blob: 63a6152be04b4f4265b25e253eadb1d4d91b8969 [file] [log] [blame]
Alexander Block31db9f72012-07-25 23:19:24 +02001/*
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>
Stephen Rothwella1857eb2012-07-27 10:11:13 +100027#include <linux/vmalloc.h>
Andy Shevchenkoed848852013-08-21 10:32:13 +030028#include <linux/string.h>
Alexander Block31db9f72012-07-25 23:19:24 +020029
30#include "send.h"
31#include "backref.h"
Filipe David Borba Manana0b947af2014-01-29 21:06:04 +000032#include "hash.h"
Alexander Block31db9f72012-07-25 23:19:24 +020033#include "locking.h"
34#include "disk-io.h"
35#include "btrfs_inode.h"
36#include "transaction.h"
37
38static int g_verbose = 0;
39
40#define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
41
42/*
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
48 */
49struct fs_path {
50 union {
51 struct {
52 char *start;
53 char *end;
Alexander Block31db9f72012-07-25 23:19:24 +020054
55 char *buf;
David Sterba1f5a7ff2014-02-03 19:23:47 +010056 unsigned short buf_len:15;
57 unsigned short reversed:1;
Alexander Block31db9f72012-07-25 23:19:24 +020058 char inline_buf[];
59 };
David Sterbaace01052014-02-05 16:17:34 +010060 /*
61 * Average path length does not exceed 200 bytes, we'll have
62 * better packing in the slab and higher chance to satisfy
63 * a allocation later during send.
64 */
65 char pad[256];
Alexander Block31db9f72012-07-25 23:19:24 +020066 };
67};
68#define FS_PATH_INLINE_SIZE \
69 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
70
71
72/* reused for each extent */
73struct clone_root {
74 struct btrfs_root *root;
75 u64 ino;
76 u64 offset;
77
78 u64 found_refs;
79};
80
81#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
82#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
83
84struct send_ctx {
85 struct file *send_filp;
86 loff_t send_off;
87 char *send_buf;
88 u32 send_size;
89 u32 send_max_size;
90 u64 total_send_size;
91 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
Mark Fashehcb95e7b2013-02-04 20:54:57 +000092 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
Alexander Block31db9f72012-07-25 23:19:24 +020093
Alexander Block31db9f72012-07-25 23:19:24 +020094 struct btrfs_root *send_root;
95 struct btrfs_root *parent_root;
96 struct clone_root *clone_roots;
97 int clone_roots_cnt;
98
99 /* current state of the compare_tree call */
100 struct btrfs_path *left_path;
101 struct btrfs_path *right_path;
102 struct btrfs_key *cmp_key;
103
104 /*
105 * infos of the currently processed inode. In case of deleted inodes,
106 * these are the values from the deleted inode.
107 */
108 u64 cur_ino;
109 u64 cur_inode_gen;
110 int cur_inode_new;
111 int cur_inode_new_gen;
112 int cur_inode_deleted;
Alexander Block31db9f72012-07-25 23:19:24 +0200113 u64 cur_inode_size;
114 u64 cur_inode_mode;
Liu Bo644d1942014-02-27 17:29:01 +0800115 u64 cur_inode_rdev;
Josef Bacik16e75492013-10-22 12:18:51 -0400116 u64 cur_inode_last_extent;
Alexander Block31db9f72012-07-25 23:19:24 +0200117
118 u64 send_progress;
119
120 struct list_head new_refs;
121 struct list_head deleted_refs;
122
123 struct radix_tree_root name_cache;
124 struct list_head name_cache_list;
125 int name_cache_size;
126
Liu Bo2131bcd32014-03-05 10:07:35 +0800127 struct file_ra_state ra;
128
Alexander Block31db9f72012-07-25 23:19:24 +0200129 char *read_buf;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +0000130
131 /*
132 * We process inodes by their increasing order, so if before an
133 * incremental send we reverse the parent/child relationship of
134 * directories such that a directory with a lower inode number was
135 * the parent of a directory with a higher inode number, and the one
136 * becoming the new parent got renamed too, we can't rename/move the
137 * directory with lower inode number when we finish processing it - we
138 * must process the directory with higher inode number first, then
139 * rename/move it and then rename/move the directory with lower inode
140 * number. Example follows.
141 *
142 * Tree state when the first send was performed:
143 *
144 * .
145 * |-- a (ino 257)
146 * |-- b (ino 258)
147 * |
148 * |
149 * |-- c (ino 259)
150 * | |-- d (ino 260)
151 * |
152 * |-- c2 (ino 261)
153 *
154 * Tree state when the second (incremental) send is performed:
155 *
156 * .
157 * |-- a (ino 257)
158 * |-- b (ino 258)
159 * |-- c2 (ino 261)
160 * |-- d2 (ino 260)
161 * |-- cc (ino 259)
162 *
163 * The sequence of steps that lead to the second state was:
164 *
165 * mv /a/b/c/d /a/b/c2/d2
166 * mv /a/b/c /a/b/c2/d2/cc
167 *
168 * "c" has lower inode number, but we can't move it (2nd mv operation)
169 * before we move "d", which has higher inode number.
170 *
171 * So we just memorize which move/rename operations must be performed
172 * later when their respective parent is processed and moved/renamed.
173 */
174
175 /* Indexed by parent directory inode number. */
176 struct rb_root pending_dir_moves;
177
178 /*
179 * Reverse index, indexed by the inode number of a directory that
180 * is waiting for the move/rename of its immediate parent before its
181 * own move/rename can be performed.
182 */
183 struct rb_root waiting_dir_moves;
Filipe Manana9dc44212014-02-19 14:31:44 +0000184
185 /*
186 * A directory that is going to be rm'ed might have a child directory
187 * which is in the pending directory moves index above. In this case,
188 * the directory can only be removed after the move/rename of its child
189 * is performed. Example:
190 *
191 * Parent snapshot:
192 *
193 * . (ino 256)
194 * |-- a/ (ino 257)
195 * |-- b/ (ino 258)
196 * |-- c/ (ino 259)
197 * | |-- x/ (ino 260)
198 * |
199 * |-- y/ (ino 261)
200 *
201 * Send snapshot:
202 *
203 * . (ino 256)
204 * |-- a/ (ino 257)
205 * |-- b/ (ino 258)
206 * |-- YY/ (ino 261)
207 * |-- x/ (ino 260)
208 *
209 * Sequence of steps that lead to the send snapshot:
210 * rm -f /a/b/c/foo.txt
211 * mv /a/b/y /a/b/YY
212 * mv /a/b/c/x /a/b/YY
213 * rmdir /a/b/c
214 *
215 * When the child is processed, its move/rename is delayed until its
216 * parent is processed (as explained above), but all other operations
217 * like update utimes, chown, chgrp, etc, are performed and the paths
218 * that it uses for those operations must use the orphanized name of
219 * its parent (the directory we're going to rm later), so we need to
220 * memorize that name.
221 *
222 * Indexed by the inode number of the directory to be deleted.
223 */
224 struct rb_root orphan_dirs;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +0000225};
226
227struct pending_dir_move {
228 struct rb_node node;
229 struct list_head list;
230 u64 parent_ino;
231 u64 ino;
232 u64 gen;
Filipe Manana84471e22015-02-28 22:29:22 +0000233 bool is_orphan;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +0000234 struct list_head update_refs;
235};
236
237struct waiting_dir_move {
238 struct rb_node node;
239 u64 ino;
Filipe Manana9dc44212014-02-19 14:31:44 +0000240 /*
241 * There might be some directory that could not be removed because it
242 * was waiting for this directory inode to be moved first. Therefore
243 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
244 */
245 u64 rmdir_ino;
Filipe Manana8b191a62015-04-09 14:09:14 +0100246 bool orphanized;
Filipe Manana9dc44212014-02-19 14:31:44 +0000247};
248
249struct orphan_dir_info {
250 struct rb_node node;
251 u64 ino;
252 u64 gen;
Alexander Block31db9f72012-07-25 23:19:24 +0200253};
254
255struct name_cache_entry {
256 struct list_head list;
Alexander Block7e0926f2012-07-28 14:20:58 +0200257 /*
258 * radix_tree has only 32bit entries but we need to handle 64bit inums.
259 * We use the lower 32bit of the 64bit inum to store it in the tree. If
260 * more then one inum would fall into the same entry, we use radix_list
261 * to store the additional entries. radix_list is also used to store
262 * entries where two entries have the same inum but different
263 * generations.
264 */
265 struct list_head radix_list;
Alexander Block31db9f72012-07-25 23:19:24 +0200266 u64 ino;
267 u64 gen;
268 u64 parent_ino;
269 u64 parent_gen;
270 int ret;
271 int need_later_update;
272 int name_len;
273 char name[];
274};
275
Filipe David Borba Manana9f037402014-01-22 10:00:53 +0000276static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
277
Filipe Manana9dc44212014-02-19 14:31:44 +0000278static struct waiting_dir_move *
279get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
280
281static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
282
Josef Bacik16e75492013-10-22 12:18:51 -0400283static int need_send_hole(struct send_ctx *sctx)
284{
285 return (sctx->parent_root && !sctx->cur_inode_new &&
286 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
287 S_ISREG(sctx->cur_inode_mode));
288}
289
Alexander Block31db9f72012-07-25 23:19:24 +0200290static void fs_path_reset(struct fs_path *p)
291{
292 if (p->reversed) {
293 p->start = p->buf + p->buf_len - 1;
294 p->end = p->start;
295 *p->start = 0;
296 } else {
297 p->start = p->buf;
298 p->end = p->start;
299 *p->start = 0;
300 }
301}
302
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000303static struct fs_path *fs_path_alloc(void)
Alexander Block31db9f72012-07-25 23:19:24 +0200304{
305 struct fs_path *p;
306
307 p = kmalloc(sizeof(*p), GFP_NOFS);
308 if (!p)
309 return NULL;
310 p->reversed = 0;
Alexander Block31db9f72012-07-25 23:19:24 +0200311 p->buf = p->inline_buf;
312 p->buf_len = FS_PATH_INLINE_SIZE;
313 fs_path_reset(p);
314 return p;
315}
316
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000317static struct fs_path *fs_path_alloc_reversed(void)
Alexander Block31db9f72012-07-25 23:19:24 +0200318{
319 struct fs_path *p;
320
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000321 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +0200322 if (!p)
323 return NULL;
324 p->reversed = 1;
325 fs_path_reset(p);
326 return p;
327}
328
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000329static void fs_path_free(struct fs_path *p)
Alexander Block31db9f72012-07-25 23:19:24 +0200330{
331 if (!p)
332 return;
David Sterbaace01052014-02-05 16:17:34 +0100333 if (p->buf != p->inline_buf)
334 kfree(p->buf);
Alexander Block31db9f72012-07-25 23:19:24 +0200335 kfree(p);
336}
337
338static int fs_path_len(struct fs_path *p)
339{
340 return p->end - p->start;
341}
342
343static int fs_path_ensure_buf(struct fs_path *p, int len)
344{
345 char *tmp_buf;
346 int path_len;
347 int old_buf_len;
348
349 len++;
350
351 if (p->buf_len >= len)
352 return 0;
353
Chris Masoncfd4a532014-04-26 05:02:03 -0700354 if (len > PATH_MAX) {
355 WARN_ON(1);
356 return -ENOMEM;
357 }
358
David Sterba1b2782c2014-02-25 19:32:59 +0100359 path_len = p->end - p->start;
360 old_buf_len = p->buf_len;
361
David Sterbaace01052014-02-05 16:17:34 +0100362 /*
363 * First time the inline_buf does not suffice
364 */
Filipe Manana01a9a8a2014-05-21 17:38:13 +0100365 if (p->buf == p->inline_buf) {
David Sterba9c9ca002014-02-25 19:33:08 +0100366 tmp_buf = kmalloc(len, GFP_NOFS);
Filipe Manana01a9a8a2014-05-21 17:38:13 +0100367 if (tmp_buf)
368 memcpy(tmp_buf, p->buf, old_buf_len);
369 } else {
David Sterba9c9ca002014-02-25 19:33:08 +0100370 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
Filipe Manana01a9a8a2014-05-21 17:38:13 +0100371 }
David Sterba9c9ca002014-02-25 19:33:08 +0100372 if (!tmp_buf)
373 return -ENOMEM;
374 p->buf = tmp_buf;
375 /*
376 * The real size of the buffer is bigger, this will let the fast path
377 * happen most of the time
378 */
379 p->buf_len = ksize(p->buf);
David Sterbaace01052014-02-05 16:17:34 +0100380
Alexander Block31db9f72012-07-25 23:19:24 +0200381 if (p->reversed) {
382 tmp_buf = p->buf + old_buf_len - path_len - 1;
383 p->end = p->buf + p->buf_len - 1;
384 p->start = p->end - path_len;
385 memmove(p->start, tmp_buf, path_len + 1);
386 } else {
387 p->start = p->buf;
388 p->end = p->start + path_len;
389 }
390 return 0;
391}
392
David Sterbab23ab572014-02-03 19:23:19 +0100393static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
394 char **prepared)
Alexander Block31db9f72012-07-25 23:19:24 +0200395{
396 int ret;
397 int new_len;
398
399 new_len = p->end - p->start + name_len;
400 if (p->start != p->end)
401 new_len++;
402 ret = fs_path_ensure_buf(p, new_len);
403 if (ret < 0)
404 goto out;
405
406 if (p->reversed) {
407 if (p->start != p->end)
408 *--p->start = '/';
409 p->start -= name_len;
David Sterbab23ab572014-02-03 19:23:19 +0100410 *prepared = p->start;
Alexander Block31db9f72012-07-25 23:19:24 +0200411 } else {
412 if (p->start != p->end)
413 *p->end++ = '/';
David Sterbab23ab572014-02-03 19:23:19 +0100414 *prepared = p->end;
Alexander Block31db9f72012-07-25 23:19:24 +0200415 p->end += name_len;
416 *p->end = 0;
417 }
418
419out:
420 return ret;
421}
422
423static int fs_path_add(struct fs_path *p, const char *name, int name_len)
424{
425 int ret;
David Sterbab23ab572014-02-03 19:23:19 +0100426 char *prepared;
Alexander Block31db9f72012-07-25 23:19:24 +0200427
David Sterbab23ab572014-02-03 19:23:19 +0100428 ret = fs_path_prepare_for_add(p, name_len, &prepared);
Alexander Block31db9f72012-07-25 23:19:24 +0200429 if (ret < 0)
430 goto out;
David Sterbab23ab572014-02-03 19:23:19 +0100431 memcpy(prepared, name, name_len);
Alexander Block31db9f72012-07-25 23:19:24 +0200432
433out:
434 return ret;
435}
436
437static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
438{
439 int ret;
David Sterbab23ab572014-02-03 19:23:19 +0100440 char *prepared;
Alexander Block31db9f72012-07-25 23:19:24 +0200441
David Sterbab23ab572014-02-03 19:23:19 +0100442 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
Alexander Block31db9f72012-07-25 23:19:24 +0200443 if (ret < 0)
444 goto out;
David Sterbab23ab572014-02-03 19:23:19 +0100445 memcpy(prepared, p2->start, p2->end - p2->start);
Alexander Block31db9f72012-07-25 23:19:24 +0200446
447out:
448 return ret;
449}
450
451static int fs_path_add_from_extent_buffer(struct fs_path *p,
452 struct extent_buffer *eb,
453 unsigned long off, int len)
454{
455 int ret;
David Sterbab23ab572014-02-03 19:23:19 +0100456 char *prepared;
Alexander Block31db9f72012-07-25 23:19:24 +0200457
David Sterbab23ab572014-02-03 19:23:19 +0100458 ret = fs_path_prepare_for_add(p, len, &prepared);
Alexander Block31db9f72012-07-25 23:19:24 +0200459 if (ret < 0)
460 goto out;
461
David Sterbab23ab572014-02-03 19:23:19 +0100462 read_extent_buffer(eb, prepared, off, len);
Alexander Block31db9f72012-07-25 23:19:24 +0200463
464out:
465 return ret;
466}
467
Alexander Block31db9f72012-07-25 23:19:24 +0200468static int fs_path_copy(struct fs_path *p, struct fs_path *from)
469{
470 int ret;
471
472 p->reversed = from->reversed;
473 fs_path_reset(p);
474
475 ret = fs_path_add_path(p, from);
476
477 return ret;
478}
479
480
481static void fs_path_unreverse(struct fs_path *p)
482{
483 char *tmp;
484 int len;
485
486 if (!p->reversed)
487 return;
488
489 tmp = p->start;
490 len = p->end - p->start;
491 p->start = p->buf;
492 p->end = p->start + len;
493 memmove(p->start, tmp, len + 1);
494 p->reversed = 0;
495}
496
497static struct btrfs_path *alloc_path_for_send(void)
498{
499 struct btrfs_path *path;
500
501 path = btrfs_alloc_path();
502 if (!path)
503 return NULL;
504 path->search_commit_root = 1;
505 path->skip_locking = 1;
Josef Bacik3f8a18c2014-03-28 17:16:01 -0400506 path->need_commit_sem = 1;
Alexander Block31db9f72012-07-25 23:19:24 +0200507 return path;
508}
509
Eric Sandeen48a3b632013-04-25 20:41:01 +0000510static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
Alexander Block31db9f72012-07-25 23:19:24 +0200511{
512 int ret;
513 mm_segment_t old_fs;
514 u32 pos = 0;
515
516 old_fs = get_fs();
517 set_fs(KERNEL_DS);
518
519 while (pos < len) {
Fabian Frederickd447d0d2014-07-15 21:17:17 +0200520 ret = vfs_write(filp, (__force const char __user *)buf + pos,
521 len - pos, off);
Alexander Block31db9f72012-07-25 23:19:24 +0200522 /* TODO handle that correctly */
523 /*if (ret == -ERESTARTSYS) {
524 continue;
525 }*/
526 if (ret < 0)
527 goto out;
528 if (ret == 0) {
529 ret = -EIO;
530 goto out;
531 }
532 pos += ret;
533 }
534
535 ret = 0;
536
537out:
538 set_fs(old_fs);
539 return ret;
540}
541
542static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
543{
544 struct btrfs_tlv_header *hdr;
545 int total_len = sizeof(*hdr) + len;
546 int left = sctx->send_max_size - sctx->send_size;
547
548 if (unlikely(left < total_len))
549 return -EOVERFLOW;
550
551 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
552 hdr->tlv_type = cpu_to_le16(attr);
553 hdr->tlv_len = cpu_to_le16(len);
554 memcpy(hdr + 1, data, len);
555 sctx->send_size += total_len;
556
557 return 0;
558}
559
David Sterba95bc79d2013-12-16 17:34:10 +0100560#define TLV_PUT_DEFINE_INT(bits) \
561 static int tlv_put_u##bits(struct send_ctx *sctx, \
562 u##bits attr, u##bits value) \
563 { \
564 __le##bits __tmp = cpu_to_le##bits(value); \
565 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
566 }
Alexander Block31db9f72012-07-25 23:19:24 +0200567
David Sterba95bc79d2013-12-16 17:34:10 +0100568TLV_PUT_DEFINE_INT(64)
Alexander Block31db9f72012-07-25 23:19:24 +0200569
570static int tlv_put_string(struct send_ctx *sctx, u16 attr,
571 const char *str, int len)
572{
573 if (len == -1)
574 len = strlen(str);
575 return tlv_put(sctx, attr, str, len);
576}
577
578static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
579 const u8 *uuid)
580{
581 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
582}
583
Alexander Block31db9f72012-07-25 23:19:24 +0200584static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
585 struct extent_buffer *eb,
586 struct btrfs_timespec *ts)
587{
588 struct btrfs_timespec bts;
589 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
590 return tlv_put(sctx, attr, &bts, sizeof(bts));
591}
592
593
594#define TLV_PUT(sctx, attrtype, attrlen, data) \
595 do { \
596 ret = tlv_put(sctx, attrtype, attrlen, data); \
597 if (ret < 0) \
598 goto tlv_put_failure; \
599 } while (0)
600
601#define TLV_PUT_INT(sctx, attrtype, bits, value) \
602 do { \
603 ret = tlv_put_u##bits(sctx, attrtype, value); \
604 if (ret < 0) \
605 goto tlv_put_failure; \
606 } while (0)
607
608#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
609#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
610#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
611#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
612#define TLV_PUT_STRING(sctx, attrtype, str, len) \
613 do { \
614 ret = tlv_put_string(sctx, attrtype, str, len); \
615 if (ret < 0) \
616 goto tlv_put_failure; \
617 } while (0)
618#define TLV_PUT_PATH(sctx, attrtype, p) \
619 do { \
620 ret = tlv_put_string(sctx, attrtype, p->start, \
621 p->end - p->start); \
622 if (ret < 0) \
623 goto tlv_put_failure; \
624 } while(0)
625#define TLV_PUT_UUID(sctx, attrtype, uuid) \
626 do { \
627 ret = tlv_put_uuid(sctx, attrtype, uuid); \
628 if (ret < 0) \
629 goto tlv_put_failure; \
630 } while (0)
Alexander Block31db9f72012-07-25 23:19:24 +0200631#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
632 do { \
633 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
634 if (ret < 0) \
635 goto tlv_put_failure; \
636 } while (0)
637
638static int send_header(struct send_ctx *sctx)
639{
640 struct btrfs_stream_header hdr;
641
642 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
643 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
644
Anand Jain1bcea352012-09-14 00:04:21 -0600645 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
646 &sctx->send_off);
Alexander Block31db9f72012-07-25 23:19:24 +0200647}
648
649/*
650 * For each command/item we want to send to userspace, we call this function.
651 */
652static int begin_cmd(struct send_ctx *sctx, int cmd)
653{
654 struct btrfs_cmd_header *hdr;
655
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +0530656 if (WARN_ON(!sctx->send_buf))
Alexander Block31db9f72012-07-25 23:19:24 +0200657 return -EINVAL;
Alexander Block31db9f72012-07-25 23:19:24 +0200658
659 BUG_ON(sctx->send_size);
660
661 sctx->send_size += sizeof(*hdr);
662 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
663 hdr->cmd = cpu_to_le16(cmd);
664
665 return 0;
666}
667
668static int send_cmd(struct send_ctx *sctx)
669{
670 int ret;
671 struct btrfs_cmd_header *hdr;
672 u32 crc;
673
674 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
675 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
676 hdr->crc = 0;
677
Filipe David Borba Manana0b947af2014-01-29 21:06:04 +0000678 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
Alexander Block31db9f72012-07-25 23:19:24 +0200679 hdr->crc = cpu_to_le32(crc);
680
Anand Jain1bcea352012-09-14 00:04:21 -0600681 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
682 &sctx->send_off);
Alexander Block31db9f72012-07-25 23:19:24 +0200683
684 sctx->total_send_size += sctx->send_size;
685 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
686 sctx->send_size = 0;
687
688 return ret;
689}
690
691/*
692 * Sends a move instruction to user space
693 */
694static int send_rename(struct send_ctx *sctx,
695 struct fs_path *from, struct fs_path *to)
696{
697 int ret;
698
699verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
700
701 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
702 if (ret < 0)
703 goto out;
704
705 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
706 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
707
708 ret = send_cmd(sctx);
709
710tlv_put_failure:
711out:
712 return ret;
713}
714
715/*
716 * Sends a link instruction to user space
717 */
718static int send_link(struct send_ctx *sctx,
719 struct fs_path *path, struct fs_path *lnk)
720{
721 int ret;
722
723verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
724
725 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
726 if (ret < 0)
727 goto out;
728
729 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
730 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
731
732 ret = send_cmd(sctx);
733
734tlv_put_failure:
735out:
736 return ret;
737}
738
739/*
740 * Sends an unlink instruction to user space
741 */
742static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
743{
744 int ret;
745
746verbose_printk("btrfs: send_unlink %s\n", path->start);
747
748 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
749 if (ret < 0)
750 goto out;
751
752 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
753
754 ret = send_cmd(sctx);
755
756tlv_put_failure:
757out:
758 return ret;
759}
760
761/*
762 * Sends a rmdir instruction to user space
763 */
764static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
765{
766 int ret;
767
768verbose_printk("btrfs: send_rmdir %s\n", path->start);
769
770 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
771 if (ret < 0)
772 goto out;
773
774 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
775
776 ret = send_cmd(sctx);
777
778tlv_put_failure:
779out:
780 return ret;
781}
782
783/*
784 * Helper function to retrieve some fields from an inode item.
