?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux/fpc-iii.git] / fs / btrfs / send.c
blob9dde9717c1b9264124d007184bc4c4d60276d99c
1 /*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/bsearch.h>
20 #include <linux/fs.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
29
30 #include "send.h"
31 #include "backref.h"
32 #include "hash.h"
33 #include "locking.h"
34 #include "disk-io.h"
35 #include "btrfs_inode.h"
36 #include "transaction.h"
37
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 = btrfs_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 if (compressed != BTRFS_COMPRESS_NONE) {
1336 /*
1337 * Offsets given by iterate_extent_inodes() are relative
1338 * to the start of the extent, we need to add logical
1339 * offset from the file extent item.
1340 * (See why at backref.c:check_extent_in_eb())
1341 */
1342 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1343 fi);
1344 }
1345 *found = cur_clone_root;
1346 ret = 0;
1347 } else {
1348 ret = -ENOENT;
1349 }
1350
1351 out:
1352 btrfs_free_path(tmp_path);
1353 kfree(backref_ctx);
1354 return ret;
1355 }
1356
1357 static int read_symlink(struct btrfs_root *root,
1358 u64 ino,
1359 struct fs_path *dest)
1360 {
1361 int ret;
1362 struct btrfs_path *path;
1363 struct btrfs_key key;
1364 struct btrfs_file_extent_item *ei;
1365 u8 type;
1366 u8 compression;
1367 unsigned long off;
1368 int len;
1369
1370 path = alloc_path_for_send();
1371 if (!path)
1372 return -ENOMEM;
1373
1374 key.objectid = ino;
1375 key.type = BTRFS_EXTENT_DATA_KEY;
1376 key.offset = 0;
1377 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1378 if (ret < 0)
1379 goto out;
1380 BUG_ON(ret);
1381
1382 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1383 struct btrfs_file_extent_item);
1384 type = btrfs_file_extent_type(path->nodes[0], ei);
1385 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1386 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1387 BUG_ON(compression);
1388
1389 off = btrfs_file_extent_inline_start(ei);
1390 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1391
1392 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1393
1394 out:
1395 btrfs_free_path(path);
1396 return ret;
1397 }
1398
1399 /*
1400 * Helper function to generate a file name that is unique in the root of
1401 * send_root and parent_root. This is used to generate names for orphan inodes.
1402 */
1403 static int gen_unique_name(struct send_ctx *sctx,
1404 u64 ino, u64 gen,
1405 struct fs_path *dest)
1406 {
1407 int ret = 0;
1408 struct btrfs_path *path;
1409 struct btrfs_dir_item *di;
1410 char tmp[64];
1411 int len;
1412 u64 idx = 0;
1413
1414 path = alloc_path_for_send();
1415 if (!path)
1416 return -ENOMEM;
1417
1418 while (1) {
1419 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1420 ino, gen, idx);
1421 if (len >= sizeof(tmp)) {
1422 /* should really not happen */
1423 ret = -EOVERFLOW;
1424 goto out;
1425 }
1426
1427 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1428 path, BTRFS_FIRST_FREE_OBJECTID,
1429 tmp, strlen(tmp), 0);
1430 btrfs_release_path(path);
1431 if (IS_ERR(di)) {
1432 ret = PTR_ERR(di);
1433 goto out;
1434 }
1435 if (di) {
1436 /* not unique, try again */
1437 idx++;
1438 continue;
1439 }
1440
1441 if (!sctx->parent_root) {
1442 /* unique */
1443 ret = 0;
1444 break;
1445 }
1446
1447 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1448 path, BTRFS_FIRST_FREE_OBJECTID,
1449 tmp, strlen(tmp), 0);
1450 btrfs_release_path(path);
1451 if (IS_ERR(di)) {
1452 ret = PTR_ERR(di);
1453 goto out;
1454 }
1455 if (di) {
1456 /* not unique, try again */
1457 idx++;
1458 continue;
1459 }
1460 /* unique */
1461 break;
1462 }
1463
1464 ret = fs_path_add(dest, tmp, strlen(tmp));
1465
1466 out:
1467 btrfs_free_path(path);
1468 return ret;
1469 }
1470
1471 enum inode_state {
1472 inode_state_no_change,
1473 inode_state_will_create,
1474 inode_state_did_create,
1475 inode_state_will_delete,
1476 inode_state_did_delete,
1477 };
1478
1479 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1480 {
1481 int ret;
1482 int left_ret;
1483 int right_ret;
1484 u64 left_gen;
1485 u64 right_gen;
1486
1487 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1488 NULL, NULL);
1489 if (ret < 0 && ret != -ENOENT)
1490 goto out;
1491 left_ret = ret;
1492
1493 if (!sctx->parent_root) {
1494 right_ret = -ENOENT;
1495 } else {
1496 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1497 NULL, NULL, NULL, NULL);
1498 if (ret < 0 && ret != -ENOENT)
1499 goto out;
1500 right_ret = ret;
1501 }
1502
1503 if (!left_ret && !right_ret) {
1504 if (left_gen == gen && right_gen == gen) {
1505 ret = inode_state_no_change;
1506 } else if (left_gen == gen) {
1507 if (ino < sctx->send_progress)
1508 ret = inode_state_did_create;
1509 else
1510 ret = inode_state_will_create;
1511 } else if (right_gen == gen) {
1512 if (ino < sctx->send_progress)
1513 ret = inode_state_did_delete;
1514 else
1515 ret = inode_state_will_delete;
1516 } else {
1517 ret = -ENOENT;
1518 }
1519 } else if (!left_ret) {
1520 if (left_gen == gen) {
1521 if (ino < sctx->send_progress)
1522 ret = inode_state_did_create;
1523 else
1524 ret = inode_state_will_create;
1525 } else {
1526 ret = -ENOENT;
1527 }
1528 } else if (!right_ret) {
1529 if (right_gen == gen) {
1530 if (ino < sctx->send_progress)
1531 ret = inode_state_did_delete;
1532 else
1533 ret = inode_state_will_delete;
1534 } else {
1535 ret = -ENOENT;
1536 }
1537 } else {
1538 ret = -ENOENT;
1539 }
1540
1541 out:
1542 return ret;
1543 }
1544
1545 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1546 {
1547 int ret;
1548
1549 ret = get_cur_inode_state(sctx, ino, gen);
1550 if (ret < 0)
1551 goto out;
1552
1553 if (ret == inode_state_no_change ||
1554 ret == inode_state_did_create ||
1555 ret == inode_state_will_delete)
1556 ret = 1;
1557 else
1558 ret = 0;
1559
1560 out:
1561 return ret;
1562 }
1563
1564 /*
1565 * Helper function to lookup a dir item in a dir.
1566 */
1567 static int lookup_dir_item_inode(struct btrfs_root *root,
1568 u64 dir, const char *name, int name_len,
1569 u64 *found_inode,
1570 u8 *found_type)
1571 {
1572 int ret = 0;
1573 struct btrfs_dir_item *di;
1574 struct btrfs_key key;
1575 struct btrfs_path *path;
1576
1577 path = alloc_path_for_send();
1578 if (!path)
1579 return -ENOMEM;
1580
1581 di = btrfs_lookup_dir_item(NULL, root, path,
1582 dir, name, name_len, 0);
1583 if (!di) {
1584 ret = -ENOENT;
1585 goto out;
1586 }
1587 if (IS_ERR(di)) {
1588 ret = PTR_ERR(di);
1589 goto out;
1590 }
1591 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1592 *found_inode = key.objectid;
1593 *found_type = btrfs_dir_type(path->nodes[0], di);
1594
1595 out:
1596 btrfs_free_path(path);
1597 return ret;
1598 }
1599
1600 /*
1601 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1602 * generation of the parent dir and the name of the dir entry.
1603 */
1604 static int get_first_ref(struct btrfs_root *root, u64 ino,
1605 u64 *dir, u64 *dir_gen, struct fs_path *name)
1606 {
1607 int ret;
1608 struct btrfs_key key;
1609 struct btrfs_key found_key;
1610 struct btrfs_path *path;
1611 int len;
1612 u64 parent_dir;
1613
1614 path = alloc_path_for_send();
1615 if (!path)
1616 return -ENOMEM;
1617
1618 key.objectid = ino;
1619 key.type = BTRFS_INODE_REF_KEY;
1620 key.offset = 0;
1621
1622 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1623 if (ret < 0)
1624 goto out;
1625 if (!ret)
1626 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1627 path->slots[0]);
1628 if (ret || found_key.objectid != ino ||
1629 (found_key.type != BTRFS_INODE_REF_KEY &&
1630 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1631 ret = -ENOENT;
1632 goto out;
1633 }
1634
1635 if (key.type == BTRFS_INODE_REF_KEY) {
1636 struct btrfs_inode_ref *iref;
1637 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1638 struct btrfs_inode_ref);
1639 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1640 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1641 (unsigned long)(iref + 1),
1642 len);
1643 parent_dir = found_key.offset;
1644 } else {
1645 struct btrfs_inode_extref *extref;
1646 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1647 struct btrfs_inode_extref);
1648 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1649 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1650 (unsigned long)&extref->name, len);
1651 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1652 }
1653 if (ret < 0)
1654 goto out;
1655 btrfs_release_path(path);
1656
1657 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1658 NULL, NULL);
1659 if (ret < 0)
1660 goto out;
1661
1662 *dir = parent_dir;
1663
1664 out:
1665 btrfs_free_path(path);
1666 return ret;
1667 }
1668
1669 static int is_first_ref(struct btrfs_root *root,
1670 u64 ino, u64 dir,
1671 const char *name, int name_len)
1672 {
1673 int ret;
1674 struct fs_path *tmp_name;
1675 u64 tmp_dir;
1676 u64 tmp_dir_gen;
1677
1678 tmp_name = fs_path_alloc();
1679 if (!tmp_name)
1680 return -ENOMEM;
1681
1682 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1683 if (ret < 0)
1684 goto out;
1685
1686 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1687 ret = 0;
1688 goto out;
1689 }
1690
1691 ret = !memcmp(tmp_name->start, name, name_len);
1692
1693 out:
1694 fs_path_free(tmp_name);
1695 return ret;
1696 }
1697
1698 /*
1699 * Used by process_recorded_refs to determine if a new ref would overwrite an
1700 * already existing ref. In case it detects an overwrite, it returns the
1701 * inode/gen in who_ino/who_gen.
1702 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1703 * to make sure later references to the overwritten inode are possible.
1704 * Orphanizing is however only required for the first ref of an inode.
1705 * process_recorded_refs does an additional is_first_ref check to see if
1706 * orphanizing is really required.
1707 */
1708 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1709 const char *name, int name_len,
1710 u64 *who_ino, u64 *who_gen)
1711 {
1712 int ret = 0;
1713 u64 gen;
1714 u64 other_inode = 0;
1715 u8 other_type = 0;
1716
1717 if (!sctx->parent_root)
1718 goto out;
1719
1720 ret = is_inode_existent(sctx, dir, dir_gen);
1721 if (ret <= 0)
1722 goto out;
1723
1724 /*
1725 * If we have a parent root we need to verify that the parent dir was
1726 * not delted and then re-created, if it was then we have no overwrite
1727 * and we can just unlink this entry.
1728 */
1729 if (sctx->parent_root) {
1730 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1731 NULL, NULL, NULL);
1732 if (ret < 0 && ret != -ENOENT)
1733 goto out;
1734 if (ret) {
1735 ret = 0;
1736 goto out;
1737 }
1738 if (gen != dir_gen)
1739 goto out;
1740 }
1741
1742 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1743 &other_inode, &other_type);
1744 if (ret < 0 && ret != -ENOENT)
1745 goto out;
1746 if (ret) {
1747 ret = 0;
1748 goto out;
1749 }
1750
1751 /*
1752 * Check if the overwritten ref was already processed. If yes, the ref
1753 * was already unlinked/moved, so we can safely assume that we will not
1754 * overwrite anything at this point in time.
1755 */
1756 if (other_inode > sctx->send_progress) {
1757 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1758 who_gen, NULL, NULL, NULL, NULL);
1759 if (ret < 0)
1760 goto out;
1761
1762 ret = 1;
1763 *who_ino = other_inode;
1764 } else {
1765 ret = 0;
1766 }
1767
1768 out:
1769 return ret;
1770 }
1771
1772 /*
1773 * Checks if the ref was overwritten by an already processed inode. This is
1774 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1775 * thus the orphan name needs be used.
1776 * process_recorded_refs also uses it to avoid unlinking of refs that were
1777 * overwritten.
1778 */
1779 static int did_overwrite_ref(struct send_ctx *sctx,
1780 u64 dir, u64 dir_gen,
1781 u64 ino, u64 ino_gen,
1782 const char *name, int name_len)
1783 {
1784 int ret = 0;
1785 u64 gen;
1786 u64 ow_inode;
1787 u8 other_type;
1788
1789 if (!sctx->parent_root)
1790 goto out;
1791
1792 ret = is_inode_existent(sctx, dir, dir_gen);
1793 if (ret <= 0)
1794 goto out;
1795
1796 /* check if the ref was overwritten by another ref */
1797 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1798 &ow_inode, &other_type);
1799 if (ret < 0 && ret != -ENOENT)
1800 goto out;
1801 if (ret) {
1802 /* was never and will never be overwritten */
1803 ret = 0;
1804 goto out;
1805 }
1806
1807 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1808 NULL, NULL);
1809 if (ret < 0)
1810 goto out;
1811
1812 if (ow_inode == ino && gen == ino_gen) {
1813 ret = 0;
1814 goto out;
1815 }
1816
1817 /* we know that it is or will be overwritten. check this now */
1818 if (ow_inode < sctx->send_progress)
1819 ret = 1;
1820 else
1821 ret = 0;
1822
1823 out:
1824 return ret;
1825 }
1826
1827 /*
1828 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1829 * that got overwritten. This is used by process_recorded_refs to determine
1830 * if it has to use the path as returned by get_cur_path or the orphan name.
1831 */
1832 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1833 {
1834 int ret = 0;
1835 struct fs_path *name = NULL;
1836 u64 dir;
1837 u64 dir_gen;
1838
1839 if (!sctx->parent_root)
1840 goto out;
1841
1842 name = fs_path_alloc();
1843 if (!name)
1844 return -ENOMEM;
1845
1846 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1847 if (ret < 0)
1848 goto out;
1849
1850 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1851 name->start, fs_path_len(name));
1852
1853 out:
1854 fs_path_free(name);
1855 return ret;
1856 }
1857
1858 /*
1859 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1860 * so we need to do some special handling in case we have clashes. This function
1861 * takes care of this with the help of name_cache_entry::radix_list.
1862 * In case of error, nce is kfreed.
