?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ASoC: neo1973_wm8753: Post gta01 support removal cleanup
[linux-2.6/btrfs-unstable.git] / fs / btrfs / send.c
blob730dce395858a6e4c86a43b4a8fa41779d719e0d
1 /*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/bsearch.h>
20 #include <linux/fs.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/crc32c.h>
28 #include <linux/vmalloc.h>
29 #include <linux/string.h>
30
31 #include "send.h"
32 #include "backref.h"
33 #include "locking.h"
34 #include "disk-io.h"
35 #include "btrfs_inode.h"
36 #include "transaction.h"
37
38 static 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 */
49 struct fs_path {
50 union {
51 struct {
52 char *start;
53 char *end;
54 char *prepared;
55
56 char *buf;
57 int buf_len;
58 unsigned int reversed:1;
59 unsigned int virtual_mem:1;
60 char inline_buf[];
61 };
62 char pad[PAGE_SIZE];
63 };
64 };
65 #define FS_PATH_INLINE_SIZE \
66 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
67
68
69 /* reused for each extent */
70 struct clone_root {
71 struct btrfs_root *root;
72 u64 ino;
73 u64 offset;
74
75 u64 found_refs;
76 };
77
78 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
79 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
80
81 struct send_ctx {
82 struct file *send_filp;
83 loff_t send_off;
84 char *send_buf;
85 u32 send_size;
86 u32 send_max_size;
87 u64 total_send_size;
88 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
89 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
90
91 struct btrfs_root *send_root;
92 struct btrfs_root *parent_root;
93 struct clone_root *clone_roots;
94 int clone_roots_cnt;
95
96 /* current state of the compare_tree call */
97 struct btrfs_path *left_path;
98 struct btrfs_path *right_path;
99 struct btrfs_key *cmp_key;
100
101 /*
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
104 */
105 u64 cur_ino;
106 u64 cur_inode_gen;
107 int cur_inode_new;
108 int cur_inode_new_gen;
109 int cur_inode_deleted;
110 u64 cur_inode_size;
111 u64 cur_inode_mode;
112 u64 cur_inode_last_extent;
113
114 u64 send_progress;
115
116 struct list_head new_refs;
117 struct list_head deleted_refs;
118
119 struct radix_tree_root name_cache;
120 struct list_head name_cache_list;
121 int name_cache_size;
122
123 char *read_buf;
124
125 /*
126 * We process inodes by their increasing order, so if before an
127 * incremental send we reverse the parent/child relationship of
128 * directories such that a directory with a lower inode number was
129 * the parent of a directory with a higher inode number, and the one
130 * becoming the new parent got renamed too, we can't rename/move the
131 * directory with lower inode number when we finish processing it - we
132 * must process the directory with higher inode number first, then
133 * rename/move it and then rename/move the directory with lower inode
134 * number. Example follows.
135 *
136 * Tree state when the first send was performed:
137 *
138 * .
139 * |-- a (ino 257)
140 * |-- b (ino 258)
141 * |
142 * |
143 * |-- c (ino 259)
144 * | |-- d (ino 260)
145 * |
146 * |-- c2 (ino 261)
147 *
148 * Tree state when the second (incremental) send is performed:
149 *
150 * .
151 * |-- a (ino 257)
152 * |-- b (ino 258)
153 * |-- c2 (ino 261)
154 * |-- d2 (ino 260)
155 * |-- cc (ino 259)
156 *
157 * The sequence of steps that lead to the second state was:
158 *
159 * mv /a/b/c/d /a/b/c2/d2
160 * mv /a/b/c /a/b/c2/d2/cc
161 *
162 * "c" has lower inode number, but we can't move it (2nd mv operation)
163 * before we move "d", which has higher inode number.
164 *
165 * So we just memorize which move/rename operations must be performed
166 * later when their respective parent is processed and moved/renamed.
167 */
168
169 /* Indexed by parent directory inode number. */
170 struct rb_root pending_dir_moves;
171
172 /*
173 * Reverse index, indexed by the inode number of a directory that
174 * is waiting for the move/rename of its immediate parent before its
175 * own move/rename can be performed.
176 */
177 struct rb_root waiting_dir_moves;
178 };
179
180 struct pending_dir_move {
181 struct rb_node node;
182 struct list_head list;
183 u64 parent_ino;
184 u64 ino;
185 u64 gen;
186 struct list_head update_refs;
187 };
188
189 struct waiting_dir_move {
190 struct rb_node node;
191 u64 ino;
192 };
193
194 struct name_cache_entry {
195 struct list_head list;
196 /*
197 * radix_tree has only 32bit entries but we need to handle 64bit inums.
198 * We use the lower 32bit of the 64bit inum to store it in the tree. If
199 * more then one inum would fall into the same entry, we use radix_list
200 * to store the additional entries. radix_list is also used to store
201 * entries where two entries have the same inum but different
202 * generations.
203 */
204 struct list_head radix_list;
205 u64 ino;
206 u64 gen;
207 u64 parent_ino;
208 u64 parent_gen;
209 int ret;
210 int need_later_update;
211 int name_len;
212 char name[];
213 };
214
215 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
216
217 static int need_send_hole(struct send_ctx *sctx)
218 {
219 return (sctx->parent_root && !sctx->cur_inode_new &&
220 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
221 S_ISREG(sctx->cur_inode_mode));
222 }
223
224 static void fs_path_reset(struct fs_path *p)
225 {
226 if (p->reversed) {
227 p->start = p->buf + p->buf_len - 1;
228 p->end = p->start;
229 *p->start = 0;
230 } else {
231 p->start = p->buf;
232 p->end = p->start;
233 *p->start = 0;
234 }
235 }
236
237 static struct fs_path *fs_path_alloc(void)
238 {
239 struct fs_path *p;
240
241 p = kmalloc(sizeof(*p), GFP_NOFS);
242 if (!p)
243 return NULL;
244 p->reversed = 0;
245 p->virtual_mem = 0;
246 p->buf = p->inline_buf;
247 p->buf_len = FS_PATH_INLINE_SIZE;
248 fs_path_reset(p);
249 return p;
250 }
251
252 static struct fs_path *fs_path_alloc_reversed(void)
253 {
254 struct fs_path *p;
255
256 p = fs_path_alloc();
257 if (!p)
258 return NULL;
259 p->reversed = 1;
260 fs_path_reset(p);
261 return p;
262 }
263
264 static void fs_path_free(struct fs_path *p)
265 {
266 if (!p)
267 return;
268 if (p->buf != p->inline_buf) {
269 if (p->virtual_mem)
270 vfree(p->buf);
271 else
272 kfree(p->buf);
273 }
274 kfree(p);
275 }
276
277 static int fs_path_len(struct fs_path *p)
278 {
279 return p->end - p->start;
280 }
281
282 static int fs_path_ensure_buf(struct fs_path *p, int len)
283 {
284 char *tmp_buf;
285 int path_len;
286 int old_buf_len;
287
288 len++;
289
290 if (p->buf_len >= len)
291 return 0;
292
293 path_len = p->end - p->start;
294 old_buf_len = p->buf_len;
295 len = PAGE_ALIGN(len);
296
297 if (p->buf == p->inline_buf) {
298 tmp_buf = kmalloc(len, GFP_NOFS | __GFP_NOWARN);
299 if (!tmp_buf) {
300 tmp_buf = vmalloc(len);
301 if (!tmp_buf)
302 return -ENOMEM;
303 p->virtual_mem = 1;
304 }
305 memcpy(tmp_buf, p->buf, p->buf_len);
306 p->buf = tmp_buf;
307 p->buf_len = len;
308 } else {
309 if (p->virtual_mem) {
310 tmp_buf = vmalloc(len);
311 if (!tmp_buf)
312 return -ENOMEM;
313 memcpy(tmp_buf, p->buf, p->buf_len);
314 vfree(p->buf);
315 } else {
316 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
317 if (!tmp_buf) {
318 tmp_buf = vmalloc(len);
319 if (!tmp_buf)
320 return -ENOMEM;
321 memcpy(tmp_buf, p->buf, p->buf_len);
322 kfree(p->buf);
323 p->virtual_mem = 1;
324 }
325 }
326 p->buf = tmp_buf;
327 p->buf_len = len;
328 }
329 if (p->reversed) {
330 tmp_buf = p->buf + old_buf_len - path_len - 1;
331 p->end = p->buf + p->buf_len - 1;
332 p->start = p->end - path_len;
333 memmove(p->start, tmp_buf, path_len + 1);
334 } else {
335 p->start = p->buf;
336 p->end = p->start + path_len;
337 }
338 return 0;
339 }
340
341 static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
342 {
343 int ret;
344 int new_len;
345
346 new_len = p->end - p->start + name_len;
347 if (p->start != p->end)
348 new_len++;
349 ret = fs_path_ensure_buf(p, new_len);
350 if (ret < 0)
351 goto out;
352
353 if (p->reversed) {
354 if (p->start != p->end)
355 *--p->start = '/';
356 p->start -= name_len;
357 p->prepared = p->start;
358 } else {
359 if (p->start != p->end)
360 *p->end++ = '/';
361 p->prepared = p->end;
362 p->end += name_len;
363 *p->end = 0;
364 }
365
366 out:
367 return ret;
368 }
369
370 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
371 {
372 int ret;
373
374 ret = fs_path_prepare_for_add(p, name_len);
375 if (ret < 0)
376 goto out;
377 memcpy(p->prepared, name, name_len);
378 p->prepared = NULL;
379
380 out:
381 return ret;
382 }
383
384 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
385 {
386 int ret;
387
388 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
389 if (ret < 0)
390 goto out;
391 memcpy(p->prepared, p2->start, p2->end - p2->start);
392 p->prepared = NULL;
393
394 out:
395 return ret;
396 }
397
398 static int fs_path_add_from_extent_buffer(struct fs_path *p,
399 struct extent_buffer *eb,
400 unsigned long off, int len)
401 {
402 int ret;
403
404 ret = fs_path_prepare_for_add(p, len);
405 if (ret < 0)
406 goto out;
407
408 read_extent_buffer(eb, p->prepared, off, len);
409 p->prepared = NULL;
410
411 out:
412 return ret;
413 }
414
415 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
416 {
417 int ret;
418
419 p->reversed = from->reversed;
420 fs_path_reset(p);
421
422 ret = fs_path_add_path(p, from);
423
424 return ret;
425 }
426
427
428 static void fs_path_unreverse(struct fs_path *p)
429 {
430 char *tmp;
431 int len;
432
433 if (!p->reversed)
434 return;
435
436 tmp = p->start;
437 len = p->end - p->start;
438 p->start = p->buf;
439 p->end = p->start + len;
440 memmove(p->start, tmp, len + 1);
441 p->reversed = 0;
442 }
443
444 static struct btrfs_path *alloc_path_for_send(void)
445 {
446 struct btrfs_path *path;
447
448 path = btrfs_alloc_path();
449 if (!path)
450 return NULL;
451 path->search_commit_root = 1;
452 path->skip_locking = 1;
453 return path;
454 }
455
456 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
457 {
458 int ret;
459 mm_segment_t old_fs;
460 u32 pos = 0;
461
462 old_fs = get_fs();
463 set_fs(KERNEL_DS);
464
465 while (pos < len) {
466 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
467 /* TODO handle that correctly */
468 /*if (ret == -ERESTARTSYS) {
469 continue;
470 }*/
471 if (ret < 0)
472 goto out;
473 if (ret == 0) {
474 ret = -EIO;
475 goto out;
476 }
477 pos += ret;
478 }
479
480 ret = 0;
481
482 out:
483 set_fs(old_fs);
484 return ret;
485 }
486
487 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
488 {
489 struct btrfs_tlv_header *hdr;
490 int total_len = sizeof(*hdr) + len;
491 int left = sctx->send_max_size - sctx->send_size;
492
493 if (unlikely(left < total_len))
494 return -EOVERFLOW;
495
496 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
497 hdr->tlv_type = cpu_to_le16(attr);
498 hdr->tlv_len = cpu_to_le16(len);
499 memcpy(hdr + 1, data, len);
500 sctx->send_size += total_len;
501
502 return 0;
503 }
504
505 #define TLV_PUT_DEFINE_INT(bits) \
506 static int tlv_put_u##bits(struct send_ctx *sctx, \
507 u##bits attr, u##bits value) \
508 { \
509 __le##bits __tmp = cpu_to_le##bits(value); \
510 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
511 }
512
513 TLV_PUT_DEFINE_INT(64)
514
515 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
516 const char *str, int len)
517 {
518 if (len == -1)
519 len = strlen(str);
520 return tlv_put(sctx, attr, str, len);
521 }
522
523 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
524 const u8 *uuid)
525 {
526 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
527 }
528
529 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
530 struct extent_buffer *eb,
531 struct btrfs_timespec *ts)
532 {
533 struct btrfs_timespec bts;
534 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
535 return tlv_put(sctx, attr, &bts, sizeof(bts));
536 }
537
538
539 #define TLV_PUT(sctx, attrtype, attrlen, data) \
540 do { \
541 ret = tlv_put(sctx, attrtype, attrlen, data); \
542 if (ret < 0) \
543 goto tlv_put_failure; \
544 } while (0)
545
546 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
547 do { \
548 ret = tlv_put_u##bits(sctx, attrtype, value); \
549 if (ret < 0) \
550 goto tlv_put_failure; \
551 } while (0)
552
553 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
554 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
555 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
556 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
557 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
558 do { \
559 ret = tlv_put_string(sctx, attrtype, str, len); \
560 if (ret < 0) \
561 goto tlv_put_failure; \
562 } while (0)
563 #define TLV_PUT_PATH(sctx, attrtype, p) \
564 do { \
565 ret = tlv_put_string(sctx, attrtype, p->start, \
566 p->end - p->start); \
567 if (ret < 0) \
568 goto tlv_put_failure; \
569 } while(0)
570 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
571 do { \
572 ret = tlv_put_uuid(sctx, attrtype, uuid); \
573 if (ret < 0) \
574 goto tlv_put_failure; \
575 } while (0)
576 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
577 do { \
578 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
579 if (ret < 0) \
580 goto tlv_put_failure; \
581 } while (0)
582
583 static int send_header(struct send_ctx *sctx)
584 {
585 struct btrfs_stream_header hdr;
586
587 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
588 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
589
590 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
591 &sctx->send_off);
592 }
593
594 /*
595 * For each command/item we want to send to userspace, we call this function.
596 */
597 static int begin_cmd(struct send_ctx *sctx, int cmd)
598 {
599 struct btrfs_cmd_header *hdr;
600
601 if (WARN_ON(!sctx->send_buf))
602 return -EINVAL;
603
604 BUG_ON(sctx->send_size);
605
606 sctx->send_size += sizeof(*hdr);
607 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
608 hdr->cmd = cpu_to_le16(cmd);
609
610 return 0;
611 }
612
613 static int send_cmd(struct send_ctx *sctx)
614 {
615 int ret;
616 struct btrfs_cmd_header *hdr;
617 u32 crc;
618
619 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
620 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
621 hdr->crc = 0;
622
623 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
624 hdr->crc = cpu_to_le32(crc);
625
626 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
627 &sctx->send_off);
628
629 sctx->total_send_size += sctx->send_size;
630 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
631 sctx->send_size = 0;
632
633 return ret;
634 }
635
636 /*
637 * Sends a move instruction to user space
638 */
639 static int send_rename(struct send_ctx *sctx,
640 struct fs_path *from, struct fs_path *to)
641 {
642 int ret;
643
644 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
645
646 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
647 if (ret < 0)
648 goto out;
649
650 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
651 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
652
653 ret = send_cmd(sctx);
654
655 tlv_put_failure:
656 out:
657 return ret;
658 }
659
660 /*
661 * Sends a link instruction to user space
662 */
663 static int send_link(struct send_ctx *sctx,
664 struct fs_path *path, struct fs_path *lnk)
665 {
666 int ret;
667
668 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
669
670 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
671 if (ret < 0)
672 goto out;
673
674 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
675 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
676
677 ret = send_cmd(sctx);
678
679 tlv_put_failure:
680 out:
681 return ret;
682 }
683
684 /*
685 * Sends an unlink instruction to user space
686 */
687 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
688 {
689 int ret;
690
691 verbose_printk("btrfs: send_unlink %s\n", path->start);
692
693 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
694 if (ret < 0)
695 goto out;
696
697 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
698
699 ret = send_cmd(sctx);
700
701 tlv_put_failure:
702 out:
703 return ret;
704 }
705
706 /*
707 * Sends a rmdir instruction to user space
708 */
709 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
710 {
711 int ret;
712
713 verbose_printk("btrfs: send_rmdir %s\n", path->start);
714
715 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
716 if (ret < 0)
717 goto out;
718
719 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
720
721 ret = send_cmd(sctx);
722
723 tlv_put_failure:
724 out:
725 return ret;
726 }
727
728 /*
729 * Helper function to retrieve some fields from an inode item.
730 */
731 static int get_inode_info(struct btrfs_root *root,
732 u64 ino, u64 *size, u64 *gen,
733 u64 *mode, u64 *uid, u64 *gid,
734 u64 *rdev)
735 {
736 int ret;
737 struct btrfs_inode_item *ii;
738 struct btrfs_key key;
739 struct btrfs_path *path;
740
741 path = alloc_path_for_send();
742 if (!path)
743 return -ENOMEM;
744
745 key.objectid = ino;
746 key.type = BTRFS_INODE_ITEM_KEY;
747 key.offset = 0;
748 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
749 if (ret < 0)
750 goto out;
751 if (ret) {
752 ret = -ENOENT;
753 goto out;
754 }
755
756 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
757 struct btrfs_inode_item);
758 if (size)
759 *size = btrfs_inode_size(path->nodes[0], ii);
760 if (gen)
761 *gen = btrfs_inode_generation(path->nodes[0], ii);
762 if (mode)
763 *mode = btrfs_inode_mode(path->nodes[0], ii);
764 if (uid)
765 *uid = btrfs_inode_uid(path->nodes[0], ii);
766 if (gid)
767 *gid = btrfs_inode_gid(path->nodes[0], ii);
768 if (rdev)
769 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
770
771 out:
772 btrfs_free_path(path);
773 return ret;
774 }
775
776 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
777 struct fs_path *p,
778 void *ctx);
779
780 /*
781 * Helper function to iterate the entries in ONE btrfs_inode_ref or
782 * btrfs_inode_extref.
783 * The iterate callback may return a non zero value to stop iteration. This can
784 * be a negative value for error codes or 1 to simply stop it.
785 *
786 * path must point to the INODE_REF or INODE_EXTREF when called.