785 */
Josef Bacik3f8a18c2014-03-28 17:16:01 -0400786static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
787 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
788 u64 *gid, u64 *rdev)
Alexander Block31db9f72012-07-25 23:19:24 +0200789{
790 int ret;
791 struct btrfs_inode_item *ii;
792 struct btrfs_key key;
Alexander Block31db9f72012-07-25 23:19:24 +0200793
794 key.objectid = ino;
795 key.type = BTRFS_INODE_ITEM_KEY;
796 key.offset = 0;
797 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Alexander Block31db9f72012-07-25 23:19:24 +0200798 if (ret) {
Josef Bacik3f8a18c2014-03-28 17:16:01 -0400799 if (ret > 0)
800 ret = -ENOENT;
801 return ret;
Alexander Block31db9f72012-07-25 23:19:24 +0200802 }
803
804 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
805 struct btrfs_inode_item);
806 if (size)
807 *size = btrfs_inode_size(path->nodes[0], ii);
808 if (gen)
809 *gen = btrfs_inode_generation(path->nodes[0], ii);
810 if (mode)
811 *mode = btrfs_inode_mode(path->nodes[0], ii);
812 if (uid)
813 *uid = btrfs_inode_uid(path->nodes[0], ii);
814 if (gid)
815 *gid = btrfs_inode_gid(path->nodes[0], ii);
Alexander Block85a7b332012-07-26 23:39:10 +0200816 if (rdev)
817 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
Alexander Block31db9f72012-07-25 23:19:24 +0200818
Josef Bacik3f8a18c2014-03-28 17:16:01 -0400819 return ret;
820}
821
822static int get_inode_info(struct btrfs_root *root,
823 u64 ino, u64 *size, u64 *gen,
824 u64 *mode, u64 *uid, u64 *gid,
825 u64 *rdev)
826{
827 struct btrfs_path *path;
828 int ret;
829
830 path = alloc_path_for_send();
831 if (!path)
832 return -ENOMEM;
833 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
834 rdev);
Alexander Block31db9f72012-07-25 23:19:24 +0200835 btrfs_free_path(path);
836 return ret;
837}
838
839typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
840 struct fs_path *p,
841 void *ctx);
842
843/*
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000844 * Helper function to iterate the entries in ONE btrfs_inode_ref or
845 * btrfs_inode_extref.
Alexander Block31db9f72012-07-25 23:19:24 +0200846 * The iterate callback may return a non zero value to stop iteration. This can
847 * be a negative value for error codes or 1 to simply stop it.
848 *
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000849 * path must point to the INODE_REF or INODE_EXTREF when called.
Alexander Block31db9f72012-07-25 23:19:24 +0200850 */
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000851static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
Alexander Block31db9f72012-07-25 23:19:24 +0200852 struct btrfs_key *found_key, int resolve,
853 iterate_inode_ref_t iterate, void *ctx)
854{
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000855 struct extent_buffer *eb = path->nodes[0];
Alexander Block31db9f72012-07-25 23:19:24 +0200856 struct btrfs_item *item;
857 struct btrfs_inode_ref *iref;
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000858 struct btrfs_inode_extref *extref;
Alexander Block31db9f72012-07-25 23:19:24 +0200859 struct btrfs_path *tmp_path;
860 struct fs_path *p;
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000861 u32 cur = 0;
Alexander Block31db9f72012-07-25 23:19:24 +0200862 u32 total;
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000863 int slot = path->slots[0];
Alexander Block31db9f72012-07-25 23:19:24 +0200864 u32 name_len;
865 char *start;
866 int ret = 0;
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000867 int num = 0;
Alexander Block31db9f72012-07-25 23:19:24 +0200868 int index;
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000869 u64 dir;
870 unsigned long name_off;
871 unsigned long elem_size;
872 unsigned long ptr;
Alexander Block31db9f72012-07-25 23:19:24 +0200873
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000874 p = fs_path_alloc_reversed();
Alexander Block31db9f72012-07-25 23:19:24 +0200875 if (!p)
876 return -ENOMEM;
877
878 tmp_path = alloc_path_for_send();
879 if (!tmp_path) {
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000880 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +0200881 return -ENOMEM;
882 }
883
Alexander Block31db9f72012-07-25 23:19:24 +0200884
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000885 if (found_key->type == BTRFS_INODE_REF_KEY) {
886 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
887 struct btrfs_inode_ref);
Ross Kirkdd3cc162013-09-16 15:58:09 +0100888 item = btrfs_item_nr(slot);
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000889 total = btrfs_item_size(eb, item);
890 elem_size = sizeof(*iref);
891 } else {
892 ptr = btrfs_item_ptr_offset(eb, slot);
893 total = btrfs_item_size_nr(eb, slot);
894 elem_size = sizeof(*extref);
895 }
896
Alexander Block31db9f72012-07-25 23:19:24 +0200897 while (cur < total) {
898 fs_path_reset(p);
899
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000900 if (found_key->type == BTRFS_INODE_REF_KEY) {
901 iref = (struct btrfs_inode_ref *)(ptr + cur);
902 name_len = btrfs_inode_ref_name_len(eb, iref);
903 name_off = (unsigned long)(iref + 1);
904 index = btrfs_inode_ref_index(eb, iref);
905 dir = found_key->offset;
906 } else {
907 extref = (struct btrfs_inode_extref *)(ptr + cur);
908 name_len = btrfs_inode_extref_name_len(eb, extref);
909 name_off = (unsigned long)&extref->name;
910 index = btrfs_inode_extref_index(eb, extref);
911 dir = btrfs_inode_extref_parent(eb, extref);
912 }
913
Alexander Block31db9f72012-07-25 23:19:24 +0200914 if (resolve) {
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000915 start = btrfs_ref_to_path(root, tmp_path, name_len,
916 name_off, eb, dir,
917 p->buf, p->buf_len);
Alexander Block31db9f72012-07-25 23:19:24 +0200918 if (IS_ERR(start)) {
919 ret = PTR_ERR(start);
920 goto out;
921 }
922 if (start < p->buf) {
923 /* overflow , try again with larger buffer */
924 ret = fs_path_ensure_buf(p,
925 p->buf_len + p->buf - start);
926 if (ret < 0)
927 goto out;
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000928 start = btrfs_ref_to_path(root, tmp_path,
929 name_len, name_off,
930 eb, dir,
931 p->buf, p->buf_len);
Alexander Block31db9f72012-07-25 23:19:24 +0200932 if (IS_ERR(start)) {
933 ret = PTR_ERR(start);
934 goto out;
935 }
936 BUG_ON(start < p->buf);
937 }
938 p->start = start;
939 } else {
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000940 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
941 name_len);
Alexander Block31db9f72012-07-25 23:19:24 +0200942 if (ret < 0)
943 goto out;
944 }
945
Jan Schmidt96b5bd72012-10-15 08:30:45 +0000946 cur += elem_size + name_len;
947 ret = iterate(num, dir, index, p, ctx);
Alexander Block31db9f72012-07-25 23:19:24 +0200948 if (ret)
949 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +0200950 num++;
951 }
952
953out:
954 btrfs_free_path(tmp_path);
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000955 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +0200956 return ret;
957}
958
959typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
960 const char *name, int name_len,
961 const char *data, int data_len,
962 u8 type, void *ctx);
963
964/*
965 * Helper function to iterate the entries in ONE btrfs_dir_item.
966 * The iterate callback may return a non zero value to stop iteration. This can
967 * be a negative value for error codes or 1 to simply stop it.
968 *
969 * path must point to the dir item when called.
970 */
Tsutomu Itoh924794c2013-05-08 07:51:52 +0000971static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
Alexander Block31db9f72012-07-25 23:19:24 +0200972 struct btrfs_key *found_key,
973 iterate_dir_item_t iterate, void *ctx)
974{
975 int ret = 0;
976 struct extent_buffer *eb;
977 struct btrfs_item *item;
978 struct btrfs_dir_item *di;
Alexander Block31db9f72012-07-25 23:19:24 +0200979 struct btrfs_key di_key;
980 char *buf = NULL;
Filipe Manana7e3ae332014-05-23 20:15:16 +0100981 int buf_len;
Alexander Block31db9f72012-07-25 23:19:24 +0200982 u32 name_len;
983 u32 data_len;
984 u32 cur;
985 u32 len;
986 u32 total;
987 int slot;
988 int num;
989 u8 type;
990
Filipe Manana4395e0c2014-08-20 10:45:45 +0100991 /*
992 * Start with a small buffer (1 page). If later we end up needing more
993 * space, which can happen for xattrs on a fs with a leaf size greater
994 * then the page size, attempt to increase the buffer. Typically xattr
995 * values are small.
996 */
997 buf_len = PATH_MAX;
Alexander Block31db9f72012-07-25 23:19:24 +0200998 buf = kmalloc(buf_len, GFP_NOFS);
999 if (!buf) {
1000 ret = -ENOMEM;
1001 goto out;
1002 }
1003
Alexander Block31db9f72012-07-25 23:19:24 +02001004 eb = path->nodes[0];
1005 slot = path->slots[0];
Ross Kirkdd3cc162013-09-16 15:58:09 +01001006 item = btrfs_item_nr(slot);
Alexander Block31db9f72012-07-25 23:19:24 +02001007 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1008 cur = 0;
1009 len = 0;
1010 total = btrfs_item_size(eb, item);
1011
1012 num = 0;
1013 while (cur < total) {
1014 name_len = btrfs_dir_name_len(eb, di);
1015 data_len = btrfs_dir_data_len(eb, di);
1016 type = btrfs_dir_type(eb, di);
1017 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1018
Filipe Manana7e3ae332014-05-23 20:15:16 +01001019 if (type == BTRFS_FT_XATTR) {
1020 if (name_len > XATTR_NAME_MAX) {
1021 ret = -ENAMETOOLONG;
1022 goto out;
1023 }
Filipe Manana4395e0c2014-08-20 10:45:45 +01001024 if (name_len + data_len > BTRFS_MAX_XATTR_SIZE(root)) {
Filipe Manana7e3ae332014-05-23 20:15:16 +01001025 ret = -E2BIG;
1026 goto out;
1027 }
1028 } else {
1029 /*
1030 * Path too long
1031 */
Filipe Manana4395e0c2014-08-20 10:45:45 +01001032 if (name_len + data_len > PATH_MAX) {
Filipe Manana7e3ae332014-05-23 20:15:16 +01001033 ret = -ENAMETOOLONG;
1034 goto out;
1035 }
Alexander Block31db9f72012-07-25 23:19:24 +02001036 }
1037
Filipe Manana4395e0c2014-08-20 10:45:45 +01001038 if (name_len + data_len > buf_len) {
1039 buf_len = name_len + data_len;
1040 if (is_vmalloc_addr(buf)) {
1041 vfree(buf);
1042 buf = NULL;
1043 } else {
1044 char *tmp = krealloc(buf, buf_len,
1045 GFP_NOFS | __GFP_NOWARN);
1046
1047 if (!tmp)
1048 kfree(buf);
1049 buf = tmp;
1050 }
1051 if (!buf) {
1052 buf = vmalloc(buf_len);
1053 if (!buf) {
1054 ret = -ENOMEM;
1055 goto out;
1056 }
1057 }
1058 }
1059
Alexander Block31db9f72012-07-25 23:19:24 +02001060 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1061 name_len + data_len);
1062
1063 len = sizeof(*di) + name_len + data_len;
1064 di = (struct btrfs_dir_item *)((char *)di + len);
1065 cur += len;
1066
1067 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1068 data_len, type, ctx);
1069 if (ret < 0)
1070 goto out;
1071 if (ret) {
1072 ret = 0;
1073 goto out;
1074 }
1075
1076 num++;
1077 }
1078
1079out:
Filipe Manana4395e0c2014-08-20 10:45:45 +01001080 kvfree(buf);
Alexander Block31db9f72012-07-25 23:19:24 +02001081 return ret;
1082}
1083
1084static int __copy_first_ref(int num, u64 dir, int index,
1085 struct fs_path *p, void *ctx)
1086{
1087 int ret;
1088 struct fs_path *pt = ctx;
1089
1090 ret = fs_path_copy(pt, p);
1091 if (ret < 0)
1092 return ret;
1093
1094 /* we want the first only */
1095 return 1;
1096}
1097
1098/*
1099 * Retrieve the first path of an inode. If an inode has more then one
1100 * ref/hardlink, this is ignored.
1101 */
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001102static int get_inode_path(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02001103 u64 ino, struct fs_path *path)
1104{
1105 int ret;
1106 struct btrfs_key key, found_key;
1107 struct btrfs_path *p;
1108
1109 p = alloc_path_for_send();
1110 if (!p)
1111 return -ENOMEM;
1112
1113 fs_path_reset(path);
1114
1115 key.objectid = ino;
1116 key.type = BTRFS_INODE_REF_KEY;
1117 key.offset = 0;
1118
1119 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1120 if (ret < 0)
1121 goto out;
1122 if (ret) {
1123 ret = 1;
1124 goto out;
1125 }
1126 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1127 if (found_key.objectid != ino ||
Jan Schmidt96b5bd72012-10-15 08:30:45 +00001128 (found_key.type != BTRFS_INODE_REF_KEY &&
1129 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
Alexander Block31db9f72012-07-25 23:19:24 +02001130 ret = -ENOENT;
1131 goto out;
1132 }
1133
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001134 ret = iterate_inode_ref(root, p, &found_key, 1,
1135 __copy_first_ref, path);
Alexander Block31db9f72012-07-25 23:19:24 +02001136 if (ret < 0)
1137 goto out;
1138 ret = 0;
1139
1140out:
1141 btrfs_free_path(p);
1142 return ret;
1143}
1144
1145struct backref_ctx {
1146 struct send_ctx *sctx;
1147
Josef Bacik3f8a18c2014-03-28 17:16:01 -04001148 struct btrfs_path *path;
Alexander Block31db9f72012-07-25 23:19:24 +02001149 /* number of total found references */
1150 u64 found;
1151
1152 /*
1153 * used for clones found in send_root. clones found behind cur_objectid
1154 * and cur_offset are not considered as allowed clones.
1155 */
1156 u64 cur_objectid;
1157 u64 cur_offset;
1158
1159 /* may be truncated in case it's the last extent in a file */
1160 u64 extent_len;
1161
Filipe Manana619d8c42015-05-03 01:56:00 +01001162 /* data offset in the file extent item */
1163 u64 data_offset;
1164
Alexander Block31db9f72012-07-25 23:19:24 +02001165 /* Just to check for bugs in backref resolving */
Alexander Blockee849c02012-07-28 12:42:05 +02001166 int found_itself;
Alexander Block31db9f72012-07-25 23:19:24 +02001167};
1168
1169static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1170{
Jan Schmidt995e01b2012-08-13 02:52:38 -06001171 u64 root = (u64)(uintptr_t)key;
Alexander Block31db9f72012-07-25 23:19:24 +02001172 struct clone_root *cr = (struct clone_root *)elt;
1173
1174 if (root < cr->root->objectid)
1175 return -1;
1176 if (root > cr->root->objectid)
1177 return 1;
1178 return 0;
1179}
1180
1181static int __clone_root_cmp_sort(const void *e1, const void *e2)
1182{
1183 struct clone_root *cr1 = (struct clone_root *)e1;
1184 struct clone_root *cr2 = (struct clone_root *)e2;
1185
1186 if (cr1->root->objectid < cr2->root->objectid)
1187 return -1;
1188 if (cr1->root->objectid > cr2->root->objectid)
1189 return 1;
1190 return 0;
1191}
1192
1193/*
1194 * Called for every backref that is found for the current extent.
Alexander Block766702e2012-07-28 14:11:31 +02001195 * Results are collected in sctx->clone_roots->ino/offset/found_refs
Alexander Block31db9f72012-07-25 23:19:24 +02001196 */
1197static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1198{
1199 struct backref_ctx *bctx = ctx_;
1200 struct clone_root *found;
1201 int ret;
1202 u64 i_size;
1203
1204 /* First check if the root is in the list of accepted clone sources */
Jan Schmidt995e01b2012-08-13 02:52:38 -06001205 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
Alexander Block31db9f72012-07-25 23:19:24 +02001206 bctx->sctx->clone_roots_cnt,
1207 sizeof(struct clone_root),
1208 __clone_root_cmp_bsearch);
1209 if (!found)
1210 return 0;
1211
1212 if (found->root == bctx->sctx->send_root &&
1213 ino == bctx->cur_objectid &&
1214 offset == bctx->cur_offset) {
Alexander Blockee849c02012-07-28 12:42:05 +02001215 bctx->found_itself = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02001216 }
1217
1218 /*
Alexander Block766702e2012-07-28 14:11:31 +02001219 * There are inodes that have extents that lie behind its i_size. Don't
Alexander Block31db9f72012-07-25 23:19:24 +02001220 * accept clones from these extents.
1221 */
Josef Bacik3f8a18c2014-03-28 17:16:01 -04001222 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1223 NULL, NULL, NULL);
1224 btrfs_release_path(bctx->path);
Alexander Block31db9f72012-07-25 23:19:24 +02001225 if (ret < 0)
1226 return ret;
1227
Filipe Manana619d8c42015-05-03 01:56:00 +01001228 if (offset + bctx->data_offset + bctx->extent_len > i_size)
Alexander Block31db9f72012-07-25 23:19:24 +02001229 return 0;
1230
1231 /*
1232 * Make sure we don't consider clones from send_root that are
1233 * behind the current inode/offset.
1234 */
1235 if (found->root == bctx->sctx->send_root) {
1236 /*
1237 * TODO for the moment we don't accept clones from the inode
1238 * that is currently send. We may change this when
1239 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1240 * file.
1241 */
1242 if (ino >= bctx->cur_objectid)
1243 return 0;
Alexander Blocke938c8a2012-07-28 16:33:49 +02001244#if 0
1245 if (ino > bctx->cur_objectid)
Alexander Block31db9f72012-07-25 23:19:24 +02001246 return 0;
Alexander Blocke938c8a2012-07-28 16:33:49 +02001247 if (offset + bctx->extent_len > bctx->cur_offset)
1248 return 0;
1249#endif
Alexander Block31db9f72012-07-25 23:19:24 +02001250 }
1251
1252 bctx->found++;
1253 found->found_refs++;
1254 if (ino < found->ino) {
1255 found->ino = ino;
1256 found->offset = offset;
1257 } else if (found->ino == ino) {
1258 /*
1259 * same extent found more then once in the same file.
1260 */
1261 if (found->offset > offset + bctx->extent_len)
1262 found->offset = offset;
1263 }
1264
1265 return 0;
1266}
1267
1268/*
Alexander Block766702e2012-07-28 14:11:31 +02001269 * Given an inode, offset and extent item, it finds a good clone for a clone
1270 * instruction. Returns -ENOENT when none could be found. The function makes
1271 * sure that the returned clone is usable at the point where sending is at the
1272 * moment. This means, that no clones are accepted which lie behind the current
1273 * inode+offset.
1274 *
Alexander Block31db9f72012-07-25 23:19:24 +02001275 * path must point to the extent item when called.
1276 */
1277static int find_extent_clone(struct send_ctx *sctx,
1278 struct btrfs_path *path,
1279 u64 ino, u64 data_offset,
1280 u64 ino_size,
1281 struct clone_root **found)
1282{
1283 int ret;
1284 int extent_type;
1285 u64 logical;
Chris Mason74dd17f2012-08-07 16:25:13 -04001286 u64 disk_byte;
Alexander Block31db9f72012-07-25 23:19:24 +02001287 u64 num_bytes;
1288 u64 extent_item_pos;
Liu Bo69917e42012-09-07 20:01:28 -06001289 u64 flags = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02001290 struct btrfs_file_extent_item *fi;
1291 struct extent_buffer *eb = path->nodes[0];
Alexander Block35075bb2012-07-28 12:44:34 +02001292 struct backref_ctx *backref_ctx = NULL;
Alexander Block31db9f72012-07-25 23:19:24 +02001293 struct clone_root *cur_clone_root;
1294 struct btrfs_key found_key;
1295 struct btrfs_path *tmp_path;
Chris Mason74dd17f2012-08-07 16:25:13 -04001296 int compressed;
Alexander Block31db9f72012-07-25 23:19:24 +02001297 u32 i;
1298
1299 tmp_path = alloc_path_for_send();
1300 if (!tmp_path)
1301 return -ENOMEM;
1302
Josef Bacik3f8a18c2014-03-28 17:16:01 -04001303 /* We only use this path under the commit sem */
1304 tmp_path->need_commit_sem = 0;
1305
Alexander Block35075bb2012-07-28 12:44:34 +02001306 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1307 if (!backref_ctx) {
1308 ret = -ENOMEM;
1309 goto out;
1310 }
1311
Josef Bacik3f8a18c2014-03-28 17:16:01 -04001312 backref_ctx->path = tmp_path;
1313
Alexander Block31db9f72012-07-25 23:19:24 +02001314 if (data_offset >= ino_size) {
1315 /*
1316 * There may be extents that lie behind the file's size.
1317 * I at least had this in combination with snapshotting while
1318 * writing large files.
1319 */
1320 ret = 0;
1321 goto out;
1322 }
1323
1324 fi = btrfs_item_ptr(eb, path->slots[0],
1325 struct btrfs_file_extent_item);
1326 extent_type = btrfs_file_extent_type(eb, fi);
1327 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1328 ret = -ENOENT;
1329 goto out;
1330 }
Chris Mason74dd17f2012-08-07 16:25:13 -04001331 compressed = btrfs_file_extent_compression(eb, fi);
Alexander Block31db9f72012-07-25 23:19:24 +02001332
1333 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
Chris Mason74dd17f2012-08-07 16:25:13 -04001334 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1335 if (disk_byte == 0) {
Alexander Block31db9f72012-07-25 23:19:24 +02001336 ret = -ENOENT;
1337 goto out;
1338 }
Chris Mason74dd17f2012-08-07 16:25:13 -04001339 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
Alexander Block31db9f72012-07-25 23:19:24 +02001340
Josef Bacik9e351cc2014-03-13 15:42:13 -04001341 down_read(&sctx->send_root->fs_info->commit_root_sem);
Liu Bo69917e42012-09-07 20:01:28 -06001342 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1343 &found_key, &flags);
Josef Bacik9e351cc2014-03-13 15:42:13 -04001344 up_read(&sctx->send_root->fs_info->commit_root_sem);
Alexander Block31db9f72012-07-25 23:19:24 +02001345 btrfs_release_path(tmp_path);
1346
1347 if (ret < 0)
1348 goto out;
Liu Bo69917e42012-09-07 20:01:28 -06001349 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
Alexander Block31db9f72012-07-25 23:19:24 +02001350 ret = -EIO;
1351 goto out;
1352 }
1353
1354 /*
1355 * Setup the clone roots.
