Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/bsearch.h>
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/sort.h>
23#include <linux/mount.h>
24#include <linux/xattr.h>
25#include <linux/posix_acl_xattr.h>
26#include <linux/radix-tree.h>
27#include <linux/vmalloc.h>
28#include <linux/string.h>
29
30#include "send.h"
31#include "backref.h"
32#include "hash.h"
33#include "locking.h"
34#include "disk-io.h"
35#include "btrfs_inode.h"
36#include "transaction.h"
37
38static int g_verbose = 0;
39
40#define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
41
42/*
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
48 */
49struct fs_path {
50 union {
51 struct {
52 char *start;
53 char *end;
54
55 char *buf;
56 unsigned short buf_len:15;
57 unsigned short reversed:1;
58 char inline_buf[];
59 };
60 /*
61 * Average path length does not exceed 200 bytes, we'll have
62 * better packing in the slab and higher chance to satisfy
63 * a allocation later during send.
64 */
65 char pad[256];
66 };
67};
68#define FS_PATH_INLINE_SIZE \
69 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
70
71
72/* reused for each extent */
73struct clone_root {
74 struct btrfs_root *root;
75 u64 ino;
76 u64 offset;
77
78 u64 found_refs;
79};
80
81#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
82#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
83
84struct send_ctx {
85 struct file *send_filp;
86 loff_t send_off;
87 char *send_buf;
88 u32 send_size;
89 u32 send_max_size;
90 u64 total_send_size;
91 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
92 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
93
94 struct btrfs_root *send_root;
95 struct btrfs_root *parent_root;
96 struct clone_root *clone_roots;
97 int clone_roots_cnt;
98
99 /* current state of the compare_tree call */
100 struct btrfs_path *left_path;
101 struct btrfs_path *right_path;
102 struct btrfs_key *cmp_key;
103
104 /*
105 * infos of the currently processed inode. In case of deleted inodes,
106 * these are the values from the deleted inode.
107 */
108 u64 cur_ino;
109 u64 cur_inode_gen;
110 int cur_inode_new;
111 int cur_inode_new_gen;
112 int cur_inode_deleted;
113 u64 cur_inode_size;
114 u64 cur_inode_mode;
115 u64 cur_inode_rdev;
116 u64 cur_inode_last_extent;
117
118 u64 send_progress;
119
120 struct list_head new_refs;
121 struct list_head deleted_refs;
122
123 struct radix_tree_root name_cache;
124 struct list_head name_cache_list;
125 int name_cache_size;
126
127 struct file_ra_state ra;
128
129 char *read_buf;
130
131 /*
132 * We process inodes by their increasing order, so if before an
133 * incremental send we reverse the parent/child relationship of
134 * directories such that a directory with a lower inode number was
135 * the parent of a directory with a higher inode number, and the one
136 * becoming the new parent got renamed too, we can't rename/move the
137 * directory with lower inode number when we finish processing it - we
138 * must process the directory with higher inode number first, then
139 * rename/move it and then rename/move the directory with lower inode
140 * number. Example follows.
141 *
142 * Tree state when the first send was performed:
143 *
144 * .
145 * |-- a (ino 257)
146 * |-- b (ino 258)
147 * |
148 * |
149 * |-- c (ino 259)
150 * | |-- d (ino 260)
151 * |
152 * |-- c2 (ino 261)
153 *
154 * Tree state when the second (incremental) send is performed:
155 *
156 * .
157 * |-- a (ino 257)
158 * |-- b (ino 258)
159 * |-- c2 (ino 261)
160 * |-- d2 (ino 260)
161 * |-- cc (ino 259)
162 *
163 * The sequence of steps that lead to the second state was:
164 *
165 * mv /a/b/c/d /a/b/c2/d2
166 * mv /a/b/c /a/b/c2/d2/cc
167 *
168 * "c" has lower inode number, but we can't move it (2nd mv operation)
169 * before we move "d", which has higher inode number.
170 *
171 * So we just memorize which move/rename operations must be performed
172 * later when their respective parent is processed and moved/renamed.
173 */
174
175 /* Indexed by parent directory inode number. */
176 struct rb_root pending_dir_moves;
177
178 /*
179 * Reverse index, indexed by the inode number of a directory that
180 * is waiting for the move/rename of its immediate parent before its
181 * own move/rename can be performed.
182 */
183 struct rb_root waiting_dir_moves;
184
185 /*
186 * A directory that is going to be rm'ed might have a child directory
187 * which is in the pending directory moves index above. In this case,
188 * the directory can only be removed after the move/rename of its child
189 * is performed. Example:
190 *
191 * Parent snapshot:
192 *
193 * . (ino 256)
194 * |-- a/ (ino 257)
195 * |-- b/ (ino 258)
196 * |-- c/ (ino 259)
197 * | |-- x/ (ino 260)
198 * |
199 * |-- y/ (ino 261)
200 *
201 * Send snapshot:
202 *
203 * . (ino 256)
204 * |-- a/ (ino 257)
205 * |-- b/ (ino 258)
206 * |-- YY/ (ino 261)
207 * |-- x/ (ino 260)
208 *
209 * Sequence of steps that lead to the send snapshot:
210 * rm -f /a/b/c/foo.txt
211 * mv /a/b/y /a/b/YY
212 * mv /a/b/c/x /a/b/YY
213 * rmdir /a/b/c
214 *
215 * When the child is processed, its move/rename is delayed until its
216 * parent is processed (as explained above), but all other operations
217 * like update utimes, chown, chgrp, etc, are performed and the paths
218 * that it uses for those operations must use the orphanized name of
219 * its parent (the directory we're going to rm later), so we need to
220 * memorize that name.
221 *
222 * Indexed by the inode number of the directory to be deleted.
223 */
224 struct rb_root orphan_dirs;
225};
226
227struct pending_dir_move {
228 struct rb_node node;
229 struct list_head list;
230 u64 parent_ino;
231 u64 ino;
232 u64 gen;
233 struct list_head update_refs;
234};
235
236struct waiting_dir_move {
237 struct rb_node node;
238 u64 ino;
239 /*
240 * There might be some directory that could not be removed because it
241 * was waiting for this directory inode to be moved first. Therefore
242 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
243 */
244 u64 rmdir_ino;
245};
246
247struct orphan_dir_info {
248 struct rb_node node;
249 u64 ino;
250 u64 gen;
251};
252
253struct name_cache_entry {
254 struct list_head list;
255 /*
256 * radix_tree has only 32bit entries but we need to handle 64bit inums.
257 * We use the lower 32bit of the 64bit inum to store it in the tree. If
258 * more then one inum would fall into the same entry, we use radix_list
259 * to store the additional entries. radix_list is also used to store
260 * entries where two entries have the same inum but different
261 * generations.
262 */
263 struct list_head radix_list;
264 u64 ino;
265 u64 gen;
266 u64 parent_ino;
267 u64 parent_gen;
268 int ret;
269 int need_later_update;
270 int name_len;
271 char name[];
272};
273
274static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
275
276static struct waiting_dir_move *
277get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
278
279static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
280
281static int need_send_hole(struct send_ctx *sctx)
282{
283 return (sctx->parent_root && !sctx->cur_inode_new &&
284 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
285 S_ISREG(sctx->cur_inode_mode));
286}
287
288static void fs_path_reset(struct fs_path *p)
289{
290 if (p->reversed) {
291 p->start = p->buf + p->buf_len - 1;
292 p->end = p->start;
293 *p->start = 0;
294 } else {
295 p->start = p->buf;
296 p->end = p->start;
297 *p->start = 0;
298 }
299}
300
301static struct fs_path *fs_path_alloc(void)
302{
303 struct fs_path *p;
304
305 p = kmalloc(sizeof(*p), GFP_NOFS);
306 if (!p)
307 return NULL;
308 p->reversed = 0;
309 p->buf = p->inline_buf;
310 p->buf_len = FS_PATH_INLINE_SIZE;
311 fs_path_reset(p);
312 return p;
313}
314
315static struct fs_path *fs_path_alloc_reversed(void)
316{
317 struct fs_path *p;
318
319 p = fs_path_alloc();
320 if (!p)
321 return NULL;
322 p->reversed = 1;
323 fs_path_reset(p);
324 return p;
325}
326
327static void fs_path_free(struct fs_path *p)
328{
329 if (!p)
330 return;
331 if (p->buf != p->inline_buf)
332 kfree(p->buf);
333 kfree(p);
334}
335
336static int fs_path_len(struct fs_path *p)
337{
338 return p->end - p->start;
339}
340
341static int fs_path_ensure_buf(struct fs_path *p, int len)
342{
343 char *tmp_buf;
344 int path_len;
345 int old_buf_len;
346
347 len++;
348
349 if (p->buf_len >= len)
350 return 0;
351
352 if (len > PATH_MAX) {
353 WARN_ON(1);
354 return -ENOMEM;
355 }
356
357 path_len = p->end - p->start;
358 old_buf_len = p->buf_len;
359
360 /*
361 * First time the inline_buf does not suffice
362 */
363 if (p->buf == p->inline_buf) {
364 tmp_buf = kmalloc(len, GFP_NOFS);
365 if (tmp_buf)
366 memcpy(tmp_buf, p->buf, old_buf_len);
367 } else {
368 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
369 }
370 if (!tmp_buf)
371 return -ENOMEM;
372 p->buf = tmp_buf;
373 /*
374 * The real size of the buffer is bigger, this will let the fast path
375 * happen most of the time
376 */
377 p->buf_len = ksize(p->buf);
378
379 if (p->reversed) {
380 tmp_buf = p->buf + old_buf_len - path_len - 1;
381 p->end = p->buf + p->buf_len - 1;
382 p->start = p->end - path_len;
383 memmove(p->start, tmp_buf, path_len + 1);
384 } else {
385 p->start = p->buf;
386 p->end = p->start + path_len;
387 }
388 return 0;
389}
390
391static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
392 char **prepared)
393{
394 int ret;
395 int new_len;
396
397 new_len = p->end - p->start + name_len;
398 if (p->start != p->end)
399 new_len++;
400 ret = fs_path_ensure_buf(p, new_len);
401 if (ret < 0)
402 goto out;
403
404 if (p->reversed) {
405 if (p->start != p->end)
406 *--p->start = '/';
407 p->start -= name_len;
408 *prepared = p->start;
409 } else {
410 if (p->start != p->end)
411 *p->end++ = '/';
412 *prepared = p->end;
413 p->end += name_len;
414 *p->end = 0;
415 }
416
417out:
418 return ret;
419}
420
421static int fs_path_add(struct fs_path *p, const char *name, int name_len)
422{
423 int ret;
424 char *prepared;
425
426 ret = fs_path_prepare_for_add(p, name_len, &prepared);
427 if (ret < 0)
428 goto out;
429 memcpy(prepared, name, name_len);
430
431out:
432 return ret;
433}
434
435static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
436{
437 int ret;
438 char *prepared;
439
440 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
441 if (ret < 0)
442 goto out;
443 memcpy(prepared, p2->start, p2->end - p2->start);
444
445out:
446 return ret;
447}
448
449static int fs_path_add_from_extent_buffer(struct fs_path *p,
450 struct extent_buffer *eb,
451 unsigned long off, int len)
452{
453 int ret;
454 char *prepared;
455
456 ret = fs_path_prepare_for_add(p, len, &prepared);
457 if (ret < 0)
458 goto out;
459
460 read_extent_buffer(eb, prepared, off, len);
461
462out:
463 return ret;
464}
465
466static int fs_path_copy(struct fs_path *p, struct fs_path *from)
467{
468 int ret;
469
470 p->reversed = from->reversed;
471 fs_path_reset(p);
472
473 ret = fs_path_add_path(p, from);
474
475 return ret;
476}
477
478
479static void fs_path_unreverse(struct fs_path *p)
480{
481 char *tmp;
482 int len;
483
484 if (!p->reversed)
485 return;
486
487 tmp = p->start;
488 len = p->end - p->start;
489 p->start = p->buf;
490 p->end = p->start + len;
491 memmove(p->start, tmp, len + 1);
492 p->reversed = 0;
493}
494
495static struct btrfs_path *alloc_path_for_send(void)
496{
497 struct btrfs_path *path;
498
499 path = btrfs_alloc_path();
500 if (!path)
501 return NULL;
502 path->search_commit_root = 1;
503 path->skip_locking = 1;
504 path->need_commit_sem = 1;
505 return path;
506}
507
508static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
509{
510 int ret;
511 mm_segment_t old_fs;
512 u32 pos = 0;
513
514 old_fs = get_fs();
515 set_fs(KERNEL_DS);
516
517 while (pos < len) {
518 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
519 /* TODO handle that correctly */
520 /*if (ret == -ERESTARTSYS) {
521 continue;
522 }*/
523 if (ret < 0)
524 goto out;
525 if (ret == 0) {
526 ret = -EIO;
527 goto out;
528 }
529 pos += ret;
530 }
531
532 ret = 0;
533
534out:
535 set_fs(old_fs);
536 return ret;
537}
538
539static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
540{
541 struct btrfs_tlv_header *hdr;
542 int total_len = sizeof(*hdr) + len;
543 int left = sctx->send_max_size - sctx->send_size;
544
545 if (unlikely(left < total_len))
546 return -EOVERFLOW;
547
548 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
549 hdr->tlv_type = cpu_to_le16(attr);
550 hdr->tlv_len = cpu_to_le16(len);
551 memcpy(hdr + 1, data, len);
552 sctx->send_size += total_len;
553
554 return 0;
555}
556
557#define TLV_PUT_DEFINE_INT(bits) \
558 static int tlv_put_u##bits(struct send_ctx *sctx, \
559 u##bits attr, u##bits value) \
560 { \
561 __le##bits __tmp = cpu_to_le##bits(value); \
562 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
563 }
564
565TLV_PUT_DEFINE_INT(64)
566
567static int tlv_put_string(struct send_ctx *sctx, u16 attr,
568 const char *str, int len)
569{
570 if (len == -1)
571 len = strlen(str);
572 return tlv_put(sctx, attr, str, len);
573}
574
575static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
576 const u8 *uuid)
577{
578 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
579}
580
581static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
582 struct extent_buffer *eb,
583 struct btrfs_timespec *ts)
584{
585 struct btrfs_timespec bts;
586 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
587 return tlv_put(sctx, attr, &bts, sizeof(bts));
588}
589
590
591#define TLV_PUT(sctx, attrtype, attrlen, data) \
592 do { \
593 ret = tlv_put(sctx, attrtype, attrlen, data); \
594 if (ret < 0) \
595 goto tlv_put_failure; \
596 } while (0)
597
598#define TLV_PUT_INT(sctx, attrtype, bits, value) \
599 do { \
600 ret = tlv_put_u##bits(sctx, attrtype, value); \
601 if (ret < 0) \
602 goto tlv_put_failure; \
603 } while (0)
604
605#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
606#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
607#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
608#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
609#define TLV_PUT_STRING(sctx, attrtype, str, len) \
610 do { \
611 ret = tlv_put_string(sctx, attrtype, str, len); \
612 if (ret < 0) \
613 goto tlv_put_failure; \
614 } while (0)
615#define TLV_PUT_PATH(sctx, attrtype, p) \
616 do { \
617 ret = tlv_put_string(sctx, attrtype, p->start, \
618 p->end - p->start); \
619 if (ret < 0) \
620 goto tlv_put_failure; \
621 } while(0)
622#define TLV_PUT_UUID(sctx, attrtype, uuid) \
623 do { \
624 ret = tlv_put_uuid(sctx, attrtype, uuid); \
625 if (ret < 0) \
626 goto tlv_put_failure; \
627 } while (0)
628#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
629 do { \
630 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
631 if (ret < 0) \
632 goto tlv_put_failure; \
633 } while (0)
634
635static int send_header(struct send_ctx *sctx)
636{
637 struct btrfs_stream_header hdr;
638
639 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
640 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
641
642 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
643 &sctx->send_off);
644}
645
646/*
647 * For each command/item we want to send to userspace, we call this function.
648 */
649static int begin_cmd(struct send_ctx *sctx, int cmd)
650{
651 struct btrfs_cmd_header *hdr;
652
653 if (WARN_ON(!sctx->send_buf))
654 return -EINVAL;
655
656 BUG_ON(sctx->send_size);
657
658 sctx->send_size += sizeof(*hdr);
659 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
660 hdr->cmd = cpu_to_le16(cmd);
661
662 return 0;
663}
664
665static int send_cmd(struct send_ctx *sctx)
666{
667 int ret;
668 struct btrfs_cmd_header *hdr;
669 u32 crc;
670
671 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
672 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
673 hdr->crc = 0;
674
675 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
676 hdr->crc = cpu_to_le32(crc);
677
678 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
679 &sctx->send_off);
680
681 sctx->total_send_size += sctx->send_size;
682 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
683 sctx->send_size = 0;
684
685 return ret;
686}
687
688/*
689 * Sends a move instruction to user space
690 */
691static int send_rename(struct send_ctx *sctx,
692 struct fs_path *from, struct fs_path *to)
693{
694 int ret;
695
696verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
697
698 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
699 if (ret < 0)
700 goto out;
701
702 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
703 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
704
705 ret = send_cmd(sctx);
706
707tlv_put_failure:
708out:
709 return ret;
710}
711
712/*
713 * Sends a link instruction to user space
714 */
715static int send_link(struct send_ctx *sctx,
716 struct fs_path *path, struct fs_path *lnk)
717{
718 int ret;
719
720verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
721
722 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
723 if (ret < 0)
724 goto out;
725
726 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
727 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
728
729 ret = send_cmd(sctx);
730
731tlv_put_failure:
732out:
733 return ret;
734}
735
736/*
737 * Sends an unlink instruction to user space
738 */
739static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
740{
741 int ret;
742
743verbose_printk("btrfs: send_unlink %s\n", path->start);
744
745 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
746 if (ret < 0)
747 goto out;
748
749 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
750
751 ret = send_cmd(sctx);
752
753tlv_put_failure:
754out:
755 return ret;
756}
757
758/*
759 * Sends a rmdir instruction to user space
760 */
761static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
762{
763 int ret;
764
765verbose_printk("btrfs: send_rmdir %s\n", path->start);
766
767 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
768 if (ret < 0)
769 goto out;
770
771 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
772
773 ret = send_cmd(sctx);
774
775tlv_put_failure:
776out:
777 return ret;
778}
779
780/*
781 * Helper function to retrieve some fields from an inode item.
782 */
783static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
784 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
785 u64 *gid, u64 *rdev)
786{
787 int ret;
788 struct btrfs_inode_item *ii;
789 struct btrfs_key key;
790
791 key.objectid = ino;
792 key.type = BTRFS_INODE_ITEM_KEY;
793 key.offset = 0;
794 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
795 if (ret) {
796 if (ret > 0)
797 ret = -ENOENT;
798 return ret;
799 }
800
801 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
802 struct btrfs_inode_item);
803 if (size)
804 *size = btrfs_inode_size(path->nodes[0], ii);
805 if (gen)
806 *gen = btrfs_inode_generation(path->nodes[0], ii);
807 if (mode)
808 *mode = btrfs_inode_mode(path->nodes[0], ii);
809 if (uid)
810 *uid = btrfs_inode_uid(path->nodes[0], ii);
811 if (gid)
812 *gid = btrfs_inode_gid(path->nodes[0], ii);
813 if (rdev)
814 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
815
816 return ret;
817}
818
819static int get_inode_info(struct btrfs_root *root,
820 u64 ino, u64 *size, u64 *gen,
821 u64 *mode, u64 *uid, u64 *gid,
822 u64 *rdev)
823{
824 struct btrfs_path *path;
825 int ret;
826
827 path = alloc_path_for_send();
828 if (!path)
829 return -ENOMEM;
830 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
831 rdev);
832 btrfs_free_path(path);
833 return ret;
834}
835
836typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
837 struct fs_path *p,
838 void *ctx);
839
840/*
841 * Helper function to iterate the entries in ONE btrfs_inode_ref or
842 * btrfs_inode_extref.
