 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6329 key.type = BTRFS_INODE_ITEM_KEY;
6330 key.offset = 0;
6331
6332 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6333 if (ret < 0)
6334 goto out;
6335 if (ret)
6336 goto out_finish;
6337
6338 while (1) {
6339 eb = path->nodes[0];
6340 slot = path->slots[0];
6341 btrfs_item_key_to_cpu(eb, &found_key, slot);
6342
6343 ret = changed_cb(path, NULL, &found_key,
6344 BTRFS_COMPARE_TREE_NEW, sctx);
6345 if (ret < 0)
6346 goto out;
6347
6348 key.objectid = found_key.objectid;
6349 key.type = found_key.type;
6350 key.offset = found_key.offset + 1;
6351
6352 ret = btrfs_next_item(send_root, path);
6353 if (ret < 0)
6354 goto out;
6355 if (ret) {
6356 ret = 0;
6357 break;
6358 }
6359 }
6360
6361out_finish:
6362 ret = finish_inode_if_needed(sctx, 1);
6363
6364out:
6365 btrfs_free_path(path);
6366 return ret;
6367}
6368
6369static int send_subvol(struct send_ctx *sctx)
6370{
6371 int ret;
6372
6373 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6374 ret = send_header(sctx);
6375 if (ret < 0)
6376 goto out;
6377 }
6378
6379 ret = send_subvol_begin(sctx);
6380 if (ret < 0)
6381 goto out;
6382
6383 if (sctx->parent_root) {
6384 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6385 changed_cb, sctx);
6386 if (ret < 0)
6387 goto out;
6388 ret = finish_inode_if_needed(sctx, 1);
6389 if (ret < 0)
6390 goto out;
6391 } else {
6392 ret = full_send_tree(sctx);
6393 if (ret < 0)
6394 goto out;
6395 }
6396
6397out:
6398 free_recorded_refs(sctx);
6399 return ret;
6400}
6401
6402/*
6403 * If orphan cleanup did remove any orphans from a root, it means the tree
6404 * was modified and therefore the commit root is not the same as the current
6405 * root anymore. This is a problem, because send uses the commit root and
6406 * therefore can see inode items that don't exist in the current root anymore,
6407 * and for example make calls to btrfs_iget, which will do tree lookups based
6408 * on the current root and not on the commit root. Those lookups will fail,
6409 * returning a -ESTALE error, and making send fail with that error. So make
6410 * sure a send does not see any orphans we have just removed, and that it will
6411 * see the same inodes regardless of whether a transaction commit happened
6412 * before it started (meaning that the commit root will be the same as the
6413 * current root) or not.
6414 */
6415static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6416{
6417 int i;
6418 struct btrfs_trans_handle *trans = NULL;
6419
6420again:
6421 if (sctx->parent_root &&
6422 sctx->parent_root->node != sctx->parent_root->commit_root)
6423 goto commit_trans;
6424
6425 for (i = 0; i < sctx->clone_roots_cnt; i++)
6426 if (sctx->clone_roots[i].root->node !=
6427 sctx->clone_roots[i].root->commit_root)
6428 goto commit_trans;
6429
6430 if (trans)
6431 return btrfs_end_transaction(trans);
6432
6433 return 0;
6434
6435commit_trans:
6436 /* Use any root, all fs roots will get their commit roots updated. */
6437 if (!trans) {
6438 trans = btrfs_join_transaction(sctx->send_root);
6439 if (IS_ERR(trans))
6440 return PTR_ERR(trans);
6441 goto again;
6442 }
6443
6444 return btrfs_commit_transaction(trans);
6445}
6446
6447static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6448{
6449 spin_lock(&root->root_item_lock);
6450 root->send_in_progress--;
6451 /*
6452 * Not much left to do, we don't know why it's unbalanced and
6453 * can't blindly reset it to 0.
6454 */
6455 if (root->send_in_progress < 0)
6456 btrfs_err(root->fs_info,
6457 "send_in_progres unbalanced %d root %llu",
6458 root->send_in_progress, root->root_key.objectid);
6459 spin_unlock(&root->root_item_lock);
6460}
6461
6462long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
6463{
6464 int ret = 0;
6465 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
6466 struct btrfs_fs_info *fs_info = send_root->fs_info;
6467 struct btrfs_root *clone_root;
6468 struct btrfs_key key;
6469 struct send_ctx *sctx = NULL;
6470 u32 i;
6471 u64 *clone_sources_tmp = NULL;
6472 int clone_sources_to_rollback = 0;
6473 unsigned alloc_size;
6474 int sort_clone_roots = 0;
6475 int index;
6476
6477 if (!capable(CAP_SYS_ADMIN))
6478 return -EPERM;
6479
6480 /*
6481 * The subvolume must remain read-only during send, protect against
6482 * making it RW. This also protects against deletion.