1356 */
1357 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1358 cur_clone_root = sctx->clone_roots + i;
1359 cur_clone_root->ino = (u64)-1;
1360 cur_clone_root->offset = 0;
1361 cur_clone_root->found_refs = 0;
1362 }
1363
Alexander Block35075bb2012-07-28 12:44:34 +02001364 backref_ctx->sctx = sctx;
1365 backref_ctx->found = 0;
1366 backref_ctx->cur_objectid = ino;
1367 backref_ctx->cur_offset = data_offset;
1368 backref_ctx->found_itself = 0;
1369 backref_ctx->extent_len = num_bytes;
Filipe Manana619d8c42015-05-03 01:56:00 +01001370 /*
1371 * For non-compressed extents iterate_extent_inodes() gives us extent
1372 * offsets that already take into account the data offset, but not for
1373 * compressed extents, since the offset is logical and not relative to
1374 * the physical extent locations. We must take this into account to
1375 * avoid sending clone offsets that go beyond the source file's size,
1376 * which would result in the clone ioctl failing with -EINVAL on the
1377 * receiving end.
1378 */
1379 if (compressed == BTRFS_COMPRESS_NONE)
1380 backref_ctx->data_offset = 0;
1381 else
1382 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
Alexander Block31db9f72012-07-25 23:19:24 +02001383
1384 /*
1385 * The last extent of a file may be too large due to page alignment.
1386 * We need to adjust extent_len in this case so that the checks in
1387 * __iterate_backrefs work.
1388 */
1389 if (data_offset + num_bytes >= ino_size)
Alexander Block35075bb2012-07-28 12:44:34 +02001390 backref_ctx->extent_len = ino_size - data_offset;
Alexander Block31db9f72012-07-25 23:19:24 +02001391
1392 /*
1393 * Now collect all backrefs.
1394 */
Chris Mason74dd17f2012-08-07 16:25:13 -04001395 if (compressed == BTRFS_COMPRESS_NONE)
1396 extent_item_pos = logical - found_key.objectid;
1397 else
1398 extent_item_pos = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02001399 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1400 found_key.objectid, extent_item_pos, 1,
Alexander Block35075bb2012-07-28 12:44:34 +02001401 __iterate_backrefs, backref_ctx);
Chris Mason74dd17f2012-08-07 16:25:13 -04001402
Alexander Block31db9f72012-07-25 23:19:24 +02001403 if (ret < 0)
1404 goto out;
1405
Alexander Block35075bb2012-07-28 12:44:34 +02001406 if (!backref_ctx->found_itself) {
Alexander Block31db9f72012-07-25 23:19:24 +02001407 /* found a bug in backref code? */
1408 ret = -EIO;
Frank Holtonefe120a2013-12-20 11:37:06 -05001409 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
Alexander Block31db9f72012-07-25 23:19:24 +02001410 "send_root. inode=%llu, offset=%llu, "
David Sterba351fd352014-05-15 16:48:20 +02001411 "disk_byte=%llu found extent=%llu",
Chris Mason74dd17f2012-08-07 16:25:13 -04001412 ino, data_offset, disk_byte, found_key.objectid);
Alexander Block31db9f72012-07-25 23:19:24 +02001413 goto out;
1414 }
1415
1416verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1417 "ino=%llu, "
1418 "num_bytes=%llu, logical=%llu\n",
1419 data_offset, ino, num_bytes, logical);
1420
Alexander Block35075bb2012-07-28 12:44:34 +02001421 if (!backref_ctx->found)
Alexander Block31db9f72012-07-25 23:19:24 +02001422 verbose_printk("btrfs: no clones found\n");
1423
1424 cur_clone_root = NULL;
1425 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1426 if (sctx->clone_roots[i].found_refs) {
1427 if (!cur_clone_root)
1428 cur_clone_root = sctx->clone_roots + i;
1429 else if (sctx->clone_roots[i].root == sctx->send_root)
1430 /* prefer clones from send_root over others */
1431 cur_clone_root = sctx->clone_roots + i;
Alexander Block31db9f72012-07-25 23:19:24 +02001432 }
1433
1434 }
1435
1436 if (cur_clone_root) {
1437 *found = cur_clone_root;
1438 ret = 0;
1439 } else {
1440 ret = -ENOENT;
1441 }
1442
1443out:
1444 btrfs_free_path(tmp_path);
Alexander Block35075bb2012-07-28 12:44:34 +02001445 kfree(backref_ctx);
Alexander Block31db9f72012-07-25 23:19:24 +02001446 return ret;
1447}
1448
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001449static int read_symlink(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02001450 u64 ino,
1451 struct fs_path *dest)
1452{
1453 int ret;
1454 struct btrfs_path *path;
1455 struct btrfs_key key;
1456 struct btrfs_file_extent_item *ei;
1457 u8 type;
1458 u8 compression;
1459 unsigned long off;
1460 int len;
1461
1462 path = alloc_path_for_send();
1463 if (!path)
1464 return -ENOMEM;
1465
1466 key.objectid = ino;
1467 key.type = BTRFS_EXTENT_DATA_KEY;
1468 key.offset = 0;
1469 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1470 if (ret < 0)
1471 goto out;
Filipe Mananaa8797192015-12-31 18:07:59 +00001472 if (ret) {
1473 /*
1474 * An empty symlink inode. Can happen in rare error paths when
1475 * creating a symlink (transaction committed before the inode
1476 * eviction handler removed the symlink inode items and a crash
1477 * happened in between or the subvol was snapshoted in between).
1478 * Print an informative message to dmesg/syslog so that the user
1479 * can delete the symlink.
1480 */
1481 btrfs_err(root->fs_info,
1482 "Found empty symlink inode %llu at root %llu",
1483 ino, root->root_key.objectid);
1484 ret = -EIO;
1485 goto out;
1486 }
Alexander Block31db9f72012-07-25 23:19:24 +02001487
1488 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1489 struct btrfs_file_extent_item);
1490 type = btrfs_file_extent_type(path->nodes[0], ei);
1491 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1492 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1493 BUG_ON(compression);
1494
1495 off = btrfs_file_extent_inline_start(ei);
Chris Mason514ac8a2014-01-03 21:07:00 -08001496 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
Alexander Block31db9f72012-07-25 23:19:24 +02001497
1498 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
Alexander Block31db9f72012-07-25 23:19:24 +02001499
1500out:
1501 btrfs_free_path(path);
1502 return ret;
1503}
1504
1505/*
1506 * Helper function to generate a file name that is unique in the root of
1507 * send_root and parent_root. This is used to generate names for orphan inodes.
1508 */
1509static int gen_unique_name(struct send_ctx *sctx,
1510 u64 ino, u64 gen,
1511 struct fs_path *dest)
1512{
1513 int ret = 0;
1514 struct btrfs_path *path;
1515 struct btrfs_dir_item *di;
1516 char tmp[64];
1517 int len;
1518 u64 idx = 0;
1519
1520 path = alloc_path_for_send();
1521 if (!path)
1522 return -ENOMEM;
1523
1524 while (1) {
Filipe David Borba Mananaf74b86d2014-01-21 23:36:38 +00001525 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
Alexander Block31db9f72012-07-25 23:19:24 +02001526 ino, gen, idx);
David Sterba64792f22014-02-03 18:24:09 +01001527 ASSERT(len < sizeof(tmp));
Alexander Block31db9f72012-07-25 23:19:24 +02001528
1529 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1530 path, BTRFS_FIRST_FREE_OBJECTID,
1531 tmp, strlen(tmp), 0);
1532 btrfs_release_path(path);
1533 if (IS_ERR(di)) {
1534 ret = PTR_ERR(di);
1535 goto out;
1536 }
1537 if (di) {
1538 /* not unique, try again */
1539 idx++;
1540 continue;
1541 }
1542
1543 if (!sctx->parent_root) {
1544 /* unique */
1545 ret = 0;
1546 break;
1547 }
1548
1549 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1550 path, BTRFS_FIRST_FREE_OBJECTID,
1551 tmp, strlen(tmp), 0);
1552 btrfs_release_path(path);
1553 if (IS_ERR(di)) {
1554 ret = PTR_ERR(di);
1555 goto out;
1556 }
1557 if (di) {
1558 /* not unique, try again */
1559 idx++;
1560 continue;
1561 }
1562 /* unique */
1563 break;
1564 }
1565
1566 ret = fs_path_add(dest, tmp, strlen(tmp));
1567
1568out:
1569 btrfs_free_path(path);
1570 return ret;
1571}
1572
1573enum inode_state {
1574 inode_state_no_change,
1575 inode_state_will_create,
1576 inode_state_did_create,
1577 inode_state_will_delete,
1578 inode_state_did_delete,
1579};
1580
1581static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1582{
1583 int ret;
1584 int left_ret;
1585 int right_ret;
1586 u64 left_gen;
1587 u64 right_gen;
1588
1589 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02001590 NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02001591 if (ret < 0 && ret != -ENOENT)
1592 goto out;
1593 left_ret = ret;
1594
1595 if (!sctx->parent_root) {
1596 right_ret = -ENOENT;
1597 } else {
1598 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
Alexander Block85a7b332012-07-26 23:39:10 +02001599 NULL, NULL, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02001600 if (ret < 0 && ret != -ENOENT)
1601 goto out;
1602 right_ret = ret;
1603 }
1604
1605 if (!left_ret && !right_ret) {
Alexander Blocke938c8a2012-07-28 16:33:49 +02001606 if (left_gen == gen && right_gen == gen) {
Alexander Block31db9f72012-07-25 23:19:24 +02001607 ret = inode_state_no_change;
Alexander Blocke938c8a2012-07-28 16:33:49 +02001608 } else if (left_gen == gen) {
Alexander Block31db9f72012-07-25 23:19:24 +02001609 if (ino < sctx->send_progress)
1610 ret = inode_state_did_create;
1611 else
1612 ret = inode_state_will_create;
1613 } else if (right_gen == gen) {
1614 if (ino < sctx->send_progress)
1615 ret = inode_state_did_delete;
1616 else
1617 ret = inode_state_will_delete;
1618 } else {
1619 ret = -ENOENT;
1620 }
1621 } else if (!left_ret) {
1622 if (left_gen == gen) {
1623 if (ino < sctx->send_progress)
1624 ret = inode_state_did_create;
1625 else
1626 ret = inode_state_will_create;
1627 } else {
1628 ret = -ENOENT;
1629 }
1630 } else if (!right_ret) {
1631 if (right_gen == gen) {
1632 if (ino < sctx->send_progress)
1633 ret = inode_state_did_delete;
1634 else
1635 ret = inode_state_will_delete;
1636 } else {
1637 ret = -ENOENT;
1638 }
1639 } else {
1640 ret = -ENOENT;
1641 }
1642
1643out:
1644 return ret;
1645}
1646
1647static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1648{
1649 int ret;
1650
1651 ret = get_cur_inode_state(sctx, ino, gen);
1652 if (ret < 0)
1653 goto out;
1654
1655 if (ret == inode_state_no_change ||
1656 ret == inode_state_did_create ||
1657 ret == inode_state_will_delete)
1658 ret = 1;
1659 else
1660 ret = 0;
1661
1662out:
1663 return ret;
1664}
1665
1666/*
1667 * Helper function to lookup a dir item in a dir.
1668 */
1669static int lookup_dir_item_inode(struct btrfs_root *root,
1670 u64 dir, const char *name, int name_len,
1671 u64 *found_inode,
1672 u8 *found_type)
1673{
1674 int ret = 0;
1675 struct btrfs_dir_item *di;
1676 struct btrfs_key key;
1677 struct btrfs_path *path;
1678
1679 path = alloc_path_for_send();
1680 if (!path)
1681 return -ENOMEM;
1682
1683 di = btrfs_lookup_dir_item(NULL, root, path,
1684 dir, name, name_len, 0);
1685 if (!di) {
1686 ret = -ENOENT;
1687 goto out;
1688 }
1689 if (IS_ERR(di)) {
1690 ret = PTR_ERR(di);
1691 goto out;
1692 }
1693 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
Filipe Manana1af56072014-05-25 04:49:24 +01001694 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1695 ret = -ENOENT;
1696 goto out;
1697 }
Alexander Block31db9f72012-07-25 23:19:24 +02001698 *found_inode = key.objectid;
1699 *found_type = btrfs_dir_type(path->nodes[0], di);
1700
1701out:
1702 btrfs_free_path(path);
1703 return ret;
1704}
1705
Alexander Block766702e2012-07-28 14:11:31 +02001706/*
1707 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1708 * generation of the parent dir and the name of the dir entry.
1709 */
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001710static int get_first_ref(struct btrfs_root *root, u64 ino,
Alexander Block31db9f72012-07-25 23:19:24 +02001711 u64 *dir, u64 *dir_gen, struct fs_path *name)
1712{
1713 int ret;
1714 struct btrfs_key key;
1715 struct btrfs_key found_key;
1716 struct btrfs_path *path;
Alexander Block31db9f72012-07-25 23:19:24 +02001717 int len;
Jan Schmidt96b5bd72012-10-15 08:30:45 +00001718 u64 parent_dir;
Alexander Block31db9f72012-07-25 23:19:24 +02001719
1720 path = alloc_path_for_send();
1721 if (!path)
1722 return -ENOMEM;
1723
1724 key.objectid = ino;
1725 key.type = BTRFS_INODE_REF_KEY;
1726 key.offset = 0;
1727
1728 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1729 if (ret < 0)
1730 goto out;
1731 if (!ret)
1732 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1733 path->slots[0]);
Jan Schmidt96b5bd72012-10-15 08:30:45 +00001734 if (ret || found_key.objectid != ino ||
1735 (found_key.type != BTRFS_INODE_REF_KEY &&
1736 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
Alexander Block31db9f72012-07-25 23:19:24 +02001737 ret = -ENOENT;
1738 goto out;
1739 }
1740
Filipe Manana51a60252014-05-13 22:01:02 +01001741 if (found_key.type == BTRFS_INODE_REF_KEY) {
Jan Schmidt96b5bd72012-10-15 08:30:45 +00001742 struct btrfs_inode_ref *iref;
1743 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1744 struct btrfs_inode_ref);
1745 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1746 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1747 (unsigned long)(iref + 1),
1748 len);
1749 parent_dir = found_key.offset;
1750 } else {
1751 struct btrfs_inode_extref *extref;
1752 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1753 struct btrfs_inode_extref);
1754 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1755 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1756 (unsigned long)&extref->name, len);
1757 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1758 }
Alexander Block31db9f72012-07-25 23:19:24 +02001759 if (ret < 0)
1760 goto out;
1761 btrfs_release_path(path);
1762
Filipe Mananab46ab972014-03-21 12:46:54 +00001763 if (dir_gen) {
1764 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1765 NULL, NULL, NULL);
1766 if (ret < 0)
1767 goto out;
1768 }
Alexander Block31db9f72012-07-25 23:19:24 +02001769
Jan Schmidt96b5bd72012-10-15 08:30:45 +00001770 *dir = parent_dir;
Alexander Block31db9f72012-07-25 23:19:24 +02001771
1772out:
1773 btrfs_free_path(path);
1774 return ret;
1775}
1776
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001777static int is_first_ref(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02001778 u64 ino, u64 dir,
1779 const char *name, int name_len)
1780{
1781 int ret;
1782 struct fs_path *tmp_name;
1783 u64 tmp_dir;
Alexander Block31db9f72012-07-25 23:19:24 +02001784
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001785 tmp_name = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02001786 if (!tmp_name)
1787 return -ENOMEM;
1788
Filipe Mananab46ab972014-03-21 12:46:54 +00001789 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
Alexander Block31db9f72012-07-25 23:19:24 +02001790 if (ret < 0)
1791 goto out;
1792
Alexander Blockb9291af2012-07-28 11:07:18 +02001793 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
Alexander Block31db9f72012-07-25 23:19:24 +02001794 ret = 0;
1795 goto out;
1796 }
1797
Alexander Blocke938c8a2012-07-28 16:33:49 +02001798 ret = !memcmp(tmp_name->start, name, name_len);
Alexander Block31db9f72012-07-25 23:19:24 +02001799
1800out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001801 fs_path_free(tmp_name);
Alexander Block31db9f72012-07-25 23:19:24 +02001802 return ret;
1803}
1804
Alexander Block766702e2012-07-28 14:11:31 +02001805/*
1806 * Used by process_recorded_refs to determine if a new ref would overwrite an
1807 * already existing ref. In case it detects an overwrite, it returns the
1808 * inode/gen in who_ino/who_gen.
1809 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1810 * to make sure later references to the overwritten inode are possible.
1811 * Orphanizing is however only required for the first ref of an inode.
1812 * process_recorded_refs does an additional is_first_ref check to see if
1813 * orphanizing is really required.
1814 */
Alexander Block31db9f72012-07-25 23:19:24 +02001815static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1816 const char *name, int name_len,
1817 u64 *who_ino, u64 *who_gen)
1818{
1819 int ret = 0;
Josef Bacikebdad912013-08-06 16:47:48 -04001820 u64 gen;
Alexander Block31db9f72012-07-25 23:19:24 +02001821 u64 other_inode = 0;
1822 u8 other_type = 0;
1823
1824 if (!sctx->parent_root)
1825 goto out;
1826
1827 ret = is_inode_existent(sctx, dir, dir_gen);
1828 if (ret <= 0)
1829 goto out;
1830
Josef Bacikebdad912013-08-06 16:47:48 -04001831 /*
1832 * If we have a parent root we need to verify that the parent dir was
1833 * not delted and then re-created, if it was then we have no overwrite
1834 * and we can just unlink this entry.
1835 */
1836 if (sctx->parent_root) {
1837 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1838 NULL, NULL, NULL);
1839 if (ret < 0 && ret != -ENOENT)
1840 goto out;
1841 if (ret) {
1842 ret = 0;
1843 goto out;
1844 }
1845 if (gen != dir_gen)
1846 goto out;
1847 }
1848
Alexander Block31db9f72012-07-25 23:19:24 +02001849 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1850 &other_inode, &other_type);
1851 if (ret < 0 && ret != -ENOENT)
1852 goto out;
1853 if (ret) {
1854 ret = 0;
1855 goto out;
1856 }
1857
Alexander Block766702e2012-07-28 14:11:31 +02001858 /*
1859 * Check if the overwritten ref was already processed. If yes, the ref
1860 * was already unlinked/moved, so we can safely assume that we will not
1861 * overwrite anything at this point in time.
1862 */
Alexander Block31db9f72012-07-25 23:19:24 +02001863 if (other_inode > sctx->send_progress) {
1864 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02001865 who_gen, NULL, NULL, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02001866 if (ret < 0)
1867 goto out;
1868
1869 ret = 1;
1870 *who_ino = other_inode;
1871 } else {
1872 ret = 0;
1873 }
1874
1875out:
1876 return ret;
1877}
1878
Alexander Block766702e2012-07-28 14:11:31 +02001879/*
1880 * Checks if the ref was overwritten by an already processed inode. This is
1881 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1882 * thus the orphan name needs be used.
1883 * process_recorded_refs also uses it to avoid unlinking of refs that were
1884 * overwritten.
1885 */
Alexander Block31db9f72012-07-25 23:19:24 +02001886static int did_overwrite_ref(struct send_ctx *sctx,
1887 u64 dir, u64 dir_gen,
1888 u64 ino, u64 ino_gen,
1889 const char *name, int name_len)
1890{
1891 int ret = 0;
1892 u64 gen;
1893 u64 ow_inode;
1894 u8 other_type;
1895
1896 if (!sctx->parent_root)
1897 goto out;
1898
1899 ret = is_inode_existent(sctx, dir, dir_gen);
1900 if (ret <= 0)
1901 goto out;
1902
1903 /* check if the ref was overwritten by another ref */
1904 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1905 &ow_inode, &other_type);
1906 if (ret < 0 && ret != -ENOENT)
1907 goto out;
1908 if (ret) {
1909 /* was never and will never be overwritten */
1910 ret = 0;
1911 goto out;
1912 }
1913
1914 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02001915 NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02001916 if (ret < 0)
1917 goto out;
1918
1919 if (ow_inode == ino && gen == ino_gen) {
1920 ret = 0;
1921 goto out;
1922 }
1923
Filipe Manana8b191a62015-04-09 14:09:14 +01001924 /*
1925 * We know that it is or will be overwritten. Check this now.
1926 * The current inode being processed might have been the one that caused
Filipe Mananab786f162015-09-26 15:30:23 +01001927 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1928 * the current inode being processed.
Filipe Manana8b191a62015-04-09 14:09:14 +01001929 */
Filipe Mananab786f162015-09-26 15:30:23 +01001930 if ((ow_inode < sctx->send_progress) ||
1931 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1932 gen == sctx->cur_inode_gen))
Alexander Block31db9f72012-07-25 23:19:24 +02001933 ret = 1;
1934 else
1935 ret = 0;
1936
1937out:
1938 return ret;
1939}
1940
Alexander Block766702e2012-07-28 14:11:31 +02001941/*
1942 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1943 * that got overwritten. This is used by process_recorded_refs to determine
1944 * if it has to use the path as returned by get_cur_path or the orphan name.
1945 */
Alexander Block31db9f72012-07-25 23:19:24 +02001946static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1947{
1948 int ret = 0;
1949 struct fs_path *name = NULL;
1950 u64 dir;
1951 u64 dir_gen;
1952
1953 if (!sctx->parent_root)
1954 goto out;
1955
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001956 name = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02001957 if (!name)
1958 return -ENOMEM;
1959
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001960 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02001961 if (ret < 0)
1962 goto out;
1963
1964 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1965 name->start, fs_path_len(name));
Alexander Block31db9f72012-07-25 23:19:24 +02001966
1967out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001968 fs_path_free(name);
Alexander Block31db9f72012-07-25 23:19:24 +02001969 return ret;
1970}
1971
Alexander Block766702e2012-07-28 14:11:31 +02001972/*
1973 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1974 * so we need to do some special handling in case we have clashes. This function
1975 * takes care of this with the help of name_cache_entry::radix_list.
Alexander Block5dc67d02012-08-01 12:07:43 +02001976 * In case of error, nce is kfreed.