787 */
788 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
789 struct btrfs_key *found_key, int resolve,
790 iterate_inode_ref_t iterate, void *ctx)
791 {
792 struct extent_buffer *eb = path->nodes[0];
793 struct btrfs_item *item;
794 struct btrfs_inode_ref *iref;
795 struct btrfs_inode_extref *extref;
796 struct btrfs_path *tmp_path;
797 struct fs_path *p;
798 u32 cur = 0;
799 u32 total;
800 int slot = path->slots[0];
801 u32 name_len;
802 char *start;
803 int ret = 0;
804 int num = 0;
805 int index;
806 u64 dir;
807 unsigned long name_off;
808 unsigned long elem_size;
809 unsigned long ptr;
810
811 p = fs_path_alloc_reversed();
812 if (!p)
813 return -ENOMEM;
814
815 tmp_path = alloc_path_for_send();
816 if (!tmp_path) {
817 fs_path_free(p);
818 return -ENOMEM;
819 }
820
821
822 if (found_key->type == BTRFS_INODE_REF_KEY) {
823 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
824 struct btrfs_inode_ref);
825 item = btrfs_item_nr(slot);
826 total = btrfs_item_size(eb, item);
827 elem_size = sizeof(*iref);
828 } else {
829 ptr = btrfs_item_ptr_offset(eb, slot);
830 total = btrfs_item_size_nr(eb, slot);
831 elem_size = sizeof(*extref);
832 }
833
834 while (cur < total) {
835 fs_path_reset(p);
836
837 if (found_key->type == BTRFS_INODE_REF_KEY) {
838 iref = (struct btrfs_inode_ref *)(ptr + cur);
839 name_len = btrfs_inode_ref_name_len(eb, iref);
840 name_off = (unsigned long)(iref + 1);
841 index = btrfs_inode_ref_index(eb, iref);
842 dir = found_key->offset;
843 } else {
844 extref = (struct btrfs_inode_extref *)(ptr + cur);
845 name_len = btrfs_inode_extref_name_len(eb, extref);
846 name_off = (unsigned long)&extref->name;
847 index = btrfs_inode_extref_index(eb, extref);
848 dir = btrfs_inode_extref_parent(eb, extref);
849 }
850
851 if (resolve) {
852 start = btrfs_ref_to_path(root, tmp_path, name_len,
853 name_off, eb, dir,
854 p->buf, p->buf_len);
855 if (IS_ERR(start)) {
856 ret = PTR_ERR(start);
857 goto out;
858 }
859 if (start < p->buf) {
860 /* overflow , try again with larger buffer */
861 ret = fs_path_ensure_buf(p,
862 p->buf_len + p->buf - start);
863 if (ret < 0)
864 goto out;
865 start = btrfs_ref_to_path(root, tmp_path,
866 name_len, name_off,
867 eb, dir,
868 p->buf, p->buf_len);
869 if (IS_ERR(start)) {
870 ret = PTR_ERR(start);
871 goto out;
872 }
873 BUG_ON(start < p->buf);
874 }
875 p->start = start;
876 } else {
877 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
878 name_len);
879 if (ret < 0)
880 goto out;
881 }
882
883 cur += elem_size + name_len;
884 ret = iterate(num, dir, index, p, ctx);
885 if (ret)
886 goto out;
887 num++;
888 }
889
890 out:
891 btrfs_free_path(tmp_path);
892 fs_path_free(p);
893 return ret;
894 }
895
896 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
897 const char *name, int name_len,
898 const char *data, int data_len,
899 u8 type, void *ctx);
900
901 /*
902 * Helper function to iterate the entries in ONE btrfs_dir_item.
903 * The iterate callback may return a non zero value to stop iteration. This can
904 * be a negative value for error codes or 1 to simply stop it.
905 *
906 * path must point to the dir item when called.
907 */
908 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
909 struct btrfs_key *found_key,
910 iterate_dir_item_t iterate, void *ctx)
911 {
912 int ret = 0;
913 struct extent_buffer *eb;
914 struct btrfs_item *item;
915 struct btrfs_dir_item *di;
916 struct btrfs_key di_key;
917 char *buf = NULL;
918 char *buf2 = NULL;
919 int buf_len;
920 int buf_virtual = 0;
921 u32 name_len;
922 u32 data_len;
923 u32 cur;
924 u32 len;
925 u32 total;
926 int slot;
927 int num;
928 u8 type;
929
930 buf_len = PAGE_SIZE;
931 buf = kmalloc(buf_len, GFP_NOFS);
932 if (!buf) {
933 ret = -ENOMEM;
934 goto out;
935 }
936
937 eb = path->nodes[0];
938 slot = path->slots[0];
939 item = btrfs_item_nr(slot);
940 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
941 cur = 0;
942 len = 0;
943 total = btrfs_item_size(eb, item);
944
945 num = 0;
946 while (cur < total) {
947 name_len = btrfs_dir_name_len(eb, di);
948 data_len = btrfs_dir_data_len(eb, di);
949 type = btrfs_dir_type(eb, di);
950 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
951
952 if (name_len + data_len > buf_len) {
953 buf_len = PAGE_ALIGN(name_len + data_len);
954 if (buf_virtual) {
955 buf2 = vmalloc(buf_len);
956 if (!buf2) {
957 ret = -ENOMEM;
958 goto out;
959 }
960 vfree(buf);
961 } else {
962 buf2 = krealloc(buf, buf_len, GFP_NOFS);
963 if (!buf2) {
964 buf2 = vmalloc(buf_len);
965 if (!buf2) {
966 ret = -ENOMEM;
967 goto out;
968 }
969 kfree(buf);
970 buf_virtual = 1;
971 }
972 }
973
974 buf = buf2;
975 buf2 = NULL;
976 }
977
978 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
979 name_len + data_len);
980
981 len = sizeof(*di) + name_len + data_len;
982 di = (struct btrfs_dir_item *)((char *)di + len);
983 cur += len;
984
985 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
986 data_len, type, ctx);
987 if (ret < 0)
988 goto out;
989 if (ret) {
990 ret = 0;
991 goto out;
992 }
993
994 num++;
995 }
996
997 out:
998 if (buf_virtual)
999 vfree(buf);
1000 else
1001 kfree(buf);
1002 return ret;
1003 }
1004
1005 static int __copy_first_ref(int num, u64 dir, int index,
1006 struct fs_path *p, void *ctx)
1007 {
1008 int ret;
1009 struct fs_path *pt = ctx;
1010
1011 ret = fs_path_copy(pt, p);
1012 if (ret < 0)
1013 return ret;
1014
1015 /* we want the first only */
1016 return 1;
1017 }
1018
1019 /*
1020 * Retrieve the first path of an inode. If an inode has more then one
1021 * ref/hardlink, this is ignored.
1022 */
1023 static int get_inode_path(struct btrfs_root *root,
1024 u64 ino, struct fs_path *path)
1025 {
1026 int ret;
1027 struct btrfs_key key, found_key;
1028 struct btrfs_path *p;
1029
1030 p = alloc_path_for_send();
1031 if (!p)
1032 return -ENOMEM;
1033
1034 fs_path_reset(path);
1035
1036 key.objectid = ino;
1037 key.type = BTRFS_INODE_REF_KEY;
1038 key.offset = 0;
1039
1040 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1041 if (ret < 0)
1042 goto out;
1043 if (ret) {
1044 ret = 1;
1045 goto out;
1046 }
1047 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1048 if (found_key.objectid != ino ||
1049 (found_key.type != BTRFS_INODE_REF_KEY &&
1050 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1051 ret = -ENOENT;
1052 goto out;
1053 }
1054
1055 ret = iterate_inode_ref(root, p, &found_key, 1,
1056 __copy_first_ref, path);
1057 if (ret < 0)
1058 goto out;
1059 ret = 0;
1060
1061 out:
1062 btrfs_free_path(p);
1063 return ret;
1064 }
1065
1066 struct backref_ctx {
1067 struct send_ctx *sctx;
1068
1069 /* number of total found references */
1070 u64 found;
1071
1072 /*
1073 * used for clones found in send_root. clones found behind cur_objectid
1074 * and cur_offset are not considered as allowed clones.
1075 */
1076 u64 cur_objectid;
1077 u64 cur_offset;
1078
1079 /* may be truncated in case it's the last extent in a file */
1080 u64 extent_len;
1081
1082 /* Just to check for bugs in backref resolving */
1083 int found_itself;
1084 };
1085
1086 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1087 {
1088 u64 root = (u64)(uintptr_t)key;
1089 struct clone_root *cr = (struct clone_root *)elt;
1090
1091 if (root < cr->root->objectid)
1092 return -1;
1093 if (root > cr->root->objectid)
1094 return 1;
1095 return 0;
1096 }
1097
1098 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1099 {
1100 struct clone_root *cr1 = (struct clone_root *)e1;
1101 struct clone_root *cr2 = (struct clone_root *)e2;
1102
1103 if (cr1->root->objectid < cr2->root->objectid)
1104 return -1;
1105 if (cr1->root->objectid > cr2->root->objectid)
1106 return 1;
1107 return 0;
1108 }
1109
1110 /*
1111 * Called for every backref that is found for the current extent.
1112 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1113 */
1114 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1115 {
1116 struct backref_ctx *bctx = ctx_;
1117 struct clone_root *found;
1118 int ret;
1119 u64 i_size;
1120
1121 /* First check if the root is in the list of accepted clone sources */
1122 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1123 bctx->sctx->clone_roots_cnt,
1124 sizeof(struct clone_root),
1125 __clone_root_cmp_bsearch);
1126 if (!found)
1127 return 0;
1128
1129 if (found->root == bctx->sctx->send_root &&
1130 ino == bctx->cur_objectid &&
1131 offset == bctx->cur_offset) {
1132 bctx->found_itself = 1;
1133 }
1134
1135 /*
1136 * There are inodes that have extents that lie behind its i_size. Don't
1137 * accept clones from these extents.
1138 */
1139 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1140 NULL);
1141 if (ret < 0)
1142 return ret;
1143
1144 if (offset + bctx->extent_len > i_size)
1145 return 0;
1146
1147 /*
1148 * Make sure we don't consider clones from send_root that are
1149 * behind the current inode/offset.
1150 */
1151 if (found->root == bctx->sctx->send_root) {
1152 /*
1153 * TODO for the moment we don't accept clones from the inode
1154 * that is currently send. We may change this when
1155 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1156 * file.
1157 */
1158 if (ino >= bctx->cur_objectid)
1159 return 0;
1160 #if 0
1161 if (ino > bctx->cur_objectid)
1162 return 0;
1163 if (offset + bctx->extent_len > bctx->cur_offset)
1164 return 0;
1165 #endif
1166 }
1167
1168 bctx->found++;
1169 found->found_refs++;
1170 if (ino < found->ino) {
1171 found->ino = ino;
1172 found->offset = offset;
1173 } else if (found->ino == ino) {
1174 /*
1175 * same extent found more then once in the same file.
1176 */
1177 if (found->offset > offset + bctx->extent_len)
1178 found->offset = offset;
1179 }
1180
1181 return 0;
1182 }
1183
1184 /*
1185 * Given an inode, offset and extent item, it finds a good clone for a clone
1186 * instruction. Returns -ENOENT when none could be found. The function makes
1187 * sure that the returned clone is usable at the point where sending is at the
1188 * moment. This means, that no clones are accepted which lie behind the current
1189 * inode+offset.
1190 *
1191 * path must point to the extent item when called.
1192 */
1193 static int find_extent_clone(struct send_ctx *sctx,
1194 struct btrfs_path *path,
1195 u64 ino, u64 data_offset,
1196 u64 ino_size,
1197 struct clone_root **found)
1198 {
1199 int ret;
1200 int extent_type;
1201 u64 logical;
1202 u64 disk_byte;
1203 u64 num_bytes;
1204 u64 extent_item_pos;
1205 u64 flags = 0;
1206 struct btrfs_file_extent_item *fi;
1207 struct extent_buffer *eb = path->nodes[0];
1208 struct backref_ctx *backref_ctx = NULL;
1209 struct clone_root *cur_clone_root;
1210 struct btrfs_key found_key;
1211 struct btrfs_path *tmp_path;
1212 int compressed;
1213 u32 i;
1214
1215 tmp_path = alloc_path_for_send();
1216 if (!tmp_path)
1217 return -ENOMEM;
1218
1219 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1220 if (!backref_ctx) {
1221 ret = -ENOMEM;
1222 goto out;
1223 }
1224
1225 if (data_offset >= ino_size) {
1226 /*
1227 * There may be extents that lie behind the file's size.
1228 * I at least had this in combination with snapshotting while
1229 * writing large files.
1230 */
1231 ret = 0;
1232 goto out;
1233 }
1234
1235 fi = btrfs_item_ptr(eb, path->slots[0],
1236 struct btrfs_file_extent_item);
1237 extent_type = btrfs_file_extent_type(eb, fi);
1238 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1239 ret = -ENOENT;
1240 goto out;
1241 }
1242 compressed = btrfs_file_extent_compression(eb, fi);
1243
1244 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1245 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1246 if (disk_byte == 0) {
1247 ret = -ENOENT;
1248 goto out;
1249 }
1250 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1251
1252 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1253 &found_key, &flags);
1254 btrfs_release_path(tmp_path);
1255
1256 if (ret < 0)
1257 goto out;
1258 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1259 ret = -EIO;
1260 goto out;
1261 }
1262
1263 /*
1264 * Setup the clone roots.
1265 */
1266 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1267 cur_clone_root = sctx->clone_roots + i;
1268 cur_clone_root->ino = (u64)-1;
1269 cur_clone_root->offset = 0;
1270 cur_clone_root->found_refs = 0;
1271 }
1272
1273 backref_ctx->sctx = sctx;
1274 backref_ctx->found = 0;
1275 backref_ctx->cur_objectid = ino;
1276 backref_ctx->cur_offset = data_offset;
1277 backref_ctx->found_itself = 0;
1278 backref_ctx->extent_len = num_bytes;
1279
1280 /*
1281 * The last extent of a file may be too large due to page alignment.
1282 * We need to adjust extent_len in this case so that the checks in
1283 * __iterate_backrefs work.
1284 */
1285 if (data_offset + num_bytes >= ino_size)
1286 backref_ctx->extent_len = ino_size - data_offset;
1287
1288 /*
1289 * Now collect all backrefs.
1290 */
1291 if (compressed == BTRFS_COMPRESS_NONE)
1292 extent_item_pos = logical - found_key.objectid;
1293 else
1294 extent_item_pos = 0;
1295
1296 extent_item_pos = logical - found_key.objectid;
1297 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1298 found_key.objectid, extent_item_pos, 1,
1299 __iterate_backrefs, backref_ctx);
1300
1301 if (ret < 0)
1302 goto out;
1303
1304 if (!backref_ctx->found_itself) {
1305 /* found a bug in backref code? */
1306 ret = -EIO;
1307 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1308 "send_root. inode=%llu, offset=%llu, "
1309 "disk_byte=%llu found extent=%llu\n",
1310 ino, data_offset, disk_byte, found_key.objectid);
1311 goto out;
1312 }
1313
1314 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1315 "ino=%llu, "
1316 "num_bytes=%llu, logical=%llu\n",
1317 data_offset, ino, num_bytes, logical);
1318
1319 if (!backref_ctx->found)
1320 verbose_printk("btrfs: no clones found\n");
1321
1322 cur_clone_root = NULL;
1323 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1324 if (sctx->clone_roots[i].found_refs) {
1325 if (!cur_clone_root)
1326 cur_clone_root = sctx->clone_roots + i;
1327 else if (sctx->clone_roots[i].root == sctx->send_root)
1328 /* prefer clones from send_root over others */
1329 cur_clone_root = sctx->clone_roots + i;
1330 }
1331
1332 }
1333
1334 if (cur_clone_root) {
1335 *found = cur_clone_root;
1336 ret = 0;
1337 } else {
1338 ret = -ENOENT;
1339 }
1340
1341 out:
1342 btrfs_free_path(tmp_path);
1343 kfree(backref_ctx);
1344 return ret;
1345 }
1346
1347 static int read_symlink(struct btrfs_root *root,
1348 u64 ino,
1349 struct fs_path *dest)
1350 {
1351 int ret;
1352 struct btrfs_path *path;
1353 struct btrfs_key key;
1354 struct btrfs_file_extent_item *ei;
1355 u8 type;
1356 u8 compression;
1357 unsigned long off;
1358 int len;
1359
1360 path = alloc_path_for_send();
1361 if (!path)
1362 return -ENOMEM;
1363
1364 key.objectid = ino;
1365 key.type = BTRFS_EXTENT_DATA_KEY;
1366 key.offset = 0;
1367 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1368 if (ret < 0)
1369 goto out;
1370 BUG_ON(ret);
1371
1372 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1373 struct btrfs_file_extent_item);
1374 type = btrfs_file_extent_type(path->nodes[0], ei);
1375 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1376 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1377 BUG_ON(compression);
1378
1379 off = btrfs_file_extent_inline_start(ei);
1380 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1381
1382 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1383
1384 out:
1385 btrfs_free_path(path);
1386 return ret;
1387 }
1388
1389 /*
1390 * Helper function to generate a file name that is unique in the root of
1391 * send_root and parent_root. This is used to generate names for orphan inodes.