843 * The iterate callback may return a non zero value to stop iteration. This can
844 * be a negative value for error codes or 1 to simply stop it.
845 *
846 * path must point to the INODE_REF or INODE_EXTREF when called.
847 */
848static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
849 struct btrfs_key *found_key, int resolve,
850 iterate_inode_ref_t iterate, void *ctx)
851{
852 struct extent_buffer *eb = path->nodes[0];
853 struct btrfs_item *item;
854 struct btrfs_inode_ref *iref;
855 struct btrfs_inode_extref *extref;
856 struct btrfs_path *tmp_path;
857 struct fs_path *p;
858 u32 cur = 0;
859 u32 total;
860 int slot = path->slots[0];
861 u32 name_len;
862 char *start;
863 int ret = 0;
864 int num = 0;
865 int index;
866 u64 dir;
867 unsigned long name_off;
868 unsigned long elem_size;
869 unsigned long ptr;
870
871 p = fs_path_alloc_reversed();
872 if (!p)
873 return -ENOMEM;
874
875 tmp_path = alloc_path_for_send();
876 if (!tmp_path) {
877 fs_path_free(p);
878 return -ENOMEM;
879 }
880
881
882 if (found_key->type == BTRFS_INODE_REF_KEY) {
883 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
884 struct btrfs_inode_ref);
885 item = btrfs_item_nr(slot);
886 total = btrfs_item_size(eb, item);
887 elem_size = sizeof(*iref);
888 } else {
889 ptr = btrfs_item_ptr_offset(eb, slot);
890 total = btrfs_item_size_nr(eb, slot);
891 elem_size = sizeof(*extref);
892 }
893
894 while (cur < total) {
895 fs_path_reset(p);
896
897 if (found_key->type == BTRFS_INODE_REF_KEY) {
898 iref = (struct btrfs_inode_ref *)(ptr + cur);
899 name_len = btrfs_inode_ref_name_len(eb, iref);
900 name_off = (unsigned long)(iref + 1);
901 index = btrfs_inode_ref_index(eb, iref);
902 dir = found_key->offset;
903 } else {
904 extref = (struct btrfs_inode_extref *)(ptr + cur);
905 name_len = btrfs_inode_extref_name_len(eb, extref);
906 name_off = (unsigned long)&extref->name;
907 index = btrfs_inode_extref_index(eb, extref);
908 dir = btrfs_inode_extref_parent(eb, extref);
909 }
910
911 if (resolve) {
912 start = btrfs_ref_to_path(root, tmp_path, name_len,
913 name_off, eb, dir,
914 p->buf, p->buf_len);
915 if (IS_ERR(start)) {
916 ret = PTR_ERR(start);
917 goto out;
918 }
919 if (start < p->buf) {
920 /* overflow , try again with larger buffer */
921 ret = fs_path_ensure_buf(p,
922 p->buf_len + p->buf - start);
923 if (ret < 0)
924 goto out;
925 start = btrfs_ref_to_path(root, tmp_path,
926 name_len, name_off,
927 eb, dir,
928 p->buf, p->buf_len);
929 if (IS_ERR(start)) {
930 ret = PTR_ERR(start);
931 goto out;
932 }
933 BUG_ON(start < p->buf);
934 }
935 p->start = start;
936 } else {
937 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
938 name_len);
939 if (ret < 0)
940 goto out;
941 }
942
943 cur += elem_size + name_len;
944 ret = iterate(num, dir, index, p, ctx);
945 if (ret)
946 goto out;
947 num++;
948 }
949
950out:
951 btrfs_free_path(tmp_path);
952 fs_path_free(p);
953 return ret;
954}
955
956typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
957 const char *name, int name_len,
958 const char *data, int data_len,
959 u8 type, void *ctx);
960
961/*
962 * Helper function to iterate the entries in ONE btrfs_dir_item.
963 * The iterate callback may return a non zero value to stop iteration. This can
964 * be a negative value for error codes or 1 to simply stop it.
965 *
966 * path must point to the dir item when called.
967 */
968static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
969 struct btrfs_key *found_key,
970 iterate_dir_item_t iterate, void *ctx)
971{
972 int ret = 0;
973 struct extent_buffer *eb;
974 struct btrfs_item *item;
975 struct btrfs_dir_item *di;
976 struct btrfs_key di_key;
977 char *buf = NULL;
978 const int buf_len = PATH_MAX;
979 u32 name_len;
980 u32 data_len;
981 u32 cur;
982 u32 len;
983 u32 total;
984 int slot;
985 int num;
986 u8 type;
987
988 buf = kmalloc(buf_len, GFP_NOFS);
989 if (!buf) {
990 ret = -ENOMEM;
991 goto out;
992 }
993
994 eb = path->nodes[0];
995 slot = path->slots[0];
996 item = btrfs_item_nr(slot);
997 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
998 cur = 0;
999 len = 0;
1000 total = btrfs_item_size(eb, item);
1001
1002 num = 0;
1003 while (cur < total) {
1004 name_len = btrfs_dir_name_len(eb, di);
1005 data_len = btrfs_dir_data_len(eb, di);
1006 type = btrfs_dir_type(eb, di);
1007 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1008
1009 /*
1010 * Path too long
1011 */
1012 if (name_len + data_len > buf_len) {
1013 ret = -ENAMETOOLONG;
1014 goto out;
1015 }
1016
1017 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1018 name_len + data_len);
1019
1020 len = sizeof(*di) + name_len + data_len;
1021 di = (struct btrfs_dir_item *)((char *)di + len);
1022 cur += len;
1023
1024 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1025 data_len, type, ctx);
1026 if (ret < 0)
1027 goto out;
1028 if (ret) {
1029 ret = 0;
1030 goto out;
1031 }
1032
1033 num++;
1034 }
1035
1036out:
1037 kfree(buf);
1038 return ret;
1039}
1040
1041static int __copy_first_ref(int num, u64 dir, int index,
1042 struct fs_path *p, void *ctx)
1043{
1044 int ret;
1045 struct fs_path *pt = ctx;
1046
1047 ret = fs_path_copy(pt, p);
1048 if (ret < 0)
1049 return ret;
1050
1051 /* we want the first only */
1052 return 1;
1053}
1054
1055/*
1056 * Retrieve the first path of an inode. If an inode has more then one
1057 * ref/hardlink, this is ignored.
1058 */
1059static int get_inode_path(struct btrfs_root *root,
1060 u64 ino, struct fs_path *path)
1061{
1062 int ret;
1063 struct btrfs_key key, found_key;
1064 struct btrfs_path *p;
1065
1066 p = alloc_path_for_send();
1067 if (!p)
1068 return -ENOMEM;
1069
1070 fs_path_reset(path);
1071
1072 key.objectid = ino;
1073 key.type = BTRFS_INODE_REF_KEY;
1074 key.offset = 0;
1075
1076 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1077 if (ret < 0)
1078 goto out;
1079 if (ret) {
1080 ret = 1;
1081 goto out;
1082 }
1083 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1084 if (found_key.objectid != ino ||
1085 (found_key.type != BTRFS_INODE_REF_KEY &&
1086 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1087 ret = -ENOENT;
1088 goto out;
1089 }
1090
1091 ret = iterate_inode_ref(root, p, &found_key, 1,
1092 __copy_first_ref, path);
1093 if (ret < 0)
1094 goto out;
1095 ret = 0;
1096
1097out:
1098 btrfs_free_path(p);
1099 return ret;
1100}
1101
1102struct backref_ctx {
1103 struct send_ctx *sctx;
1104
1105 struct btrfs_path *path;
1106 /* number of total found references */
1107 u64 found;
1108
1109 /*
1110 * used for clones found in send_root. clones found behind cur_objectid
1111 * and cur_offset are not considered as allowed clones.
1112 */
1113 u64 cur_objectid;
1114 u64 cur_offset;
1115
1116 /* may be truncated in case it's the last extent in a file */
1117 u64 extent_len;
1118
1119 /* Just to check for bugs in backref resolving */
1120 int found_itself;
1121};
1122
1123static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1124{
1125 u64 root = (u64)(uintptr_t)key;
1126 struct clone_root *cr = (struct clone_root *)elt;
1127
1128 if (root < cr->root->objectid)
1129 return -1;
1130 if (root > cr->root->objectid)
1131 return 1;
1132 return 0;
1133}
1134
1135static int __clone_root_cmp_sort(const void *e1, const void *e2)
1136{
1137 struct clone_root *cr1 = (struct clone_root *)e1;
1138 struct clone_root *cr2 = (struct clone_root *)e2;
1139
1140 if (cr1->root->objectid < cr2->root->objectid)
1141 return -1;
1142 if (cr1->root->objectid > cr2->root->objectid)
1143 return 1;
1144 return 0;
1145}
1146
1147/*
1148 * Called for every backref that is found for the current extent.
1149 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1150 */
1151static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1152{
1153 struct backref_ctx *bctx = ctx_;
1154 struct clone_root *found;
1155 int ret;
1156 u64 i_size;
1157
1158 /* First check if the root is in the list of accepted clone sources */
1159 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1160 bctx->sctx->clone_roots_cnt,
1161 sizeof(struct clone_root),
1162 __clone_root_cmp_bsearch);
1163 if (!found)
1164 return 0;
1165
1166 if (found->root == bctx->sctx->send_root &&
1167 ino == bctx->cur_objectid &&
1168 offset == bctx->cur_offset) {
1169 bctx->found_itself = 1;
1170 }
1171
1172 /*
1173 * There are inodes that have extents that lie behind its i_size. Don't
1174 * accept clones from these extents.
1175 */
1176 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1177 NULL, NULL, NULL);
1178 btrfs_release_path(bctx->path);
1179 if (ret < 0)
1180 return ret;
1181
1182 if (offset + bctx->extent_len > i_size)
1183 return 0;
1184
1185 /*
1186 * Make sure we don't consider clones from send_root that are
1187 * behind the current inode/offset.
1188 */
1189 if (found->root == bctx->sctx->send_root) {
1190 /*
1191 * TODO for the moment we don't accept clones from the inode
1192 * that is currently send. We may change this when
1193 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1194 * file.
1195 */
1196 if (ino >= bctx->cur_objectid)
1197 return 0;
1198#if 0
1199 if (ino > bctx->cur_objectid)
1200 return 0;
1201 if (offset + bctx->extent_len > bctx->cur_offset)
1202 return 0;
1203#endif
1204 }
1205
1206 bctx->found++;
1207 found->found_refs++;
1208 if (ino < found->ino) {
1209 found->ino = ino;
1210 found->offset = offset;
1211 } else if (found->ino == ino) {
1212 /*
1213 * same extent found more then once in the same file.
1214 */
1215 if (found->offset > offset + bctx->extent_len)
1216 found->offset = offset;
1217 }
1218
1219 return 0;
1220}
1221
1222/*
1223 * Given an inode, offset and extent item, it finds a good clone for a clone
1224 * instruction. Returns -ENOENT when none could be found. The function makes
1225 * sure that the returned clone is usable at the point where sending is at the
1226 * moment. This means, that no clones are accepted which lie behind the current
1227 * inode+offset.
1228 *
1229 * path must point to the extent item when called.
1230 */
1231static int find_extent_clone(struct send_ctx *sctx,
1232 struct btrfs_path *path,
1233 u64 ino, u64 data_offset,
1234 u64 ino_size,
1235 struct clone_root **found)
1236{
1237 int ret;
1238 int extent_type;
1239 u64 logical;
1240 u64 disk_byte;
1241 u64 num_bytes;
1242 u64 extent_item_pos;
1243 u64 flags = 0;
1244 struct btrfs_file_extent_item *fi;
1245 struct extent_buffer *eb = path->nodes[0];
1246 struct backref_ctx *backref_ctx = NULL;
1247 struct clone_root *cur_clone_root;
1248 struct btrfs_key found_key;
1249 struct btrfs_path *tmp_path;
1250 int compressed;
1251 u32 i;
1252
1253 tmp_path = alloc_path_for_send();
1254 if (!tmp_path)
1255 return -ENOMEM;
1256
1257 /* We only use this path under the commit sem */
1258 tmp_path->need_commit_sem = 0;
1259
1260 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1261 if (!backref_ctx) {
1262 ret = -ENOMEM;
1263 goto out;
1264 }
1265
1266 backref_ctx->path = tmp_path;
1267
1268 if (data_offset >= ino_size) {
1269 /*
1270 * There may be extents that lie behind the file's size.
1271 * I at least had this in combination with snapshotting while
1272 * writing large files.
1273 */
1274 ret = 0;
1275 goto out;
1276 }
1277
1278 fi = btrfs_item_ptr(eb, path->slots[0],
1279 struct btrfs_file_extent_item);
1280 extent_type = btrfs_file_extent_type(eb, fi);
1281 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1282 ret = -ENOENT;
1283 goto out;
1284 }
1285 compressed = btrfs_file_extent_compression(eb, fi);
1286
1287 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1288 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1289 if (disk_byte == 0) {
1290 ret = -ENOENT;
1291 goto out;
1292 }
1293 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1294
1295 down_read(&sctx->send_root->fs_info->commit_root_sem);
1296 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1297 &found_key, &flags);
1298 up_read(&sctx->send_root->fs_info->commit_root_sem);
1299 btrfs_release_path(tmp_path);
1300
1301 if (ret < 0)
1302 goto out;
1303 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1304 ret = -EIO;
1305 goto out;
1306 }
1307
1308 /*
1309 * Setup the clone roots.
1310 */
1311 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1312 cur_clone_root = sctx->clone_roots + i;
1313 cur_clone_root->ino = (u64)-1;
1314 cur_clone_root->offset = 0;
1315 cur_clone_root->found_refs = 0;
1316 }
1317
1318 backref_ctx->sctx = sctx;
1319 backref_ctx->found = 0;
1320 backref_ctx->cur_objectid = ino;
1321 backref_ctx->cur_offset = data_offset;
1322 backref_ctx->found_itself = 0;
1323 backref_ctx->extent_len = num_bytes;
1324
1325 /*
1326 * The last extent of a file may be too large due to page alignment.
1327 * We need to adjust extent_len in this case so that the checks in
1328 * __iterate_backrefs work.
1329 */
1330 if (data_offset + num_bytes >= ino_size)
1331 backref_ctx->extent_len = ino_size - data_offset;
1332
1333 /*
1334 * Now collect all backrefs.
1335 */
1336 if (compressed == BTRFS_COMPRESS_NONE)
1337 extent_item_pos = logical - found_key.objectid;
1338 else
1339 extent_item_pos = 0;
1340 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1341 found_key.objectid, extent_item_pos, 1,
1342 __iterate_backrefs, backref_ctx);
1343
1344 if (ret < 0)
1345 goto out;
1346
1347 if (!backref_ctx->found_itself) {
1348 /* found a bug in backref code? */
1349 ret = -EIO;
1350 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1351 "send_root. inode=%llu, offset=%llu, "
1352 "disk_byte=%llu found extent=%llu\n",
1353 ino, data_offset, disk_byte, found_key.objectid);
1354 goto out;
1355 }
1356
1357verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1358 "ino=%llu, "
1359 "num_bytes=%llu, logical=%llu\n",
1360 data_offset, ino, num_bytes, logical);
1361
1362 if (!backref_ctx->found)
1363 verbose_printk("btrfs: no clones found\n");
1364
1365 cur_clone_root = NULL;
1366 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1367 if (sctx->clone_roots[i].found_refs) {
1368 if (!cur_clone_root)
1369 cur_clone_root = sctx->clone_roots + i;
1370 else if (sctx->clone_roots[i].root == sctx->send_root)
1371 /* prefer clones from send_root over others */
1372 cur_clone_root = sctx->clone_roots + i;
1373 }
1374
1375 }
1376
1377 if (cur_clone_root) {
1378 if (compressed != BTRFS_COMPRESS_NONE) {
1379 /*
1380 * Offsets given by iterate_extent_inodes() are relative
1381 * to the start of the extent, we need to add logical
1382 * offset from the file extent item.
1383 * (See why at backref.c:check_extent_in_eb())
1384 */
1385 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1386 fi);
1387 }
1388 *found = cur_clone_root;
1389 ret = 0;
1390 } else {
1391 ret = -ENOENT;
1392 }
1393
1394out:
1395 btrfs_free_path(tmp_path);
1396 kfree(backref_ctx);
1397 return ret;
1398}
1399
1400static int read_symlink(struct btrfs_root *root,
1401 u64 ino,
1402 struct fs_path *dest)
1403{
1404 int ret;
1405 struct btrfs_path *path;
1406 struct btrfs_key key;
1407 struct btrfs_file_extent_item *ei;
1408 u8 type;
1409 u8 compression;
1410 unsigned long off;
1411 int len;
1412
1413 path = alloc_path_for_send();
1414 if (!path)
1415 return -ENOMEM;
1416
1417 key.objectid = ino;
1418 key.type = BTRFS_EXTENT_DATA_KEY;
1419 key.offset = 0;
1420 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1421 if (ret < 0)
1422 goto out;
1423 BUG_ON(ret);
1424
1425 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1426 struct btrfs_file_extent_item);
1427 type = btrfs_file_extent_type(path->nodes[0], ei);
1428 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1429 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1430 BUG_ON(compression);
1431
1432 off = btrfs_file_extent_inline_start(ei);
1433 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1434
1435 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1436
1437out:
1438 btrfs_free_path(path);
1439 return ret;
1440}
1441
1442/*
1443 * Helper function to generate a file name that is unique in the root of
1444 * send_root and parent_root. This is used to generate names for orphan inodes.
1445 */
1446static int gen_unique_name(struct send_ctx *sctx,
1447 u64 ino, u64 gen,
1448 struct fs_path *dest)
1449{
1450 int ret = 0;
1451 struct btrfs_path *path;
1452 struct btrfs_dir_item *di;
1453 char tmp[64];
1454 int len;
1455 u64 idx = 0;
1456
1457 path = alloc_path_for_send();
1458 if (!path)
1459 return -ENOMEM;
1460
1461 while (1) {
1462 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1463 ino, gen, idx);
1464 ASSERT(len < sizeof(tmp));
1465
1466 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1467 path, BTRFS_FIRST_FREE_OBJECTID,
1468 tmp, strlen(tmp), 0);
1469 btrfs_release_path(path);
1470 if (IS_ERR(di)) {
1471 ret = PTR_ERR(di);
1472 goto out;
1473 }
1474 if (di) {
1475 /* not unique, try again */
1476 idx++;
1477 continue;
1478 }
1479
1480 if (!sctx->parent_root) {
1481 /* unique */
1482 ret = 0;
1483 break;
1484 }
1485
1486 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1487 path, BTRFS_FIRST_FREE_OBJECTID,
1488 tmp, strlen(tmp), 0);
1489 btrfs_release_path(path);
1490 if (IS_ERR(di)) {
1491 ret = PTR_ERR(di);
1492 goto out;
1493 }
1494 if (di) {
1495 /* not unique, try again */
1496 idx++;
1497 continue;
1498 }
1499 /* unique */
1500 break;
1501 }
1502
1503 ret = fs_path_add(dest, tmp, strlen(tmp));
1504
1505out:
1506 btrfs_free_path(path);
1507 return ret;
1508}
1509
1510enum inode_state {
1511 inode_state_no_change,
1512 inode_state_will_create,
1513 inode_state_did_create,
1514 inode_state_will_delete,
1515 inode_state_did_delete,
1516};
1517
1518static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1519{
1520 int ret;
1521 int left_ret;
1522 int right_ret;
1523 u64 left_gen;
1524 u64 right_gen;
1525
1526 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1527 NULL, NULL);
1528 if (ret < 0 && ret != -ENOENT)
1529 goto out;
1530 left_ret = ret;
1531
1532 if (!sctx->parent_root) {
1533 right_ret = -ENOENT;
1534 } else {
1535 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1536 NULL, NULL, NULL, NULL);
1537 if (ret < 0 && ret != -ENOENT)
1538 goto out;
1539 right_ret = ret;
1540 }
1541
1542 if (!left_ret && !right_ret) {
1543 if (left_gen == gen && right_gen == gen) {
1544 ret = inode_state_no_change;
1545 } else if (left_gen == gen) {
1546 if (ino < sctx->send_progress)
1547 ret = inode_state_did_create;
1548 else
1549 ret = inode_state_will_create;
1550 } else if (right_gen == gen) {
1551 if (ino < sctx->send_progress)
1552 ret = inode_state_did_delete;
1553 else
1554 ret = inode_state_will_delete;
1555 } else {
1556 ret = -ENOENT;
1557 }
1558 } else if (!left_ret) {
1559 if (left_gen == gen) {
1560 if (ino < sctx->send_progress)
1561 ret = inode_state_did_create;
1562 else
1563 ret = inode_state_will_create;
1564 } else {
1565 ret = -ENOENT;
1566 }
1567 } else if (!right_ret) {
1568 if (right_gen == gen) {
1569 if (ino < sctx->send_progress)
1570 ret = inode_state_did_delete;
1571 else
1572 ret = inode_state_will_delete;
1573 } else {
1574 ret = -ENOENT;
1575 }
1576 } else {
1577 ret = -ENOENT;
1578 }
1579
1580out:
1581 return ret;
1582}
1583
1584static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1585{
1586 int ret;
1587
1588 ret = get_cur_inode_state(sctx, ino, gen);
1589 if (ret < 0)
1590 goto out;
1591
1592 if (ret == inode_state_no_change ||
1593 ret == inode_state_did_create ||
1594 ret == inode_state_will_delete)
1595 ret = 1;
1596 else
1597 ret = 0;
1598
1599out:
1600 return ret;
1601}
1602
1603/*
1604 * Helper function to lookup a dir item in a dir.
