?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: use normal return path for root == send_root case
[linux-2.6.git] / fs / btrfs / send.c
blobd2a4ee9125df84afc685fc49d81d53e481c75439
1 /*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/bsearch.h>
20 #include <linux/fs.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/crc32c.h>
28 #include <linux/vmalloc.h>
29
30 #include "send.h"
31 #include "backref.h"
32 #include "locking.h"
33 #include "disk-io.h"
34 #include "btrfs_inode.h"
35 #include "transaction.h"
36
37 static int g_verbose = 0;
38
39 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
40
41 /*
42 * A fs_path is a helper to dynamically build path names with unknown size.
43 * It reallocates the internal buffer on demand.
44 * It allows fast adding of path elements on the right side (normal path) and
45 * fast adding to the left side (reversed path). A reversed path can also be
46 * unreversed if needed.
47 */
48 struct fs_path {
49 union {
50 struct {
51 char *start;
52 char *end;
53 char *prepared;
54
55 char *buf;
56 int buf_len;
57 int reversed:1;
58 int virtual_mem:1;
59 char inline_buf[];
60 };
61 char pad[PAGE_SIZE];
62 };
63 };
64 #define FS_PATH_INLINE_SIZE \
65 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
66
67
68 /* reused for each extent */
69 struct clone_root {
70 struct btrfs_root *root;
71 u64 ino;
72 u64 offset;
73
74 u64 found_refs;
75 };
76
77 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
78 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
79
80 struct send_ctx {
81 struct file *send_filp;
82 loff_t send_off;
83 char *send_buf;
84 u32 send_size;
85 u32 send_max_size;
86 u64 total_send_size;
87 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
88
89 struct vfsmount *mnt;
90
91 struct btrfs_root *send_root;
92 struct btrfs_root *parent_root;
93 struct clone_root *clone_roots;
94 int clone_roots_cnt;
95
96 /* current state of the compare_tree call */
97 struct btrfs_path *left_path;
98 struct btrfs_path *right_path;
99 struct btrfs_key *cmp_key;
100
101 /*
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
104 */
105 u64 cur_ino;
106 u64 cur_inode_gen;
107 int cur_inode_new;
108 int cur_inode_new_gen;
109 int cur_inode_deleted;
110 u64 cur_inode_size;
111 u64 cur_inode_mode;
112
113 u64 send_progress;
114
115 struct list_head new_refs;
116 struct list_head deleted_refs;
117
118 struct radix_tree_root name_cache;
119 struct list_head name_cache_list;
120 int name_cache_size;
121
122 struct file *cur_inode_filp;
123 char *read_buf;
124 };
125
126 struct name_cache_entry {
127 struct list_head list;
128 u64 ino;
129 u64 gen;
130 u64 parent_ino;
131 u64 parent_gen;
132 int ret;
133 int need_later_update;
134 int name_len;
135 char name[];
136 };
137
138 static void fs_path_reset(struct fs_path *p)
139 {
140 if (p->reversed) {
141 p->start = p->buf + p->buf_len - 1;
142 p->end = p->start;
143 *p->start = 0;
144 } else {
145 p->start = p->buf;
146 p->end = p->start;
147 *p->start = 0;
148 }
149 }
150
151 static struct fs_path *fs_path_alloc(struct send_ctx *sctx)
152 {
153 struct fs_path *p;
154
155 p = kmalloc(sizeof(*p), GFP_NOFS);
156 if (!p)
157 return NULL;
158 p->reversed = 0;
159 p->virtual_mem = 0;
160 p->buf = p->inline_buf;
161 p->buf_len = FS_PATH_INLINE_SIZE;
162 fs_path_reset(p);
163 return p;
164 }
165
166 static struct fs_path *fs_path_alloc_reversed(struct send_ctx *sctx)
167 {
168 struct fs_path *p;
169
170 p = fs_path_alloc(sctx);
171 if (!p)
172 return NULL;
173 p->reversed = 1;
174 fs_path_reset(p);
175 return p;
176 }
177
178 static void fs_path_free(struct send_ctx *sctx, struct fs_path *p)
179 {
180 if (!p)
181 return;
182 if (p->buf != p->inline_buf) {
183 if (p->virtual_mem)
184 vfree(p->buf);
185 else
186 kfree(p->buf);
187 }
188 kfree(p);
189 }
190
191 static int fs_path_len(struct fs_path *p)
192 {
193 return p->end - p->start;
194 }
195
196 static int fs_path_ensure_buf(struct fs_path *p, int len)
197 {
198 char *tmp_buf;
199 int path_len;
200 int old_buf_len;
201
202 len++;
203
204 if (p->buf_len >= len)
205 return 0;
206
207 path_len = p->end - p->start;
208 old_buf_len = p->buf_len;
209 len = PAGE_ALIGN(len);
210
211 if (p->buf == p->inline_buf) {
212 tmp_buf = kmalloc(len, GFP_NOFS);
213 if (!tmp_buf) {
214 tmp_buf = vmalloc(len);
215 if (!tmp_buf)
216 return -ENOMEM;
217 p->virtual_mem = 1;
218 }
219 memcpy(tmp_buf, p->buf, p->buf_len);
220 p->buf = tmp_buf;
221 p->buf_len = len;
222 } else {
223 if (p->virtual_mem) {
224 tmp_buf = vmalloc(len);
225 if (!tmp_buf)
226 return -ENOMEM;
227 memcpy(tmp_buf, p->buf, p->buf_len);
228 vfree(p->buf);
229 } else {
230 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
231 if (!tmp_buf) {
232 tmp_buf = vmalloc(len);
233 if (!tmp_buf)
234 return -ENOMEM;
235 memcpy(tmp_buf, p->buf, p->buf_len);
236 kfree(p->buf);
237 p->virtual_mem = 1;
238 }
239 }
240 p->buf = tmp_buf;
241 p->buf_len = len;
242 }
243 if (p->reversed) {
244 tmp_buf = p->buf + old_buf_len - path_len - 1;
245 p->end = p->buf + p->buf_len - 1;
246 p->start = p->end - path_len;
247 memmove(p->start, tmp_buf, path_len + 1);
248 } else {
249 p->start = p->buf;
250 p->end = p->start + path_len;
251 }
252 return 0;
253 }
254
255 static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
256 {
257 int ret;
258 int new_len;
259
260 new_len = p->end - p->start + name_len;
261 if (p->start != p->end)
262 new_len++;
263 ret = fs_path_ensure_buf(p, new_len);
264 if (ret < 0)
265 goto out;
266
267 if (p->reversed) {
268 if (p->start != p->end)
269 *--p->start = '/';
270 p->start -= name_len;
271 p->prepared = p->start;
272 } else {
273 if (p->start != p->end)
274 *p->end++ = '/';
275 p->prepared = p->end;
276 p->end += name_len;
277 *p->end = 0;
278 }
279
280 out:
281 return ret;
282 }
283
284 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
285 {
286 int ret;
287
288 ret = fs_path_prepare_for_add(p, name_len);
289 if (ret < 0)
290 goto out;
291 memcpy(p->prepared, name, name_len);
292 p->prepared = NULL;
293
294 out:
295 return ret;
296 }
297
298 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
299 {
300 int ret;
301
302 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
303 if (ret < 0)
304 goto out;
305 memcpy(p->prepared, p2->start, p2->end - p2->start);
306 p->prepared = NULL;
307
308 out:
309 return ret;
310 }
311
312 static int fs_path_add_from_extent_buffer(struct fs_path *p,
313 struct extent_buffer *eb,
314 unsigned long off, int len)
315 {
316 int ret;
317
318 ret = fs_path_prepare_for_add(p, len);
319 if (ret < 0)
320 goto out;
321
322 read_extent_buffer(eb, p->prepared, off, len);
323 p->prepared = NULL;
324
325 out:
326 return ret;
327 }
328
329 #if 0
330 static void fs_path_remove(struct fs_path *p)
331 {
332 BUG_ON(p->reversed);
333 while (p->start != p->end && *p->end != '/')
334 p->end--;
335 *p->end = 0;
336 }
337 #endif
338
339 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
340 {
341 int ret;
342
343 p->reversed = from->reversed;
344 fs_path_reset(p);
345
346 ret = fs_path_add_path(p, from);
347
348 return ret;
349 }
350
351
352 static void fs_path_unreverse(struct fs_path *p)
353 {
354 char *tmp;
355 int len;
356
357 if (!p->reversed)
358 return;
359
360 tmp = p->start;
361 len = p->end - p->start;
362 p->start = p->buf;
363 p->end = p->start + len;
364 memmove(p->start, tmp, len + 1);
365 p->reversed = 0;
366 }
367
368 static struct btrfs_path *alloc_path_for_send(void)
369 {
370 struct btrfs_path *path;
371
372 path = btrfs_alloc_path();
373 if (!path)
374 return NULL;
375 path->search_commit_root = 1;
376 path->skip_locking = 1;
377 return path;
378 }
379
380 static int write_buf(struct send_ctx *sctx, const void *buf, u32 len)
381 {
382 int ret;
383 mm_segment_t old_fs;
384 u32 pos = 0;
385
386 old_fs = get_fs();
387 set_fs(KERNEL_DS);
388
389 while (pos < len) {
390 ret = vfs_write(sctx->send_filp, (char *)buf + pos, len - pos,
391 &sctx->send_off);
392 /* TODO handle that correctly */
393 /*if (ret == -ERESTARTSYS) {
394 continue;
395 }*/
396 if (ret < 0)
397 goto out;
398 if (ret == 0) {
399 ret = -EIO;
400 goto out;
401 }
402 pos += ret;
403 }
404
405 ret = 0;
406
407 out:
408 set_fs(old_fs);
409 return ret;
410 }
411
412 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
413 {
414 struct btrfs_tlv_header *hdr;
415 int total_len = sizeof(*hdr) + len;
416 int left = sctx->send_max_size - sctx->send_size;
417
418 if (unlikely(left < total_len))
419 return -EOVERFLOW;
420
421 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
422 hdr->tlv_type = cpu_to_le16(attr);
423 hdr->tlv_len = cpu_to_le16(len);
424 memcpy(hdr + 1, data, len);
425 sctx->send_size += total_len;
426
427 return 0;
428 }
429
430 #if 0
431 static int tlv_put_u8(struct send_ctx *sctx, u16 attr, u8 value)
432 {
433 return tlv_put(sctx, attr, &value, sizeof(value));
434 }
435
436 static int tlv_put_u16(struct send_ctx *sctx, u16 attr, u16 value)
437 {
438 __le16 tmp = cpu_to_le16(value);
439 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
440 }
441
442 static int tlv_put_u32(struct send_ctx *sctx, u16 attr, u32 value)
443 {
444 __le32 tmp = cpu_to_le32(value);
445 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
446 }
447 #endif
448
449 static int tlv_put_u64(struct send_ctx *sctx, u16 attr, u64 value)
450 {
451 __le64 tmp = cpu_to_le64(value);
452 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
453 }
454
455 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
456 const char *str, int len)
457 {
458 if (len == -1)
459 len = strlen(str);
460 return tlv_put(sctx, attr, str, len);
461 }
462
463 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
464 const u8 *uuid)
465 {
466 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
467 }
468
469 #if 0
470 static int tlv_put_timespec(struct send_ctx *sctx, u16 attr,
471 struct timespec *ts)
472 {
473 struct btrfs_timespec bts;
474 bts.sec = cpu_to_le64(ts->tv_sec);
475 bts.nsec = cpu_to_le32(ts->tv_nsec);
476 return tlv_put(sctx, attr, &bts, sizeof(bts));
477 }
478 #endif
479
480 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
481 struct extent_buffer *eb,
482 struct btrfs_timespec *ts)
483 {
484 struct btrfs_timespec bts;
485 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
486 return tlv_put(sctx, attr, &bts, sizeof(bts));
487 }
488
489
490 #define TLV_PUT(sctx, attrtype, attrlen, data) \
491 do { \
492 ret = tlv_put(sctx, attrtype, attrlen, data); \
493 if (ret < 0) \
494 goto tlv_put_failure; \
495 } while (0)
496
497 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
498 do { \
499 ret = tlv_put_u##bits(sctx, attrtype, value); \
500 if (ret < 0) \
501 goto tlv_put_failure; \
502 } while (0)
503
504 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
505 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
506 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
507 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
508 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
509 do { \
510 ret = tlv_put_string(sctx, attrtype, str, len); \
511 if (ret < 0) \
512 goto tlv_put_failure; \
513 } while (0)
514 #define TLV_PUT_PATH(sctx, attrtype, p) \
515 do { \
516 ret = tlv_put_string(sctx, attrtype, p->start, \
517 p->end - p->start); \
518 if (ret < 0) \
519 goto tlv_put_failure; \
520 } while(0)
521 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
522 do { \
523 ret = tlv_put_uuid(sctx, attrtype, uuid); \
524 if (ret < 0) \
525 goto tlv_put_failure; \
526 } while (0)
527 #define TLV_PUT_TIMESPEC(sctx, attrtype, ts) \
528 do { \
529 ret = tlv_put_timespec(sctx, attrtype, ts); \
530 if (ret < 0) \
531 goto tlv_put_failure; \
532 } while (0)
533 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
534 do { \
535 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
536 if (ret < 0) \
537 goto tlv_put_failure; \
538 } while (0)
539
540 static int send_header(struct send_ctx *sctx)
541 {
542 struct btrfs_stream_header hdr;
543
544 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
545 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
546
547 return write_buf(sctx, &hdr, sizeof(hdr));
548 }
549
550 /*
551 * For each command/item we want to send to userspace, we call this function.
552 */
553 static int begin_cmd(struct send_ctx *sctx, int cmd)
554 {
555 struct btrfs_cmd_header *hdr;
556
557 if (!sctx->send_buf) {
558 WARN_ON(1);
559 return -EINVAL;
560 }
561
562 BUG_ON(sctx->send_size);
563
564 sctx->send_size += sizeof(*hdr);
565 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
566 hdr->cmd = cpu_to_le16(cmd);
567
568 return 0;
569 }
570
571 static int send_cmd(struct send_ctx *sctx)
572 {
573 int ret;
574 struct btrfs_cmd_header *hdr;
575 u32 crc;
576
577 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
578 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
579 hdr->crc = 0;
580
581 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
582 hdr->crc = cpu_to_le32(crc);
583
584 ret = write_buf(sctx, sctx->send_buf, sctx->send_size);
585
586 sctx->total_send_size += sctx->send_size;
587 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
588 sctx->send_size = 0;
589
590 return ret;
591 }
592
593 /*
594 * Sends a move instruction to user space
595 */
596 static int send_rename(struct send_ctx *sctx,
597 struct fs_path *from, struct fs_path *to)
598 {
599 int ret;
600
601 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
602
603 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
604 if (ret < 0)
605 goto out;
606
607 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
608 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
609
610 ret = send_cmd(sctx);
611
612 tlv_put_failure:
613 out:
614 return ret;
615 }
616
617 /*
618 * Sends a link instruction to user space
619 */
620 static int send_link(struct send_ctx *sctx,
621 struct fs_path *path, struct fs_path *lnk)
622 {
623 int ret;
624
625 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
626
627 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
628 if (ret < 0)
629 goto out;
630
631 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
632 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
633
634 ret = send_cmd(sctx);
635
636 tlv_put_failure:
637 out:
638 return ret;
639 }
640
641 /*
642 * Sends an unlink instruction to user space
643 */
644 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
645 {
646 int ret;
647
648 verbose_printk("btrfs: send_unlink %s\n", path->start);
649
650 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
651 if (ret < 0)
652 goto out;
653
654 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
655
656 ret = send_cmd(sctx);
657
658 tlv_put_failure:
659 out:
660 return ret;
661 }
662
663 /*
664 * Sends a rmdir instruction to user space
665 */
666 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
667 {
668 int ret;
669
670 verbose_printk("btrfs: send_rmdir %s\n", path->start);
671
672 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
673 if (ret < 0)
674 goto out;
675
676 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
677
678 ret = send_cmd(sctx);
679
680 tlv_put_failure:
681 out:
682 return ret;
683 }
684
685 /*
686 * Helper function to retrieve some fields from an inode item.