1392 */
1393 static int gen_unique_name(struct send_ctx *sctx,
1394 u64 ino, u64 gen,
1395 struct fs_path *dest)
1396 {
1397 int ret = 0;
1398 struct btrfs_path *path;
1399 struct btrfs_dir_item *di;
1400 char tmp[64];
1401 int len;
1402 u64 idx = 0;
1403
1404 path = alloc_path_for_send();
1405 if (!path)
1406 return -ENOMEM;
1407
1408 while (1) {
1409 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1410 ino, gen, idx);
1411 if (len >= sizeof(tmp)) {
1412 /* should really not happen */
1413 ret = -EOVERFLOW;
1414 goto out;
1415 }
1416
1417 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1418 path, BTRFS_FIRST_FREE_OBJECTID,
1419 tmp, strlen(tmp), 0);
1420 btrfs_release_path(path);
1421 if (IS_ERR(di)) {
1422 ret = PTR_ERR(di);
1423 goto out;
1424 }
1425 if (di) {
1426 /* not unique, try again */
1427 idx++;
1428 continue;
1429 }
1430
1431 if (!sctx->parent_root) {
1432 /* unique */
1433 ret = 0;
1434 break;
1435 }
1436
1437 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1438 path, BTRFS_FIRST_FREE_OBJECTID,
1439 tmp, strlen(tmp), 0);
1440 btrfs_release_path(path);
1441 if (IS_ERR(di)) {
1442 ret = PTR_ERR(di);
1443 goto out;
1444 }
1445 if (di) {
1446 /* not unique, try again */
1447 idx++;
1448 continue;
1449 }
1450 /* unique */
1451 break;
1452 }
1453
1454 ret = fs_path_add(dest, tmp, strlen(tmp));
1455
1456 out:
1457 btrfs_free_path(path);
1458 return ret;
1459 }
1460
1461 enum inode_state {
1462 inode_state_no_change,
1463 inode_state_will_create,
1464 inode_state_did_create,
1465 inode_state_will_delete,
1466 inode_state_did_delete,
1467 };
1468
1469 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1470 {
1471 int ret;
1472 int left_ret;
1473 int right_ret;
1474 u64 left_gen;
1475 u64 right_gen;
1476
1477 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1478 NULL, NULL);
1479 if (ret < 0 && ret != -ENOENT)
1480 goto out;
1481 left_ret = ret;
1482
1483 if (!sctx->parent_root) {
1484 right_ret = -ENOENT;
1485 } else {
1486 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1487 NULL, NULL, NULL, NULL);
1488 if (ret < 0 && ret != -ENOENT)
1489 goto out;
1490 right_ret = ret;
1491 }
1492
1493 if (!left_ret && !right_ret) {
1494 if (left_gen == gen && right_gen == gen) {
1495 ret = inode_state_no_change;
1496 } else if (left_gen == gen) {
1497 if (ino < sctx->send_progress)
1498 ret = inode_state_did_create;
1499 else
1500 ret = inode_state_will_create;
1501 } else if (right_gen == gen) {
1502 if (ino < sctx->send_progress)
1503 ret = inode_state_did_delete;
1504 else
1505 ret = inode_state_will_delete;
1506 } else {
1507 ret = -ENOENT;
1508 }
1509 } else if (!left_ret) {
1510 if (left_gen == gen) {
1511 if (ino < sctx->send_progress)
1512 ret = inode_state_did_create;
1513 else
1514 ret = inode_state_will_create;
1515 } else {
1516 ret = -ENOENT;
1517 }
1518 } else if (!right_ret) {
1519 if (right_gen == gen) {
1520 if (ino < sctx->send_progress)
1521 ret = inode_state_did_delete;
1522 else
1523 ret = inode_state_will_delete;
1524 } else {
1525 ret = -ENOENT;
1526 }
1527 } else {
1528 ret = -ENOENT;
1529 }
1530
1531 out:
1532 return ret;
1533 }
1534
1535 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1536 {
1537 int ret;
1538
1539 ret = get_cur_inode_state(sctx, ino, gen);
1540 if (ret < 0)
1541 goto out;
1542
1543 if (ret == inode_state_no_change ||
1544 ret == inode_state_did_create ||
1545 ret == inode_state_will_delete)
1546 ret = 1;
1547 else
1548 ret = 0;
1549
1550 out:
1551 return ret;
1552 }
1553
1554 /*
1555 * Helper function to lookup a dir item in a dir.
1556 */
1557 static int lookup_dir_item_inode(struct btrfs_root *root,
1558 u64 dir, const char *name, int name_len,
1559 u64 *found_inode,
1560 u8 *found_type)
1561 {
1562 int ret = 0;
1563 struct btrfs_dir_item *di;
1564 struct btrfs_key key;
1565 struct btrfs_path *path;
1566
1567 path = alloc_path_for_send();
1568 if (!path)
1569 return -ENOMEM;
1570
1571 di = btrfs_lookup_dir_item(NULL, root, path,
1572 dir, name, name_len, 0);
1573 if (!di) {
1574 ret = -ENOENT;
1575 goto out;
1576 }
1577 if (IS_ERR(di)) {
1578 ret = PTR_ERR(di);
1579 goto out;
1580 }
1581 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1582 *found_inode = key.objectid;
1583 *found_type = btrfs_dir_type(path->nodes[0], di);
1584
1585 out:
1586 btrfs_free_path(path);
1587 return ret;
1588 }
1589
1590 /*
1591 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1592 * generation of the parent dir and the name of the dir entry.
1593 */
1594 static int get_first_ref(struct btrfs_root *root, u64 ino,
1595 u64 *dir, u64 *dir_gen, struct fs_path *name)
1596 {
1597 int ret;
1598 struct btrfs_key key;
1599 struct btrfs_key found_key;
1600 struct btrfs_path *path;
1601 int len;
1602 u64 parent_dir;
1603
1604 path = alloc_path_for_send();
1605 if (!path)
1606 return -ENOMEM;
1607
1608 key.objectid = ino;
1609 key.type = BTRFS_INODE_REF_KEY;
1610 key.offset = 0;
1611
1612 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1613 if (ret < 0)
1614 goto out;
1615 if (!ret)
1616 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1617 path->slots[0]);
1618 if (ret || found_key.objectid != ino ||
1619 (found_key.type != BTRFS_INODE_REF_KEY &&
1620 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1621 ret = -ENOENT;
1622 goto out;
1623 }
1624
1625 if (key.type == BTRFS_INODE_REF_KEY) {
1626 struct btrfs_inode_ref *iref;
1627 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1628 struct btrfs_inode_ref);
1629 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1630 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1631 (unsigned long)(iref + 1),
1632 len);
1633 parent_dir = found_key.offset;
1634 } else {
1635 struct btrfs_inode_extref *extref;
1636 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1637 struct btrfs_inode_extref);
1638 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1639 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1640 (unsigned long)&extref->name, len);
1641 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1642 }
1643 if (ret < 0)
1644 goto out;
1645 btrfs_release_path(path);
1646
1647 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1648 NULL, NULL);
1649 if (ret < 0)
1650 goto out;
1651
1652 *dir = parent_dir;
1653
1654 out:
1655 btrfs_free_path(path);
1656 return ret;
1657 }
1658
1659 static int is_first_ref(struct btrfs_root *root,
1660 u64 ino, u64 dir,
1661 const char *name, int name_len)
1662 {
1663 int ret;
1664 struct fs_path *tmp_name;
1665 u64 tmp_dir;
1666 u64 tmp_dir_gen;
1667
1668 tmp_name = fs_path_alloc();
1669 if (!tmp_name)
1670 return -ENOMEM;
1671
1672 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1673 if (ret < 0)
1674 goto out;
1675
1676 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1677 ret = 0;
1678 goto out;
1679 }
1680
1681 ret = !memcmp(tmp_name->start, name, name_len);
1682
1683 out:
1684 fs_path_free(tmp_name);
1685 return ret;
1686 }
1687
1688 /*
1689 * Used by process_recorded_refs to determine if a new ref would overwrite an
1690 * already existing ref. In case it detects an overwrite, it returns the
1691 * inode/gen in who_ino/who_gen.
1692 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1693 * to make sure later references to the overwritten inode are possible.
1694 * Orphanizing is however only required for the first ref of an inode.
1695 * process_recorded_refs does an additional is_first_ref check to see if
1696 * orphanizing is really required.
1697 */
1698 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1699 const char *name, int name_len,
1700 u64 *who_ino, u64 *who_gen)
1701 {
1702 int ret = 0;
1703 u64 gen;
1704 u64 other_inode = 0;
1705 u8 other_type = 0;
1706
1707 if (!sctx->parent_root)
1708 goto out;
1709
1710 ret = is_inode_existent(sctx, dir, dir_gen);
1711 if (ret <= 0)
1712 goto out;
1713
1714 /*
1715 * If we have a parent root we need to verify that the parent dir was
1716 * not delted and then re-created, if it was then we have no overwrite
1717 * and we can just unlink this entry.
1718 */
1719 if (sctx->parent_root) {
1720 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1721 NULL, NULL, NULL);
1722 if (ret < 0 && ret != -ENOENT)
1723 goto out;
1724 if (ret) {
1725 ret = 0;
1726 goto out;
1727 }
1728 if (gen != dir_gen)
1729 goto out;
1730 }
1731
1732 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1733 &other_inode, &other_type);
1734 if (ret < 0 && ret != -ENOENT)
1735 goto out;
1736 if (ret) {
1737 ret = 0;
1738 goto out;
1739 }
1740
1741 /*
1742 * Check if the overwritten ref was already processed. If yes, the ref
1743 * was already unlinked/moved, so we can safely assume that we will not
1744 * overwrite anything at this point in time.
1745 */
1746 if (other_inode > sctx->send_progress) {
1747 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1748 who_gen, NULL, NULL, NULL, NULL);
1749 if (ret < 0)
1750 goto out;
1751
1752 ret = 1;
1753 *who_ino = other_inode;
1754 } else {
1755 ret = 0;
1756 }
1757
1758 out:
1759 return ret;
1760 }
1761
1762 /*
1763 * Checks if the ref was overwritten by an already processed inode. This is
1764 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1765 * thus the orphan name needs be used.
1766 * process_recorded_refs also uses it to avoid unlinking of refs that were
1767 * overwritten.
1768 */
1769 static int did_overwrite_ref(struct send_ctx *sctx,
1770 u64 dir, u64 dir_gen,
1771 u64 ino, u64 ino_gen,
1772 const char *name, int name_len)
1773 {
1774 int ret = 0;
1775 u64 gen;
1776 u64 ow_inode;
1777 u8 other_type;
1778
1779 if (!sctx->parent_root)
1780 goto out;
1781
1782 ret = is_inode_existent(sctx, dir, dir_gen);
1783 if (ret <= 0)
1784 goto out;
1785
1786 /* check if the ref was overwritten by another ref */
1787 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1788 &ow_inode, &other_type);
1789 if (ret < 0 && ret != -ENOENT)
1790 goto out;
1791 if (ret) {
1792 /* was never and will never be overwritten */
1793 ret = 0;
1794 goto out;
1795 }
1796
1797 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1798 NULL, NULL);
1799 if (ret < 0)
1800 goto out;
1801
1802 if (ow_inode == ino && gen == ino_gen) {
1803 ret = 0;
1804 goto out;
1805 }
1806
1807 /* we know that it is or will be overwritten. check this now */
1808 if (ow_inode < sctx->send_progress)
1809 ret = 1;
1810 else
1811 ret = 0;
1812
1813 out:
1814 return ret;
1815 }
1816
1817 /*
1818 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1819 * that got overwritten. This is used by process_recorded_refs to determine
1820 * if it has to use the path as returned by get_cur_path or the orphan name.
1821 */
1822 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1823 {
1824 int ret = 0;
1825 struct fs_path *name = NULL;
1826 u64 dir;
1827 u64 dir_gen;
1828
1829 if (!sctx->parent_root)
1830 goto out;
1831
1832 name = fs_path_alloc();
1833 if (!name)
1834 return -ENOMEM;
1835
1836 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1837 if (ret < 0)
1838 goto out;
1839
1840 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1841 name->start, fs_path_len(name));
1842
1843 out:
1844 fs_path_free(name);
1845 return ret;
1846 }
1847
1848 /*
1849 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1850 * so we need to do some special handling in case we have clashes. This function
1851 * takes care of this with the help of name_cache_entry::radix_list.
1852 * In case of error, nce is kfreed.
1853 */
1854 static int name_cache_insert(struct send_ctx *sctx,
1855 struct name_cache_entry *nce)
1856 {
1857 int ret = 0;
1858 struct list_head *nce_head;
1859
1860 nce_head = radix_tree_lookup(&sctx->name_cache,
1861 (unsigned long)nce->ino);
1862 if (!nce_head) {
1863 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1864 if (!nce_head) {
1865 kfree(nce);
1866 return -ENOMEM;
1867 }
1868 INIT_LIST_HEAD(nce_head);
1869
1870 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1871 if (ret < 0) {
1872 kfree(nce_head);
1873 kfree(nce);
1874 return ret;
1875 }
1876 }
1877 list_add_tail(&nce->radix_list, nce_head);
1878 list_add_tail(&nce->list, &sctx->name_cache_list);
1879 sctx->name_cache_size++;
1880
1881 return ret;
1882 }
1883
1884 static void name_cache_delete(struct send_ctx *sctx,
1885 struct name_cache_entry *nce)
1886 {
1887 struct list_head *nce_head;
1888
1889 nce_head = radix_tree_lookup(&sctx->name_cache,
1890 (unsigned long)nce->ino);
1891 BUG_ON(!nce_head);
1892
1893 list_del(&nce->radix_list);
1894 list_del(&nce->list);
1895 sctx->name_cache_size--;
1896
1897 if (list_empty(nce_head)) {
1898 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1899 kfree(nce_head);
1900 }
1901 }
1902
1903 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1904 u64 ino, u64 gen)
1905 {
1906 struct list_head *nce_head;
1907 struct name_cache_entry *cur;
1908
1909 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1910 if (!nce_head)
1911 return NULL;
1912
1913 list_for_each_entry(cur, nce_head, radix_list) {
1914 if (cur->ino == ino && cur->gen == gen)
1915 return cur;
1916 }
1917 return NULL;
1918 }
1919
1920 /*
1921 * Removes the entry from the list and adds it back to the end. This marks the
1922 * entry as recently used so that name_cache_clean_unused does not remove it.
1923 */
1924 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1925 {
1926 list_del(&nce->list);
1927 list_add_tail(&nce->list, &sctx->name_cache_list);
1928 }
1929
1930 /*
1931 * Remove some entries from the beginning of name_cache_list.
1932 */
1933 static void name_cache_clean_unused(struct send_ctx *sctx)
1934 {
1935 struct name_cache_entry *nce;
1936
1937 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1938 return;
1939
1940 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1941 nce = list_entry(sctx->name_cache_list.next,
1942 struct name_cache_entry, list);
1943 name_cache_delete(sctx, nce);
1944 kfree(nce);
1945 }
1946 }
1947
1948 static void name_cache_free(struct send_ctx *sctx)
1949 {
1950 struct name_cache_entry *nce;
1951
1952 while (!list_empty(&sctx->name_cache_list)) {
1953 nce = list_entry(sctx->name_cache_list.next,
1954 struct name_cache_entry, list);
1955 name_cache_delete(sctx, nce);
1956 kfree(nce);
1957 }
1958 }
1959
1960 /*
1961 * Used by get_cur_path for each ref up to the root.
1962 * Returns 0 if it succeeded.
1963 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1964 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1965 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1966 * Returns <0 in case of error.
1967 */
1968 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1969 u64 ino, u64 gen,
1970 int skip_name_cache,
1971 u64 *parent_ino,
1972 u64 *parent_gen,
1973 struct fs_path *dest)
1974 {
1975 int ret;
1976 int nce_ret;
1977 struct btrfs_path *path = NULL;
1978 struct name_cache_entry *nce = NULL;
1979
1980 if (skip_name_cache)
1981 goto get_ref;
1982 /*
1983 * First check if we already did a call to this function with the same
1984 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1985 * return the cached result.
1986 */
1987 nce = name_cache_search(sctx, ino, gen);
1988 if (nce) {
1989 if (ino < sctx->send_progress && nce->need_later_update) {
1990 name_cache_delete(sctx, nce);
1991 kfree(nce);
1992 nce = NULL;
1993 } else {
1994 name_cache_used(sctx, nce);
1995 *parent_ino = nce->parent_ino;
1996 *parent_gen = nce->parent_gen;
1997 ret = fs_path_add(dest, nce->name, nce->name_len);
1998 if (ret < 0)
1999 goto out;
2000 ret = nce->ret;
2001 goto out;
2002 }
2003 }
2004
2005 path = alloc_path_for_send();
2006 if (!path)
2007 return -ENOMEM;
2008
2009 /*
2010 * If the inode is not existent yet, add the orphan name and return 1.
2011 * This should only happen for the parent dir that we determine in
2012 * __record_new_ref
2013 */
2014 ret = is_inode_existent(sctx, ino, gen);
2015 if (ret < 0)
2016 goto out;
2017
2018 if (!ret) {
2019 ret = gen_unique_name(sctx, ino, gen, dest);
2020 if (ret < 0)
2021 goto out;
2022 ret = 1;
2023 goto out_cache;
2024 }
2025
2026 get_ref:
2027 /*
2028 * Depending on whether the inode was already processed or not, use
2029 * send_root or parent_root for ref lookup.
2030 */
2031 if (ino < sctx->send_progress && !skip_name_cache)
2032 ret = get_first_ref(sctx->send_root, ino,
2033 parent_ino, parent_gen, dest);
2034 else
2035 ret = get_first_ref(sctx->parent_root, ino,
2036 parent_ino, parent_gen, dest);
2037 if (ret < 0)
2038 goto out;
2039
2040 /*
2041 * Check if the ref was overwritten by an inode's ref that was processed
2042 * earlier. If yes, treat as orphan and return 1.
2043 */
2044 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2045 dest->start, dest->end - dest->start);
2046 if (ret < 0)
2047 goto out;
2048 if (ret) {
2049 fs_path_reset(dest);
2050 ret = gen_unique_name(sctx, ino, gen, dest);
2051 if (ret < 0)
2052 goto out;
2053 ret = 1;
2054 }
2055 if (skip_name_cache)
2056 goto out;
2057
2058 out_cache:
2059 /*
2060 * Store the result of the lookup in the name cache.
2061 */
2062 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2063 if (!nce) {
2064 ret = -ENOMEM;
2065 goto out;
2066 }
2067
2068 nce->ino = ino;
2069 nce->gen = gen;
2070 nce->parent_ino = *parent_ino;
2071 nce->parent_gen = *parent_gen;
2072 nce->name_len = fs_path_len(dest);
2073 nce->ret = ret;
2074 strcpy(nce->name, dest->start);
2075
2076 if (ino < sctx->send_progress)
2077 nce->need_later_update = 0;
2078 else
2079 nce->need_later_update = 1;
2080
2081 nce_ret = name_cache_insert(sctx, nce);
2082 if (nce_ret < 0)
2083 ret = nce_ret;
2084 name_cache_clean_unused(sctx);
2085
2086 out:
2087 btrfs_free_path(path);
2088 return ret;
2089 }
2090
2091 /*
2092 * Magic happens here. This function returns the first ref to an inode as it
2093 * would look like while receiving the stream at this point in time.
2094 * We walk the path up to the root. For every inode in between, we check if it
2095 * was already processed/sent. If yes, we continue with the parent as found
2096 * in send_root. If not, we continue with the parent as found in parent_root.
2097 * If we encounter an inode that was deleted at this point in time, we use the
2098 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2099 * that were not created yet and overwritten inodes/refs.
2100 *
2101 * When do we have have orphan inodes:
2102 * 1. When an inode is freshly created and thus no valid refs are available yet
2103 * 2. When a directory lost all it's refs (deleted) but still has dir items
2104 * inside which were not processed yet (pending for move/delete). If anyone
2105 * tried to get the path to the dir items, it would get a path inside that
2106 * orphan directory.
2107 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2108 * of an unprocessed inode. If in that case the first ref would be
2109 * overwritten, the overwritten inode gets "orphanized". Later when we
2110 * process this overwritten inode, it is restored at a new place by moving
2111 * the orphan inode.
2112 *
2113 * sctx->send_progress tells this function at which point in time receiving
2114 * would be.