1863 */
1864 static int name_cache_insert(struct send_ctx *sctx,
1865 struct name_cache_entry *nce)
1866 {
1867 int ret = 0;
1868 struct list_head *nce_head;
1869
1870 nce_head = radix_tree_lookup(&sctx->name_cache,
1871 (unsigned long)nce->ino);
1872 if (!nce_head) {
1873 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1874 if (!nce_head) {
1875 kfree(nce);
1876 return -ENOMEM;
1877 }
1878 INIT_LIST_HEAD(nce_head);
1879
1880 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1881 if (ret < 0) {
1882 kfree(nce_head);
1883 kfree(nce);
1884 return ret;
1885 }
1886 }
1887 list_add_tail(&nce->radix_list, nce_head);
1888 list_add_tail(&nce->list, &sctx->name_cache_list);
1889 sctx->name_cache_size++;
1890
1891 return ret;
1892 }
1893
1894 static void name_cache_delete(struct send_ctx *sctx,
1895 struct name_cache_entry *nce)
1896 {
1897 struct list_head *nce_head;
1898
1899 nce_head = radix_tree_lookup(&sctx->name_cache,
1900 (unsigned long)nce->ino);
1901 BUG_ON(!nce_head);
1902
1903 list_del(&nce->radix_list);
1904 list_del(&nce->list);
1905 sctx->name_cache_size--;
1906
1907 if (list_empty(nce_head)) {
1908 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1909 kfree(nce_head);
1910 }
1911 }
1912
1913 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1914 u64 ino, u64 gen)
1915 {
1916 struct list_head *nce_head;
1917 struct name_cache_entry *cur;
1918
1919 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1920 if (!nce_head)
1921 return NULL;
1922
1923 list_for_each_entry(cur, nce_head, radix_list) {
1924 if (cur->ino == ino && cur->gen == gen)
1925 return cur;
1926 }
1927 return NULL;
1928 }
1929
1930 /*
1931 * Removes the entry from the list and adds it back to the end. This marks the
1932 * entry as recently used so that name_cache_clean_unused does not remove it.
1933 */
1934 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1935 {
1936 list_del(&nce->list);
1937 list_add_tail(&nce->list, &sctx->name_cache_list);
1938 }
1939
1940 /*
1941 * Remove some entries from the beginning of name_cache_list.
1942 */
1943 static void name_cache_clean_unused(struct send_ctx *sctx)
1944 {
1945 struct name_cache_entry *nce;
1946
1947 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1948 return;
1949
1950 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1951 nce = list_entry(sctx->name_cache_list.next,
1952 struct name_cache_entry, list);
1953 name_cache_delete(sctx, nce);
1954 kfree(nce);
1955 }
1956 }
1957
1958 static void name_cache_free(struct send_ctx *sctx)
1959 {
1960 struct name_cache_entry *nce;
1961
1962 while (!list_empty(&sctx->name_cache_list)) {
1963 nce = list_entry(sctx->name_cache_list.next,
1964 struct name_cache_entry, list);
1965 name_cache_delete(sctx, nce);
1966 kfree(nce);
1967 }
1968 }
1969
1970 /*
1971 * Used by get_cur_path for each ref up to the root.
1972 * Returns 0 if it succeeded.
1973 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1974 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1975 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1976 * Returns <0 in case of error.
1977 */
1978 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1979 u64 ino, u64 gen,
1980 int skip_name_cache,
1981 u64 *parent_ino,
1982 u64 *parent_gen,
1983 struct fs_path *dest)
1984 {
1985 int ret;
1986 int nce_ret;
1987 struct btrfs_path *path = NULL;
1988 struct name_cache_entry *nce = NULL;
1989
1990 if (skip_name_cache)
1991 goto get_ref;
1992 /*
1993 * First check if we already did a call to this function with the same
1994 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1995 * return the cached result.
1996 */
1997 nce = name_cache_search(sctx, ino, gen);
1998 if (nce) {
1999 if (ino < sctx->send_progress && nce->need_later_update) {
2000 name_cache_delete(sctx, nce);
2001 kfree(nce);
2002 nce = NULL;
2003 } else {
2004 name_cache_used(sctx, nce);
2005 *parent_ino = nce->parent_ino;
2006 *parent_gen = nce->parent_gen;
2007 ret = fs_path_add(dest, nce->name, nce->name_len);
2008 if (ret < 0)
2009 goto out;
2010 ret = nce->ret;
2011 goto out;
2012 }
2013 }
2014
2015 path = alloc_path_for_send();
2016 if (!path)
2017 return -ENOMEM;
2018
2019 /*
2020 * If the inode is not existent yet, add the orphan name and return 1.
2021 * This should only happen for the parent dir that we determine in
2022 * __record_new_ref
2023 */
2024 ret = is_inode_existent(sctx, ino, gen);
2025 if (ret < 0)
2026 goto out;
2027
2028 if (!ret) {
2029 ret = gen_unique_name(sctx, ino, gen, dest);
2030 if (ret < 0)
2031 goto out;
2032 ret = 1;
2033 goto out_cache;
2034 }
2035
2036 get_ref:
2037 /*
2038 * Depending on whether the inode was already processed or not, use
2039 * send_root or parent_root for ref lookup.
2040 */
2041 if (ino < sctx->send_progress && !skip_name_cache)
2042 ret = get_first_ref(sctx->send_root, ino,
2043 parent_ino, parent_gen, dest);
2044 else
2045 ret = get_first_ref(sctx->parent_root, ino,
2046 parent_ino, parent_gen, dest);
2047 if (ret < 0)
2048 goto out;
2049
2050 /*
2051 * Check if the ref was overwritten by an inode's ref that was processed
2052 * earlier. If yes, treat as orphan and return 1.
2053 */
2054 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2055 dest->start, dest->end - dest->start);
2056 if (ret < 0)
2057 goto out;
2058 if (ret) {
2059 fs_path_reset(dest);
2060 ret = gen_unique_name(sctx, ino, gen, dest);
2061 if (ret < 0)
2062 goto out;
2063 ret = 1;
2064 }
2065 if (skip_name_cache)
2066 goto out;
2067
2068 out_cache:
2069 /*
2070 * Store the result of the lookup in the name cache.
2071 */
2072 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2073 if (!nce) {
2074 ret = -ENOMEM;
2075 goto out;
2076 }
2077
2078 nce->ino = ino;
2079 nce->gen = gen;
2080 nce->parent_ino = *parent_ino;
2081 nce->parent_gen = *parent_gen;
2082 nce->name_len = fs_path_len(dest);
2083 nce->ret = ret;
2084 strcpy(nce->name, dest->start);
2085
2086 if (ino < sctx->send_progress)
2087 nce->need_later_update = 0;
2088 else
2089 nce->need_later_update = 1;
2090
2091 nce_ret = name_cache_insert(sctx, nce);
2092 if (nce_ret < 0)
2093 ret = nce_ret;
2094 name_cache_clean_unused(sctx);
2095
2096 out:
2097 btrfs_free_path(path);
2098 return ret;
2099 }
2100
2101 /*
2102 * Magic happens here. This function returns the first ref to an inode as it
2103 * would look like while receiving the stream at this point in time.
2104 * We walk the path up to the root. For every inode in between, we check if it
2105 * was already processed/sent. If yes, we continue with the parent as found
2106 * in send_root. If not, we continue with the parent as found in parent_root.
2107 * If we encounter an inode that was deleted at this point in time, we use the
2108 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2109 * that were not created yet and overwritten inodes/refs.
2110 *
2111 * When do we have have orphan inodes:
2112 * 1. When an inode is freshly created and thus no valid refs are available yet
2113 * 2. When a directory lost all it's refs (deleted) but still has dir items
2114 * inside which were not processed yet (pending for move/delete). If anyone
2115 * tried to get the path to the dir items, it would get a path inside that
2116 * orphan directory.
2117 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2118 * of an unprocessed inode. If in that case the first ref would be
2119 * overwritten, the overwritten inode gets "orphanized". Later when we
2120 * process this overwritten inode, it is restored at a new place by moving
2121 * the orphan inode.
2122 *
2123 * sctx->send_progress tells this function at which point in time receiving
2124 * would be.
2125 */
2126 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2127 struct fs_path *dest)
2128 {
2129 int ret = 0;
2130 struct fs_path *name = NULL;
2131 u64 parent_inode = 0;
2132 u64 parent_gen = 0;
2133 int stop = 0;
2134 u64 start_ino = ino;
2135 u64 start_gen = gen;
2136 int skip_name_cache = 0;
2137
2138 name = fs_path_alloc();
2139 if (!name) {
2140 ret = -ENOMEM;
2141 goto out;
2142 }
2143
2144 if (is_waiting_for_move(sctx, ino))
2145 skip_name_cache = 1;
2146
2147 again:
2148 dest->reversed = 1;
2149 fs_path_reset(dest);
2150
2151 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2152 fs_path_reset(name);
2153
2154 ret = __get_cur_name_and_parent(sctx, ino, gen, skip_name_cache,
2155 &parent_inode, &parent_gen, name);
2156 if (ret < 0)
2157 goto out;
2158 if (ret)
2159 stop = 1;
2160
2161 if (!skip_name_cache &&
2162 is_waiting_for_move(sctx, parent_inode)) {
2163 ino = start_ino;
2164 gen = start_gen;
2165 stop = 0;
2166 skip_name_cache = 1;
2167 goto again;
2168 }
2169
2170 ret = fs_path_add_path(dest, name);
2171 if (ret < 0)
2172 goto out;
2173
2174 ino = parent_inode;
2175 gen = parent_gen;
2176 }
2177
2178 out:
2179 fs_path_free(name);
2180 if (!ret)
2181 fs_path_unreverse(dest);
2182 return ret;
2183 }
2184
2185 /*
2186 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2187 */
2188 static int send_subvol_begin(struct send_ctx *sctx)
2189 {
2190 int ret;
2191 struct btrfs_root *send_root = sctx->send_root;
2192 struct btrfs_root *parent_root = sctx->parent_root;
2193 struct btrfs_path *path;
2194 struct btrfs_key key;
2195 struct btrfs_root_ref *ref;
2196 struct extent_buffer *leaf;
2197 char *name = NULL;
2198 int namelen;
2199
2200 path = btrfs_alloc_path();
2201 if (!path)
2202 return -ENOMEM;
2203
2204 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2205 if (!name) {
2206 btrfs_free_path(path);
2207 return -ENOMEM;
2208 }
2209
2210 key.objectid = send_root->objectid;
2211 key.type = BTRFS_ROOT_BACKREF_KEY;
2212 key.offset = 0;
2213
2214 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2215 &key, path, 1, 0);
2216 if (ret < 0)
2217 goto out;
2218 if (ret) {
2219 ret = -ENOENT;
2220 goto out;
2221 }
2222
2223 leaf = path->nodes[0];
2224 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2225 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2226 key.objectid != send_root->objectid) {
2227 ret = -ENOENT;
2228 goto out;
2229 }
2230 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2231 namelen = btrfs_root_ref_name_len(leaf, ref);
2232 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2233 btrfs_release_path(path);
2234
2235 if (parent_root) {
2236 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2237 if (ret < 0)
2238 goto out;
2239 } else {
2240 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2241 if (ret < 0)
2242 goto out;
2243 }
2244
2245 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2246 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2247 sctx->send_root->root_item.uuid);
2248 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2249 le64_to_cpu(sctx->send_root->root_item.ctransid));
2250 if (parent_root) {
2251 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2252 sctx->parent_root->root_item.uuid);
2253 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2254 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2255 }
2256
2257 ret = send_cmd(sctx);
2258
2259 tlv_put_failure:
2260 out:
2261 btrfs_free_path(path);
2262 kfree(name);
2263 return ret;
2264 }
2265
2266 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2267 {
2268 int ret = 0;
2269 struct fs_path *p;
2270
2271 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2272
2273 p = fs_path_alloc();
2274 if (!p)
2275 return -ENOMEM;
2276
2277 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2278 if (ret < 0)
2279 goto out;
2280
2281 ret = get_cur_path(sctx, ino, gen, p);
2282 if (ret < 0)
2283 goto out;
2284 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2285 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2286
2287 ret = send_cmd(sctx);
2288
2289 tlv_put_failure:
2290 out:
2291 fs_path_free(p);
2292 return ret;
2293 }
2294
2295 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2296 {
2297 int ret = 0;
2298 struct fs_path *p;
2299
2300 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2301
2302 p = fs_path_alloc();
2303 if (!p)
2304 return -ENOMEM;
2305
2306 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2307 if (ret < 0)
2308 goto out;
2309
2310 ret = get_cur_path(sctx, ino, gen, p);
2311 if (ret < 0)
2312 goto out;
2313 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2314 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2315
2316 ret = send_cmd(sctx);
2317
2318 tlv_put_failure:
2319 out:
2320 fs_path_free(p);
2321 return ret;
2322 }
2323
2324 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2325 {
2326 int ret = 0;
2327 struct fs_path *p;
2328
2329 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2330
2331 p = fs_path_alloc();
2332 if (!p)
2333 return -ENOMEM;
2334
2335 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2336 if (ret < 0)
2337 goto out;
2338
2339 ret = get_cur_path(sctx, ino, gen, p);
2340 if (ret < 0)
2341 goto out;
2342 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2343 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2344 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2345
2346 ret = send_cmd(sctx);
2347
2348 tlv_put_failure:
2349 out:
2350 fs_path_free(p);
2351 return ret;
2352 }
2353
2354 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2355 {
2356 int ret = 0;
2357 struct fs_path *p = NULL;
2358 struct btrfs_inode_item *ii;
2359 struct btrfs_path *path = NULL;
2360 struct extent_buffer *eb;
2361 struct btrfs_key key;
2362 int slot;
2363
2364 verbose_printk("btrfs: send_utimes %llu\n", ino);
2365
2366 p = fs_path_alloc();
2367 if (!p)
2368 return -ENOMEM;
2369
2370 path = alloc_path_for_send();
2371 if (!path) {
2372 ret = -ENOMEM;
2373 goto out;
2374 }
2375
2376 key.objectid = ino;
2377 key.type = BTRFS_INODE_ITEM_KEY;
2378 key.offset = 0;
2379 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2380 if (ret < 0)
2381 goto out;
2382
2383 eb = path->nodes[0];
2384 slot = path->slots[0];
2385 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2386
2387 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2388 if (ret < 0)
2389 goto out;
2390
2391 ret = get_cur_path(sctx, ino, gen, p);
2392 if (ret < 0)
2393 goto out;
2394 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2395 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2396 btrfs_inode_atime(ii));
2397 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2398 btrfs_inode_mtime(ii));
2399 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2400 btrfs_inode_ctime(ii));
2401 /* TODO Add otime support when the otime patches get into upstream */
2402
2403 ret = send_cmd(sctx);
2404
2405 tlv_put_failure:
2406 out:
2407 fs_path_free(p);
2408 btrfs_free_path(path);
2409 return ret;
2410 }
2411
2412 /*
2413 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2414 * a valid path yet because we did not process the refs yet. So, the inode
2415 * is created as orphan.