687 */
688 static int get_inode_info(struct btrfs_root *root,
689 u64 ino, u64 *size, u64 *gen,
690 u64 *mode, u64 *uid, u64 *gid,
691 u64 *rdev)
692 {
693 int ret;
694 struct btrfs_inode_item *ii;
695 struct btrfs_key key;
696 struct btrfs_path *path;
697
698 path = alloc_path_for_send();
699 if (!path)
700 return -ENOMEM;
701
702 key.objectid = ino;
703 key.type = BTRFS_INODE_ITEM_KEY;
704 key.offset = 0;
705 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
706 if (ret < 0)
707 goto out;
708 if (ret) {
709 ret = -ENOENT;
710 goto out;
711 }
712
713 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
714 struct btrfs_inode_item);
715 if (size)
716 *size = btrfs_inode_size(path->nodes[0], ii);
717 if (gen)
718 *gen = btrfs_inode_generation(path->nodes[0], ii);
719 if (mode)
720 *mode = btrfs_inode_mode(path->nodes[0], ii);
721 if (uid)
722 *uid = btrfs_inode_uid(path->nodes[0], ii);
723 if (gid)
724 *gid = btrfs_inode_gid(path->nodes[0], ii);
725 if (rdev)
726 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
727
728 out:
729 btrfs_free_path(path);
730 return ret;
731 }
732
733 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
734 struct fs_path *p,
735 void *ctx);
736
737 /*
738 * Helper function to iterate the entries in ONE btrfs_inode_ref.
739 * The iterate callback may return a non zero value to stop iteration. This can
740 * be a negative value for error codes or 1 to simply stop it.
741 *
742 * path must point to the INODE_REF when called.
743 */
744 static int iterate_inode_ref(struct send_ctx *sctx,
745 struct btrfs_root *root, struct btrfs_path *path,
746 struct btrfs_key *found_key, int resolve,
747 iterate_inode_ref_t iterate, void *ctx)
748 {
749 struct extent_buffer *eb;
750 struct btrfs_item *item;
751 struct btrfs_inode_ref *iref;
752 struct btrfs_path *tmp_path;
753 struct fs_path *p;
754 u32 cur;
755 u32 len;
756 u32 total;
757 int slot;
758 u32 name_len;
759 char *start;
760 int ret = 0;
761 int num;
762 int index;
763
764 p = fs_path_alloc_reversed(sctx);
765 if (!p)
766 return -ENOMEM;
767
768 tmp_path = alloc_path_for_send();
769 if (!tmp_path) {
770 fs_path_free(sctx, p);
771 return -ENOMEM;
772 }
773
774 eb = path->nodes[0];
775 slot = path->slots[0];
776 item = btrfs_item_nr(eb, slot);
777 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
778 cur = 0;
779 len = 0;
780 total = btrfs_item_size(eb, item);
781
782 num = 0;
783 while (cur < total) {
784 fs_path_reset(p);
785
786 name_len = btrfs_inode_ref_name_len(eb, iref);
787 index = btrfs_inode_ref_index(eb, iref);
788 if (resolve) {
789 start = btrfs_iref_to_path(root, tmp_path, iref, eb,
790 found_key->offset, p->buf,
791 p->buf_len);
792 if (IS_ERR(start)) {
793 ret = PTR_ERR(start);
794 goto out;
795 }
796 if (start < p->buf) {
797 /* overflow , try again with larger buffer */
798 ret = fs_path_ensure_buf(p,
799 p->buf_len + p->buf - start);
800 if (ret < 0)
801 goto out;
802 start = btrfs_iref_to_path(root, tmp_path, iref,
803 eb, found_key->offset, p->buf,
804 p->buf_len);
805 if (IS_ERR(start)) {
806 ret = PTR_ERR(start);
807 goto out;
808 }
809 BUG_ON(start < p->buf);
810 }
811 p->start = start;
812 } else {
813 ret = fs_path_add_from_extent_buffer(p, eb,
814 (unsigned long)(iref + 1), name_len);
815 if (ret < 0)
816 goto out;
817 }
818
819
820 len = sizeof(*iref) + name_len;
821 iref = (struct btrfs_inode_ref *)((char *)iref + len);
822 cur += len;
823
824 ret = iterate(num, found_key->offset, index, p, ctx);
825 if (ret)
826 goto out;
827
828 num++;
829 }
830
831 out:
832 btrfs_free_path(tmp_path);
833 fs_path_free(sctx, p);
834 return ret;
835 }
836
837 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
838 const char *name, int name_len,
839 const char *data, int data_len,
840 u8 type, void *ctx);
841
842 /*
843 * Helper function to iterate the entries in ONE btrfs_dir_item.
844 * The iterate callback may return a non zero value to stop iteration. This can
845 * be a negative value for error codes or 1 to simply stop it.
846 *
847 * path must point to the dir item when called.
848 */
849 static int iterate_dir_item(struct send_ctx *sctx,
850 struct btrfs_root *root, struct btrfs_path *path,
851 struct btrfs_key *found_key,
852 iterate_dir_item_t iterate, void *ctx)
853 {
854 int ret = 0;
855 struct extent_buffer *eb;
856 struct btrfs_item *item;
857 struct btrfs_dir_item *di;
858 struct btrfs_path *tmp_path = NULL;
859 struct btrfs_key di_key;
860 char *buf = NULL;
861 char *buf2 = NULL;
862 int buf_len;
863 int buf_virtual = 0;
864 u32 name_len;
865 u32 data_len;
866 u32 cur;
867 u32 len;
868 u32 total;
869 int slot;
870 int num;
871 u8 type;
872
873 buf_len = PAGE_SIZE;
874 buf = kmalloc(buf_len, GFP_NOFS);
875 if (!buf) {
876 ret = -ENOMEM;
877 goto out;
878 }
879
880 tmp_path = alloc_path_for_send();
881 if (!tmp_path) {
882 ret = -ENOMEM;
883 goto out;
884 }
885
886 eb = path->nodes[0];
887 slot = path->slots[0];
888 item = btrfs_item_nr(eb, slot);
889 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
890 cur = 0;
891 len = 0;
892 total = btrfs_item_size(eb, item);
893
894 num = 0;
895 while (cur < total) {
896 name_len = btrfs_dir_name_len(eb, di);
897 data_len = btrfs_dir_data_len(eb, di);
898 type = btrfs_dir_type(eb, di);
899 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
900
901 if (name_len + data_len > buf_len) {
902 buf_len = PAGE_ALIGN(name_len + data_len);
903 if (buf_virtual) {
904 buf2 = vmalloc(buf_len);
905 if (!buf2) {
906 ret = -ENOMEM;
907 goto out;
908 }
909 vfree(buf);
910 } else {
911 buf2 = krealloc(buf, buf_len, GFP_NOFS);
912 if (!buf2) {
913 buf2 = vmalloc(buf_len);
914 if (!buf2) {
915 ret = -ENOMEM;
916 goto out;
917 }
918 kfree(buf);
919 buf_virtual = 1;
920 }
921 }
922
923 buf = buf2;
924 buf2 = NULL;
925 }
926
927 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
928 name_len + data_len);
929
930 len = sizeof(*di) + name_len + data_len;
931 di = (struct btrfs_dir_item *)((char *)di + len);
932 cur += len;
933
934 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
935 data_len, type, ctx);
936 if (ret < 0)
937 goto out;
938 if (ret) {
939 ret = 0;
940 goto out;
941 }
942
943 num++;
944 }
945
946 out:
947 btrfs_free_path(tmp_path);
948 if (buf_virtual)
949 vfree(buf);
950 else
951 kfree(buf);
952 return ret;
953 }
954
955 static int __copy_first_ref(int num, u64 dir, int index,
956 struct fs_path *p, void *ctx)
957 {
958 int ret;
959 struct fs_path *pt = ctx;
960
961 ret = fs_path_copy(pt, p);
962 if (ret < 0)
963 return ret;
964
965 /* we want the first only */
966 return 1;
967 }
968
969 /*
970 * Retrieve the first path of an inode. If an inode has more then one
971 * ref/hardlink, this is ignored.
972 */
973 static int get_inode_path(struct send_ctx *sctx, struct btrfs_root *root,
974 u64 ino, struct fs_path *path)
975 {
976 int ret;
977 struct btrfs_key key, found_key;
978 struct btrfs_path *p;
979
980 p = alloc_path_for_send();
981 if (!p)
982 return -ENOMEM;
983
984 fs_path_reset(path);
985
986 key.objectid = ino;
987 key.type = BTRFS_INODE_REF_KEY;
988 key.offset = 0;
989
990 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
991 if (ret < 0)
992 goto out;
993 if (ret) {
994 ret = 1;
995 goto out;
996 }
997 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
998 if (found_key.objectid != ino ||
999 found_key.type != BTRFS_INODE_REF_KEY) {
1000 ret = -ENOENT;
1001 goto out;
1002 }
1003
1004 ret = iterate_inode_ref(sctx, root, p, &found_key, 1,
1005 __copy_first_ref, path);
1006 if (ret < 0)
1007 goto out;
1008 ret = 0;
1009
1010 out:
1011 btrfs_free_path(p);
1012 return ret;
1013 }
1014
1015 struct backref_ctx {
1016 struct send_ctx *sctx;
1017
1018 /* number of total found references */
1019 u64 found;
1020
1021 /*
1022 * used for clones found in send_root. clones found behind cur_objectid
1023 * and cur_offset are not considered as allowed clones.
1024 */
1025 u64 cur_objectid;
1026 u64 cur_offset;
1027
1028 /* may be truncated in case it's the last extent in a file */
1029 u64 extent_len;
1030
1031 /* Just to check for bugs in backref resolving */
1032 int found_itself;
1033 };
1034
1035 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1036 {
1037 u64 root = (u64)key;
1038 struct clone_root *cr = (struct clone_root *)elt;
1039
1040 if (root < cr->root->objectid)
1041 return -1;
1042 if (root > cr->root->objectid)
1043 return 1;
1044 return 0;
1045 }
1046
1047 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1048 {
1049 struct clone_root *cr1 = (struct clone_root *)e1;
1050 struct clone_root *cr2 = (struct clone_root *)e2;
1051
1052 if (cr1->root->objectid < cr2->root->objectid)
1053 return -1;
1054 if (cr1->root->objectid > cr2->root->objectid)
1055 return 1;
1056 return 0;
1057 }
1058
1059 /*
1060 * Called for every backref that is found for the current extent.
1061 */
1062 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1063 {
1064 struct backref_ctx *bctx = ctx_;
1065 struct clone_root *found;
1066 int ret;
1067 u64 i_size;
1068
1069 /* First check if the root is in the list of accepted clone sources */
1070 found = bsearch((void *)root, bctx->sctx->clone_roots,
1071 bctx->sctx->clone_roots_cnt,
1072 sizeof(struct clone_root),
1073 __clone_root_cmp_bsearch);
1074 if (!found)
1075 return 0;
1076
1077 if (found->root == bctx->sctx->send_root &&
1078 ino == bctx->cur_objectid &&
1079 offset == bctx->cur_offset) {
1080 bctx->found_itself = 1;
1081 }
1082
1083 /*
1084 * There are inodes that have extents that lie behind it's i_size. Don't
1085 * accept clones from these extents.
1086 */
1087 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1088 NULL);
1089 if (ret < 0)
1090 return ret;
1091
1092 if (offset + bctx->extent_len > i_size)
1093 return 0;
1094
1095 /*
1096 * Make sure we don't consider clones from send_root that are
1097 * behind the current inode/offset.
1098 */
1099 if (found->root == bctx->sctx->send_root) {
1100 /*
1101 * TODO for the moment we don't accept clones from the inode
1102 * that is currently send. We may change this when
1103 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1104 * file.
1105 */
1106 if (ino >= bctx->cur_objectid)
1107 return 0;
1108 /*if (ino > ctx->cur_objectid)
1109 return 0;
1110 if (offset + ctx->extent_len > ctx->cur_offset)
1111 return 0;*/
1112 }
1113
1114 bctx->found++;
1115 found->found_refs++;
1116 if (ino < found->ino) {
1117 found->ino = ino;
1118 found->offset = offset;
1119 } else if (found->ino == ino) {
1120 /*
1121 * same extent found more then once in the same file.
1122 */
1123 if (found->offset > offset + bctx->extent_len)
1124 found->offset = offset;
1125 }
1126
1127 return 0;
1128 }
1129
1130 /*
1131 * path must point to the extent item when called.
1132 */
1133 static int find_extent_clone(struct send_ctx *sctx,
1134 struct btrfs_path *path,
1135 u64 ino, u64 data_offset,
1136 u64 ino_size,
1137 struct clone_root **found)
1138 {
1139 int ret;
1140 int extent_type;
1141 u64 logical;
1142 u64 num_bytes;
1143 u64 extent_item_pos;
1144 struct btrfs_file_extent_item *fi;
1145 struct extent_buffer *eb = path->nodes[0];
1146 struct backref_ctx *backref_ctx = NULL;
1147 struct clone_root *cur_clone_root;
1148 struct btrfs_key found_key;
1149 struct btrfs_path *tmp_path;
1150 u32 i;
1151
1152 tmp_path = alloc_path_for_send();
1153 if (!tmp_path)
1154 return -ENOMEM;
1155
1156 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1157 if (!backref_ctx) {
1158 ret = -ENOMEM;
1159 goto out;
1160 }
1161
1162 if (data_offset >= ino_size) {
1163 /*
1164 * There may be extents that lie behind the file's size.
1165 * I at least had this in combination with snapshotting while
1166 * writing large files.
1167 */
1168 ret = 0;
1169 goto out;
1170 }
1171
1172 fi = btrfs_item_ptr(eb, path->slots[0],
1173 struct btrfs_file_extent_item);
1174 extent_type = btrfs_file_extent_type(eb, fi);
1175 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1176 ret = -ENOENT;
1177 goto out;
1178 }
1179
1180 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1181 logical = btrfs_file_extent_disk_bytenr(eb, fi);
1182 if (logical == 0) {
1183 ret = -ENOENT;
1184 goto out;
1185 }
1186 logical += btrfs_file_extent_offset(eb, fi);
1187
1188 ret = extent_from_logical(sctx->send_root->fs_info,
1189 logical, tmp_path, &found_key);
1190 btrfs_release_path(tmp_path);
1191
1192 if (ret < 0)
1193 goto out;
1194 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1195 ret = -EIO;
1196 goto out;
1197 }
1198
1199 /*
1200 * Setup the clone roots.
1201 */
1202 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1203 cur_clone_root = sctx->clone_roots + i;
1204 cur_clone_root->ino = (u64)-1;
1205 cur_clone_root->offset = 0;
1206 cur_clone_root->found_refs = 0;
1207 }
1208
1209 backref_ctx->sctx = sctx;
1210 backref_ctx->found = 0;
1211 backref_ctx->cur_objectid = ino;
1212 backref_ctx->cur_offset = data_offset;
1213 backref_ctx->found_itself = 0;
1214 backref_ctx->extent_len = num_bytes;
1215
1216 /*
1217 * The last extent of a file may be too large due to page alignment.