2115 */
2116 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2117 struct fs_path *dest)
2118 {
2119 int ret = 0;
2120 struct fs_path *name = NULL;
2121 u64 parent_inode = 0;
2122 u64 parent_gen = 0;
2123 int stop = 0;
2124 u64 start_ino = ino;
2125 u64 start_gen = gen;
2126 int skip_name_cache = 0;
2127
2128 name = fs_path_alloc();
2129 if (!name) {
2130 ret = -ENOMEM;
2131 goto out;
2132 }
2133
2134 if (is_waiting_for_move(sctx, ino))
2135 skip_name_cache = 1;
2136
2137 again:
2138 dest->reversed = 1;
2139 fs_path_reset(dest);
2140
2141 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2142 fs_path_reset(name);
2143
2144 ret = __get_cur_name_and_parent(sctx, ino, gen, skip_name_cache,
2145 &parent_inode, &parent_gen, name);
2146 if (ret < 0)
2147 goto out;
2148 if (ret)
2149 stop = 1;
2150
2151 if (!skip_name_cache &&
2152 is_waiting_for_move(sctx, parent_inode)) {
2153 ino = start_ino;
2154 gen = start_gen;
2155 stop = 0;
2156 skip_name_cache = 1;
2157 goto again;
2158 }
2159
2160 ret = fs_path_add_path(dest, name);
2161 if (ret < 0)
2162 goto out;
2163
2164 ino = parent_inode;
2165 gen = parent_gen;
2166 }
2167
2168 out:
2169 fs_path_free(name);
2170 if (!ret)
2171 fs_path_unreverse(dest);
2172 return ret;
2173 }
2174
2175 /*
2176 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2177 */
2178 static int send_subvol_begin(struct send_ctx *sctx)
2179 {
2180 int ret;
2181 struct btrfs_root *send_root = sctx->send_root;
2182 struct btrfs_root *parent_root = sctx->parent_root;
2183 struct btrfs_path *path;
2184 struct btrfs_key key;
2185 struct btrfs_root_ref *ref;
2186 struct extent_buffer *leaf;
2187 char *name = NULL;
2188 int namelen;
2189
2190 path = btrfs_alloc_path();
2191 if (!path)
2192 return -ENOMEM;
2193
2194 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2195 if (!name) {
2196 btrfs_free_path(path);
2197 return -ENOMEM;
2198 }
2199
2200 key.objectid = send_root->objectid;
2201 key.type = BTRFS_ROOT_BACKREF_KEY;
2202 key.offset = 0;
2203
2204 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2205 &key, path, 1, 0);
2206 if (ret < 0)
2207 goto out;
2208 if (ret) {
2209 ret = -ENOENT;
2210 goto out;
2211 }
2212
2213 leaf = path->nodes[0];
2214 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2215 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2216 key.objectid != send_root->objectid) {
2217 ret = -ENOENT;
2218 goto out;
2219 }
2220 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2221 namelen = btrfs_root_ref_name_len(leaf, ref);
2222 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2223 btrfs_release_path(path);
2224
2225 if (parent_root) {
2226 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2227 if (ret < 0)
2228 goto out;
2229 } else {
2230 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2231 if (ret < 0)
2232 goto out;
2233 }
2234
2235 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2236 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2237 sctx->send_root->root_item.uuid);
2238 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2239 le64_to_cpu(sctx->send_root->root_item.ctransid));
2240 if (parent_root) {
2241 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2242 sctx->parent_root->root_item.uuid);
2243 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2244 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2245 }
2246
2247 ret = send_cmd(sctx);
2248
2249 tlv_put_failure:
2250 out:
2251 btrfs_free_path(path);
2252 kfree(name);
2253 return ret;
2254 }
2255
2256 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2257 {
2258 int ret = 0;
2259 struct fs_path *p;
2260
2261 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2262
2263 p = fs_path_alloc();
2264 if (!p)
2265 return -ENOMEM;
2266
2267 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2268 if (ret < 0)
2269 goto out;
2270
2271 ret = get_cur_path(sctx, ino, gen, p);
2272 if (ret < 0)
2273 goto out;
2274 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2275 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2276
2277 ret = send_cmd(sctx);
2278
2279 tlv_put_failure:
2280 out:
2281 fs_path_free(p);
2282 return ret;
2283 }
2284
2285 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2286 {
2287 int ret = 0;
2288 struct fs_path *p;
2289
2290 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2291
2292 p = fs_path_alloc();
2293 if (!p)
2294 return -ENOMEM;
2295
2296 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2297 if (ret < 0)
2298 goto out;
2299
2300 ret = get_cur_path(sctx, ino, gen, p);
2301 if (ret < 0)
2302 goto out;
2303 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2304 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2305
2306 ret = send_cmd(sctx);
2307
2308 tlv_put_failure:
2309 out:
2310 fs_path_free(p);
2311 return ret;
2312 }
2313
2314 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2315 {
2316 int ret = 0;
2317 struct fs_path *p;
2318
2319 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2320
2321 p = fs_path_alloc();
2322 if (!p)
2323 return -ENOMEM;
2324
2325 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2326 if (ret < 0)
2327 goto out;
2328
2329 ret = get_cur_path(sctx, ino, gen, p);
2330 if (ret < 0)
2331 goto out;
2332 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2333 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2334 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2335
2336 ret = send_cmd(sctx);
2337
2338 tlv_put_failure:
2339 out:
2340 fs_path_free(p);
2341 return ret;
2342 }
2343
2344 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2345 {
2346 int ret = 0;
2347 struct fs_path *p = NULL;
2348 struct btrfs_inode_item *ii;
2349 struct btrfs_path *path = NULL;
2350 struct extent_buffer *eb;
2351 struct btrfs_key key;
2352 int slot;
2353
2354 verbose_printk("btrfs: send_utimes %llu\n", ino);
2355
2356 p = fs_path_alloc();
2357 if (!p)
2358 return -ENOMEM;
2359
2360 path = alloc_path_for_send();
2361 if (!path) {
2362 ret = -ENOMEM;
2363 goto out;
2364 }
2365
2366 key.objectid = ino;
2367 key.type = BTRFS_INODE_ITEM_KEY;
2368 key.offset = 0;
2369 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2370 if (ret < 0)
2371 goto out;
2372
2373 eb = path->nodes[0];
2374 slot = path->slots[0];
2375 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2376
2377 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2378 if (ret < 0)
2379 goto out;
2380
2381 ret = get_cur_path(sctx, ino, gen, p);
2382 if (ret < 0)
2383 goto out;
2384 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2385 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2386 btrfs_inode_atime(ii));
2387 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2388 btrfs_inode_mtime(ii));
2389 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2390 btrfs_inode_ctime(ii));
2391 /* TODO Add otime support when the otime patches get into upstream */
2392
2393 ret = send_cmd(sctx);
2394
2395 tlv_put_failure:
2396 out:
2397 fs_path_free(p);
2398 btrfs_free_path(path);
2399 return ret;
2400 }
2401
2402 /*
2403 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2404 * a valid path yet because we did not process the refs yet. So, the inode
2405 * is created as orphan.
2406 */
2407 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2408 {
2409 int ret = 0;
2410 struct fs_path *p;
2411 int cmd;
2412 u64 gen;
2413 u64 mode;
2414 u64 rdev;
2415
2416 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2417
2418 p = fs_path_alloc();
2419 if (!p)
2420 return -ENOMEM;
2421
2422 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2423 NULL, &rdev);
2424 if (ret < 0)
2425 goto out;
2426
2427 if (S_ISREG(mode)) {
2428 cmd = BTRFS_SEND_C_MKFILE;
2429 } else if (S_ISDIR(mode)) {
2430 cmd = BTRFS_SEND_C_MKDIR;
2431 } else if (S_ISLNK(mode)) {
2432 cmd = BTRFS_SEND_C_SYMLINK;
2433 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2434 cmd = BTRFS_SEND_C_MKNOD;
2435 } else if (S_ISFIFO(mode)) {
2436 cmd = BTRFS_SEND_C_MKFIFO;
2437 } else if (S_ISSOCK(mode)) {
2438 cmd = BTRFS_SEND_C_MKSOCK;
2439 } else {
2440 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2441 (int)(mode & S_IFMT));
2442 ret = -ENOTSUPP;
2443 goto out;
2444 }
2445
2446 ret = begin_cmd(sctx, cmd);
2447 if (ret < 0)
2448 goto out;
2449
2450 ret = gen_unique_name(sctx, ino, gen, p);
2451 if (ret < 0)
2452 goto out;
2453
2454 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2455 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2456
2457 if (S_ISLNK(mode)) {
2458 fs_path_reset(p);
2459 ret = read_symlink(sctx->send_root, ino, p);
2460 if (ret < 0)
2461 goto out;
2462 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2463 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2464 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2465 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2466 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2467 }
2468
2469 ret = send_cmd(sctx);
2470 if (ret < 0)
2471 goto out;
2472
2473
2474 tlv_put_failure:
2475 out:
2476 fs_path_free(p);
2477 return ret;
2478 }
2479
2480 /*
2481 * We need some special handling for inodes that get processed before the parent
2482 * directory got created. See process_recorded_refs for details.
2483 * This function does the check if we already created the dir out of order.
2484 */
2485 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2486 {
2487 int ret = 0;
2488 struct btrfs_path *path = NULL;
2489 struct btrfs_key key;
2490 struct btrfs_key found_key;
2491 struct btrfs_key di_key;
2492 struct extent_buffer *eb;
2493 struct btrfs_dir_item *di;
2494 int slot;
2495
2496 path = alloc_path_for_send();
2497 if (!path) {
2498 ret = -ENOMEM;
2499 goto out;
2500 }
2501
2502 key.objectid = dir;
2503 key.type = BTRFS_DIR_INDEX_KEY;
2504 key.offset = 0;
2505 while (1) {
2506 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2507 1, 0);
2508 if (ret < 0)
2509 goto out;
2510 if (!ret) {
2511 eb = path->nodes[0];
2512 slot = path->slots[0];
2513 btrfs_item_key_to_cpu(eb, &found_key, slot);
2514 }
2515 if (ret || found_key.objectid != key.objectid ||
2516 found_key.type != key.type) {
2517 ret = 0;
2518 goto out;
2519 }
2520
2521 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2522 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2523
2524 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2525 di_key.objectid < sctx->send_progress) {
2526 ret = 1;
2527 goto out;
2528 }
2529
2530 key.offset = found_key.offset + 1;
2531 btrfs_release_path(path);
2532 }
2533
2534 out:
2535 btrfs_free_path(path);
2536 return ret;
2537 }
2538
2539 /*
2540 * Only creates the inode if it is:
2541 * 1. Not a directory
2542 * 2. Or a directory which was not created already due to out of order
2543 * directories. See did_create_dir and process_recorded_refs for details.
2544 */
2545 static int send_create_inode_if_needed(struct send_ctx *sctx)
2546 {
2547 int ret;
2548
2549 if (S_ISDIR(sctx->cur_inode_mode)) {
2550 ret = did_create_dir(sctx, sctx->cur_ino);
2551 if (ret < 0)
2552 goto out;
2553 if (ret) {
2554 ret = 0;
2555 goto out;
2556 }
2557 }
2558
2559 ret = send_create_inode(sctx, sctx->cur_ino);
2560 if (ret < 0)
2561 goto out;
2562
2563 out:
2564 return ret;
2565 }
2566
2567 struct recorded_ref {
2568 struct list_head list;
2569 char *dir_path;
2570 char *name;
2571 struct fs_path *full_path;
2572 u64 dir;
2573 u64 dir_gen;
2574 int dir_path_len;
2575 int name_len;
2576 };
2577
2578 /*
2579 * We need to process new refs before deleted refs, but compare_tree gives us
2580 * everything mixed. So we first record all refs and later process them.
2581 * This function is a helper to record one ref.
2582 */
2583 static int record_ref(struct list_head *head, u64 dir,
2584 u64 dir_gen, struct fs_path *path)
2585 {
2586 struct recorded_ref *ref;
2587
2588 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2589 if (!ref)
2590 return -ENOMEM;
2591
2592 ref->dir = dir;
2593 ref->dir_gen = dir_gen;
2594 ref->full_path = path;
2595
2596 ref->name = (char *)kbasename(ref->full_path->start);
2597 ref->name_len = ref->full_path->end - ref->name;
2598 ref->dir_path = ref->full_path->start;
2599 if (ref->name == ref->full_path->start)
2600 ref->dir_path_len = 0;
2601 else
2602 ref->dir_path_len = ref->full_path->end -
2603 ref->full_path->start - 1 - ref->name_len;
2604
2605 list_add_tail(&ref->list, head);
2606 return 0;
2607 }
2608
2609 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2610 {
2611 struct recorded_ref *new;
2612
2613 new = kmalloc(sizeof(*ref), GFP_NOFS);
2614 if (!new)
2615 return -ENOMEM;
2616
2617 new->dir = ref->dir;
2618 new->dir_gen = ref->dir_gen;
2619 new->full_path = NULL;
2620 INIT_LIST_HEAD(&new->list);
2621 list_add_tail(&new->list, list);
2622 return 0;
2623 }
2624
2625 static void __free_recorded_refs(struct list_head *head)
2626 {
2627 struct recorded_ref *cur;
2628
2629 while (!list_empty(head)) {
2630 cur = list_entry(head->next, struct recorded_ref, list);
2631 fs_path_free(cur->full_path);
2632 list_del(&cur->list);
2633 kfree(cur);
2634 }
2635 }
2636
2637 static void free_recorded_refs(struct send_ctx *sctx)
2638 {
2639 __free_recorded_refs(&sctx->new_refs);
2640 __free_recorded_refs(&sctx->deleted_refs);
2641 }
2642
2643 /*
2644 * Renames/moves a file/dir to its orphan name. Used when the first
2645 * ref of an unprocessed inode gets overwritten and for all non empty
2646 * directories.
2647 */
2648 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2649 struct fs_path *path)
2650 {
2651 int ret;
2652 struct fs_path *orphan;
2653
2654 orphan = fs_path_alloc();
2655 if (!orphan)
2656 return -ENOMEM;
2657
2658 ret = gen_unique_name(sctx, ino, gen, orphan);
2659 if (ret < 0)
2660 goto out;
2661
2662 ret = send_rename(sctx, path, orphan);
2663
2664 out:
2665 fs_path_free(orphan);
2666 return ret;
2667 }
2668
2669 /*
2670 * Returns 1 if a directory can be removed at this point in time.
2671 * We check this by iterating all dir items and checking if the inode behind
2672 * the dir item was already processed.
2673 */
2674 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2675 {
2676 int ret = 0;
2677 struct btrfs_root *root = sctx->parent_root;
2678 struct btrfs_path *path;
2679 struct btrfs_key key;
2680 struct btrfs_key found_key;
2681 struct btrfs_key loc;
2682 struct btrfs_dir_item *di;
2683
2684 /*
2685 * Don't try to rmdir the top/root subvolume dir.
2686 */
2687 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2688 return 0;
2689
2690 path = alloc_path_for_send();
2691 if (!path)
2692 return -ENOMEM;
2693
2694 key.objectid = dir;
2695 key.type = BTRFS_DIR_INDEX_KEY;
2696 key.offset = 0;
2697
2698 while (1) {
2699 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2700 if (ret < 0)
2701 goto out;
2702 if (!ret) {
2703 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2704 path->slots[0]);
2705 }
2706 if (ret || found_key.objectid != key.objectid ||
2707 found_key.type != key.type) {
2708 break;
2709 }
2710
2711 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2712 struct btrfs_dir_item);
2713 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2714
2715 if (loc.objectid > send_progress) {
2716 ret = 0;
2717 goto out;
2718 }
2719
2720 btrfs_release_path(path);
2721 key.offset = found_key.offset + 1;
2722 }
2723
2724 ret = 1;
2725
2726 out:
2727 btrfs_free_path(path);
2728 return ret;
2729 }
2730
2731 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2732 {
2733 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2734 struct waiting_dir_move *entry;
2735
2736 while (n) {
2737 entry = rb_entry(n, struct waiting_dir_move, node);
2738 if (ino < entry->ino)
2739 n = n->rb_left;
2740 else if (ino > entry->ino)
2741 n = n->rb_right;
2742 else
2743 return 1;
2744 }
2745 return 0;
2746 }
2747
2748 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2749 {
2750 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2751 struct rb_node *parent = NULL;
2752 struct waiting_dir_move *entry, *dm;
2753
2754 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2755 if (!dm)
2756 return -ENOMEM;
2757 dm->ino = ino;
2758
2759 while (*p) {
2760 parent = *p;
2761 entry = rb_entry(parent, struct waiting_dir_move, node);
2762 if (ino < entry->ino) {
2763 p = &(*p)->rb_left;
2764 } else if (ino > entry->ino) {
2765 p = &(*p)->rb_right;
2766 } else {
2767 kfree(dm);
2768 return -EEXIST;
2769 }
2770 }
2771
2772 rb_link_node(&dm->node, parent, p);
2773 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2774 return 0;
2775 }
2776
2777 #ifdef CONFIG_BTRFS_ASSERT
2778
2779 static int del_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2780 {
2781 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2782 struct waiting_dir_move *entry;
2783
2784 while (n) {
2785 entry = rb_entry(n, struct waiting_dir_move, node);
2786 if (ino < entry->ino) {
2787 n = n->rb_left;
2788 } else if (ino > entry->ino) {
2789 n = n->rb_right;
2790 } else {
2791 rb_erase(&entry->node, &sctx->waiting_dir_moves);
2792 kfree(entry);
2793 return 0;
2794 }
2795 }
2796 return -ENOENT;
2797 }
2798
2799 #endif
2800
2801 static int add_pending_dir_move(struct send_ctx *sctx, u64 parent_ino)
2802 {
2803 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2804 struct rb_node *parent = NULL;
2805 struct pending_dir_move *entry, *pm;
2806 struct recorded_ref *cur;
2807 int exists = 0;
2808 int ret;
2809
2810 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2811 if (!pm)
2812 return -ENOMEM;
2813 pm->parent_ino = parent_ino;
2814 pm->ino = sctx->cur_ino;
2815 pm->gen = sctx->cur_inode_gen;
2816 INIT_LIST_HEAD(&pm->list);
2817 INIT_LIST_HEAD(&pm->update_refs);
2818 RB_CLEAR_NODE(&pm->node);
2819
2820 while (*p) {
2821 parent = *p;
2822 entry = rb_entry(parent, struct pending_dir_move, node);
2823 if (parent_ino < entry->parent_ino) {
2824 p = &(*p)->rb_left;
2825 } else if (parent_ino > entry->parent_ino) {
2826 p = &(*p)->rb_right;
2827 } else {
2828 exists = 1;
2829 break;
2830 }
2831 }
2832
2833 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2834 ret = dup_ref(cur, &pm->update_refs);
2835 if (ret < 0)
2836 goto out;
2837 }
2838 list_for_each_entry(cur, &sctx->new_refs, list) {
2839 ret = dup_ref(cur, &pm->update_refs);
2840 if (ret < 0)
2841 goto out;
2842 }
2843
2844 ret = add_waiting_dir_move(sctx, pm->ino);
2845 if (ret)
2846 goto out;
2847
2848 if (exists) {
2849 list_add_tail(&pm->list, &entry->list);
2850 } else {
2851 rb_link_node(&pm->node, parent, p);
2852 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
2853 }
2854 ret = 0;
2855 out:
2856 if (ret) {
2857 __free_recorded_refs(&pm->update_refs);
2858 kfree(pm);
2859 }
2860 return ret;
2861 }
2862
2863 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
2864 u64 parent_ino)
2865 {
2866 struct rb_node *n = sctx->pending_dir_moves.rb_node;
2867 struct pending_dir_move *entry;
2868
2869 while (n) {
2870 entry = rb_entry(n, struct pending_dir_move, node);
2871 if (parent_ino < entry->parent_ino)
2872 n = n->rb_left;
2873 else if (parent_ino > entry->parent_ino)
2874 n = n->rb_right;
2875 else
2876 return entry;
2877 }
2878 return NULL;
2879 }
2880
2881 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
2882 {
2883 struct fs_path *from_path = NULL;
2884 struct fs_path *to_path = NULL;
2885 u64 orig_progress = sctx->send_progress;
2886 struct recorded_ref *cur;
2887 int ret;
2888
2889 from_path = fs_path_alloc();
2890 if (!from_path)
2891 return -ENOMEM;
2892
2893 sctx->send_progress = pm->ino;
2894 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
2895 if (ret < 0)
2896 goto out;
2897
2898 to_path = fs_path_alloc();
2899 if (!to_path) {
2900 ret = -ENOMEM;
2901 goto out;
2902 }
2903
2904 sctx->send_progress = sctx->cur_ino + 1;
2905 ASSERT(del_waiting_dir_move(sctx, pm->ino) == 0);
2906 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
2907 if (ret < 0)
2908 goto out;
2909
2910 ret = send_rename(sctx, from_path, to_path);
2911 if (ret < 0)
2912 goto out;
2913
2914 ret = send_utimes(sctx, pm->ino, pm->gen);
2915 if (ret < 0)
2916 goto out;
2917
2918 /*
2919 * After rename/move, need to update the utimes of both new parent(s)
2920 * and old parent(s).