Alexander Block766702e2012-07-28 14:11:31 +02001977 */
Alexander Block31db9f72012-07-25 23:19:24 +02001978static int name_cache_insert(struct send_ctx *sctx,
1979 struct name_cache_entry *nce)
1980{
1981 int ret = 0;
Alexander Block7e0926f2012-07-28 14:20:58 +02001982 struct list_head *nce_head;
Alexander Block31db9f72012-07-25 23:19:24 +02001983
Alexander Block7e0926f2012-07-28 14:20:58 +02001984 nce_head = radix_tree_lookup(&sctx->name_cache,
1985 (unsigned long)nce->ino);
1986 if (!nce_head) {
1987 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
Tsutomu Itohcfa7a9c2012-12-17 06:38:51 +00001988 if (!nce_head) {
1989 kfree(nce);
Alexander Block31db9f72012-07-25 23:19:24 +02001990 return -ENOMEM;
Tsutomu Itohcfa7a9c2012-12-17 06:38:51 +00001991 }
Alexander Block7e0926f2012-07-28 14:20:58 +02001992 INIT_LIST_HEAD(nce_head);
Alexander Block31db9f72012-07-25 23:19:24 +02001993
Alexander Block7e0926f2012-07-28 14:20:58 +02001994 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
Alexander Block5dc67d02012-08-01 12:07:43 +02001995 if (ret < 0) {
1996 kfree(nce_head);
1997 kfree(nce);
Alexander Block31db9f72012-07-25 23:19:24 +02001998 return ret;
Alexander Block5dc67d02012-08-01 12:07:43 +02001999 }
Alexander Block31db9f72012-07-25 23:19:24 +02002000 }
Alexander Block7e0926f2012-07-28 14:20:58 +02002001 list_add_tail(&nce->radix_list, nce_head);
Alexander Block31db9f72012-07-25 23:19:24 +02002002 list_add_tail(&nce->list, &sctx->name_cache_list);
2003 sctx->name_cache_size++;
2004
2005 return ret;
2006}
2007
2008static void name_cache_delete(struct send_ctx *sctx,
2009 struct name_cache_entry *nce)
2010{
Alexander Block7e0926f2012-07-28 14:20:58 +02002011 struct list_head *nce_head;
Alexander Block31db9f72012-07-25 23:19:24 +02002012
Alexander Block7e0926f2012-07-28 14:20:58 +02002013 nce_head = radix_tree_lookup(&sctx->name_cache,
2014 (unsigned long)nce->ino);
David Sterba57fb8912014-02-03 19:24:40 +01002015 if (!nce_head) {
2016 btrfs_err(sctx->send_root->fs_info,
2017 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2018 nce->ino, sctx->name_cache_size);
2019 }
Alexander Block31db9f72012-07-25 23:19:24 +02002020
Alexander Block7e0926f2012-07-28 14:20:58 +02002021 list_del(&nce->radix_list);
Alexander Block31db9f72012-07-25 23:19:24 +02002022 list_del(&nce->list);
Alexander Block31db9f72012-07-25 23:19:24 +02002023 sctx->name_cache_size--;
Alexander Block7e0926f2012-07-28 14:20:58 +02002024
David Sterba57fb8912014-02-03 19:24:40 +01002025 /*
2026 * We may not get to the final release of nce_head if the lookup fails
2027 */
2028 if (nce_head && list_empty(nce_head)) {
Alexander Block7e0926f2012-07-28 14:20:58 +02002029 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2030 kfree(nce_head);
2031 }
Alexander Block31db9f72012-07-25 23:19:24 +02002032}
2033
2034static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2035 u64 ino, u64 gen)
2036{
Alexander Block7e0926f2012-07-28 14:20:58 +02002037 struct list_head *nce_head;
2038 struct name_cache_entry *cur;
Alexander Block31db9f72012-07-25 23:19:24 +02002039
Alexander Block7e0926f2012-07-28 14:20:58 +02002040 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2041 if (!nce_head)
Alexander Block31db9f72012-07-25 23:19:24 +02002042 return NULL;
2043
Alexander Block7e0926f2012-07-28 14:20:58 +02002044 list_for_each_entry(cur, nce_head, radix_list) {
2045 if (cur->ino == ino && cur->gen == gen)
2046 return cur;
2047 }
Alexander Block31db9f72012-07-25 23:19:24 +02002048 return NULL;
2049}
2050
Alexander Block766702e2012-07-28 14:11:31 +02002051/*
2052 * Removes the entry from the list and adds it back to the end. This marks the
2053 * entry as recently used so that name_cache_clean_unused does not remove it.
2054 */
Alexander Block31db9f72012-07-25 23:19:24 +02002055static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2056{
2057 list_del(&nce->list);
2058 list_add_tail(&nce->list, &sctx->name_cache_list);
2059}
2060
Alexander Block766702e2012-07-28 14:11:31 +02002061/*
2062 * Remove some entries from the beginning of name_cache_list.
2063 */
Alexander Block31db9f72012-07-25 23:19:24 +02002064static void name_cache_clean_unused(struct send_ctx *sctx)
2065{
2066 struct name_cache_entry *nce;
2067
2068 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2069 return;
2070
2071 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2072 nce = list_entry(sctx->name_cache_list.next,
2073 struct name_cache_entry, list);
2074 name_cache_delete(sctx, nce);
2075 kfree(nce);
2076 }
2077}
2078
2079static void name_cache_free(struct send_ctx *sctx)
2080{
2081 struct name_cache_entry *nce;
Alexander Block31db9f72012-07-25 23:19:24 +02002082
Alexander Blocke938c8a2012-07-28 16:33:49 +02002083 while (!list_empty(&sctx->name_cache_list)) {
2084 nce = list_entry(sctx->name_cache_list.next,
2085 struct name_cache_entry, list);
Alexander Block31db9f72012-07-25 23:19:24 +02002086 name_cache_delete(sctx, nce);
Alexander Block17589bd2012-07-28 14:13:35 +02002087 kfree(nce);
Alexander Block31db9f72012-07-25 23:19:24 +02002088 }
2089}
2090
Alexander Block766702e2012-07-28 14:11:31 +02002091/*
2092 * Used by get_cur_path for each ref up to the root.
2093 * Returns 0 if it succeeded.
2094 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2095 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2096 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2097 * Returns <0 in case of error.
2098 */
Alexander Block31db9f72012-07-25 23:19:24 +02002099static int __get_cur_name_and_parent(struct send_ctx *sctx,
2100 u64 ino, u64 gen,
2101 u64 *parent_ino,
2102 u64 *parent_gen,
2103 struct fs_path *dest)
2104{
2105 int ret;
2106 int nce_ret;
Alexander Block31db9f72012-07-25 23:19:24 +02002107 struct name_cache_entry *nce = NULL;
2108
Alexander Block766702e2012-07-28 14:11:31 +02002109 /*
2110 * First check if we already did a call to this function with the same
2111 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2112 * return the cached result.
2113 */
Alexander Block31db9f72012-07-25 23:19:24 +02002114 nce = name_cache_search(sctx, ino, gen);
2115 if (nce) {
2116 if (ino < sctx->send_progress && nce->need_later_update) {
2117 name_cache_delete(sctx, nce);
2118 kfree(nce);
2119 nce = NULL;
2120 } else {
2121 name_cache_used(sctx, nce);
2122 *parent_ino = nce->parent_ino;
2123 *parent_gen = nce->parent_gen;
2124 ret = fs_path_add(dest, nce->name, nce->name_len);
2125 if (ret < 0)
2126 goto out;
2127 ret = nce->ret;
2128 goto out;
2129 }
2130 }
2131
Alexander Block766702e2012-07-28 14:11:31 +02002132 /*
2133 * If the inode is not existent yet, add the orphan name and return 1.
2134 * This should only happen for the parent dir that we determine in
2135 * __record_new_ref
2136 */
Alexander Block31db9f72012-07-25 23:19:24 +02002137 ret = is_inode_existent(sctx, ino, gen);
2138 if (ret < 0)
2139 goto out;
2140
2141 if (!ret) {
2142 ret = gen_unique_name(sctx, ino, gen, dest);
2143 if (ret < 0)
2144 goto out;
2145 ret = 1;
2146 goto out_cache;
2147 }
2148
Alexander Block766702e2012-07-28 14:11:31 +02002149 /*
2150 * Depending on whether the inode was already processed or not, use
2151 * send_root or parent_root for ref lookup.
2152 */
Filipe Mananabf0d1f42014-02-21 00:01:32 +00002153 if (ino < sctx->send_progress)
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002154 ret = get_first_ref(sctx->send_root, ino,
2155 parent_ino, parent_gen, dest);
Alexander Block31db9f72012-07-25 23:19:24 +02002156 else
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002157 ret = get_first_ref(sctx->parent_root, ino,
2158 parent_ino, parent_gen, dest);
Alexander Block31db9f72012-07-25 23:19:24 +02002159 if (ret < 0)
2160 goto out;
2161
Alexander Block766702e2012-07-28 14:11:31 +02002162 /*
2163 * Check if the ref was overwritten by an inode's ref that was processed
2164 * earlier. If yes, treat as orphan and return 1.
2165 */
Alexander Block31db9f72012-07-25 23:19:24 +02002166 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2167 dest->start, dest->end - dest->start);
2168 if (ret < 0)
2169 goto out;
2170 if (ret) {
2171 fs_path_reset(dest);
2172 ret = gen_unique_name(sctx, ino, gen, dest);
2173 if (ret < 0)
2174 goto out;
2175 ret = 1;
2176 }
2177
2178out_cache:
Alexander Block766702e2012-07-28 14:11:31 +02002179 /*
2180 * Store the result of the lookup in the name cache.
2181 */
Alexander Block31db9f72012-07-25 23:19:24 +02002182 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2183 if (!nce) {
2184 ret = -ENOMEM;
2185 goto out;
2186 }
2187
2188 nce->ino = ino;
2189 nce->gen = gen;
2190 nce->parent_ino = *parent_ino;
2191 nce->parent_gen = *parent_gen;
2192 nce->name_len = fs_path_len(dest);
2193 nce->ret = ret;
2194 strcpy(nce->name, dest->start);
Alexander Block31db9f72012-07-25 23:19:24 +02002195
2196 if (ino < sctx->send_progress)
2197 nce->need_later_update = 0;
2198 else
2199 nce->need_later_update = 1;
2200
2201 nce_ret = name_cache_insert(sctx, nce);
2202 if (nce_ret < 0)
2203 ret = nce_ret;
2204 name_cache_clean_unused(sctx);
2205
2206out:
Alexander Block31db9f72012-07-25 23:19:24 +02002207 return ret;
2208}
2209
2210/*
2211 * Magic happens here. This function returns the first ref to an inode as it
2212 * would look like while receiving the stream at this point in time.
2213 * We walk the path up to the root. For every inode in between, we check if it
2214 * was already processed/sent. If yes, we continue with the parent as found
2215 * in send_root. If not, we continue with the parent as found in parent_root.
2216 * If we encounter an inode that was deleted at this point in time, we use the
2217 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2218 * that were not created yet and overwritten inodes/refs.
2219 *
2220 * When do we have have orphan inodes:
2221 * 1. When an inode is freshly created and thus no valid refs are available yet
2222 * 2. When a directory lost all it's refs (deleted) but still has dir items
2223 * inside which were not processed yet (pending for move/delete). If anyone
2224 * tried to get the path to the dir items, it would get a path inside that
2225 * orphan directory.
2226 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2227 * of an unprocessed inode. If in that case the first ref would be
2228 * overwritten, the overwritten inode gets "orphanized". Later when we
2229 * process this overwritten inode, it is restored at a new place by moving
2230 * the orphan inode.
2231 *
2232 * sctx->send_progress tells this function at which point in time receiving
2233 * would be.
2234 */
2235static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2236 struct fs_path *dest)
2237{
2238 int ret = 0;
2239 struct fs_path *name = NULL;
2240 u64 parent_inode = 0;
2241 u64 parent_gen = 0;
2242 int stop = 0;
2243
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002244 name = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002245 if (!name) {
2246 ret = -ENOMEM;
2247 goto out;
2248 }
2249
2250 dest->reversed = 1;
2251 fs_path_reset(dest);
2252
2253 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
Filipe Manana8b191a62015-04-09 14:09:14 +01002254 struct waiting_dir_move *wdm;
2255
Alexander Block31db9f72012-07-25 23:19:24 +02002256 fs_path_reset(name);
2257
Filipe Manana9dc44212014-02-19 14:31:44 +00002258 if (is_waiting_for_rm(sctx, ino)) {
2259 ret = gen_unique_name(sctx, ino, gen, name);
2260 if (ret < 0)
2261 goto out;
2262 ret = fs_path_add_path(dest, name);
2263 break;
2264 }
2265
Filipe Manana8b191a62015-04-09 14:09:14 +01002266 wdm = get_waiting_dir_move(sctx, ino);
2267 if (wdm && wdm->orphanized) {
2268 ret = gen_unique_name(sctx, ino, gen, name);
2269 stop = 1;
2270 } else if (wdm) {
Filipe Mananabf0d1f42014-02-21 00:01:32 +00002271 ret = get_first_ref(sctx->parent_root, ino,
2272 &parent_inode, &parent_gen, name);
2273 } else {
2274 ret = __get_cur_name_and_parent(sctx, ino, gen,
2275 &parent_inode,
2276 &parent_gen, name);
2277 if (ret)
2278 stop = 1;
2279 }
2280
Alexander Block31db9f72012-07-25 23:19:24 +02002281 if (ret < 0)
2282 goto out;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002283
Alexander Block31db9f72012-07-25 23:19:24 +02002284 ret = fs_path_add_path(dest, name);
2285 if (ret < 0)
2286 goto out;
2287
2288 ino = parent_inode;
2289 gen = parent_gen;
2290 }
2291
2292out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002293 fs_path_free(name);
Alexander Block31db9f72012-07-25 23:19:24 +02002294 if (!ret)
2295 fs_path_unreverse(dest);
2296 return ret;
2297}
2298
2299/*
Alexander Block31db9f72012-07-25 23:19:24 +02002300 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2301 */
2302static int send_subvol_begin(struct send_ctx *sctx)
2303{
2304 int ret;
2305 struct btrfs_root *send_root = sctx->send_root;
2306 struct btrfs_root *parent_root = sctx->parent_root;
2307 struct btrfs_path *path;
2308 struct btrfs_key key;
2309 struct btrfs_root_ref *ref;
2310 struct extent_buffer *leaf;
2311 char *name = NULL;
2312 int namelen;
2313
Wang Shilongffcfaf82014-01-15 00:26:43 +08002314 path = btrfs_alloc_path();
Alexander Block31db9f72012-07-25 23:19:24 +02002315 if (!path)
2316 return -ENOMEM;
2317
2318 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2319 if (!name) {
2320 btrfs_free_path(path);
2321 return -ENOMEM;
2322 }
2323
2324 key.objectid = send_root->objectid;
2325 key.type = BTRFS_ROOT_BACKREF_KEY;
2326 key.offset = 0;
2327
2328 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2329 &key, path, 1, 0);
2330 if (ret < 0)
2331 goto out;
2332 if (ret) {
2333 ret = -ENOENT;
2334 goto out;
2335 }
2336
2337 leaf = path->nodes[0];
2338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2339 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2340 key.objectid != send_root->objectid) {
2341 ret = -ENOENT;
2342 goto out;
2343 }
2344 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2345 namelen = btrfs_root_ref_name_len(leaf, ref);
2346 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2347 btrfs_release_path(path);
2348
Alexander Block31db9f72012-07-25 23:19:24 +02002349 if (parent_root) {
2350 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2351 if (ret < 0)
2352 goto out;
2353 } else {
2354 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2355 if (ret < 0)
2356 goto out;
2357 }
2358
2359 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
Robin Ruedeb96b1db2015-09-30 21:23:33 +02002360
2361 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2362 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2363 sctx->send_root->root_item.received_uuid);
2364 else
2365 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2366 sctx->send_root->root_item.uuid);
2367
Alexander Block31db9f72012-07-25 23:19:24 +02002368 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
Filipe David Borba Manana5a0f4e22013-12-03 15:55:48 +00002369 le64_to_cpu(sctx->send_root->root_item.ctransid));
Alexander Block31db9f72012-07-25 23:19:24 +02002370 if (parent_root) {
Josef Bacik37b8d272015-06-04 17:17:25 -04002371 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2372 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2373 parent_root->root_item.received_uuid);
2374 else
2375 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2376 parent_root->root_item.uuid);
Alexander Block31db9f72012-07-25 23:19:24 +02002377 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
Filipe David Borba Manana5a0f4e22013-12-03 15:55:48 +00002378 le64_to_cpu(sctx->parent_root->root_item.ctransid));
Alexander Block31db9f72012-07-25 23:19:24 +02002379 }
2380
2381 ret = send_cmd(sctx);
2382
2383tlv_put_failure:
2384out:
2385 btrfs_free_path(path);
2386 kfree(name);
2387 return ret;
2388}
2389
2390static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2391{
2392 int ret = 0;
2393 struct fs_path *p;
2394
2395verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2396
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002397 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002398 if (!p)
2399 return -ENOMEM;
2400
2401 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2402 if (ret < 0)
2403 goto out;
2404
2405 ret = get_cur_path(sctx, ino, gen, p);
2406 if (ret < 0)
2407 goto out;
2408 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2409 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2410
2411 ret = send_cmd(sctx);
2412
2413tlv_put_failure:
2414out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002415 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002416 return ret;
2417}
2418
2419static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2420{
2421 int ret = 0;
2422 struct fs_path *p;
2423
2424verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2425
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002426 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002427 if (!p)
2428 return -ENOMEM;
2429
2430 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2431 if (ret < 0)
2432 goto out;
2433
2434 ret = get_cur_path(sctx, ino, gen, p);
2435 if (ret < 0)
2436 goto out;
2437 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2438 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2439
2440 ret = send_cmd(sctx);
2441
2442tlv_put_failure:
2443out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002444 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002445 return ret;
2446}
2447
2448static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2449{
2450 int ret = 0;
2451 struct fs_path *p;
2452
2453verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2454
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002455 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002456 if (!p)
2457 return -ENOMEM;
2458
2459 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2460 if (ret < 0)
2461 goto out;
2462
2463 ret = get_cur_path(sctx, ino, gen, p);
2464 if (ret < 0)
2465 goto out;
2466 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2467 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2468 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2469
2470 ret = send_cmd(sctx);
2471
2472tlv_put_failure:
2473out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002474 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002475 return ret;
2476}
2477
2478static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2479{
2480 int ret = 0;
2481 struct fs_path *p = NULL;
2482 struct btrfs_inode_item *ii;
2483 struct btrfs_path *path = NULL;
2484 struct extent_buffer *eb;
2485 struct btrfs_key key;
2486 int slot;
2487
2488verbose_printk("btrfs: send_utimes %llu\n", ino);
2489
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002490 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002491 if (!p)
2492 return -ENOMEM;
2493
2494 path = alloc_path_for_send();
2495 if (!path) {
2496 ret = -ENOMEM;
2497 goto out;
2498 }
2499
2500 key.objectid = ino;
2501 key.type = BTRFS_INODE_ITEM_KEY;
2502 key.offset = 0;
2503 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2504 if (ret < 0)
2505 goto out;
2506
2507 eb = path->nodes[0];
2508 slot = path->slots[0];
2509 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2510
2511 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2512 if (ret < 0)
2513 goto out;
2514
2515 ret = get_cur_path(sctx, ino, gen, p);
2516 if (ret < 0)
2517 goto out;
2518 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
David Sterbaa937b972014-12-12 17:39:12 +01002519 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2520 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2521 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
Alexander Block766702e2012-07-28 14:11:31 +02002522 /* TODO Add otime support when the otime patches get into upstream */
Alexander Block31db9f72012-07-25 23:19:24 +02002523
2524 ret = send_cmd(sctx);
2525
2526tlv_put_failure:
2527out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002528 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002529 btrfs_free_path(path);
2530 return ret;
2531}
2532
2533/*
2534 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2535 * a valid path yet because we did not process the refs yet. So, the inode
2536 * is created as orphan.
2537 */
Alexander Block1f4692d2012-07-28 10:42:24 +02002538static int send_create_inode(struct send_ctx *sctx, u64 ino)
Alexander Block31db9f72012-07-25 23:19:24 +02002539{
2540 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02002541 struct fs_path *p;
Alexander Block31db9f72012-07-25 23:19:24 +02002542 int cmd;
Alexander Block1f4692d2012-07-28 10:42:24 +02002543 u64 gen;
Alexander Block31db9f72012-07-25 23:19:24 +02002544 u64 mode;
Alexander Block1f4692d2012-07-28 10:42:24 +02002545 u64 rdev;
Alexander Block31db9f72012-07-25 23:19:24 +02002546
Alexander Block1f4692d2012-07-28 10:42:24 +02002547verbose_printk("btrfs: send_create_inode %llu\n", ino);
Alexander Block31db9f72012-07-25 23:19:24 +02002548
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002549 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002550 if (!p)
2551 return -ENOMEM;
2552
Liu Bo644d1942014-02-27 17:29:01 +08002553 if (ino != sctx->cur_ino) {
2554 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2555 NULL, NULL, &rdev);
2556 if (ret < 0)
2557 goto out;
2558 } else {
2559 gen = sctx->cur_inode_gen;
2560 mode = sctx->cur_inode_mode;
2561 rdev = sctx->cur_inode_rdev;
2562 }
Alexander Block31db9f72012-07-25 23:19:24 +02002563
Alexander Blocke938c8a2012-07-28 16:33:49 +02002564 if (S_ISREG(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002565 cmd = BTRFS_SEND_C_MKFILE;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002566 } else if (S_ISDIR(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002567 cmd = BTRFS_SEND_C_MKDIR;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002568 } else if (S_ISLNK(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002569 cmd = BTRFS_SEND_C_SYMLINK;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002570 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002571 cmd = BTRFS_SEND_C_MKNOD;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002572 } else if (S_ISFIFO(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002573 cmd = BTRFS_SEND_C_MKFIFO;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002574 } else if (S_ISSOCK(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002575 cmd = BTRFS_SEND_C_MKSOCK;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002576 } else {
David Sterbaf14d1042015-10-08 11:37:06 +02002577 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
Alexander Block31db9f72012-07-25 23:19:24 +02002578 (int)(mode & S_IFMT));
2579 ret = -ENOTSUPP;
2580 goto out;
2581 }
2582
2583 ret = begin_cmd(sctx, cmd);
2584 if (ret < 0)
2585 goto out;
2586
Alexander Block1f4692d2012-07-28 10:42:24 +02002587 ret = gen_unique_name(sctx, ino, gen, p);
Alexander Block31db9f72012-07-25 23:19:24 +02002588 if (ret < 0)
2589 goto out;
2590
2591 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
Alexander Block1f4692d2012-07-28 10:42:24 +02002592 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
Alexander Block31db9f72012-07-25 23:19:24 +02002593
2594 if (S_ISLNK(mode)) {
2595 fs_path_reset(p);
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002596 ret = read_symlink(sctx->send_root, ino, p);
Alexander Block31db9f72012-07-25 23:19:24 +02002597 if (ret < 0)
2598 goto out;
2599 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2600 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2601 S_ISFIFO(mode) || S_ISSOCK(mode)) {
Arne Jansend79e5042012-10-15 18:28:46 +00002602 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2603 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
Alexander Block31db9f72012-07-25 23:19:24 +02002604 }
2605
2606 ret = send_cmd(sctx);
2607 if (ret < 0)
2608 goto out;
2609
2610
2611tlv_put_failure:
2612out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002613 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002614 return ret;
2615}
2616
Alexander Block1f4692d2012-07-28 10:42:24 +02002617/*
2618 * We need some special handling for inodes that get processed before the parent
2619 * directory got created. See process_recorded_refs for details.
2620 * This function does the check if we already created the dir out of order.