1218 * We need to adjust extent_len in this case so that the checks in
1219 * __iterate_backrefs work.
1220 */
1221 if (data_offset + num_bytes >= ino_size)
1222 backref_ctx->extent_len = ino_size - data_offset;
1223
1224 /*
1225 * Now collect all backrefs.
1226 */
1227 extent_item_pos = logical - found_key.objectid;
1228 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1229 found_key.objectid, extent_item_pos, 1,
1230 __iterate_backrefs, backref_ctx);
1231 if (ret < 0)
1232 goto out;
1233
1234 if (!backref_ctx->found_itself) {
1235 /* found a bug in backref code? */
1236 ret = -EIO;
1237 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1238 "send_root. inode=%llu, offset=%llu, "
1239 "logical=%llu\n",
1240 ino, data_offset, logical);
1241 goto out;
1242 }
1243
1244 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1245 "ino=%llu, "
1246 "num_bytes=%llu, logical=%llu\n",
1247 data_offset, ino, num_bytes, logical);
1248
1249 if (!backref_ctx->found)
1250 verbose_printk("btrfs: no clones found\n");
1251
1252 cur_clone_root = NULL;
1253 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1254 if (sctx->clone_roots[i].found_refs) {
1255 if (!cur_clone_root)
1256 cur_clone_root = sctx->clone_roots + i;
1257 else if (sctx->clone_roots[i].root == sctx->send_root)
1258 /* prefer clones from send_root over others */
1259 cur_clone_root = sctx->clone_roots + i;
1260 break;
1261 }
1262
1263 }
1264
1265 if (cur_clone_root) {
1266 *found = cur_clone_root;
1267 ret = 0;
1268 } else {
1269 ret = -ENOENT;
1270 }
1271
1272 out:
1273 btrfs_free_path(tmp_path);
1274 kfree(backref_ctx);
1275 return ret;
1276 }
1277
1278 static int read_symlink(struct send_ctx *sctx,
1279 struct btrfs_root *root,
1280 u64 ino,
1281 struct fs_path *dest)
1282 {
1283 int ret;
1284 struct btrfs_path *path;
1285 struct btrfs_key key;
1286 struct btrfs_file_extent_item *ei;
1287 u8 type;
1288 u8 compression;
1289 unsigned long off;
1290 int len;
1291
1292 path = alloc_path_for_send();
1293 if (!path)
1294 return -ENOMEM;
1295
1296 key.objectid = ino;
1297 key.type = BTRFS_EXTENT_DATA_KEY;
1298 key.offset = 0;
1299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1300 if (ret < 0)
1301 goto out;
1302 BUG_ON(ret);
1303
1304 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1305 struct btrfs_file_extent_item);
1306 type = btrfs_file_extent_type(path->nodes[0], ei);
1307 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1308 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1309 BUG_ON(compression);
1310
1311 off = btrfs_file_extent_inline_start(ei);
1312 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1313
1314 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1315 if (ret < 0)
1316 goto out;
1317
1318 out:
1319 btrfs_free_path(path);
1320 return ret;
1321 }
1322
1323 /*
1324 * Helper function to generate a file name that is unique in the root of
1325 * send_root and parent_root. This is used to generate names for orphan inodes.
1326 */
1327 static int gen_unique_name(struct send_ctx *sctx,
1328 u64 ino, u64 gen,
1329 struct fs_path *dest)
1330 {
1331 int ret = 0;
1332 struct btrfs_path *path;
1333 struct btrfs_dir_item *di;
1334 char tmp[64];
1335 int len;
1336 u64 idx = 0;
1337
1338 path = alloc_path_for_send();
1339 if (!path)
1340 return -ENOMEM;
1341
1342 while (1) {
1343 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1344 ino, gen, idx);
1345 if (len >= sizeof(tmp)) {
1346 /* should really not happen */
1347 ret = -EOVERFLOW;
1348 goto out;
1349 }
1350
1351 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1352 path, BTRFS_FIRST_FREE_OBJECTID,
1353 tmp, strlen(tmp), 0);
1354 btrfs_release_path(path);
1355 if (IS_ERR(di)) {
1356 ret = PTR_ERR(di);
1357 goto out;
1358 }
1359 if (di) {
1360 /* not unique, try again */
1361 idx++;
1362 continue;
1363 }
1364
1365 if (!sctx->parent_root) {
1366 /* unique */
1367 ret = 0;
1368 break;
1369 }
1370
1371 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1372 path, BTRFS_FIRST_FREE_OBJECTID,
1373 tmp, strlen(tmp), 0);
1374 btrfs_release_path(path);
1375 if (IS_ERR(di)) {
1376 ret = PTR_ERR(di);
1377 goto out;
1378 }
1379 if (di) {
1380 /* not unique, try again */
1381 idx++;
1382 continue;
1383 }
1384 /* unique */
1385 break;
1386 }
1387
1388 ret = fs_path_add(dest, tmp, strlen(tmp));
1389
1390 out:
1391 btrfs_free_path(path);
1392 return ret;
1393 }
1394
1395 enum inode_state {
1396 inode_state_no_change,
1397 inode_state_will_create,
1398 inode_state_did_create,
1399 inode_state_will_delete,
1400 inode_state_did_delete,
1401 };
1402
1403 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1404 {
1405 int ret;
1406 int left_ret;
1407 int right_ret;
1408 u64 left_gen;
1409 u64 right_gen;
1410
1411 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1412 NULL, NULL);
1413 if (ret < 0 && ret != -ENOENT)
1414 goto out;
1415 left_ret = ret;
1416
1417 if (!sctx->parent_root) {
1418 right_ret = -ENOENT;
1419 } else {
1420 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1421 NULL, NULL, NULL, NULL);
1422 if (ret < 0 && ret != -ENOENT)
1423 goto out;
1424 right_ret = ret;
1425 }
1426
1427 if (!left_ret && !right_ret) {
1428 if (left_gen == gen && right_gen == gen)
1429 ret = inode_state_no_change;
1430 else if (left_gen == gen) {
1431 if (ino < sctx->send_progress)
1432 ret = inode_state_did_create;
1433 else
1434 ret = inode_state_will_create;
1435 } else if (right_gen == gen) {
1436 if (ino < sctx->send_progress)
1437 ret = inode_state_did_delete;
1438 else
1439 ret = inode_state_will_delete;
1440 } else {
1441 ret = -ENOENT;
1442 }
1443 } else if (!left_ret) {
1444 if (left_gen == gen) {
1445 if (ino < sctx->send_progress)
1446 ret = inode_state_did_create;
1447 else
1448 ret = inode_state_will_create;
1449 } else {
1450 ret = -ENOENT;
1451 }
1452 } else if (!right_ret) {
1453 if (right_gen == gen) {
1454 if (ino < sctx->send_progress)
1455 ret = inode_state_did_delete;
1456 else
1457 ret = inode_state_will_delete;
1458 } else {
1459 ret = -ENOENT;
1460 }
1461 } else {
1462 ret = -ENOENT;
1463 }
1464
1465 out:
1466 return ret;
1467 }
1468
1469 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1470 {
1471 int ret;
1472
1473 ret = get_cur_inode_state(sctx, ino, gen);
1474 if (ret < 0)
1475 goto out;
1476
1477 if (ret == inode_state_no_change ||
1478 ret == inode_state_did_create ||
1479 ret == inode_state_will_delete)
1480 ret = 1;
1481 else
1482 ret = 0;
1483
1484 out:
1485 return ret;
1486 }
1487
1488 /*
1489 * Helper function to lookup a dir item in a dir.
1490 */
1491 static int lookup_dir_item_inode(struct btrfs_root *root,
1492 u64 dir, const char *name, int name_len,
1493 u64 *found_inode,
1494 u8 *found_type)
1495 {
1496 int ret = 0;
1497 struct btrfs_dir_item *di;
1498 struct btrfs_key key;
1499 struct btrfs_path *path;
1500
1501 path = alloc_path_for_send();
1502 if (!path)
1503 return -ENOMEM;
1504
1505 di = btrfs_lookup_dir_item(NULL, root, path,
1506 dir, name, name_len, 0);
1507 if (!di) {
1508 ret = -ENOENT;
1509 goto out;
1510 }
1511 if (IS_ERR(di)) {
1512 ret = PTR_ERR(di);
1513 goto out;
1514 }
1515 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1516 *found_inode = key.objectid;
1517 *found_type = btrfs_dir_type(path->nodes[0], di);
1518
1519 out:
1520 btrfs_free_path(path);
1521 return ret;
1522 }
1523
1524 static int get_first_ref(struct send_ctx *sctx,
1525 struct btrfs_root *root, u64 ino,
1526 u64 *dir, u64 *dir_gen, struct fs_path *name)
1527 {
1528 int ret;
1529 struct btrfs_key key;
1530 struct btrfs_key found_key;
1531 struct btrfs_path *path;
1532 struct btrfs_inode_ref *iref;
1533 int len;
1534
1535 path = alloc_path_for_send();
1536 if (!path)
1537 return -ENOMEM;
1538
1539 key.objectid = ino;
1540 key.type = BTRFS_INODE_REF_KEY;
1541 key.offset = 0;
1542
1543 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1544 if (ret < 0)
1545 goto out;
1546 if (!ret)
1547 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1548 path->slots[0]);
1549 if (ret || found_key.objectid != key.objectid ||
1550 found_key.type != key.type) {
1551 ret = -ENOENT;
1552 goto out;
1553 }
1554
1555 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1556 struct btrfs_inode_ref);
1557 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1558 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1559 (unsigned long)(iref + 1), len);
1560 if (ret < 0)
1561 goto out;
1562 btrfs_release_path(path);
1563
1564 ret = get_inode_info(root, found_key.offset, NULL, dir_gen, NULL, NULL,
1565 NULL, NULL);
1566 if (ret < 0)
1567 goto out;
1568
1569 *dir = found_key.offset;
1570
1571 out:
1572 btrfs_free_path(path);
1573 return ret;
1574 }
1575
1576 static int is_first_ref(struct send_ctx *sctx,
1577 struct btrfs_root *root,
1578 u64 ino, u64 dir,
1579 const char *name, int name_len)
1580 {
1581 int ret;
1582 struct fs_path *tmp_name;
1583 u64 tmp_dir;
1584 u64 tmp_dir_gen;
1585
1586 tmp_name = fs_path_alloc(sctx);
1587 if (!tmp_name)
1588 return -ENOMEM;
1589
1590 ret = get_first_ref(sctx, root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1591 if (ret < 0)
1592 goto out;
1593
1594 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1595 ret = 0;
1596 goto out;
1597 }
1598
1599 ret = memcmp(tmp_name->start, name, name_len);
1600 if (ret)
1601 ret = 0;
1602 else
1603 ret = 1;
1604
1605 out:
1606 fs_path_free(sctx, tmp_name);
1607 return ret;
1608 }
1609
1610 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1611 const char *name, int name_len,
1612 u64 *who_ino, u64 *who_gen)
1613 {
1614 int ret = 0;
1615 u64 other_inode = 0;
1616 u8 other_type = 0;
1617
1618 if (!sctx->parent_root)
1619 goto out;
1620
1621 ret = is_inode_existent(sctx, dir, dir_gen);
1622 if (ret <= 0)
1623 goto out;
1624
1625 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1626 &other_inode, &other_type);
1627 if (ret < 0 && ret != -ENOENT)
1628 goto out;
1629 if (ret) {
1630 ret = 0;
1631 goto out;
1632 }
1633
1634 if (other_inode > sctx->send_progress) {
1635 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1636 who_gen, NULL, NULL, NULL, NULL);
1637 if (ret < 0)
1638 goto out;
1639
1640 ret = 1;
1641 *who_ino = other_inode;
1642 } else {
1643 ret = 0;
1644 }
1645
1646 out:
1647 return ret;
1648 }
1649
1650 static int did_overwrite_ref(struct send_ctx *sctx,
1651 u64 dir, u64 dir_gen,
1652 u64 ino, u64 ino_gen,
1653 const char *name, int name_len)
1654 {
1655 int ret = 0;
1656 u64 gen;
1657 u64 ow_inode;
1658 u8 other_type;
1659
1660 if (!sctx->parent_root)
1661 goto out;
1662
1663 ret = is_inode_existent(sctx, dir, dir_gen);
1664 if (ret <= 0)
1665 goto out;
1666
1667 /* check if the ref was overwritten by another ref */
1668 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1669 &ow_inode, &other_type);
1670 if (ret < 0 && ret != -ENOENT)
1671 goto out;
1672 if (ret) {
1673 /* was never and will never be overwritten */
1674 ret = 0;
1675 goto out;
1676 }
1677
1678 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1679 NULL, NULL);
1680 if (ret < 0)
1681 goto out;
1682
1683 if (ow_inode == ino && gen == ino_gen) {
1684 ret = 0;
1685 goto out;
1686 }
1687
1688 /* we know that it is or will be overwritten. check this now */
1689 if (ow_inode < sctx->send_progress)
1690 ret = 1;
1691 else
1692 ret = 0;
1693
1694 out:
1695 return ret;
1696 }
1697
1698 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1699 {
1700 int ret = 0;
1701 struct fs_path *name = NULL;
1702 u64 dir;
1703 u64 dir_gen;
1704
1705 if (!sctx->parent_root)
1706 goto out;
1707
1708 name = fs_path_alloc(sctx);
1709 if (!name)
1710 return -ENOMEM;
1711
1712 ret = get_first_ref(sctx, sctx->parent_root, ino, &dir, &dir_gen, name);
1713 if (ret < 0)
1714 goto out;
1715
1716 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1717 name->start, fs_path_len(name));
1718 if (ret < 0)
1719 goto out;
1720
1721 out:
1722 fs_path_free(sctx, name);
1723 return ret;
1724 }
1725
1726 static int name_cache_insert(struct send_ctx *sctx,
1727 struct name_cache_entry *nce)
1728 {
1729 int ret = 0;
1730 struct name_cache_entry **ncea;
1731
1732 ncea = radix_tree_lookup(&sctx->name_cache, nce->ino);
1733 if (ncea) {
1734 if (!ncea[0])
1735 ncea[0] = nce;
1736 else if (!ncea[1])
1737 ncea[1] = nce;
1738 else
1739 BUG();
1740 } else {
1741 ncea = kmalloc(sizeof(void *) * 2, GFP_NOFS);
1742 if (!ncea)
1743 return -ENOMEM;
1744
1745 ncea[0] = nce;
1746 ncea[1] = NULL;
1747 ret = radix_tree_insert(&sctx->name_cache, nce->ino, ncea);
1748 if (ret < 0)
1749 return ret;
1750 }
1751 list_add_tail(&nce->list, &sctx->name_cache_list);
1752 sctx->name_cache_size++;
1753
1754 return ret;
1755 }
1756
1757 static void name_cache_delete(struct send_ctx *sctx,
1758 struct name_cache_entry *nce)
1759 {
1760 struct name_cache_entry **ncea;
1761
1762 ncea = radix_tree_lookup(&sctx->name_cache, nce->ino);
1763 BUG_ON(!ncea);
1764
1765 if (ncea[0] == nce)
1766 ncea[0] = NULL;
1767 else if (ncea[1] == nce)
1768 ncea[1] = NULL;
1769 else
1770 BUG();
1771
1772 if (!ncea[0] && !ncea[1]) {
1773 radix_tree_delete(&sctx->name_cache, nce->ino);
1774 kfree(ncea);
1775 }
1776
1777 list_del(&nce->list);
1778
1779 sctx->name_cache_size--;
1780 }
1781
1782 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1783 u64 ino, u64 gen)
1784 {