6483 */
6484 spin_lock(&send_root->root_item_lock);
6485 send_root->send_in_progress++;
6486 spin_unlock(&send_root->root_item_lock);
6487
6488 /*
6489 * This is done when we lookup the root, it should already be complete
6490 * by the time we get here.
6491 */
6492 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6493
6494 /*
6495 * Userspace tools do the checks and warn the user if it's
6496 * not RO.
6497 */
6498 if (!btrfs_root_readonly(send_root)) {
6499 ret = -EPERM;
6500 goto out;
6501 }
6502
6503 /*
6504 * Check that we don't overflow at later allocations, we request
6505 * clone_sources_count + 1 items, and compare to unsigned long inside
6506 * access_ok.
6507 */
6508 if (arg->clone_sources_count >
6509 ULONG_MAX / sizeof(struct clone_root) - 1) {
6510 ret = -EINVAL;
6511 goto out;
6512 }
6513
6514 if (!access_ok(VERIFY_READ, arg->clone_sources,
6515 sizeof(*arg->clone_sources) *
6516 arg->clone_sources_count)) {
6517 ret = -EFAULT;
6518 goto out;
6519 }
6520
6521 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
6522 ret = -EINVAL;
6523 goto out;
6524 }
6525
6526 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
6527 if (!sctx) {
6528 ret = -ENOMEM;
6529 goto out;
6530 }
6531
6532 INIT_LIST_HEAD(&sctx->new_refs);
6533 INIT_LIST_HEAD(&sctx->deleted_refs);
6534 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
6535 INIT_LIST_HEAD(&sctx->name_cache_list);
6536
6537 sctx->flags = arg->flags;
6538
6539 sctx->send_filp = fget(arg->send_fd);
6540 if (!sctx->send_filp) {
6541 ret = -EBADF;
6542 goto out;
6543 }
6544
6545 sctx->send_root = send_root;
6546 /*
6547 * Unlikely but possible, if the subvolume is marked for deletion but
6548 * is slow to remove the directory entry, send can still be started
6549 */
6550 if (btrfs_root_dead(sctx->send_root)) {
6551 ret = -EPERM;
6552 goto out;
6553 }
6554
6555 sctx->clone_roots_cnt = arg->clone_sources_count;
6556
6557 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6558 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
6559 if (!sctx->send_buf) {
6560 ret = -ENOMEM;
6561 goto out;
6562 }
6563
6564 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
6565 if (!sctx->read_buf) {
6566 ret = -ENOMEM;
6567 goto out;
6568 }
6569
6570 sctx->pending_dir_moves = RB_ROOT;
6571 sctx->waiting_dir_moves = RB_ROOT;
6572 sctx->orphan_dirs = RB_ROOT;
6573
6574 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6575
6576 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
6577 if (!sctx->clone_roots) {
6578 ret = -ENOMEM;
6579 goto out;
6580 }
6581
6582 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6583
6584 if (arg->clone_sources_count) {
6585 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
6586 if (!clone_sources_tmp) {
6587 ret = -ENOMEM;
6588 goto out;
6589 }
6590
6591 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
6592 alloc_size);
6593 if (ret) {
6594 ret = -EFAULT;
6595 goto out;
6596 }
6597
6598 for (i = 0; i < arg->clone_sources_count; i++) {
6599 key.objectid = clone_sources_tmp[i];
6600 key.type = BTRFS_ROOT_ITEM_KEY;
6601 key.offset = (u64)-1;
6602
6603 index = srcu_read_lock(&fs_info->subvol_srcu);
6604
6605 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
6606 if (IS_ERR(clone_root)) {
6607 srcu_read_unlock(&fs_info->subvol_srcu, index);
6608 ret = PTR_ERR(clone_root);
6609 goto out;
6610 }
6611 spin_lock(&clone_root->root_item_lock);
6612 if (!btrfs_root_readonly(clone_root) ||
6613 btrfs_root_dead(clone_root)) {
6614 spin_unlock(&clone_root->root_item_lock);
6615 srcu_read_unlock(&fs_info->subvol_srcu, index);
6616 ret = -EPERM;
6617 goto out;
6618 }
6619 clone_root->send_in_progress++;
6620 spin_unlock(&clone_root->root_item_lock);
6621 srcu_read_unlock(&fs_info->subvol_srcu, index);
6622
6623 sctx->clone_roots[i].root = clone_root;
6624 clone_sources_to_rollback = i + 1;