2416 */
2417 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2418 {
2419 int ret = 0;
2420 struct fs_path *p;
2421 int cmd;
2422 u64 gen;
2423 u64 mode;
2424 u64 rdev;
2425
2426 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2427
2428 p = fs_path_alloc();
2429 if (!p)
2430 return -ENOMEM;
2431
2432 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2433 NULL, &rdev);
2434 if (ret < 0)
2435 goto out;
2436
2437 if (S_ISREG(mode)) {
2438 cmd = BTRFS_SEND_C_MKFILE;
2439 } else if (S_ISDIR(mode)) {
2440 cmd = BTRFS_SEND_C_MKDIR;
2441 } else if (S_ISLNK(mode)) {
2442 cmd = BTRFS_SEND_C_SYMLINK;
2443 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2444 cmd = BTRFS_SEND_C_MKNOD;
2445 } else if (S_ISFIFO(mode)) {
2446 cmd = BTRFS_SEND_C_MKFIFO;
2447 } else if (S_ISSOCK(mode)) {
2448 cmd = BTRFS_SEND_C_MKSOCK;
2449 } else {
2450 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2451 (int)(mode & S_IFMT));
2452 ret = -ENOTSUPP;
2453 goto out;
2454 }
2455
2456 ret = begin_cmd(sctx, cmd);
2457 if (ret < 0)
2458 goto out;
2459
2460 ret = gen_unique_name(sctx, ino, gen, p);
2461 if (ret < 0)
2462 goto out;
2463
2464 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2465 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2466
2467 if (S_ISLNK(mode)) {
2468 fs_path_reset(p);
2469 ret = read_symlink(sctx->send_root, ino, p);
2470 if (ret < 0)
2471 goto out;
2472 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2473 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2474 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2475 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2476 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2477 }
2478
2479 ret = send_cmd(sctx);
2480 if (ret < 0)
2481 goto out;
2482
2483
2484 tlv_put_failure:
2485 out:
2486 fs_path_free(p);
2487 return ret;
2488 }
2489
2490 /*
2491 * We need some special handling for inodes that get processed before the parent
2492 * directory got created. See process_recorded_refs for details.
2493 * This function does the check if we already created the dir out of order.
2494 */
2495 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2496 {
2497 int ret = 0;
2498 struct btrfs_path *path = NULL;
2499 struct btrfs_key key;
2500 struct btrfs_key found_key;
2501 struct btrfs_key di_key;
2502 struct extent_buffer *eb;
2503 struct btrfs_dir_item *di;
2504 int slot;
2505
2506 path = alloc_path_for_send();
2507 if (!path) {
2508 ret = -ENOMEM;
2509 goto out;
2510 }
2511
2512 key.objectid = dir;
2513 key.type = BTRFS_DIR_INDEX_KEY;
2514 key.offset = 0;
2515 while (1) {
2516 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2517 1, 0);
2518 if (ret < 0)
2519 goto out;
2520 if (!ret) {
2521 eb = path->nodes[0];
2522 slot = path->slots[0];
2523 btrfs_item_key_to_cpu(eb, &found_key, slot);
2524 }
2525 if (ret || found_key.objectid != key.objectid ||
2526 found_key.type != key.type) {
2527 ret = 0;
2528 goto out;
2529 }
2530
2531 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2532 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2533
2534 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2535 di_key.objectid < sctx->send_progress) {
2536 ret = 1;
2537 goto out;
2538 }
2539
2540 key.offset = found_key.offset + 1;
2541 btrfs_release_path(path);
2542 }
2543
2544 out:
2545 btrfs_free_path(path);
2546 return ret;
2547 }
2548
2549 /*
2550 * Only creates the inode if it is:
2551 * 1. Not a directory
2552 * 2. Or a directory which was not created already due to out of order
2553 * directories. See did_create_dir and process_recorded_refs for details.
2554 */
2555 static int send_create_inode_if_needed(struct send_ctx *sctx)
2556 {
2557 int ret;
2558
2559 if (S_ISDIR(sctx->cur_inode_mode)) {
2560 ret = did_create_dir(sctx, sctx->cur_ino);
2561 if (ret < 0)
2562 goto out;
2563 if (ret) {
2564 ret = 0;
2565 goto out;
2566 }
2567 }
2568
2569 ret = send_create_inode(sctx, sctx->cur_ino);
2570 if (ret < 0)
2571 goto out;
2572
2573 out:
2574 return ret;
2575 }
2576
2577 struct recorded_ref {
2578 struct list_head list;
2579 char *dir_path;
2580 char *name;
2581 struct fs_path *full_path;
2582 u64 dir;
2583 u64 dir_gen;
2584 int dir_path_len;
2585 int name_len;
2586 };
2587
2588 /*
2589 * We need to process new refs before deleted refs, but compare_tree gives us
2590 * everything mixed. So we first record all refs and later process them.
2591 * This function is a helper to record one ref.
2592 */
2593 static int record_ref(struct list_head *head, u64 dir,
2594 u64 dir_gen, struct fs_path *path)
2595 {
2596 struct recorded_ref *ref;
2597
2598 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2599 if (!ref)
2600 return -ENOMEM;
2601
2602 ref->dir = dir;
2603 ref->dir_gen = dir_gen;
2604 ref->full_path = path;
2605
2606 ref->name = (char *)kbasename(ref->full_path->start);
2607 ref->name_len = ref->full_path->end - ref->name;
2608 ref->dir_path = ref->full_path->start;
2609 if (ref->name == ref->full_path->start)
2610 ref->dir_path_len = 0;
2611 else
2612 ref->dir_path_len = ref->full_path->end -
2613 ref->full_path->start - 1 - ref->name_len;
2614
2615 list_add_tail(&ref->list, head);
2616 return 0;
2617 }
2618
2619 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2620 {
2621 struct recorded_ref *new;
2622
2623 new = kmalloc(sizeof(*ref), GFP_NOFS);
2624 if (!new)
2625 return -ENOMEM;
2626
2627 new->dir = ref->dir;
2628 new->dir_gen = ref->dir_gen;
2629 new->full_path = NULL;
2630 INIT_LIST_HEAD(&new->list);
2631 list_add_tail(&new->list, list);
2632 return 0;
2633 }
2634
2635 static void __free_recorded_refs(struct list_head *head)
2636 {
2637 struct recorded_ref *cur;
2638
2639 while (!list_empty(head)) {
2640 cur = list_entry(head->next, struct recorded_ref, list);
2641 fs_path_free(cur->full_path);
2642 list_del(&cur->list);
2643 kfree(cur);
2644 }
2645 }
2646
2647 static void free_recorded_refs(struct send_ctx *sctx)
2648 {
2649 __free_recorded_refs(&sctx->new_refs);
2650 __free_recorded_refs(&sctx->deleted_refs);
2651 }
2652
2653 /*
2654 * Renames/moves a file/dir to its orphan name. Used when the first
2655 * ref of an unprocessed inode gets overwritten and for all non empty
2656 * directories.
2657 */
2658 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2659 struct fs_path *path)
2660 {
2661 int ret;
2662 struct fs_path *orphan;
2663
2664 orphan = fs_path_alloc();
2665 if (!orphan)
2666 return -ENOMEM;
2667
2668 ret = gen_unique_name(sctx, ino, gen, orphan);
2669 if (ret < 0)
2670 goto out;
2671
2672 ret = send_rename(sctx, path, orphan);
2673
2674 out:
2675 fs_path_free(orphan);
2676 return ret;
2677 }
2678
2679 /*
2680 * Returns 1 if a directory can be removed at this point in time.
2681 * We check this by iterating all dir items and checking if the inode behind
2682 * the dir item was already processed.
2683 */
2684 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2685 {
2686 int ret = 0;
2687 struct btrfs_root *root = sctx->parent_root;
2688 struct btrfs_path *path;
2689 struct btrfs_key key;
2690 struct btrfs_key found_key;
2691 struct btrfs_key loc;
2692 struct btrfs_dir_item *di;
2693
2694 /*
2695 * Don't try to rmdir the top/root subvolume dir.
2696 */
2697 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2698 return 0;
2699
2700 path = alloc_path_for_send();
2701 if (!path)
2702 return -ENOMEM;
2703
2704 key.objectid = dir;
2705 key.type = BTRFS_DIR_INDEX_KEY;
2706 key.offset = 0;
2707
2708 while (1) {
2709 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2710 if (ret < 0)
2711 goto out;
2712 if (!ret) {
2713 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2714 path->slots[0]);
2715 }
2716 if (ret || found_key.objectid != key.objectid ||
2717 found_key.type != key.type) {
2718 break;
2719 }
2720
2721 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2722 struct btrfs_dir_item);
2723 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2724
2725 if (loc.objectid > send_progress) {
2726 ret = 0;
2727 goto out;
2728 }
2729
2730 btrfs_release_path(path);
2731 key.offset = found_key.offset + 1;
2732 }
2733
2734 ret = 1;
2735
2736 out:
2737 btrfs_free_path(path);
2738 return ret;
2739 }
2740
2741 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2742 {
2743 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2744 struct waiting_dir_move *entry;
2745
2746 while (n) {
2747 entry = rb_entry(n, struct waiting_dir_move, node);
2748 if (ino < entry->ino)
2749 n = n->rb_left;
2750 else if (ino > entry->ino)
2751 n = n->rb_right;
2752 else
2753 return 1;
2754 }
2755 return 0;
2756 }
2757
2758 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2759 {
2760 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2761 struct rb_node *parent = NULL;
2762 struct waiting_dir_move *entry, *dm;
2763
2764 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2765 if (!dm)
2766 return -ENOMEM;
2767 dm->ino = ino;
2768
2769 while (*p) {
2770 parent = *p;
2771 entry = rb_entry(parent, struct waiting_dir_move, node);
2772 if (ino < entry->ino) {
2773 p = &(*p)->rb_left;
2774 } else if (ino > entry->ino) {
2775 p = &(*p)->rb_right;
2776 } else {
2777 kfree(dm);
2778 return -EEXIST;
2779 }
2780 }
2781
2782 rb_link_node(&dm->node, parent, p);
2783 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2784 return 0;
2785 }
2786
2787 static int del_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2788 {
2789 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2790 struct waiting_dir_move *entry;
2791
2792 while (n) {
2793 entry = rb_entry(n, struct waiting_dir_move, node);
2794 if (ino < entry->ino) {
2795 n = n->rb_left;
2796 } else if (ino > entry->ino) {
2797 n = n->rb_right;
2798 } else {
2799 rb_erase(&entry->node, &sctx->waiting_dir_moves);
2800 kfree(entry);
2801 return 0;
2802 }
2803 }
2804 return -ENOENT;
2805 }
2806
2807 static int add_pending_dir_move(struct send_ctx *sctx, u64 parent_ino)
2808 {
2809 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2810 struct rb_node *parent = NULL;
2811 struct pending_dir_move *entry, *pm;
2812 struct recorded_ref *cur;
2813 int exists = 0;
2814 int ret;
2815
2816 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2817 if (!pm)
2818 return -ENOMEM;
2819 pm->parent_ino = parent_ino;
2820 pm->ino = sctx->cur_ino;
2821 pm->gen = sctx->cur_inode_gen;
2822 INIT_LIST_HEAD(&pm->list);
2823 INIT_LIST_HEAD(&pm->update_refs);
2824 RB_CLEAR_NODE(&pm->node);
2825
2826 while (*p) {
2827 parent = *p;
2828 entry = rb_entry(parent, struct pending_dir_move, node);
2829 if (parent_ino < entry->parent_ino) {
2830 p = &(*p)->rb_left;
2831 } else if (parent_ino > entry->parent_ino) {
2832 p = &(*p)->rb_right;
2833 } else {
2834 exists = 1;
2835 break;
2836 }
2837 }
2838
2839 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2840 ret = dup_ref(cur, &pm->update_refs);
2841 if (ret < 0)
2842 goto out;
2843 }
2844 list_for_each_entry(cur, &sctx->new_refs, list) {
2845 ret = dup_ref(cur, &pm->update_refs);
2846 if (ret < 0)
2847 goto out;
2848 }
2849
2850 ret = add_waiting_dir_move(sctx, pm->ino);
2851 if (ret)
2852 goto out;
2853
2854 if (exists) {
2855 list_add_tail(&pm->list, &entry->list);
2856 } else {
2857 rb_link_node(&pm->node, parent, p);
2858 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
2859 }
2860 ret = 0;
2861 out:
2862 if (ret) {
2863 __free_recorded_refs(&pm->update_refs);
2864 kfree(pm);
2865 }
2866 return ret;
2867 }
2868
2869 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
2870 u64 parent_ino)
2871 {
2872 struct rb_node *n = sctx->pending_dir_moves.rb_node;
2873 struct pending_dir_move *entry;
2874
2875 while (n) {
2876 entry = rb_entry(n, struct pending_dir_move, node);
2877 if (parent_ino < entry->parent_ino)
2878 n = n->rb_left;
2879 else if (parent_ino > entry->parent_ino)
2880 n = n->rb_right;
2881 else
2882 return entry;
2883 }
2884 return NULL;
2885 }
2886
2887 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
2888 {
2889 struct fs_path *from_path = NULL;
2890 struct fs_path *to_path = NULL;
2891 u64 orig_progress = sctx->send_progress;
2892 struct recorded_ref *cur;
2893 int ret;
2894
2895 from_path = fs_path_alloc();
2896 if (!from_path)
2897 return -ENOMEM;
2898
2899 sctx->send_progress = pm->ino;
2900 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
2901 if (ret < 0)
2902 goto out;
2903
2904 to_path = fs_path_alloc();
2905 if (!to_path) {
2906 ret = -ENOMEM;
2907 goto out;
2908 }
2909
2910 sctx->send_progress = sctx->cur_ino + 1;
2911 ret = del_waiting_dir_move(sctx, pm->ino);
2912 ASSERT(ret == 0);
2913
2914 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
2915 if (ret < 0)
2916 goto out;
2917
2918 ret = send_rename(sctx, from_path, to_path);
2919 if (ret < 0)
2920 goto out;
2921
2922 ret = send_utimes(sctx, pm->ino, pm->gen);
2923 if (ret < 0)
2924 goto out;
2925
2926 /*
2927 * After rename/move, need to update the utimes of both new parent(s)
2928 * and old parent(s).