2921 */
2922 list_for_each_entry(cur, &pm->update_refs, list) {
2923 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
2924 if (ret < 0)
2925 goto out;
2926 }
2927
2928 out:
2929 fs_path_free(from_path);
2930 fs_path_free(to_path);
2931 sctx->send_progress = orig_progress;
2932
2933 return ret;
2934 }
2935
2936 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
2937 {
2938 if (!list_empty(&m->list))
2939 list_del(&m->list);
2940 if (!RB_EMPTY_NODE(&m->node))
2941 rb_erase(&m->node, &sctx->pending_dir_moves);
2942 __free_recorded_refs(&m->update_refs);
2943 kfree(m);
2944 }
2945
2946 static void tail_append_pending_moves(struct pending_dir_move *moves,
2947 struct list_head *stack)
2948 {
2949 if (list_empty(&moves->list)) {
2950 list_add_tail(&moves->list, stack);
2951 } else {
2952 LIST_HEAD(list);
2953 list_splice_init(&moves->list, &list);
2954 list_add_tail(&moves->list, stack);
2955 list_splice_tail(&list, stack);
2956 }
2957 }
2958
2959 static int apply_children_dir_moves(struct send_ctx *sctx)
2960 {
2961 struct pending_dir_move *pm;
2962 struct list_head stack;
2963 u64 parent_ino = sctx->cur_ino;
2964 int ret = 0;
2965
2966 pm = get_pending_dir_moves(sctx, parent_ino);
2967 if (!pm)
2968 return 0;
2969
2970 INIT_LIST_HEAD(&stack);
2971 tail_append_pending_moves(pm, &stack);
2972
2973 while (!list_empty(&stack)) {
2974 pm = list_first_entry(&stack, struct pending_dir_move, list);
2975 parent_ino = pm->ino;
2976 ret = apply_dir_move(sctx, pm);
2977 free_pending_move(sctx, pm);
2978 if (ret)
2979 goto out;
2980 pm = get_pending_dir_moves(sctx, parent_ino);
2981 if (pm)
2982 tail_append_pending_moves(pm, &stack);
2983 }
2984 return 0;
2985
2986 out:
2987 while (!list_empty(&stack)) {
2988 pm = list_first_entry(&stack, struct pending_dir_move, list);
2989 free_pending_move(sctx, pm);
2990 }
2991 return ret;
2992 }
2993
2994 static int wait_for_parent_move(struct send_ctx *sctx,
2995 struct recorded_ref *parent_ref)
2996 {
2997 int ret;
2998 u64 ino = parent_ref->dir;
2999 u64 parent_ino_before, parent_ino_after;
3000 u64 new_gen, old_gen;
3001 struct fs_path *path_before = NULL;
3002 struct fs_path *path_after = NULL;
3003 int len1, len2;
3004
3005 if (parent_ref->dir <= sctx->cur_ino)
3006 return 0;
3007
3008 if (is_waiting_for_move(sctx, ino))
3009 return 1;
3010
3011 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3012 NULL, NULL, NULL, NULL);
3013 if (ret == -ENOENT)
3014 return 0;
3015 else if (ret < 0)
3016 return ret;
3017
3018 ret = get_inode_info(sctx->send_root, ino, NULL, &new_gen,
3019 NULL, NULL, NULL, NULL);
3020 if (ret < 0)
3021 return ret;
3022
3023 if (new_gen != old_gen)
3024 return 0;
3025
3026 path_before = fs_path_alloc();
3027 if (!path_before)
3028 return -ENOMEM;
3029
3030 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3031 NULL, path_before);
3032 if (ret == -ENOENT) {
3033 ret = 0;
3034 goto out;
3035 } else if (ret < 0) {
3036 goto out;
3037 }
3038
3039 path_after = fs_path_alloc();
3040 if (!path_after) {
3041 ret = -ENOMEM;
3042 goto out;
3043 }
3044
3045 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3046 NULL, path_after);
3047 if (ret == -ENOENT) {
3048 ret = 0;
3049 goto out;
3050 } else if (ret < 0) {
3051 goto out;
3052 }
3053
3054 len1 = fs_path_len(path_before);
3055 len2 = fs_path_len(path_after);
3056 if ((parent_ino_before != parent_ino_after) && (len1 != len2 ||
3057 memcmp(path_before->start, path_after->start, len1))) {
3058 ret = 1;
3059 goto out;
3060 }
3061 ret = 0;
3062
3063 out:
3064 fs_path_free(path_before);
3065 fs_path_free(path_after);
3066
3067 return ret;
3068 }
3069
3070 /*
3071 * This does all the move/link/unlink/rmdir magic.
3072 */
3073 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3074 {
3075 int ret = 0;
3076 struct recorded_ref *cur;
3077 struct recorded_ref *cur2;
3078 struct list_head check_dirs;
3079 struct fs_path *valid_path = NULL;
3080 u64 ow_inode = 0;
3081 u64 ow_gen;
3082 int did_overwrite = 0;
3083 int is_orphan = 0;
3084
3085 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3086
3087 /*
3088 * This should never happen as the root dir always has the same ref
3089 * which is always '..'
3090 */
3091 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3092 INIT_LIST_HEAD(&check_dirs);
3093
3094 valid_path = fs_path_alloc();
3095 if (!valid_path) {
3096 ret = -ENOMEM;
3097 goto out;
3098 }
3099
3100 /*
3101 * First, check if the first ref of the current inode was overwritten
3102 * before. If yes, we know that the current inode was already orphanized
3103 * and thus use the orphan name. If not, we can use get_cur_path to
3104 * get the path of the first ref as it would like while receiving at
3105 * this point in time.
3106 * New inodes are always orphan at the beginning, so force to use the
3107 * orphan name in this case.
3108 * The first ref is stored in valid_path and will be updated if it
3109 * gets moved around.
3110 */
3111 if (!sctx->cur_inode_new) {
3112 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3113 sctx->cur_inode_gen);
3114 if (ret < 0)
3115 goto out;
3116 if (ret)
3117 did_overwrite = 1;
3118 }
3119 if (sctx->cur_inode_new || did_overwrite) {
3120 ret = gen_unique_name(sctx, sctx->cur_ino,
3121 sctx->cur_inode_gen, valid_path);
3122 if (ret < 0)
3123 goto out;
3124 is_orphan = 1;
3125 } else {
3126 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3127 valid_path);
3128 if (ret < 0)
3129 goto out;
3130 }
3131
3132 list_for_each_entry(cur, &sctx->new_refs, list) {
3133 /*
3134 * We may have refs where the parent directory does not exist
3135 * yet. This happens if the parent directories inum is higher
3136 * the the current inum. To handle this case, we create the
3137 * parent directory out of order. But we need to check if this
3138 * did already happen before due to other refs in the same dir.
3139 */
3140 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3141 if (ret < 0)
3142 goto out;
3143 if (ret == inode_state_will_create) {
3144 ret = 0;
3145 /*
3146 * First check if any of the current inodes refs did
3147 * already create the dir.
3148 */
3149 list_for_each_entry(cur2, &sctx->new_refs, list) {
3150 if (cur == cur2)
3151 break;
3152 if (cur2->dir == cur->dir) {
3153 ret = 1;
3154 break;
3155 }
3156 }
3157
3158 /*
3159 * If that did not happen, check if a previous inode
3160 * did already create the dir.
3161 */
3162 if (!ret)
3163 ret = did_create_dir(sctx, cur->dir);
3164 if (ret < 0)
3165 goto out;
3166 if (!ret) {
3167 ret = send_create_inode(sctx, cur->dir);
3168 if (ret < 0)
3169 goto out;
3170 }
3171 }
3172
3173 /*
3174 * Check if this new ref would overwrite the first ref of
3175 * another unprocessed inode. If yes, orphanize the
3176 * overwritten inode. If we find an overwritten ref that is
3177 * not the first ref, simply unlink it.
3178 */
3179 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3180 cur->name, cur->name_len,
3181 &ow_inode, &ow_gen);
3182 if (ret < 0)
3183 goto out;
3184 if (ret) {
3185 ret = is_first_ref(sctx->parent_root,
3186 ow_inode, cur->dir, cur->name,
3187 cur->name_len);
3188 if (ret < 0)
3189 goto out;
3190 if (ret) {
3191 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3192 cur->full_path);
3193 if (ret < 0)
3194 goto out;
3195 } else {
3196 ret = send_unlink(sctx, cur->full_path);
3197 if (ret < 0)
3198 goto out;
3199 }
3200 }
3201
3202 /*
3203 * link/move the ref to the new place. If we have an orphan
3204 * inode, move it and update valid_path. If not, link or move
3205 * it depending on the inode mode.
3206 */
3207 if (is_orphan) {
3208 ret = send_rename(sctx, valid_path, cur->full_path);
3209 if (ret < 0)
3210 goto out;
3211 is_orphan = 0;
3212 ret = fs_path_copy(valid_path, cur->full_path);
3213 if (ret < 0)
3214 goto out;
3215 } else {
3216 if (S_ISDIR(sctx->cur_inode_mode)) {
3217 /*
3218 * Dirs can't be linked, so move it. For moved
3219 * dirs, we always have one new and one deleted
3220 * ref. The deleted ref is ignored later.
3221 */
3222 if (wait_for_parent_move(sctx, cur)) {
3223 ret = add_pending_dir_move(sctx,
3224 cur->dir);
3225 *pending_move = 1;
3226 } else {
3227 ret = send_rename(sctx, valid_path,
3228 cur->full_path);
3229 if (!ret)
3230 ret = fs_path_copy(valid_path,
3231 cur->full_path);
3232 }
3233 if (ret < 0)
3234 goto out;
3235 } else {
3236 ret = send_link(sctx, cur->full_path,
3237 valid_path);
3238 if (ret < 0)
3239 goto out;
3240 }
3241 }
3242 ret = dup_ref(cur, &check_dirs);
3243 if (ret < 0)
3244 goto out;
3245 }
3246
3247 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3248 /*
3249 * Check if we can already rmdir the directory. If not,
3250 * orphanize it. For every dir item inside that gets deleted
3251 * later, we do this check again and rmdir it then if possible.
3252 * See the use of check_dirs for more details.
3253 */
3254 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
3255 if (ret < 0)
3256 goto out;
3257 if (ret) {
3258 ret = send_rmdir(sctx, valid_path);
3259 if (ret < 0)
3260 goto out;
3261 } else if (!is_orphan) {
3262 ret = orphanize_inode(sctx, sctx->cur_ino,
3263 sctx->cur_inode_gen, valid_path);
3264 if (ret < 0)
3265 goto out;
3266 is_orphan = 1;
3267 }
3268
3269 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3270 ret = dup_ref(cur, &check_dirs);
3271 if (ret < 0)
3272 goto out;
3273 }
3274 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3275 !list_empty(&sctx->deleted_refs)) {
3276 /*
3277 * We have a moved dir. Add the old parent to check_dirs
3278 */
3279 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3280 list);
3281 ret = dup_ref(cur, &check_dirs);
3282 if (ret < 0)
3283 goto out;
3284 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3285 /*
3286 * We have a non dir inode. Go through all deleted refs and
3287 * unlink them if they were not already overwritten by other
3288 * inodes.
3289 */
3290 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3291 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3292 sctx->cur_ino, sctx->cur_inode_gen,
3293 cur->name, cur->name_len);
3294 if (ret < 0)
3295 goto out;
3296 if (!ret) {
3297 ret = send_unlink(sctx, cur->full_path);
3298 if (ret < 0)
3299 goto out;
3300 }
3301 ret = dup_ref(cur, &check_dirs);
3302 if (ret < 0)
3303 goto out;
3304 }
3305 /*
3306 * If the inode is still orphan, unlink the orphan. This may
3307 * happen when a previous inode did overwrite the first ref
3308 * of this inode and no new refs were added for the current
3309 * inode. Unlinking does not mean that the inode is deleted in
3310 * all cases. There may still be links to this inode in other
3311 * places.
3312 */
3313 if (is_orphan) {
3314 ret = send_unlink(sctx, valid_path);
3315 if (ret < 0)
3316 goto out;
3317 }
3318 }
3319
3320 /*
3321 * We did collect all parent dirs where cur_inode was once located. We
3322 * now go through all these dirs and check if they are pending for
3323 * deletion and if it's finally possible to perform the rmdir now.
3324 * We also update the inode stats of the parent dirs here.
3325 */
3326 list_for_each_entry(cur, &check_dirs, list) {
3327 /*
3328 * In case we had refs into dirs that were not processed yet,
3329 * we don't need to do the utime and rmdir logic for these dirs.
3330 * The dir will be processed later.
3331 */
3332 if (cur->dir > sctx->cur_ino)
3333 continue;
3334
3335 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3336 if (ret < 0)
3337 goto out;
3338
3339 if (ret == inode_state_did_create ||
3340 ret == inode_state_no_change) {
3341 /* TODO delayed utimes */
3342 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3343 if (ret < 0)
3344 goto out;
3345 } else if (ret == inode_state_did_delete) {
3346 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
3347 if (ret < 0)
3348 goto out;
3349 if (ret) {
3350 ret = get_cur_path(sctx, cur->dir,
3351 cur->dir_gen, valid_path);
3352 if (ret < 0)
3353 goto out;
3354 ret = send_rmdir(sctx, valid_path);
3355 if (ret < 0)
3356 goto out;
3357 }
3358 }
3359 }
3360
3361 ret = 0;
3362
3363 out:
3364 __free_recorded_refs(&check_dirs);
3365 free_recorded_refs(sctx);
3366 fs_path_free(valid_path);
3367 return ret;
3368 }
3369
3370 static int __record_new_ref(int num, u64 dir, int index,
3371 struct fs_path *name,
3372 void *ctx)
3373 {
3374 int ret = 0;
3375 struct send_ctx *sctx = ctx;
3376 struct fs_path *p;
3377 u64 gen;
3378
3379 p = fs_path_alloc();
3380 if (!p)
3381 return -ENOMEM;
3382
3383 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
3384 NULL, NULL);
3385 if (ret < 0)
3386 goto out;
3387
3388 ret = get_cur_path(sctx, dir, gen, p);
3389 if (ret < 0)
3390 goto out;
3391 ret = fs_path_add_path(p, name);
3392 if (ret < 0)
3393 goto out;
3394
3395 ret = record_ref(&sctx->new_refs, dir, gen, p);
3396
3397 out:
3398 if (ret)
3399 fs_path_free(p);
3400 return ret;
3401 }
3402
3403 static int __record_deleted_ref(int num, u64 dir, int index,
3404 struct fs_path *name,
3405 void *ctx)
3406 {
3407 int ret = 0;
3408 struct send_ctx *sctx = ctx;
3409 struct fs_path *p;
3410 u64 gen;
3411
3412 p = fs_path_alloc();
3413 if (!p)
3414 return -ENOMEM;
3415
3416 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3417 NULL, NULL);
3418 if (ret < 0)
3419 goto out;
3420
3421 ret = get_cur_path(sctx, dir, gen, p);
3422 if (ret < 0)
3423 goto out;
3424 ret = fs_path_add_path(p, name);
3425 if (ret < 0)
3426 goto out;
3427
3428 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3429
3430 out:
3431 if (ret)
3432 fs_path_free(p);
3433 return ret;
3434 }
3435
3436 static int record_new_ref(struct send_ctx *sctx)
3437 {
3438 int ret;
3439
3440 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3441 sctx->cmp_key, 0, __record_new_ref, sctx);
3442 if (ret < 0)
3443 goto out;
3444 ret = 0;
3445
3446 out:
3447 return ret;
3448 }
3449
3450 static int record_deleted_ref(struct send_ctx *sctx)
3451 {
3452 int ret;
3453
3454 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3455 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3456 if (ret < 0)
3457 goto out;
3458 ret = 0;
3459
3460 out:
3461 return ret;
3462 }
3463
3464 struct find_ref_ctx {
3465 u64 dir;
3466 u64 dir_gen;
3467 struct btrfs_root *root;
3468 struct fs_path *name;
3469 int found_idx;
3470 };
3471
3472 static int __find_iref(int num, u64 dir, int index,
3473 struct fs_path *name,
3474 void *ctx_)
3475 {
3476 struct find_ref_ctx *ctx = ctx_;
3477 u64 dir_gen;
3478 int ret;
3479
3480 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3481 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3482 /*
3483 * To avoid doing extra lookups we'll only do this if everything
3484 * else matches.