1605 */
1606static int lookup_dir_item_inode(struct btrfs_root *root,
1607 u64 dir, const char *name, int name_len,
1608 u64 *found_inode,
1609 u8 *found_type)
1610{
1611 int ret = 0;
1612 struct btrfs_dir_item *di;
1613 struct btrfs_key key;
1614 struct btrfs_path *path;
1615
1616 path = alloc_path_for_send();
1617 if (!path)
1618 return -ENOMEM;
1619
1620 di = btrfs_lookup_dir_item(NULL, root, path,
1621 dir, name, name_len, 0);
1622 if (!di) {
1623 ret = -ENOENT;
1624 goto out;
1625 }
1626 if (IS_ERR(di)) {
1627 ret = PTR_ERR(di);
1628 goto out;
1629 }
1630 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1631 *found_inode = key.objectid;
1632 *found_type = btrfs_dir_type(path->nodes[0], di);
1633
1634out:
1635 btrfs_free_path(path);
1636 return ret;
1637}
1638
1639/*
1640 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1641 * generation of the parent dir and the name of the dir entry.
1642 */
1643static int get_first_ref(struct btrfs_root *root, u64 ino,
1644 u64 *dir, u64 *dir_gen, struct fs_path *name)
1645{
1646 int ret;
1647 struct btrfs_key key;
1648 struct btrfs_key found_key;
1649 struct btrfs_path *path;
1650 int len;
1651 u64 parent_dir;
1652
1653 path = alloc_path_for_send();
1654 if (!path)
1655 return -ENOMEM;
1656
1657 key.objectid = ino;
1658 key.type = BTRFS_INODE_REF_KEY;
1659 key.offset = 0;
1660
1661 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1662 if (ret < 0)
1663 goto out;
1664 if (!ret)
1665 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1666 path->slots[0]);
1667 if (ret || found_key.objectid != ino ||
1668 (found_key.type != BTRFS_INODE_REF_KEY &&
1669 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1670 ret = -ENOENT;
1671 goto out;
1672 }
1673
1674 if (found_key.type == BTRFS_INODE_REF_KEY) {
1675 struct btrfs_inode_ref *iref;
1676 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1677 struct btrfs_inode_ref);
1678 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1679 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1680 (unsigned long)(iref + 1),
1681 len);
1682 parent_dir = found_key.offset;
1683 } else {
1684 struct btrfs_inode_extref *extref;
1685 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1686 struct btrfs_inode_extref);
1687 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1688 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1689 (unsigned long)&extref->name, len);
1690 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1691 }
1692 if (ret < 0)
1693 goto out;
1694 btrfs_release_path(path);
1695
1696 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1697 NULL, NULL);
1698 if (ret < 0)
1699 goto out;
1700
1701 *dir = parent_dir;
1702
1703out:
1704 btrfs_free_path(path);
1705 return ret;
1706}
1707
1708static int is_first_ref(struct btrfs_root *root,
1709 u64 ino, u64 dir,
1710 const char *name, int name_len)
1711{
1712 int ret;
1713 struct fs_path *tmp_name;
1714 u64 tmp_dir;
1715 u64 tmp_dir_gen;
1716
1717 tmp_name = fs_path_alloc();
1718 if (!tmp_name)
1719 return -ENOMEM;
1720
1721 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1722 if (ret < 0)
1723 goto out;
1724
1725 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1726 ret = 0;
1727 goto out;
1728 }
1729
1730 ret = !memcmp(tmp_name->start, name, name_len);
1731
1732out:
1733 fs_path_free(tmp_name);
1734 return ret;
1735}
1736
1737/*
1738 * Used by process_recorded_refs to determine if a new ref would overwrite an
1739 * already existing ref. In case it detects an overwrite, it returns the
1740 * inode/gen in who_ino/who_gen.
1741 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1742 * to make sure later references to the overwritten inode are possible.
1743 * Orphanizing is however only required for the first ref of an inode.
1744 * process_recorded_refs does an additional is_first_ref check to see if
1745 * orphanizing is really required.
1746 */
1747static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1748 const char *name, int name_len,
1749 u64 *who_ino, u64 *who_gen)
1750{
1751 int ret = 0;
1752 u64 gen;
1753 u64 other_inode = 0;
1754 u8 other_type = 0;
1755
1756 if (!sctx->parent_root)
1757 goto out;
1758
1759 ret = is_inode_existent(sctx, dir, dir_gen);
1760 if (ret <= 0)
1761 goto out;
1762
1763 /*
1764 * If we have a parent root we need to verify that the parent dir was
1765 * not delted and then re-created, if it was then we have no overwrite
1766 * and we can just unlink this entry.
1767 */
1768 if (sctx->parent_root) {
1769 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1770 NULL, NULL, NULL);
1771 if (ret < 0 && ret != -ENOENT)
1772 goto out;
1773 if (ret) {
1774 ret = 0;
1775 goto out;
1776 }
1777 if (gen != dir_gen)
1778 goto out;
1779 }
1780
1781 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1782 &other_inode, &other_type);
1783 if (ret < 0 && ret != -ENOENT)
1784 goto out;
1785 if (ret) {
1786 ret = 0;
1787 goto out;
1788 }
1789
1790 /*
1791 * Check if the overwritten ref was already processed. If yes, the ref
1792 * was already unlinked/moved, so we can safely assume that we will not
1793 * overwrite anything at this point in time.
1794 */
1795 if (other_inode > sctx->send_progress) {
1796 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1797 who_gen, NULL, NULL, NULL, NULL);
1798 if (ret < 0)
1799 goto out;
1800
1801 ret = 1;
1802 *who_ino = other_inode;
1803 } else {
1804 ret = 0;
1805 }
1806
1807out:
1808 return ret;
1809}
1810
1811/*
1812 * Checks if the ref was overwritten by an already processed inode. This is
1813 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1814 * thus the orphan name needs be used.
1815 * process_recorded_refs also uses it to avoid unlinking of refs that were
1816 * overwritten.
1817 */
1818static int did_overwrite_ref(struct send_ctx *sctx,
1819 u64 dir, u64 dir_gen,
1820 u64 ino, u64 ino_gen,
1821 const char *name, int name_len)
1822{
1823 int ret = 0;
1824 u64 gen;
1825 u64 ow_inode;
1826 u8 other_type;
1827
1828 if (!sctx->parent_root)
1829 goto out;
1830
1831 ret = is_inode_existent(sctx, dir, dir_gen);
1832 if (ret <= 0)
1833 goto out;
1834
1835 /* check if the ref was overwritten by another ref */
1836 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1837 &ow_inode, &other_type);
1838 if (ret < 0 && ret != -ENOENT)
1839 goto out;
1840 if (ret) {
1841 /* was never and will never be overwritten */
1842 ret = 0;
1843 goto out;
1844 }
1845
1846 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1847 NULL, NULL);
1848 if (ret < 0)
1849 goto out;
1850
1851 if (ow_inode == ino && gen == ino_gen) {
1852 ret = 0;
1853 goto out;
1854 }
1855
1856 /* we know that it is or will be overwritten. check this now */
1857 if (ow_inode < sctx->send_progress)
1858 ret = 1;
1859 else
1860 ret = 0;
1861
1862out:
1863 return ret;
1864}
1865
1866/*
1867 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1868 * that got overwritten. This is used by process_recorded_refs to determine
1869 * if it has to use the path as returned by get_cur_path or the orphan name.
1870 */
1871static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1872{
1873 int ret = 0;
1874 struct fs_path *name = NULL;
1875 u64 dir;
1876 u64 dir_gen;
1877
1878 if (!sctx->parent_root)
1879 goto out;
1880
1881 name = fs_path_alloc();
1882 if (!name)
1883 return -ENOMEM;
1884
1885 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1886 if (ret < 0)
1887 goto out;
1888
1889 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1890 name->start, fs_path_len(name));
1891
1892out:
1893 fs_path_free(name);
1894 return ret;
1895}
1896
1897/*
1898 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1899 * so we need to do some special handling in case we have clashes. This function
1900 * takes care of this with the help of name_cache_entry::radix_list.
1901 * In case of error, nce is kfreed.
1902 */
1903static int name_cache_insert(struct send_ctx *sctx,
1904 struct name_cache_entry *nce)
1905{
1906 int ret = 0;
1907 struct list_head *nce_head;
1908
1909 nce_head = radix_tree_lookup(&sctx->name_cache,
1910 (unsigned long)nce->ino);
1911 if (!nce_head) {
1912 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1913 if (!nce_head) {
1914 kfree(nce);
1915 return -ENOMEM;
1916 }
1917 INIT_LIST_HEAD(nce_head);
1918
1919 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1920 if (ret < 0) {
1921 kfree(nce_head);
1922 kfree(nce);
1923 return ret;
1924 }
1925 }
1926 list_add_tail(&nce->radix_list, nce_head);
1927 list_add_tail(&nce->list, &sctx->name_cache_list);
1928 sctx->name_cache_size++;
1929
1930 return ret;
1931}
1932
1933static void name_cache_delete(struct send_ctx *sctx,
1934 struct name_cache_entry *nce)
1935{
1936 struct list_head *nce_head;
1937
1938 nce_head = radix_tree_lookup(&sctx->name_cache,
1939 (unsigned long)nce->ino);
1940 if (!nce_head) {
1941 btrfs_err(sctx->send_root->fs_info,
1942 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
1943 nce->ino, sctx->name_cache_size);
1944 }
1945
1946 list_del(&nce->radix_list);
1947 list_del(&nce->list);
1948 sctx->name_cache_size--;
1949
1950 /*
1951 * We may not get to the final release of nce_head if the lookup fails
1952 */
1953 if (nce_head && list_empty(nce_head)) {
1954 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1955 kfree(nce_head);
1956 }
1957}
1958
1959static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1960 u64 ino, u64 gen)
1961{
1962 struct list_head *nce_head;
1963 struct name_cache_entry *cur;
1964
1965 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1966 if (!nce_head)
1967 return NULL;
1968
1969 list_for_each_entry(cur, nce_head, radix_list) {
1970 if (cur->ino == ino && cur->gen == gen)
1971 return cur;
1972 }
1973 return NULL;
1974}
1975
1976/*
1977 * Removes the entry from the list and adds it back to the end. This marks the
1978 * entry as recently used so that name_cache_clean_unused does not remove it.
1979 */
1980static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1981{
1982 list_del(&nce->list);
1983 list_add_tail(&nce->list, &sctx->name_cache_list);
1984}
1985
1986/*
1987 * Remove some entries from the beginning of name_cache_list.
1988 */
1989static void name_cache_clean_unused(struct send_ctx *sctx)
1990{
1991 struct name_cache_entry *nce;
1992
1993 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1994 return;
1995
1996 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1997 nce = list_entry(sctx->name_cache_list.next,
1998 struct name_cache_entry, list);
1999 name_cache_delete(sctx, nce);
2000 kfree(nce);
2001 }
2002}
2003
2004static void name_cache_free(struct send_ctx *sctx)
2005{
2006 struct name_cache_entry *nce;
2007
2008 while (!list_empty(&sctx->name_cache_list)) {
2009 nce = list_entry(sctx->name_cache_list.next,
2010 struct name_cache_entry, list);
2011 name_cache_delete(sctx, nce);
2012 kfree(nce);
2013 }
2014}
2015
2016/*
2017 * Used by get_cur_path for each ref up to the root.
2018 * Returns 0 if it succeeded.
2019 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2020 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2021 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2022 * Returns <0 in case of error.
2023 */
2024static int __get_cur_name_and_parent(struct send_ctx *sctx,
2025 u64 ino, u64 gen,
2026 u64 *parent_ino,
2027 u64 *parent_gen,
2028 struct fs_path *dest)
2029{
2030 int ret;
2031 int nce_ret;
2032 struct btrfs_path *path = NULL;
2033 struct name_cache_entry *nce = NULL;
2034
2035 /*
2036 * First check if we already did a call to this function with the same
2037 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2038 * return the cached result.
2039 */
2040 nce = name_cache_search(sctx, ino, gen);
2041 if (nce) {
2042 if (ino < sctx->send_progress && nce->need_later_update) {
2043 name_cache_delete(sctx, nce);
2044 kfree(nce);
2045 nce = NULL;
2046 } else {
2047 name_cache_used(sctx, nce);
2048 *parent_ino = nce->parent_ino;
2049 *parent_gen = nce->parent_gen;
2050 ret = fs_path_add(dest, nce->name, nce->name_len);
2051 if (ret < 0)
2052 goto out;
2053 ret = nce->ret;
2054 goto out;
2055 }
2056 }
2057
2058 path = alloc_path_for_send();
2059 if (!path)
2060 return -ENOMEM;
2061
2062 /*
2063 * If the inode is not existent yet, add the orphan name and return 1.
2064 * This should only happen for the parent dir that we determine in
2065 * __record_new_ref
2066 */
2067 ret = is_inode_existent(sctx, ino, gen);
2068 if (ret < 0)
2069 goto out;
2070
2071 if (!ret) {
2072 ret = gen_unique_name(sctx, ino, gen, dest);
2073 if (ret < 0)
2074 goto out;
2075 ret = 1;
2076 goto out_cache;
2077 }
2078
2079 /*
2080 * Depending on whether the inode was already processed or not, use
2081 * send_root or parent_root for ref lookup.
2082 */
2083 if (ino < sctx->send_progress)
2084 ret = get_first_ref(sctx->send_root, ino,
2085 parent_ino, parent_gen, dest);
2086 else
2087 ret = get_first_ref(sctx->parent_root, ino,
2088 parent_ino, parent_gen, dest);
2089 if (ret < 0)
2090 goto out;
2091
2092 /*
2093 * Check if the ref was overwritten by an inode's ref that was processed
2094 * earlier. If yes, treat as orphan and return 1.
2095 */
2096 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2097 dest->start, dest->end - dest->start);
2098 if (ret < 0)
2099 goto out;
2100 if (ret) {
2101 fs_path_reset(dest);
2102 ret = gen_unique_name(sctx, ino, gen, dest);
2103 if (ret < 0)
2104 goto out;
2105 ret = 1;
2106 }
2107
2108out_cache:
2109 /*
2110 * Store the result of the lookup in the name cache.
2111 */
2112 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2113 if (!nce) {
2114 ret = -ENOMEM;
2115 goto out;
2116 }
2117
2118 nce->ino = ino;
2119 nce->gen = gen;
2120 nce->parent_ino = *parent_ino;
2121 nce->parent_gen = *parent_gen;
2122 nce->name_len = fs_path_len(dest);
2123 nce->ret = ret;
2124 strcpy(nce->name, dest->start);
2125
2126 if (ino < sctx->send_progress)
2127 nce->need_later_update = 0;
2128 else
2129 nce->need_later_update = 1;
2130
2131 nce_ret = name_cache_insert(sctx, nce);
2132 if (nce_ret < 0)
2133 ret = nce_ret;
2134 name_cache_clean_unused(sctx);
2135
2136out:
2137 btrfs_free_path(path);
2138 return ret;
2139}
2140
2141/*
2142 * Magic happens here. This function returns the first ref to an inode as it
2143 * would look like while receiving the stream at this point in time.
2144 * We walk the path up to the root. For every inode in between, we check if it
2145 * was already processed/sent. If yes, we continue with the parent as found
2146 * in send_root. If not, we continue with the parent as found in parent_root.
2147 * If we encounter an inode that was deleted at this point in time, we use the
2148 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2149 * that were not created yet and overwritten inodes/refs.
2150 *
2151 * When do we have have orphan inodes:
2152 * 1. When an inode is freshly created and thus no valid refs are available yet
2153 * 2. When a directory lost all it's refs (deleted) but still has dir items
2154 * inside which were not processed yet (pending for move/delete). If anyone
2155 * tried to get the path to the dir items, it would get a path inside that
2156 * orphan directory.
2157 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2158 * of an unprocessed inode. If in that case the first ref would be
2159 * overwritten, the overwritten inode gets "orphanized". Later when we
2160 * process this overwritten inode, it is restored at a new place by moving
2161 * the orphan inode.
2162 *
2163 * sctx->send_progress tells this function at which point in time receiving
2164 * would be.