2929 */
2930 list_for_each_entry(cur, &pm->update_refs, list) {
2931 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
2932 if (ret < 0)
2933 goto out;
2934 }
2935
2936 out:
2937 fs_path_free(from_path);
2938 fs_path_free(to_path);
2939 sctx->send_progress = orig_progress;
2940
2941 return ret;
2942 }
2943
2944 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
2945 {
2946 if (!list_empty(&m->list))
2947 list_del(&m->list);
2948 if (!RB_EMPTY_NODE(&m->node))
2949 rb_erase(&m->node, &sctx->pending_dir_moves);
2950 __free_recorded_refs(&m->update_refs);
2951 kfree(m);
2952 }
2953
2954 static void tail_append_pending_moves(struct pending_dir_move *moves,
2955 struct list_head *stack)
2956 {
2957 if (list_empty(&moves->list)) {
2958 list_add_tail(&moves->list, stack);
2959 } else {
2960 LIST_HEAD(list);
2961 list_splice_init(&moves->list, &list);
2962 list_add_tail(&moves->list, stack);
2963 list_splice_tail(&list, stack);
2964 }
2965 }
2966
2967 static int apply_children_dir_moves(struct send_ctx *sctx)
2968 {
2969 struct pending_dir_move *pm;
2970 struct list_head stack;
2971 u64 parent_ino = sctx->cur_ino;
2972 int ret = 0;
2973
2974 pm = get_pending_dir_moves(sctx, parent_ino);
2975 if (!pm)
2976 return 0;
2977
2978 INIT_LIST_HEAD(&stack);
2979 tail_append_pending_moves(pm, &stack);
2980
2981 while (!list_empty(&stack)) {
2982 pm = list_first_entry(&stack, struct pending_dir_move, list);
2983 parent_ino = pm->ino;
2984 ret = apply_dir_move(sctx, pm);
2985 free_pending_move(sctx, pm);
2986 if (ret)
2987 goto out;
2988 pm = get_pending_dir_moves(sctx, parent_ino);
2989 if (pm)
2990 tail_append_pending_moves(pm, &stack);
2991 }
2992 return 0;
2993
2994 out:
2995 while (!list_empty(&stack)) {
2996 pm = list_first_entry(&stack, struct pending_dir_move, list);
2997 free_pending_move(sctx, pm);
2998 }
2999 return ret;
3000 }
3001
3002 static int wait_for_parent_move(struct send_ctx *sctx,
3003 struct recorded_ref *parent_ref)
3004 {
3005 int ret;
3006 u64 ino = parent_ref->dir;
3007 u64 parent_ino_before, parent_ino_after;
3008 u64 new_gen, old_gen;
3009 struct fs_path *path_before = NULL;
3010 struct fs_path *path_after = NULL;
3011 int len1, len2;
3012
3013 if (parent_ref->dir <= sctx->cur_ino)
3014 return 0;
3015
3016 if (is_waiting_for_move(sctx, ino))
3017 return 1;
3018
3019 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3020 NULL, NULL, NULL, NULL);
3021 if (ret == -ENOENT)
3022 return 0;
3023 else if (ret < 0)
3024 return ret;
3025
3026 ret = get_inode_info(sctx->send_root, ino, NULL, &new_gen,
3027 NULL, NULL, NULL, NULL);
3028 if (ret < 0)
3029 return ret;
3030
3031 if (new_gen != old_gen)
3032 return 0;
3033
3034 path_before = fs_path_alloc();
3035 if (!path_before)
3036 return -ENOMEM;
3037
3038 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3039 NULL, path_before);
3040 if (ret == -ENOENT) {
3041 ret = 0;
3042 goto out;
3043 } else if (ret < 0) {
3044 goto out;
3045 }
3046
3047 path_after = fs_path_alloc();
3048 if (!path_after) {
3049 ret = -ENOMEM;
3050 goto out;
3051 }
3052
3053 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3054 NULL, path_after);
3055 if (ret == -ENOENT) {
3056 ret = 0;
3057 goto out;
3058 } else if (ret < 0) {
3059 goto out;
3060 }
3061
3062 len1 = fs_path_len(path_before);
3063 len2 = fs_path_len(path_after);
3064 if ((parent_ino_before != parent_ino_after) && (len1 != len2 ||
3065 memcmp(path_before->start, path_after->start, len1))) {
3066 ret = 1;
3067 goto out;
3068 }
3069 ret = 0;
3070
3071 out:
3072 fs_path_free(path_before);
3073 fs_path_free(path_after);
3074
3075 return ret;
3076 }
3077
3078 /*
3079 * This does all the move/link/unlink/rmdir magic.
3080 */
3081 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3082 {
3083 int ret = 0;
3084 struct recorded_ref *cur;
3085 struct recorded_ref *cur2;
3086 struct list_head check_dirs;
3087 struct fs_path *valid_path = NULL;
3088 u64 ow_inode = 0;
3089 u64 ow_gen;
3090 int did_overwrite = 0;
3091 int is_orphan = 0;
3092
3093 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3094
3095 /*
3096 * This should never happen as the root dir always has the same ref
3097 * which is always '..'
3098 */
3099 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3100 INIT_LIST_HEAD(&check_dirs);
3101
3102 valid_path = fs_path_alloc();
3103 if (!valid_path) {
3104 ret = -ENOMEM;
3105 goto out;
3106 }
3107
3108 /*
3109 * First, check if the first ref of the current inode was overwritten
3110 * before. If yes, we know that the current inode was already orphanized
3111 * and thus use the orphan name. If not, we can use get_cur_path to
3112 * get the path of the first ref as it would like while receiving at
3113 * this point in time.
3114 * New inodes are always orphan at the beginning, so force to use the
3115 * orphan name in this case.
3116 * The first ref is stored in valid_path and will be updated if it
3117 * gets moved around.
3118 */
3119 if (!sctx->cur_inode_new) {
3120 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3121 sctx->cur_inode_gen);
3122 if (ret < 0)
3123 goto out;
3124 if (ret)
3125 did_overwrite = 1;
3126 }
3127 if (sctx->cur_inode_new || did_overwrite) {
3128 ret = gen_unique_name(sctx, sctx->cur_ino,
3129 sctx->cur_inode_gen, valid_path);
3130 if (ret < 0)
3131 goto out;
3132 is_orphan = 1;
3133 } else {
3134 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3135 valid_path);
3136 if (ret < 0)
3137 goto out;
3138 }
3139
3140 list_for_each_entry(cur, &sctx->new_refs, list) {
3141 /*
3142 * We may have refs where the parent directory does not exist
3143 * yet. This happens if the parent directories inum is higher
3144 * the the current inum. To handle this case, we create the
3145 * parent directory out of order. But we need to check if this
3146 * did already happen before due to other refs in the same dir.
3147 */
3148 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3149 if (ret < 0)
3150 goto out;
3151 if (ret == inode_state_will_create) {
3152 ret = 0;
3153 /*
3154 * First check if any of the current inodes refs did
3155 * already create the dir.
3156 */
3157 list_for_each_entry(cur2, &sctx->new_refs, list) {
3158 if (cur == cur2)
3159 break;
3160 if (cur2->dir == cur->dir) {
3161 ret = 1;
3162 break;
3163 }
3164 }
3165
3166 /*
3167 * If that did not happen, check if a previous inode
3168 * did already create the dir.
3169 */
3170 if (!ret)
3171 ret = did_create_dir(sctx, cur->dir);
3172 if (ret < 0)
3173 goto out;
3174 if (!ret) {
3175 ret = send_create_inode(sctx, cur->dir);
3176 if (ret < 0)
3177 goto out;
3178 }
3179 }
3180
3181 /*
3182 * Check if this new ref would overwrite the first ref of
3183 * another unprocessed inode. If yes, orphanize the
3184 * overwritten inode. If we find an overwritten ref that is
3185 * not the first ref, simply unlink it.
3186 */
3187 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3188 cur->name, cur->name_len,
3189 &ow_inode, &ow_gen);
3190 if (ret < 0)
3191 goto out;
3192 if (ret) {
3193 ret = is_first_ref(sctx->parent_root,
3194 ow_inode, cur->dir, cur->name,
3195 cur->name_len);
3196 if (ret < 0)
3197 goto out;
3198 if (ret) {
3199 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3200 cur->full_path);
3201 if (ret < 0)
3202 goto out;
3203 } else {
3204 ret = send_unlink(sctx, cur->full_path);
3205 if (ret < 0)
3206 goto out;
3207 }
3208 }
3209
3210 /*
3211 * link/move the ref to the new place. If we have an orphan
3212 * inode, move it and update valid_path. If not, link or move
3213 * it depending on the inode mode.
3214 */
3215 if (is_orphan) {
3216 ret = send_rename(sctx, valid_path, cur->full_path);
3217 if (ret < 0)
3218 goto out;
3219 is_orphan = 0;
3220 ret = fs_path_copy(valid_path, cur->full_path);
3221 if (ret < 0)
3222 goto out;
3223 } else {
3224 if (S_ISDIR(sctx->cur_inode_mode)) {
3225 /*
3226 * Dirs can't be linked, so move it. For moved
3227 * dirs, we always have one new and one deleted
3228 * ref. The deleted ref is ignored later.
3229 */
3230 if (wait_for_parent_move(sctx, cur)) {
3231 ret = add_pending_dir_move(sctx,
3232 cur->dir);
3233 *pending_move = 1;
3234 } else {
3235 ret = send_rename(sctx, valid_path,
3236 cur->full_path);
3237 if (!ret)
3238 ret = fs_path_copy(valid_path,
3239 cur->full_path);
3240 }
3241 if (ret < 0)
3242 goto out;
3243 } else {
3244 ret = send_link(sctx, cur->full_path,
3245 valid_path);
3246 if (ret < 0)
3247 goto out;
3248 }
3249 }
3250 ret = dup_ref(cur, &check_dirs);
3251 if (ret < 0)
3252 goto out;
3253 }
3254
3255 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3256 /*
3257 * Check if we can already rmdir the directory. If not,
3258 * orphanize it. For every dir item inside that gets deleted
3259 * later, we do this check again and rmdir it then if possible.
3260 * See the use of check_dirs for more details.
3261 */
3262 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
3263 if (ret < 0)
3264 goto out;
3265 if (ret) {
3266 ret = send_rmdir(sctx, valid_path);
3267 if (ret < 0)
3268 goto out;
3269 } else if (!is_orphan) {
3270 ret = orphanize_inode(sctx, sctx->cur_ino,
3271 sctx->cur_inode_gen, valid_path);
3272 if (ret < 0)
3273 goto out;
3274 is_orphan = 1;
3275 }
3276
3277 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3278 ret = dup_ref(cur, &check_dirs);
3279 if (ret < 0)
3280 goto out;
3281 }
3282 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3283 !list_empty(&sctx->deleted_refs)) {
3284 /*
3285 * We have a moved dir. Add the old parent to check_dirs
3286 */
3287 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3288 list);
3289 ret = dup_ref(cur, &check_dirs);
3290 if (ret < 0)
3291 goto out;
3292 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3293 /*
3294 * We have a non dir inode. Go through all deleted refs and
3295 * unlink them if they were not already overwritten by other
3296 * inodes.
3297 */
3298 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3299 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3300 sctx->cur_ino, sctx->cur_inode_gen,
3301 cur->name, cur->name_len);
3302 if (ret < 0)
3303 goto out;
3304 if (!ret) {
3305 ret = send_unlink(sctx, cur->full_path);
3306 if (ret < 0)
3307 goto out;
3308 }
3309 ret = dup_ref(cur, &check_dirs);
3310 if (ret < 0)
3311 goto out;
3312 }
3313 /*
3314 * If the inode is still orphan, unlink the orphan. This may
3315 * happen when a previous inode did overwrite the first ref
3316 * of this inode and no new refs were added for the current
3317 * inode. Unlinking does not mean that the inode is deleted in
3318 * all cases. There may still be links to this inode in other
3319 * places.
3320 */
3321 if (is_orphan) {
3322 ret = send_unlink(sctx, valid_path);
3323 if (ret < 0)
3324 goto out;
3325 }
3326 }
3327
3328 /*
3329 * We did collect all parent dirs where cur_inode was once located. We
3330 * now go through all these dirs and check if they are pending for
3331 * deletion and if it's finally possible to perform the rmdir now.
3332 * We also update the inode stats of the parent dirs here.
3333 */
3334 list_for_each_entry(cur, &check_dirs, list) {
3335 /*
3336 * In case we had refs into dirs that were not processed yet,
3337 * we don't need to do the utime and rmdir logic for these dirs.
3338 * The dir will be processed later.
3339 */
3340 if (cur->dir > sctx->cur_ino)
3341 continue;
3342
3343 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3344 if (ret < 0)
3345 goto out;
3346
3347 if (ret == inode_state_did_create ||
3348 ret == inode_state_no_change) {
3349 /* TODO delayed utimes */
3350 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3351 if (ret < 0)
3352 goto out;
3353 } else if (ret == inode_state_did_delete) {
3354 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
3355 if (ret < 0)
3356 goto out;
3357 if (ret) {
3358 ret = get_cur_path(sctx, cur->dir,
3359 cur->dir_gen, valid_path);
3360 if (ret < 0)
3361 goto out;
3362 ret = send_rmdir(sctx, valid_path);
3363 if (ret < 0)
3364 goto out;
3365 }
3366 }
3367 }
3368
3369 ret = 0;
3370
3371 out:
3372 __free_recorded_refs(&check_dirs);
3373 free_recorded_refs(sctx);
3374 fs_path_free(valid_path);
3375 return ret;
3376 }
3377
3378 static int __record_new_ref(int num, u64 dir, int index,
3379 struct fs_path *name,
3380 void *ctx)
3381 {
3382 int ret = 0;
3383 struct send_ctx *sctx = ctx;
3384 struct fs_path *p;
3385 u64 gen;
3386
3387 p = fs_path_alloc();
3388 if (!p)
3389 return -ENOMEM;
3390
3391 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
3392 NULL, NULL);
3393 if (ret < 0)
3394 goto out;
3395
3396 ret = get_cur_path(sctx, dir, gen, p);
3397 if (ret < 0)
3398 goto out;
3399 ret = fs_path_add_path(p, name);
3400 if (ret < 0)
3401 goto out;
3402
3403 ret = record_ref(&sctx->new_refs, dir, gen, p);
3404
3405 out:
3406 if (ret)
3407 fs_path_free(p);
3408 return ret;
3409 }
3410
3411 static int __record_deleted_ref(int num, u64 dir, int index,
3412 struct fs_path *name,
3413 void *ctx)
3414 {
3415 int ret = 0;
3416 struct send_ctx *sctx = ctx;
3417 struct fs_path *p;
3418 u64 gen;
3419
3420 p = fs_path_alloc();
3421 if (!p)
3422 return -ENOMEM;
3423
3424 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3425 NULL, NULL);
3426 if (ret < 0)
3427 goto out;
3428
3429 ret = get_cur_path(sctx, dir, gen, p);
3430 if (ret < 0)
3431 goto out;
3432 ret = fs_path_add_path(p, name);
3433 if (ret < 0)
3434 goto out;
3435
3436 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3437
3438 out:
3439 if (ret)
3440 fs_path_free(p);
3441 return ret;
3442 }
3443
3444 static int record_new_ref(struct send_ctx *sctx)
3445 {
3446 int ret;
3447
3448 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3449 sctx->cmp_key, 0, __record_new_ref, sctx);
3450 if (ret < 0)
3451 goto out;
3452 ret = 0;
3453
3454 out:
3455 return ret;
3456 }
3457
3458 static int record_deleted_ref(struct send_ctx *sctx)
3459 {
3460 int ret;
3461
3462 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3463 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3464 if (ret < 0)
3465 goto out;
3466 ret = 0;
3467
3468 out:
3469 return ret;
3470 }
3471
3472 struct find_ref_ctx {
3473 u64 dir;
3474 u64 dir_gen;
3475 struct btrfs_root *root;
3476 struct fs_path *name;
3477 int found_idx;
3478 };
3479
3480 static int __find_iref(int num, u64 dir, int index,
3481 struct fs_path *name,
3482 void *ctx_)
3483 {
3484 struct find_ref_ctx *ctx = ctx_;
3485 u64 dir_gen;
3486 int ret;
3487
3488 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3489 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3490 /*
3491 * To avoid doing extra lookups we'll only do this if everything
3492 * else matches.