3485 */
3486 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3487 NULL, NULL, NULL);
3488 if (ret)
3489 return ret;
3490 if (dir_gen != ctx->dir_gen)
3491 return 0;
3492 ctx->found_idx = num;
3493 return 1;
3494 }
3495 return 0;
3496 }
3497
3498 static int find_iref(struct btrfs_root *root,
3499 struct btrfs_path *path,
3500 struct btrfs_key *key,
3501 u64 dir, u64 dir_gen, struct fs_path *name)
3502 {
3503 int ret;
3504 struct find_ref_ctx ctx;
3505
3506 ctx.dir = dir;
3507 ctx.name = name;
3508 ctx.dir_gen = dir_gen;
3509 ctx.found_idx = -1;
3510 ctx.root = root;
3511
3512 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3513 if (ret < 0)
3514 return ret;
3515
3516 if (ctx.found_idx == -1)
3517 return -ENOENT;
3518
3519 return ctx.found_idx;
3520 }
3521
3522 static int __record_changed_new_ref(int num, u64 dir, int index,
3523 struct fs_path *name,
3524 void *ctx)
3525 {
3526 u64 dir_gen;
3527 int ret;
3528 struct send_ctx *sctx = ctx;
3529
3530 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3531 NULL, NULL, NULL);
3532 if (ret)
3533 return ret;
3534
3535 ret = find_iref(sctx->parent_root, sctx->right_path,
3536 sctx->cmp_key, dir, dir_gen, name);
3537 if (ret == -ENOENT)
3538 ret = __record_new_ref(num, dir, index, name, sctx);
3539 else if (ret > 0)
3540 ret = 0;
3541
3542 return ret;
3543 }
3544
3545 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3546 struct fs_path *name,
3547 void *ctx)
3548 {
3549 u64 dir_gen;
3550 int ret;
3551 struct send_ctx *sctx = ctx;
3552
3553 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3554 NULL, NULL, NULL);
3555 if (ret)
3556 return ret;
3557
3558 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
3559 dir, dir_gen, name);
3560 if (ret == -ENOENT)
3561 ret = __record_deleted_ref(num, dir, index, name, sctx);
3562 else if (ret > 0)
3563 ret = 0;
3564
3565 return ret;
3566 }
3567
3568 static int record_changed_ref(struct send_ctx *sctx)
3569 {
3570 int ret = 0;
3571
3572 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3573 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3574 if (ret < 0)
3575 goto out;
3576 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3577 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3578 if (ret < 0)
3579 goto out;
3580 ret = 0;
3581
3582 out:
3583 return ret;
3584 }
3585
3586 /*
3587 * Record and process all refs at once. Needed when an inode changes the
3588 * generation number, which means that it was deleted and recreated.
3589 */
3590 static int process_all_refs(struct send_ctx *sctx,
3591 enum btrfs_compare_tree_result cmd)
3592 {
3593 int ret;
3594 struct btrfs_root *root;
3595 struct btrfs_path *path;
3596 struct btrfs_key key;
3597 struct btrfs_key found_key;
3598 struct extent_buffer *eb;
3599 int slot;
3600 iterate_inode_ref_t cb;
3601 int pending_move = 0;
3602
3603 path = alloc_path_for_send();
3604 if (!path)
3605 return -ENOMEM;
3606
3607 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3608 root = sctx->send_root;
3609 cb = __record_new_ref;
3610 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3611 root = sctx->parent_root;
3612 cb = __record_deleted_ref;
3613 } else {
3614 BUG();
3615 }
3616
3617 key.objectid = sctx->cmp_key->objectid;
3618 key.type = BTRFS_INODE_REF_KEY;
3619 key.offset = 0;
3620 while (1) {
3621 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3622 if (ret < 0)
3623 goto out;
3624 if (ret)
3625 break;
3626
3627 eb = path->nodes[0];
3628 slot = path->slots[0];
3629 btrfs_item_key_to_cpu(eb, &found_key, slot);
3630
3631 if (found_key.objectid != key.objectid ||
3632 (found_key.type != BTRFS_INODE_REF_KEY &&
3633 found_key.type != BTRFS_INODE_EXTREF_KEY))
3634 break;
3635
3636 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
3637 btrfs_release_path(path);
3638 if (ret < 0)
3639 goto out;
3640
3641 key.offset = found_key.offset + 1;
3642 }
3643 btrfs_release_path(path);
3644
3645 ret = process_recorded_refs(sctx, &pending_move);
3646 /* Only applicable to an incremental send. */
3647 ASSERT(pending_move == 0);
3648
3649 out:
3650 btrfs_free_path(path);
3651 return ret;
3652 }
3653
3654 static int send_set_xattr(struct send_ctx *sctx,
3655 struct fs_path *path,
3656 const char *name, int name_len,
3657 const char *data, int data_len)
3658 {
3659 int ret = 0;
3660
3661 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3662 if (ret < 0)
3663 goto out;
3664
3665 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3666 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3667 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3668
3669 ret = send_cmd(sctx);
3670
3671 tlv_put_failure:
3672 out:
3673 return ret;
3674 }
3675
3676 static int send_remove_xattr(struct send_ctx *sctx,
3677 struct fs_path *path,
3678 const char *name, int name_len)
3679 {
3680 int ret = 0;
3681
3682 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3683 if (ret < 0)
3684 goto out;
3685
3686 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3687 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3688
3689 ret = send_cmd(sctx);
3690
3691 tlv_put_failure:
3692 out:
3693 return ret;
3694 }
3695
3696 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3697 const char *name, int name_len,
3698 const char *data, int data_len,
3699 u8 type, void *ctx)
3700 {
3701 int ret;
3702 struct send_ctx *sctx = ctx;
3703 struct fs_path *p;
3704 posix_acl_xattr_header dummy_acl;
3705
3706 p = fs_path_alloc();
3707 if (!p)
3708 return -ENOMEM;
3709
3710 /*
3711 * This hack is needed because empty acl's are stored as zero byte
3712 * data in xattrs. Problem with that is, that receiving these zero byte
3713 * acl's will fail later. To fix this, we send a dummy acl list that
3714 * only contains the version number and no entries.
3715 */
3716 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3717 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3718 if (data_len == 0) {
3719 dummy_acl.a_version =
3720 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3721 data = (char *)&dummy_acl;
3722 data_len = sizeof(dummy_acl);
3723 }
3724 }
3725
3726 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3727 if (ret < 0)
3728 goto out;
3729
3730 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3731
3732 out:
3733 fs_path_free(p);
3734 return ret;
3735 }
3736
3737 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3738 const char *name, int name_len,
3739 const char *data, int data_len,
3740 u8 type, void *ctx)
3741 {
3742 int ret;
3743 struct send_ctx *sctx = ctx;
3744 struct fs_path *p;
3745
3746 p = fs_path_alloc();
3747 if (!p)
3748 return -ENOMEM;
3749
3750 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3751 if (ret < 0)
3752 goto out;
3753
3754 ret = send_remove_xattr(sctx, p, name, name_len);
3755
3756 out:
3757 fs_path_free(p);
3758 return ret;
3759 }
3760
3761 static int process_new_xattr(struct send_ctx *sctx)
3762 {
3763 int ret = 0;
3764
3765 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3766 sctx->cmp_key, __process_new_xattr, sctx);
3767
3768 return ret;
3769 }
3770
3771 static int process_deleted_xattr(struct send_ctx *sctx)
3772 {
3773 int ret;
3774
3775 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3776 sctx->cmp_key, __process_deleted_xattr, sctx);
3777
3778 return ret;
3779 }
3780
3781 struct find_xattr_ctx {
3782 const char *name;
3783 int name_len;
3784 int found_idx;
3785 char *found_data;
3786 int found_data_len;
3787 };
3788
3789 static int __find_xattr(int num, struct btrfs_key *di_key,
3790 const char *name, int name_len,
3791 const char *data, int data_len,
3792 u8 type, void *vctx)
3793 {
3794 struct find_xattr_ctx *ctx = vctx;
3795
3796 if (name_len == ctx->name_len &&
3797 strncmp(name, ctx->name, name_len) == 0) {
3798 ctx->found_idx = num;
3799 ctx->found_data_len = data_len;
3800 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
3801 if (!ctx->found_data)
3802 return -ENOMEM;
3803 return 1;
3804 }
3805 return 0;
3806 }
3807
3808 static int find_xattr(struct btrfs_root *root,
3809 struct btrfs_path *path,
3810 struct btrfs_key *key,
3811 const char *name, int name_len,
3812 char **data, int *data_len)
3813 {
3814 int ret;
3815 struct find_xattr_ctx ctx;
3816
3817 ctx.name = name;
3818 ctx.name_len = name_len;
3819 ctx.found_idx = -1;
3820 ctx.found_data = NULL;
3821 ctx.found_data_len = 0;
3822
3823 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
3824 if (ret < 0)
3825 return ret;
3826
3827 if (ctx.found_idx == -1)
3828 return -ENOENT;
3829 if (data) {
3830 *data = ctx.found_data;
3831 *data_len = ctx.found_data_len;
3832 } else {
3833 kfree(ctx.found_data);
3834 }
3835 return ctx.found_idx;
3836 }
3837
3838
3839 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3840 const char *name, int name_len,
3841 const char *data, int data_len,
3842 u8 type, void *ctx)
3843 {
3844 int ret;
3845 struct send_ctx *sctx = ctx;
3846 char *found_data = NULL;
3847 int found_data_len = 0;
3848
3849 ret = find_xattr(sctx->parent_root, sctx->right_path,
3850 sctx->cmp_key, name, name_len, &found_data,
3851 &found_data_len);
3852 if (ret == -ENOENT) {
3853 ret = __process_new_xattr(num, di_key, name, name_len, data,
3854 data_len, type, ctx);
3855 } else if (ret >= 0) {
3856 if (data_len != found_data_len ||
3857 memcmp(data, found_data, data_len)) {
3858 ret = __process_new_xattr(num, di_key, name, name_len,
3859 data, data_len, type, ctx);
3860 } else {
3861 ret = 0;
3862 }
3863 }
3864
3865 kfree(found_data);
3866 return ret;
3867 }
3868
3869 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3870 const char *name, int name_len,
3871 const char *data, int data_len,
3872 u8 type, void *ctx)
3873 {
3874 int ret;
3875 struct send_ctx *sctx = ctx;
3876
3877 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3878 name, name_len, NULL, NULL);
3879 if (ret == -ENOENT)
3880 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3881 data_len, type, ctx);
3882 else if (ret >= 0)
3883 ret = 0;
3884
3885 return ret;
3886 }
3887
3888 static int process_changed_xattr(struct send_ctx *sctx)
3889 {
3890 int ret = 0;
3891
3892 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3893 sctx->cmp_key, __process_changed_new_xattr, sctx);
3894 if (ret < 0)
3895 goto out;
3896 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3897 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3898
3899 out:
3900 return ret;
3901 }
3902
3903 static int process_all_new_xattrs(struct send_ctx *sctx)
3904 {
3905 int ret;
3906 struct btrfs_root *root;
3907 struct btrfs_path *path;
3908 struct btrfs_key key;
3909 struct btrfs_key found_key;
3910 struct extent_buffer *eb;
3911 int slot;
3912
3913 path = alloc_path_for_send();
3914 if (!path)
3915 return -ENOMEM;
3916
3917 root = sctx->send_root;
3918
3919 key.objectid = sctx->cmp_key->objectid;
3920 key.type = BTRFS_XATTR_ITEM_KEY;
3921 key.offset = 0;
3922 while (1) {
3923 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3924 if (ret < 0)
3925 goto out;
3926 if (ret) {
3927 ret = 0;
3928 goto out;
3929 }
3930
3931 eb = path->nodes[0];
3932 slot = path->slots[0];
3933 btrfs_item_key_to_cpu(eb, &found_key, slot);
3934
3935 if (found_key.objectid != key.objectid ||
3936 found_key.type != key.type) {
3937 ret = 0;
3938 goto out;
3939 }
3940
3941 ret = iterate_dir_item(root, path, &found_key,
3942 __process_new_xattr, sctx);
3943 if (ret < 0)
3944 goto out;
3945
3946 btrfs_release_path(path);
3947 key.offset = found_key.offset + 1;
3948 }
3949
3950 out:
3951 btrfs_free_path(path);
3952 return ret;
3953 }
3954
3955 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3956 {
3957 struct btrfs_root *root = sctx->send_root;
3958 struct btrfs_fs_info *fs_info = root->fs_info;
3959 struct inode *inode;
3960 struct page *page;
3961 char *addr;
3962 struct btrfs_key key;
3963 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3964 pgoff_t last_index;
3965 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3966 ssize_t ret = 0;
3967
3968 key.objectid = sctx->cur_ino;
3969 key.type = BTRFS_INODE_ITEM_KEY;
3970 key.offset = 0;
3971
3972 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3973 if (IS_ERR(inode))
3974 return PTR_ERR(inode);
3975
3976 if (offset + len > i_size_read(inode)) {
3977 if (offset > i_size_read(inode))
3978 len = 0;
3979 else
3980 len = offset - i_size_read(inode);
3981 }
3982 if (len == 0)
3983 goto out;
3984
3985 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
3986 while (index <= last_index) {
3987 unsigned cur_len = min_t(unsigned, len,
3988 PAGE_CACHE_SIZE - pg_offset);
3989 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3990 if (!page) {
3991 ret = -ENOMEM;
3992 break;
3993 }
3994
3995 if (!PageUptodate(page)) {
3996 btrfs_readpage(NULL, page);
3997 lock_page(page);
3998 if (!PageUptodate(page)) {
3999 unlock_page(page);
4000 page_cache_release(page);
4001 ret = -EIO;
4002 break;
4003 }
4004 }
4005
4006 addr = kmap(page);
4007 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4008 kunmap(page);
4009 unlock_page(page);
4010 page_cache_release(page);
4011 index++;
4012 pg_offset = 0;
4013 len -= cur_len;
4014 ret += cur_len;
4015 }
4016 out:
4017 iput(inode);
4018 return ret;
4019 }
4020
4021 /*
4022 * Read some bytes from the current inode/file and send a write command to
4023 * user space.
4024 */
4025 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4026 {
4027 int ret = 0;
4028 struct fs_path *p;
4029 ssize_t num_read = 0;
4030
4031 p = fs_path_alloc();
4032 if (!p)
4033 return -ENOMEM;
4034
4035 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4036
4037 num_read = fill_read_buf(sctx, offset, len);
4038 if (num_read <= 0) {
4039 if (num_read < 0)
4040 ret = num_read;
4041 goto out;
4042 }
4043
4044 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4045 if (ret < 0)
4046 goto out;
4047
4048 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4049 if (ret < 0)
4050 goto out;
4051
4052 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4053 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4054 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4055
4056 ret = send_cmd(sctx);
4057
4058 tlv_put_failure:
4059 out:
4060 fs_path_free(p);
4061 if (ret < 0)
4062 return ret;
4063 return num_read;
4064 }
4065
4066 /*
4067 * Send a clone command to user space.
4068 */
4069 static int send_clone(struct send_ctx *sctx,
4070 u64 offset, u32 len,
4071 struct clone_root *clone_root)
4072 {
4073 int ret = 0;
4074 struct fs_path *p;
4075 u64 gen;
4076
4077 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4078 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4079 clone_root->root->objectid, clone_root->ino,
4080 clone_root->offset);
4081
4082 p = fs_path_alloc();
4083 if (!p)
4084 return -ENOMEM;
4085
4086 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4087 if (ret < 0)
4088 goto out;
4089
4090 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4091 if (ret < 0)
4092 goto out;
4093
4094 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4095 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4096 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4097
4098 if (clone_root->root == sctx->send_root) {
4099 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4100 &gen, NULL, NULL, NULL, NULL);
4101 if (ret < 0)
4102 goto out;
4103 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4104 } else {
4105 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4106 }
4107 if (ret < 0)
4108 goto out;
4109
4110 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4111 clone_root->root->root_item.uuid);
4112 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4113 le64_to_cpu(clone_root->root->root_item.ctransid));
4114 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4115 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4116 clone_root->offset);
4117
4118 ret = send_cmd(sctx);
4119
4120 tlv_put_failure:
4121 out:
4122 fs_path_free(p);
4123 return ret;
4124 }
4125
4126 /*
4127 * Send an update extent command to user space.
4128 */
4129 static int send_update_extent(struct send_ctx *sctx,
4130 u64 offset, u32 len)
4131 {
4132 int ret = 0;
4133 struct fs_path *p;
4134
4135 p = fs_path_alloc();
4136 if (!p)
4137 return -ENOMEM;
4138
4139 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4140 if (ret < 0)
4141 goto out;
4142
4143 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4144 if (ret < 0)
4145 goto out;
4146
4147 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4148 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4149 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4150
4151 ret = send_cmd(sctx);
4152
4153 tlv_put_failure:
4154 out:
4155 fs_path_free(p);
4156 return ret;
4157 }
4158
4159 static int send_hole(struct send_ctx *sctx, u64 end)
4160 {
4161 struct fs_path *p = NULL;
4162 u64 offset = sctx->cur_inode_last_extent;
4163 u64 len;
4164 int ret = 0;
4165
4166 p = fs_path_alloc();
4167 if (!p)
4168 return -ENOMEM;
4169 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4170 while (offset < end) {
4171 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4172
4173 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4174 if (ret < 0)
4175 break;
4176 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4177 if (ret < 0)
4178 break;
4179 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4180 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4181 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4182 ret = send_cmd(sctx);
4183 if (ret < 0)
4184 break;
4185 offset += len;
4186 }
4187 tlv_put_failure:
4188 fs_path_free(p);
4189 return ret;
4190 }
4191
4192 static int send_write_or_clone(struct send_ctx *sctx,
4193 struct btrfs_path *path,
4194 struct btrfs_key *key,
4195 struct clone_root *clone_root)
4196 {
4197 int ret = 0;
4198 struct btrfs_file_extent_item *ei;
4199 u64 offset = key->offset;
4200 u64 pos = 0;
4201 u64 len;
4202 u32 l;
4203 u8 type;
4204 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4205
4206 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4207 struct btrfs_file_extent_item);
4208 type = btrfs_file_extent_type(path->nodes[0], ei);
4209 if (type == BTRFS_FILE_EXTENT_INLINE) {
4210 len = btrfs_file_extent_inline_len(path->nodes[0],
4211 path->slots[0], ei);
4212 /*
4213 * it is possible the inline item won't cover the whole page,
4214 * but there may be items after this page. Make
4215 * sure to send the whole thing
4216 */
4217 len = PAGE_CACHE_ALIGN(len);
4218 } else {
4219 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4220 }
4221
4222 if (offset + len > sctx->cur_inode_size)
4223 len = sctx->cur_inode_size - offset;
4224 if (len == 0) {
4225 ret = 0;
4226 goto out;
4227 }
4228
4229 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4230 ret = send_clone(sctx, offset, len, clone_root);
4231 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4232 ret = send_update_extent(sctx, offset, len);
4233 } else {
4234 while (pos < len) {
4235 l = len - pos;
4236 if (l > BTRFS_SEND_READ_SIZE)
4237 l = BTRFS_SEND_READ_SIZE;
4238 ret = send_write(sctx, pos + offset, l);
4239 if (ret < 0)
4240 goto out;
4241 if (!ret)
4242 break;
4243 pos += ret;
4244 }
4245 ret = 0;
4246 }
4247 out:
4248 return ret;
4249 }
4250
4251 static int is_extent_unchanged(struct send_ctx *sctx,
4252 struct btrfs_path *left_path,
4253 struct btrfs_key *ekey)
4254 {
4255 int ret = 0;
4256 struct btrfs_key key;
4257 struct btrfs_path *path = NULL;
4258 struct extent_buffer *eb;
4259 int slot;
4260 struct btrfs_key found_key;
4261 struct btrfs_file_extent_item *ei;
4262 u64 left_disknr;
4263 u64 right_disknr;
4264 u64 left_offset;
4265 u64 right_offset;
4266 u64 left_offset_fixed;
4267 u64 left_len;
4268 u64 right_len;
4269 u64 left_gen;
4270 u64 right_gen;
4271 u8 left_type;
4272 u8 right_type;
4273
4274 path = alloc_path_for_send();
4275 if (!path)
4276 return -ENOMEM;
4277
4278 eb = left_path->nodes[0];
4279 slot = left_path->slots[0];
4280 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4281 left_type = btrfs_file_extent_type(eb, ei);
4282
4283 if (left_type != BTRFS_FILE_EXTENT_REG) {
4284 ret = 0;
4285 goto out;
4286 }
4287 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4288 left_len = btrfs_file_extent_num_bytes(eb, ei);
4289 left_offset = btrfs_file_extent_offset(eb, ei);
4290 left_gen = btrfs_file_extent_generation(eb, ei);
4291
4292 /*
4293 * Following comments will refer to these graphics. L is the left
4294 * extents which we are checking at the moment. 1-8 are the right
4295 * extents that we iterate.