1785 struct name_cache_entry **ncea;
1786
1787 ncea = radix_tree_lookup(&sctx->name_cache, ino);
1788 if (!ncea)
1789 return NULL;
1790
1791 if (ncea[0] && ncea[0]->gen == gen)
1792 return ncea[0];
1793 else if (ncea[1] && ncea[1]->gen == gen)
1794 return ncea[1];
1795 return NULL;
1796 }
1797
1798 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1799 {
1800 list_del(&nce->list);
1801 list_add_tail(&nce->list, &sctx->name_cache_list);
1802 }
1803
1804 static void name_cache_clean_unused(struct send_ctx *sctx)
1805 {
1806 struct name_cache_entry *nce;
1807
1808 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1809 return;
1810
1811 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1812 nce = list_entry(sctx->name_cache_list.next,
1813 struct name_cache_entry, list);
1814 name_cache_delete(sctx, nce);
1815 kfree(nce);
1816 }
1817 }
1818
1819 static void name_cache_free(struct send_ctx *sctx)
1820 {
1821 struct name_cache_entry *nce;
1822 struct name_cache_entry *tmp;
1823
1824 list_for_each_entry_safe(nce, tmp, &sctx->name_cache_list, list) {
1825 name_cache_delete(sctx, nce);
1826 }
1827 }
1828
1829 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1830 u64 ino, u64 gen,
1831 u64 *parent_ino,
1832 u64 *parent_gen,
1833 struct fs_path *dest)
1834 {
1835 int ret;
1836 int nce_ret;
1837 struct btrfs_path *path = NULL;
1838 struct name_cache_entry *nce = NULL;
1839
1840 nce = name_cache_search(sctx, ino, gen);
1841 if (nce) {
1842 if (ino < sctx->send_progress && nce->need_later_update) {
1843 name_cache_delete(sctx, nce);
1844 kfree(nce);
1845 nce = NULL;
1846 } else {
1847 name_cache_used(sctx, nce);
1848 *parent_ino = nce->parent_ino;
1849 *parent_gen = nce->parent_gen;
1850 ret = fs_path_add(dest, nce->name, nce->name_len);
1851 if (ret < 0)
1852 goto out;
1853 ret = nce->ret;
1854 goto out;
1855 }
1856 }
1857
1858 path = alloc_path_for_send();
1859 if (!path)
1860 return -ENOMEM;
1861
1862 ret = is_inode_existent(sctx, ino, gen);
1863 if (ret < 0)
1864 goto out;
1865
1866 if (!ret) {
1867 ret = gen_unique_name(sctx, ino, gen, dest);
1868 if (ret < 0)
1869 goto out;
1870 ret = 1;
1871 goto out_cache;
1872 }
1873
1874 if (ino < sctx->send_progress)
1875 ret = get_first_ref(sctx, sctx->send_root, ino,
1876 parent_ino, parent_gen, dest);
1877 else
1878 ret = get_first_ref(sctx, sctx->parent_root, ino,
1879 parent_ino, parent_gen, dest);
1880 if (ret < 0)
1881 goto out;
1882
1883 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
1884 dest->start, dest->end - dest->start);
1885 if (ret < 0)
1886 goto out;
1887 if (ret) {
1888 fs_path_reset(dest);
1889 ret = gen_unique_name(sctx, ino, gen, dest);
1890 if (ret < 0)
1891 goto out;
1892 ret = 1;
1893 }
1894
1895 out_cache:
1896 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
1897 if (!nce) {
1898 ret = -ENOMEM;
1899 goto out;
1900 }
1901
1902 nce->ino = ino;
1903 nce->gen = gen;
1904 nce->parent_ino = *parent_ino;
1905 nce->parent_gen = *parent_gen;
1906 nce->name_len = fs_path_len(dest);
1907 nce->ret = ret;
1908 strcpy(nce->name, dest->start);
1909
1910 if (ino < sctx->send_progress)
1911 nce->need_later_update = 0;
1912 else
1913 nce->need_later_update = 1;
1914
1915 nce_ret = name_cache_insert(sctx, nce);
1916 if (nce_ret < 0)
1917 ret = nce_ret;
1918 name_cache_clean_unused(sctx);
1919
1920 out:
1921 btrfs_free_path(path);
1922 return ret;
1923 }
1924
1925 /*
1926 * Magic happens here. This function returns the first ref to an inode as it
1927 * would look like while receiving the stream at this point in time.
1928 * We walk the path up to the root. For every inode in between, we check if it
1929 * was already processed/sent. If yes, we continue with the parent as found
1930 * in send_root. If not, we continue with the parent as found in parent_root.
1931 * If we encounter an inode that was deleted at this point in time, we use the
1932 * inodes "orphan" name instead of the real name and stop. Same with new inodes
1933 * that were not created yet and overwritten inodes/refs.
1934 *
1935 * When do we have have orphan inodes:
1936 * 1. When an inode is freshly created and thus no valid refs are available yet
1937 * 2. When a directory lost all it's refs (deleted) but still has dir items
1938 * inside which were not processed yet (pending for move/delete). If anyone
1939 * tried to get the path to the dir items, it would get a path inside that
1940 * orphan directory.
1941 * 3. When an inode is moved around or gets new links, it may overwrite the ref
1942 * of an unprocessed inode. If in that case the first ref would be
1943 * overwritten, the overwritten inode gets "orphanized". Later when we
1944 * process this overwritten inode, it is restored at a new place by moving
1945 * the orphan inode.
1946 *
1947 * sctx->send_progress tells this function at which point in time receiving
1948 * would be.
1949 */
1950 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
1951 struct fs_path *dest)
1952 {
1953 int ret = 0;
1954 struct fs_path *name = NULL;
1955 u64 parent_inode = 0;
1956 u64 parent_gen = 0;
1957 int stop = 0;
1958
1959 name = fs_path_alloc(sctx);
1960 if (!name) {
1961 ret = -ENOMEM;
1962 goto out;
1963 }
1964
1965 dest->reversed = 1;
1966 fs_path_reset(dest);
1967
1968 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
1969 fs_path_reset(name);
1970
1971 ret = __get_cur_name_and_parent(sctx, ino, gen,
1972 &parent_inode, &parent_gen, name);
1973 if (ret < 0)
1974 goto out;
1975 if (ret)
1976 stop = 1;
1977
1978 ret = fs_path_add_path(dest, name);
1979 if (ret < 0)
1980 goto out;
1981
1982 ino = parent_inode;
1983 gen = parent_gen;
1984 }
1985
1986 out:
1987 fs_path_free(sctx, name);
1988 if (!ret)
1989 fs_path_unreverse(dest);
1990 return ret;
1991 }
1992
1993 /*
1994 * Called for regular files when sending extents data. Opens a struct file
1995 * to read from the file.
1996 */
1997 static int open_cur_inode_file(struct send_ctx *sctx)
1998 {
1999 int ret = 0;
2000 struct btrfs_key key;
2001 struct path path;
2002 struct inode *inode;
2003 struct dentry *dentry;
2004 struct file *filp;
2005 int new = 0;
2006
2007 if (sctx->cur_inode_filp)
2008 goto out;
2009
2010 key.objectid = sctx->cur_ino;
2011 key.type = BTRFS_INODE_ITEM_KEY;
2012 key.offset = 0;
2013
2014 inode = btrfs_iget(sctx->send_root->fs_info->sb, &key, sctx->send_root,
2015 &new);
2016 if (IS_ERR(inode)) {
2017 ret = PTR_ERR(inode);
2018 goto out;
2019 }
2020
2021 dentry = d_obtain_alias(inode);
2022 inode = NULL;
2023 if (IS_ERR(dentry)) {
2024 ret = PTR_ERR(dentry);
2025 goto out;
2026 }
2027
2028 path.mnt = sctx->mnt;
2029 path.dentry = dentry;
2030 filp = dentry_open(&path, O_RDONLY | O_LARGEFILE, current_cred());
2031 dput(dentry);
2032 dentry = NULL;
2033 if (IS_ERR(filp)) {
2034 ret = PTR_ERR(filp);
2035 goto out;
2036 }
2037 sctx->cur_inode_filp = filp;
2038
2039 out:
2040 /*
2041 * no xxxput required here as every vfs op
2042 * does it by itself on failure
2043 */
2044 return ret;
2045 }
2046
2047 /*
2048 * Closes the struct file that was created in open_cur_inode_file
2049 */
2050 static int close_cur_inode_file(struct send_ctx *sctx)
2051 {
2052 int ret = 0;
2053
2054 if (!sctx->cur_inode_filp)
2055 goto out;
2056
2057 ret = filp_close(sctx->cur_inode_filp, NULL);
2058 sctx->cur_inode_filp = NULL;
2059
2060 out:
2061 return ret;
2062 }
2063
2064 /*
2065 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2066 */
2067 static int send_subvol_begin(struct send_ctx *sctx)
2068 {
2069 int ret;
2070 struct btrfs_root *send_root = sctx->send_root;
2071 struct btrfs_root *parent_root = sctx->parent_root;
2072 struct btrfs_path *path;
2073 struct btrfs_key key;
2074 struct btrfs_root_ref *ref;
2075 struct extent_buffer *leaf;
2076 char *name = NULL;
2077 int namelen;
2078
2079 path = alloc_path_for_send();
2080 if (!path)
2081 return -ENOMEM;
2082
2083 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2084 if (!name) {
2085 btrfs_free_path(path);
2086 return -ENOMEM;
2087 }
2088
2089 key.objectid = send_root->objectid;
2090 key.type = BTRFS_ROOT_BACKREF_KEY;
2091 key.offset = 0;
2092
2093 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2094 &key, path, 1, 0);
2095 if (ret < 0)
2096 goto out;
2097 if (ret) {
2098 ret = -ENOENT;
2099 goto out;
2100 }
2101
2102 leaf = path->nodes[0];
2103 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2104 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2105 key.objectid != send_root->objectid) {
2106 ret = -ENOENT;
2107 goto out;
2108 }
2109 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2110 namelen = btrfs_root_ref_name_len(leaf, ref);
2111 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2112 btrfs_release_path(path);
2113
2114 if (ret < 0)
2115 goto out;
2116
2117 if (parent_root) {
2118 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2119 if (ret < 0)
2120 goto out;
2121 } else {
2122 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2123 if (ret < 0)
2124 goto out;
2125 }
2126
2127 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2128 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2129 sctx->send_root->root_item.uuid);
2130 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2131 sctx->send_root->root_item.ctransid);
2132 if (parent_root) {
2133 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2134 sctx->parent_root->root_item.uuid);
2135 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2136 sctx->parent_root->root_item.ctransid);
2137 }
2138
2139 ret = send_cmd(sctx);
2140
2141 tlv_put_failure:
2142 out:
2143 btrfs_free_path(path);
2144 kfree(name);
2145 return ret;
2146 }
2147
2148 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2149 {
2150 int ret = 0;
2151 struct fs_path *p;
2152
2153 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2154
2155 p = fs_path_alloc(sctx);
2156 if (!p)
2157 return -ENOMEM;
2158
2159 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2160 if (ret < 0)
2161 goto out;
2162
2163 ret = get_cur_path(sctx, ino, gen, p);
2164 if (ret < 0)
2165 goto out;
2166 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2167 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2168
2169 ret = send_cmd(sctx);
2170
2171 tlv_put_failure:
2172 out:
2173 fs_path_free(sctx, p);
2174 return ret;
2175 }
2176
2177 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2178 {
2179 int ret = 0;
2180 struct fs_path *p;
2181
2182 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2183
2184 p = fs_path_alloc(sctx);
2185 if (!p)
2186 return -ENOMEM;
2187
2188 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2189 if (ret < 0)
2190 goto out;
2191
2192 ret = get_cur_path(sctx, ino, gen, p);
2193 if (ret < 0)
2194 goto out;
2195 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2196 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2197
2198 ret = send_cmd(sctx);
2199
2200 tlv_put_failure:
2201 out:
2202 fs_path_free(sctx, p);
2203 return ret;
2204 }
2205
2206 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2207 {
2208 int ret = 0;
2209 struct fs_path *p;
2210
2211 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2212
2213 p = fs_path_alloc(sctx);
2214 if (!p)
2215 return -ENOMEM;
2216
2217 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2218 if (ret < 0)
2219 goto out;
2220
2221 ret = get_cur_path(sctx, ino, gen, p);
2222 if (ret < 0)
2223 goto out;
2224 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2225 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2226 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2227
2228 ret = send_cmd(sctx);
2229
2230 tlv_put_failure:
2231 out:
2232 fs_path_free(sctx, p);
2233 return ret;
2234 }
2235
2236 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2237 {
2238 int ret = 0;
2239 struct fs_path *p = NULL;
2240 struct btrfs_inode_item *ii;
2241 struct btrfs_path *path = NULL;
2242 struct extent_buffer *eb;
2243 struct btrfs_key key;
2244 int slot;
2245
2246 verbose_printk("btrfs: send_utimes %llu\n", ino);
2247
2248 p = fs_path_alloc(sctx);
2249 if (!p)
2250 return -ENOMEM;
2251
2252 path = alloc_path_for_send();
2253 if (!path) {
2254 ret = -ENOMEM;
2255 goto out;
2256 }
2257
2258 key.objectid = ino;
2259 key.type = BTRFS_INODE_ITEM_KEY;
2260 key.offset = 0;
2261 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2262 if (ret < 0)
2263 goto out;
2264
2265 eb = path->nodes[0];
2266 slot = path->slots[0];
2267 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2268
2269 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2270 if (ret < 0)
2271 goto out;
2272
2273 ret = get_cur_path(sctx, ino, gen, p);
2274 if (ret < 0)
2275 goto out;
2276 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2277 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2278 btrfs_inode_atime(ii));
2279 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2280 btrfs_inode_mtime(ii));
2281 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2282 btrfs_inode_ctime(ii));
2283 /* TODO otime? */
2284
2285 ret = send_cmd(sctx);
2286
2287 tlv_put_failure:
2288 out:
2289 fs_path_free(sctx, p);
2290 btrfs_free_path(path);
2291 return ret;
2292 }
2293
2294 /*
2295 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2296 * a valid path yet because we did not process the refs yet. So, the inode
2297 * is created as orphan.