2165 */
2166static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2167 struct fs_path *dest)
2168{
2169 int ret = 0;
2170 struct fs_path *name = NULL;
2171 u64 parent_inode = 0;
2172 u64 parent_gen = 0;
2173 int stop = 0;
2174
2175 name = fs_path_alloc();
2176 if (!name) {
2177 ret = -ENOMEM;
2178 goto out;
2179 }
2180
2181 dest->reversed = 1;
2182 fs_path_reset(dest);
2183
2184 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2185 fs_path_reset(name);
2186
2187 if (is_waiting_for_rm(sctx, ino)) {
2188 ret = gen_unique_name(sctx, ino, gen, name);
2189 if (ret < 0)
2190 goto out;
2191 ret = fs_path_add_path(dest, name);
2192 break;
2193 }
2194
2195 if (is_waiting_for_move(sctx, ino)) {
2196 ret = get_first_ref(sctx->parent_root, ino,
2197 &parent_inode, &parent_gen, name);
2198 } else {
2199 ret = __get_cur_name_and_parent(sctx, ino, gen,
2200 &parent_inode,
2201 &parent_gen, name);
2202 if (ret)
2203 stop = 1;
2204 }
2205
2206 if (ret < 0)
2207 goto out;
2208
2209 ret = fs_path_add_path(dest, name);
2210 if (ret < 0)
2211 goto out;
2212
2213 ino = parent_inode;
2214 gen = parent_gen;
2215 }
2216
2217out:
2218 fs_path_free(name);
2219 if (!ret)
2220 fs_path_unreverse(dest);
2221 return ret;
2222}
2223
2224/*
2225 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2226 */
2227static int send_subvol_begin(struct send_ctx *sctx)
2228{
2229 int ret;
2230 struct btrfs_root *send_root = sctx->send_root;
2231 struct btrfs_root *parent_root = sctx->parent_root;
2232 struct btrfs_path *path;
2233 struct btrfs_key key;
2234 struct btrfs_root_ref *ref;
2235 struct extent_buffer *leaf;
2236 char *name = NULL;
2237 int namelen;
2238
2239 path = btrfs_alloc_path();
2240 if (!path)
2241 return -ENOMEM;
2242
2243 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2244 if (!name) {
2245 btrfs_free_path(path);
2246 return -ENOMEM;
2247 }
2248
2249 key.objectid = send_root->objectid;
2250 key.type = BTRFS_ROOT_BACKREF_KEY;
2251 key.offset = 0;
2252
2253 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2254 &key, path, 1, 0);
2255 if (ret < 0)
2256 goto out;
2257 if (ret) {
2258 ret = -ENOENT;
2259 goto out;
2260 }
2261
2262 leaf = path->nodes[0];
2263 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2264 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2265 key.objectid != send_root->objectid) {
2266 ret = -ENOENT;
2267 goto out;
2268 }
2269 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2270 namelen = btrfs_root_ref_name_len(leaf, ref);
2271 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2272 btrfs_release_path(path);
2273
2274 if (parent_root) {
2275 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2276 if (ret < 0)
2277 goto out;
2278 } else {
2279 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2280 if (ret < 0)
2281 goto out;
2282 }
2283
2284 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2285 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2286 sctx->send_root->root_item.uuid);
2287 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2288 le64_to_cpu(sctx->send_root->root_item.ctransid));
2289 if (parent_root) {
2290 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2291 sctx->parent_root->root_item.uuid);
2292 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2293 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2294 }
2295
2296 ret = send_cmd(sctx);
2297
2298tlv_put_failure:
2299out:
2300 btrfs_free_path(path);
2301 kfree(name);
2302 return ret;
2303}
2304
2305static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2306{
2307 int ret = 0;
2308 struct fs_path *p;
2309
2310verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2311
2312 p = fs_path_alloc();
2313 if (!p)
2314 return -ENOMEM;
2315
2316 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2317 if (ret < 0)
2318 goto out;
2319
2320 ret = get_cur_path(sctx, ino, gen, p);
2321 if (ret < 0)
2322 goto out;
2323 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2324 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2325
2326 ret = send_cmd(sctx);
2327
2328tlv_put_failure:
2329out:
2330 fs_path_free(p);
2331 return ret;
2332}
2333
2334static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2335{
2336 int ret = 0;
2337 struct fs_path *p;
2338
2339verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2340
2341 p = fs_path_alloc();
2342 if (!p)
2343 return -ENOMEM;
2344
2345 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2346 if (ret < 0)
2347 goto out;
2348
2349 ret = get_cur_path(sctx, ino, gen, p);
2350 if (ret < 0)
2351 goto out;
2352 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2353 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2354
2355 ret = send_cmd(sctx);
2356
2357tlv_put_failure:
2358out:
2359 fs_path_free(p);
2360 return ret;
2361}
2362
2363static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2364{
2365 int ret = 0;
2366 struct fs_path *p;
2367
2368verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2369
2370 p = fs_path_alloc();
2371 if (!p)
2372 return -ENOMEM;
2373
2374 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2375 if (ret < 0)
2376 goto out;
2377
2378 ret = get_cur_path(sctx, ino, gen, p);
2379 if (ret < 0)
2380 goto out;
2381 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2382 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2383 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2384
2385 ret = send_cmd(sctx);
2386
2387tlv_put_failure:
2388out:
2389 fs_path_free(p);
2390 return ret;
2391}
2392
2393static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2394{
2395 int ret = 0;
2396 struct fs_path *p = NULL;
2397 struct btrfs_inode_item *ii;
2398 struct btrfs_path *path = NULL;
2399 struct extent_buffer *eb;
2400 struct btrfs_key key;
2401 int slot;
2402
2403verbose_printk("btrfs: send_utimes %llu\n", ino);
2404
2405 p = fs_path_alloc();
2406 if (!p)
2407 return -ENOMEM;
2408
2409 path = alloc_path_for_send();
2410 if (!path) {
2411 ret = -ENOMEM;
2412 goto out;
2413 }
2414
2415 key.objectid = ino;
2416 key.type = BTRFS_INODE_ITEM_KEY;
2417 key.offset = 0;
2418 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2419 if (ret < 0)
2420 goto out;
2421
2422 eb = path->nodes[0];
2423 slot = path->slots[0];
2424 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2425
2426 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2427 if (ret < 0)
2428 goto out;
2429
2430 ret = get_cur_path(sctx, ino, gen, p);
2431 if (ret < 0)
2432 goto out;
2433 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2434 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2435 btrfs_inode_atime(ii));
2436 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2437 btrfs_inode_mtime(ii));
2438 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2439 btrfs_inode_ctime(ii));
2440 /* TODO Add otime support when the otime patches get into upstream */
2441
2442 ret = send_cmd(sctx);
2443
2444tlv_put_failure:
2445out:
2446 fs_path_free(p);
2447 btrfs_free_path(path);
2448 return ret;
2449}
2450
2451/*
2452 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2453 * a valid path yet because we did not process the refs yet. So, the inode
2454 * is created as orphan.
2455 */
2456static int send_create_inode(struct send_ctx *sctx, u64 ino)
2457{
2458 int ret = 0;
2459 struct fs_path *p;
2460 int cmd;
2461 u64 gen;
2462 u64 mode;
2463 u64 rdev;
2464
2465verbose_printk("btrfs: send_create_inode %llu\n", ino);
2466
2467 p = fs_path_alloc();
2468 if (!p)
2469 return -ENOMEM;
2470
2471 if (ino != sctx->cur_ino) {
2472 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2473 NULL, NULL, &rdev);
2474 if (ret < 0)
2475 goto out;
2476 } else {
2477 gen = sctx->cur_inode_gen;
2478 mode = sctx->cur_inode_mode;
2479 rdev = sctx->cur_inode_rdev;
2480 }
2481
2482 if (S_ISREG(mode)) {
2483 cmd = BTRFS_SEND_C_MKFILE;
2484 } else if (S_ISDIR(mode)) {
2485 cmd = BTRFS_SEND_C_MKDIR;
2486 } else if (S_ISLNK(mode)) {
2487 cmd = BTRFS_SEND_C_SYMLINK;
2488 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2489 cmd = BTRFS_SEND_C_MKNOD;
2490 } else if (S_ISFIFO(mode)) {
2491 cmd = BTRFS_SEND_C_MKFIFO;
2492 } else if (S_ISSOCK(mode)) {
2493 cmd = BTRFS_SEND_C_MKSOCK;
2494 } else {
2495 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2496 (int)(mode & S_IFMT));
2497 ret = -ENOTSUPP;
2498 goto out;
2499 }
2500
2501 ret = begin_cmd(sctx, cmd);
2502 if (ret < 0)
2503 goto out;
2504
2505 ret = gen_unique_name(sctx, ino, gen, p);
2506 if (ret < 0)
2507 goto out;
2508
2509 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2510 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2511
2512 if (S_ISLNK(mode)) {
2513 fs_path_reset(p);
2514 ret = read_symlink(sctx->send_root, ino, p);
2515 if (ret < 0)
2516 goto out;
2517 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2518 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2519 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2520 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2521 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2522 }
2523
2524 ret = send_cmd(sctx);
2525 if (ret < 0)
2526 goto out;
2527
2528
2529tlv_put_failure:
2530out:
2531 fs_path_free(p);
2532 return ret;
2533}
2534
2535/*
2536 * We need some special handling for inodes that get processed before the parent
2537 * directory got created. See process_recorded_refs for details.
2538 * This function does the check if we already created the dir out of order.
2539 */
2540static int did_create_dir(struct send_ctx *sctx, u64 dir)
2541{
2542 int ret = 0;
2543 struct btrfs_path *path = NULL;
2544 struct btrfs_key key;
2545 struct btrfs_key found_key;
2546 struct btrfs_key di_key;
2547 struct extent_buffer *eb;
2548 struct btrfs_dir_item *di;
2549 int slot;
2550
2551 path = alloc_path_for_send();
2552 if (!path) {
2553 ret = -ENOMEM;
2554 goto out;
2555 }
2556
2557 key.objectid = dir;
2558 key.type = BTRFS_DIR_INDEX_KEY;
2559 key.offset = 0;
2560 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2561 if (ret < 0)
2562 goto out;
2563
2564 while (1) {
2565 eb = path->nodes[0];
2566 slot = path->slots[0];
2567 if (slot >= btrfs_header_nritems(eb)) {
2568 ret = btrfs_next_leaf(sctx->send_root, path);
2569 if (ret < 0) {
2570 goto out;
2571 } else if (ret > 0) {
2572 ret = 0;
2573 break;
2574 }
2575 continue;
2576 }
2577
2578 btrfs_item_key_to_cpu(eb, &found_key, slot);
2579 if (found_key.objectid != key.objectid ||
2580 found_key.type != key.type) {
2581 ret = 0;
2582 goto out;
2583 }
2584
2585 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2586 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2587
2588 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2589 di_key.objectid < sctx->send_progress) {
2590 ret = 1;
2591 goto out;
2592 }
2593
2594 path->slots[0]++;
2595 }
2596
2597out:
2598 btrfs_free_path(path);
2599 return ret;
2600}
2601
2602/*
2603 * Only creates the inode if it is:
2604 * 1. Not a directory
2605 * 2. Or a directory which was not created already due to out of order
2606 * directories. See did_create_dir and process_recorded_refs for details.
2607 */
2608static int send_create_inode_if_needed(struct send_ctx *sctx)
2609{
2610 int ret;
2611
2612 if (S_ISDIR(sctx->cur_inode_mode)) {
2613 ret = did_create_dir(sctx, sctx->cur_ino);
2614 if (ret < 0)
2615 goto out;
2616 if (ret) {
2617 ret = 0;
2618 goto out;
2619 }
2620 }
2621
2622 ret = send_create_inode(sctx, sctx->cur_ino);
2623 if (ret < 0)
2624 goto out;
2625
2626out:
2627 return ret;
2628}
2629
2630struct recorded_ref {
2631 struct list_head list;
2632 char *dir_path;
2633 char *name;
2634 struct fs_path *full_path;
2635 u64 dir;
2636 u64 dir_gen;
2637 int dir_path_len;
2638 int name_len;
2639};
2640
2641/*
2642 * We need to process new refs before deleted refs, but compare_tree gives us
2643 * everything mixed. So we first record all refs and later process them.
2644 * This function is a helper to record one ref.
2645 */
2646static int __record_ref(struct list_head *head, u64 dir,
2647 u64 dir_gen, struct fs_path *path)
2648{
2649 struct recorded_ref *ref;
2650
2651 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2652 if (!ref)
2653 return -ENOMEM;
2654
2655 ref->dir = dir;
2656 ref->dir_gen = dir_gen;
2657 ref->full_path = path;
2658
2659 ref->name = (char *)kbasename(ref->full_path->start);
2660 ref->name_len = ref->full_path->end - ref->name;
2661 ref->dir_path = ref->full_path->start;
2662 if (ref->name == ref->full_path->start)
2663 ref->dir_path_len = 0;
2664 else
2665 ref->dir_path_len = ref->full_path->end -
2666 ref->full_path->start - 1 - ref->name_len;
2667
2668 list_add_tail(&ref->list, head);
2669 return 0;
2670}
2671
2672static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2673{
2674 struct recorded_ref *new;
2675
2676 new = kmalloc(sizeof(*ref), GFP_NOFS);
2677 if (!new)
2678 return -ENOMEM;
2679
2680 new->dir = ref->dir;
2681 new->dir_gen = ref->dir_gen;
2682 new->full_path = NULL;
2683 INIT_LIST_HEAD(&new->list);
2684 list_add_tail(&new->list, list);
2685 return 0;
2686}
2687
2688static void __free_recorded_refs(struct list_head *head)
2689{
2690 struct recorded_ref *cur;
2691
2692 while (!list_empty(head)) {
2693 cur = list_entry(head->next, struct recorded_ref, list);
2694 fs_path_free(cur->full_path);
2695 list_del(&cur->list);
2696 kfree(cur);
2697 }
2698}
2699
2700static void free_recorded_refs(struct send_ctx *sctx)
2701{
2702 __free_recorded_refs(&sctx->new_refs);
2703 __free_recorded_refs(&sctx->deleted_refs);
2704}
2705
2706/*
2707 * Renames/moves a file/dir to its orphan name. Used when the first
2708 * ref of an unprocessed inode gets overwritten and for all non empty
2709 * directories.
2710 */
2711static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2712 struct fs_path *path)
2713{
2714 int ret;
2715 struct fs_path *orphan;
2716
2717 orphan = fs_path_alloc();
2718 if (!orphan)
2719 return -ENOMEM;
2720
2721 ret = gen_unique_name(sctx, ino, gen, orphan);
2722 if (ret < 0)
2723 goto out;
2724
2725 ret = send_rename(sctx, path, orphan);
2726
2727out:
2728 fs_path_free(orphan);
2729 return ret;
2730}
2731
2732static struct orphan_dir_info *
2733add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2734{
2735 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2736 struct rb_node *parent = NULL;
2737 struct orphan_dir_info *entry, *odi;
2738
2739 odi = kmalloc(sizeof(*odi), GFP_NOFS);
2740 if (!odi)
2741 return ERR_PTR(-ENOMEM);
2742 odi->ino = dir_ino;
2743 odi->gen = 0;
2744
2745 while (*p) {
2746 parent = *p;
2747 entry = rb_entry(parent, struct orphan_dir_info, node);
2748 if (dir_ino < entry->ino) {
2749 p = &(*p)->rb_left;
2750 } else if (dir_ino > entry->ino) {
2751 p = &(*p)->rb_right;
2752 } else {
2753 kfree(odi);
2754 return entry;
2755 }
2756 }
2757
2758 rb_link_node(&odi->node, parent, p);
2759 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2760 return odi;
2761}
2762
2763static struct orphan_dir_info *
2764get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2765{
2766 struct rb_node *n = sctx->orphan_dirs.rb_node;
2767 struct orphan_dir_info *entry;
2768
2769 while (n) {
2770 entry = rb_entry(n, struct orphan_dir_info, node);
2771 if (dir_ino < entry->ino)
2772 n = n->rb_left;
2773 else if (dir_ino > entry->ino)
2774 n = n->rb_right;
2775 else
2776 return entry;
2777 }
2778 return NULL;
2779}
2780
2781static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2782{
2783 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2784
2785 return odi != NULL;
2786}
2787
2788static void free_orphan_dir_info(struct send_ctx *sctx,
2789 struct orphan_dir_info *odi)
2790{
2791 if (!odi)
2792 return;
2793 rb_erase(&odi->node, &sctx->orphan_dirs);
2794 kfree(odi);
2795}
2796
2797/*
2798 * Returns 1 if a directory can be removed at this point in time.
2799 * We check this by iterating all dir items and checking if the inode behind
2800 * the dir item was already processed.
2801 */
2802static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2803 u64 send_progress)
2804{
2805 int ret = 0;
2806 struct btrfs_root *root = sctx->parent_root;
2807 struct btrfs_path *path;
2808 struct btrfs_key key;
2809 struct btrfs_key found_key;
2810 struct btrfs_key loc;
2811 struct btrfs_dir_item *di;
2812
2813 /*
2814 * Don't try to rmdir the top/root subvolume dir.
2815 */
2816 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2817 return 0;
2818
2819 path = alloc_path_for_send();
2820 if (!path)
2821 return -ENOMEM;
2822
2823 key.objectid = dir;
2824 key.type = BTRFS_DIR_INDEX_KEY;
2825 key.offset = 0;
2826 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2827 if (ret < 0)
2828 goto out;
2829
2830 while (1) {
2831 struct waiting_dir_move *dm;
2832
2833 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2834 ret = btrfs_next_leaf(root, path);
2835 if (ret < 0)
2836 goto out;
2837 else if (ret > 0)
2838 break;
2839 continue;
2840 }
2841 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2842 path->slots[0]);
2843 if (found_key.objectid != key.objectid ||
2844 found_key.type != key.type)
2845 break;
2846
2847 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2848 struct btrfs_dir_item);
2849 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2850
2851 dm = get_waiting_dir_move(sctx, loc.objectid);
2852 if (dm) {
2853 struct orphan_dir_info *odi;
2854
2855 odi = add_orphan_dir_info(sctx, dir);
2856 if (IS_ERR(odi)) {
2857 ret = PTR_ERR(odi);
2858 goto out;
2859 }
2860 odi->gen = dir_gen;
2861 dm->rmdir_ino = dir;
2862 ret = 0;
2863 goto out;
2864 }
2865
2866 if (loc.objectid > send_progress) {
2867 ret = 0;
2868 goto out;
2869 }
2870
2871 path->slots[0]++;
2872 }
2873
2874 ret = 1;
2875
2876out:
2877 btrfs_free_path(path);
2878 return ret;
2879}
2880
2881static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2882{
2883 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2884
2885 return entry != NULL;
2886}
2887
2888static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2889{
2890 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2891 struct rb_node *parent = NULL;
2892 struct waiting_dir_move *entry, *dm;
2893
2894 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2895 if (!dm)
2896 return -ENOMEM;
2897 dm->ino = ino;
2898 dm->rmdir_ino = 0;
2899
2900 while (*p) {
2901 parent = *p;
2902 entry = rb_entry(parent, struct waiting_dir_move, node);
2903 if (ino < entry->ino) {
2904 p = &(*p)->rb_left;
2905 } else if (ino > entry->ino) {
2906 p = &(*p)->rb_right;
2907 } else {
2908 kfree(dm);
2909 return -EEXIST;
2910 }
2911 }
2912
2913 rb_link_node(&dm->node, parent, p);
2914 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2915 return 0;
2916}
2917
2918static struct waiting_dir_move *
2919get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2920{
2921 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2922 struct waiting_dir_move *entry;
2923
2924 while (n) {
2925 entry = rb_entry(n, struct waiting_dir_move, node);
2926 if (ino < entry->ino)
2927 n = n->rb_left;
2928 else if (ino > entry->ino)
2929 n = n->rb_right;
2930 else
2931 return entry;
2932 }
2933 return NULL;
2934}
2935
2936static void free_waiting_dir_move(struct send_ctx *sctx,
2937 struct waiting_dir_move *dm)
2938{
2939 if (!dm)
2940 return;
2941 rb_erase(&dm->node, &sctx->waiting_dir_moves);
2942 kfree(dm);
2943}
2944
2945static int add_pending_dir_move(struct send_ctx *sctx,
2946 u64 ino,
2947 u64 ino_gen,
2948 u64 parent_ino)
2949{
2950 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2951 struct rb_node *parent = NULL;
2952 struct pending_dir_move *entry = NULL, *pm;
2953 struct recorded_ref *cur;
2954 int exists = 0;
2955 int ret;
2956
2957 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2958 if (!pm)
2959 return -ENOMEM;
2960 pm->parent_ino = parent_ino;
2961 pm->ino = ino;
2962 pm->gen = ino_gen;
2963 INIT_LIST_HEAD(&pm->list);
2964 INIT_LIST_HEAD(&pm->update_refs);
2965 RB_CLEAR_NODE(&pm->node);
2966
2967 while (*p) {
2968 parent = *p;
2969 entry = rb_entry(parent, struct pending_dir_move, node);
2970 if (parent_ino < entry->parent_ino) {
2971 p = &(*p)->rb_left;
2972 } else if (parent_ino > entry->parent_ino) {
2973 p = &(*p)->rb_right;
2974 } else {
2975 exists = 1;
2976 break;
2977 }
2978 }
2979
2980 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2981 ret = dup_ref(cur, &pm->update_refs);
2982 if (ret < 0)
2983 goto out;
2984 }
2985 list_for_each_entry(cur, &sctx->new_refs, list) {
2986 ret = dup_ref(cur, &pm->update_refs);
2987 if (ret < 0)
2988 goto out;
2989 }
2990
2991 ret = add_waiting_dir_move(sctx, pm->ino);
2992 if (ret)
2993 goto out;
2994
2995 if (exists) {
2996 list_add_tail(&pm->list, &entry->list);
2997 } else {
2998 rb_link_node(&pm->node, parent, p);
2999 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3000 }
3001 ret = 0;
3002out:
3003 if (ret) {
3004 __free_recorded_refs(&pm->update_refs);
3005 kfree(pm);
3006 }
3007 return ret;
3008}
3009
3010static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3011 u64 parent_ino)
3012{
3013 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3014 struct pending_dir_move *entry;
3015
3016 while (n) {
3017 entry = rb_entry(n, struct pending_dir_move, node);
3018 if (parent_ino < entry->parent_ino)
3019 n = n->rb_left;
3020 else if (parent_ino > entry->parent_ino)
3021 n = n->rb_right;
3022 else
3023 return entry;
3024 }
3025 return NULL;
3026}
3027
3028static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3029{
3030 struct fs_path *from_path = NULL;
3031 struct fs_path *to_path = NULL;
3032 struct fs_path *name = NULL;
3033 u64 orig_progress = sctx->send_progress;
3034 struct recorded_ref *cur;
3035 u64 parent_ino, parent_gen;
3036 struct waiting_dir_move *dm = NULL;
3037 u64 rmdir_ino = 0;
3038 int ret;
3039
3040 name = fs_path_alloc();
3041 from_path = fs_path_alloc();
3042 if (!name || !from_path) {
3043 ret = -ENOMEM;
3044 goto out;
3045 }
3046
3047 dm = get_waiting_dir_move(sctx, pm->ino);
3048 ASSERT(dm);
3049 rmdir_ino = dm->rmdir_ino;
3050 free_waiting_dir_move(sctx, dm);
3051
3052 ret = get_first_ref(sctx->parent_root, pm->ino,
3053 &parent_ino, &parent_gen, name);
3054 if (ret < 0)
3055 goto out;
3056
3057 if (parent_ino == sctx->cur_ino) {
3058 /* child only renamed, not moved */
3059 ASSERT(parent_gen == sctx->cur_inode_gen);
3060 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3061 from_path);
3062 if (ret < 0)
3063 goto out;
3064 ret = fs_path_add_path(from_path, name);
3065 if (ret < 0)
3066 goto out;
3067 } else {
3068 /* child moved and maybe renamed too */
3069 sctx->send_progress = pm->ino;
3070 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
3071 if (ret < 0)
3072 goto out;
3073 }
3074
3075 fs_path_free(name);
3076 name = NULL;
3077
3078 to_path = fs_path_alloc();
3079 if (!to_path) {
3080 ret = -ENOMEM;
3081 goto out;
3082 }
3083
3084 sctx->send_progress = sctx->cur_ino + 1;
3085 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3086 if (ret < 0)
3087 goto out;
3088
3089 ret = send_rename(sctx, from_path, to_path);
3090 if (ret < 0)
3091 goto out;
3092
3093 if (rmdir_ino) {
3094 struct orphan_dir_info *odi;
3095
3096 odi = get_orphan_dir_info(sctx, rmdir_ino);
3097 if (!odi) {
3098 /* already deleted */
3099 goto finish;
3100 }
3101 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1);
3102 if (ret < 0)
3103 goto out;
3104 if (!ret)
3105 goto finish;
3106
3107 name = fs_path_alloc();
3108 if (!name) {
3109 ret = -ENOMEM;
3110 goto out;
3111 }
3112 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3113 if (ret < 0)
3114 goto out;
3115 ret = send_rmdir(sctx, name);
3116 if (ret < 0)
3117 goto out;
3118 free_orphan_dir_info(sctx, odi);
3119 }
3120
3121finish:
3122 ret = send_utimes(sctx, pm->ino, pm->gen);
3123 if (ret < 0)
3124 goto out;
3125
3126 /*
3127 * After rename/move, need to update the utimes of both new parent(s)
3128 * and old parent(s).