4296 *
4297 * |-----L-----|
4298 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4299 *
4300 * |-----L-----|
4301 * |--1--|-2b-|...(same as above)
4302 *
4303 * Alternative situation. Happens on files where extents got split.
4304 * |-----L-----|
4305 * |-----------7-----------|-6-|
4306 *
4307 * Alternative situation. Happens on files which got larger.
4308 * |-----L-----|
4309 * |-8-|
4310 * Nothing follows after 8.
4311 */
4312
4313 key.objectid = ekey->objectid;
4314 key.type = BTRFS_EXTENT_DATA_KEY;
4315 key.offset = ekey->offset;
4316 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4317 if (ret < 0)
4318 goto out;
4319 if (ret) {
4320 ret = 0;
4321 goto out;
4322 }
4323
4324 /*
4325 * Handle special case where the right side has no extents at all.
4326 */
4327 eb = path->nodes[0];
4328 slot = path->slots[0];
4329 btrfs_item_key_to_cpu(eb, &found_key, slot);
4330 if (found_key.objectid != key.objectid ||
4331 found_key.type != key.type) {
4332 /* If we're a hole then just pretend nothing changed */
4333 ret = (left_disknr) ? 0 : 1;
4334 goto out;
4335 }
4336
4337 /*
4338 * We're now on 2a, 2b or 7.
4339 */
4340 key = found_key;
4341 while (key.offset < ekey->offset + left_len) {
4342 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4343 right_type = btrfs_file_extent_type(eb, ei);
4344 if (right_type != BTRFS_FILE_EXTENT_REG) {
4345 ret = 0;
4346 goto out;
4347 }
4348
4349 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4350 right_len = btrfs_file_extent_num_bytes(eb, ei);
4351 right_offset = btrfs_file_extent_offset(eb, ei);
4352 right_gen = btrfs_file_extent_generation(eb, ei);
4353
4354 /*
4355 * Are we at extent 8? If yes, we know the extent is changed.
4356 * This may only happen on the first iteration.
4357 */
4358 if (found_key.offset + right_len <= ekey->offset) {
4359 /* If we're a hole just pretend nothing changed */
4360 ret = (left_disknr) ? 0 : 1;
4361 goto out;
4362 }
4363
4364 left_offset_fixed = left_offset;
4365 if (key.offset < ekey->offset) {
4366 /* Fix the right offset for 2a and 7. */
4367 right_offset += ekey->offset - key.offset;
4368 } else {
4369 /* Fix the left offset for all behind 2a and 2b */
4370 left_offset_fixed += key.offset - ekey->offset;
4371 }
4372
4373 /*
4374 * Check if we have the same extent.
4375 */
4376 if (left_disknr != right_disknr ||
4377 left_offset_fixed != right_offset ||
4378 left_gen != right_gen) {
4379 ret = 0;
4380 goto out;
4381 }
4382
4383 /*
4384 * Go to the next extent.
4385 */
4386 ret = btrfs_next_item(sctx->parent_root, path);
4387 if (ret < 0)
4388 goto out;
4389 if (!ret) {
4390 eb = path->nodes[0];
4391 slot = path->slots[0];
4392 btrfs_item_key_to_cpu(eb, &found_key, slot);
4393 }
4394 if (ret || found_key.objectid != key.objectid ||
4395 found_key.type != key.type) {
4396 key.offset += right_len;
4397 break;
4398 }
4399 if (found_key.offset != key.offset + right_len) {
4400 ret = 0;
4401 goto out;
4402 }
4403 key = found_key;
4404 }
4405
4406 /*
4407 * We're now behind the left extent (treat as unchanged) or at the end
4408 * of the right side (treat as changed).
4409 */
4410 if (key.offset >= ekey->offset + left_len)
4411 ret = 1;
4412 else
4413 ret = 0;
4414
4415
4416 out:
4417 btrfs_free_path(path);
4418 return ret;
4419 }
4420
4421 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4422 {
4423 struct btrfs_path *path;
4424 struct btrfs_root *root = sctx->send_root;
4425 struct btrfs_file_extent_item *fi;
4426 struct btrfs_key key;
4427 u64 extent_end;
4428 u8 type;
4429 int ret;
4430
4431 path = alloc_path_for_send();
4432 if (!path)
4433 return -ENOMEM;
4434
4435 sctx->cur_inode_last_extent = 0;
4436
4437 key.objectid = sctx->cur_ino;
4438 key.type = BTRFS_EXTENT_DATA_KEY;
4439 key.offset = offset;
4440 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4441 if (ret < 0)
4442 goto out;
4443 ret = 0;
4444 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4445 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4446 goto out;
4447
4448 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4449 struct btrfs_file_extent_item);
4450 type = btrfs_file_extent_type(path->nodes[0], fi);
4451 if (type == BTRFS_FILE_EXTENT_INLINE) {
4452 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4453 path->slots[0], fi);
4454 extent_end = ALIGN(key.offset + size,
4455 sctx->send_root->sectorsize);
4456 } else {
4457 extent_end = key.offset +
4458 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4459 }
4460 sctx->cur_inode_last_extent = extent_end;
4461 out:
4462 btrfs_free_path(path);
4463 return ret;
4464 }
4465
4466 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4467 struct btrfs_key *key)
4468 {
4469 struct btrfs_file_extent_item *fi;
4470 u64 extent_end;
4471 u8 type;
4472 int ret = 0;
4473
4474 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4475 return 0;
4476
4477 if (sctx->cur_inode_last_extent == (u64)-1) {
4478 ret = get_last_extent(sctx, key->offset - 1);
4479 if (ret)
4480 return ret;
4481 }
4482
4483 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4484 struct btrfs_file_extent_item);
4485 type = btrfs_file_extent_type(path->nodes[0], fi);
4486 if (type == BTRFS_FILE_EXTENT_INLINE) {
4487 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4488 path->slots[0], fi);
4489 extent_end = ALIGN(key->offset + size,
4490 sctx->send_root->sectorsize);
4491 } else {
4492 extent_end = key->offset +
4493 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4494 }
4495
4496 if (path->slots[0] == 0 &&
4497 sctx->cur_inode_last_extent < key->offset) {
4498 /*
4499 * We might have skipped entire leafs that contained only
4500 * file extent items for our current inode. These leafs have
4501 * a generation number smaller (older) than the one in the
4502 * current leaf and the leaf our last extent came from, and
4503 * are located between these 2 leafs.
4504 */
4505 ret = get_last_extent(sctx, key->offset - 1);
4506 if (ret)
4507 return ret;
4508 }
4509
4510 if (sctx->cur_inode_last_extent < key->offset)
4511 ret = send_hole(sctx, key->offset);
4512 sctx->cur_inode_last_extent = extent_end;
4513 return ret;
4514 }
4515
4516 static int process_extent(struct send_ctx *sctx,
4517 struct btrfs_path *path,
4518 struct btrfs_key *key)
4519 {
4520 struct clone_root *found_clone = NULL;
4521 int ret = 0;
4522
4523 if (S_ISLNK(sctx->cur_inode_mode))
4524 return 0;
4525
4526 if (sctx->parent_root && !sctx->cur_inode_new) {
4527 ret = is_extent_unchanged(sctx, path, key);
4528 if (ret < 0)
4529 goto out;
4530 if (ret) {
4531 ret = 0;
4532 goto out_hole;
4533 }
4534 } else {
4535 struct btrfs_file_extent_item *ei;
4536 u8 type;
4537
4538 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4539 struct btrfs_file_extent_item);
4540 type = btrfs_file_extent_type(path->nodes[0], ei);
4541 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4542 type == BTRFS_FILE_EXTENT_REG) {
4543 /*
4544 * The send spec does not have a prealloc command yet,
4545 * so just leave a hole for prealloc'ed extents until
4546 * we have enough commands queued up to justify rev'ing
4547 * the send spec.
4548 */
4549 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4550 ret = 0;
4551 goto out;
4552 }
4553
4554 /* Have a hole, just skip it. */
4555 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4556 ret = 0;
4557 goto out;
4558 }
4559 }
4560 }
4561
4562 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4563 sctx->cur_inode_size, &found_clone);
4564 if (ret != -ENOENT && ret < 0)
4565 goto out;
4566
4567 ret = send_write_or_clone(sctx, path, key, found_clone);
4568 if (ret)
4569 goto out;
4570 out_hole:
4571 ret = maybe_send_hole(sctx, path, key);
4572 out:
4573 return ret;
4574 }
4575
4576 static int process_all_extents(struct send_ctx *sctx)
4577 {
4578 int ret;
4579 struct btrfs_root *root;
4580 struct btrfs_path *path;
4581 struct btrfs_key key;
4582 struct btrfs_key found_key;
4583 struct extent_buffer *eb;
4584 int slot;
4585
4586 root = sctx->send_root;
4587 path = alloc_path_for_send();
4588 if (!path)
4589 return -ENOMEM;
4590
4591 key.objectid = sctx->cmp_key->objectid;
4592 key.type = BTRFS_EXTENT_DATA_KEY;
4593 key.offset = 0;
4594 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4595 if (ret < 0)
4596 goto out;
4597
4598 while (1) {
4599 eb = path->nodes[0];
4600 slot = path->slots[0];
4601
4602 if (slot >= btrfs_header_nritems(eb)) {
4603 ret = btrfs_next_leaf(root, path);
4604 if (ret < 0) {
4605 goto out;
4606 } else if (ret > 0) {
4607 ret = 0;
4608 break;
4609 }
4610 continue;
4611 }
4612
4613 btrfs_item_key_to_cpu(eb, &found_key, slot);
4614
4615 if (found_key.objectid != key.objectid ||
4616 found_key.type != key.type) {
4617 ret = 0;
4618 goto out;
4619 }
4620
4621 ret = process_extent(sctx, path, &found_key);
4622 if (ret < 0)
4623 goto out;
4624
4625 path->slots[0]++;
4626 }
4627
4628 out:
4629 btrfs_free_path(path);
4630 return ret;
4631 }
4632
4633 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4634 int *pending_move,
4635 int *refs_processed)
4636 {
4637 int ret = 0;
4638
4639 if (sctx->cur_ino == 0)
4640 goto out;
4641 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4642 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4643 goto out;
4644 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4645 goto out;
4646
4647 ret = process_recorded_refs(sctx, pending_move);
4648 if (ret < 0)
4649 goto out;
4650
4651 *refs_processed = 1;
4652 out:
4653 return ret;
4654 }
4655
4656 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4657 {
4658 int ret = 0;
4659 u64 left_mode;
4660 u64 left_uid;
4661 u64 left_gid;
4662 u64 right_mode;
4663 u64 right_uid;
4664 u64 right_gid;
4665 int need_chmod = 0;
4666 int need_chown = 0;
4667 int pending_move = 0;
4668 int refs_processed = 0;
4669
4670 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4671 &refs_processed);
4672 if (ret < 0)
4673 goto out;
4674
4675 /*
4676 * We have processed the refs and thus need to advance send_progress.
4677 * Now, calls to get_cur_xxx will take the updated refs of the current
4678 * inode into account.
4679 *
4680 * On the other hand, if our current inode is a directory and couldn't
4681 * be moved/renamed because its parent was renamed/moved too and it has
4682 * a higher inode number, we can only move/rename our current inode
4683 * after we moved/renamed its parent. Therefore in this case operate on
4684 * the old path (pre move/rename) of our current inode, and the
4685 * move/rename will be performed later.
4686 */
4687 if (refs_processed && !pending_move)
4688 sctx->send_progress = sctx->cur_ino + 1;
4689
4690 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4691 goto out;
4692 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4693 goto out;
4694
4695 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4696 &left_mode, &left_uid, &left_gid, NULL);
4697 if (ret < 0)
4698 goto out;
4699
4700 if (!sctx->parent_root || sctx->cur_inode_new) {
4701 need_chown = 1;
4702 if (!S_ISLNK(sctx->cur_inode_mode))
4703 need_chmod = 1;
4704 } else {
4705 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4706 NULL, NULL, &right_mode, &right_uid,
4707 &right_gid, NULL);
4708 if (ret < 0)
4709 goto out;
4710
4711 if (left_uid != right_uid || left_gid != right_gid)
4712 need_chown = 1;
4713 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4714 need_chmod = 1;
4715 }
4716
4717 if (S_ISREG(sctx->cur_inode_mode)) {
4718 if (need_send_hole(sctx)) {
4719 if (sctx->cur_inode_last_extent == (u64)-1) {
4720 ret = get_last_extent(sctx, (u64)-1);
4721 if (ret)
4722 goto out;
4723 }
4724 if (sctx->cur_inode_last_extent <
4725 sctx->cur_inode_size) {
4726 ret = send_hole(sctx, sctx->cur_inode_size);
4727 if (ret)
4728 goto out;
4729 }
4730 }
4731 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4732 sctx->cur_inode_size);
4733 if (ret < 0)
4734 goto out;
4735 }
4736
4737 if (need_chown) {
4738 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4739 left_uid, left_gid);
4740 if (ret < 0)
4741 goto out;
4742 }
4743 if (need_chmod) {
4744 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4745 left_mode);
4746 if (ret < 0)
4747 goto out;
4748 }
4749
4750 /*
4751 * If other directory inodes depended on our current directory
4752 * inode's move/rename, now do their move/rename operations.
4753 */
4754 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
4755 ret = apply_children_dir_moves(sctx);
4756 if (ret)
4757 goto out;
4758 }
4759
4760 /*
4761 * Need to send that every time, no matter if it actually
4762 * changed between the two trees as we have done changes to
4763 * the inode before.
4764 */
4765 sctx->send_progress = sctx->cur_ino + 1;
4766 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4767 if (ret < 0)
4768 goto out;
4769
4770 out:
4771 return ret;
4772 }
4773
4774 static int changed_inode(struct send_ctx *sctx,
4775 enum btrfs_compare_tree_result result)
4776 {
4777 int ret = 0;
4778 struct btrfs_key *key = sctx->cmp_key;
4779 struct btrfs_inode_item *left_ii = NULL;
4780 struct btrfs_inode_item *right_ii = NULL;
4781 u64 left_gen = 0;
4782 u64 right_gen = 0;
4783
4784 sctx->cur_ino = key->objectid;
4785 sctx->cur_inode_new_gen = 0;
4786 sctx->cur_inode_last_extent = (u64)-1;
4787
4788 /*
4789 * Set send_progress to current inode. This will tell all get_cur_xxx
4790 * functions that the current inode's refs are not updated yet. Later,
4791 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4792 */
4793 sctx->send_progress = sctx->cur_ino;
4794
4795 if (result == BTRFS_COMPARE_TREE_NEW ||
4796 result == BTRFS_COMPARE_TREE_CHANGED) {
4797 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4798 sctx->left_path->slots[0],
4799 struct btrfs_inode_item);
4800 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4801 left_ii);
4802 } else {
4803 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4804 sctx->right_path->slots[0],
4805 struct btrfs_inode_item);
4806 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4807 right_ii);
4808 }
4809 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4810 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4811 sctx->right_path->slots[0],
4812 struct btrfs_inode_item);
4813
4814 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4815 right_ii);
4816
4817 /*
4818 * The cur_ino = root dir case is special here. We can't treat
4819 * the inode as deleted+reused because it would generate a
4820 * stream that tries to delete/mkdir the root dir.
4821 */
4822 if (left_gen != right_gen &&
4823 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4824 sctx->cur_inode_new_gen = 1;
4825 }
4826
4827 if (result == BTRFS_COMPARE_TREE_NEW) {
4828 sctx->cur_inode_gen = left_gen;
4829 sctx->cur_inode_new = 1;
4830 sctx->cur_inode_deleted = 0;
4831 sctx->cur_inode_size = btrfs_inode_size(
4832 sctx->left_path->nodes[0], left_ii);
4833 sctx->cur_inode_mode = btrfs_inode_mode(
4834 sctx->left_path->nodes[0], left_ii);
4835 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4836 ret = send_create_inode_if_needed(sctx);
4837 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4838 sctx->cur_inode_gen = right_gen;
4839 sctx->cur_inode_new = 0;
4840 sctx->cur_inode_deleted = 1;
4841 sctx->cur_inode_size = btrfs_inode_size(
4842 sctx->right_path->nodes[0], right_ii);
4843 sctx->cur_inode_mode = btrfs_inode_mode(
4844 sctx->right_path->nodes[0], right_ii);
4845 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4846 /*
4847 * We need to do some special handling in case the inode was
4848 * reported as changed with a changed generation number. This
4849 * means that the original inode was deleted and new inode
4850 * reused the same inum. So we have to treat the old inode as
4851 * deleted and the new one as new.
4852 */
4853 if (sctx->cur_inode_new_gen) {
4854 /*
4855 * First, process the inode as if it was deleted.
4856 */
4857 sctx->cur_inode_gen = right_gen;
4858 sctx->cur_inode_new = 0;
4859 sctx->cur_inode_deleted = 1;
4860 sctx->cur_inode_size = btrfs_inode_size(
4861 sctx->right_path->nodes[0], right_ii);
4862 sctx->cur_inode_mode = btrfs_inode_mode(
4863 sctx->right_path->nodes[0], right_ii);
4864 ret = process_all_refs(sctx,
4865 BTRFS_COMPARE_TREE_DELETED);
4866 if (ret < 0)
4867 goto out;
4868
4869 /*
4870 * Now process the inode as if it was new.
4871 */
4872 sctx->cur_inode_gen = left_gen;
4873 sctx->cur_inode_new = 1;
4874 sctx->cur_inode_deleted = 0;
4875 sctx->cur_inode_size = btrfs_inode_size(
4876 sctx->left_path->nodes[0], left_ii);
4877 sctx->cur_inode_mode = btrfs_inode_mode(
4878 sctx->left_path->nodes[0], left_ii);
4879 ret = send_create_inode_if_needed(sctx);
4880 if (ret < 0)
4881 goto out;
4882
4883 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4884 if (ret < 0)
4885 goto out;
4886 /*
4887 * Advance send_progress now as we did not get into
4888 * process_recorded_refs_if_needed in the new_gen case.