2621 */
2622static int did_create_dir(struct send_ctx *sctx, u64 dir)
2623{
2624 int ret = 0;
2625 struct btrfs_path *path = NULL;
2626 struct btrfs_key key;
2627 struct btrfs_key found_key;
2628 struct btrfs_key di_key;
2629 struct extent_buffer *eb;
2630 struct btrfs_dir_item *di;
2631 int slot;
2632
2633 path = alloc_path_for_send();
2634 if (!path) {
2635 ret = -ENOMEM;
2636 goto out;
2637 }
2638
2639 key.objectid = dir;
2640 key.type = BTRFS_DIR_INDEX_KEY;
2641 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002642 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2643 if (ret < 0)
2644 goto out;
2645
Alexander Block1f4692d2012-07-28 10:42:24 +02002646 while (1) {
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002647 eb = path->nodes[0];
2648 slot = path->slots[0];
2649 if (slot >= btrfs_header_nritems(eb)) {
2650 ret = btrfs_next_leaf(sctx->send_root, path);
2651 if (ret < 0) {
2652 goto out;
2653 } else if (ret > 0) {
2654 ret = 0;
2655 break;
2656 }
2657 continue;
Alexander Block1f4692d2012-07-28 10:42:24 +02002658 }
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002659
2660 btrfs_item_key_to_cpu(eb, &found_key, slot);
2661 if (found_key.objectid != key.objectid ||
Alexander Block1f4692d2012-07-28 10:42:24 +02002662 found_key.type != key.type) {
2663 ret = 0;
2664 goto out;
2665 }
2666
2667 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2668 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2669
Josef Bacika0525412013-08-12 10:56:14 -04002670 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2671 di_key.objectid < sctx->send_progress) {
Alexander Block1f4692d2012-07-28 10:42:24 +02002672 ret = 1;
2673 goto out;
2674 }
2675
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002676 path->slots[0]++;
Alexander Block1f4692d2012-07-28 10:42:24 +02002677 }
2678
2679out:
2680 btrfs_free_path(path);
2681 return ret;
2682}
2683
2684/*
2685 * Only creates the inode if it is:
2686 * 1. Not a directory
2687 * 2. Or a directory which was not created already due to out of order
2688 * directories. See did_create_dir and process_recorded_refs for details.
2689 */
2690static int send_create_inode_if_needed(struct send_ctx *sctx)
2691{
2692 int ret;
2693
2694 if (S_ISDIR(sctx->cur_inode_mode)) {
2695 ret = did_create_dir(sctx, sctx->cur_ino);
2696 if (ret < 0)
2697 goto out;
2698 if (ret) {
2699 ret = 0;
2700 goto out;
2701 }
2702 }
2703
2704 ret = send_create_inode(sctx, sctx->cur_ino);
2705 if (ret < 0)
2706 goto out;
2707
2708out:
2709 return ret;
2710}
2711
Alexander Block31db9f72012-07-25 23:19:24 +02002712struct recorded_ref {
2713 struct list_head list;
2714 char *dir_path;
2715 char *name;
2716 struct fs_path *full_path;
2717 u64 dir;
2718 u64 dir_gen;
2719 int dir_path_len;
2720 int name_len;
2721};
2722
2723/*
2724 * We need to process new refs before deleted refs, but compare_tree gives us
2725 * everything mixed. So we first record all refs and later process them.
2726 * This function is a helper to record one ref.
2727 */
Liu Boa4d96d62014-03-03 21:31:03 +08002728static int __record_ref(struct list_head *head, u64 dir,
Alexander Block31db9f72012-07-25 23:19:24 +02002729 u64 dir_gen, struct fs_path *path)
2730{
2731 struct recorded_ref *ref;
Alexander Block31db9f72012-07-25 23:19:24 +02002732
2733 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2734 if (!ref)
2735 return -ENOMEM;
2736
2737 ref->dir = dir;
2738 ref->dir_gen = dir_gen;
2739 ref->full_path = path;
2740
Andy Shevchenkoed848852013-08-21 10:32:13 +03002741 ref->name = (char *)kbasename(ref->full_path->start);
2742 ref->name_len = ref->full_path->end - ref->name;
2743 ref->dir_path = ref->full_path->start;
2744 if (ref->name == ref->full_path->start)
Alexander Block31db9f72012-07-25 23:19:24 +02002745 ref->dir_path_len = 0;
Andy Shevchenkoed848852013-08-21 10:32:13 +03002746 else
Alexander Block31db9f72012-07-25 23:19:24 +02002747 ref->dir_path_len = ref->full_path->end -
2748 ref->full_path->start - 1 - ref->name_len;
Alexander Block31db9f72012-07-25 23:19:24 +02002749
2750 list_add_tail(&ref->list, head);
2751 return 0;
2752}
2753
Josef Bacikba5e8f22013-08-16 16:52:55 -04002754static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2755{
2756 struct recorded_ref *new;
2757
2758 new = kmalloc(sizeof(*ref), GFP_NOFS);
2759 if (!new)
2760 return -ENOMEM;
2761
2762 new->dir = ref->dir;
2763 new->dir_gen = ref->dir_gen;
2764 new->full_path = NULL;
2765 INIT_LIST_HEAD(&new->list);
2766 list_add_tail(&new->list, list);
2767 return 0;
2768}
2769
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002770static void __free_recorded_refs(struct list_head *head)
Alexander Block31db9f72012-07-25 23:19:24 +02002771{
2772 struct recorded_ref *cur;
Alexander Block31db9f72012-07-25 23:19:24 +02002773
Alexander Blocke938c8a2012-07-28 16:33:49 +02002774 while (!list_empty(head)) {
2775 cur = list_entry(head->next, struct recorded_ref, list);
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002776 fs_path_free(cur->full_path);
Alexander Blocke938c8a2012-07-28 16:33:49 +02002777 list_del(&cur->list);
Alexander Block31db9f72012-07-25 23:19:24 +02002778 kfree(cur);
2779 }
Alexander Block31db9f72012-07-25 23:19:24 +02002780}
2781
2782static void free_recorded_refs(struct send_ctx *sctx)
2783{
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002784 __free_recorded_refs(&sctx->new_refs);
2785 __free_recorded_refs(&sctx->deleted_refs);
Alexander Block31db9f72012-07-25 23:19:24 +02002786}
2787
2788/*
Alexander Block766702e2012-07-28 14:11:31 +02002789 * Renames/moves a file/dir to its orphan name. Used when the first
Alexander Block31db9f72012-07-25 23:19:24 +02002790 * ref of an unprocessed inode gets overwritten and for all non empty
2791 * directories.
2792 */
2793static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2794 struct fs_path *path)
2795{
2796 int ret;
2797 struct fs_path *orphan;
2798
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002799 orphan = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002800 if (!orphan)
2801 return -ENOMEM;
2802
2803 ret = gen_unique_name(sctx, ino, gen, orphan);
2804 if (ret < 0)
2805 goto out;
2806
2807 ret = send_rename(sctx, path, orphan);
2808
2809out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002810 fs_path_free(orphan);
Alexander Block31db9f72012-07-25 23:19:24 +02002811 return ret;
2812}
2813
Filipe Manana9dc44212014-02-19 14:31:44 +00002814static struct orphan_dir_info *
2815add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2816{
2817 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2818 struct rb_node *parent = NULL;
2819 struct orphan_dir_info *entry, *odi;
2820
2821 odi = kmalloc(sizeof(*odi), GFP_NOFS);
2822 if (!odi)
2823 return ERR_PTR(-ENOMEM);
2824 odi->ino = dir_ino;
2825 odi->gen = 0;
2826
2827 while (*p) {
2828 parent = *p;
2829 entry = rb_entry(parent, struct orphan_dir_info, node);
2830 if (dir_ino < entry->ino) {
2831 p = &(*p)->rb_left;
2832 } else if (dir_ino > entry->ino) {
2833 p = &(*p)->rb_right;
2834 } else {
2835 kfree(odi);
2836 return entry;
2837 }
2838 }
2839
2840 rb_link_node(&odi->node, parent, p);
2841 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2842 return odi;
2843}
2844
2845static struct orphan_dir_info *
2846get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2847{
2848 struct rb_node *n = sctx->orphan_dirs.rb_node;
2849 struct orphan_dir_info *entry;
2850
2851 while (n) {
2852 entry = rb_entry(n, struct orphan_dir_info, node);
2853 if (dir_ino < entry->ino)
2854 n = n->rb_left;
2855 else if (dir_ino > entry->ino)
2856 n = n->rb_right;
2857 else
2858 return entry;
2859 }
2860 return NULL;
2861}
2862
2863static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2864{
2865 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2866
2867 return odi != NULL;
2868}
2869
2870static void free_orphan_dir_info(struct send_ctx *sctx,
2871 struct orphan_dir_info *odi)
2872{
2873 if (!odi)
2874 return;
2875 rb_erase(&odi->node, &sctx->orphan_dirs);
2876 kfree(odi);
2877}
2878
Alexander Block31db9f72012-07-25 23:19:24 +02002879/*
2880 * Returns 1 if a directory can be removed at this point in time.
2881 * We check this by iterating all dir items and checking if the inode behind
2882 * the dir item was already processed.
2883 */
Filipe Manana9dc44212014-02-19 14:31:44 +00002884static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2885 u64 send_progress)
Alexander Block31db9f72012-07-25 23:19:24 +02002886{
2887 int ret = 0;
2888 struct btrfs_root *root = sctx->parent_root;
2889 struct btrfs_path *path;
2890 struct btrfs_key key;
2891 struct btrfs_key found_key;
2892 struct btrfs_key loc;
2893 struct btrfs_dir_item *di;
2894
Alexander Block6d85ed02012-08-01 14:48:59 +02002895 /*
2896 * Don't try to rmdir the top/root subvolume dir.
2897 */
2898 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2899 return 0;
2900
Alexander Block31db9f72012-07-25 23:19:24 +02002901 path = alloc_path_for_send();
2902 if (!path)
2903 return -ENOMEM;
2904
2905 key.objectid = dir;
2906 key.type = BTRFS_DIR_INDEX_KEY;
2907 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002908 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2909 if (ret < 0)
2910 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02002911
2912 while (1) {
Filipe Manana9dc44212014-02-19 14:31:44 +00002913 struct waiting_dir_move *dm;
2914
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002915 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2916 ret = btrfs_next_leaf(root, path);
2917 if (ret < 0)
2918 goto out;
2919 else if (ret > 0)
2920 break;
2921 continue;
Alexander Block31db9f72012-07-25 23:19:24 +02002922 }
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002923 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2924 path->slots[0]);
2925 if (found_key.objectid != key.objectid ||
2926 found_key.type != key.type)
Alexander Block31db9f72012-07-25 23:19:24 +02002927 break;
Alexander Block31db9f72012-07-25 23:19:24 +02002928
2929 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2930 struct btrfs_dir_item);
2931 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2932
Filipe Manana9dc44212014-02-19 14:31:44 +00002933 dm = get_waiting_dir_move(sctx, loc.objectid);
2934 if (dm) {
2935 struct orphan_dir_info *odi;
2936
2937 odi = add_orphan_dir_info(sctx, dir);
2938 if (IS_ERR(odi)) {
2939 ret = PTR_ERR(odi);
2940 goto out;
2941 }
2942 odi->gen = dir_gen;
2943 dm->rmdir_ino = dir;
2944 ret = 0;
2945 goto out;
2946 }
2947
Alexander Block31db9f72012-07-25 23:19:24 +02002948 if (loc.objectid > send_progress) {
2949 ret = 0;
2950 goto out;
2951 }
2952
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002953 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02002954 }
2955
2956 ret = 1;
2957
2958out:
2959 btrfs_free_path(path);
2960 return ret;
2961}
2962
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002963static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2964{
Filipe Manana9dc44212014-02-19 14:31:44 +00002965 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002966
Filipe Manana9dc44212014-02-19 14:31:44 +00002967 return entry != NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002968}
2969
Filipe Manana8b191a62015-04-09 14:09:14 +01002970static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002971{
2972 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2973 struct rb_node *parent = NULL;
2974 struct waiting_dir_move *entry, *dm;
2975
2976 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2977 if (!dm)
2978 return -ENOMEM;
2979 dm->ino = ino;
Filipe Manana9dc44212014-02-19 14:31:44 +00002980 dm->rmdir_ino = 0;
Filipe Manana8b191a62015-04-09 14:09:14 +01002981 dm->orphanized = orphanized;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002982
2983 while (*p) {
2984 parent = *p;
2985 entry = rb_entry(parent, struct waiting_dir_move, node);
2986 if (ino < entry->ino) {
2987 p = &(*p)->rb_left;
2988 } else if (ino > entry->ino) {
2989 p = &(*p)->rb_right;
2990 } else {
2991 kfree(dm);
2992 return -EEXIST;
2993 }
2994 }
2995
2996 rb_link_node(&dm->node, parent, p);
2997 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2998 return 0;
2999}
3000
Filipe Manana9dc44212014-02-19 14:31:44 +00003001static struct waiting_dir_move *
3002get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003003{
3004 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3005 struct waiting_dir_move *entry;
3006
3007 while (n) {
3008 entry = rb_entry(n, struct waiting_dir_move, node);
Filipe Manana9dc44212014-02-19 14:31:44 +00003009 if (ino < entry->ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003010 n = n->rb_left;
Filipe Manana9dc44212014-02-19 14:31:44 +00003011 else if (ino > entry->ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003012 n = n->rb_right;
Filipe Manana9dc44212014-02-19 14:31:44 +00003013 else
3014 return entry;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003015 }
Filipe Manana9dc44212014-02-19 14:31:44 +00003016 return NULL;
3017}
3018
3019static void free_waiting_dir_move(struct send_ctx *sctx,
3020 struct waiting_dir_move *dm)
3021{
3022 if (!dm)
3023 return;
3024 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3025 kfree(dm);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003026}
3027
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003028static int add_pending_dir_move(struct send_ctx *sctx,
3029 u64 ino,
3030 u64 ino_gen,
Filipe Mananaf9594922014-03-27 20:14:01 +00003031 u64 parent_ino,
3032 struct list_head *new_refs,
Filipe Manana84471e22015-02-28 22:29:22 +00003033 struct list_head *deleted_refs,
3034 const bool is_orphan)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003035{
3036 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3037 struct rb_node *parent = NULL;
Chris Mason73b802f2014-03-21 15:30:44 -07003038 struct pending_dir_move *entry = NULL, *pm;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003039 struct recorded_ref *cur;
3040 int exists = 0;
3041 int ret;
3042
3043 pm = kmalloc(sizeof(*pm), GFP_NOFS);
3044 if (!pm)
3045 return -ENOMEM;
3046 pm->parent_ino = parent_ino;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003047 pm->ino = ino;
3048 pm->gen = ino_gen;
Filipe Manana84471e22015-02-28 22:29:22 +00003049 pm->is_orphan = is_orphan;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003050 INIT_LIST_HEAD(&pm->list);
3051 INIT_LIST_HEAD(&pm->update_refs);
3052 RB_CLEAR_NODE(&pm->node);
3053
3054 while (*p) {
3055 parent = *p;
3056 entry = rb_entry(parent, struct pending_dir_move, node);
3057 if (parent_ino < entry->parent_ino) {
3058 p = &(*p)->rb_left;
3059 } else if (parent_ino > entry->parent_ino) {
3060 p = &(*p)->rb_right;
3061 } else {
3062 exists = 1;
3063 break;
3064 }
3065 }
3066
Filipe Mananaf9594922014-03-27 20:14:01 +00003067 list_for_each_entry(cur, deleted_refs, list) {
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003068 ret = dup_ref(cur, &pm->update_refs);
3069 if (ret < 0)
3070 goto out;
3071 }
Filipe Mananaf9594922014-03-27 20:14:01 +00003072 list_for_each_entry(cur, new_refs, list) {
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003073 ret = dup_ref(cur, &pm->update_refs);
3074 if (ret < 0)
3075 goto out;
3076 }
3077
Filipe Manana8b191a62015-04-09 14:09:14 +01003078 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003079 if (ret)
3080 goto out;
3081
3082 if (exists) {
3083 list_add_tail(&pm->list, &entry->list);
3084 } else {
3085 rb_link_node(&pm->node, parent, p);
3086 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3087 }
3088 ret = 0;
3089out:
3090 if (ret) {
3091 __free_recorded_refs(&pm->update_refs);
3092 kfree(pm);
3093 }
3094 return ret;
3095}
3096
3097static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3098 u64 parent_ino)
3099{
3100 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3101 struct pending_dir_move *entry;
3102
3103 while (n) {
3104 entry = rb_entry(n, struct pending_dir_move, node);
3105 if (parent_ino < entry->parent_ino)
3106 n = n->rb_left;
3107 else if (parent_ino > entry->parent_ino)
3108 n = n->rb_right;
3109 else
3110 return entry;
3111 }
3112 return NULL;
3113}
3114
3115static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3116{
3117 struct fs_path *from_path = NULL;
3118 struct fs_path *to_path = NULL;
Filipe Manana2b863a12014-02-16 13:43:11 +00003119 struct fs_path *name = NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003120 u64 orig_progress = sctx->send_progress;
3121 struct recorded_ref *cur;
Filipe Manana2b863a12014-02-16 13:43:11 +00003122 u64 parent_ino, parent_gen;
Filipe Manana9dc44212014-02-19 14:31:44 +00003123 struct waiting_dir_move *dm = NULL;
3124 u64 rmdir_ino = 0;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003125 int ret;
3126
Filipe Manana2b863a12014-02-16 13:43:11 +00003127 name = fs_path_alloc();
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003128 from_path = fs_path_alloc();
Filipe Manana2b863a12014-02-16 13:43:11 +00003129 if (!name || !from_path) {
3130 ret = -ENOMEM;
3131 goto out;
3132 }
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003133
Filipe Manana9dc44212014-02-19 14:31:44 +00003134 dm = get_waiting_dir_move(sctx, pm->ino);
3135 ASSERT(dm);
3136 rmdir_ino = dm->rmdir_ino;
3137 free_waiting_dir_move(sctx, dm);
Filipe Manana2b863a12014-02-16 13:43:11 +00003138
Filipe Manana84471e22015-02-28 22:29:22 +00003139 if (pm->is_orphan) {
3140 ret = gen_unique_name(sctx, pm->ino,
3141 pm->gen, from_path);
3142 } else {
3143 ret = get_first_ref(sctx->parent_root, pm->ino,
3144 &parent_ino, &parent_gen, name);
3145 if (ret < 0)
3146 goto out;
3147 ret = get_cur_path(sctx, parent_ino, parent_gen,
3148 from_path);
3149 if (ret < 0)
3150 goto out;
3151 ret = fs_path_add_path(from_path, name);
3152 }
Filipe Mananac992ec92014-03-22 17:15:24 +00003153 if (ret < 0)
3154 goto out;
3155
Filipe Mananaf9594922014-03-27 20:14:01 +00003156 sctx->send_progress = sctx->cur_ino + 1;
Filipe Mananac992ec92014-03-22 17:15:24 +00003157 fs_path_reset(name);
3158 to_path = name;
3159 name = NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003160 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3161 if (ret < 0)
3162 goto out;
3163
3164 ret = send_rename(sctx, from_path, to_path);
3165 if (ret < 0)
3166 goto out;
3167
Filipe Manana9dc44212014-02-19 14:31:44 +00003168 if (rmdir_ino) {
3169 struct orphan_dir_info *odi;
3170
3171 odi = get_orphan_dir_info(sctx, rmdir_ino);
3172 if (!odi) {
3173 /* already deleted */
3174 goto finish;
3175 }
3176 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1);
3177 if (ret < 0)
3178 goto out;
3179 if (!ret)
3180 goto finish;
3181
3182 name = fs_path_alloc();
3183 if (!name) {
3184 ret = -ENOMEM;
3185 goto out;
3186 }
3187 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3188 if (ret < 0)
3189 goto out;
3190 ret = send_rmdir(sctx, name);
3191 if (ret < 0)
3192 goto out;
3193 free_orphan_dir_info(sctx, odi);
3194 }
3195
3196finish:
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003197 ret = send_utimes(sctx, pm->ino, pm->gen);
3198 if (ret < 0)
3199 goto out;
3200
3201 /*
3202 * After rename/move, need to update the utimes of both new parent(s)
3203 * and old parent(s).
3204 */
3205 list_for_each_entry(cur, &pm->update_refs, list) {
Filipe Manana9dc44212014-02-19 14:31:44 +00003206 if (cur->dir == rmdir_ino)
3207 continue;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003208 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3209 if (ret < 0)
3210 goto out;
3211 }
3212
3213out:
Filipe Manana2b863a12014-02-16 13:43:11 +00003214 fs_path_free(name);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003215 fs_path_free(from_path);
3216 fs_path_free(to_path);
3217 sctx->send_progress = orig_progress;
3218
3219 return ret;
3220}
3221
3222static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3223{
3224 if (!list_empty(&m->list))
3225 list_del(&m->list);
3226 if (!RB_EMPTY_NODE(&m->node))
3227 rb_erase(&m->node, &sctx->pending_dir_moves);
3228 __free_recorded_refs(&m->update_refs);
3229 kfree(m);
3230}
3231
3232static void tail_append_pending_moves(struct pending_dir_move *moves,
3233 struct list_head *stack)
3234{
3235 if (list_empty(&moves->list)) {
3236 list_add_tail(&moves->list, stack);
3237 } else {
3238 LIST_HEAD(list);
3239 list_splice_init(&moves->list, &list);
3240 list_add_tail(&moves->list, stack);
3241 list_splice_tail(&list, stack);
3242 }
3243}
3244
3245static int apply_children_dir_moves(struct send_ctx *sctx)
3246{
3247 struct pending_dir_move *pm;
3248 struct list_head stack;
3249 u64 parent_ino = sctx->cur_ino;
3250 int ret = 0;
3251
3252 pm = get_pending_dir_moves(sctx, parent_ino);
3253 if (!pm)
3254 return 0;
3255
3256 INIT_LIST_HEAD(&stack);
3257 tail_append_pending_moves(pm, &stack);
3258
3259 while (!list_empty(&stack)) {
3260 pm = list_first_entry(&stack, struct pending_dir_move, list);
3261 parent_ino = pm->ino;
3262 ret = apply_dir_move(sctx, pm);
3263 free_pending_move(sctx, pm);
3264 if (ret)
3265 goto out;
3266 pm = get_pending_dir_moves(sctx, parent_ino);
3267 if (pm)
3268 tail_append_pending_moves(pm, &stack);
3269 }
3270 return 0;
3271
3272out:
3273 while (!list_empty(&stack)) {
3274 pm = list_first_entry(&stack, struct pending_dir_move, list);
3275 free_pending_move(sctx, pm);
3276 }
3277 return ret;
3278}
3279
Filipe Manana84471e22015-02-28 22:29:22 +00003280/*
3281 * We might need to delay a directory rename even when no ancestor directory
3282 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3283 * renamed. This happens when we rename a directory to the old name (the name
3284 * in the parent root) of some other unrelated directory that got its rename
3285 * delayed due to some ancestor with higher number that got renamed.
3286 *
3287 * Example:
3288 *
3289 * Parent snapshot:
3290 * . (ino 256)
3291 * |---- a/ (ino 257)
3292 * | |---- file (ino 260)
3293 * |
3294 * |---- b/ (ino 258)
3295 * |---- c/ (ino 259)
3296 *
3297 * Send snapshot:
3298 * . (ino 256)
3299 * |---- a/ (ino 258)
3300 * |---- x/ (ino 259)
3301 * |---- y/ (ino 257)
3302 * |----- file (ino 260)
3303 *
3304 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3305 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3306 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3307 * must issue is:
3308 *
3309 * 1 - rename 259 from 'c' to 'x'
3310 * 2 - rename 257 from 'a' to 'x/y'
3311 * 3 - rename 258 from 'b' to 'a'
3312 *
3313 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3314 * be done right away and < 0 on error.