3129 */
3130 list_for_each_entry(cur, &pm->update_refs, list) {
3131 if (cur->dir == rmdir_ino)
3132 continue;
3133 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3134 if (ret < 0)
3135 goto out;
3136 }
3137
3138out:
3139 fs_path_free(name);
3140 fs_path_free(from_path);
3141 fs_path_free(to_path);
3142 sctx->send_progress = orig_progress;
3143
3144 return ret;
3145}
3146
3147static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3148{
3149 if (!list_empty(&m->list))
3150 list_del(&m->list);
3151 if (!RB_EMPTY_NODE(&m->node))
3152 rb_erase(&m->node, &sctx->pending_dir_moves);
3153 __free_recorded_refs(&m->update_refs);
3154 kfree(m);
3155}
3156
3157static void tail_append_pending_moves(struct pending_dir_move *moves,
3158 struct list_head *stack)
3159{
3160 if (list_empty(&moves->list)) {
3161 list_add_tail(&moves->list, stack);
3162 } else {
3163 LIST_HEAD(list);
3164 list_splice_init(&moves->list, &list);
3165 list_add_tail(&moves->list, stack);
3166 list_splice_tail(&list, stack);
3167 }
3168}
3169
3170static int apply_children_dir_moves(struct send_ctx *sctx)
3171{
3172 struct pending_dir_move *pm;
3173 struct list_head stack;
3174 u64 parent_ino = sctx->cur_ino;
3175 int ret = 0;
3176
3177 pm = get_pending_dir_moves(sctx, parent_ino);
3178 if (!pm)
3179 return 0;
3180
3181 INIT_LIST_HEAD(&stack);
3182 tail_append_pending_moves(pm, &stack);
3183
3184 while (!list_empty(&stack)) {
3185 pm = list_first_entry(&stack, struct pending_dir_move, list);
3186 parent_ino = pm->ino;
3187 ret = apply_dir_move(sctx, pm);
3188 free_pending_move(sctx, pm);
3189 if (ret)
3190 goto out;
3191 pm = get_pending_dir_moves(sctx, parent_ino);
3192 if (pm)
3193 tail_append_pending_moves(pm, &stack);
3194 }
3195 return 0;
3196
3197out:
3198 while (!list_empty(&stack)) {
3199 pm = list_first_entry(&stack, struct pending_dir_move, list);
3200 free_pending_move(sctx, pm);
3201 }
3202 return ret;
3203}
3204
3205static int wait_for_parent_move(struct send_ctx *sctx,
3206 struct recorded_ref *parent_ref)
3207{
3208 int ret;
3209 u64 ino = parent_ref->dir;
3210 u64 parent_ino_before, parent_ino_after;
3211 u64 old_gen;
3212 struct fs_path *path_before = NULL;
3213 struct fs_path *path_after = NULL;
3214 int len1, len2;
3215 int register_upper_dirs;
3216 u64 gen;
3217
3218 if (is_waiting_for_move(sctx, ino))
3219 return 1;
3220
3221 if (parent_ref->dir <= sctx->cur_ino)
3222 return 0;
3223
3224 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3225 NULL, NULL, NULL, NULL);
3226 if (ret == -ENOENT)
3227 return 0;
3228 else if (ret < 0)
3229 return ret;
3230
3231 if (parent_ref->dir_gen != old_gen)
3232 return 0;
3233
3234 path_before = fs_path_alloc();
3235 if (!path_before)
3236 return -ENOMEM;
3237
3238 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3239 NULL, path_before);
3240 if (ret == -ENOENT) {
3241 ret = 0;
3242 goto out;
3243 } else if (ret < 0) {
3244 goto out;
3245 }
3246
3247 path_after = fs_path_alloc();
3248 if (!path_after) {
3249 ret = -ENOMEM;
3250 goto out;
3251 }
3252
3253 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3254 &gen, path_after);
3255 if (ret == -ENOENT) {
3256 ret = 0;
3257 goto out;
3258 } else if (ret < 0) {
3259 goto out;
3260 }
3261
3262 len1 = fs_path_len(path_before);
3263 len2 = fs_path_len(path_after);
3264 if (parent_ino_before != parent_ino_after || len1 != len2 ||
3265 memcmp(path_before->start, path_after->start, len1)) {
3266 ret = 1;
3267 goto out;
3268 }
3269 ret = 0;
3270
3271 /*
3272 * Ok, our new most direct ancestor has a higher inode number but
3273 * wasn't moved/renamed. So maybe some of the new ancestors higher in
3274 * the hierarchy have an higher inode number too *and* were renamed
3275 * or moved - in this case we need to wait for the ancestor's rename
3276 * or move operation before we can do the move/rename for the current
3277 * inode.
3278 */
3279 register_upper_dirs = 0;
3280 ino = parent_ino_after;
3281again:
3282 while ((ret == 0 || register_upper_dirs) && ino > sctx->cur_ino) {
3283 u64 parent_gen;
3284
3285 fs_path_reset(path_before);
3286 fs_path_reset(path_after);
3287
3288 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3289 &parent_gen, path_after);
3290 if (ret < 0)
3291 goto out;
3292 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3293 NULL, path_before);
3294 if (ret == -ENOENT) {
3295 ret = 0;
3296 break;
3297 } else if (ret < 0) {
3298 goto out;
3299 }
3300
3301 len1 = fs_path_len(path_before);
3302 len2 = fs_path_len(path_after);
3303 if (parent_ino_before != parent_ino_after || len1 != len2 ||
3304 memcmp(path_before->start, path_after->start, len1)) {
3305 ret = 1;
3306 if (register_upper_dirs) {
3307 break;
3308 } else {
3309 register_upper_dirs = 1;
3310 ino = parent_ref->dir;
3311 gen = parent_ref->dir_gen;
3312 goto again;
3313 }
3314 } else if (register_upper_dirs) {
3315 ret = add_pending_dir_move(sctx, ino, gen,
3316 parent_ino_after);
3317 if (ret < 0 && ret != -EEXIST)
3318 goto out;
3319 }
3320
3321 ino = parent_ino_after;
3322 gen = parent_gen;
3323 }
3324
3325out:
3326 fs_path_free(path_before);
3327 fs_path_free(path_after);
3328
3329 return ret;
3330}
3331
3332/*
3333 * This does all the move/link/unlink/rmdir magic.
3334 */
3335static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3336{
3337 int ret = 0;
3338 struct recorded_ref *cur;
3339 struct recorded_ref *cur2;
3340 struct list_head check_dirs;
3341 struct fs_path *valid_path = NULL;
3342 u64 ow_inode = 0;
3343 u64 ow_gen;
3344 int did_overwrite = 0;
3345 int is_orphan = 0;
3346 u64 last_dir_ino_rm = 0;
3347
3348verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3349
3350 /*
3351 * This should never happen as the root dir always has the same ref
3352 * which is always '..'
3353 */
3354 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3355 INIT_LIST_HEAD(&check_dirs);
3356
3357 valid_path = fs_path_alloc();
3358 if (!valid_path) {
3359 ret = -ENOMEM;
3360 goto out;
3361 }
3362
3363 /*
3364 * First, check if the first ref of the current inode was overwritten
3365 * before. If yes, we know that the current inode was already orphanized
3366 * and thus use the orphan name. If not, we can use get_cur_path to
3367 * get the path of the first ref as it would like while receiving at
3368 * this point in time.
3369 * New inodes are always orphan at the beginning, so force to use the
3370 * orphan name in this case.
3371 * The first ref is stored in valid_path and will be updated if it
3372 * gets moved around.
3373 */
3374 if (!sctx->cur_inode_new) {
3375 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3376 sctx->cur_inode_gen);
3377 if (ret < 0)
3378 goto out;
3379 if (ret)
3380 did_overwrite = 1;
3381 }
3382 if (sctx->cur_inode_new || did_overwrite) {
3383 ret = gen_unique_name(sctx, sctx->cur_ino,
3384 sctx->cur_inode_gen, valid_path);
3385 if (ret < 0)
3386 goto out;
3387 is_orphan = 1;
3388 } else {
3389 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3390 valid_path);
3391 if (ret < 0)
3392 goto out;
3393 }
3394
3395 list_for_each_entry(cur, &sctx->new_refs, list) {
3396 /*
3397 * We may have refs where the parent directory does not exist
3398 * yet. This happens if the parent directories inum is higher
3399 * the the current inum. To handle this case, we create the
3400 * parent directory out of order. But we need to check if this
3401 * did already happen before due to other refs in the same dir.
3402 */
3403 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3404 if (ret < 0)
3405 goto out;
3406 if (ret == inode_state_will_create) {
3407 ret = 0;
3408 /*
3409 * First check if any of the current inodes refs did
3410 * already create the dir.
3411 */
3412 list_for_each_entry(cur2, &sctx->new_refs, list) {
3413 if (cur == cur2)
3414 break;
3415 if (cur2->dir == cur->dir) {
3416 ret = 1;
3417 break;
3418 }
3419 }
3420
3421 /*
3422 * If that did not happen, check if a previous inode
3423 * did already create the dir.
3424 */
3425 if (!ret)
3426 ret = did_create_dir(sctx, cur->dir);
3427 if (ret < 0)
3428 goto out;
3429 if (!ret) {
3430 ret = send_create_inode(sctx, cur->dir);
3431 if (ret < 0)
3432 goto out;
3433 }
3434 }
3435
3436 /*
3437 * Check if this new ref would overwrite the first ref of
3438 * another unprocessed inode. If yes, orphanize the
3439 * overwritten inode. If we find an overwritten ref that is
3440 * not the first ref, simply unlink it.
3441 */
3442 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3443 cur->name, cur->name_len,
3444 &ow_inode, &ow_gen);
3445 if (ret < 0)
3446 goto out;
3447 if (ret) {
3448 ret = is_first_ref(sctx->parent_root,
3449 ow_inode, cur->dir, cur->name,
3450 cur->name_len);
3451 if (ret < 0)
3452 goto out;
3453 if (ret) {
3454 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3455 cur->full_path);
3456 if (ret < 0)
3457 goto out;
3458 } else {
3459 ret = send_unlink(sctx, cur->full_path);
3460 if (ret < 0)
3461 goto out;
3462 }
3463 }
3464
3465 /*
3466 * link/move the ref to the new place. If we have an orphan
3467 * inode, move it and update valid_path. If not, link or move
3468 * it depending on the inode mode.
3469 */
3470 if (is_orphan) {
3471 ret = send_rename(sctx, valid_path, cur->full_path);
3472 if (ret < 0)
3473 goto out;
3474 is_orphan = 0;
3475 ret = fs_path_copy(valid_path, cur->full_path);
3476 if (ret < 0)
3477 goto out;
3478 } else {
3479 if (S_ISDIR(sctx->cur_inode_mode)) {
3480 /*
3481 * Dirs can't be linked, so move it. For moved
3482 * dirs, we always have one new and one deleted
3483 * ref. The deleted ref is ignored later.
3484 */
3485 ret = wait_for_parent_move(sctx, cur);
3486 if (ret < 0)
3487 goto out;
3488 if (ret) {
3489 ret = add_pending_dir_move(sctx,
3490 sctx->cur_ino,
3491 sctx->cur_inode_gen,
3492 cur->dir);
3493 *pending_move = 1;
3494 } else {
3495 ret = send_rename(sctx, valid_path,
3496 cur->full_path);
3497 if (!ret)
3498 ret = fs_path_copy(valid_path,
3499 cur->full_path);
3500 }
3501 if (ret < 0)
3502 goto out;
3503 } else {
3504 ret = send_link(sctx, cur->full_path,
3505 valid_path);
3506 if (ret < 0)
3507 goto out;
3508 }
3509 }
3510 ret = dup_ref(cur, &check_dirs);
3511 if (ret < 0)
3512 goto out;
3513 }
3514
3515 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3516 /*
3517 * Check if we can already rmdir the directory. If not,
3518 * orphanize it. For every dir item inside that gets deleted
3519 * later, we do this check again and rmdir it then if possible.
3520 * See the use of check_dirs for more details.
3521 */
3522 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3523 sctx->cur_ino);
3524 if (ret < 0)
3525 goto out;
3526 if (ret) {
3527 ret = send_rmdir(sctx, valid_path);
3528 if (ret < 0)
3529 goto out;
3530 } else if (!is_orphan) {
3531 ret = orphanize_inode(sctx, sctx->cur_ino,
3532 sctx->cur_inode_gen, valid_path);
3533 if (ret < 0)
3534 goto out;
3535 is_orphan = 1;
3536 }
3537
3538 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3539 ret = dup_ref(cur, &check_dirs);
3540 if (ret < 0)
3541 goto out;
3542 }
3543 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3544 !list_empty(&sctx->deleted_refs)) {
3545 /*
3546 * We have a moved dir. Add the old parent to check_dirs
3547 */
3548 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3549 list);
3550 ret = dup_ref(cur, &check_dirs);
3551 if (ret < 0)
3552 goto out;
3553 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3554 /*
3555 * We have a non dir inode. Go through all deleted refs and
3556 * unlink them if they were not already overwritten by other
3557 * inodes.
3558 */
3559 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3560 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3561 sctx->cur_ino, sctx->cur_inode_gen,
3562 cur->name, cur->name_len);
3563 if (ret < 0)
3564 goto out;
3565 if (!ret) {
3566 ret = send_unlink(sctx, cur->full_path);
3567 if (ret < 0)
3568 goto out;
3569 }
3570 ret = dup_ref(cur, &check_dirs);
3571 if (ret < 0)
3572 goto out;
3573 }
3574 /*
3575 * If the inode is still orphan, unlink the orphan. This may
3576 * happen when a previous inode did overwrite the first ref
3577 * of this inode and no new refs were added for the current
3578 * inode. Unlinking does not mean that the inode is deleted in
3579 * all cases. There may still be links to this inode in other
3580 * places.
3581 */
3582 if (is_orphan) {
3583 ret = send_unlink(sctx, valid_path);
3584 if (ret < 0)
3585 goto out;
3586 }
3587 }
3588
3589 /*
3590 * We did collect all parent dirs where cur_inode was once located. We
3591 * now go through all these dirs and check if they are pending for
3592 * deletion and if it's finally possible to perform the rmdir now.
3593 * We also update the inode stats of the parent dirs here.
3594 */
3595 list_for_each_entry(cur, &check_dirs, list) {
3596 /*
3597 * In case we had refs into dirs that were not processed yet,
3598 * we don't need to do the utime and rmdir logic for these dirs.
3599 * The dir will be processed later.
3600 */
3601 if (cur->dir > sctx->cur_ino)
3602 continue;
3603
3604 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3605 if (ret < 0)
3606 goto out;
3607
3608 if (ret == inode_state_did_create ||
3609 ret == inode_state_no_change) {
3610 /* TODO delayed utimes */
3611 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3612 if (ret < 0)
3613 goto out;
3614 } else if (ret == inode_state_did_delete &&
3615 cur->dir != last_dir_ino_rm) {
3616 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3617 sctx->cur_ino);
3618 if (ret < 0)
3619 goto out;
3620 if (ret) {
3621 ret = get_cur_path(sctx, cur->dir,
3622 cur->dir_gen, valid_path);
3623 if (ret < 0)
3624 goto out;
3625 ret = send_rmdir(sctx, valid_path);
3626 if (ret < 0)
3627 goto out;
3628 last_dir_ino_rm = cur->dir;
3629 }
3630 }
3631 }
3632
3633 ret = 0;
3634
3635out:
3636 __free_recorded_refs(&check_dirs);
3637 free_recorded_refs(sctx);
3638 fs_path_free(valid_path);
3639 return ret;
3640}
3641
3642static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3643 struct fs_path *name, void *ctx, struct list_head *refs)
3644{
3645 int ret = 0;
3646 struct send_ctx *sctx = ctx;
3647 struct fs_path *p;
3648 u64 gen;
3649
3650 p = fs_path_alloc();
3651 if (!p)
3652 return -ENOMEM;
3653
3654 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
3655 NULL, NULL);
3656 if (ret < 0)
3657 goto out;
3658
3659 ret = get_cur_path(sctx, dir, gen, p);
3660 if (ret < 0)
3661 goto out;
3662 ret = fs_path_add_path(p, name);
3663 if (ret < 0)
3664 goto out;
3665
3666 ret = __record_ref(refs, dir, gen, p);
3667
3668out:
3669 if (ret)
3670 fs_path_free(p);
3671 return ret;
3672}
3673
3674static int __record_new_ref(int num, u64 dir, int index,
3675 struct fs_path *name,
3676 void *ctx)
3677{
3678 struct send_ctx *sctx = ctx;
3679 return record_ref(sctx->send_root, num, dir, index, name,
3680 ctx, &sctx->new_refs);
3681}
3682
3683
3684static int __record_deleted_ref(int num, u64 dir, int index,
3685 struct fs_path *name,
3686 void *ctx)
3687{
3688 struct send_ctx *sctx = ctx;
3689 return record_ref(sctx->parent_root, num, dir, index, name,
3690 ctx, &sctx->deleted_refs);
3691}
3692
3693static int record_new_ref(struct send_ctx *sctx)
3694{
3695 int ret;
3696
3697 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3698 sctx->cmp_key, 0, __record_new_ref, sctx);
3699 if (ret < 0)
3700 goto out;
3701 ret = 0;
3702
3703out:
3704 return ret;
3705}
3706
3707static int record_deleted_ref(struct send_ctx *sctx)
3708{
3709 int ret;
3710
3711 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3712 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3713 if (ret < 0)
3714 goto out;
3715 ret = 0;
3716
3717out:
3718 return ret;
3719}
3720
3721struct find_ref_ctx {
3722 u64 dir;
3723 u64 dir_gen;
3724 struct btrfs_root *root;
3725 struct fs_path *name;
3726 int found_idx;
3727};
3728
3729static int __find_iref(int num, u64 dir, int index,
3730 struct fs_path *name,
3731 void *ctx_)
3732{
3733 struct find_ref_ctx *ctx = ctx_;
3734 u64 dir_gen;
3735 int ret;
3736
3737 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3738 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3739 /*
3740 * To avoid doing extra lookups we'll only do this if everything
3741 * else matches.