3493 */
3494 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3495 NULL, NULL, NULL);
3496 if (ret)
3497 return ret;
3498 if (dir_gen != ctx->dir_gen)
3499 return 0;
3500 ctx->found_idx = num;
3501 return 1;
3502 }
3503 return 0;
3504 }
3505
3506 static int find_iref(struct btrfs_root *root,
3507 struct btrfs_path *path,
3508 struct btrfs_key *key,
3509 u64 dir, u64 dir_gen, struct fs_path *name)
3510 {
3511 int ret;
3512 struct find_ref_ctx ctx;
3513
3514 ctx.dir = dir;
3515 ctx.name = name;
3516 ctx.dir_gen = dir_gen;
3517 ctx.found_idx = -1;
3518 ctx.root = root;
3519
3520 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3521 if (ret < 0)
3522 return ret;
3523
3524 if (ctx.found_idx == -1)
3525 return -ENOENT;
3526
3527 return ctx.found_idx;
3528 }
3529
3530 static int __record_changed_new_ref(int num, u64 dir, int index,
3531 struct fs_path *name,
3532 void *ctx)
3533 {
3534 u64 dir_gen;
3535 int ret;
3536 struct send_ctx *sctx = ctx;
3537
3538 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3539 NULL, NULL, NULL);
3540 if (ret)
3541 return ret;
3542
3543 ret = find_iref(sctx->parent_root, sctx->right_path,
3544 sctx->cmp_key, dir, dir_gen, name);
3545 if (ret == -ENOENT)
3546 ret = __record_new_ref(num, dir, index, name, sctx);
3547 else if (ret > 0)
3548 ret = 0;
3549
3550 return ret;
3551 }
3552
3553 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3554 struct fs_path *name,
3555 void *ctx)
3556 {
3557 u64 dir_gen;
3558 int ret;
3559 struct send_ctx *sctx = ctx;
3560
3561 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3562 NULL, NULL, NULL);
3563 if (ret)
3564 return ret;
3565
3566 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
3567 dir, dir_gen, name);
3568 if (ret == -ENOENT)
3569 ret = __record_deleted_ref(num, dir, index, name, sctx);
3570 else if (ret > 0)
3571 ret = 0;
3572
3573 return ret;
3574 }
3575
3576 static int record_changed_ref(struct send_ctx *sctx)
3577 {
3578 int ret = 0;
3579
3580 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3581 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3582 if (ret < 0)
3583 goto out;
3584 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3585 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3586 if (ret < 0)
3587 goto out;
3588 ret = 0;
3589
3590 out:
3591 return ret;
3592 }
3593
3594 /*
3595 * Record and process all refs at once. Needed when an inode changes the
3596 * generation number, which means that it was deleted and recreated.
3597 */
3598 static int process_all_refs(struct send_ctx *sctx,
3599 enum btrfs_compare_tree_result cmd)
3600 {
3601 int ret;
3602 struct btrfs_root *root;
3603 struct btrfs_path *path;
3604 struct btrfs_key key;
3605 struct btrfs_key found_key;
3606 struct extent_buffer *eb;
3607 int slot;
3608 iterate_inode_ref_t cb;
3609 int pending_move = 0;
3610
3611 path = alloc_path_for_send();
3612 if (!path)
3613 return -ENOMEM;
3614
3615 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3616 root = sctx->send_root;
3617 cb = __record_new_ref;
3618 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3619 root = sctx->parent_root;
3620 cb = __record_deleted_ref;
3621 } else {
3622 BUG();
3623 }
3624
3625 key.objectid = sctx->cmp_key->objectid;
3626 key.type = BTRFS_INODE_REF_KEY;
3627 key.offset = 0;
3628 while (1) {
3629 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3630 if (ret < 0)
3631 goto out;
3632 if (ret)
3633 break;
3634
3635 eb = path->nodes[0];
3636 slot = path->slots[0];
3637 btrfs_item_key_to_cpu(eb, &found_key, slot);
3638
3639 if (found_key.objectid != key.objectid ||
3640 (found_key.type != BTRFS_INODE_REF_KEY &&
3641 found_key.type != BTRFS_INODE_EXTREF_KEY))
3642 break;
3643
3644 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
3645 btrfs_release_path(path);
3646 if (ret < 0)
3647 goto out;
3648
3649 key.offset = found_key.offset + 1;
3650 }
3651 btrfs_release_path(path);
3652
3653 ret = process_recorded_refs(sctx, &pending_move);
3654 /* Only applicable to an incremental send. */
3655 ASSERT(pending_move == 0);
3656
3657 out:
3658 btrfs_free_path(path);
3659 return ret;
3660 }
3661
3662 static int send_set_xattr(struct send_ctx *sctx,
3663 struct fs_path *path,
3664 const char *name, int name_len,
3665 const char *data, int data_len)
3666 {
3667 int ret = 0;
3668
3669 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3670 if (ret < 0)
3671 goto out;
3672
3673 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3674 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3675 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3676
3677 ret = send_cmd(sctx);
3678
3679 tlv_put_failure:
3680 out:
3681 return ret;
3682 }
3683
3684 static int send_remove_xattr(struct send_ctx *sctx,
3685 struct fs_path *path,
3686 const char *name, int name_len)
3687 {
3688 int ret = 0;
3689
3690 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3691 if (ret < 0)
3692 goto out;
3693
3694 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3695 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3696
3697 ret = send_cmd(sctx);
3698
3699 tlv_put_failure:
3700 out:
3701 return ret;
3702 }
3703
3704 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3705 const char *name, int name_len,
3706 const char *data, int data_len,
3707 u8 type, void *ctx)
3708 {
3709 int ret;
3710 struct send_ctx *sctx = ctx;
3711 struct fs_path *p;
3712 posix_acl_xattr_header dummy_acl;
3713
3714 p = fs_path_alloc();
3715 if (!p)
3716 return -ENOMEM;
3717
3718 /*
3719 * This hack is needed because empty acl's are stored as zero byte
3720 * data in xattrs. Problem with that is, that receiving these zero byte
3721 * acl's will fail later. To fix this, we send a dummy acl list that
3722 * only contains the version number and no entries.
3723 */
3724 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3725 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3726 if (data_len == 0) {
3727 dummy_acl.a_version =
3728 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3729 data = (char *)&dummy_acl;
3730 data_len = sizeof(dummy_acl);
3731 }
3732 }
3733
3734 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3735 if (ret < 0)
3736 goto out;
3737
3738 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3739
3740 out:
3741 fs_path_free(p);
3742 return ret;
3743 }
3744
3745 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3746 const char *name, int name_len,
3747 const char *data, int data_len,
3748 u8 type, void *ctx)
3749 {
3750 int ret;
3751 struct send_ctx *sctx = ctx;
3752 struct fs_path *p;
3753
3754 p = fs_path_alloc();
3755 if (!p)
3756 return -ENOMEM;
3757
3758 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3759 if (ret < 0)
3760 goto out;
3761
3762 ret = send_remove_xattr(sctx, p, name, name_len);
3763
3764 out:
3765 fs_path_free(p);
3766 return ret;
3767 }
3768
3769 static int process_new_xattr(struct send_ctx *sctx)
3770 {
3771 int ret = 0;
3772
3773 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3774 sctx->cmp_key, __process_new_xattr, sctx);
3775
3776 return ret;
3777 }
3778
3779 static int process_deleted_xattr(struct send_ctx *sctx)
3780 {
3781 int ret;
3782
3783 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3784 sctx->cmp_key, __process_deleted_xattr, sctx);
3785
3786 return ret;
3787 }
3788
3789 struct find_xattr_ctx {
3790 const char *name;
3791 int name_len;
3792 int found_idx;
3793 char *found_data;
3794 int found_data_len;
3795 };
3796
3797 static int __find_xattr(int num, struct btrfs_key *di_key,
3798 const char *name, int name_len,
3799 const char *data, int data_len,
3800 u8 type, void *vctx)
3801 {
3802 struct find_xattr_ctx *ctx = vctx;
3803
3804 if (name_len == ctx->name_len &&
3805 strncmp(name, ctx->name, name_len) == 0) {
3806 ctx->found_idx = num;
3807 ctx->found_data_len = data_len;
3808 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
3809 if (!ctx->found_data)
3810 return -ENOMEM;
3811 return 1;
3812 }
3813 return 0;
3814 }
3815
3816 static int find_xattr(struct btrfs_root *root,
3817 struct btrfs_path *path,
3818 struct btrfs_key *key,
3819 const char *name, int name_len,
3820 char **data, int *data_len)
3821 {
3822 int ret;
3823 struct find_xattr_ctx ctx;
3824
3825 ctx.name = name;
3826 ctx.name_len = name_len;
3827 ctx.found_idx = -1;
3828 ctx.found_data = NULL;
3829 ctx.found_data_len = 0;
3830
3831 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
3832 if (ret < 0)
3833 return ret;
3834
3835 if (ctx.found_idx == -1)
3836 return -ENOENT;
3837 if (data) {
3838 *data = ctx.found_data;
3839 *data_len = ctx.found_data_len;
3840 } else {
3841 kfree(ctx.found_data);
3842 }
3843 return ctx.found_idx;
3844 }
3845
3846
3847 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3848 const char *name, int name_len,
3849 const char *data, int data_len,
3850 u8 type, void *ctx)
3851 {
3852 int ret;
3853 struct send_ctx *sctx = ctx;
3854 char *found_data = NULL;
3855 int found_data_len = 0;
3856
3857 ret = find_xattr(sctx->parent_root, sctx->right_path,
3858 sctx->cmp_key, name, name_len, &found_data,
3859 &found_data_len);
3860 if (ret == -ENOENT) {
3861 ret = __process_new_xattr(num, di_key, name, name_len, data,
3862 data_len, type, ctx);
3863 } else if (ret >= 0) {
3864 if (data_len != found_data_len ||
3865 memcmp(data, found_data, data_len)) {
3866 ret = __process_new_xattr(num, di_key, name, name_len,
3867 data, data_len, type, ctx);
3868 } else {
3869 ret = 0;
3870 }
3871 }
3872
3873 kfree(found_data);
3874 return ret;
3875 }
3876
3877 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3878 const char *name, int name_len,
3879 const char *data, int data_len,
3880 u8 type, void *ctx)
3881 {
3882 int ret;
3883 struct send_ctx *sctx = ctx;
3884
3885 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3886 name, name_len, NULL, NULL);
3887 if (ret == -ENOENT)
3888 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3889 data_len, type, ctx);
3890 else if (ret >= 0)
3891 ret = 0;
3892
3893 return ret;
3894 }
3895
3896 static int process_changed_xattr(struct send_ctx *sctx)
3897 {
3898 int ret = 0;
3899
3900 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3901 sctx->cmp_key, __process_changed_new_xattr, sctx);
3902 if (ret < 0)
3903 goto out;
3904 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3905 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3906
3907 out:
3908 return ret;
3909 }
3910
3911 static int process_all_new_xattrs(struct send_ctx *sctx)
3912 {
3913 int ret;
3914 struct btrfs_root *root;
3915 struct btrfs_path *path;
3916 struct btrfs_key key;
3917 struct btrfs_key found_key;
3918 struct extent_buffer *eb;
3919 int slot;
3920
3921 path = alloc_path_for_send();
3922 if (!path)
3923 return -ENOMEM;
3924
3925 root = sctx->send_root;
3926
3927 key.objectid = sctx->cmp_key->objectid;
3928 key.type = BTRFS_XATTR_ITEM_KEY;
3929 key.offset = 0;
3930 while (1) {
3931 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3932 if (ret < 0)
3933 goto out;
3934 if (ret) {
3935 ret = 0;
3936 goto out;
3937 }
3938
3939 eb = path->nodes[0];
3940 slot = path->slots[0];
3941 btrfs_item_key_to_cpu(eb, &found_key, slot);
3942
3943 if (found_key.objectid != key.objectid ||
3944 found_key.type != key.type) {
3945 ret = 0;
3946 goto out;
3947 }
3948
3949 ret = iterate_dir_item(root, path, &found_key,
3950 __process_new_xattr, sctx);
3951 if (ret < 0)
3952 goto out;
3953
3954 btrfs_release_path(path);
3955 key.offset = found_key.offset + 1;
3956 }
3957
3958 out:
3959 btrfs_free_path(path);
3960 return ret;
3961 }
3962
3963 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3964 {
3965 struct btrfs_root *root = sctx->send_root;
3966 struct btrfs_fs_info *fs_info = root->fs_info;
3967 struct inode *inode;
3968 struct page *page;
3969 char *addr;
3970 struct btrfs_key key;
3971 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3972 pgoff_t last_index;
3973 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3974 ssize_t ret = 0;
3975
3976 key.objectid = sctx->cur_ino;
3977 key.type = BTRFS_INODE_ITEM_KEY;
3978 key.offset = 0;
3979
3980 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3981 if (IS_ERR(inode))
3982 return PTR_ERR(inode);
3983
3984 if (offset + len > i_size_read(inode)) {
3985 if (offset > i_size_read(inode))
3986 len = 0;
3987 else
3988 len = offset - i_size_read(inode);
3989 }
3990 if (len == 0)
3991 goto out;
3992
3993 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
3994 while (index <= last_index) {
3995 unsigned cur_len = min_t(unsigned, len,
3996 PAGE_CACHE_SIZE - pg_offset);
3997 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3998 if (!page) {
3999 ret = -ENOMEM;
4000 break;
4001 }
4002
4003 if (!PageUptodate(page)) {
4004 btrfs_readpage(NULL, page);
4005 lock_page(page);
4006 if (!PageUptodate(page)) {
4007 unlock_page(page);
4008 page_cache_release(page);
4009 ret = -EIO;
4010 break;
4011 }
4012 }
4013
4014 addr = kmap(page);
4015 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4016 kunmap(page);
4017 unlock_page(page);
4018 page_cache_release(page);
4019 index++;
4020 pg_offset = 0;
4021 len -= cur_len;
4022 ret += cur_len;
4023 }
4024 out:
4025 iput(inode);
4026 return ret;
4027 }
4028
4029 /*
4030 * Read some bytes from the current inode/file and send a write command to
4031 * user space.