3315 */
3316static int wait_for_dest_dir_move(struct send_ctx *sctx,
3317 struct recorded_ref *parent_ref,
3318 const bool is_orphan)
3319{
3320 struct btrfs_path *path;
3321 struct btrfs_key key;
3322 struct btrfs_key di_key;
3323 struct btrfs_dir_item *di;
3324 u64 left_gen;
3325 u64 right_gen;
3326 int ret = 0;
3327
3328 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3329 return 0;
3330
3331 path = alloc_path_for_send();
3332 if (!path)
3333 return -ENOMEM;
3334
3335 key.objectid = parent_ref->dir;
3336 key.type = BTRFS_DIR_ITEM_KEY;
3337 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3338
3339 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3340 if (ret < 0) {
3341 goto out;
3342 } else if (ret > 0) {
3343 ret = 0;
3344 goto out;
3345 }
3346
3347 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3348 parent_ref->name, parent_ref->name_len);
3349 if (!di) {
3350 ret = 0;
3351 goto out;
3352 }
3353 /*
3354 * di_key.objectid has the number of the inode that has a dentry in the
3355 * parent directory with the same name that sctx->cur_ino is being
3356 * renamed to. We need to check if that inode is in the send root as
3357 * well and if it is currently marked as an inode with a pending rename,
3358 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3359 * that it happens after that other inode is renamed.
3360 */
3361 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3362 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3363 ret = 0;
3364 goto out;
3365 }
3366
3367 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3368 &left_gen, NULL, NULL, NULL, NULL);
3369 if (ret < 0)
3370 goto out;
3371 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3372 &right_gen, NULL, NULL, NULL, NULL);
3373 if (ret < 0) {
3374 if (ret == -ENOENT)
3375 ret = 0;
3376 goto out;
3377 }
3378
3379 /* Different inode, no need to delay the rename of sctx->cur_ino */
3380 if (right_gen != left_gen) {
3381 ret = 0;
3382 goto out;
3383 }
3384
3385 if (is_waiting_for_move(sctx, di_key.objectid)) {
3386 ret = add_pending_dir_move(sctx,
3387 sctx->cur_ino,
3388 sctx->cur_inode_gen,
3389 di_key.objectid,
3390 &sctx->new_refs,
3391 &sctx->deleted_refs,
3392 is_orphan);
3393 if (!ret)
3394 ret = 1;
3395 }
3396out:
3397 btrfs_free_path(path);
3398 return ret;
3399}
3400
Filipe Manana80aa6022015-03-27 17:50:45 +00003401/*
3402 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3403 * Return 1 if true, 0 if false and < 0 on error.
3404 */
3405static int is_ancestor(struct btrfs_root *root,
3406 const u64 ino1,
3407 const u64 ino1_gen,
3408 const u64 ino2,
3409 struct fs_path *fs_path)
3410{
3411 u64 ino = ino2;
3412
3413 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3414 int ret;
3415 u64 parent;
3416 u64 parent_gen;
3417
3418 fs_path_reset(fs_path);
3419 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3420 if (ret < 0) {
3421 if (ret == -ENOENT && ino == ino2)
3422 ret = 0;
3423 return ret;
3424 }
3425 if (parent == ino1)
3426 return parent_gen == ino1_gen ? 1 : 0;
3427 ino = parent;
3428 }
3429 return 0;
3430}
3431
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003432static int wait_for_parent_move(struct send_ctx *sctx,
Filipe Manana8b191a62015-04-09 14:09:14 +01003433 struct recorded_ref *parent_ref,
3434 const bool is_orphan)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003435{
Filipe Mananaf9594922014-03-27 20:14:01 +00003436 int ret = 0;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003437 u64 ino = parent_ref->dir;
3438 u64 parent_ino_before, parent_ino_after;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003439 struct fs_path *path_before = NULL;
3440 struct fs_path *path_after = NULL;
3441 int len1, len2;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003442
3443 path_after = fs_path_alloc();
Filipe Mananaf9594922014-03-27 20:14:01 +00003444 path_before = fs_path_alloc();
3445 if (!path_after || !path_before) {
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003446 ret = -ENOMEM;
3447 goto out;
3448 }
3449
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003450 /*
Filipe Mananaf9594922014-03-27 20:14:01 +00003451 * Our current directory inode may not yet be renamed/moved because some
3452 * ancestor (immediate or not) has to be renamed/moved first. So find if
3453 * such ancestor exists and make sure our own rename/move happens after
Filipe Manana80aa6022015-03-27 17:50:45 +00003454 * that ancestor is processed to avoid path build infinite loops (done
3455 * at get_cur_path()).
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003456 */
Filipe Mananaf9594922014-03-27 20:14:01 +00003457 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3458 if (is_waiting_for_move(sctx, ino)) {
Filipe Manana80aa6022015-03-27 17:50:45 +00003459 /*
3460 * If the current inode is an ancestor of ino in the
3461 * parent root, we need to delay the rename of the
3462 * current inode, otherwise don't delayed the rename
3463 * because we can end up with a circular dependency
3464 * of renames, resulting in some directories never
3465 * getting the respective rename operations issued in
3466 * the send stream or getting into infinite path build
3467 * loops.
3468 */
3469 ret = is_ancestor(sctx->parent_root,
3470 sctx->cur_ino, sctx->cur_inode_gen,
3471 ino, path_before);
Filipe Mananaf9594922014-03-27 20:14:01 +00003472 break;
3473 }
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003474
3475 fs_path_reset(path_before);
3476 fs_path_reset(path_after);
3477
3478 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
Filipe Mananaf9594922014-03-27 20:14:01 +00003479 NULL, path_after);
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003480 if (ret < 0)
3481 goto out;
3482 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3483 NULL, path_before);
Filipe Mananaf9594922014-03-27 20:14:01 +00003484 if (ret < 0 && ret != -ENOENT) {
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003485 goto out;
Filipe Mananaf9594922014-03-27 20:14:01 +00003486 } else if (ret == -ENOENT) {
Filipe Mananabf8e8ca2014-10-02 19:17:32 +01003487 ret = 0;
Filipe Mananaf9594922014-03-27 20:14:01 +00003488 break;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003489 }
3490
3491 len1 = fs_path_len(path_before);
3492 len2 = fs_path_len(path_after);
Filipe Mananaf9594922014-03-27 20:14:01 +00003493 if (ino > sctx->cur_ino &&
3494 (parent_ino_before != parent_ino_after || len1 != len2 ||
3495 memcmp(path_before->start, path_after->start, len1))) {
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003496 ret = 1;
Filipe Mananaf9594922014-03-27 20:14:01 +00003497 break;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003498 }
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003499 ino = parent_ino_after;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003500 }
3501
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003502out:
3503 fs_path_free(path_before);
3504 fs_path_free(path_after);
3505
Filipe Mananaf9594922014-03-27 20:14:01 +00003506 if (ret == 1) {
3507 ret = add_pending_dir_move(sctx,
3508 sctx->cur_ino,
3509 sctx->cur_inode_gen,
3510 ino,
3511 &sctx->new_refs,
Filipe Manana84471e22015-02-28 22:29:22 +00003512 &sctx->deleted_refs,
Filipe Manana8b191a62015-04-09 14:09:14 +01003513 is_orphan);
Filipe Mananaf9594922014-03-27 20:14:01 +00003514 if (!ret)
3515 ret = 1;
3516 }
3517
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003518 return ret;
3519}
3520
Alexander Block31db9f72012-07-25 23:19:24 +02003521/*
3522 * This does all the move/link/unlink/rmdir magic.
3523 */
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003524static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
Alexander Block31db9f72012-07-25 23:19:24 +02003525{
3526 int ret = 0;
3527 struct recorded_ref *cur;
Alexander Block1f4692d2012-07-28 10:42:24 +02003528 struct recorded_ref *cur2;
Josef Bacikba5e8f22013-08-16 16:52:55 -04003529 struct list_head check_dirs;
Alexander Block31db9f72012-07-25 23:19:24 +02003530 struct fs_path *valid_path = NULL;
Chris Masonb24baf62012-07-25 19:21:10 -04003531 u64 ow_inode = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02003532 u64 ow_gen;
3533 int did_overwrite = 0;
3534 int is_orphan = 0;
Filipe Manana29d6d302014-02-16 21:01:39 +00003535 u64 last_dir_ino_rm = 0;
Filipe Manana84471e22015-02-28 22:29:22 +00003536 bool can_rename = true;
Alexander Block31db9f72012-07-25 23:19:24 +02003537
3538verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3539
Alexander Block6d85ed02012-08-01 14:48:59 +02003540 /*
3541 * This should never happen as the root dir always has the same ref
3542 * which is always '..'
3543 */
3544 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
Josef Bacikba5e8f22013-08-16 16:52:55 -04003545 INIT_LIST_HEAD(&check_dirs);
Alexander Block6d85ed02012-08-01 14:48:59 +02003546
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003547 valid_path = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02003548 if (!valid_path) {
3549 ret = -ENOMEM;
3550 goto out;
3551 }
3552
Alexander Block31db9f72012-07-25 23:19:24 +02003553 /*
3554 * First, check if the first ref of the current inode was overwritten
3555 * before. If yes, we know that the current inode was already orphanized
3556 * and thus use the orphan name. If not, we can use get_cur_path to
3557 * get the path of the first ref as it would like while receiving at
3558 * this point in time.
3559 * New inodes are always orphan at the beginning, so force to use the
3560 * orphan name in this case.
3561 * The first ref is stored in valid_path and will be updated if it
3562 * gets moved around.
3563 */
3564 if (!sctx->cur_inode_new) {
3565 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3566 sctx->cur_inode_gen);
3567 if (ret < 0)
3568 goto out;
3569 if (ret)
3570 did_overwrite = 1;
3571 }
3572 if (sctx->cur_inode_new || did_overwrite) {
3573 ret = gen_unique_name(sctx, sctx->cur_ino,
3574 sctx->cur_inode_gen, valid_path);
3575 if (ret < 0)
3576 goto out;
3577 is_orphan = 1;
3578 } else {
3579 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3580 valid_path);
3581 if (ret < 0)
3582 goto out;
3583 }
3584
3585 list_for_each_entry(cur, &sctx->new_refs, list) {
3586 /*
Alexander Block1f4692d2012-07-28 10:42:24 +02003587 * We may have refs where the parent directory does not exist
3588 * yet. This happens if the parent directories inum is higher
3589 * the the current inum. To handle this case, we create the
3590 * parent directory out of order. But we need to check if this
3591 * did already happen before due to other refs in the same dir.
3592 */
3593 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3594 if (ret < 0)
3595 goto out;
3596 if (ret == inode_state_will_create) {
3597 ret = 0;
3598 /*
3599 * First check if any of the current inodes refs did
3600 * already create the dir.
3601 */
3602 list_for_each_entry(cur2, &sctx->new_refs, list) {
3603 if (cur == cur2)
3604 break;
3605 if (cur2->dir == cur->dir) {
3606 ret = 1;
3607 break;
3608 }
3609 }
3610
3611 /*
3612 * If that did not happen, check if a previous inode
3613 * did already create the dir.
3614 */
3615 if (!ret)
3616 ret = did_create_dir(sctx, cur->dir);
3617 if (ret < 0)
3618 goto out;
3619 if (!ret) {
3620 ret = send_create_inode(sctx, cur->dir);
3621 if (ret < 0)
3622 goto out;
3623 }
3624 }
3625
3626 /*
Alexander Block31db9f72012-07-25 23:19:24 +02003627 * Check if this new ref would overwrite the first ref of
3628 * another unprocessed inode. If yes, orphanize the
3629 * overwritten inode. If we find an overwritten ref that is
3630 * not the first ref, simply unlink it.
3631 */
3632 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3633 cur->name, cur->name_len,
3634 &ow_inode, &ow_gen);
3635 if (ret < 0)
3636 goto out;
3637 if (ret) {
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003638 ret = is_first_ref(sctx->parent_root,
3639 ow_inode, cur->dir, cur->name,
3640 cur->name_len);
Alexander Block31db9f72012-07-25 23:19:24 +02003641 if (ret < 0)
3642 goto out;
3643 if (ret) {
Filipe Manana8996a482015-03-12 15:16:20 +00003644 struct name_cache_entry *nce;
3645
Alexander Block31db9f72012-07-25 23:19:24 +02003646 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3647 cur->full_path);
3648 if (ret < 0)
3649 goto out;
Filipe Manana8996a482015-03-12 15:16:20 +00003650 /*
3651 * Make sure we clear our orphanized inode's
3652 * name from the name cache. This is because the
3653 * inode ow_inode might be an ancestor of some
3654 * other inode that will be orphanized as well
3655 * later and has an inode number greater than
3656 * sctx->send_progress. We need to prevent
3657 * future name lookups from using the old name
3658 * and get instead the orphan name.
3659 */
3660 nce = name_cache_search(sctx, ow_inode, ow_gen);
3661 if (nce) {
3662 name_cache_delete(sctx, nce);
3663 kfree(nce);
3664 }
Alexander Block31db9f72012-07-25 23:19:24 +02003665 } else {
3666 ret = send_unlink(sctx, cur->full_path);
3667 if (ret < 0)
3668 goto out;
3669 }
3670 }
3671
Filipe Manana84471e22015-02-28 22:29:22 +00003672 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3673 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3674 if (ret < 0)
3675 goto out;
3676 if (ret == 1) {
3677 can_rename = false;
3678 *pending_move = 1;
3679 }
3680 }
3681
Filipe Manana8b191a62015-04-09 14:09:14 +01003682 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3683 can_rename) {
3684 ret = wait_for_parent_move(sctx, cur, is_orphan);
3685 if (ret < 0)
3686 goto out;
3687 if (ret == 1) {
3688 can_rename = false;
3689 *pending_move = 1;
3690 }
3691 }
3692
Alexander Block31db9f72012-07-25 23:19:24 +02003693 /*
3694 * link/move the ref to the new place. If we have an orphan
3695 * inode, move it and update valid_path. If not, link or move
3696 * it depending on the inode mode.
3697 */
Filipe Manana84471e22015-02-28 22:29:22 +00003698 if (is_orphan && can_rename) {
Alexander Block31db9f72012-07-25 23:19:24 +02003699 ret = send_rename(sctx, valid_path, cur->full_path);
3700 if (ret < 0)
3701 goto out;
3702 is_orphan = 0;
3703 ret = fs_path_copy(valid_path, cur->full_path);
3704 if (ret < 0)
3705 goto out;
Filipe Manana84471e22015-02-28 22:29:22 +00003706 } else if (can_rename) {
Alexander Block31db9f72012-07-25 23:19:24 +02003707 if (S_ISDIR(sctx->cur_inode_mode)) {
3708 /*
3709 * Dirs can't be linked, so move it. For moved
3710 * dirs, we always have one new and one deleted
3711 * ref. The deleted ref is ignored later.
3712 */
Filipe Manana8b191a62015-04-09 14:09:14 +01003713 ret = send_rename(sctx, valid_path,
3714 cur->full_path);
3715 if (!ret)
3716 ret = fs_path_copy(valid_path,
3717 cur->full_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003718 if (ret < 0)
3719 goto out;
3720 } else {
3721 ret = send_link(sctx, cur->full_path,
3722 valid_path);
3723 if (ret < 0)
3724 goto out;
3725 }
3726 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04003727 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003728 if (ret < 0)
3729 goto out;
3730 }
3731
3732 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3733 /*
3734 * Check if we can already rmdir the directory. If not,
3735 * orphanize it. For every dir item inside that gets deleted
3736 * later, we do this check again and rmdir it then if possible.
3737 * See the use of check_dirs for more details.
3738 */
Filipe Manana9dc44212014-02-19 14:31:44 +00003739 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3740 sctx->cur_ino);
Alexander Block31db9f72012-07-25 23:19:24 +02003741 if (ret < 0)
3742 goto out;
3743 if (ret) {
3744 ret = send_rmdir(sctx, valid_path);
3745 if (ret < 0)
3746 goto out;
3747 } else if (!is_orphan) {
3748 ret = orphanize_inode(sctx, sctx->cur_ino,
3749 sctx->cur_inode_gen, valid_path);
3750 if (ret < 0)
3751 goto out;
3752 is_orphan = 1;
3753 }
3754
3755 list_for_each_entry(cur, &sctx->deleted_refs, list) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04003756 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003757 if (ret < 0)
3758 goto out;
3759 }
Alexander Blockccf16262012-07-28 11:46:29 +02003760 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3761 !list_empty(&sctx->deleted_refs)) {
3762 /*
3763 * We have a moved dir. Add the old parent to check_dirs
3764 */
3765 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3766 list);
Josef Bacikba5e8f22013-08-16 16:52:55 -04003767 ret = dup_ref(cur, &check_dirs);
Alexander Blockccf16262012-07-28 11:46:29 +02003768 if (ret < 0)
3769 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003770 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3771 /*
3772 * We have a non dir inode. Go through all deleted refs and
3773 * unlink them if they were not already overwritten by other
3774 * inodes.
3775 */
3776 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3777 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3778 sctx->cur_ino, sctx->cur_inode_gen,
3779 cur->name, cur->name_len);
3780 if (ret < 0)
3781 goto out;
3782 if (!ret) {
Alexander Block1f4692d2012-07-28 10:42:24 +02003783 ret = send_unlink(sctx, cur->full_path);
3784 if (ret < 0)
3785 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003786 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04003787 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003788 if (ret < 0)
3789 goto out;
3790 }
Alexander Block31db9f72012-07-25 23:19:24 +02003791 /*
3792 * If the inode is still orphan, unlink the orphan. This may
3793 * happen when a previous inode did overwrite the first ref
3794 * of this inode and no new refs were added for the current
Alexander Block766702e2012-07-28 14:11:31 +02003795 * inode. Unlinking does not mean that the inode is deleted in
3796 * all cases. There may still be links to this inode in other
3797 * places.
Alexander Block31db9f72012-07-25 23:19:24 +02003798 */
Alexander Block1f4692d2012-07-28 10:42:24 +02003799 if (is_orphan) {
Alexander Block31db9f72012-07-25 23:19:24 +02003800 ret = send_unlink(sctx, valid_path);
3801 if (ret < 0)
3802 goto out;
3803 }
3804 }
3805
3806 /*
3807 * We did collect all parent dirs where cur_inode was once located. We
3808 * now go through all these dirs and check if they are pending for
3809 * deletion and if it's finally possible to perform the rmdir now.
3810 * We also update the inode stats of the parent dirs here.
3811 */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003812 list_for_each_entry(cur, &check_dirs, list) {
Alexander Block766702e2012-07-28 14:11:31 +02003813 /*
3814 * In case we had refs into dirs that were not processed yet,
3815 * we don't need to do the utime and rmdir logic for these dirs.
3816 * The dir will be processed later.
3817 */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003818 if (cur->dir > sctx->cur_ino)
Alexander Block31db9f72012-07-25 23:19:24 +02003819 continue;
3820
Josef Bacikba5e8f22013-08-16 16:52:55 -04003821 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
Alexander Block31db9f72012-07-25 23:19:24 +02003822 if (ret < 0)
3823 goto out;
3824
3825 if (ret == inode_state_did_create ||
3826 ret == inode_state_no_change) {
3827 /* TODO delayed utimes */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003828 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
Alexander Block31db9f72012-07-25 23:19:24 +02003829 if (ret < 0)
3830 goto out;
Filipe Manana29d6d302014-02-16 21:01:39 +00003831 } else if (ret == inode_state_did_delete &&
3832 cur->dir != last_dir_ino_rm) {
Filipe Manana9dc44212014-02-19 14:31:44 +00003833 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3834 sctx->cur_ino);
Alexander Block31db9f72012-07-25 23:19:24 +02003835 if (ret < 0)
3836 goto out;
3837 if (ret) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04003838 ret = get_cur_path(sctx, cur->dir,
3839 cur->dir_gen, valid_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003840 if (ret < 0)
3841 goto out;
3842 ret = send_rmdir(sctx, valid_path);
3843 if (ret < 0)
3844 goto out;
Filipe Manana29d6d302014-02-16 21:01:39 +00003845 last_dir_ino_rm = cur->dir;
Alexander Block31db9f72012-07-25 23:19:24 +02003846 }
3847 }
3848 }
3849
Alexander Block31db9f72012-07-25 23:19:24 +02003850 ret = 0;
3851
3852out:
Josef Bacikba5e8f22013-08-16 16:52:55 -04003853 __free_recorded_refs(&check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003854 free_recorded_refs(sctx);
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003855 fs_path_free(valid_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003856 return ret;
3857}
3858
Liu Boa4d96d62014-03-03 21:31:03 +08003859static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3860 struct fs_path *name, void *ctx, struct list_head *refs)
Alexander Block31db9f72012-07-25 23:19:24 +02003861{
3862 int ret = 0;
3863 struct send_ctx *sctx = ctx;
3864 struct fs_path *p;
3865 u64 gen;
3866
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003867 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02003868 if (!p)
3869 return -ENOMEM;
3870
Liu Boa4d96d62014-03-03 21:31:03 +08003871 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02003872 NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02003873 if (ret < 0)
3874 goto out;
3875
Alexander Block31db9f72012-07-25 23:19:24 +02003876 ret = get_cur_path(sctx, dir, gen, p);
3877 if (ret < 0)
3878 goto out;
3879 ret = fs_path_add_path(p, name);
3880 if (ret < 0)
3881 goto out;
3882
Liu Boa4d96d62014-03-03 21:31:03 +08003883 ret = __record_ref(refs, dir, gen, p);
Alexander Block31db9f72012-07-25 23:19:24 +02003884
3885out:
3886 if (ret)
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003887 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02003888 return ret;
3889}
3890
Liu Boa4d96d62014-03-03 21:31:03 +08003891static int __record_new_ref(int num, u64 dir, int index,
3892 struct fs_path *name,
3893 void *ctx)
3894{
3895 struct send_ctx *sctx = ctx;
3896 return record_ref(sctx->send_root, num, dir, index, name,
3897 ctx, &sctx->new_refs);
3898}
3899
3900
Alexander Block31db9f72012-07-25 23:19:24 +02003901static int __record_deleted_ref(int num, u64 dir, int index,
3902 struct fs_path *name,
3903 void *ctx)
3904{
Alexander Block31db9f72012-07-25 23:19:24 +02003905 struct send_ctx *sctx = ctx;
Liu Boa4d96d62014-03-03 21:31:03 +08003906 return record_ref(sctx->parent_root, num, dir, index, name,
3907 ctx, &sctx->deleted_refs);
Alexander Block31db9f72012-07-25 23:19:24 +02003908}
3909
3910static int record_new_ref(struct send_ctx *sctx)
3911{
3912 int ret;
3913
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003914 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3915 sctx->cmp_key, 0, __record_new_ref, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02003916 if (ret < 0)
3917 goto out;
3918 ret = 0;
3919
3920out:
3921 return ret;
3922}
3923
3924static int record_deleted_ref(struct send_ctx *sctx)
3925{
3926 int ret;
3927
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003928 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3929 sctx->cmp_key, 0, __record_deleted_ref, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02003930 if (ret < 0)
3931 goto out;
3932 ret = 0;
3933
3934out:
3935 return ret;
3936}
3937
3938struct find_ref_ctx {
3939 u64 dir;
Josef Bacikba5e8f22013-08-16 16:52:55 -04003940 u64 dir_gen;
3941 struct btrfs_root *root;
Alexander Block31db9f72012-07-25 23:19:24 +02003942 struct fs_path *name;
3943 int found_idx;
3944};
3945
3946static int __find_iref(int num, u64 dir, int index,
3947 struct fs_path *name,
3948 void *ctx_)
3949{
3950 struct find_ref_ctx *ctx = ctx_;
Josef Bacikba5e8f22013-08-16 16:52:55 -04003951 u64 dir_gen;
3952 int ret;
Alexander Block31db9f72012-07-25 23:19:24 +02003953
3954 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3955 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04003956 /*
3957 * To avoid doing extra lookups we'll only do this if everything
3958 * else matches.