3742 */
3743 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3744 NULL, NULL, NULL);
3745 if (ret)
3746 return ret;
3747 if (dir_gen != ctx->dir_gen)
3748 return 0;
3749 ctx->found_idx = num;
3750 return 1;
3751 }
3752 return 0;
3753}
3754
3755static int find_iref(struct btrfs_root *root,
3756 struct btrfs_path *path,
3757 struct btrfs_key *key,
3758 u64 dir, u64 dir_gen, struct fs_path *name)
3759{
3760 int ret;
3761 struct find_ref_ctx ctx;
3762
3763 ctx.dir = dir;
3764 ctx.name = name;
3765 ctx.dir_gen = dir_gen;
3766 ctx.found_idx = -1;
3767 ctx.root = root;
3768
3769 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3770 if (ret < 0)
3771 return ret;
3772
3773 if (ctx.found_idx == -1)
3774 return -ENOENT;
3775
3776 return ctx.found_idx;
3777}
3778
3779static int __record_changed_new_ref(int num, u64 dir, int index,
3780 struct fs_path *name,
3781 void *ctx)
3782{
3783 u64 dir_gen;
3784 int ret;
3785 struct send_ctx *sctx = ctx;
3786
3787 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3788 NULL, NULL, NULL);
3789 if (ret)
3790 return ret;
3791
3792 ret = find_iref(sctx->parent_root, sctx->right_path,
3793 sctx->cmp_key, dir, dir_gen, name);
3794 if (ret == -ENOENT)
3795 ret = __record_new_ref(num, dir, index, name, sctx);
3796 else if (ret > 0)
3797 ret = 0;
3798
3799 return ret;
3800}
3801
3802static int __record_changed_deleted_ref(int num, u64 dir, int index,
3803 struct fs_path *name,
3804 void *ctx)
3805{
3806 u64 dir_gen;
3807 int ret;
3808 struct send_ctx *sctx = ctx;
3809
3810 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3811 NULL, NULL, NULL);
3812 if (ret)
3813 return ret;
3814
3815 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
3816 dir, dir_gen, name);
3817 if (ret == -ENOENT)
3818 ret = __record_deleted_ref(num, dir, index, name, sctx);
3819 else if (ret > 0)
3820 ret = 0;
3821
3822 return ret;
3823}
3824
3825static int record_changed_ref(struct send_ctx *sctx)
3826{
3827 int ret = 0;
3828
3829 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3830 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3831 if (ret < 0)
3832 goto out;
3833 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3834 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3835 if (ret < 0)
3836 goto out;
3837 ret = 0;
3838
3839out:
3840 return ret;
3841}
3842
3843/*
3844 * Record and process all refs at once. Needed when an inode changes the
3845 * generation number, which means that it was deleted and recreated.
3846 */
3847static int process_all_refs(struct send_ctx *sctx,
3848 enum btrfs_compare_tree_result cmd)
3849{
3850 int ret;
3851 struct btrfs_root *root;
3852 struct btrfs_path *path;
3853 struct btrfs_key key;
3854 struct btrfs_key found_key;
3855 struct extent_buffer *eb;
3856 int slot;
3857 iterate_inode_ref_t cb;
3858 int pending_move = 0;
3859
3860 path = alloc_path_for_send();
3861 if (!path)
3862 return -ENOMEM;
3863
3864 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3865 root = sctx->send_root;
3866 cb = __record_new_ref;
3867 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3868 root = sctx->parent_root;
3869 cb = __record_deleted_ref;
3870 } else {
3871 btrfs_err(sctx->send_root->fs_info,
3872 "Wrong command %d in process_all_refs", cmd);
3873 ret = -EINVAL;
3874 goto out;
3875 }
3876
3877 key.objectid = sctx->cmp_key->objectid;
3878 key.type = BTRFS_INODE_REF_KEY;
3879 key.offset = 0;
3880 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3881 if (ret < 0)
3882 goto out;
3883
3884 while (1) {
3885 eb = path->nodes[0];
3886 slot = path->slots[0];
3887 if (slot >= btrfs_header_nritems(eb)) {
3888 ret = btrfs_next_leaf(root, path);
3889 if (ret < 0)
3890 goto out;
3891 else if (ret > 0)
3892 break;
3893 continue;
3894 }
3895
3896 btrfs_item_key_to_cpu(eb, &found_key, slot);
3897
3898 if (found_key.objectid != key.objectid ||
3899 (found_key.type != BTRFS_INODE_REF_KEY &&
3900 found_key.type != BTRFS_INODE_EXTREF_KEY))
3901 break;
3902
3903 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
3904 if (ret < 0)
3905 goto out;
3906
3907 path->slots[0]++;
3908 }
3909 btrfs_release_path(path);
3910
3911 ret = process_recorded_refs(sctx, &pending_move);
3912 /* Only applicable to an incremental send. */
3913 ASSERT(pending_move == 0);
3914
3915out:
3916 btrfs_free_path(path);
3917 return ret;
3918}
3919
3920static int send_set_xattr(struct send_ctx *sctx,
3921 struct fs_path *path,
3922 const char *name, int name_len,
3923 const char *data, int data_len)
3924{
3925 int ret = 0;
3926
3927 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3928 if (ret < 0)
3929 goto out;
3930
3931 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3932 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3933 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3934
3935 ret = send_cmd(sctx);
3936
3937tlv_put_failure:
3938out:
3939 return ret;
3940}
3941
3942static int send_remove_xattr(struct send_ctx *sctx,
3943 struct fs_path *path,
3944 const char *name, int name_len)
3945{
3946 int ret = 0;
3947
3948 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3949 if (ret < 0)
3950 goto out;
3951
3952 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3953 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3954
3955 ret = send_cmd(sctx);
3956
3957tlv_put_failure:
3958out:
3959 return ret;
3960}
3961
3962static int __process_new_xattr(int num, struct btrfs_key *di_key,
3963 const char *name, int name_len,
3964 const char *data, int data_len,
3965 u8 type, void *ctx)
3966{
3967 int ret;
3968 struct send_ctx *sctx = ctx;
3969 struct fs_path *p;
3970 posix_acl_xattr_header dummy_acl;
3971
3972 p = fs_path_alloc();
3973 if (!p)
3974 return -ENOMEM;
3975
3976 /*
3977 * This hack is needed because empty acl's are stored as zero byte
3978 * data in xattrs. Problem with that is, that receiving these zero byte
3979 * acl's will fail later. To fix this, we send a dummy acl list that
3980 * only contains the version number and no entries.
3981 */
3982 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3983 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3984 if (data_len == 0) {
3985 dummy_acl.a_version =
3986 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3987 data = (char *)&dummy_acl;
3988 data_len = sizeof(dummy_acl);
3989 }
3990 }
3991
3992 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3993 if (ret < 0)
3994 goto out;
3995
3996 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3997
3998out:
3999 fs_path_free(p);
4000 return ret;
4001}
4002
4003static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4004 const char *name, int name_len,
4005 const char *data, int data_len,
4006 u8 type, void *ctx)
4007{
4008 int ret;
4009 struct send_ctx *sctx = ctx;
4010 struct fs_path *p;
4011
4012 p = fs_path_alloc();
4013 if (!p)
4014 return -ENOMEM;
4015
4016 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4017 if (ret < 0)
4018 goto out;
4019
4020 ret = send_remove_xattr(sctx, p, name, name_len);
4021
4022out:
4023 fs_path_free(p);
4024 return ret;
4025}
4026
4027static int process_new_xattr(struct send_ctx *sctx)
4028{
4029 int ret = 0;
4030
4031 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4032 sctx->cmp_key, __process_new_xattr, sctx);
4033
4034 return ret;
4035}
4036
4037static int process_deleted_xattr(struct send_ctx *sctx)
4038{
4039 int ret;
4040
4041 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4042 sctx->cmp_key, __process_deleted_xattr, sctx);
4043
4044 return ret;
4045}
4046
4047struct find_xattr_ctx {
4048 const char *name;
4049 int name_len;
4050 int found_idx;
4051 char *found_data;
4052 int found_data_len;
4053};
4054
4055static int __find_xattr(int num, struct btrfs_key *di_key,
4056 const char *name, int name_len,
4057 const char *data, int data_len,
4058 u8 type, void *vctx)
4059{
4060 struct find_xattr_ctx *ctx = vctx;
4061
4062 if (name_len == ctx->name_len &&
4063 strncmp(name, ctx->name, name_len) == 0) {
4064 ctx->found_idx = num;
4065 ctx->found_data_len = data_len;
4066 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
4067 if (!ctx->found_data)
4068 return -ENOMEM;
4069 return 1;
4070 }
4071 return 0;
4072}
4073
4074static int find_xattr(struct btrfs_root *root,
4075 struct btrfs_path *path,
4076 struct btrfs_key *key,
4077 const char *name, int name_len,
4078 char **data, int *data_len)
4079{
4080 int ret;
4081 struct find_xattr_ctx ctx;
4082
4083 ctx.name = name;
4084 ctx.name_len = name_len;
4085 ctx.found_idx = -1;
4086 ctx.found_data = NULL;
4087 ctx.found_data_len = 0;
4088
4089 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
4090 if (ret < 0)
4091 return ret;
4092
4093 if (ctx.found_idx == -1)
4094 return -ENOENT;
4095 if (data) {
4096 *data = ctx.found_data;
4097 *data_len = ctx.found_data_len;
4098 } else {
4099 kfree(ctx.found_data);
4100 }
4101 return ctx.found_idx;
4102}
4103
4104
4105static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4106 const char *name, int name_len,
4107 const char *data, int data_len,
4108 u8 type, void *ctx)
4109{
4110 int ret;
4111 struct send_ctx *sctx = ctx;
4112 char *found_data = NULL;
4113 int found_data_len = 0;
4114
4115 ret = find_xattr(sctx->parent_root, sctx->right_path,
4116 sctx->cmp_key, name, name_len, &found_data,
4117 &found_data_len);
4118 if (ret == -ENOENT) {
4119 ret = __process_new_xattr(num, di_key, name, name_len, data,
4120 data_len, type, ctx);
4121 } else if (ret >= 0) {
4122 if (data_len != found_data_len ||
4123 memcmp(data, found_data, data_len)) {
4124 ret = __process_new_xattr(num, di_key, name, name_len,
4125 data, data_len, type, ctx);
4126 } else {
4127 ret = 0;
4128 }
4129 }
4130
4131 kfree(found_data);
4132 return ret;
4133}
4134
4135static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4136 const char *name, int name_len,
4137 const char *data, int data_len,
4138 u8 type, void *ctx)
4139{
4140 int ret;
4141 struct send_ctx *sctx = ctx;
4142
4143 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4144 name, name_len, NULL, NULL);
4145 if (ret == -ENOENT)
4146 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4147 data_len, type, ctx);
4148 else if (ret >= 0)
4149 ret = 0;
4150
4151 return ret;
4152}
4153
4154static int process_changed_xattr(struct send_ctx *sctx)
4155{
4156 int ret = 0;
4157
4158 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4159 sctx->cmp_key, __process_changed_new_xattr, sctx);
4160 if (ret < 0)
4161 goto out;
4162 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4163 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4164
4165out:
4166 return ret;
4167}
4168
4169static int process_all_new_xattrs(struct send_ctx *sctx)
4170{
4171 int ret;
4172 struct btrfs_root *root;
4173 struct btrfs_path *path;
4174 struct btrfs_key key;
4175 struct btrfs_key found_key;
4176 struct extent_buffer *eb;
4177 int slot;
4178
4179 path = alloc_path_for_send();
4180 if (!path)
4181 return -ENOMEM;
4182
4183 root = sctx->send_root;
4184
4185 key.objectid = sctx->cmp_key->objectid;
4186 key.type = BTRFS_XATTR_ITEM_KEY;
4187 key.offset = 0;
4188 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4189 if (ret < 0)
4190 goto out;
4191
4192 while (1) {
4193 eb = path->nodes[0];
4194 slot = path->slots[0];
4195 if (slot >= btrfs_header_nritems(eb)) {
4196 ret = btrfs_next_leaf(root, path);
4197 if (ret < 0) {
4198 goto out;
4199 } else if (ret > 0) {
4200 ret = 0;
4201 break;
4202 }
4203 continue;
4204 }
4205
4206 btrfs_item_key_to_cpu(eb, &found_key, slot);
4207 if (found_key.objectid != key.objectid ||
4208 found_key.type != key.type) {
4209 ret = 0;
4210 goto out;
4211 }
4212
4213 ret = iterate_dir_item(root, path, &found_key,
4214 __process_new_xattr, sctx);
4215 if (ret < 0)
4216 goto out;
4217
4218 path->slots[0]++;
4219 }
4220
4221out:
4222 btrfs_free_path(path);
4223 return ret;
4224}
4225
4226static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4227{
4228 struct btrfs_root *root = sctx->send_root;
4229 struct btrfs_fs_info *fs_info = root->fs_info;
4230 struct inode *inode;
4231 struct page *page;
4232 char *addr;
4233 struct btrfs_key key;
4234 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
4235 pgoff_t last_index;
4236 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
4237 ssize_t ret = 0;
4238
4239 key.objectid = sctx->cur_ino;
4240 key.type = BTRFS_INODE_ITEM_KEY;
4241 key.offset = 0;
4242
4243 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4244 if (IS_ERR(inode))
4245 return PTR_ERR(inode);
4246
4247 if (offset + len > i_size_read(inode)) {
4248 if (offset > i_size_read(inode))
4249 len = 0;
4250 else
4251 len = offset - i_size_read(inode);
4252 }
4253 if (len == 0)
4254 goto out;
4255
4256 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
4257
4258 /* initial readahead */
4259 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4260 file_ra_state_init(&sctx->ra, inode->i_mapping);
4261 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4262 last_index - index + 1);
4263
4264 while (index <= last_index) {
4265 unsigned cur_len = min_t(unsigned, len,
4266 PAGE_CACHE_SIZE - pg_offset);
4267 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4268 if (!page) {
4269 ret = -ENOMEM;
4270 break;
4271 }
4272
4273 if (!PageUptodate(page)) {
4274 btrfs_readpage(NULL, page);
4275 lock_page(page);
4276 if (!PageUptodate(page)) {
4277 unlock_page(page);
4278 page_cache_release(page);
4279 ret = -EIO;
4280 break;
4281 }
4282 }
4283
4284 addr = kmap(page);
4285 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4286 kunmap(page);
4287 unlock_page(page);
4288 page_cache_release(page);
4289 index++;
4290 pg_offset = 0;
4291 len -= cur_len;
4292 ret += cur_len;
4293 }
4294out:
4295 iput(inode);
4296 return ret;
4297}
4298
4299/*
4300 * Read some bytes from the current inode/file and send a write command to
4301 * user space.
4302 */
4303static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4304{
4305 int ret = 0;
4306 struct fs_path *p;
4307 ssize_t num_read = 0;
4308
4309 p = fs_path_alloc();
4310 if (!p)
4311 return -ENOMEM;
4312
4313verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4314
4315 num_read = fill_read_buf(sctx, offset, len);
4316 if (num_read <= 0) {
4317 if (num_read < 0)
4318 ret = num_read;
4319 goto out;
4320 }
4321
4322 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4323 if (ret < 0)
4324 goto out;
4325
4326 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4327 if (ret < 0)
4328 goto out;
4329
4330 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4331 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4332 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4333
4334 ret = send_cmd(sctx);
4335
4336tlv_put_failure:
4337out:
4338 fs_path_free(p);
4339 if (ret < 0)
4340 return ret;
4341 return num_read;
4342}
4343
4344/*
4345 * Send a clone command to user space.
4346 */
4347static int send_clone(struct send_ctx *sctx,
4348 u64 offset, u32 len,
4349 struct clone_root *clone_root)
4350{
4351 int ret = 0;
4352 struct fs_path *p;
4353 u64 gen;
4354
4355verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4356 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4357 clone_root->root->objectid, clone_root->ino,
4358 clone_root->offset);
4359
4360 p = fs_path_alloc();
4361 if (!p)
4362 return -ENOMEM;
4363
4364 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4365 if (ret < 0)
4366 goto out;
4367
4368 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4369 if (ret < 0)
4370 goto out;
4371
4372 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4373 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4374 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4375
4376 if (clone_root->root == sctx->send_root) {
4377 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4378 &gen, NULL, NULL, NULL, NULL);
4379 if (ret < 0)
4380 goto out;
4381 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4382 } else {
4383 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4384 }
4385 if (ret < 0)
4386 goto out;
4387
4388 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4389 clone_root->root->root_item.uuid);
4390 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4391 le64_to_cpu(clone_root->root->root_item.ctransid));
4392 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4393 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4394 clone_root->offset);
4395
4396 ret = send_cmd(sctx);
4397
4398tlv_put_failure:
4399out:
4400 fs_path_free(p);
4401 return ret;
4402}
4403
4404/*
4405 * Send an update extent command to user space.