4032 */
4033 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4034 {
4035 int ret = 0;
4036 struct fs_path *p;
4037 ssize_t num_read = 0;
4038
4039 p = fs_path_alloc();
4040 if (!p)
4041 return -ENOMEM;
4042
4043 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4044
4045 num_read = fill_read_buf(sctx, offset, len);
4046 if (num_read <= 0) {
4047 if (num_read < 0)
4048 ret = num_read;
4049 goto out;
4050 }
4051
4052 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4053 if (ret < 0)
4054 goto out;
4055
4056 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4057 if (ret < 0)
4058 goto out;
4059
4060 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4061 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4062 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4063
4064 ret = send_cmd(sctx);
4065
4066 tlv_put_failure:
4067 out:
4068 fs_path_free(p);
4069 if (ret < 0)
4070 return ret;
4071 return num_read;
4072 }
4073
4074 /*
4075 * Send a clone command to user space.
4076 */
4077 static int send_clone(struct send_ctx *sctx,
4078 u64 offset, u32 len,
4079 struct clone_root *clone_root)
4080 {
4081 int ret = 0;
4082 struct fs_path *p;
4083 u64 gen;
4084
4085 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4086 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4087 clone_root->root->objectid, clone_root->ino,
4088 clone_root->offset);
4089
4090 p = fs_path_alloc();
4091 if (!p)
4092 return -ENOMEM;
4093
4094 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4095 if (ret < 0)
4096 goto out;
4097
4098 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4099 if (ret < 0)
4100 goto out;
4101
4102 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4103 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4104 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4105
4106 if (clone_root->root == sctx->send_root) {
4107 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4108 &gen, NULL, NULL, NULL, NULL);
4109 if (ret < 0)
4110 goto out;
4111 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4112 } else {
4113 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4114 }
4115 if (ret < 0)
4116 goto out;
4117
4118 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4119 clone_root->root->root_item.uuid);
4120 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4121 le64_to_cpu(clone_root->root->root_item.ctransid));
4122 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4123 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4124 clone_root->offset);
4125
4126 ret = send_cmd(sctx);
4127
4128 tlv_put_failure:
4129 out:
4130 fs_path_free(p);
4131 return ret;
4132 }
4133
4134 /*
4135 * Send an update extent command to user space.
4136 */
4137 static int send_update_extent(struct send_ctx *sctx,
4138 u64 offset, u32 len)
4139 {
4140 int ret = 0;
4141 struct fs_path *p;
4142
4143 p = fs_path_alloc();
4144 if (!p)
4145 return -ENOMEM;
4146
4147 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4148 if (ret < 0)
4149 goto out;
4150
4151 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4152 if (ret < 0)
4153 goto out;
4154
4155 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4156 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4157 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4158
4159 ret = send_cmd(sctx);
4160
4161 tlv_put_failure:
4162 out:
4163 fs_path_free(p);
4164 return ret;
4165 }
4166
4167 static int send_hole(struct send_ctx *sctx, u64 end)
4168 {
4169 struct fs_path *p = NULL;
4170 u64 offset = sctx->cur_inode_last_extent;
4171 u64 len;
4172 int ret = 0;
4173
4174 p = fs_path_alloc();
4175 if (!p)
4176 return -ENOMEM;
4177 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4178 while (offset < end) {
4179 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4180
4181 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4182 if (ret < 0)
4183 break;
4184 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4185 if (ret < 0)
4186 break;
4187 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4188 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4189 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4190 ret = send_cmd(sctx);
4191 if (ret < 0)
4192 break;
4193 offset += len;
4194 }
4195 tlv_put_failure:
4196 fs_path_free(p);
4197 return ret;
4198 }
4199
4200 static int send_write_or_clone(struct send_ctx *sctx,
4201 struct btrfs_path *path,
4202 struct btrfs_key *key,
4203 struct clone_root *clone_root)
4204 {
4205 int ret = 0;
4206 struct btrfs_file_extent_item *ei;
4207 u64 offset = key->offset;
4208 u64 pos = 0;
4209 u64 len;
4210 u32 l;
4211 u8 type;
4212 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4213
4214 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4215 struct btrfs_file_extent_item);
4216 type = btrfs_file_extent_type(path->nodes[0], ei);
4217 if (type == BTRFS_FILE_EXTENT_INLINE) {
4218 len = btrfs_file_extent_inline_len(path->nodes[0],
4219 path->slots[0], ei);
4220 /*
4221 * it is possible the inline item won't cover the whole page,
4222 * but there may be items after this page. Make
4223 * sure to send the whole thing
4224 */
4225 len = PAGE_CACHE_ALIGN(len);
4226 } else {
4227 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4228 }
4229
4230 if (offset + len > sctx->cur_inode_size)
4231 len = sctx->cur_inode_size - offset;
4232 if (len == 0) {
4233 ret = 0;
4234 goto out;
4235 }
4236
4237 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4238 ret = send_clone(sctx, offset, len, clone_root);
4239 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4240 ret = send_update_extent(sctx, offset, len);
4241 } else {
4242 while (pos < len) {
4243 l = len - pos;
4244 if (l > BTRFS_SEND_READ_SIZE)
4245 l = BTRFS_SEND_READ_SIZE;
4246 ret = send_write(sctx, pos + offset, l);
4247 if (ret < 0)
4248 goto out;
4249 if (!ret)
4250 break;
4251 pos += ret;
4252 }
4253 ret = 0;
4254 }
4255 out:
4256 return ret;
4257 }
4258
4259 static int is_extent_unchanged(struct send_ctx *sctx,
4260 struct btrfs_path *left_path,
4261 struct btrfs_key *ekey)
4262 {
4263 int ret = 0;
4264 struct btrfs_key key;
4265 struct btrfs_path *path = NULL;
4266 struct extent_buffer *eb;
4267 int slot;
4268 struct btrfs_key found_key;
4269 struct btrfs_file_extent_item *ei;
4270 u64 left_disknr;
4271 u64 right_disknr;
4272 u64 left_offset;
4273 u64 right_offset;
4274 u64 left_offset_fixed;
4275 u64 left_len;
4276 u64 right_len;
4277 u64 left_gen;
4278 u64 right_gen;
4279 u8 left_type;
4280 u8 right_type;
4281
4282 path = alloc_path_for_send();
4283 if (!path)
4284 return -ENOMEM;
4285
4286 eb = left_path->nodes[0];
4287 slot = left_path->slots[0];
4288 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4289 left_type = btrfs_file_extent_type(eb, ei);
4290
4291 if (left_type != BTRFS_FILE_EXTENT_REG) {
4292 ret = 0;
4293 goto out;
4294 }
4295 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4296 left_len = btrfs_file_extent_num_bytes(eb, ei);
4297 left_offset = btrfs_file_extent_offset(eb, ei);
4298 left_gen = btrfs_file_extent_generation(eb, ei);
4299
4300 /*
4301 * Following comments will refer to these graphics. L is the left
4302 * extents which we are checking at the moment. 1-8 are the right
4303 * extents that we iterate.
4304 *
4305 * |-----L-----|
4306 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4307 *
4308 * |-----L-----|
4309 * |--1--|-2b-|...(same as above)
4310 *
4311 * Alternative situation. Happens on files where extents got split.
4312 * |-----L-----|
4313 * |-----------7-----------|-6-|
4314 *
4315 * Alternative situation. Happens on files which got larger.
4316 * |-----L-----|
4317 * |-8-|
4318 * Nothing follows after 8.
4319 */
4320
4321 key.objectid = ekey->objectid;
4322 key.type = BTRFS_EXTENT_DATA_KEY;
4323 key.offset = ekey->offset;
4324 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4325 if (ret < 0)
4326 goto out;
4327 if (ret) {
4328 ret = 0;
4329 goto out;
4330 }
4331
4332 /*
4333 * Handle special case where the right side has no extents at all.
4334 */
4335 eb = path->nodes[0];
4336 slot = path->slots[0];
4337 btrfs_item_key_to_cpu(eb, &found_key, slot);
4338 if (found_key.objectid != key.objectid ||
4339 found_key.type != key.type) {
4340 /* If we're a hole then just pretend nothing changed */
4341 ret = (left_disknr) ? 0 : 1;
4342 goto out;
4343 }
4344
4345 /*
4346 * We're now on 2a, 2b or 7.
4347 */
4348 key = found_key;
4349 while (key.offset < ekey->offset + left_len) {
4350 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4351 right_type = btrfs_file_extent_type(eb, ei);
4352 if (right_type != BTRFS_FILE_EXTENT_REG) {
4353 ret = 0;
4354 goto out;
4355 }
4356
4357 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4358 right_len = btrfs_file_extent_num_bytes(eb, ei);
4359 right_offset = btrfs_file_extent_offset(eb, ei);
4360 right_gen = btrfs_file_extent_generation(eb, ei);
4361
4362 /*
4363 * Are we at extent 8? If yes, we know the extent is changed.
4364 * This may only happen on the first iteration.
4365 */
4366 if (found_key.offset + right_len <= ekey->offset) {
4367 /* If we're a hole just pretend nothing changed */
4368 ret = (left_disknr) ? 0 : 1;
4369 goto out;
4370 }
4371
4372 left_offset_fixed = left_offset;
4373 if (key.offset < ekey->offset) {
4374 /* Fix the right offset for 2a and 7. */
4375 right_offset += ekey->offset - key.offset;
4376 } else {
4377 /* Fix the left offset for all behind 2a and 2b */
4378 left_offset_fixed += key.offset - ekey->offset;
4379 }
4380
4381 /*
4382 * Check if we have the same extent.
4383 */
4384 if (left_disknr != right_disknr ||
4385 left_offset_fixed != right_offset ||
4386 left_gen != right_gen) {
4387 ret = 0;
4388 goto out;
4389 }
4390
4391 /*
4392 * Go to the next extent.
4393 */
4394 ret = btrfs_next_item(sctx->parent_root, path);
4395 if (ret < 0)
4396 goto out;
4397 if (!ret) {
4398 eb = path->nodes[0];
4399 slot = path->slots[0];
4400 btrfs_item_key_to_cpu(eb, &found_key, slot);
4401 }
4402 if (ret || found_key.objectid != key.objectid ||
4403 found_key.type != key.type) {
4404 key.offset += right_len;
4405 break;
4406 }
4407 if (found_key.offset != key.offset + right_len) {
4408 ret = 0;
4409 goto out;
4410 }
4411 key = found_key;
4412 }
4413
4414 /*
4415 * We're now behind the left extent (treat as unchanged) or at the end
4416 * of the right side (treat as changed).
4417 */
4418 if (key.offset >= ekey->offset + left_len)
4419 ret = 1;
4420 else
4421 ret = 0;
4422
4423
4424 out:
4425 btrfs_free_path(path);
4426 return ret;
4427 }
4428
4429 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4430 {
4431 struct btrfs_path *path;
4432 struct btrfs_root *root = sctx->send_root;
4433 struct btrfs_file_extent_item *fi;
4434 struct btrfs_key key;
4435 u64 extent_end;
4436 u8 type;
4437 int ret;
4438
4439 path = alloc_path_for_send();
4440 if (!path)
4441 return -ENOMEM;
4442
4443 sctx->cur_inode_last_extent = 0;
4444
4445 key.objectid = sctx->cur_ino;
4446 key.type = BTRFS_EXTENT_DATA_KEY;
4447 key.offset = offset;
4448 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4449 if (ret < 0)
4450 goto out;
4451 ret = 0;
4452 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4453 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4454 goto out;
4455
4456 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4457 struct btrfs_file_extent_item);
4458 type = btrfs_file_extent_type(path->nodes[0], fi);
4459 if (type == BTRFS_FILE_EXTENT_INLINE) {
4460 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4461 path->slots[0], fi);
4462 extent_end = ALIGN(key.offset + size,
4463 sctx->send_root->sectorsize);
4464 } else {
4465 extent_end = key.offset +
4466 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4467 }
4468 sctx->cur_inode_last_extent = extent_end;
4469 out:
4470 btrfs_free_path(path);
4471 return ret;
4472 }
4473
4474 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4475 struct btrfs_key *key)
4476 {
4477 struct btrfs_file_extent_item *fi;
4478 u64 extent_end;
4479 u8 type;
4480 int ret = 0;
4481
4482 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4483 return 0;
4484
4485 if (sctx->cur_inode_last_extent == (u64)-1) {
4486 ret = get_last_extent(sctx, key->offset - 1);
4487 if (ret)
4488 return ret;
4489 }
4490
4491 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4492 struct btrfs_file_extent_item);
4493 type = btrfs_file_extent_type(path->nodes[0], fi);
4494 if (type == BTRFS_FILE_EXTENT_INLINE) {
4495 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4496 path->slots[0], fi);
4497 extent_end = ALIGN(key->offset + size,
4498 sctx->send_root->sectorsize);
4499 } else {
4500 extent_end = key->offset +
4501 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4502 }
4503
4504 if (path->slots[0] == 0 &&
4505 sctx->cur_inode_last_extent < key->offset) {
4506 /*
4507 * We might have skipped entire leafs that contained only
4508 * file extent items for our current inode. These leafs have
4509 * a generation number smaller (older) than the one in the
4510 * current leaf and the leaf our last extent came from, and
4511 * are located between these 2 leafs.
4512 */
4513 ret = get_last_extent(sctx, key->offset - 1);
4514 if (ret)
4515 return ret;
4516 }
4517
4518 if (sctx->cur_inode_last_extent < key->offset)
4519 ret = send_hole(sctx, key->offset);
4520 sctx->cur_inode_last_extent = extent_end;
4521 return ret;
4522 }
4523
4524 static int process_extent(struct send_ctx *sctx,
4525 struct btrfs_path *path,
4526 struct btrfs_key *key)
4527 {
4528 struct clone_root *found_clone = NULL;
4529 int ret = 0;
4530
4531 if (S_ISLNK(sctx->cur_inode_mode))
4532 return 0;
4533
4534 if (sctx->parent_root && !sctx->cur_inode_new) {
4535 ret = is_extent_unchanged(sctx, path, key);
4536 if (ret < 0)
4537 goto out;
4538 if (ret) {
4539 ret = 0;
4540 goto out_hole;
4541 }
4542 } else {
4543 struct btrfs_file_extent_item *ei;
4544 u8 type;
4545
4546 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4547 struct btrfs_file_extent_item);
4548 type = btrfs_file_extent_type(path->nodes[0], ei);
4549 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4550 type == BTRFS_FILE_EXTENT_REG) {
4551 /*
4552 * The send spec does not have a prealloc command yet,
4553 * so just leave a hole for prealloc'ed extents until
4554 * we have enough commands queued up to justify rev'ing
4555 * the send spec.