3959 */
3960 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3961 NULL, NULL, NULL);
3962 if (ret)
3963 return ret;
3964 if (dir_gen != ctx->dir_gen)
3965 return 0;
Alexander Block31db9f72012-07-25 23:19:24 +02003966 ctx->found_idx = num;
3967 return 1;
3968 }
3969 return 0;
3970}
3971
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003972static int find_iref(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02003973 struct btrfs_path *path,
3974 struct btrfs_key *key,
Josef Bacikba5e8f22013-08-16 16:52:55 -04003975 u64 dir, u64 dir_gen, struct fs_path *name)
Alexander Block31db9f72012-07-25 23:19:24 +02003976{
3977 int ret;
3978 struct find_ref_ctx ctx;
3979
3980 ctx.dir = dir;
3981 ctx.name = name;
Josef Bacikba5e8f22013-08-16 16:52:55 -04003982 ctx.dir_gen = dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02003983 ctx.found_idx = -1;
Josef Bacikba5e8f22013-08-16 16:52:55 -04003984 ctx.root = root;
Alexander Block31db9f72012-07-25 23:19:24 +02003985
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003986 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
Alexander Block31db9f72012-07-25 23:19:24 +02003987 if (ret < 0)
3988 return ret;
3989
3990 if (ctx.found_idx == -1)
3991 return -ENOENT;
3992
3993 return ctx.found_idx;
3994}
3995
3996static int __record_changed_new_ref(int num, u64 dir, int index,
3997 struct fs_path *name,
3998 void *ctx)
3999{
Josef Bacikba5e8f22013-08-16 16:52:55 -04004000 u64 dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004001 int ret;
4002 struct send_ctx *sctx = ctx;
4003
Josef Bacikba5e8f22013-08-16 16:52:55 -04004004 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4005 NULL, NULL, NULL);
4006 if (ret)
4007 return ret;
4008
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004009 ret = find_iref(sctx->parent_root, sctx->right_path,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004010 sctx->cmp_key, dir, dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02004011 if (ret == -ENOENT)
4012 ret = __record_new_ref(num, dir, index, name, sctx);
4013 else if (ret > 0)
4014 ret = 0;
4015
4016 return ret;
4017}
4018
4019static int __record_changed_deleted_ref(int num, u64 dir, int index,
4020 struct fs_path *name,
4021 void *ctx)
4022{
Josef Bacikba5e8f22013-08-16 16:52:55 -04004023 u64 dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004024 int ret;
4025 struct send_ctx *sctx = ctx;
4026
Josef Bacikba5e8f22013-08-16 16:52:55 -04004027 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4028 NULL, NULL, NULL);
4029 if (ret)
4030 return ret;
4031
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004032 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004033 dir, dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02004034 if (ret == -ENOENT)
4035 ret = __record_deleted_ref(num, dir, index, name, sctx);
4036 else if (ret > 0)
4037 ret = 0;
4038
4039 return ret;
4040}
4041
4042static int record_changed_ref(struct send_ctx *sctx)
4043{
4044 int ret = 0;
4045
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004046 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004047 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4048 if (ret < 0)
4049 goto out;
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004050 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004051 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4052 if (ret < 0)
4053 goto out;
4054 ret = 0;
4055
4056out:
4057 return ret;
4058}
4059
4060/*
4061 * Record and process all refs at once. Needed when an inode changes the
4062 * generation number, which means that it was deleted and recreated.
4063 */
4064static int process_all_refs(struct send_ctx *sctx,
4065 enum btrfs_compare_tree_result cmd)
4066{
4067 int ret;
4068 struct btrfs_root *root;
4069 struct btrfs_path *path;
4070 struct btrfs_key key;
4071 struct btrfs_key found_key;
4072 struct extent_buffer *eb;
4073 int slot;
4074 iterate_inode_ref_t cb;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00004075 int pending_move = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004076
4077 path = alloc_path_for_send();
4078 if (!path)
4079 return -ENOMEM;
4080
4081 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4082 root = sctx->send_root;
4083 cb = __record_new_ref;
4084 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4085 root = sctx->parent_root;
4086 cb = __record_deleted_ref;
4087 } else {
David Sterba4d1a63b2014-02-03 19:24:19 +01004088 btrfs_err(sctx->send_root->fs_info,
4089 "Wrong command %d in process_all_refs", cmd);
4090 ret = -EINVAL;
4091 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004092 }
4093
4094 key.objectid = sctx->cmp_key->objectid;
4095 key.type = BTRFS_INODE_REF_KEY;
4096 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004097 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4098 if (ret < 0)
4099 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004100
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004101 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02004102 eb = path->nodes[0];
4103 slot = path->slots[0];
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004104 if (slot >= btrfs_header_nritems(eb)) {
4105 ret = btrfs_next_leaf(root, path);
4106 if (ret < 0)
4107 goto out;
4108 else if (ret > 0)
4109 break;
4110 continue;
4111 }
4112
Alexander Block31db9f72012-07-25 23:19:24 +02004113 btrfs_item_key_to_cpu(eb, &found_key, slot);
4114
4115 if (found_key.objectid != key.objectid ||
Jan Schmidt96b5bd72012-10-15 08:30:45 +00004116 (found_key.type != BTRFS_INODE_REF_KEY &&
4117 found_key.type != BTRFS_INODE_EXTREF_KEY))
Alexander Block31db9f72012-07-25 23:19:24 +02004118 break;
Alexander Block31db9f72012-07-25 23:19:24 +02004119
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004120 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004121 if (ret < 0)
4122 goto out;
4123
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004124 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02004125 }
Alexander Blocke938c8a2012-07-28 16:33:49 +02004126 btrfs_release_path(path);
Alexander Block31db9f72012-07-25 23:19:24 +02004127
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00004128 ret = process_recorded_refs(sctx, &pending_move);
4129 /* Only applicable to an incremental send. */
4130 ASSERT(pending_move == 0);
Alexander Block31db9f72012-07-25 23:19:24 +02004131
4132out:
4133 btrfs_free_path(path);
4134 return ret;
4135}
4136
4137static int send_set_xattr(struct send_ctx *sctx,
4138 struct fs_path *path,
4139 const char *name, int name_len,
4140 const char *data, int data_len)
4141{
4142 int ret = 0;
4143
4144 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4145 if (ret < 0)
4146 goto out;
4147
4148 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4149 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4150 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4151
4152 ret = send_cmd(sctx);
4153
4154tlv_put_failure:
4155out:
4156 return ret;
4157}
4158
4159static int send_remove_xattr(struct send_ctx *sctx,
4160 struct fs_path *path,
4161 const char *name, int name_len)
4162{
4163 int ret = 0;
4164
4165 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4166 if (ret < 0)
4167 goto out;
4168
4169 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4170 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4171
4172 ret = send_cmd(sctx);
4173
4174tlv_put_failure:
4175out:
4176 return ret;
4177}
4178
4179static int __process_new_xattr(int num, struct btrfs_key *di_key,
4180 const char *name, int name_len,
4181 const char *data, int data_len,
4182 u8 type, void *ctx)
4183{
4184 int ret;
4185 struct send_ctx *sctx = ctx;
4186 struct fs_path *p;
4187 posix_acl_xattr_header dummy_acl;
4188
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004189 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004190 if (!p)
4191 return -ENOMEM;
4192
4193 /*
4194 * This hack is needed because empty acl's are stored as zero byte
4195 * data in xattrs. Problem with that is, that receiving these zero byte
4196 * acl's will fail later. To fix this, we send a dummy acl list that
4197 * only contains the version number and no entries.
4198 */
4199 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4200 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4201 if (data_len == 0) {
4202 dummy_acl.a_version =
4203 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4204 data = (char *)&dummy_acl;
4205 data_len = sizeof(dummy_acl);
4206 }
4207 }
4208
4209 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4210 if (ret < 0)
4211 goto out;
4212
4213 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4214
4215out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004216 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004217 return ret;
4218}
4219
4220static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4221 const char *name, int name_len,
4222 const char *data, int data_len,
4223 u8 type, void *ctx)
4224{
4225 int ret;
4226 struct send_ctx *sctx = ctx;
4227 struct fs_path *p;
4228
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004229 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004230 if (!p)
4231 return -ENOMEM;
4232
4233 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4234 if (ret < 0)
4235 goto out;
4236
4237 ret = send_remove_xattr(sctx, p, name, name_len);
4238
4239out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004240 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004241 return ret;
4242}
4243
4244static int process_new_xattr(struct send_ctx *sctx)
4245{
4246 int ret = 0;
4247
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004248 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4249 sctx->cmp_key, __process_new_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004250
4251 return ret;
4252}
4253
4254static int process_deleted_xattr(struct send_ctx *sctx)
4255{
4256 int ret;
4257
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004258 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4259 sctx->cmp_key, __process_deleted_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004260
4261 return ret;
4262}
4263
4264struct find_xattr_ctx {
4265 const char *name;
4266 int name_len;
4267 int found_idx;
4268 char *found_data;
4269 int found_data_len;
4270};
4271
4272static int __find_xattr(int num, struct btrfs_key *di_key,
4273 const char *name, int name_len,
4274 const char *data, int data_len,
4275 u8 type, void *vctx)
4276{
4277 struct find_xattr_ctx *ctx = vctx;
4278
4279 if (name_len == ctx->name_len &&
4280 strncmp(name, ctx->name, name_len) == 0) {
4281 ctx->found_idx = num;
4282 ctx->found_data_len = data_len;
Thomas Meyera5959bc2013-06-01 09:37:50 +00004283 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
Alexander Block31db9f72012-07-25 23:19:24 +02004284 if (!ctx->found_data)
4285 return -ENOMEM;
Alexander Block31db9f72012-07-25 23:19:24 +02004286 return 1;
4287 }
4288 return 0;
4289}
4290
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004291static int find_xattr(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02004292 struct btrfs_path *path,
4293 struct btrfs_key *key,
4294 const char *name, int name_len,
4295 char **data, int *data_len)
4296{
4297 int ret;
4298 struct find_xattr_ctx ctx;
4299
4300 ctx.name = name;
4301 ctx.name_len = name_len;
4302 ctx.found_idx = -1;
4303 ctx.found_data = NULL;
4304 ctx.found_data_len = 0;
4305
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004306 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004307 if (ret < 0)
4308 return ret;
4309
4310 if (ctx.found_idx == -1)
4311 return -ENOENT;
4312 if (data) {
4313 *data = ctx.found_data;
4314 *data_len = ctx.found_data_len;
4315 } else {
4316 kfree(ctx.found_data);
4317 }
4318 return ctx.found_idx;
4319}
4320
4321
4322static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4323 const char *name, int name_len,
4324 const char *data, int data_len,
4325 u8 type, void *ctx)
4326{
4327 int ret;
4328 struct send_ctx *sctx = ctx;
4329 char *found_data = NULL;
4330 int found_data_len = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004331
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004332 ret = find_xattr(sctx->parent_root, sctx->right_path,
4333 sctx->cmp_key, name, name_len, &found_data,
4334 &found_data_len);
Alexander Block31db9f72012-07-25 23:19:24 +02004335 if (ret == -ENOENT) {
4336 ret = __process_new_xattr(num, di_key, name, name_len, data,
4337 data_len, type, ctx);
4338 } else if (ret >= 0) {
4339 if (data_len != found_data_len ||
4340 memcmp(data, found_data, data_len)) {
4341 ret = __process_new_xattr(num, di_key, name, name_len,
4342 data, data_len, type, ctx);
4343 } else {
4344 ret = 0;
4345 }
4346 }
4347
4348 kfree(found_data);
Alexander Block31db9f72012-07-25 23:19:24 +02004349 return ret;
4350}
4351
4352static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4353 const char *name, int name_len,
4354 const char *data, int data_len,
4355 u8 type, void *ctx)
4356{
4357 int ret;
4358 struct send_ctx *sctx = ctx;
4359
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004360 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4361 name, name_len, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02004362 if (ret == -ENOENT)
4363 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4364 data_len, type, ctx);
4365 else if (ret >= 0)
4366 ret = 0;
4367
4368 return ret;
4369}
4370
4371static int process_changed_xattr(struct send_ctx *sctx)
4372{
4373 int ret = 0;
4374
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004375 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004376 sctx->cmp_key, __process_changed_new_xattr, sctx);
4377 if (ret < 0)
4378 goto out;
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004379 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004380 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4381
4382out:
4383 return ret;
4384}
4385
4386static int process_all_new_xattrs(struct send_ctx *sctx)
4387{
4388 int ret;
4389 struct btrfs_root *root;
4390 struct btrfs_path *path;
4391 struct btrfs_key key;
4392 struct btrfs_key found_key;
4393 struct extent_buffer *eb;
4394 int slot;
4395
4396 path = alloc_path_for_send();
4397 if (!path)
4398 return -ENOMEM;
4399
4400 root = sctx->send_root;
4401
4402 key.objectid = sctx->cmp_key->objectid;
4403 key.type = BTRFS_XATTR_ITEM_KEY;
4404 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004405 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4406 if (ret < 0)
4407 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004408
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004409 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02004410 eb = path->nodes[0];
4411 slot = path->slots[0];
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004412 if (slot >= btrfs_header_nritems(eb)) {
4413 ret = btrfs_next_leaf(root, path);
4414 if (ret < 0) {
4415 goto out;
4416 } else if (ret > 0) {
4417 ret = 0;
4418 break;
4419 }
4420 continue;
4421 }
Alexander Block31db9f72012-07-25 23:19:24 +02004422
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004423 btrfs_item_key_to_cpu(eb, &found_key, slot);
Alexander Block31db9f72012-07-25 23:19:24 +02004424 if (found_key.objectid != key.objectid ||
4425 found_key.type != key.type) {
4426 ret = 0;
4427 goto out;
4428 }
4429
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004430 ret = iterate_dir_item(root, path, &found_key,
4431 __process_new_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004432 if (ret < 0)
4433 goto out;
4434
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004435 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02004436 }
4437
4438out:
4439 btrfs_free_path(path);
4440 return ret;
4441}
4442
Josef Baciked259092013-10-25 11:36:01 -04004443static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4444{
4445 struct btrfs_root *root = sctx->send_root;
4446 struct btrfs_fs_info *fs_info = root->fs_info;
4447 struct inode *inode;
4448 struct page *page;
4449 char *addr;
4450 struct btrfs_key key;
4451 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
4452 pgoff_t last_index;
4453 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
4454 ssize_t ret = 0;
4455
4456 key.objectid = sctx->cur_ino;
4457 key.type = BTRFS_INODE_ITEM_KEY;
4458 key.offset = 0;
4459
4460 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4461 if (IS_ERR(inode))
4462 return PTR_ERR(inode);
4463
4464 if (offset + len > i_size_read(inode)) {
4465 if (offset > i_size_read(inode))
4466 len = 0;
4467 else
4468 len = offset - i_size_read(inode);
4469 }
4470 if (len == 0)
4471 goto out;
4472
4473 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
Liu Bo2131bcd32014-03-05 10:07:35 +08004474
4475 /* initial readahead */
4476 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4477 file_ra_state_init(&sctx->ra, inode->i_mapping);
4478 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4479 last_index - index + 1);
4480
Josef Baciked259092013-10-25 11:36:01 -04004481 while (index <= last_index) {
4482 unsigned cur_len = min_t(unsigned, len,
4483 PAGE_CACHE_SIZE - pg_offset);
4484 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4485 if (!page) {
4486 ret = -ENOMEM;
4487 break;
4488 }
4489
4490 if (!PageUptodate(page)) {
4491 btrfs_readpage(NULL, page);
4492 lock_page(page);
4493 if (!PageUptodate(page)) {
4494 unlock_page(page);
4495 page_cache_release(page);
4496 ret = -EIO;
4497 break;
4498 }
4499 }
4500
4501 addr = kmap(page);
4502 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4503 kunmap(page);
4504 unlock_page(page);
4505 page_cache_release(page);
4506 index++;
4507 pg_offset = 0;
4508 len -= cur_len;
4509 ret += cur_len;
4510 }
4511out:
4512 iput(inode);
4513 return ret;
4514}
4515
Alexander Block31db9f72012-07-25 23:19:24 +02004516/*
4517 * Read some bytes from the current inode/file and send a write command to
4518 * user space.
4519 */
4520static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4521{
4522 int ret = 0;
4523 struct fs_path *p;
Josef Baciked259092013-10-25 11:36:01 -04004524 ssize_t num_read = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004525
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004526 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004527 if (!p)
4528 return -ENOMEM;
4529
Alexander Block31db9f72012-07-25 23:19:24 +02004530verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4531
Josef Baciked259092013-10-25 11:36:01 -04004532 num_read = fill_read_buf(sctx, offset, len);
4533 if (num_read <= 0) {
4534 if (num_read < 0)
4535 ret = num_read;
Alexander Block31db9f72012-07-25 23:19:24 +02004536 goto out;
Josef Baciked259092013-10-25 11:36:01 -04004537 }
Alexander Block31db9f72012-07-25 23:19:24 +02004538
4539 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4540 if (ret < 0)
4541 goto out;
4542
4543 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4544 if (ret < 0)
4545 goto out;
4546
4547 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4548 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
Alexander Blocke938c8a2012-07-28 16:33:49 +02004549 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
Alexander Block31db9f72012-07-25 23:19:24 +02004550
4551 ret = send_cmd(sctx);
4552
4553tlv_put_failure:
4554out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004555 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004556 if (ret < 0)
4557 return ret;
Alexander Blocke938c8a2012-07-28 16:33:49 +02004558 return num_read;
Alexander Block31db9f72012-07-25 23:19:24 +02004559}
4560
4561/*
4562 * Send a clone command to user space.
4563 */
4564static int send_clone(struct send_ctx *sctx,
4565 u64 offset, u32 len,
4566 struct clone_root *clone_root)
4567{
4568 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004569 struct fs_path *p;
4570 u64 gen;
4571
4572verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4573 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4574 clone_root->root->objectid, clone_root->ino,
4575 clone_root->offset);
4576
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004577 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004578 if (!p)
4579 return -ENOMEM;
4580
4581 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4582 if (ret < 0)
4583 goto out;
4584
4585 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4586 if (ret < 0)
4587 goto out;
4588
4589 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4590 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4591 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4592
Alexander Blocke938c8a2012-07-28 16:33:49 +02004593 if (clone_root->root == sctx->send_root) {
Alexander Block31db9f72012-07-25 23:19:24 +02004594 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02004595 &gen, NULL, NULL, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02004596 if (ret < 0)
4597 goto out;
4598 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4599 } else {
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004600 ret = get_inode_path(clone_root->root, clone_root->ino, p);
Alexander Block31db9f72012-07-25 23:19:24 +02004601 }
4602 if (ret < 0)
4603 goto out;
4604
Josef Bacik37b8d272015-06-04 17:17:25 -04004605 /*
4606 * If the parent we're using has a received_uuid set then use that as
4607 * our clone source as that is what we will look for when doing a
4608 * receive.
4609 *
4610 * This covers the case that we create a snapshot off of a received
4611 * subvolume and then use that as the parent and try to receive on a
4612 * different host.
4613 */
4614 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4615 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4616 clone_root->root->root_item.received_uuid);
4617 else
4618 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4619 clone_root->root->root_item.uuid);
Alexander Block31db9f72012-07-25 23:19:24 +02004620 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
Filipe David Borba Manana5a0f4e22013-12-03 15:55:48 +00004621 le64_to_cpu(clone_root->root->root_item.ctransid));
Alexander Block31db9f72012-07-25 23:19:24 +02004622 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4623 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4624 clone_root->offset);
4625
4626 ret = send_cmd(sctx);
4627
4628tlv_put_failure:
4629out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004630 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004631 return ret;
4632}
4633
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004634/*
4635 * Send an update extent command to user space.
4636 */
4637static int send_update_extent(struct send_ctx *sctx,
4638 u64 offset, u32 len)
4639{
4640 int ret = 0;
4641 struct fs_path *p;
4642
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004643 p = fs_path_alloc();
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004644 if (!p)
4645 return -ENOMEM;
4646
4647 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4648 if (ret < 0)
4649 goto out;
4650
4651 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4652 if (ret < 0)
4653 goto out;
4654
4655 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4656 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4657 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4658
4659 ret = send_cmd(sctx);
4660
4661tlv_put_failure:
4662out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004663 fs_path_free(p);
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004664 return ret;
4665}
4666
Josef Bacik16e75492013-10-22 12:18:51 -04004667static int send_hole(struct send_ctx *sctx, u64 end)
4668{
4669 struct fs_path *p = NULL;
4670 u64 offset = sctx->cur_inode_last_extent;
4671 u64 len;
4672 int ret = 0;
4673
4674 p = fs_path_alloc();
4675 if (!p)
4676 return -ENOMEM;
Filipe Mananac715e152014-03-31 14:52:14 +01004677 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4678 if (ret < 0)
4679 goto tlv_put_failure;
Josef Bacik16e75492013-10-22 12:18:51 -04004680 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4681 while (offset < end) {
4682 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4683
4684 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4685 if (ret < 0)
4686 break;
Josef Bacik16e75492013-10-22 12:18:51 -04004687 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4688 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4689 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4690 ret = send_cmd(sctx);
4691 if (ret < 0)
4692 break;
4693 offset += len;
4694 }
4695tlv_put_failure:
4696 fs_path_free(p);
4697 return ret;
4698}
4699
Filipe Mananad906d492015-10-02 10:47:34 +01004700static int send_extent_data(struct send_ctx *sctx,
4701 const u64 offset,
4702 const u64 len)
4703{
4704 u64 sent = 0;
4705
4706 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4707 return send_update_extent(sctx, offset, len);
4708
4709 while (sent < len) {
4710 u64 size = len - sent;
4711 int ret;
4712
4713 if (size > BTRFS_SEND_READ_SIZE)
4714 size = BTRFS_SEND_READ_SIZE;
4715 ret = send_write(sctx, offset + sent, size);
4716 if (ret < 0)
4717 return ret;
4718 if (!ret)
4719 break;
4720 sent += ret;
4721 }
4722 return 0;
4723}
4724
4725static int clone_range(struct send_ctx *sctx,
4726 struct clone_root *clone_root,
4727 const u64 disk_byte,
4728 u64 data_offset,
4729 u64 offset,
4730 u64 len)
4731{
4732 struct btrfs_path *path;
4733 struct btrfs_key key;
4734 int ret;
4735
4736 path = alloc_path_for_send();
4737 if (!path)
4738 return -ENOMEM;
4739
4740 /*
4741 * We can't send a clone operation for the entire range if we find
4742 * extent items in the respective range in the source file that
4743 * refer to different extents or if we find holes.
4744 * So check for that and do a mix of clone and regular write/copy
4745 * operations if needed.
4746 *
4747 * Example:
4748 *
4749 * mkfs.btrfs -f /dev/sda
4750 * mount /dev/sda /mnt
4751 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
4752 * cp --reflink=always /mnt/foo /mnt/bar
4753 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
4754 * btrfs subvolume snapshot -r /mnt /mnt/snap
4755 *
4756 * If when we send the snapshot and we are processing file bar (which
4757 * has a higher inode number than foo) we blindly send a clone operation
4758 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
4759 * a file bar that matches the content of file foo - iow, doesn't match
4760 * the content from bar in the original filesystem.