4406 */
4407static int send_update_extent(struct send_ctx *sctx,
4408 u64 offset, u32 len)
4409{
4410 int ret = 0;
4411 struct fs_path *p;
4412
4413 p = fs_path_alloc();
4414 if (!p)
4415 return -ENOMEM;
4416
4417 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4418 if (ret < 0)
4419 goto out;
4420
4421 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4422 if (ret < 0)
4423 goto out;
4424
4425 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4426 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4427 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4428
4429 ret = send_cmd(sctx);
4430
4431tlv_put_failure:
4432out:
4433 fs_path_free(p);
4434 return ret;
4435}
4436
4437static int send_hole(struct send_ctx *sctx, u64 end)
4438{
4439 struct fs_path *p = NULL;
4440 u64 offset = sctx->cur_inode_last_extent;
4441 u64 len;
4442 int ret = 0;
4443
4444 p = fs_path_alloc();
4445 if (!p)
4446 return -ENOMEM;
4447 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4448 if (ret < 0)
4449 goto tlv_put_failure;
4450 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4451 while (offset < end) {
4452 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4453
4454 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4455 if (ret < 0)
4456 break;
4457 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4458 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4459 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4460 ret = send_cmd(sctx);
4461 if (ret < 0)
4462 break;
4463 offset += len;
4464 }
4465tlv_put_failure:
4466 fs_path_free(p);
4467 return ret;
4468}
4469
4470static int send_write_or_clone(struct send_ctx *sctx,
4471 struct btrfs_path *path,
4472 struct btrfs_key *key,
4473 struct clone_root *clone_root)
4474{
4475 int ret = 0;
4476 struct btrfs_file_extent_item *ei;
4477 u64 offset = key->offset;
4478 u64 pos = 0;
4479 u64 len;
4480 u32 l;
4481 u8 type;
4482 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4483
4484 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4485 struct btrfs_file_extent_item);
4486 type = btrfs_file_extent_type(path->nodes[0], ei);
4487 if (type == BTRFS_FILE_EXTENT_INLINE) {
4488 len = btrfs_file_extent_inline_len(path->nodes[0],
4489 path->slots[0], ei);
4490 /*
4491 * it is possible the inline item won't cover the whole page,
4492 * but there may be items after this page. Make
4493 * sure to send the whole thing
4494 */
4495 len = PAGE_CACHE_ALIGN(len);
4496 } else {
4497 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4498 }
4499
4500 if (offset + len > sctx->cur_inode_size)
4501 len = sctx->cur_inode_size - offset;
4502 if (len == 0) {
4503 ret = 0;
4504 goto out;
4505 }
4506
4507 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4508 ret = send_clone(sctx, offset, len, clone_root);
4509 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4510 ret = send_update_extent(sctx, offset, len);
4511 } else {
4512 while (pos < len) {
4513 l = len - pos;
4514 if (l > BTRFS_SEND_READ_SIZE)
4515 l = BTRFS_SEND_READ_SIZE;
4516 ret = send_write(sctx, pos + offset, l);
4517 if (ret < 0)
4518 goto out;
4519 if (!ret)
4520 break;
4521 pos += ret;
4522 }
4523 ret = 0;
4524 }
4525out:
4526 return ret;
4527}
4528
4529static int is_extent_unchanged(struct send_ctx *sctx,
4530 struct btrfs_path *left_path,
4531 struct btrfs_key *ekey)
4532{
4533 int ret = 0;
4534 struct btrfs_key key;
4535 struct btrfs_path *path = NULL;
4536 struct extent_buffer *eb;
4537 int slot;
4538 struct btrfs_key found_key;
4539 struct btrfs_file_extent_item *ei;
4540 u64 left_disknr;
4541 u64 right_disknr;
4542 u64 left_offset;
4543 u64 right_offset;
4544 u64 left_offset_fixed;
4545 u64 left_len;
4546 u64 right_len;
4547 u64 left_gen;
4548 u64 right_gen;
4549 u8 left_type;
4550 u8 right_type;
4551
4552 path = alloc_path_for_send();
4553 if (!path)
4554 return -ENOMEM;
4555
4556 eb = left_path->nodes[0];
4557 slot = left_path->slots[0];
4558 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4559 left_type = btrfs_file_extent_type(eb, ei);
4560
4561 if (left_type != BTRFS_FILE_EXTENT_REG) {
4562 ret = 0;
4563 goto out;
4564 }
4565 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4566 left_len = btrfs_file_extent_num_bytes(eb, ei);
4567 left_offset = btrfs_file_extent_offset(eb, ei);
4568 left_gen = btrfs_file_extent_generation(eb, ei);
4569
4570 /*
4571 * Following comments will refer to these graphics. L is the left
4572 * extents which we are checking at the moment. 1-8 are the right
4573 * extents that we iterate.
4574 *
4575 * |-----L-----|
4576 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4577 *
4578 * |-----L-----|
4579 * |--1--|-2b-|...(same as above)
4580 *
4581 * Alternative situation. Happens on files where extents got split.
4582 * |-----L-----|
4583 * |-----------7-----------|-6-|
4584 *
4585 * Alternative situation. Happens on files which got larger.
4586 * |-----L-----|
4587 * |-8-|
4588 * Nothing follows after 8.
4589 */
4590
4591 key.objectid = ekey->objectid;
4592 key.type = BTRFS_EXTENT_DATA_KEY;
4593 key.offset = ekey->offset;
4594 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4595 if (ret < 0)
4596 goto out;
4597 if (ret) {
4598 ret = 0;
4599 goto out;
4600 }
4601
4602 /*
4603 * Handle special case where the right side has no extents at all.
4604 */
4605 eb = path->nodes[0];
4606 slot = path->slots[0];
4607 btrfs_item_key_to_cpu(eb, &found_key, slot);
4608 if (found_key.objectid != key.objectid ||
4609 found_key.type != key.type) {
4610 /* If we're a hole then just pretend nothing changed */
4611 ret = (left_disknr) ? 0 : 1;
4612 goto out;
4613 }
4614
4615 /*
4616 * We're now on 2a, 2b or 7.
4617 */
4618 key = found_key;
4619 while (key.offset < ekey->offset + left_len) {
4620 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4621 right_type = btrfs_file_extent_type(eb, ei);
4622 if (right_type != BTRFS_FILE_EXTENT_REG) {
4623 ret = 0;
4624 goto out;
4625 }
4626
4627 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4628 right_len = btrfs_file_extent_num_bytes(eb, ei);
4629 right_offset = btrfs_file_extent_offset(eb, ei);
4630 right_gen = btrfs_file_extent_generation(eb, ei);
4631
4632 /*
4633 * Are we at extent 8? If yes, we know the extent is changed.
4634 * This may only happen on the first iteration.
4635 */
4636 if (found_key.offset + right_len <= ekey->offset) {
4637 /* If we're a hole just pretend nothing changed */
4638 ret = (left_disknr) ? 0 : 1;
4639 goto out;
4640 }
4641
4642 left_offset_fixed = left_offset;
4643 if (key.offset < ekey->offset) {
4644 /* Fix the right offset for 2a and 7. */
4645 right_offset += ekey->offset - key.offset;
4646 } else {
4647 /* Fix the left offset for all behind 2a and 2b */
4648 left_offset_fixed += key.offset - ekey->offset;
4649 }
4650
4651 /*
4652 * Check if we have the same extent.
4653 */
4654 if (left_disknr != right_disknr ||
4655 left_offset_fixed != right_offset ||
4656 left_gen != right_gen) {
4657 ret = 0;
4658 goto out;
4659 }
4660
4661 /*
4662 * Go to the next extent.
4663 */
4664 ret = btrfs_next_item(sctx->parent_root, path);
4665 if (ret < 0)
4666 goto out;
4667 if (!ret) {
4668 eb = path->nodes[0];
4669 slot = path->slots[0];
4670 btrfs_item_key_to_cpu(eb, &found_key, slot);
4671 }
4672 if (ret || found_key.objectid != key.objectid ||
4673 found_key.type != key.type) {
4674 key.offset += right_len;
4675 break;
4676 }
4677 if (found_key.offset != key.offset + right_len) {
4678 ret = 0;
4679 goto out;
4680 }
4681 key = found_key;
4682 }
4683
4684 /*
4685 * We're now behind the left extent (treat as unchanged) or at the end
4686 * of the right side (treat as changed).
4687 */
4688 if (key.offset >= ekey->offset + left_len)
4689 ret = 1;
4690 else
4691 ret = 0;
4692
4693
4694out:
4695 btrfs_free_path(path);
4696 return ret;
4697}
4698
4699static int get_last_extent(struct send_ctx *sctx, u64 offset)
4700{
4701 struct btrfs_path *path;
4702 struct btrfs_root *root = sctx->send_root;
4703 struct btrfs_file_extent_item *fi;
4704 struct btrfs_key key;
4705 u64 extent_end;
4706 u8 type;
4707 int ret;
4708
4709 path = alloc_path_for_send();
4710 if (!path)
4711 return -ENOMEM;
4712
4713 sctx->cur_inode_last_extent = 0;
4714
4715 key.objectid = sctx->cur_ino;
4716 key.type = BTRFS_EXTENT_DATA_KEY;
4717 key.offset = offset;
4718 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4719 if (ret < 0)
4720 goto out;
4721 ret = 0;
4722 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4723 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4724 goto out;
4725
4726 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4727 struct btrfs_file_extent_item);
4728 type = btrfs_file_extent_type(path->nodes[0], fi);
4729 if (type == BTRFS_FILE_EXTENT_INLINE) {
4730 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4731 path->slots[0], fi);
4732 extent_end = ALIGN(key.offset + size,
4733 sctx->send_root->sectorsize);
4734 } else {
4735 extent_end = key.offset +
4736 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4737 }
4738 sctx->cur_inode_last_extent = extent_end;
4739out:
4740 btrfs_free_path(path);
4741 return ret;
4742}
4743
4744static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4745 struct btrfs_key *key)
4746{
4747 struct btrfs_file_extent_item *fi;
4748 u64 extent_end;
4749 u8 type;
4750 int ret = 0;
4751
4752 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4753 return 0;
4754
4755 if (sctx->cur_inode_last_extent == (u64)-1) {
4756 ret = get_last_extent(sctx, key->offset - 1);
4757 if (ret)
4758 return ret;
4759 }
4760
4761 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4762 struct btrfs_file_extent_item);
4763 type = btrfs_file_extent_type(path->nodes[0], fi);
4764 if (type == BTRFS_FILE_EXTENT_INLINE) {
4765 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4766 path->slots[0], fi);
4767 extent_end = ALIGN(key->offset + size,
4768 sctx->send_root->sectorsize);
4769 } else {
4770 extent_end = key->offset +
4771 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4772 }
4773
4774 if (path->slots[0] == 0 &&
4775 sctx->cur_inode_last_extent < key->offset) {
4776 /*
4777 * We might have skipped entire leafs that contained only
4778 * file extent items for our current inode. These leafs have
4779 * a generation number smaller (older) than the one in the
4780 * current leaf and the leaf our last extent came from, and
4781 * are located between these 2 leafs.
4782 */
4783 ret = get_last_extent(sctx, key->offset - 1);
4784 if (ret)
4785 return ret;
4786 }
4787
4788 if (sctx->cur_inode_last_extent < key->offset)
4789 ret = send_hole(sctx, key->offset);
4790 sctx->cur_inode_last_extent = extent_end;
4791 return ret;
4792}
4793
4794static int process_extent(struct send_ctx *sctx,
4795 struct btrfs_path *path,
4796 struct btrfs_key *key)
4797{
4798 struct clone_root *found_clone = NULL;
4799 int ret = 0;
4800
4801 if (S_ISLNK(sctx->cur_inode_mode))
4802 return 0;
4803
4804 if (sctx->parent_root && !sctx->cur_inode_new) {
4805 ret = is_extent_unchanged(sctx, path, key);
4806 if (ret < 0)
4807 goto out;
4808 if (ret) {
4809 ret = 0;
4810 goto out_hole;
4811 }
4812 } else {
4813 struct btrfs_file_extent_item *ei;
4814 u8 type;
4815
4816 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4817 struct btrfs_file_extent_item);
4818 type = btrfs_file_extent_type(path->nodes[0], ei);
4819 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4820 type == BTRFS_FILE_EXTENT_REG) {
4821 /*
4822 * The send spec does not have a prealloc command yet,
4823 * so just leave a hole for prealloc'ed extents until
4824 * we have enough commands queued up to justify rev'ing
4825 * the send spec.
4826 */
4827 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4828 ret = 0;
4829 goto out;
4830 }
4831
4832 /* Have a hole, just skip it. */
4833 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4834 ret = 0;
4835 goto out;
4836 }
4837 }
4838 }
4839
4840 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4841 sctx->cur_inode_size, &found_clone);
4842 if (ret != -ENOENT && ret < 0)
4843 goto out;
4844
4845 ret = send_write_or_clone(sctx, path, key, found_clone);
4846 if (ret)
4847 goto out;
4848out_hole:
4849 ret = maybe_send_hole(sctx, path, key);
4850out:
4851 return ret;
4852}
4853
4854static int process_all_extents(struct send_ctx *sctx)
4855{
4856 int ret;
4857 struct btrfs_root *root;
4858 struct btrfs_path *path;
4859 struct btrfs_key key;
4860 struct btrfs_key found_key;
4861 struct extent_buffer *eb;
4862 int slot;
4863
4864 root = sctx->send_root;
4865 path = alloc_path_for_send();
4866 if (!path)
4867 return -ENOMEM;
4868
4869 key.objectid = sctx->cmp_key->objectid;
4870 key.type = BTRFS_EXTENT_DATA_KEY;
4871 key.offset = 0;
4872 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4873 if (ret < 0)
4874 goto out;
4875
4876 while (1) {
4877 eb = path->nodes[0];
4878 slot = path->slots[0];
4879
4880 if (slot >= btrfs_header_nritems(eb)) {
4881 ret = btrfs_next_leaf(root, path);
4882 if (ret < 0) {
4883 goto out;
4884 } else if (ret > 0) {
4885 ret = 0;
4886 break;
4887 }
4888 continue;
4889 }
4890
4891 btrfs_item_key_to_cpu(eb, &found_key, slot);
4892
4893 if (found_key.objectid != key.objectid ||
4894 found_key.type != key.type) {
4895 ret = 0;
4896 goto out;
4897 }
4898
4899 ret = process_extent(sctx, path, &found_key);
4900 if (ret < 0)
4901 goto out;
4902
4903 path->slots[0]++;
4904 }
4905
4906out:
4907 btrfs_free_path(path);
4908 return ret;
4909}
4910
4911static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4912 int *pending_move,
4913 int *refs_processed)
4914{
4915 int ret = 0;
4916
4917 if (sctx->cur_ino == 0)
4918 goto out;
4919 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4920 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4921 goto out;
4922 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4923 goto out;
4924
4925 ret = process_recorded_refs(sctx, pending_move);
4926 if (ret < 0)
4927 goto out;
4928
4929 *refs_processed = 1;
4930out:
4931 return ret;
4932}
4933
4934static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4935{
4936 int ret = 0;
4937 u64 left_mode;
4938 u64 left_uid;
4939 u64 left_gid;
4940 u64 right_mode;
4941 u64 right_uid;
4942 u64 right_gid;
4943 int need_chmod = 0;
4944 int need_chown = 0;
4945 int pending_move = 0;
4946 int refs_processed = 0;
4947
4948 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4949 &refs_processed);
4950 if (ret < 0)
4951 goto out;
4952
4953 /*
4954 * We have processed the refs and thus need to advance send_progress.
4955 * Now, calls to get_cur_xxx will take the updated refs of the current
4956 * inode into account.
4957 *
4958 * On the other hand, if our current inode is a directory and couldn't
4959 * be moved/renamed because its parent was renamed/moved too and it has
4960 * a higher inode number, we can only move/rename our current inode
4961 * after we moved/renamed its parent. Therefore in this case operate on
4962 * the old path (pre move/rename) of our current inode, and the
4963 * move/rename will be performed later.
4964 */
4965 if (refs_processed && !pending_move)
4966 sctx->send_progress = sctx->cur_ino + 1;
4967
4968 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4969 goto out;
4970 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4971 goto out;
4972
4973 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4974 &left_mode, &left_uid, &left_gid, NULL);
4975 if (ret < 0)
4976 goto out;
4977
4978 if (!sctx->parent_root || sctx->cur_inode_new) {
4979 need_chown = 1;
4980 if (!S_ISLNK(sctx->cur_inode_mode))
4981 need_chmod = 1;
4982 } else {
4983 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4984 NULL, NULL, &right_mode, &right_uid,
4985 &right_gid, NULL);
4986 if (ret < 0)
4987 goto out;
4988
4989 if (left_uid != right_uid || left_gid != right_gid)
4990 need_chown = 1;
4991 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4992 need_chmod = 1;
4993 }
4994
4995 if (S_ISREG(sctx->cur_inode_mode)) {
4996 if (need_send_hole(sctx)) {
4997 if (sctx->cur_inode_last_extent == (u64)-1 ||
4998 sctx->cur_inode_last_extent <
4999 sctx->cur_inode_size) {
5000 ret = get_last_extent(sctx, (u64)-1);
5001 if (ret)
5002 goto out;
5003 }
5004 if (sctx->cur_inode_last_extent <
5005 sctx->cur_inode_size) {
5006 ret = send_hole(sctx, sctx->cur_inode_size);
5007 if (ret)
5008 goto out;
5009 }
5010 }
5011 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5012 sctx->cur_inode_size);
5013 if (ret < 0)
5014 goto out;
5015 }
5016
5017 if (need_chown) {
5018 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5019 left_uid, left_gid);
5020 if (ret < 0)
5021 goto out;
5022 }
5023 if (need_chmod) {
5024 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5025 left_mode);
5026 if (ret < 0)
5027 goto out;
5028 }
5029
5030 /*
5031 * If other directory inodes depended on our current directory
5032 * inode's move/rename, now do their move/rename operations.
5033 */
5034 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5035 ret = apply_children_dir_moves(sctx);
5036 if (ret)
5037 goto out;
5038 /*
5039 * Need to send that every time, no matter if it actually
5040 * changed between the two trees as we have done changes to
5041 * the inode before. If our inode is a directory and it's
5042 * waiting to be moved/renamed, we will send its utimes when
5043 * it's moved/renamed, therefore we don't need to do it here.
5044 */
5045 sctx->send_progress = sctx->cur_ino + 1;
5046 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5047 if (ret < 0)
5048 goto out;
5049 }
5050
5051out:
5052 return ret;
5053}
5054
5055static int changed_inode(struct send_ctx *sctx,
5056 enum btrfs_compare_tree_result result)
5057{
5058 int ret = 0;
5059 struct btrfs_key *key = sctx->cmp_key;
5060 struct btrfs_inode_item *left_ii = NULL;
5061 struct btrfs_inode_item *right_ii = NULL;
5062 u64 left_gen = 0;
5063 u64 right_gen = 0;
5064
5065 sctx->cur_ino = key->objectid;
5066 sctx->cur_inode_new_gen = 0;
5067 sctx->cur_inode_last_extent = (u64)-1;
5068
5069 /*
5070 * Set send_progress to current inode. This will tell all get_cur_xxx
5071 * functions that the current inode's refs are not updated yet. Later,
5072 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5073 */
5074 sctx->send_progress = sctx->cur_ino;
5075
5076 if (result == BTRFS_COMPARE_TREE_NEW ||
5077 result == BTRFS_COMPARE_TREE_CHANGED) {
5078 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5079 sctx->left_path->slots[0],
5080 struct btrfs_inode_item);
5081 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5082 left_ii);
5083 } else {
5084 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5085 sctx->right_path->slots[0],
5086 struct btrfs_inode_item);
5087 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5088 right_ii);
5089 }
5090 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5091 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5092 sctx->right_path->slots[0],
5093 struct btrfs_inode_item);
5094
5095 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5096 right_ii);
5097
5098 /*
5099 * The cur_ino = root dir case is special here. We can't treat
5100 * the inode as deleted+reused because it would generate a
5101 * stream that tries to delete/mkdir the root dir.
5102 */
5103 if (left_gen != right_gen &&
5104 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5105 sctx->cur_inode_new_gen = 1;
5106 }
5107
5108 if (result == BTRFS_COMPARE_TREE_NEW) {
5109 sctx->cur_inode_gen = left_gen;
5110 sctx->cur_inode_new = 1;
5111 sctx->cur_inode_deleted = 0;
5112 sctx->cur_inode_size = btrfs_inode_size(
5113 sctx->left_path->nodes[0], left_ii);
5114 sctx->cur_inode_mode = btrfs_inode_mode(
5115 sctx->left_path->nodes[0], left_ii);
5116 sctx->cur_inode_rdev = btrfs_inode_rdev(
5117 sctx->left_path->nodes[0], left_ii);
5118 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5119 ret = send_create_inode_if_needed(sctx);
5120 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5121 sctx->cur_inode_gen = right_gen;
5122 sctx->cur_inode_new = 0;
5123 sctx->cur_inode_deleted = 1;
5124 sctx->cur_inode_size = btrfs_inode_size(
5125 sctx->right_path->nodes[0], right_ii);
5126 sctx->cur_inode_mode = btrfs_inode_mode(
5127 sctx->right_path->nodes[0], right_ii);
5128 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5129 /*
5130 * We need to do some special handling in case the inode was
5131 * reported as changed with a changed generation number. This
5132 * means that the original inode was deleted and new inode
5133 * reused the same inum. So we have to treat the old inode as
5134 * deleted and the new one as new.