2298 */
2299 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2300 {
2301 int ret = 0;
2302 struct fs_path *p;
2303 int cmd;
2304 u64 gen;
2305 u64 mode;
2306 u64 rdev;
2307
2308 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2309
2310 p = fs_path_alloc(sctx);
2311 if (!p)
2312 return -ENOMEM;
2313
2314 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2315 NULL, &rdev);
2316 if (ret < 0)
2317 goto out;
2318
2319 if (S_ISREG(mode))
2320 cmd = BTRFS_SEND_C_MKFILE;
2321 else if (S_ISDIR(mode))
2322 cmd = BTRFS_SEND_C_MKDIR;
2323 else if (S_ISLNK(mode))
2324 cmd = BTRFS_SEND_C_SYMLINK;
2325 else if (S_ISCHR(mode) || S_ISBLK(mode))
2326 cmd = BTRFS_SEND_C_MKNOD;
2327 else if (S_ISFIFO(mode))
2328 cmd = BTRFS_SEND_C_MKFIFO;
2329 else if (S_ISSOCK(mode))
2330 cmd = BTRFS_SEND_C_MKSOCK;
2331 else {
2332 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2333 (int)(mode & S_IFMT));
2334 ret = -ENOTSUPP;
2335 goto out;
2336 }
2337
2338 ret = begin_cmd(sctx, cmd);
2339 if (ret < 0)
2340 goto out;
2341
2342 ret = gen_unique_name(sctx, ino, gen, p);
2343 if (ret < 0)
2344 goto out;
2345
2346 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2347 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2348
2349 if (S_ISLNK(mode)) {
2350 fs_path_reset(p);
2351 ret = read_symlink(sctx, sctx->send_root, ino, p);
2352 if (ret < 0)
2353 goto out;
2354 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2355 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2356 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2357 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, rdev);
2358 }
2359
2360 ret = send_cmd(sctx);
2361 if (ret < 0)
2362 goto out;
2363
2364
2365 tlv_put_failure:
2366 out:
2367 fs_path_free(sctx, p);
2368 return ret;
2369 }
2370
2371 /*
2372 * We need some special handling for inodes that get processed before the parent
2373 * directory got created. See process_recorded_refs for details.
2374 * This function does the check if we already created the dir out of order.
2375 */
2376 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2377 {
2378 int ret = 0;
2379 struct btrfs_path *path = NULL;
2380 struct btrfs_key key;
2381 struct btrfs_key found_key;
2382 struct btrfs_key di_key;
2383 struct extent_buffer *eb;
2384 struct btrfs_dir_item *di;
2385 int slot;
2386
2387 path = alloc_path_for_send();
2388 if (!path) {
2389 ret = -ENOMEM;
2390 goto out;
2391 }
2392
2393 key.objectid = dir;
2394 key.type = BTRFS_DIR_INDEX_KEY;
2395 key.offset = 0;
2396 while (1) {
2397 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2398 1, 0);
2399 if (ret < 0)
2400 goto out;
2401 if (!ret) {
2402 eb = path->nodes[0];
2403 slot = path->slots[0];
2404 btrfs_item_key_to_cpu(eb, &found_key, slot);
2405 }
2406 if (ret || found_key.objectid != key.objectid ||
2407 found_key.type != key.type) {
2408 ret = 0;
2409 goto out;
2410 }
2411
2412 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2413 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2414
2415 if (di_key.objectid < sctx->send_progress) {
2416 ret = 1;
2417 goto out;
2418 }
2419
2420 key.offset = found_key.offset + 1;
2421 btrfs_release_path(path);
2422 }
2423
2424 out:
2425 btrfs_free_path(path);
2426 return ret;
2427 }
2428
2429 /*
2430 * Only creates the inode if it is:
2431 * 1. Not a directory
2432 * 2. Or a directory which was not created already due to out of order
2433 * directories. See did_create_dir and process_recorded_refs for details.
2434 */
2435 static int send_create_inode_if_needed(struct send_ctx *sctx)
2436 {
2437 int ret;
2438
2439 if (S_ISDIR(sctx->cur_inode_mode)) {
2440 ret = did_create_dir(sctx, sctx->cur_ino);
2441 if (ret < 0)
2442 goto out;
2443 if (ret) {
2444 ret = 0;
2445 goto out;
2446 }
2447 }
2448
2449 ret = send_create_inode(sctx, sctx->cur_ino);
2450 if (ret < 0)
2451 goto out;
2452
2453 out:
2454 return ret;
2455 }
2456
2457 struct recorded_ref {
2458 struct list_head list;
2459 char *dir_path;
2460 char *name;
2461 struct fs_path *full_path;
2462 u64 dir;
2463 u64 dir_gen;
2464 int dir_path_len;
2465 int name_len;
2466 };
2467
2468 /*
2469 * We need to process new refs before deleted refs, but compare_tree gives us
2470 * everything mixed. So we first record all refs and later process them.
2471 * This function is a helper to record one ref.
2472 */
2473 static int record_ref(struct list_head *head, u64 dir,
2474 u64 dir_gen, struct fs_path *path)
2475 {
2476 struct recorded_ref *ref;
2477 char *tmp;
2478
2479 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2480 if (!ref)
2481 return -ENOMEM;
2482
2483 ref->dir = dir;
2484 ref->dir_gen = dir_gen;
2485 ref->full_path = path;
2486
2487 tmp = strrchr(ref->full_path->start, '/');
2488 if (!tmp) {
2489 ref->name_len = ref->full_path->end - ref->full_path->start;
2490 ref->name = ref->full_path->start;
2491 ref->dir_path_len = 0;
2492 ref->dir_path = ref->full_path->start;
2493 } else {
2494 tmp++;
2495 ref->name_len = ref->full_path->end - tmp;
2496 ref->name = tmp;
2497 ref->dir_path = ref->full_path->start;
2498 ref->dir_path_len = ref->full_path->end -
2499 ref->full_path->start - 1 - ref->name_len;
2500 }
2501
2502 list_add_tail(&ref->list, head);
2503 return 0;
2504 }
2505
2506 static void __free_recorded_refs(struct send_ctx *sctx, struct list_head *head)
2507 {
2508 struct recorded_ref *cur;
2509 struct recorded_ref *tmp;
2510
2511 list_for_each_entry_safe(cur, tmp, head, list) {
2512 fs_path_free(sctx, cur->full_path);
2513 kfree(cur);
2514 }
2515 INIT_LIST_HEAD(head);
2516 }
2517
2518 static void free_recorded_refs(struct send_ctx *sctx)
2519 {
2520 __free_recorded_refs(sctx, &sctx->new_refs);
2521 __free_recorded_refs(sctx, &sctx->deleted_refs);
2522 }
2523
2524 /*
2525 * Renames/moves a file/dir to it's orphan name. Used when the first
2526 * ref of an unprocessed inode gets overwritten and for all non empty
2527 * directories.
2528 */
2529 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2530 struct fs_path *path)
2531 {
2532 int ret;
2533 struct fs_path *orphan;
2534
2535 orphan = fs_path_alloc(sctx);
2536 if (!orphan)
2537 return -ENOMEM;
2538
2539 ret = gen_unique_name(sctx, ino, gen, orphan);
2540 if (ret < 0)
2541 goto out;
2542
2543 ret = send_rename(sctx, path, orphan);
2544
2545 out:
2546 fs_path_free(sctx, orphan);
2547 return ret;
2548 }
2549
2550 /*
2551 * Returns 1 if a directory can be removed at this point in time.
2552 * We check this by iterating all dir items and checking if the inode behind
2553 * the dir item was already processed.
2554 */
2555 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2556 {
2557 int ret = 0;
2558 struct btrfs_root *root = sctx->parent_root;
2559 struct btrfs_path *path;
2560 struct btrfs_key key;
2561 struct btrfs_key found_key;
2562 struct btrfs_key loc;
2563 struct btrfs_dir_item *di;
2564
2565 path = alloc_path_for_send();
2566 if (!path)
2567 return -ENOMEM;
2568
2569 key.objectid = dir;
2570 key.type = BTRFS_DIR_INDEX_KEY;
2571 key.offset = 0;
2572
2573 while (1) {
2574 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2575 if (ret < 0)
2576 goto out;
2577 if (!ret) {
2578 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2579 path->slots[0]);
2580 }
2581 if (ret || found_key.objectid != key.objectid ||
2582 found_key.type != key.type) {
2583 break;
2584 }
2585
2586 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2587 struct btrfs_dir_item);
2588 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2589
2590 if (loc.objectid > send_progress) {
2591 ret = 0;
2592 goto out;
2593 }
2594
2595 btrfs_release_path(path);
2596 key.offset = found_key.offset + 1;
2597 }
2598
2599 ret = 1;
2600
2601 out:
2602 btrfs_free_path(path);
2603 return ret;
2604 }
2605
2606 /*
2607 * This does all the move/link/unlink/rmdir magic.
2608 */
2609 static int process_recorded_refs(struct send_ctx *sctx)
2610 {
2611 int ret = 0;
2612 struct recorded_ref *cur;
2613 struct recorded_ref *cur2;
2614 struct ulist *check_dirs = NULL;
2615 struct ulist_iterator uit;
2616 struct ulist_node *un;
2617 struct fs_path *valid_path = NULL;
2618 u64 ow_inode = 0;
2619 u64 ow_gen;
2620 int did_overwrite = 0;
2621 int is_orphan = 0;
2622
2623 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2624
2625 valid_path = fs_path_alloc(sctx);
2626 if (!valid_path) {
2627 ret = -ENOMEM;
2628 goto out;
2629 }
2630
2631 check_dirs = ulist_alloc(GFP_NOFS);
2632 if (!check_dirs) {
2633 ret = -ENOMEM;
2634 goto out;
2635 }
2636
2637 /*
2638 * First, check if the first ref of the current inode was overwritten
2639 * before. If yes, we know that the current inode was already orphanized
2640 * and thus use the orphan name. If not, we can use get_cur_path to
2641 * get the path of the first ref as it would like while receiving at
2642 * this point in time.
2643 * New inodes are always orphan at the beginning, so force to use the
2644 * orphan name in this case.
2645 * The first ref is stored in valid_path and will be updated if it
2646 * gets moved around.
2647 */
2648 if (!sctx->cur_inode_new) {
2649 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2650 sctx->cur_inode_gen);
2651 if (ret < 0)
2652 goto out;
2653 if (ret)
2654 did_overwrite = 1;
2655 }
2656 if (sctx->cur_inode_new || did_overwrite) {
2657 ret = gen_unique_name(sctx, sctx->cur_ino,
2658 sctx->cur_inode_gen, valid_path);
2659 if (ret < 0)
2660 goto out;
2661 is_orphan = 1;
2662 } else {
2663 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2664 valid_path);
2665 if (ret < 0)
2666 goto out;
2667 }
2668
2669 list_for_each_entry(cur, &sctx->new_refs, list) {
2670 /*
2671 * We may have refs where the parent directory does not exist
2672 * yet. This happens if the parent directories inum is higher
2673 * the the current inum. To handle this case, we create the
2674 * parent directory out of order. But we need to check if this
2675 * did already happen before due to other refs in the same dir.
2676 */
2677 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
2678 if (ret < 0)
2679 goto out;
2680 if (ret == inode_state_will_create) {
2681 ret = 0;
2682 /*
2683 * First check if any of the current inodes refs did
2684 * already create the dir.
2685 */
2686 list_for_each_entry(cur2, &sctx->new_refs, list) {
2687 if (cur == cur2)
2688 break;
2689 if (cur2->dir == cur->dir) {
2690 ret = 1;
2691 break;
2692 }
2693 }
2694
2695 /*
2696 * If that did not happen, check if a previous inode
2697 * did already create the dir.
2698 */
2699 if (!ret)
2700 ret = did_create_dir(sctx, cur->dir);
2701 if (ret < 0)
2702 goto out;
2703 if (!ret) {
2704 ret = send_create_inode(sctx, cur->dir);
2705 if (ret < 0)
2706 goto out;
2707 }
2708 }
2709
2710 /*
2711 * Check if this new ref would overwrite the first ref of
2712 * another unprocessed inode. If yes, orphanize the
2713 * overwritten inode. If we find an overwritten ref that is
2714 * not the first ref, simply unlink it.
2715 */
2716 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2717 cur->name, cur->name_len,
2718 &ow_inode, &ow_gen);
2719 if (ret < 0)
2720 goto out;
2721 if (ret) {
2722 ret = is_first_ref(sctx, sctx->parent_root,
2723 ow_inode, cur->dir, cur->name,
2724 cur->name_len);
2725 if (ret < 0)
2726 goto out;
2727 if (ret) {
2728 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2729 cur->full_path);
2730 if (ret < 0)
2731 goto out;
2732 } else {
2733 ret = send_unlink(sctx, cur->full_path);
2734 if (ret < 0)
2735 goto out;
2736 }
2737 }
2738
2739 /*
2740 * link/move the ref to the new place. If we have an orphan
2741 * inode, move it and update valid_path. If not, link or move
2742 * it depending on the inode mode.
2743 */
2744 if (is_orphan) {
2745 ret = send_rename(sctx, valid_path, cur->full_path);
2746 if (ret < 0)
2747 goto out;
2748 is_orphan = 0;
2749 ret = fs_path_copy(valid_path, cur->full_path);
2750 if (ret < 0)
2751 goto out;
2752 } else {
2753 if (S_ISDIR(sctx->cur_inode_mode)) {
2754 /*
2755 * Dirs can't be linked, so move it. For moved
2756 * dirs, we always have one new and one deleted
2757 * ref. The deleted ref is ignored later.
2758 */
2759 ret = send_rename(sctx, valid_path,
2760 cur->full_path);
2761 if (ret < 0)
2762 goto out;
2763 ret = fs_path_copy(valid_path, cur->full_path);
2764 if (ret < 0)
2765 goto out;
2766 } else {
2767 ret = send_link(sctx, cur->full_path,
2768 valid_path);
2769 if (ret < 0)
2770 goto out;
2771 }
2772 }
2773 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2774 GFP_NOFS);
2775 if (ret < 0)
2776 goto out;
2777 }
2778
2779 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2780 /*
2781 * Check if we can already rmdir the directory. If not,
2782 * orphanize it. For every dir item inside that gets deleted
2783 * later, we do this check again and rmdir it then if possible.
2784 * See the use of check_dirs for more details.
2785 */
2786 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2787 if (ret < 0)
2788 goto out;
2789 if (ret) {
2790 ret = send_rmdir(sctx, valid_path);
2791 if (ret < 0)
2792 goto out;
2793 } else if (!is_orphan) {
2794 ret = orphanize_inode(sctx, sctx->cur_ino,
2795 sctx->cur_inode_gen, valid_path);
2796 if (ret < 0)
2797 goto out;
2798 is_orphan = 1;
2799 }
2800
2801 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2802 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2803 GFP_NOFS);
2804 if (ret < 0)
2805 goto out;
2806 }
2807 } else if (S_ISDIR(sctx->cur_inode_mode) &&
2808 !list_empty(&sctx->deleted_refs)) {
2809 /*
2810 * We have a moved dir. Add the old parent to check_dirs
2811 */
2812 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
2813 list);
2814 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2815 GFP_NOFS);
2816 if (ret < 0)
2817 goto out;
2818 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2819 /*
2820 * We have a non dir inode. Go through all deleted refs and
2821 * unlink them if they were not already overwritten by other
2822 * inodes.
2823 */
2824 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2825 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2826 sctx->cur_ino, sctx->cur_inode_gen,
2827 cur->name, cur->name_len);
2828 if (ret < 0)
2829 goto out;
2830 if (!ret) {
2831 ret = send_unlink(sctx, cur->full_path);
2832 if (ret < 0)
2833 goto out;
2834 }
2835 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2836 GFP_NOFS);
2837 if (ret < 0)
2838 goto out;
2839 }
2840
2841 /*
2842 * If the inode is still orphan, unlink the orphan. This may
2843 * happen when a previous inode did overwrite the first ref
2844 * of this inode and no new refs were added for the current
2845 * inode.
2846 */
2847 if (is_orphan) {
2848 ret = send_unlink(sctx, valid_path);
2849 if (ret < 0)
2850 goto out;
2851 }
2852 }
2853
2854 /*
2855 * We did collect all parent dirs where cur_inode was once located. We
2856 * now go through all these dirs and check if they are pending for
2857 * deletion and if it's finally possible to perform the rmdir now.