6625 }
6626 kvfree(clone_sources_tmp);
6627 clone_sources_tmp = NULL;
6628 }
6629
6630 if (arg->parent_root) {
6631 key.objectid = arg->parent_root;
6632 key.type = BTRFS_ROOT_ITEM_KEY;
6633 key.offset = (u64)-1;
6634
6635 index = srcu_read_lock(&fs_info->subvol_srcu);
6636
6637 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
6638 if (IS_ERR(sctx->parent_root)) {
6639 srcu_read_unlock(&fs_info->subvol_srcu, index);
6640 ret = PTR_ERR(sctx->parent_root);
6641 goto out;
6642 }
6643
6644 spin_lock(&sctx->parent_root->root_item_lock);
6645 sctx->parent_root->send_in_progress++;
6646 if (!btrfs_root_readonly(sctx->parent_root) ||
6647 btrfs_root_dead(sctx->parent_root)) {
6648 spin_unlock(&sctx->parent_root->root_item_lock);
6649 srcu_read_unlock(&fs_info->subvol_srcu, index);
6650 ret = -EPERM;
6651 goto out;
6652 }
6653 spin_unlock(&sctx->parent_root->root_item_lock);
6654
6655 srcu_read_unlock(&fs_info->subvol_srcu, index);
6656 }
6657
6658 /*
6659 * Clones from send_root are allowed, but only if the clone source
6660 * is behind the current send position. This is checked while searching
6661 * for possible clone sources.
6662 */
6663 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6664
6665 /* We do a bsearch later */
6666 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6667 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6668 NULL);
6669 sort_clone_roots = 1;
6670
6671 ret = ensure_commit_roots_uptodate(sctx);
6672 if (ret)
6673 goto out;
6674
6675 current->journal_info = BTRFS_SEND_TRANS_STUB;
6676 ret = send_subvol(sctx);
6677 current->journal_info = NULL;
6678 if (ret < 0)
6679 goto out;
6680
6681 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6682 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6683 if (ret < 0)
6684 goto out;
6685 ret = send_cmd(sctx);
6686 if (ret < 0)
6687 goto out;
6688 }
6689
6690out:
6691 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6692 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6693 struct rb_node *n;
6694 struct pending_dir_move *pm;
6695
6696 n = rb_first(&sctx->pending_dir_moves);
6697 pm = rb_entry(n, struct pending_dir_move, node);
6698 while (!list_empty(&pm->list)) {
6699 struct pending_dir_move *pm2;
6700
6701 pm2 = list_first_entry(&pm->list,
6702 struct pending_dir_move, list);
6703 free_pending_move(sctx, pm2);
6704 }
6705 free_pending_move(sctx, pm);
6706 }
6707
6708 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6709 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6710 struct rb_node *n;
6711 struct waiting_dir_move *dm;
6712
6713 n = rb_first(&sctx->waiting_dir_moves);
6714 dm = rb_entry(n, struct waiting_dir_move, node);
6715 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6716 kfree(dm);
6717 }
6718
6719 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6720 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6721 struct rb_node *n;
6722 struct orphan_dir_info *odi;
6723
6724 n = rb_first(&sctx->orphan_dirs);
6725 odi = rb_entry(n, struct orphan_dir_info, node);
6726 free_orphan_dir_info(sctx, odi);
6727 }
6728
6729 if (sort_clone_roots) {
6730 for (i = 0; i < sctx->clone_roots_cnt; i++)
6731 btrfs_root_dec_send_in_progress(
6732 sctx->clone_roots[i].root);
6733 } else {
6734 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6735 btrfs_root_dec_send_in_progress(
6736 sctx->clone_roots[i].root);
6737
6738 btrfs_root_dec_send_in_progress(send_root);
6739 }
6740 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6741 btrfs_root_dec_send_in_progress(sctx->parent_root);
6742
6743 kvfree(clone_sources_tmp);
6744
6745 if (sctx) {
6746 if (sctx->send_filp)
6747 fput(sctx->send_filp);
6748
6749 kvfree(sctx->clone_roots);
6750 kvfree(sctx->send_buf);
6751 kvfree(sctx->read_buf);
6752
6753 name_cache_free(sctx);
6754
6755 kfree(sctx);
6756 }
6757
6758 return ret;
6759}