4889 */
4890 sctx->send_progress = sctx->cur_ino + 1;
4891
4892 /*
4893 * Now process all extents and xattrs of the inode as if
4894 * they were all new.
4895 */
4896 ret = process_all_extents(sctx);
4897 if (ret < 0)
4898 goto out;
4899 ret = process_all_new_xattrs(sctx);
4900 if (ret < 0)
4901 goto out;
4902 } else {
4903 sctx->cur_inode_gen = left_gen;
4904 sctx->cur_inode_new = 0;
4905 sctx->cur_inode_new_gen = 0;
4906 sctx->cur_inode_deleted = 0;
4907 sctx->cur_inode_size = btrfs_inode_size(
4908 sctx->left_path->nodes[0], left_ii);
4909 sctx->cur_inode_mode = btrfs_inode_mode(
4910 sctx->left_path->nodes[0], left_ii);
4911 }
4912 }
4913
4914 out:
4915 return ret;
4916 }
4917
4918 /*
4919 * We have to process new refs before deleted refs, but compare_trees gives us
4920 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4921 * first and later process them in process_recorded_refs.
4922 * For the cur_inode_new_gen case, we skip recording completely because
4923 * changed_inode did already initiate processing of refs. The reason for this is
4924 * that in this case, compare_tree actually compares the refs of 2 different
4925 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4926 * refs of the right tree as deleted and all refs of the left tree as new.
4927 */
4928 static int changed_ref(struct send_ctx *sctx,
4929 enum btrfs_compare_tree_result result)
4930 {
4931 int ret = 0;
4932
4933 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4934
4935 if (!sctx->cur_inode_new_gen &&
4936 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4937 if (result == BTRFS_COMPARE_TREE_NEW)
4938 ret = record_new_ref(sctx);
4939 else if (result == BTRFS_COMPARE_TREE_DELETED)
4940 ret = record_deleted_ref(sctx);
4941 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4942 ret = record_changed_ref(sctx);
4943 }
4944
4945 return ret;
4946 }
4947
4948 /*
4949 * Process new/deleted/changed xattrs. We skip processing in the
4950 * cur_inode_new_gen case because changed_inode did already initiate processing
4951 * of xattrs. The reason is the same as in changed_ref
4952 */
4953 static int changed_xattr(struct send_ctx *sctx,
4954 enum btrfs_compare_tree_result result)
4955 {
4956 int ret = 0;
4957
4958 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4959
4960 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4961 if (result == BTRFS_COMPARE_TREE_NEW)
4962 ret = process_new_xattr(sctx);
4963 else if (result == BTRFS_COMPARE_TREE_DELETED)
4964 ret = process_deleted_xattr(sctx);
4965 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4966 ret = process_changed_xattr(sctx);
4967 }
4968
4969 return ret;
4970 }
4971
4972 /*
4973 * Process new/deleted/changed extents. We skip processing in the
4974 * cur_inode_new_gen case because changed_inode did already initiate processing
4975 * of extents. The reason is the same as in changed_ref
4976 */
4977 static int changed_extent(struct send_ctx *sctx,
4978 enum btrfs_compare_tree_result result)
4979 {
4980 int ret = 0;
4981
4982 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4983
4984 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4985 if (result != BTRFS_COMPARE_TREE_DELETED)
4986 ret = process_extent(sctx, sctx->left_path,
4987 sctx->cmp_key);
4988 }
4989
4990 return ret;
4991 }
4992
4993 static int dir_changed(struct send_ctx *sctx, u64 dir)
4994 {
4995 u64 orig_gen, new_gen;
4996 int ret;
4997
4998 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
4999 NULL, NULL);
5000 if (ret)
5001 return ret;
5002
5003 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5004 NULL, NULL, NULL);
5005 if (ret)
5006 return ret;
5007
5008 return (orig_gen != new_gen) ? 1 : 0;
5009 }
5010
5011 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5012 struct btrfs_key *key)
5013 {
5014 struct btrfs_inode_extref *extref;
5015 struct extent_buffer *leaf;
5016 u64 dirid = 0, last_dirid = 0;
5017 unsigned long ptr;
5018 u32 item_size;
5019 u32 cur_offset = 0;
5020 int ref_name_len;
5021 int ret = 0;
5022
5023 /* Easy case, just check this one dirid */
5024 if (key->type == BTRFS_INODE_REF_KEY) {
5025 dirid = key->offset;
5026
5027 ret = dir_changed(sctx, dirid);
5028 goto out;
5029 }
5030
5031 leaf = path->nodes[0];
5032 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5033 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5034 while (cur_offset < item_size) {
5035 extref = (struct btrfs_inode_extref *)(ptr +
5036 cur_offset);
5037 dirid = btrfs_inode_extref_parent(leaf, extref);
5038 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5039 cur_offset += ref_name_len + sizeof(*extref);
5040 if (dirid == last_dirid)
5041 continue;
5042 ret = dir_changed(sctx, dirid);
5043 if (ret)
5044 break;
5045 last_dirid = dirid;
5046 }
5047 out:
5048 return ret;
5049 }
5050
5051 /*
5052 * Updates compare related fields in sctx and simply forwards to the actual
5053 * changed_xxx functions.
5054 */
5055 static int changed_cb(struct btrfs_root *left_root,
5056 struct btrfs_root *right_root,
5057 struct btrfs_path *left_path,
5058 struct btrfs_path *right_path,
5059 struct btrfs_key *key,
5060 enum btrfs_compare_tree_result result,
5061 void *ctx)
5062 {
5063 int ret = 0;
5064 struct send_ctx *sctx = ctx;
5065
5066 if (result == BTRFS_COMPARE_TREE_SAME) {
5067 if (key->type == BTRFS_INODE_REF_KEY ||
5068 key->type == BTRFS_INODE_EXTREF_KEY) {
5069 ret = compare_refs(sctx, left_path, key);
5070 if (!ret)
5071 return 0;
5072 if (ret < 0)
5073 return ret;
5074 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5075 return maybe_send_hole(sctx, left_path, key);
5076 } else {
5077 return 0;
5078 }
5079 result = BTRFS_COMPARE_TREE_CHANGED;
5080 ret = 0;
5081 }
5082
5083 sctx->left_path = left_path;
5084 sctx->right_path = right_path;
5085 sctx->cmp_key = key;
5086
5087 ret = finish_inode_if_needed(sctx, 0);
5088 if (ret < 0)
5089 goto out;
5090
5091 /* Ignore non-FS objects */
5092 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5093 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5094 goto out;
5095
5096 if (key->type == BTRFS_INODE_ITEM_KEY)
5097 ret = changed_inode(sctx, result);
5098 else if (key->type == BTRFS_INODE_REF_KEY ||
5099 key->type == BTRFS_INODE_EXTREF_KEY)
5100 ret = changed_ref(sctx, result);
5101 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5102 ret = changed_xattr(sctx, result);
5103 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5104 ret = changed_extent(sctx, result);
5105
5106 out:
5107 return ret;
5108 }
5109
5110 static int full_send_tree(struct send_ctx *sctx)
5111 {
5112 int ret;
5113 struct btrfs_root *send_root = sctx->send_root;
5114 struct btrfs_key key;
5115 struct btrfs_key found_key;
5116 struct btrfs_path *path;
5117 struct extent_buffer *eb;
5118 int slot;
5119 u64 start_ctransid;
5120 u64 ctransid;
5121
5122 path = alloc_path_for_send();
5123 if (!path)
5124 return -ENOMEM;
5125
5126 spin_lock(&send_root->root_item_lock);
5127 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5128 spin_unlock(&send_root->root_item_lock);
5129
5130 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5131 key.type = BTRFS_INODE_ITEM_KEY;
5132 key.offset = 0;
5133
5134 /*
5135 * Make sure the tree has not changed after re-joining. We detect this
5136 * by comparing start_ctransid and ctransid. They should always match.
5137 */
5138 spin_lock(&send_root->root_item_lock);
5139 ctransid = btrfs_root_ctransid(&send_root->root_item);
5140 spin_unlock(&send_root->root_item_lock);
5141
5142 if (ctransid != start_ctransid) {
5143 WARN(1, KERN_WARNING "BTRFS: the root that you're trying to "
5144 "send was modified in between. This is "
5145 "probably a bug.\n");
5146 ret = -EIO;
5147 goto out;
5148 }
5149
5150 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5151 if (ret < 0)
5152 goto out;
5153 if (ret)
5154 goto out_finish;
5155
5156 while (1) {
5157 eb = path->nodes[0];
5158 slot = path->slots[0];
5159 btrfs_item_key_to_cpu(eb, &found_key, slot);
5160
5161 ret = changed_cb(send_root, NULL, path, NULL,
5162 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5163 if (ret < 0)
5164 goto out;
5165
5166 key.objectid = found_key.objectid;
5167 key.type = found_key.type;
5168 key.offset = found_key.offset + 1;
5169
5170 ret = btrfs_next_item(send_root, path);
5171 if (ret < 0)
5172 goto out;
5173 if (ret) {
5174 ret = 0;
5175 break;
5176 }
5177 }
5178
5179 out_finish:
5180 ret = finish_inode_if_needed(sctx, 1);
5181
5182 out:
5183 btrfs_free_path(path);
5184 return ret;
5185 }
5186
5187 static int send_subvol(struct send_ctx *sctx)
5188 {
5189 int ret;
5190
5191 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5192 ret = send_header(sctx);
5193 if (ret < 0)
5194 goto out;
5195 }
5196
5197 ret = send_subvol_begin(sctx);
5198 if (ret < 0)
5199 goto out;
5200
5201 if (sctx->parent_root) {
5202 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5203 changed_cb, sctx);
5204 if (ret < 0)
5205 goto out;
5206 ret = finish_inode_if_needed(sctx, 1);
5207 if (ret < 0)
5208 goto out;
5209 } else {
5210 ret = full_send_tree(sctx);
5211 if (ret < 0)
5212 goto out;
5213 }
5214
5215 out:
5216 free_recorded_refs(sctx);
5217 return ret;
5218 }
5219
5220 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5221 {
5222 spin_lock(&root->root_item_lock);
5223 root->send_in_progress--;
5224 /*
5225 * Not much left to do, we don't know why it's unbalanced and
5226 * can't blindly reset it to 0.
5227 */
5228 if (root->send_in_progress < 0)
5229 btrfs_err(root->fs_info,
5230 "send_in_progres unbalanced %d root %llu\n",
5231 root->send_in_progress, root->root_key.objectid);
5232 spin_unlock(&root->root_item_lock);
5233 }
5234
5235 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5236 {
5237 int ret = 0;
5238 struct btrfs_root *send_root;
5239 struct btrfs_root *clone_root;
5240 struct btrfs_fs_info *fs_info;
5241 struct btrfs_ioctl_send_args *arg = NULL;
5242 struct btrfs_key key;
5243 struct send_ctx *sctx = NULL;
5244 u32 i;
5245 u64 *clone_sources_tmp = NULL;
5246 int clone_sources_to_rollback = 0;
5247 int sort_clone_roots = 0;
5248 int index;
5249
5250 if (!capable(CAP_SYS_ADMIN))
5251 return -EPERM;
5252
5253 send_root = BTRFS_I(file_inode(mnt_file))->root;
5254 fs_info = send_root->fs_info;
5255
5256 /*
5257 * The subvolume must remain read-only during send, protect against
5258 * making it RW.
5259 */
5260 spin_lock(&send_root->root_item_lock);
5261 send_root->send_in_progress++;
5262 spin_unlock(&send_root->root_item_lock);
5263
5264 /*
5265 * This is done when we lookup the root, it should already be complete
5266 * by the time we get here.
5267 */
5268 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5269
5270 /*
5271 * Userspace tools do the checks and warn the user if it's
5272 * not RO.
5273 */
5274 if (!btrfs_root_readonly(send_root)) {
5275 ret = -EPERM;
5276 goto out;
5277 }
5278
5279 arg = memdup_user(arg_, sizeof(*arg));
5280 if (IS_ERR(arg)) {
5281 ret = PTR_ERR(arg);
5282 arg = NULL;
5283 goto out;
5284 }
5285
5286 if (!access_ok(VERIFY_READ, arg->clone_sources,
5287 sizeof(*arg->clone_sources) *
5288 arg->clone_sources_count)) {
5289 ret = -EFAULT;
5290 goto out;
5291 }
5292
5293 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5294 ret = -EINVAL;
5295 goto out;
5296 }
5297
5298 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5299 if (!sctx) {
5300 ret = -ENOMEM;
5301 goto out;
5302 }
5303
5304 INIT_LIST_HEAD(&sctx->new_refs);
5305 INIT_LIST_HEAD(&sctx->deleted_refs);
5306 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5307 INIT_LIST_HEAD(&sctx->name_cache_list);
5308
5309 sctx->flags = arg->flags;
5310
5311 sctx->send_filp = fget(arg->send_fd);
5312 if (!sctx->send_filp) {
5313 ret = -EBADF;
5314 goto out;
5315 }
5316
5317 sctx->send_root = send_root;
5318 sctx->clone_roots_cnt = arg->clone_sources_count;
5319
5320 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5321 sctx->send_buf = vmalloc(sctx->send_max_size);
5322 if (!sctx->send_buf) {
5323 ret = -ENOMEM;
5324 goto out;
5325 }
5326
5327 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5328 if (!sctx->read_buf) {
5329 ret = -ENOMEM;
5330 goto out;
5331 }
5332
5333 sctx->pending_dir_moves = RB_ROOT;
5334 sctx->waiting_dir_moves = RB_ROOT;
5335
5336 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5337 (arg->clone_sources_count + 1));
5338 if (!sctx->clone_roots) {
5339 ret = -ENOMEM;
5340 goto out;
5341 }
5342
5343 if (arg->clone_sources_count) {
5344 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5345 sizeof(*arg->clone_sources));
5346 if (!clone_sources_tmp) {
5347 ret = -ENOMEM;
5348 goto out;
5349 }
5350
5351 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5352 arg->clone_sources_count *
5353 sizeof(*arg->clone_sources));
5354 if (ret) {
5355 ret = -EFAULT;
5356 goto out;
5357 }
5358
5359 for (i = 0; i < arg->clone_sources_count; i++) {
5360 key.objectid = clone_sources_tmp[i];
5361 key.type = BTRFS_ROOT_ITEM_KEY;
5362 key.offset = (u64)-1;
5363
5364 index = srcu_read_lock(&fs_info->subvol_srcu);
5365
5366 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5367 if (IS_ERR(clone_root)) {
5368 srcu_read_unlock(&fs_info->subvol_srcu, index);
5369 ret = PTR_ERR(clone_root);
5370 goto out;
5371 }
5372 clone_sources_to_rollback = i + 1;
5373 spin_lock(&clone_root->root_item_lock);
5374 clone_root->send_in_progress++;
5375 if (!btrfs_root_readonly(clone_root)) {
5376 spin_unlock(&clone_root->root_item_lock);
5377 srcu_read_unlock(&fs_info->subvol_srcu, index);
5378 ret = -EPERM;
5379 goto out;
5380 }
5381 spin_unlock(&clone_root->root_item_lock);
5382 srcu_read_unlock(&fs_info->subvol_srcu, index);
5383
5384 sctx->clone_roots[i].root = clone_root;
5385 }
5386 vfree(clone_sources_tmp);
5387 clone_sources_tmp = NULL;
5388 }
5389
5390 if (arg->parent_root) {
5391 key.objectid = arg->parent_root;
5392 key.type = BTRFS_ROOT_ITEM_KEY;
5393 key.offset = (u64)-1;
5394
5395 index = srcu_read_lock(&fs_info->subvol_srcu);
5396
5397 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5398 if (IS_ERR(sctx->parent_root)) {
5399 srcu_read_unlock(&fs_info->subvol_srcu, index);
5400 ret = PTR_ERR(sctx->parent_root);
5401 goto out;
5402 }
5403
5404 spin_lock(&sctx->parent_root->root_item_lock);
5405 sctx->parent_root->send_in_progress++;
5406 if (!btrfs_root_readonly(sctx->parent_root)) {
5407 spin_unlock(&sctx->parent_root->root_item_lock);
5408 srcu_read_unlock(&fs_info->subvol_srcu, index);
5409 ret = -EPERM;
5410 goto out;
5411 }
5412 spin_unlock(&sctx->parent_root->root_item_lock);
5413
5414 srcu_read_unlock(&fs_info->subvol_srcu, index);
5415 }
5416
5417 /*
5418 * Clones from send_root are allowed, but only if the clone source
5419 * is behind the current send position. This is checked while searching
5420 * for possible clone sources.
5421 */
5422 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5423
5424 /* We do a bsearch later */
5425 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5426 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5427 NULL);
5428 sort_clone_roots = 1;
5429
5430 ret = send_subvol(sctx);
5431 if (ret < 0)
5432 goto out;
5433
5434 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5435 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5436 if (ret < 0)
5437 goto out;
5438 ret = send_cmd(sctx);
5439 if (ret < 0)
5440 goto out;
5441 }
5442
5443 out:
5444 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5445 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5446 struct rb_node *n;
5447 struct pending_dir_move *pm;
5448
5449 n = rb_first(&sctx->pending_dir_moves);
5450 pm = rb_entry(n, struct pending_dir_move, node);
5451 while (!list_empty(&pm->list)) {
5452 struct pending_dir_move *pm2;
5453
5454 pm2 = list_first_entry(&pm->list,
5455 struct pending_dir_move, list);
5456 free_pending_move(sctx, pm2);
5457 }
5458 free_pending_move(sctx, pm);
5459 }
5460
5461 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5462 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5463 struct rb_node *n;
5464 struct waiting_dir_move *dm;
5465
5466 n = rb_first(&sctx->waiting_dir_moves);
5467 dm = rb_entry(n, struct waiting_dir_move, node);
5468 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5469 kfree(dm);
5470 }
5471
5472 if (sort_clone_roots) {
5473 for (i = 0; i < sctx->clone_roots_cnt; i++)
5474 btrfs_root_dec_send_in_progress(
5475 sctx->clone_roots[i].root);
5476 } else {
5477 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5478 btrfs_root_dec_send_in_progress(
5479 sctx->clone_roots[i].root);
5480
5481 btrfs_root_dec_send_in_progress(send_root);
5482 }
5483 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5484 btrfs_root_dec_send_in_progress(sctx->parent_root);
5485
5486 kfree(arg);
5487 vfree(clone_sources_tmp);
5488
5489 if (sctx) {
5490 if (sctx->send_filp)
5491 fput(sctx->send_filp);
5492
5493 vfree(sctx->clone_roots);
5494 vfree(sctx->send_buf);
5495 vfree(sctx->read_buf);
5496
5497 name_cache_free(sctx);
5498
5499 kfree(sctx);
5500 }
5501
5502 return ret;
5503 }
(deprecated) Btrfs devel tree