4556 */
4557 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4558 ret = 0;
4559 goto out;
4560 }
4561
4562 /* Have a hole, just skip it. */
4563 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4564 ret = 0;
4565 goto out;
4566 }
4567 }
4568 }
4569
4570 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4571 sctx->cur_inode_size, &found_clone);
4572 if (ret != -ENOENT && ret < 0)
4573 goto out;
4574
4575 ret = send_write_or_clone(sctx, path, key, found_clone);
4576 if (ret)
4577 goto out;
4578 out_hole:
4579 ret = maybe_send_hole(sctx, path, key);
4580 out:
4581 return ret;
4582 }
4583
4584 static int process_all_extents(struct send_ctx *sctx)
4585 {
4586 int ret;
4587 struct btrfs_root *root;
4588 struct btrfs_path *path;
4589 struct btrfs_key key;
4590 struct btrfs_key found_key;
4591 struct extent_buffer *eb;
4592 int slot;
4593
4594 root = sctx->send_root;
4595 path = alloc_path_for_send();
4596 if (!path)
4597 return -ENOMEM;
4598
4599 key.objectid = sctx->cmp_key->objectid;
4600 key.type = BTRFS_EXTENT_DATA_KEY;
4601 key.offset = 0;
4602 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4603 if (ret < 0)
4604 goto out;
4605
4606 while (1) {
4607 eb = path->nodes[0];
4608 slot = path->slots[0];
4609
4610 if (slot >= btrfs_header_nritems(eb)) {
4611 ret = btrfs_next_leaf(root, path);
4612 if (ret < 0) {
4613 goto out;
4614 } else if (ret > 0) {
4615 ret = 0;
4616 break;
4617 }
4618 continue;
4619 }
4620
4621 btrfs_item_key_to_cpu(eb, &found_key, slot);
4622
4623 if (found_key.objectid != key.objectid ||
4624 found_key.type != key.type) {
4625 ret = 0;
4626 goto out;
4627 }
4628
4629 ret = process_extent(sctx, path, &found_key);
4630 if (ret < 0)
4631 goto out;
4632
4633 path->slots[0]++;
4634 }
4635
4636 out:
4637 btrfs_free_path(path);
4638 return ret;
4639 }
4640
4641 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4642 int *pending_move,
4643 int *refs_processed)
4644 {
4645 int ret = 0;
4646
4647 if (sctx->cur_ino == 0)
4648 goto out;
4649 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4650 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4651 goto out;
4652 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4653 goto out;
4654
4655 ret = process_recorded_refs(sctx, pending_move);
4656 if (ret < 0)
4657 goto out;
4658
4659 *refs_processed = 1;
4660 out:
4661 return ret;
4662 }
4663
4664 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4665 {
4666 int ret = 0;
4667 u64 left_mode;
4668 u64 left_uid;
4669 u64 left_gid;
4670 u64 right_mode;
4671 u64 right_uid;
4672 u64 right_gid;
4673 int need_chmod = 0;
4674 int need_chown = 0;
4675 int pending_move = 0;
4676 int refs_processed = 0;
4677
4678 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4679 &refs_processed);
4680 if (ret < 0)
4681 goto out;
4682
4683 /*
4684 * We have processed the refs and thus need to advance send_progress.
4685 * Now, calls to get_cur_xxx will take the updated refs of the current
4686 * inode into account.
4687 *
4688 * On the other hand, if our current inode is a directory and couldn't
4689 * be moved/renamed because its parent was renamed/moved too and it has
4690 * a higher inode number, we can only move/rename our current inode
4691 * after we moved/renamed its parent. Therefore in this case operate on
4692 * the old path (pre move/rename) of our current inode, and the
4693 * move/rename will be performed later.
4694 */
4695 if (refs_processed && !pending_move)
4696 sctx->send_progress = sctx->cur_ino + 1;
4697
4698 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4699 goto out;
4700 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4701 goto out;
4702
4703 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4704 &left_mode, &left_uid, &left_gid, NULL);
4705 if (ret < 0)
4706 goto out;
4707
4708 if (!sctx->parent_root || sctx->cur_inode_new) {
4709 need_chown = 1;
4710 if (!S_ISLNK(sctx->cur_inode_mode))
4711 need_chmod = 1;
4712 } else {
4713 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4714 NULL, NULL, &right_mode, &right_uid,
4715 &right_gid, NULL);
4716 if (ret < 0)
4717 goto out;
4718
4719 if (left_uid != right_uid || left_gid != right_gid)
4720 need_chown = 1;
4721 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4722 need_chmod = 1;
4723 }
4724
4725 if (S_ISREG(sctx->cur_inode_mode)) {
4726 if (need_send_hole(sctx)) {
4727 if (sctx->cur_inode_last_extent == (u64)-1) {
4728 ret = get_last_extent(sctx, (u64)-1);
4729 if (ret)
4730 goto out;
4731 }
4732 if (sctx->cur_inode_last_extent <
4733 sctx->cur_inode_size) {
4734 ret = send_hole(sctx, sctx->cur_inode_size);
4735 if (ret)
4736 goto out;
4737 }
4738 }
4739 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4740 sctx->cur_inode_size);
4741 if (ret < 0)
4742 goto out;
4743 }
4744
4745 if (need_chown) {
4746 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4747 left_uid, left_gid);
4748 if (ret < 0)
4749 goto out;
4750 }
4751 if (need_chmod) {
4752 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4753 left_mode);
4754 if (ret < 0)
4755 goto out;
4756 }
4757
4758 /*
4759 * If other directory inodes depended on our current directory
4760 * inode's move/rename, now do their move/rename operations.
4761 */
4762 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
4763 ret = apply_children_dir_moves(sctx);
4764 if (ret)
4765 goto out;
4766 }
4767
4768 /*
4769 * Need to send that every time, no matter if it actually
4770 * changed between the two trees as we have done changes to
4771 * the inode before.
4772 */
4773 sctx->send_progress = sctx->cur_ino + 1;
4774 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4775 if (ret < 0)
4776 goto out;
4777
4778 out:
4779 return ret;
4780 }
4781
4782 static int changed_inode(struct send_ctx *sctx,
4783 enum btrfs_compare_tree_result result)
4784 {
4785 int ret = 0;
4786 struct btrfs_key *key = sctx->cmp_key;
4787 struct btrfs_inode_item *left_ii = NULL;
4788 struct btrfs_inode_item *right_ii = NULL;
4789 u64 left_gen = 0;
4790 u64 right_gen = 0;
4791
4792 sctx->cur_ino = key->objectid;
4793 sctx->cur_inode_new_gen = 0;
4794 sctx->cur_inode_last_extent = (u64)-1;
4795
4796 /*
4797 * Set send_progress to current inode. This will tell all get_cur_xxx
4798 * functions that the current inode's refs are not updated yet. Later,
4799 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4800 */
4801 sctx->send_progress = sctx->cur_ino;
4802
4803 if (result == BTRFS_COMPARE_TREE_NEW ||
4804 result == BTRFS_COMPARE_TREE_CHANGED) {
4805 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4806 sctx->left_path->slots[0],
4807 struct btrfs_inode_item);
4808 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4809 left_ii);
4810 } else {
4811 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4812 sctx->right_path->slots[0],
4813 struct btrfs_inode_item);
4814 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4815 right_ii);
4816 }
4817 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4818 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4819 sctx->right_path->slots[0],
4820 struct btrfs_inode_item);
4821
4822 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4823 right_ii);
4824
4825 /*
4826 * The cur_ino = root dir case is special here. We can't treat
4827 * the inode as deleted+reused because it would generate a
4828 * stream that tries to delete/mkdir the root dir.
4829 */
4830 if (left_gen != right_gen &&
4831 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4832 sctx->cur_inode_new_gen = 1;
4833 }
4834
4835 if (result == BTRFS_COMPARE_TREE_NEW) {
4836 sctx->cur_inode_gen = left_gen;
4837 sctx->cur_inode_new = 1;
4838 sctx->cur_inode_deleted = 0;
4839 sctx->cur_inode_size = btrfs_inode_size(
4840 sctx->left_path->nodes[0], left_ii);
4841 sctx->cur_inode_mode = btrfs_inode_mode(
4842 sctx->left_path->nodes[0], left_ii);
4843 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4844 ret = send_create_inode_if_needed(sctx);
4845 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4846 sctx->cur_inode_gen = right_gen;
4847 sctx->cur_inode_new = 0;
4848 sctx->cur_inode_deleted = 1;
4849 sctx->cur_inode_size = btrfs_inode_size(
4850 sctx->right_path->nodes[0], right_ii);
4851 sctx->cur_inode_mode = btrfs_inode_mode(
4852 sctx->right_path->nodes[0], right_ii);
4853 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4854 /*
4855 * We need to do some special handling in case the inode was
4856 * reported as changed with a changed generation number. This
4857 * means that the original inode was deleted and new inode
4858 * reused the same inum. So we have to treat the old inode as
4859 * deleted and the new one as new.
4860 */
4861 if (sctx->cur_inode_new_gen) {
4862 /*
4863 * First, process the inode as if it was deleted.
4864 */
4865 sctx->cur_inode_gen = right_gen;
4866 sctx->cur_inode_new = 0;
4867 sctx->cur_inode_deleted = 1;
4868 sctx->cur_inode_size = btrfs_inode_size(
4869 sctx->right_path->nodes[0], right_ii);
4870 sctx->cur_inode_mode = btrfs_inode_mode(
4871 sctx->right_path->nodes[0], right_ii);
4872 ret = process_all_refs(sctx,
4873 BTRFS_COMPARE_TREE_DELETED);
4874 if (ret < 0)
4875 goto out;
4876
4877 /*
4878 * Now process the inode as if it was new.
4879 */
4880 sctx->cur_inode_gen = left_gen;
4881 sctx->cur_inode_new = 1;
4882 sctx->cur_inode_deleted = 0;
4883 sctx->cur_inode_size = btrfs_inode_size(
4884 sctx->left_path->nodes[0], left_ii);
4885 sctx->cur_inode_mode = btrfs_inode_mode(
4886 sctx->left_path->nodes[0], left_ii);
4887 ret = send_create_inode_if_needed(sctx);
4888 if (ret < 0)
4889 goto out;
4890
4891 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4892 if (ret < 0)
4893 goto out;
4894 /*
4895 * Advance send_progress now as we did not get into
4896 * process_recorded_refs_if_needed in the new_gen case.
4897 */
4898 sctx->send_progress = sctx->cur_ino + 1;
4899
4900 /*
4901 * Now process all extents and xattrs of the inode as if
4902 * they were all new.
4903 */
4904 ret = process_all_extents(sctx);
4905 if (ret < 0)
4906 goto out;
4907 ret = process_all_new_xattrs(sctx);
4908 if (ret < 0)
4909 goto out;
4910 } else {
4911 sctx->cur_inode_gen = left_gen;
4912 sctx->cur_inode_new = 0;
4913 sctx->cur_inode_new_gen = 0;
4914 sctx->cur_inode_deleted = 0;
4915 sctx->cur_inode_size = btrfs_inode_size(
4916 sctx->left_path->nodes[0], left_ii);
4917 sctx->cur_inode_mode = btrfs_inode_mode(
4918 sctx->left_path->nodes[0], left_ii);
4919 }
4920 }
4921
4922 out:
4923 return ret;
4924 }
4925
4926 /*
4927 * We have to process new refs before deleted refs, but compare_trees gives us
4928 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4929 * first and later process them in process_recorded_refs.
4930 * For the cur_inode_new_gen case, we skip recording completely because
4931 * changed_inode did already initiate processing of refs. The reason for this is
4932 * that in this case, compare_tree actually compares the refs of 2 different
4933 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4934 * refs of the right tree as deleted and all refs of the left tree as new.
4935 */
4936 static int changed_ref(struct send_ctx *sctx,
4937 enum btrfs_compare_tree_result result)
4938 {
4939 int ret = 0;
4940
4941 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4942
4943 if (!sctx->cur_inode_new_gen &&
4944 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4945 if (result == BTRFS_COMPARE_TREE_NEW)
4946 ret = record_new_ref(sctx);
4947 else if (result == BTRFS_COMPARE_TREE_DELETED)
4948 ret = record_deleted_ref(sctx);
4949 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4950 ret = record_changed_ref(sctx);
4951 }
4952
4953 return ret;
4954 }
4955
4956 /*
4957 * Process new/deleted/changed xattrs. We skip processing in the
4958 * cur_inode_new_gen case because changed_inode did already initiate processing
4959 * of xattrs. The reason is the same as in changed_ref
4960 */
4961 static int changed_xattr(struct send_ctx *sctx,
4962 enum btrfs_compare_tree_result result)
4963 {
4964 int ret = 0;
4965
4966 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4967
4968 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4969 if (result == BTRFS_COMPARE_TREE_NEW)
4970 ret = process_new_xattr(sctx);
4971 else if (result == BTRFS_COMPARE_TREE_DELETED)
4972 ret = process_deleted_xattr(sctx);
4973 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4974 ret = process_changed_xattr(sctx);
4975 }
4976
4977 return ret;
4978 }
4979
4980 /*
4981 * Process new/deleted/changed extents. We skip processing in the
4982 * cur_inode_new_gen case because changed_inode did already initiate processing
4983 * of extents. The reason is the same as in changed_ref
4984 */
4985 static int changed_extent(struct send_ctx *sctx,
4986 enum btrfs_compare_tree_result result)
4987 {
4988 int ret = 0;
4989
4990 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4991
4992 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4993 if (result != BTRFS_COMPARE_TREE_DELETED)
4994 ret = process_extent(sctx, sctx->left_path,
4995 sctx->cmp_key);
4996 }
4997
4998 return ret;
4999 }
5000
5001 static int dir_changed(struct send_ctx *sctx, u64 dir)
5002 {
5003 u64 orig_gen, new_gen;
5004 int ret;
5005
5006 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5007 NULL, NULL);
5008 if (ret)
5009 return ret;
5010
5011 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5012 NULL, NULL, NULL);
5013 if (ret)
5014 return ret;
5015
5016 return (orig_gen != new_gen) ? 1 : 0;
5017 }
5018
5019 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5020 struct btrfs_key *key)
5021 {
5022 struct btrfs_inode_extref *extref;
5023 struct extent_buffer *leaf;
5024 u64 dirid = 0, last_dirid = 0;
5025 unsigned long ptr;
5026 u32 item_size;
5027 u32 cur_offset = 0;
5028 int ref_name_len;
5029 int ret = 0;
5030
5031 /* Easy case, just check this one dirid */
5032 if (key->type == BTRFS_INODE_REF_KEY) {
5033 dirid = key->offset;
5034
5035 ret = dir_changed(sctx, dirid);
5036 goto out;
5037 }
5038
5039 leaf = path->nodes[0];
5040 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5041 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5042 while (cur_offset < item_size) {
5043 extref = (struct btrfs_inode_extref *)(ptr +
5044 cur_offset);
5045 dirid = btrfs_inode_extref_parent(leaf, extref);
5046 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5047 cur_offset += ref_name_len + sizeof(*extref);
5048 if (dirid == last_dirid)
5049 continue;
5050 ret = dir_changed(sctx, dirid);
5051 if (ret)
5052 break;
5053 last_dirid = dirid;
5054 }
5055 out:
5056 return ret;
5057 }
5058
5059 /*
5060 * Updates compare related fields in sctx and simply forwards to the actual
5061 * changed_xxx functions.