5135 */
5136 if (sctx->cur_inode_new_gen) {
5137 /*
5138 * First, process the inode as if it was deleted.
5139 */
5140 sctx->cur_inode_gen = right_gen;
5141 sctx->cur_inode_new = 0;
5142 sctx->cur_inode_deleted = 1;
5143 sctx->cur_inode_size = btrfs_inode_size(
5144 sctx->right_path->nodes[0], right_ii);
5145 sctx->cur_inode_mode = btrfs_inode_mode(
5146 sctx->right_path->nodes[0], right_ii);
5147 ret = process_all_refs(sctx,
5148 BTRFS_COMPARE_TREE_DELETED);
5149 if (ret < 0)
5150 goto out;
5151
5152 /*
5153 * Now process the inode as if it was new.
5154 */
5155 sctx->cur_inode_gen = left_gen;
5156 sctx->cur_inode_new = 1;
5157 sctx->cur_inode_deleted = 0;
5158 sctx->cur_inode_size = btrfs_inode_size(
5159 sctx->left_path->nodes[0], left_ii);
5160 sctx->cur_inode_mode = btrfs_inode_mode(
5161 sctx->left_path->nodes[0], left_ii);
5162 sctx->cur_inode_rdev = btrfs_inode_rdev(
5163 sctx->left_path->nodes[0], left_ii);
5164 ret = send_create_inode_if_needed(sctx);
5165 if (ret < 0)
5166 goto out;
5167
5168 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5169 if (ret < 0)
5170 goto out;
5171 /*
5172 * Advance send_progress now as we did not get into
5173 * process_recorded_refs_if_needed in the new_gen case.
5174 */
5175 sctx->send_progress = sctx->cur_ino + 1;
5176
5177 /*
5178 * Now process all extents and xattrs of the inode as if
5179 * they were all new.
5180 */
5181 ret = process_all_extents(sctx);
5182 if (ret < 0)
5183 goto out;
5184 ret = process_all_new_xattrs(sctx);
5185 if (ret < 0)
5186 goto out;
5187 } else {
5188 sctx->cur_inode_gen = left_gen;
5189 sctx->cur_inode_new = 0;
5190 sctx->cur_inode_new_gen = 0;
5191 sctx->cur_inode_deleted = 0;
5192 sctx->cur_inode_size = btrfs_inode_size(
5193 sctx->left_path->nodes[0], left_ii);
5194 sctx->cur_inode_mode = btrfs_inode_mode(
5195 sctx->left_path->nodes[0], left_ii);
5196 }
5197 }
5198
5199out:
5200 return ret;
5201}
5202
5203/*
5204 * We have to process new refs before deleted refs, but compare_trees gives us
5205 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5206 * first and later process them in process_recorded_refs.
5207 * For the cur_inode_new_gen case, we skip recording completely because
5208 * changed_inode did already initiate processing of refs. The reason for this is
5209 * that in this case, compare_tree actually compares the refs of 2 different
5210 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5211 * refs of the right tree as deleted and all refs of the left tree as new.
5212 */
5213static int changed_ref(struct send_ctx *sctx,
5214 enum btrfs_compare_tree_result result)
5215{
5216 int ret = 0;
5217
5218 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5219
5220 if (!sctx->cur_inode_new_gen &&
5221 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5222 if (result == BTRFS_COMPARE_TREE_NEW)
5223 ret = record_new_ref(sctx);
5224 else if (result == BTRFS_COMPARE_TREE_DELETED)
5225 ret = record_deleted_ref(sctx);
5226 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5227 ret = record_changed_ref(sctx);
5228 }
5229
5230 return ret;
5231}
5232
5233/*
5234 * Process new/deleted/changed xattrs. We skip processing in the
5235 * cur_inode_new_gen case because changed_inode did already initiate processing
5236 * of xattrs. The reason is the same as in changed_ref
5237 */
5238static int changed_xattr(struct send_ctx *sctx,
5239 enum btrfs_compare_tree_result result)
5240{
5241 int ret = 0;
5242
5243 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5244
5245 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5246 if (result == BTRFS_COMPARE_TREE_NEW)
5247 ret = process_new_xattr(sctx);
5248 else if (result == BTRFS_COMPARE_TREE_DELETED)
5249 ret = process_deleted_xattr(sctx);
5250 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5251 ret = process_changed_xattr(sctx);
5252 }
5253
5254 return ret;
5255}
5256
5257/*
5258 * Process new/deleted/changed extents. We skip processing in the
5259 * cur_inode_new_gen case because changed_inode did already initiate processing
5260 * of extents. The reason is the same as in changed_ref
5261 */
5262static int changed_extent(struct send_ctx *sctx,
5263 enum btrfs_compare_tree_result result)
5264{
5265 int ret = 0;
5266
5267 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5268
5269 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5270 if (result != BTRFS_COMPARE_TREE_DELETED)
5271 ret = process_extent(sctx, sctx->left_path,
5272 sctx->cmp_key);
5273 }
5274
5275 return ret;
5276}
5277
5278static int dir_changed(struct send_ctx *sctx, u64 dir)
5279{
5280 u64 orig_gen, new_gen;
5281 int ret;
5282
5283 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5284 NULL, NULL);
5285 if (ret)
5286 return ret;
5287
5288 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5289 NULL, NULL, NULL);
5290 if (ret)
5291 return ret;
5292
5293 return (orig_gen != new_gen) ? 1 : 0;
5294}
5295
5296static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5297 struct btrfs_key *key)
5298{
5299 struct btrfs_inode_extref *extref;
5300 struct extent_buffer *leaf;
5301 u64 dirid = 0, last_dirid = 0;
5302 unsigned long ptr;
5303 u32 item_size;
5304 u32 cur_offset = 0;
5305 int ref_name_len;
5306 int ret = 0;
5307
5308 /* Easy case, just check this one dirid */
5309 if (key->type == BTRFS_INODE_REF_KEY) {
5310 dirid = key->offset;
5311
5312 ret = dir_changed(sctx, dirid);
5313 goto out;
5314 }
5315
5316 leaf = path->nodes[0];
5317 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5318 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5319 while (cur_offset < item_size) {
5320 extref = (struct btrfs_inode_extref *)(ptr +
5321 cur_offset);
5322 dirid = btrfs_inode_extref_parent(leaf, extref);
5323 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5324 cur_offset += ref_name_len + sizeof(*extref);
5325 if (dirid == last_dirid)
5326 continue;
5327 ret = dir_changed(sctx, dirid);
5328 if (ret)
5329 break;
5330 last_dirid = dirid;
5331 }
5332out:
5333 return ret;
5334}
5335
5336/*
5337 * Updates compare related fields in sctx and simply forwards to the actual
5338 * changed_xxx functions.
5339 */
5340static int changed_cb(struct btrfs_root *left_root,
5341 struct btrfs_root *right_root,
5342 struct btrfs_path *left_path,
5343 struct btrfs_path *right_path,
5344 struct btrfs_key *key,
5345 enum btrfs_compare_tree_result result,
5346 void *ctx)
5347{
5348 int ret = 0;
5349 struct send_ctx *sctx = ctx;
5350
5351 if (result == BTRFS_COMPARE_TREE_SAME) {
5352 if (key->type == BTRFS_INODE_REF_KEY ||
5353 key->type == BTRFS_INODE_EXTREF_KEY) {
5354 ret = compare_refs(sctx, left_path, key);
5355 if (!ret)
5356 return 0;
5357 if (ret < 0)
5358 return ret;
5359 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5360 return maybe_send_hole(sctx, left_path, key);
5361 } else {
5362 return 0;
5363 }
5364 result = BTRFS_COMPARE_TREE_CHANGED;
5365 ret = 0;
5366 }
5367
5368 sctx->left_path = left_path;
5369 sctx->right_path = right_path;
5370 sctx->cmp_key = key;
5371
5372 ret = finish_inode_if_needed(sctx, 0);
5373 if (ret < 0)
5374 goto out;
5375
5376 /* Ignore non-FS objects */
5377 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5378 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5379 goto out;
5380
5381 if (key->type == BTRFS_INODE_ITEM_KEY)
5382 ret = changed_inode(sctx, result);
5383 else if (key->type == BTRFS_INODE_REF_KEY ||
5384 key->type == BTRFS_INODE_EXTREF_KEY)
5385 ret = changed_ref(sctx, result);
5386 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5387 ret = changed_xattr(sctx, result);
5388 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5389 ret = changed_extent(sctx, result);
5390
5391out:
5392 return ret;
5393}
5394
5395static int full_send_tree(struct send_ctx *sctx)
5396{
5397 int ret;
5398 struct btrfs_root *send_root = sctx->send_root;
5399 struct btrfs_key key;
5400 struct btrfs_key found_key;
5401 struct btrfs_path *path;
5402 struct extent_buffer *eb;
5403 int slot;
5404
5405 path = alloc_path_for_send();
5406 if (!path)
5407 return -ENOMEM;
5408
5409 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5410 key.type = BTRFS_INODE_ITEM_KEY;
5411 key.offset = 0;
5412
5413 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5414 if (ret < 0)
5415 goto out;
5416 if (ret)
5417 goto out_finish;
5418
5419 while (1) {
5420 eb = path->nodes[0];
5421 slot = path->slots[0];
5422 btrfs_item_key_to_cpu(eb, &found_key, slot);
5423
5424 ret = changed_cb(send_root, NULL, path, NULL,
5425 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5426 if (ret < 0)
5427 goto out;
5428
5429 key.objectid = found_key.objectid;
5430 key.type = found_key.type;
5431 key.offset = found_key.offset + 1;
5432
5433 ret = btrfs_next_item(send_root, path);
5434 if (ret < 0)
5435 goto out;
5436 if (ret) {
5437 ret = 0;
5438 break;
5439 }
5440 }
5441
5442out_finish:
5443 ret = finish_inode_if_needed(sctx, 1);
5444
5445out:
5446 btrfs_free_path(path);
5447 return ret;
5448}
5449
5450static int send_subvol(struct send_ctx *sctx)
5451{
5452 int ret;
5453
5454 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5455 ret = send_header(sctx);
5456 if (ret < 0)
5457 goto out;
5458 }
5459
5460 ret = send_subvol_begin(sctx);
5461 if (ret < 0)
5462 goto out;
5463
5464 if (sctx->parent_root) {
5465 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5466 changed_cb, sctx);
5467 if (ret < 0)
5468 goto out;
5469 ret = finish_inode_if_needed(sctx, 1);
5470 if (ret < 0)
5471 goto out;
5472 } else {
5473 ret = full_send_tree(sctx);
5474 if (ret < 0)
5475 goto out;
5476 }
5477
5478out:
5479 free_recorded_refs(sctx);
5480 return ret;
5481}
5482
5483static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5484{
5485 spin_lock(&root->root_item_lock);
5486 root->send_in_progress--;
5487 /*
5488 * Not much left to do, we don't know why it's unbalanced and
5489 * can't blindly reset it to 0.
5490 */
5491 if (root->send_in_progress < 0)
5492 btrfs_err(root->fs_info,
5493 "send_in_progres unbalanced %d root %llu\n",
5494 root->send_in_progress, root->root_key.objectid);
5495 spin_unlock(&root->root_item_lock);
5496}
5497
5498long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5499{
5500 int ret = 0;
5501 struct btrfs_root *send_root;
5502 struct btrfs_root *clone_root;
5503 struct btrfs_fs_info *fs_info;
5504 struct btrfs_ioctl_send_args *arg = NULL;
5505 struct btrfs_key key;
5506 struct send_ctx *sctx = NULL;
5507 u32 i;
5508 u64 *clone_sources_tmp = NULL;
5509 int clone_sources_to_rollback = 0;
5510 int sort_clone_roots = 0;
5511 int index;
5512
5513 if (!capable(CAP_SYS_ADMIN))
5514 return -EPERM;
5515
5516 send_root = BTRFS_I(file_inode(mnt_file))->root;
5517 fs_info = send_root->fs_info;
5518
5519 /*
5520 * The subvolume must remain read-only during send, protect against
5521 * making it RW.
5522 */
5523 spin_lock(&send_root->root_item_lock);
5524 send_root->send_in_progress++;
5525 spin_unlock(&send_root->root_item_lock);
5526
5527 /*
5528 * This is done when we lookup the root, it should already be complete
5529 * by the time we get here.
5530 */
5531 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5532
5533 /*
5534 * Userspace tools do the checks and warn the user if it's
5535 * not RO.
5536 */
5537 if (!btrfs_root_readonly(send_root)) {
5538 ret = -EPERM;
5539 goto out;
5540 }
5541
5542 arg = memdup_user(arg_, sizeof(*arg));
5543 if (IS_ERR(arg)) {
5544 ret = PTR_ERR(arg);
5545 arg = NULL;
5546 goto out;
5547 }
5548
5549 if (!access_ok(VERIFY_READ, arg->clone_sources,
5550 sizeof(*arg->clone_sources) *
5551 arg->clone_sources_count)) {
5552 ret = -EFAULT;
5553 goto out;
5554 }
5555
5556 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5557 ret = -EINVAL;
5558 goto out;
5559 }
5560
5561 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5562 if (!sctx) {
5563 ret = -ENOMEM;
5564 goto out;
5565 }
5566
5567 INIT_LIST_HEAD(&sctx->new_refs);
5568 INIT_LIST_HEAD(&sctx->deleted_refs);
5569 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5570 INIT_LIST_HEAD(&sctx->name_cache_list);
5571
5572 sctx->flags = arg->flags;
5573
5574 sctx->send_filp = fget(arg->send_fd);
5575 if (!sctx->send_filp) {
5576 ret = -EBADF;
5577 goto out;
5578 }
5579
5580 sctx->send_root = send_root;
5581 sctx->clone_roots_cnt = arg->clone_sources_count;
5582
5583 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5584 sctx->send_buf = vmalloc(sctx->send_max_size);
5585 if (!sctx->send_buf) {
5586 ret = -ENOMEM;
5587 goto out;
5588 }
5589
5590 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5591 if (!sctx->read_buf) {
5592 ret = -ENOMEM;
5593 goto out;
5594 }
5595
5596 sctx->pending_dir_moves = RB_ROOT;
5597 sctx->waiting_dir_moves = RB_ROOT;
5598 sctx->orphan_dirs = RB_ROOT;
5599
5600 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5601 (arg->clone_sources_count + 1));
5602 if (!sctx->clone_roots) {
5603 ret = -ENOMEM;
5604 goto out;
5605 }
5606
5607 if (arg->clone_sources_count) {
5608 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5609 sizeof(*arg->clone_sources));
5610 if (!clone_sources_tmp) {
5611 ret = -ENOMEM;
5612 goto out;
5613 }
5614
5615 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5616 arg->clone_sources_count *
5617 sizeof(*arg->clone_sources));
5618 if (ret) {
5619 ret = -EFAULT;
5620 goto out;
5621 }
5622
5623 for (i = 0; i < arg->clone_sources_count; i++) {
5624 key.objectid = clone_sources_tmp[i];
5625 key.type = BTRFS_ROOT_ITEM_KEY;
5626 key.offset = (u64)-1;
5627
5628 index = srcu_read_lock(&fs_info->subvol_srcu);
5629
5630 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5631 if (IS_ERR(clone_root)) {
5632 srcu_read_unlock(&fs_info->subvol_srcu, index);
5633 ret = PTR_ERR(clone_root);
5634 goto out;
5635 }
5636 clone_sources_to_rollback = i + 1;
5637 spin_lock(&clone_root->root_item_lock);
5638 clone_root->send_in_progress++;
5639 if (!btrfs_root_readonly(clone_root)) {
5640 spin_unlock(&clone_root->root_item_lock);
5641 srcu_read_unlock(&fs_info->subvol_srcu, index);
5642 ret = -EPERM;
5643 goto out;
5644 }
5645 spin_unlock(&clone_root->root_item_lock);
5646 srcu_read_unlock(&fs_info->subvol_srcu, index);
5647
5648 sctx->clone_roots[i].root = clone_root;
5649 }
5650 vfree(clone_sources_tmp);
5651 clone_sources_tmp = NULL;
5652 }
5653
5654 if (arg->parent_root) {
5655 key.objectid = arg->parent_root;
5656 key.type = BTRFS_ROOT_ITEM_KEY;
5657 key.offset = (u64)-1;
5658
5659 index = srcu_read_lock(&fs_info->subvol_srcu);
5660
5661 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5662 if (IS_ERR(sctx->parent_root)) {
5663 srcu_read_unlock(&fs_info->subvol_srcu, index);
5664 ret = PTR_ERR(sctx->parent_root);
5665 goto out;
5666 }
5667
5668 spin_lock(&sctx->parent_root->root_item_lock);
5669 sctx->parent_root->send_in_progress++;
5670 if (!btrfs_root_readonly(sctx->parent_root)) {
5671 spin_unlock(&sctx->parent_root->root_item_lock);
5672 srcu_read_unlock(&fs_info->subvol_srcu, index);
5673 ret = -EPERM;
5674 goto out;
5675 }
5676 spin_unlock(&sctx->parent_root->root_item_lock);
5677
5678 srcu_read_unlock(&fs_info->subvol_srcu, index);
5679 }
5680
5681 /*
5682 * Clones from send_root are allowed, but only if the clone source
5683 * is behind the current send position. This is checked while searching
5684 * for possible clone sources.
5685 */
5686 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5687
5688 /* We do a bsearch later */
5689 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5690 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5691 NULL);
5692 sort_clone_roots = 1;
5693
5694 current->journal_info = (void *)BTRFS_SEND_TRANS_STUB;
5695 ret = send_subvol(sctx);
5696 current->journal_info = NULL;
5697 if (ret < 0)
5698 goto out;
5699
5700 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5701 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5702 if (ret < 0)
5703 goto out;
5704 ret = send_cmd(sctx);
5705 if (ret < 0)
5706 goto out;
5707 }
5708
5709out:
5710 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5711 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5712 struct rb_node *n;
5713 struct pending_dir_move *pm;
5714
5715 n = rb_first(&sctx->pending_dir_moves);
5716 pm = rb_entry(n, struct pending_dir_move, node);
5717 while (!list_empty(&pm->list)) {
5718 struct pending_dir_move *pm2;
5719
5720 pm2 = list_first_entry(&pm->list,
5721 struct pending_dir_move, list);
5722 free_pending_move(sctx, pm2);
5723 }
5724 free_pending_move(sctx, pm);
5725 }
5726
5727 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5728 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5729 struct rb_node *n;
5730 struct waiting_dir_move *dm;
5731
5732 n = rb_first(&sctx->waiting_dir_moves);
5733 dm = rb_entry(n, struct waiting_dir_move, node);
5734 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5735 kfree(dm);
5736 }
5737
5738 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
5739 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
5740 struct rb_node *n;
5741 struct orphan_dir_info *odi;
5742
5743 n = rb_first(&sctx->orphan_dirs);
5744 odi = rb_entry(n, struct orphan_dir_info, node);
5745 free_orphan_dir_info(sctx, odi);
5746 }
5747
5748 if (sort_clone_roots) {
5749 for (i = 0; i < sctx->clone_roots_cnt; i++)
5750 btrfs_root_dec_send_in_progress(
5751 sctx->clone_roots[i].root);
5752 } else {
5753 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5754 btrfs_root_dec_send_in_progress(
5755 sctx->clone_roots[i].root);
5756
5757 btrfs_root_dec_send_in_progress(send_root);
5758 }
5759 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5760 btrfs_root_dec_send_in_progress(sctx->parent_root);
5761
5762 kfree(arg);
5763 vfree(clone_sources_tmp);
5764
5765 if (sctx) {
5766 if (sctx->send_filp)
5767 fput(sctx->send_filp);
5768
5769 vfree(sctx->clone_roots);
5770 vfree(sctx->send_buf);
5771 vfree(sctx->read_buf);
5772
5773 name_cache_free(sctx);
5774
5775 kfree(sctx);
5776 }
5777
5778 return ret;
5779}