2858 * We also update the inode stats of the parent dirs here.
2859 */
2860 ULIST_ITER_INIT(&uit);
2861 while ((un = ulist_next(check_dirs, &uit))) {
2862 if (un->val > sctx->cur_ino)
2863 continue;
2864
2865 ret = get_cur_inode_state(sctx, un->val, un->aux);
2866 if (ret < 0)
2867 goto out;
2868
2869 if (ret == inode_state_did_create ||
2870 ret == inode_state_no_change) {
2871 /* TODO delayed utimes */
2872 ret = send_utimes(sctx, un->val, un->aux);
2873 if (ret < 0)
2874 goto out;
2875 } else if (ret == inode_state_did_delete) {
2876 ret = can_rmdir(sctx, un->val, sctx->cur_ino);
2877 if (ret < 0)
2878 goto out;
2879 if (ret) {
2880 ret = get_cur_path(sctx, un->val, un->aux,
2881 valid_path);
2882 if (ret < 0)
2883 goto out;
2884 ret = send_rmdir(sctx, valid_path);
2885 if (ret < 0)
2886 goto out;
2887 }
2888 }
2889 }
2890
2891 /*
2892 * Current inode is now at it's new position, so we must increase
2893 * send_progress
2894 */
2895 sctx->send_progress = sctx->cur_ino + 1;
2896
2897 ret = 0;
2898
2899 out:
2900 free_recorded_refs(sctx);
2901 ulist_free(check_dirs);
2902 fs_path_free(sctx, valid_path);
2903 return ret;
2904 }
2905
2906 static int __record_new_ref(int num, u64 dir, int index,
2907 struct fs_path *name,
2908 void *ctx)
2909 {
2910 int ret = 0;
2911 struct send_ctx *sctx = ctx;
2912 struct fs_path *p;
2913 u64 gen;
2914
2915 p = fs_path_alloc(sctx);
2916 if (!p)
2917 return -ENOMEM;
2918
2919 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
2920 NULL, NULL);
2921 if (ret < 0)
2922 goto out;
2923
2924 ret = get_cur_path(sctx, dir, gen, p);
2925 if (ret < 0)
2926 goto out;
2927 ret = fs_path_add_path(p, name);
2928 if (ret < 0)
2929 goto out;
2930
2931 ret = record_ref(&sctx->new_refs, dir, gen, p);
2932
2933 out:
2934 if (ret)
2935 fs_path_free(sctx, p);
2936 return ret;
2937 }
2938
2939 static int __record_deleted_ref(int num, u64 dir, int index,
2940 struct fs_path *name,
2941 void *ctx)
2942 {
2943 int ret = 0;
2944 struct send_ctx *sctx = ctx;
2945 struct fs_path *p;
2946 u64 gen;
2947
2948 p = fs_path_alloc(sctx);
2949 if (!p)
2950 return -ENOMEM;
2951
2952 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
2953 NULL, NULL);
2954 if (ret < 0)
2955 goto out;
2956
2957 ret = get_cur_path(sctx, dir, gen, p);
2958 if (ret < 0)
2959 goto out;
2960 ret = fs_path_add_path(p, name);
2961 if (ret < 0)
2962 goto out;
2963
2964 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
2965
2966 out:
2967 if (ret)
2968 fs_path_free(sctx, p);
2969 return ret;
2970 }
2971
2972 static int record_new_ref(struct send_ctx *sctx)
2973 {
2974 int ret;
2975
2976 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
2977 sctx->cmp_key, 0, __record_new_ref, sctx);
2978 if (ret < 0)
2979 goto out;
2980 ret = 0;
2981
2982 out:
2983 return ret;
2984 }
2985
2986 static int record_deleted_ref(struct send_ctx *sctx)
2987 {
2988 int ret;
2989
2990 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
2991 sctx->cmp_key, 0, __record_deleted_ref, sctx);
2992 if (ret < 0)
2993 goto out;
2994 ret = 0;
2995
2996 out:
2997 return ret;
2998 }
2999
3000 struct find_ref_ctx {
3001 u64 dir;
3002 struct fs_path *name;
3003 int found_idx;
3004 };
3005
3006 static int __find_iref(int num, u64 dir, int index,
3007 struct fs_path *name,
3008 void *ctx_)
3009 {
3010 struct find_ref_ctx *ctx = ctx_;
3011
3012 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3013 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3014 ctx->found_idx = num;
3015 return 1;
3016 }
3017 return 0;
3018 }
3019
3020 static int find_iref(struct send_ctx *sctx,
3021 struct btrfs_root *root,
3022 struct btrfs_path *path,
3023 struct btrfs_key *key,
3024 u64 dir, struct fs_path *name)
3025 {
3026 int ret;
3027 struct find_ref_ctx ctx;
3028
3029 ctx.dir = dir;
3030 ctx.name = name;
3031 ctx.found_idx = -1;
3032
3033 ret = iterate_inode_ref(sctx, root, path, key, 0, __find_iref, &ctx);
3034 if (ret < 0)
3035 return ret;
3036
3037 if (ctx.found_idx == -1)
3038 return -ENOENT;
3039
3040 return ctx.found_idx;
3041 }
3042
3043 static int __record_changed_new_ref(int num, u64 dir, int index,
3044 struct fs_path *name,
3045 void *ctx)
3046 {
3047 int ret;
3048 struct send_ctx *sctx = ctx;
3049
3050 ret = find_iref(sctx, sctx->parent_root, sctx->right_path,
3051 sctx->cmp_key, dir, name);
3052 if (ret == -ENOENT)
3053 ret = __record_new_ref(num, dir, index, name, sctx);
3054 else if (ret > 0)
3055 ret = 0;
3056
3057 return ret;
3058 }
3059
3060 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3061 struct fs_path *name,
3062 void *ctx)
3063 {
3064 int ret;
3065 struct send_ctx *sctx = ctx;
3066
3067 ret = find_iref(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3068 dir, name);
3069 if (ret == -ENOENT)
3070 ret = __record_deleted_ref(num, dir, index, name, sctx);
3071 else if (ret > 0)
3072 ret = 0;
3073
3074 return ret;
3075 }
3076
3077 static int record_changed_ref(struct send_ctx *sctx)
3078 {
3079 int ret = 0;
3080
3081 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3082 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3083 if (ret < 0)
3084 goto out;
3085 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3086 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3087 if (ret < 0)
3088 goto out;
3089 ret = 0;
3090
3091 out:
3092 return ret;
3093 }
3094
3095 /*
3096 * Record and process all refs at once. Needed when an inode changes the
3097 * generation number, which means that it was deleted and recreated.
3098 */
3099 static int process_all_refs(struct send_ctx *sctx,
3100 enum btrfs_compare_tree_result cmd)
3101 {
3102 int ret;
3103 struct btrfs_root *root;
3104 struct btrfs_path *path;
3105 struct btrfs_key key;
3106 struct btrfs_key found_key;
3107 struct extent_buffer *eb;
3108 int slot;
3109 iterate_inode_ref_t cb;
3110
3111 path = alloc_path_for_send();
3112 if (!path)
3113 return -ENOMEM;
3114
3115 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3116 root = sctx->send_root;
3117 cb = __record_new_ref;
3118 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3119 root = sctx->parent_root;
3120 cb = __record_deleted_ref;
3121 } else {
3122 BUG();
3123 }
3124
3125 key.objectid = sctx->cmp_key->objectid;
3126 key.type = BTRFS_INODE_REF_KEY;
3127 key.offset = 0;
3128 while (1) {
3129 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3130 if (ret < 0) {
3131 btrfs_release_path(path);
3132 goto out;
3133 }
3134 if (ret) {
3135 btrfs_release_path(path);
3136 break;
3137 }
3138
3139 eb = path->nodes[0];
3140 slot = path->slots[0];
3141 btrfs_item_key_to_cpu(eb, &found_key, slot);
3142
3143 if (found_key.objectid != key.objectid ||
3144 found_key.type != key.type) {
3145 btrfs_release_path(path);
3146 break;
3147 }
3148
3149 ret = iterate_inode_ref(sctx, sctx->parent_root, path,
3150 &found_key, 0, cb, sctx);
3151 btrfs_release_path(path);
3152 if (ret < 0)
3153 goto out;
3154
3155 key.offset = found_key.offset + 1;
3156 }
3157
3158 ret = process_recorded_refs(sctx);
3159
3160 out:
3161 btrfs_free_path(path);
3162 return ret;
3163 }
3164
3165 static int send_set_xattr(struct send_ctx *sctx,
3166 struct fs_path *path,
3167 const char *name, int name_len,
3168 const char *data, int data_len)
3169 {
3170 int ret = 0;
3171
3172 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3173 if (ret < 0)
3174 goto out;
3175
3176 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3177 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3178 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3179
3180 ret = send_cmd(sctx);
3181
3182 tlv_put_failure:
3183 out:
3184 return ret;
3185 }
3186
3187 static int send_remove_xattr(struct send_ctx *sctx,
3188 struct fs_path *path,
3189 const char *name, int name_len)
3190 {
3191 int ret = 0;
3192
3193 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3194 if (ret < 0)
3195 goto out;
3196
3197 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3198 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3199
3200 ret = send_cmd(sctx);
3201
3202 tlv_put_failure:
3203 out:
3204 return ret;
3205 }
3206
3207 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3208 const char *name, int name_len,
3209 const char *data, int data_len,
3210 u8 type, void *ctx)
3211 {
3212 int ret;
3213 struct send_ctx *sctx = ctx;
3214 struct fs_path *p;
3215 posix_acl_xattr_header dummy_acl;
3216
3217 p = fs_path_alloc(sctx);
3218 if (!p)
3219 return -ENOMEM;
3220
3221 /*
3222 * This hack is needed because empty acl's are stored as zero byte
3223 * data in xattrs. Problem with that is, that receiving these zero byte
3224 * acl's will fail later. To fix this, we send a dummy acl list that
3225 * only contains the version number and no entries.
3226 */
3227 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3228 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3229 if (data_len == 0) {
3230 dummy_acl.a_version =
3231 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3232 data = (char *)&dummy_acl;
3233 data_len = sizeof(dummy_acl);
3234 }
3235 }
3236
3237 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3238 if (ret < 0)
3239 goto out;
3240
3241 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3242
3243 out:
3244 fs_path_free(sctx, p);
3245 return ret;
3246 }
3247
3248 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3249 const char *name, int name_len,
3250 const char *data, int data_len,
3251 u8 type, void *ctx)
3252 {
3253 int ret;
3254 struct send_ctx *sctx = ctx;
3255 struct fs_path *p;
3256
3257 p = fs_path_alloc(sctx);
3258 if (!p)
3259 return -ENOMEM;
3260
3261 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3262 if (ret < 0)
3263 goto out;
3264
3265 ret = send_remove_xattr(sctx, p, name, name_len);
3266
3267 out:
3268 fs_path_free(sctx, p);
3269 return ret;
3270 }
3271
3272 static int process_new_xattr(struct send_ctx *sctx)
3273 {
3274 int ret = 0;
3275
3276 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3277 sctx->cmp_key, __process_new_xattr, sctx);
3278
3279 return ret;
3280 }
3281
3282 static int process_deleted_xattr(struct send_ctx *sctx)
3283 {
3284 int ret;
3285
3286 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3287 sctx->cmp_key, __process_deleted_xattr, sctx);
3288
3289 return ret;
3290 }
3291
3292 struct find_xattr_ctx {
3293 const char *name;
3294 int name_len;
3295 int found_idx;
3296 char *found_data;
3297 int found_data_len;
3298 };
3299
3300 static int __find_xattr(int num, struct btrfs_key *di_key,
3301 const char *name, int name_len,
3302 const char *data, int data_len,
3303 u8 type, void *vctx)
3304 {
3305 struct find_xattr_ctx *ctx = vctx;
3306
3307 if (name_len == ctx->name_len &&
3308 strncmp(name, ctx->name, name_len) == 0) {
3309 ctx->found_idx = num;
3310 ctx->found_data_len = data_len;
3311 ctx->found_data = kmalloc(data_len, GFP_NOFS);
3312 if (!ctx->found_data)
3313 return -ENOMEM;
3314 memcpy(ctx->found_data, data, data_len);
3315 return 1;
3316 }
3317 return 0;
3318 }
3319
3320 static int find_xattr(struct send_ctx *sctx,
3321 struct btrfs_root *root,
3322 struct btrfs_path *path,
3323 struct btrfs_key *key,
3324 const char *name, int name_len,
3325 char **data, int *data_len)
3326 {
3327 int ret;
3328 struct find_xattr_ctx ctx;
3329
3330 ctx.name = name;
3331 ctx.name_len = name_len;
3332 ctx.found_idx = -1;
3333 ctx.found_data = NULL;
3334 ctx.found_data_len = 0;
3335
3336 ret = iterate_dir_item(sctx, root, path, key, __find_xattr, &ctx);
3337 if (ret < 0)
3338 return ret;
3339
3340 if (ctx.found_idx == -1)
3341 return -ENOENT;
3342 if (data) {
3343 *data = ctx.found_data;
3344 *data_len = ctx.found_data_len;
3345 } else {
3346 kfree(ctx.found_data);
3347 }
3348 return ctx.found_idx;
3349 }
3350
3351
3352 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3353 const char *name, int name_len,
3354 const char *data, int data_len,
3355 u8 type, void *ctx)
3356 {
3357 int ret;
3358 struct send_ctx *sctx = ctx;
3359 char *found_data = NULL;
3360 int found_data_len = 0;
3361 struct fs_path *p = NULL;
3362
3363 ret = find_xattr(sctx, sctx->parent_root, sctx->right_path,
3364 sctx->cmp_key, name, name_len, &found_data,
3365 &found_data_len);
3366 if (ret == -ENOENT) {
3367 ret = __process_new_xattr(num, di_key, name, name_len, data,
3368 data_len, type, ctx);
3369 } else if (ret >= 0) {
3370 if (data_len != found_data_len ||
3371 memcmp(data, found_data, data_len)) {
3372 ret = __process_new_xattr(num, di_key, name, name_len,
3373 data, data_len, type, ctx);
3374 } else {
3375 ret = 0;
3376 }
3377 }
3378
3379 kfree(found_data);
3380 fs_path_free(sctx, p);
3381 return ret;
3382 }
3383
3384 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3385 const char *name, int name_len,
3386 const char *data, int data_len,
3387 u8 type, void *ctx)
3388 {
3389 int ret;
3390 struct send_ctx *sctx = ctx;
3391
3392 ret = find_xattr(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3393 name, name_len, NULL, NULL);
3394 if (ret == -ENOENT)
3395 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3396 data_len, type, ctx);
3397 else if (ret >= 0)
3398 ret = 0;
3399
3400 return ret;
3401 }
3402
3403 static int process_changed_xattr(struct send_ctx *sctx)
3404 {
3405 int ret = 0;
3406
3407 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3408 sctx->cmp_key, __process_changed_new_xattr, sctx);
3409 if (ret < 0)
3410 goto out;
3411 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3412 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3413
3414 out:
3415 return ret;
3416 }
3417
3418 static int process_all_new_xattrs(struct send_ctx *sctx)
3419 {
3420 int ret;
3421 struct btrfs_root *root;
3422 struct btrfs_path *path;
3423 struct btrfs_key key;
3424 struct btrfs_key found_key;
3425 struct extent_buffer *eb;
3426 int slot;
3427
3428 path = alloc_path_for_send();
3429 if (!path)
3430 return -ENOMEM;
3431
3432 root = sctx->send_root;
3433
3434 key.objectid = sctx->cmp_key->objectid;
3435 key.type = BTRFS_XATTR_ITEM_KEY;
3436 key.offset = 0;
3437 while (1) {
3438 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3439 if (ret < 0)
3440 goto out;
3441 if (ret) {
3442 ret = 0;
3443 goto out;
3444 }
3445
3446 eb = path->nodes[0];
3447 slot = path->slots[0];
3448 btrfs_item_key_to_cpu(eb, &found_key, slot);
3449
3450 if (found_key.objectid != key.objectid ||
3451 found_key.type != key.type) {
3452 ret = 0;
3453 goto out;
3454 }
3455
3456 ret = iterate_dir_item(sctx, root, path, &found_key,
3457 __process_new_xattr, sctx);
3458 if (ret < 0)
3459 goto out;
3460
3461 btrfs_release_path(path);
3462 key.offset = found_key.offset + 1;
3463 }
3464
3465 out:
3466 btrfs_free_path(path);
3467 return ret;
3468 }
3469
3470 /*
3471 * Read some bytes from the current inode/file and send a write command to
3472 * user space.