5062 */
5063 static int changed_cb(struct btrfs_root *left_root,
5064 struct btrfs_root *right_root,
5065 struct btrfs_path *left_path,
5066 struct btrfs_path *right_path,
5067 struct btrfs_key *key,
5068 enum btrfs_compare_tree_result result,
5069 void *ctx)
5070 {
5071 int ret = 0;
5072 struct send_ctx *sctx = ctx;
5073
5074 if (result == BTRFS_COMPARE_TREE_SAME) {
5075 if (key->type == BTRFS_INODE_REF_KEY ||
5076 key->type == BTRFS_INODE_EXTREF_KEY) {
5077 ret = compare_refs(sctx, left_path, key);
5078 if (!ret)
5079 return 0;
5080 if (ret < 0)
5081 return ret;
5082 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5083 return maybe_send_hole(sctx, left_path, key);
5084 } else {
5085 return 0;
5086 }
5087 result = BTRFS_COMPARE_TREE_CHANGED;
5088 ret = 0;
5089 }
5090
5091 sctx->left_path = left_path;
5092 sctx->right_path = right_path;
5093 sctx->cmp_key = key;
5094
5095 ret = finish_inode_if_needed(sctx, 0);
5096 if (ret < 0)
5097 goto out;
5098
5099 /* Ignore non-FS objects */
5100 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5101 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5102 goto out;
5103
5104 if (key->type == BTRFS_INODE_ITEM_KEY)
5105 ret = changed_inode(sctx, result);
5106 else if (key->type == BTRFS_INODE_REF_KEY ||
5107 key->type == BTRFS_INODE_EXTREF_KEY)
5108 ret = changed_ref(sctx, result);
5109 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5110 ret = changed_xattr(sctx, result);
5111 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5112 ret = changed_extent(sctx, result);
5113
5114 out:
5115 return ret;
5116 }
5117
5118 static int full_send_tree(struct send_ctx *sctx)
5119 {
5120 int ret;
5121 struct btrfs_root *send_root = sctx->send_root;
5122 struct btrfs_key key;
5123 struct btrfs_key found_key;
5124 struct btrfs_path *path;
5125 struct extent_buffer *eb;
5126 int slot;
5127 u64 start_ctransid;
5128 u64 ctransid;
5129
5130 path = alloc_path_for_send();
5131 if (!path)
5132 return -ENOMEM;
5133
5134 spin_lock(&send_root->root_item_lock);
5135 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5136 spin_unlock(&send_root->root_item_lock);
5137
5138 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5139 key.type = BTRFS_INODE_ITEM_KEY;
5140 key.offset = 0;
5141
5142 /*
5143 * Make sure the tree has not changed after re-joining. We detect this
5144 * by comparing start_ctransid and ctransid. They should always match.
5145 */
5146 spin_lock(&send_root->root_item_lock);
5147 ctransid = btrfs_root_ctransid(&send_root->root_item);
5148 spin_unlock(&send_root->root_item_lock);
5149
5150 if (ctransid != start_ctransid) {
5151 WARN(1, KERN_WARNING "BTRFS: the root that you're trying to "
5152 "send was modified in between. This is "
5153 "probably a bug.\n");
5154 ret = -EIO;
5155 goto out;
5156 }
5157
5158 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5159 if (ret < 0)
5160 goto out;
5161 if (ret)
5162 goto out_finish;
5163
5164 while (1) {
5165 eb = path->nodes[0];
5166 slot = path->slots[0];
5167 btrfs_item_key_to_cpu(eb, &found_key, slot);
5168
5169 ret = changed_cb(send_root, NULL, path, NULL,
5170 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5171 if (ret < 0)
5172 goto out;
5173
5174 key.objectid = found_key.objectid;
5175 key.type = found_key.type;
5176 key.offset = found_key.offset + 1;
5177
5178 ret = btrfs_next_item(send_root, path);
5179 if (ret < 0)
5180 goto out;
5181 if (ret) {
5182 ret = 0;
5183 break;
5184 }
5185 }
5186
5187 out_finish:
5188 ret = finish_inode_if_needed(sctx, 1);
5189
5190 out:
5191 btrfs_free_path(path);
5192 return ret;
5193 }
5194
5195 static int send_subvol(struct send_ctx *sctx)
5196 {
5197 int ret;
5198
5199 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5200 ret = send_header(sctx);
5201 if (ret < 0)
5202 goto out;
5203 }
5204
5205 ret = send_subvol_begin(sctx);
5206 if (ret < 0)
5207 goto out;
5208
5209 if (sctx->parent_root) {
5210 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5211 changed_cb, sctx);
5212 if (ret < 0)
5213 goto out;
5214 ret = finish_inode_if_needed(sctx, 1);
5215 if (ret < 0)
5216 goto out;
5217 } else {
5218 ret = full_send_tree(sctx);
5219 if (ret < 0)
5220 goto out;
5221 }
5222
5223 out:
5224 free_recorded_refs(sctx);
5225 return ret;
5226 }
5227
5228 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5229 {
5230 spin_lock(&root->root_item_lock);
5231 root->send_in_progress--;
5232 /*
5233 * Not much left to do, we don't know why it's unbalanced and
5234 * can't blindly reset it to 0.
5235 */
5236 if (root->send_in_progress < 0)
5237 btrfs_err(root->fs_info,
5238 "send_in_progres unbalanced %d root %llu\n",
5239 root->send_in_progress, root->root_key.objectid);
5240 spin_unlock(&root->root_item_lock);
5241 }
5242
5243 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5244 {
5245 int ret = 0;
5246 struct btrfs_root *send_root;
5247 struct btrfs_root *clone_root;
5248 struct btrfs_fs_info *fs_info;
5249 struct btrfs_ioctl_send_args *arg = NULL;
5250 struct btrfs_key key;
5251 struct send_ctx *sctx = NULL;
5252 u32 i;
5253 u64 *clone_sources_tmp = NULL;
5254 int clone_sources_to_rollback = 0;
5255 int sort_clone_roots = 0;
5256 int index;
5257
5258 if (!capable(CAP_SYS_ADMIN))
5259 return -EPERM;
5260
5261 send_root = BTRFS_I(file_inode(mnt_file))->root;
5262 fs_info = send_root->fs_info;
5263
5264 /*
5265 * The subvolume must remain read-only during send, protect against
5266 * making it RW.
5267 */
5268 spin_lock(&send_root->root_item_lock);
5269 send_root->send_in_progress++;
5270 spin_unlock(&send_root->root_item_lock);
5271
5272 /*
5273 * This is done when we lookup the root, it should already be complete
5274 * by the time we get here.
5275 */
5276 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5277
5278 /*
5279 * Userspace tools do the checks and warn the user if it's
5280 * not RO.
5281 */
5282 if (!btrfs_root_readonly(send_root)) {
5283 ret = -EPERM;
5284 goto out;
5285 }
5286
5287 arg = memdup_user(arg_, sizeof(*arg));
5288 if (IS_ERR(arg)) {
5289 ret = PTR_ERR(arg);
5290 arg = NULL;
5291 goto out;
5292 }
5293
5294 if (!access_ok(VERIFY_READ, arg->clone_sources,
5295 sizeof(*arg->clone_sources) *
5296 arg->clone_sources_count)) {
5297 ret = -EFAULT;
5298 goto out;
5299 }
5300
5301 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5302 ret = -EINVAL;
5303 goto out;
5304 }
5305
5306 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5307 if (!sctx) {
5308 ret = -ENOMEM;
5309 goto out;
5310 }
5311
5312 INIT_LIST_HEAD(&sctx->new_refs);
5313 INIT_LIST_HEAD(&sctx->deleted_refs);
5314 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5315 INIT_LIST_HEAD(&sctx->name_cache_list);
5316
5317 sctx->flags = arg->flags;
5318
5319 sctx->send_filp = fget(arg->send_fd);
5320 if (!sctx->send_filp) {
5321 ret = -EBADF;
5322 goto out;
5323 }
5324
5325 sctx->send_root = send_root;
5326 sctx->clone_roots_cnt = arg->clone_sources_count;
5327
5328 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5329 sctx->send_buf = vmalloc(sctx->send_max_size);
5330 if (!sctx->send_buf) {
5331 ret = -ENOMEM;
5332 goto out;
5333 }
5334
5335 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5336 if (!sctx->read_buf) {
5337 ret = -ENOMEM;
5338 goto out;
5339 }
5340
5341 sctx->pending_dir_moves = RB_ROOT;
5342 sctx->waiting_dir_moves = RB_ROOT;
5343
5344 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5345 (arg->clone_sources_count + 1));
5346 if (!sctx->clone_roots) {
5347 ret = -ENOMEM;
5348 goto out;
5349 }
5350
5351 if (arg->clone_sources_count) {
5352 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5353 sizeof(*arg->clone_sources));
5354 if (!clone_sources_tmp) {
5355 ret = -ENOMEM;
5356 goto out;
5357 }
5358
5359 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5360 arg->clone_sources_count *
5361 sizeof(*arg->clone_sources));
5362 if (ret) {
5363 ret = -EFAULT;
5364 goto out;
5365 }
5366
5367 for (i = 0; i < arg->clone_sources_count; i++) {
5368 key.objectid = clone_sources_tmp[i];
5369 key.type = BTRFS_ROOT_ITEM_KEY;
5370 key.offset = (u64)-1;
5371
5372 index = srcu_read_lock(&fs_info->subvol_srcu);
5373
5374 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5375 if (IS_ERR(clone_root)) {
5376 srcu_read_unlock(&fs_info->subvol_srcu, index);
5377 ret = PTR_ERR(clone_root);
5378 goto out;
5379 }
5380 clone_sources_to_rollback = i + 1;
5381 spin_lock(&clone_root->root_item_lock);
5382 clone_root->send_in_progress++;
5383 if (!btrfs_root_readonly(clone_root)) {
5384 spin_unlock(&clone_root->root_item_lock);
5385 srcu_read_unlock(&fs_info->subvol_srcu, index);
5386 ret = -EPERM;
5387 goto out;
5388 }
5389 spin_unlock(&clone_root->root_item_lock);
5390 srcu_read_unlock(&fs_info->subvol_srcu, index);
5391
5392 sctx->clone_roots[i].root = clone_root;
5393 }
5394 vfree(clone_sources_tmp);
5395 clone_sources_tmp = NULL;
5396 }
5397
5398 if (arg->parent_root) {
5399 key.objectid = arg->parent_root;
5400 key.type = BTRFS_ROOT_ITEM_KEY;
5401 key.offset = (u64)-1;
5402
5403 index = srcu_read_lock(&fs_info->subvol_srcu);
5404
5405 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5406 if (IS_ERR(sctx->parent_root)) {
5407 srcu_read_unlock(&fs_info->subvol_srcu, index);
5408 ret = PTR_ERR(sctx->parent_root);
5409 goto out;
5410 }
5411
5412 spin_lock(&sctx->parent_root->root_item_lock);
5413 sctx->parent_root->send_in_progress++;
5414 if (!btrfs_root_readonly(sctx->parent_root)) {
5415 spin_unlock(&sctx->parent_root->root_item_lock);
5416 srcu_read_unlock(&fs_info->subvol_srcu, index);
5417 ret = -EPERM;
5418 goto out;
5419 }
5420 spin_unlock(&sctx->parent_root->root_item_lock);
5421
5422 srcu_read_unlock(&fs_info->subvol_srcu, index);
5423 }
5424
5425 /*
5426 * Clones from send_root are allowed, but only if the clone source
5427 * is behind the current send position. This is checked while searching
5428 * for possible clone sources.
5429 */
5430 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5431
5432 /* We do a bsearch later */
5433 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5434 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5435 NULL);
5436 sort_clone_roots = 1;
5437
5438 ret = send_subvol(sctx);
5439 if (ret < 0)
5440 goto out;
5441
5442 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5443 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5444 if (ret < 0)
5445 goto out;
5446 ret = send_cmd(sctx);
5447 if (ret < 0)
5448 goto out;
5449 }
5450
5451 out:
5452 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5453 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5454 struct rb_node *n;
5455 struct pending_dir_move *pm;
5456
5457 n = rb_first(&sctx->pending_dir_moves);
5458 pm = rb_entry(n, struct pending_dir_move, node);
5459 while (!list_empty(&pm->list)) {
5460 struct pending_dir_move *pm2;
5461
5462 pm2 = list_first_entry(&pm->list,
5463 struct pending_dir_move, list);
5464 free_pending_move(sctx, pm2);
5465 }
5466 free_pending_move(sctx, pm);
5467 }
5468
5469 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5470 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5471 struct rb_node *n;
5472 struct waiting_dir_move *dm;
5473
5474 n = rb_first(&sctx->waiting_dir_moves);
5475 dm = rb_entry(n, struct waiting_dir_move, node);
5476 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5477 kfree(dm);
5478 }
5479
5480 if (sort_clone_roots) {
5481 for (i = 0; i < sctx->clone_roots_cnt; i++)
5482 btrfs_root_dec_send_in_progress(
5483 sctx->clone_roots[i].root);
5484 } else {
5485 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5486 btrfs_root_dec_send_in_progress(
5487 sctx->clone_roots[i].root);
5488
5489 btrfs_root_dec_send_in_progress(send_root);
5490 }
5491 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5492 btrfs_root_dec_send_in_progress(sctx->parent_root);
5493
5494 kfree(arg);
5495 vfree(clone_sources_tmp);
5496
5497 if (sctx) {
5498 if (sctx->send_filp)
5499 fput(sctx->send_filp);
5500
5501 vfree(sctx->clone_roots);
5502 vfree(sctx->send_buf);
5503 vfree(sctx->read_buf);
5504
5505 name_cache_free(sctx);
5506
5507 kfree(sctx);
5508 }
5509
5510 return ret;
5511 }
Linux with added FPC-III support