3473 */
3474 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3475 {
3476 int ret = 0;
3477 struct fs_path *p;
3478 loff_t pos = offset;
3479 int readed = 0;
3480 mm_segment_t old_fs;
3481
3482 p = fs_path_alloc(sctx);
3483 if (!p)
3484 return -ENOMEM;
3485
3486 /*
3487 * vfs normally only accepts user space buffers for security reasons.
3488 * we only read from the file and also only provide the read_buf buffer
3489 * to vfs. As this buffer does not come from a user space call, it's
3490 * ok to temporary allow kernel space buffers.
3491 */
3492 old_fs = get_fs();
3493 set_fs(KERNEL_DS);
3494
3495 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3496
3497 ret = open_cur_inode_file(sctx);
3498 if (ret < 0)
3499 goto out;
3500
3501 ret = vfs_read(sctx->cur_inode_filp, sctx->read_buf, len, &pos);
3502 if (ret < 0)
3503 goto out;
3504 readed = ret;
3505 if (!readed)
3506 goto out;
3507
3508 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3509 if (ret < 0)
3510 goto out;
3511
3512 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3513 if (ret < 0)
3514 goto out;
3515
3516 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3517 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3518 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, readed);
3519
3520 ret = send_cmd(sctx);
3521
3522 tlv_put_failure:
3523 out:
3524 fs_path_free(sctx, p);
3525 set_fs(old_fs);
3526 if (ret < 0)
3527 return ret;
3528 return readed;
3529 }
3530
3531 /*
3532 * Send a clone command to user space.
3533 */
3534 static int send_clone(struct send_ctx *sctx,
3535 u64 offset, u32 len,
3536 struct clone_root *clone_root)
3537 {
3538 int ret = 0;
3539 struct btrfs_root *clone_root2 = clone_root->root;
3540 struct fs_path *p;
3541 u64 gen;
3542
3543 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
3544 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
3545 clone_root->root->objectid, clone_root->ino,
3546 clone_root->offset);
3547
3548 p = fs_path_alloc(sctx);
3549 if (!p)
3550 return -ENOMEM;
3551
3552 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
3553 if (ret < 0)
3554 goto out;
3555
3556 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3557 if (ret < 0)
3558 goto out;
3559
3560 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3561 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
3562 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3563
3564 if (clone_root2 == sctx->send_root) {
3565 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
3566 &gen, NULL, NULL, NULL, NULL);
3567 if (ret < 0)
3568 goto out;
3569 ret = get_cur_path(sctx, clone_root->ino, gen, p);
3570 } else {
3571 ret = get_inode_path(sctx, clone_root2, clone_root->ino, p);
3572 }
3573 if (ret < 0)
3574 goto out;
3575
3576 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
3577 clone_root2->root_item.uuid);
3578 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
3579 clone_root2->root_item.ctransid);
3580 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3581 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3582 clone_root->offset);
3583
3584 ret = send_cmd(sctx);
3585
3586 tlv_put_failure:
3587 out:
3588 fs_path_free(sctx, p);
3589 return ret;
3590 }
3591
3592 static int send_write_or_clone(struct send_ctx *sctx,
3593 struct btrfs_path *path,
3594 struct btrfs_key *key,
3595 struct clone_root *clone_root)
3596 {
3597 int ret = 0;
3598 struct btrfs_file_extent_item *ei;
3599 u64 offset = key->offset;
3600 u64 pos = 0;
3601 u64 len;
3602 u32 l;
3603 u8 type;
3604
3605 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3606 struct btrfs_file_extent_item);
3607 type = btrfs_file_extent_type(path->nodes[0], ei);
3608 if (type == BTRFS_FILE_EXTENT_INLINE)
3609 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
3610 else
3611 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3612
3613 if (offset + len > sctx->cur_inode_size)
3614 len = sctx->cur_inode_size - offset;
3615 if (len == 0) {
3616 ret = 0;
3617 goto out;
3618 }
3619
3620 if (!clone_root) {
3621 while (pos < len) {
3622 l = len - pos;
3623 if (l > BTRFS_SEND_READ_SIZE)
3624 l = BTRFS_SEND_READ_SIZE;
3625 ret = send_write(sctx, pos + offset, l);
3626 if (ret < 0)
3627 goto out;
3628 if (!ret)
3629 break;
3630 pos += ret;
3631 }
3632 ret = 0;
3633 } else {
3634 ret = send_clone(sctx, offset, len, clone_root);
3635 }
3636
3637 out:
3638 return ret;
3639 }
3640
3641 static int is_extent_unchanged(struct send_ctx *sctx,
3642 struct btrfs_path *left_path,
3643 struct btrfs_key *ekey)
3644 {
3645 int ret = 0;
3646 struct btrfs_key key;
3647 struct btrfs_path *path = NULL;
3648 struct extent_buffer *eb;
3649 int slot;
3650 struct btrfs_key found_key;
3651 struct btrfs_file_extent_item *ei;
3652 u64 left_disknr;
3653 u64 right_disknr;
3654 u64 left_offset;
3655 u64 right_offset;
3656 u64 left_offset_fixed;
3657 u64 left_len;
3658 u64 right_len;
3659 u8 left_type;
3660 u8 right_type;
3661
3662 path = alloc_path_for_send();
3663 if (!path)
3664 return -ENOMEM;
3665
3666 eb = left_path->nodes[0];
3667 slot = left_path->slots[0];
3668
3669 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3670 left_type = btrfs_file_extent_type(eb, ei);
3671 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3672 left_len = btrfs_file_extent_num_bytes(eb, ei);
3673 left_offset = btrfs_file_extent_offset(eb, ei);
3674
3675 if (left_type != BTRFS_FILE_EXTENT_REG) {
3676 ret = 0;
3677 goto out;
3678 }
3679
3680 /*
3681 * Following comments will refer to these graphics. L is the left
3682 * extents which we are checking at the moment. 1-8 are the right
3683 * extents that we iterate.
3684 *
3685 * |-----L-----|
3686 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3687 *
3688 * |-----L-----|
3689 * |--1--|-2b-|...(same as above)
3690 *
3691 * Alternative situation. Happens on files where extents got split.
3692 * |-----L-----|
3693 * |-----------7-----------|-6-|
3694 *
3695 * Alternative situation. Happens on files which got larger.
3696 * |-----L-----|
3697 * |-8-|
3698 * Nothing follows after 8.
3699 */
3700
3701 key.objectid = ekey->objectid;
3702 key.type = BTRFS_EXTENT_DATA_KEY;
3703 key.offset = ekey->offset;
3704 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3705 if (ret < 0)
3706 goto out;
3707 if (ret) {
3708 ret = 0;
3709 goto out;
3710 }
3711
3712 /*
3713 * Handle special case where the right side has no extents at all.
3714 */
3715 eb = path->nodes[0];
3716 slot = path->slots[0];
3717 btrfs_item_key_to_cpu(eb, &found_key, slot);
3718 if (found_key.objectid != key.objectid ||
3719 found_key.type != key.type) {
3720 ret = 0;
3721 goto out;
3722 }
3723
3724 /*
3725 * We're now on 2a, 2b or 7.
3726 */
3727 key = found_key;
3728 while (key.offset < ekey->offset + left_len) {
3729 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3730 right_type = btrfs_file_extent_type(eb, ei);
3731 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3732 right_len = btrfs_file_extent_num_bytes(eb, ei);
3733 right_offset = btrfs_file_extent_offset(eb, ei);
3734
3735 if (right_type != BTRFS_FILE_EXTENT_REG) {
3736 ret = 0;
3737 goto out;
3738 }
3739
3740 /*
3741 * Are we at extent 8? If yes, we know the extent is changed.
3742 * This may only happen on the first iteration.
3743 */
3744 if (found_key.offset + right_len < ekey->offset) {
3745 ret = 0;
3746 goto out;
3747 }
3748
3749 left_offset_fixed = left_offset;
3750 if (key.offset < ekey->offset) {
3751 /* Fix the right offset for 2a and 7. */
3752 right_offset += ekey->offset - key.offset;
3753 } else {
3754 /* Fix the left offset for all behind 2a and 2b */
3755 left_offset_fixed += key.offset - ekey->offset;
3756 }
3757
3758 /*
3759 * Check if we have the same extent.
3760 */
3761 if (left_disknr + left_offset_fixed !=
3762 right_disknr + right_offset) {
3763 ret = 0;
3764 goto out;
3765 }
3766
3767 /*
3768 * Go to the next extent.
3769 */
3770 ret = btrfs_next_item(sctx->parent_root, path);
3771 if (ret < 0)
3772 goto out;
3773 if (!ret) {
3774 eb = path->nodes[0];
3775 slot = path->slots[0];
3776 btrfs_item_key_to_cpu(eb, &found_key, slot);
3777 }
3778 if (ret || found_key.objectid != key.objectid ||
3779 found_key.type != key.type) {
3780 key.offset += right_len;
3781 break;
3782 } else {
3783 if (found_key.offset != key.offset + right_len) {
3784 /* Should really not happen */
3785 ret = -EIO;
3786 goto out;
3787 }
3788 }
3789 key = found_key;
3790 }
3791
3792 /*
3793 * We're now behind the left extent (treat as unchanged) or at the end
3794 * of the right side (treat as changed).
3795 */
3796 if (key.offset >= ekey->offset + left_len)
3797 ret = 1;
3798 else
3799 ret = 0;
3800
3801
3802 out:
3803 btrfs_free_path(path);
3804 return ret;
3805 }
3806
3807 static int process_extent(struct send_ctx *sctx,
3808 struct btrfs_path *path,
3809 struct btrfs_key *key)
3810 {
3811 int ret = 0;
3812 struct clone_root *found_clone = NULL;
3813
3814 if (S_ISLNK(sctx->cur_inode_mode))
3815 return 0;
3816
3817 if (sctx->parent_root && !sctx->cur_inode_new) {
3818 ret = is_extent_unchanged(sctx, path, key);
3819 if (ret < 0)
3820 goto out;
3821 if (ret) {
3822 ret = 0;
3823 goto out;
3824 }
3825 }
3826
3827 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
3828 sctx->cur_inode_size, &found_clone);
3829 if (ret != -ENOENT && ret < 0)
3830 goto out;
3831
3832 ret = send_write_or_clone(sctx, path, key, found_clone);
3833
3834 out:
3835 return ret;
3836 }
3837
3838 static int process_all_extents(struct send_ctx *sctx)
3839 {
3840 int ret;
3841 struct btrfs_root *root;
3842 struct btrfs_path *path;
3843 struct btrfs_key key;
3844 struct btrfs_key found_key;
3845 struct extent_buffer *eb;
3846 int slot;
3847
3848 root = sctx->send_root;
3849 path = alloc_path_for_send();
3850 if (!path)
3851 return -ENOMEM;
3852
3853 key.objectid = sctx->cmp_key->objectid;
3854 key.type = BTRFS_EXTENT_DATA_KEY;
3855 key.offset = 0;
3856 while (1) {
3857 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3858 if (ret < 0)
3859 goto out;
3860 if (ret) {
3861 ret = 0;
3862 goto out;
3863 }
3864
3865 eb = path->nodes[0];
3866 slot = path->slots[0];
3867 btrfs_item_key_to_cpu(eb, &found_key, slot);
3868
3869 if (found_key.objectid != key.objectid ||
3870 found_key.type != key.type) {
3871 ret = 0;
3872 goto out;
3873 }
3874
3875 ret = process_extent(sctx, path, &found_key);
3876 if (ret < 0)
3877 goto out;
3878
3879 btrfs_release_path(path);
3880 key.offset = found_key.offset + 1;
3881 }
3882
3883 out:
3884 btrfs_free_path(path);
3885 return ret;
3886 }
3887
3888 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
3889 {
3890 int ret = 0;
3891
3892 if (sctx->cur_ino == 0)
3893 goto out;
3894 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
3895 sctx->cmp_key->type <= BTRFS_INODE_REF_KEY)
3896 goto out;
3897 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
3898 goto out;
3899
3900 ret = process_recorded_refs(sctx);
3901
3902 out:
3903 return ret;
3904 }
3905
3906 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
3907 {
3908 int ret = 0;
3909 u64 left_mode;
3910 u64 left_uid;
3911 u64 left_gid;
3912 u64 right_mode;
3913 u64 right_uid;
3914 u64 right_gid;
3915 int need_chmod = 0;
3916 int need_chown = 0;
3917
3918 ret = process_recorded_refs_if_needed(sctx, at_end);
3919 if (ret < 0)
3920 goto out;
3921
3922 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
3923 goto out;
3924 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
3925 goto out;
3926
3927 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
3928 &left_mode, &left_uid, &left_gid, NULL);
3929 if (ret < 0)
3930 goto out;
3931
3932 if (!S_ISLNK(sctx->cur_inode_mode)) {
3933 if (!sctx->parent_root || sctx->cur_inode_new) {
3934 need_chmod = 1;
3935 need_chown = 1;
3936 } else {
3937 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
3938 NULL, NULL, &right_mode, &right_uid,
3939 &right_gid, NULL);
3940 if (ret < 0)
3941 goto out;
3942
3943 if (left_uid != right_uid || left_gid != right_gid)
3944 need_chown = 1;
3945 if (left_mode != right_mode)
3946 need_chmod = 1;
3947 }
3948 }
3949
3950 if (S_ISREG(sctx->cur_inode_mode)) {
3951 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3952 sctx->cur_inode_size);
3953 if (ret < 0)
3954 goto out;
3955 }
3956
3957 if (need_chown) {
3958 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3959 left_uid, left_gid);
3960 if (ret < 0)
3961 goto out;
3962 }
3963 if (need_chmod) {
3964 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3965 left_mode);
3966 if (ret < 0)
3967 goto out;
3968 }
3969
3970 /*
3971 * Need to send that every time, no matter if it actually changed
3972 * between the two trees as we have done changes to the inode before.
3973 */
3974 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
3975 if (ret < 0)
3976 goto out;
3977
3978 out:
3979 return ret;
3980 }
3981
3982 static int changed_inode(struct send_ctx *sctx,
3983 enum btrfs_compare_tree_result result)
3984 {
3985 int ret = 0;
3986 struct btrfs_key *key = sctx->cmp_key;
3987 struct btrfs_inode_item *left_ii = NULL;
3988 struct btrfs_inode_item *right_ii = NULL;
3989 u64 left_gen = 0;
3990 u64 right_gen = 0;
3991
3992 ret = close_cur_inode_file(sctx);
3993 if (ret < 0)
3994 goto out;
3995
3996 sctx->cur_ino = key->objectid;
3997 sctx->cur_inode_new_gen = 0;
3998 sctx->send_progress = sctx->cur_ino;
3999
4000 if (result == BTRFS_COMPARE_TREE_NEW ||
4001 result == BTRFS_COMPARE_TREE_CHANGED) {
4002 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4003 sctx->left_path->slots[0],
4004 struct btrfs_inode_item);
4005 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4006 left_ii);
4007 } else {
4008 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4009 sctx->right_path->slots[0],
4010 struct btrfs_inode_item);
4011 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4012 right_ii);
4013 }
4014 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4015 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4016 sctx->right_path->slots[0],
4017 struct btrfs_inode_item);
4018
4019 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4020 right_ii);
4021 if (left_gen != right_gen)
4022 sctx->cur_inode_new_gen = 1;
4023 }
4024
4025 if (result == BTRFS_COMPARE_TREE_NEW) {
4026 sctx->cur_inode_gen = left_gen;
4027 sctx->cur_inode_new = 1;
4028 sctx->cur_inode_deleted = 0;
4029 sctx->cur_inode_size = btrfs_inode_size(
4030 sctx->left_path->nodes[0], left_ii);
4031 sctx->cur_inode_mode = btrfs_inode_mode(
4032 sctx->left_path->nodes[0], left_ii);
4033 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4034 ret = send_create_inode_if_needed(sctx);
4035 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4036 sctx->cur_inode_gen = right_gen;
4037 sctx->cur_inode_new = 0;
4038 sctx->cur_inode_deleted = 1;
4039 sctx->cur_inode_size = btrfs_inode_size(
4040 sctx->right_path->nodes[0], right_ii);
4041 sctx->cur_inode_mode = btrfs_inode_mode(
4042 sctx->right_path->nodes[0], right_ii);
4043 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4044 if (sctx->cur_inode_new_gen) {
4045 sctx->cur_inode_gen = right_gen;
4046 sctx->cur_inode_new = 0;
4047 sctx->cur_inode_deleted = 1;
4048 sctx->cur_inode_size = btrfs_inode_size(
4049 sctx->right_path->nodes[0], right_ii);
4050 sctx->cur_inode_mode = btrfs_inode_mode(
4051 sctx->right_path->nodes[0], right_ii);
4052 ret = process_all_refs(sctx,
4053 BTRFS_COMPARE_TREE_DELETED);
4054 if (ret < 0)
4055 goto out;
4056
4057 sctx->cur_inode_gen = left_gen;
4058 sctx->cur_inode_new = 1;
4059 sctx->cur_inode_deleted = 0;
4060 sctx->cur_inode_size = btrfs_inode_size(
4061 sctx->left_path->nodes[0], left_ii);
4062 sctx->cur_inode_mode = btrfs_inode_mode(
4063 sctx->left_path->nodes[0], left_ii);
4064 ret = send_create_inode_if_needed(sctx);
4065 if (ret < 0)
4066 goto out;
4067
4068 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4069 if (ret < 0)
4070 goto out;
4071 ret = process_all_extents(sctx);
4072 if (ret < 0)
4073 goto out;
4074 ret = process_all_new_xattrs(sctx);
4075 if (ret < 0)
4076 goto out;
4077 } else {
4078 sctx->cur_inode_gen = left_gen;
4079 sctx->cur_inode_new = 0;
4080 sctx->cur_inode_new_gen = 0;
4081 sctx->cur_inode_deleted = 0;
4082 sctx->cur_inode_size = btrfs_inode_size(
4083 sctx->left_path->nodes[0], left_ii);
4084 sctx->cur_inode_mode = btrfs_inode_mode(
4085 sctx->left_path->nodes[0], left_ii);
4086 }
4087 }
4088
4089 out:
4090 return ret;
4091 }
4092
4093 static int changed_ref(struct send_ctx *sctx,
4094 enum btrfs_compare_tree_result result)
4095 {
4096 int ret = 0;
4097
4098 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4099
4100 if (!sctx->cur_inode_new_gen &&
4101 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4102 if (result == BTRFS_COMPARE_TREE_NEW)
4103 ret = record_new_ref(sctx);
4104 else if (result == BTRFS_COMPARE_TREE_DELETED)
4105 ret = record_deleted_ref(sctx);
4106 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4107 ret = record_changed_ref(sctx);
4108 }
4109
4110 return ret;
4111 }
4112
4113 static int changed_xattr(struct send_ctx *sctx,
4114 enum btrfs_compare_tree_result result)
4115 {
4116 int ret = 0;
4117
4118 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4119
4120 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4121 if (result == BTRFS_COMPARE_TREE_NEW)
4122 ret = process_new_xattr(sctx);
4123 else if (result == BTRFS_COMPARE_TREE_DELETED)
4124 ret = process_deleted_xattr(sctx);
4125 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4126 ret = process_changed_xattr(sctx);
4127 }
4128
4129 return ret;
4130 }
4131
4132 static int changed_extent(struct send_ctx *sctx,
4133 enum btrfs_compare_tree_result result)
4134 {
4135 int ret = 0;
4136
4137 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4138
4139 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4140 if (result != BTRFS_COMPARE_TREE_DELETED)
4141 ret = process_extent(sctx, sctx->left_path,
4142 sctx->cmp_key);
4143 }
4144
4145 return ret;
4146 }
4147
4148
4149 static int changed_cb(struct btrfs_root *left_root,
4150 struct btrfs_root *right_root,
4151 struct btrfs_path *left_path,
4152 struct btrfs_path *right_path,
4153 struct btrfs_key *key,
4154 enum btrfs_compare_tree_result result,
4155 void *ctx)
4156 {
4157 int ret = 0;
4158 struct send_ctx *sctx = ctx;
4159
4160 sctx->left_path = left_path;
4161 sctx->right_path = right_path;
4162 sctx->cmp_key = key;
4163
4164 ret = finish_inode_if_needed(sctx, 0);
4165 if (ret < 0)
4166 goto out;
4167
4168 if (key->type == BTRFS_INODE_ITEM_KEY)
4169 ret = changed_inode(sctx, result);
4170 else if (key->type == BTRFS_INODE_REF_KEY)
4171 ret = changed_ref(sctx, result);
4172 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4173 ret = changed_xattr(sctx, result);
4174 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4175 ret = changed_extent(sctx, result);
4176
4177 out:
4178 return ret;
4179 }
4180
4181 static int full_send_tree(struct send_ctx *sctx)
4182 {
4183 int ret;
4184 struct btrfs_trans_handle *trans = NULL;
4185 struct btrfs_root *send_root = sctx->send_root;
4186 struct btrfs_key key;
4187 struct btrfs_key found_key;
4188 struct btrfs_path *path;
4189 struct extent_buffer *eb;
4190 int slot;
4191 u64 start_ctransid;
4192 u64 ctransid;
4193
4194 path = alloc_path_for_send();
4195 if (!path)
4196 return -ENOMEM;
4197
4198 spin_lock(&send_root->root_times_lock);
4199 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
4200 spin_unlock(&send_root->root_times_lock);
4201
4202 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4203 key.type = BTRFS_INODE_ITEM_KEY;
4204 key.offset = 0;
4205
4206 join_trans:
4207 /*
4208 * We need to make sure the transaction does not get committed
4209 * while we do anything on commit roots. Join a transaction to prevent
4210 * this.
4211 */
4212 trans = btrfs_join_transaction(send_root);
4213 if (IS_ERR(trans)) {
4214 ret = PTR_ERR(trans);
4215 trans = NULL;
4216 goto out;
4217 }
4218
4219 /*
4220 * Make sure the tree has not changed
4221 */
4222 spin_lock(&send_root->root_times_lock);
4223 ctransid = btrfs_root_ctransid(&send_root->root_item);
4224 spin_unlock(&send_root->root_times_lock);
4225
4226 if (ctransid != start_ctransid) {
4227 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4228 "send was modified in between. This is "
4229 "probably a bug.\n");
4230 ret = -EIO;
4231 goto out;
4232 }
4233
4234 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4235 if (ret < 0)
4236 goto out;
4237 if (ret)
4238 goto out_finish;
4239
4240 while (1) {
4241 /*
4242 * When someone want to commit while we iterate, end the
4243 * joined transaction and rejoin.
4244 */
4245 if (btrfs_should_end_transaction(trans, send_root)) {
4246 ret = btrfs_end_transaction(trans, send_root);
4247 trans = NULL;
4248 if (ret < 0)
4249 goto out;
4250 btrfs_release_path(path);
4251 goto join_trans;
4252 }
4253
4254 eb = path->nodes[0];
4255 slot = path->slots[0];
4256 btrfs_item_key_to_cpu(eb, &found_key, slot);
4257
4258 ret = changed_cb(send_root, NULL, path, NULL,
4259 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4260 if (ret < 0)
4261 goto out;
4262
4263 key.objectid = found_key.objectid;
4264 key.type = found_key.type;
4265 key.offset = found_key.offset + 1;
4266
4267 ret = btrfs_next_item(send_root, path);
4268 if (ret < 0)
4269 goto out;
4270 if (ret) {
4271 ret = 0;
4272 break;
4273 }
4274 }
4275
4276 out_finish:
4277 ret = finish_inode_if_needed(sctx, 1);
4278
4279 out:
4280 btrfs_free_path(path);
4281 if (trans) {
4282 if (!ret)
4283 ret = btrfs_end_transaction(trans, send_root);
4284 else
4285 btrfs_end_transaction(trans, send_root);
4286 }
4287 return ret;
4288 }
4289
4290 static int send_subvol(struct send_ctx *sctx)
4291 {
4292 int ret;
4293
4294 ret = send_header(sctx);
4295 if (ret < 0)
4296 goto out;
4297
4298 ret = send_subvol_begin(sctx);
4299 if (ret < 0)
4300 goto out;
4301
4302 if (sctx->parent_root) {
4303 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4304 changed_cb, sctx);
4305 if (ret < 0)
4306 goto out;
4307 ret = finish_inode_if_needed(sctx, 1);
4308 if (ret < 0)
4309 goto out;
4310 } else {
4311 ret = full_send_tree(sctx);
4312 if (ret < 0)
4313 goto out;
4314 }
4315
4316 out:
4317 if (!ret)
4318 ret = close_cur_inode_file(sctx);
4319 else
4320 close_cur_inode_file(sctx);
4321
4322 free_recorded_refs(sctx);
4323 return ret;
4324 }
4325
4326 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4327 {
4328 int ret = 0;
4329 struct btrfs_root *send_root;
4330 struct btrfs_root *clone_root;
4331 struct btrfs_fs_info *fs_info;
4332 struct btrfs_ioctl_send_args *arg = NULL;
4333 struct btrfs_key key;
4334 struct file *filp = NULL;
4335 struct send_ctx *sctx = NULL;
4336 u32 i;
4337 u64 *clone_sources_tmp = NULL;
4338
4339 if (!capable(CAP_SYS_ADMIN))
4340 return -EPERM;
4341
4342 send_root = BTRFS_I(fdentry(mnt_file)->d_inode)->root;
4343 fs_info = send_root->fs_info;
4344
4345 arg = memdup_user(arg_, sizeof(*arg));
4346 if (IS_ERR(arg)) {
4347 ret = PTR_ERR(arg);
4348 arg = NULL;
4349 goto out;
4350 }
4351
4352 if (!access_ok(VERIFY_READ, arg->clone_sources,
4353 sizeof(*arg->clone_sources *
4354 arg->clone_sources_count))) {
4355 ret = -EFAULT;
4356 goto out;
4357 }
4358
4359 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4360 if (!sctx) {
4361 ret = -ENOMEM;
4362 goto out;
4363 }
4364
4365 INIT_LIST_HEAD(&sctx->new_refs);
4366 INIT_LIST_HEAD(&sctx->deleted_refs);
4367 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4368 INIT_LIST_HEAD(&sctx->name_cache_list);
4369
4370 sctx->send_filp = fget(arg->send_fd);
4371 if (IS_ERR(sctx->send_filp)) {
4372 ret = PTR_ERR(sctx->send_filp);
4373 goto out;
4374 }
4375
4376 sctx->mnt = mnt_file->f_path.mnt;
4377
4378 sctx->send_root = send_root;
4379 sctx->clone_roots_cnt = arg->clone_sources_count;
4380
4381 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4382 sctx->send_buf = vmalloc(sctx->send_max_size);
4383 if (!sctx->send_buf) {
4384 ret = -ENOMEM;
4385 goto out;
4386 }
4387
4388 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4389 if (!sctx->read_buf) {
4390 ret = -ENOMEM;
4391 goto out;
4392 }
4393
4394 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4395 (arg->clone_sources_count + 1));
4396 if (!sctx->clone_roots) {
4397 ret = -ENOMEM;
4398 goto out;
4399 }
4400
4401 if (arg->clone_sources_count) {
4402 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4403 sizeof(*arg->clone_sources));
4404 if (!clone_sources_tmp) {
4405 ret = -ENOMEM;
4406 goto out;
4407 }
4408
4409 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4410 arg->clone_sources_count *
4411 sizeof(*arg->clone_sources));
4412 if (ret) {
4413 ret = -EFAULT;
4414 goto out;
4415 }
4416
4417 for (i = 0; i < arg->clone_sources_count; i++) {
4418 key.objectid = clone_sources_tmp[i];
4419 key.type = BTRFS_ROOT_ITEM_KEY;
4420 key.offset = (u64)-1;
4421 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
4422 if (!clone_root) {
4423 ret = -EINVAL;
4424 goto out;
4425 }
4426 if (IS_ERR(clone_root)) {
4427 ret = PTR_ERR(clone_root);
4428 goto out;
4429 }
4430 sctx->clone_roots[i].root = clone_root;
4431 }
4432 vfree(clone_sources_tmp);
4433 clone_sources_tmp = NULL;
4434 }
4435
4436 if (arg->parent_root) {
4437 key.objectid = arg->parent_root;
4438 key.type = BTRFS_ROOT_ITEM_KEY;
4439 key.offset = (u64)-1;
4440 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
4441 if (!sctx->parent_root) {
4442 ret = -EINVAL;
4443 goto out;
4444 }
4445 }
4446
4447 /*
4448 * Clones from send_root are allowed, but only if the clone source
4449 * is behind the current send position. This is checked while searching
4450 * for possible clone sources.
4451 */
4452 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4453
4454 /* We do a bsearch later */
4455 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4456 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4457 NULL);
4458
4459 ret = send_subvol(sctx);
4460 if (ret < 0)
4461 goto out;
4462
4463 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4464 if (ret < 0)
4465 goto out;
4466 ret = send_cmd(sctx);
4467 if (ret < 0)
4468 goto out;
4469
4470 out:
4471 if (filp)
4472 fput(filp);
4473 kfree(arg);
4474 vfree(clone_sources_tmp);
4475
4476 if (sctx) {
4477 if (sctx->send_filp)
4478 fput(sctx->send_filp);
4479
4480 vfree(sctx->clone_roots);
4481 vfree(sctx->send_buf);
4482 vfree(sctx->read_buf);
4483
4484 name_cache_free(sctx);
4485
4486 kfree(sctx);
4487 }
4488
4489 return ret;
4490 }
Linus' kernel tree