4761 */
4762 key.objectid = clone_root->ino;
4763 key.type = BTRFS_EXTENT_DATA_KEY;
4764 key.offset = clone_root->offset;
4765 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
4766 if (ret < 0)
4767 goto out;
4768 if (ret > 0 && path->slots[0] > 0) {
4769 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4770 if (key.objectid == clone_root->ino &&
4771 key.type == BTRFS_EXTENT_DATA_KEY)
4772 path->slots[0]--;
4773 }
4774
4775 while (true) {
4776 struct extent_buffer *leaf = path->nodes[0];
4777 int slot = path->slots[0];
4778 struct btrfs_file_extent_item *ei;
4779 u8 type;
4780 u64 ext_len;
4781 u64 clone_len;
4782
4783 if (slot >= btrfs_header_nritems(leaf)) {
4784 ret = btrfs_next_leaf(clone_root->root, path);
4785 if (ret < 0)
4786 goto out;
4787 else if (ret > 0)
4788 break;
4789 continue;
4790 }
4791
4792 btrfs_item_key_to_cpu(leaf, &key, slot);
4793
4794 /*
4795 * We might have an implicit trailing hole (NO_HOLES feature
4796 * enabled). We deal with it after leaving this loop.
4797 */
4798 if (key.objectid != clone_root->ino ||
4799 key.type != BTRFS_EXTENT_DATA_KEY)
4800 break;
4801
4802 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4803 type = btrfs_file_extent_type(leaf, ei);
4804 if (type == BTRFS_FILE_EXTENT_INLINE) {
4805 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
4806 ext_len = PAGE_CACHE_ALIGN(ext_len);
4807 } else {
4808 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
4809 }
4810
4811 if (key.offset + ext_len <= clone_root->offset)
4812 goto next;
4813
4814 if (key.offset > clone_root->offset) {
4815 /* Implicit hole, NO_HOLES feature enabled. */
4816 u64 hole_len = key.offset - clone_root->offset;
4817
4818 if (hole_len > len)
4819 hole_len = len;
4820 ret = send_extent_data(sctx, offset, hole_len);
4821 if (ret < 0)
4822 goto out;
4823
4824 len -= hole_len;
4825 if (len == 0)
4826 break;
4827 offset += hole_len;
4828 clone_root->offset += hole_len;
4829 data_offset += hole_len;
4830 }
4831
4832 if (key.offset >= clone_root->offset + len)
4833 break;
4834
4835 clone_len = min_t(u64, ext_len, len);
4836
4837 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
4838 btrfs_file_extent_offset(leaf, ei) == data_offset)
4839 ret = send_clone(sctx, offset, clone_len, clone_root);
4840 else
4841 ret = send_extent_data(sctx, offset, clone_len);
4842
4843 if (ret < 0)
4844 goto out;
4845
4846 len -= clone_len;
4847 if (len == 0)
4848 break;
4849 offset += clone_len;
4850 clone_root->offset += clone_len;
4851 data_offset += clone_len;
4852next:
4853 path->slots[0]++;
4854 }
4855
4856 if (len > 0)
4857 ret = send_extent_data(sctx, offset, len);
4858 else
4859 ret = 0;
4860out:
4861 btrfs_free_path(path);
4862 return ret;
4863}
4864
Alexander Block31db9f72012-07-25 23:19:24 +02004865static int send_write_or_clone(struct send_ctx *sctx,
4866 struct btrfs_path *path,
4867 struct btrfs_key *key,
4868 struct clone_root *clone_root)
4869{
4870 int ret = 0;
4871 struct btrfs_file_extent_item *ei;
4872 u64 offset = key->offset;
Alexander Block31db9f72012-07-25 23:19:24 +02004873 u64 len;
Alexander Block31db9f72012-07-25 23:19:24 +02004874 u8 type;
Filipe David Borba Manana28e5dd82014-01-12 02:26:28 +00004875 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
Alexander Block31db9f72012-07-25 23:19:24 +02004876
4877 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4878 struct btrfs_file_extent_item);
4879 type = btrfs_file_extent_type(path->nodes[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004880 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08004881 len = btrfs_file_extent_inline_len(path->nodes[0],
4882 path->slots[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004883 /*
4884 * it is possible the inline item won't cover the whole page,
4885 * but there may be items after this page. Make
4886 * sure to send the whole thing
4887 */
4888 len = PAGE_CACHE_ALIGN(len);
4889 } else {
Alexander Block31db9f72012-07-25 23:19:24 +02004890 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004891 }
Alexander Block31db9f72012-07-25 23:19:24 +02004892
4893 if (offset + len > sctx->cur_inode_size)
4894 len = sctx->cur_inode_size - offset;
4895 if (len == 0) {
4896 ret = 0;
4897 goto out;
4898 }
4899
Filipe David Borba Manana28e5dd82014-01-12 02:26:28 +00004900 if (clone_root && IS_ALIGNED(offset + len, bs)) {
Filipe Mananad906d492015-10-02 10:47:34 +01004901 u64 disk_byte;
4902 u64 data_offset;
4903
4904 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
4905 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
4906 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
4907 offset, len);
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004908 } else {
Filipe Mananad906d492015-10-02 10:47:34 +01004909 ret = send_extent_data(sctx, offset, len);
Alexander Block31db9f72012-07-25 23:19:24 +02004910 }
Alexander Block31db9f72012-07-25 23:19:24 +02004911out:
4912 return ret;
4913}
4914
4915static int is_extent_unchanged(struct send_ctx *sctx,
4916 struct btrfs_path *left_path,
4917 struct btrfs_key *ekey)
4918{
4919 int ret = 0;
4920 struct btrfs_key key;
4921 struct btrfs_path *path = NULL;
4922 struct extent_buffer *eb;
4923 int slot;
4924 struct btrfs_key found_key;
4925 struct btrfs_file_extent_item *ei;
4926 u64 left_disknr;
4927 u64 right_disknr;
4928 u64 left_offset;
4929 u64 right_offset;
4930 u64 left_offset_fixed;
4931 u64 left_len;
4932 u64 right_len;
Chris Mason74dd17f2012-08-07 16:25:13 -04004933 u64 left_gen;
4934 u64 right_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004935 u8 left_type;
4936 u8 right_type;
4937
4938 path = alloc_path_for_send();
4939 if (!path)
4940 return -ENOMEM;
4941
4942 eb = left_path->nodes[0];
4943 slot = left_path->slots[0];
Alexander Block31db9f72012-07-25 23:19:24 +02004944 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4945 left_type = btrfs_file_extent_type(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02004946
4947 if (left_type != BTRFS_FILE_EXTENT_REG) {
4948 ret = 0;
4949 goto out;
4950 }
Chris Mason74dd17f2012-08-07 16:25:13 -04004951 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4952 left_len = btrfs_file_extent_num_bytes(eb, ei);
4953 left_offset = btrfs_file_extent_offset(eb, ei);
4954 left_gen = btrfs_file_extent_generation(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02004955
4956 /*
4957 * Following comments will refer to these graphics. L is the left
4958 * extents which we are checking at the moment. 1-8 are the right
4959 * extents that we iterate.
4960 *
4961 * |-----L-----|
4962 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4963 *
4964 * |-----L-----|
4965 * |--1--|-2b-|...(same as above)
4966 *
4967 * Alternative situation. Happens on files where extents got split.
4968 * |-----L-----|
4969 * |-----------7-----------|-6-|
4970 *
4971 * Alternative situation. Happens on files which got larger.
4972 * |-----L-----|
4973 * |-8-|
4974 * Nothing follows after 8.
4975 */
4976
4977 key.objectid = ekey->objectid;
4978 key.type = BTRFS_EXTENT_DATA_KEY;
4979 key.offset = ekey->offset;
4980 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4981 if (ret < 0)
4982 goto out;
4983 if (ret) {
4984 ret = 0;
4985 goto out;
4986 }
4987
4988 /*
4989 * Handle special case where the right side has no extents at all.
4990 */
4991 eb = path->nodes[0];
4992 slot = path->slots[0];
4993 btrfs_item_key_to_cpu(eb, &found_key, slot);
4994 if (found_key.objectid != key.objectid ||
4995 found_key.type != key.type) {
Josef Bacik57cfd462013-08-20 15:55:39 -04004996 /* If we're a hole then just pretend nothing changed */
4997 ret = (left_disknr) ? 0 : 1;
Alexander Block31db9f72012-07-25 23:19:24 +02004998 goto out;
4999 }
5000
5001 /*
5002 * We're now on 2a, 2b or 7.
5003 */
5004 key = found_key;
5005 while (key.offset < ekey->offset + left_len) {
5006 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5007 right_type = btrfs_file_extent_type(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02005008 if (right_type != BTRFS_FILE_EXTENT_REG) {
5009 ret = 0;
5010 goto out;
5011 }
5012
Josef Bacik007d31f2013-10-31 16:49:02 -04005013 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5014 right_len = btrfs_file_extent_num_bytes(eb, ei);
5015 right_offset = btrfs_file_extent_offset(eb, ei);
5016 right_gen = btrfs_file_extent_generation(eb, ei);
5017
Alexander Block31db9f72012-07-25 23:19:24 +02005018 /*
5019 * Are we at extent 8? If yes, we know the extent is changed.
5020 * This may only happen on the first iteration.
5021 */
Alexander Blockd8347fa2012-08-01 12:49:15 +02005022 if (found_key.offset + right_len <= ekey->offset) {
Josef Bacik57cfd462013-08-20 15:55:39 -04005023 /* If we're a hole just pretend nothing changed */
5024 ret = (left_disknr) ? 0 : 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005025 goto out;
5026 }
5027
5028 left_offset_fixed = left_offset;
5029 if (key.offset < ekey->offset) {
5030 /* Fix the right offset for 2a and 7. */
5031 right_offset += ekey->offset - key.offset;
5032 } else {
5033 /* Fix the left offset for all behind 2a and 2b */
5034 left_offset_fixed += key.offset - ekey->offset;
5035 }
5036
5037 /*
5038 * Check if we have the same extent.
5039 */
Alexander Block39540962012-08-01 12:46:05 +02005040 if (left_disknr != right_disknr ||
Chris Mason74dd17f2012-08-07 16:25:13 -04005041 left_offset_fixed != right_offset ||
5042 left_gen != right_gen) {
Alexander Block31db9f72012-07-25 23:19:24 +02005043 ret = 0;
5044 goto out;
5045 }
5046
5047 /*
5048 * Go to the next extent.
5049 */
5050 ret = btrfs_next_item(sctx->parent_root, path);
5051 if (ret < 0)
5052 goto out;
5053 if (!ret) {
5054 eb = path->nodes[0];
5055 slot = path->slots[0];
5056 btrfs_item_key_to_cpu(eb, &found_key, slot);
5057 }
5058 if (ret || found_key.objectid != key.objectid ||
5059 found_key.type != key.type) {
5060 key.offset += right_len;
5061 break;
Jan Schmidtadaa4b82013-03-21 14:30:23 +00005062 }
5063 if (found_key.offset != key.offset + right_len) {
5064 ret = 0;
5065 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005066 }
5067 key = found_key;
5068 }
5069
5070 /*
5071 * We're now behind the left extent (treat as unchanged) or at the end
5072 * of the right side (treat as changed).
5073 */
5074 if (key.offset >= ekey->offset + left_len)
5075 ret = 1;
5076 else
5077 ret = 0;
5078
5079
5080out:
5081 btrfs_free_path(path);
5082 return ret;
5083}
5084
Josef Bacik16e75492013-10-22 12:18:51 -04005085static int get_last_extent(struct send_ctx *sctx, u64 offset)
5086{
5087 struct btrfs_path *path;
5088 struct btrfs_root *root = sctx->send_root;
5089 struct btrfs_file_extent_item *fi;
5090 struct btrfs_key key;
5091 u64 extent_end;
5092 u8 type;
5093 int ret;
5094
5095 path = alloc_path_for_send();
5096 if (!path)
5097 return -ENOMEM;
5098
5099 sctx->cur_inode_last_extent = 0;
5100
5101 key.objectid = sctx->cur_ino;
5102 key.type = BTRFS_EXTENT_DATA_KEY;
5103 key.offset = offset;
5104 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5105 if (ret < 0)
5106 goto out;
5107 ret = 0;
5108 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5109 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5110 goto out;
5111
5112 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5113 struct btrfs_file_extent_item);
5114 type = btrfs_file_extent_type(path->nodes[0], fi);
5115 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08005116 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5117 path->slots[0], fi);
Josef Bacik16e75492013-10-22 12:18:51 -04005118 extent_end = ALIGN(key.offset + size,
5119 sctx->send_root->sectorsize);
5120 } else {
5121 extent_end = key.offset +
5122 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5123 }
5124 sctx->cur_inode_last_extent = extent_end;
5125out:
5126 btrfs_free_path(path);
5127 return ret;
5128}
5129
5130static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5131 struct btrfs_key *key)
5132{
5133 struct btrfs_file_extent_item *fi;
5134 u64 extent_end;
5135 u8 type;
5136 int ret = 0;
5137
5138 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5139 return 0;
5140
5141 if (sctx->cur_inode_last_extent == (u64)-1) {
5142 ret = get_last_extent(sctx, key->offset - 1);
5143 if (ret)
5144 return ret;
5145 }
5146
5147 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5148 struct btrfs_file_extent_item);
5149 type = btrfs_file_extent_type(path->nodes[0], fi);
5150 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08005151 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5152 path->slots[0], fi);
Josef Bacik16e75492013-10-22 12:18:51 -04005153 extent_end = ALIGN(key->offset + size,
5154 sctx->send_root->sectorsize);
5155 } else {
5156 extent_end = key->offset +
5157 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5158 }
Filipe David Borba Mananabf54f412014-01-28 01:38:06 +00005159
5160 if (path->slots[0] == 0 &&
5161 sctx->cur_inode_last_extent < key->offset) {
5162 /*
5163 * We might have skipped entire leafs that contained only
5164 * file extent items for our current inode. These leafs have
5165 * a generation number smaller (older) than the one in the
5166 * current leaf and the leaf our last extent came from, and
5167 * are located between these 2 leafs.
5168 */
5169 ret = get_last_extent(sctx, key->offset - 1);
5170 if (ret)
5171 return ret;
5172 }
5173
Josef Bacik16e75492013-10-22 12:18:51 -04005174 if (sctx->cur_inode_last_extent < key->offset)
5175 ret = send_hole(sctx, key->offset);
5176 sctx->cur_inode_last_extent = extent_end;
5177 return ret;
5178}
5179
Alexander Block31db9f72012-07-25 23:19:24 +02005180static int process_extent(struct send_ctx *sctx,
5181 struct btrfs_path *path,
5182 struct btrfs_key *key)
5183{
Alexander Block31db9f72012-07-25 23:19:24 +02005184 struct clone_root *found_clone = NULL;
Josef Bacik57cfd462013-08-20 15:55:39 -04005185 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02005186
5187 if (S_ISLNK(sctx->cur_inode_mode))
5188 return 0;
5189
5190 if (sctx->parent_root && !sctx->cur_inode_new) {
5191 ret = is_extent_unchanged(sctx, path, key);
5192 if (ret < 0)
5193 goto out;
5194 if (ret) {
5195 ret = 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005196 goto out_hole;
Alexander Block31db9f72012-07-25 23:19:24 +02005197 }
Josef Bacik57cfd462013-08-20 15:55:39 -04005198 } else {
5199 struct btrfs_file_extent_item *ei;
5200 u8 type;
5201
5202 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5203 struct btrfs_file_extent_item);
5204 type = btrfs_file_extent_type(path->nodes[0], ei);
5205 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5206 type == BTRFS_FILE_EXTENT_REG) {
5207 /*
5208 * The send spec does not have a prealloc command yet,
5209 * so just leave a hole for prealloc'ed extents until
5210 * we have enough commands queued up to justify rev'ing
5211 * the send spec.
5212 */
5213 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5214 ret = 0;
5215 goto out;
5216 }
5217
5218 /* Have a hole, just skip it. */
5219 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5220 ret = 0;
5221 goto out;
5222 }
5223 }
Alexander Block31db9f72012-07-25 23:19:24 +02005224 }
5225
5226 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5227 sctx->cur_inode_size, &found_clone);
5228 if (ret != -ENOENT && ret < 0)
5229 goto out;
5230
5231 ret = send_write_or_clone(sctx, path, key, found_clone);
Josef Bacik16e75492013-10-22 12:18:51 -04005232 if (ret)
5233 goto out;
5234out_hole:
5235 ret = maybe_send_hole(sctx, path, key);
Alexander Block31db9f72012-07-25 23:19:24 +02005236out:
5237 return ret;
5238}
5239
5240static int process_all_extents(struct send_ctx *sctx)
5241{
5242 int ret;
5243 struct btrfs_root *root;
5244 struct btrfs_path *path;
5245 struct btrfs_key key;
5246 struct btrfs_key found_key;
5247 struct extent_buffer *eb;
5248 int slot;
5249
5250 root = sctx->send_root;
5251 path = alloc_path_for_send();
5252 if (!path)
5253 return -ENOMEM;
5254
5255 key.objectid = sctx->cmp_key->objectid;
5256 key.type = BTRFS_EXTENT_DATA_KEY;
5257 key.offset = 0;
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005258 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5259 if (ret < 0)
5260 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005261
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005262 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02005263 eb = path->nodes[0];
5264 slot = path->slots[0];
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005265
5266 if (slot >= btrfs_header_nritems(eb)) {
5267 ret = btrfs_next_leaf(root, path);
5268 if (ret < 0) {
5269 goto out;
5270 } else if (ret > 0) {
5271 ret = 0;
5272 break;
5273 }
5274 continue;
5275 }
5276
Alexander Block31db9f72012-07-25 23:19:24 +02005277 btrfs_item_key_to_cpu(eb, &found_key, slot);
5278
5279 if (found_key.objectid != key.objectid ||
5280 found_key.type != key.type) {
5281 ret = 0;
5282 goto out;
5283 }
5284
5285 ret = process_extent(sctx, path, &found_key);
5286 if (ret < 0)
5287 goto out;
5288
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005289 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02005290 }
5291
5292out:
5293 btrfs_free_path(path);
5294 return ret;
5295}
5296
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005297static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5298 int *pending_move,
5299 int *refs_processed)
Alexander Block31db9f72012-07-25 23:19:24 +02005300{
5301 int ret = 0;
5302
5303 if (sctx->cur_ino == 0)
5304 goto out;
5305 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
Jan Schmidt96b5bd72012-10-15 08:30:45 +00005306 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
Alexander Block31db9f72012-07-25 23:19:24 +02005307 goto out;
5308 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5309 goto out;
5310
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005311 ret = process_recorded_refs(sctx, pending_move);
Alexander Blocke479d9b2012-07-28 16:09:35 +02005312 if (ret < 0)
5313 goto out;
5314
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005315 *refs_processed = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005316out:
5317 return ret;
5318}
5319
5320static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5321{
5322 int ret = 0;
5323 u64 left_mode;
5324 u64 left_uid;
5325 u64 left_gid;
5326 u64 right_mode;
5327 u64 right_uid;
5328 u64 right_gid;
5329 int need_chmod = 0;
5330 int need_chown = 0;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005331 int pending_move = 0;
5332 int refs_processed = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02005333
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005334 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5335 &refs_processed);
Alexander Block31db9f72012-07-25 23:19:24 +02005336 if (ret < 0)
5337 goto out;
5338
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005339 /*
5340 * We have processed the refs and thus need to advance send_progress.
5341 * Now, calls to get_cur_xxx will take the updated refs of the current
5342 * inode into account.
5343 *
5344 * On the other hand, if our current inode is a directory and couldn't
5345 * be moved/renamed because its parent was renamed/moved too and it has
5346 * a higher inode number, we can only move/rename our current inode
5347 * after we moved/renamed its parent. Therefore in this case operate on
5348 * the old path (pre move/rename) of our current inode, and the
5349 * move/rename will be performed later.
5350 */
5351 if (refs_processed && !pending_move)
5352 sctx->send_progress = sctx->cur_ino + 1;
5353
Alexander Block31db9f72012-07-25 23:19:24 +02005354 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5355 goto out;
5356 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5357 goto out;
5358
5359 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02005360 &left_mode, &left_uid, &left_gid, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02005361 if (ret < 0)
5362 goto out;
5363
Alex Lyakase2d044f2012-10-17 13:52:47 +00005364 if (!sctx->parent_root || sctx->cur_inode_new) {
5365 need_chown = 1;
5366 if (!S_ISLNK(sctx->cur_inode_mode))
Alexander Block31db9f72012-07-25 23:19:24 +02005367 need_chmod = 1;
Alex Lyakase2d044f2012-10-17 13:52:47 +00005368 } else {
5369 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5370 NULL, NULL, &right_mode, &right_uid,
5371 &right_gid, NULL);
5372 if (ret < 0)
5373 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005374
Alex Lyakase2d044f2012-10-17 13:52:47 +00005375 if (left_uid != right_uid || left_gid != right_gid)
5376 need_chown = 1;
5377 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5378 need_chmod = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005379 }
5380
5381 if (S_ISREG(sctx->cur_inode_mode)) {
Josef Bacik16e75492013-10-22 12:18:51 -04005382 if (need_send_hole(sctx)) {
Filipe Manana766b5e52014-03-30 23:02:53 +01005383 if (sctx->cur_inode_last_extent == (u64)-1 ||
5384 sctx->cur_inode_last_extent <
5385 sctx->cur_inode_size) {
Josef Bacik16e75492013-10-22 12:18:51 -04005386 ret = get_last_extent(sctx, (u64)-1);
5387 if (ret)
5388 goto out;
5389 }
5390 if (sctx->cur_inode_last_extent <
5391 sctx->cur_inode_size) {
5392 ret = send_hole(sctx, sctx->cur_inode_size);
5393 if (ret)
5394 goto out;
5395 }
5396 }
Alexander Block31db9f72012-07-25 23:19:24 +02005397 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5398 sctx->cur_inode_size);
5399 if (ret < 0)
5400 goto out;
5401 }
5402
5403 if (need_chown) {
5404 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5405 left_uid, left_gid);
5406 if (ret < 0)
5407 goto out;
5408 }
5409 if (need_chmod) {
5410 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5411 left_mode);
5412 if (ret < 0)
5413 goto out;
5414 }
5415
5416 /*
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005417 * If other directory inodes depended on our current directory
5418 * inode's move/rename, now do their move/rename operations.
Alexander Block31db9f72012-07-25 23:19:24 +02005419 */
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005420 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5421 ret = apply_children_dir_moves(sctx);
5422 if (ret)
5423 goto out;
Filipe Mananafcbd2152014-03-03 12:28:40 +00005424 /*
5425 * Need to send that every time, no matter if it actually
5426 * changed between the two trees as we have done changes to
5427 * the inode before. If our inode is a directory and it's
5428 * waiting to be moved/renamed, we will send its utimes when
5429 * it's moved/renamed, therefore we don't need to do it here.
5430 */
5431 sctx->send_progress = sctx->cur_ino + 1;
5432 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5433 if (ret < 0)
5434 goto out;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005435 }
5436
Alexander Block31db9f72012-07-25 23:19:24 +02005437out:
5438 return ret;
5439}
5440
5441static int changed_inode(struct send_ctx *sctx,
5442 enum btrfs_compare_tree_result result)
5443{
5444 int ret = 0;
5445 struct btrfs_key *key = sctx->cmp_key;
5446 struct btrfs_inode_item *left_ii = NULL;
5447 struct btrfs_inode_item *right_ii = NULL;
5448 u64 left_gen = 0;
5449 u64 right_gen = 0;
5450
Alexander Block31db9f72012-07-25 23:19:24 +02005451 sctx->cur_ino = key->objectid;
5452 sctx->cur_inode_new_gen = 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005453 sctx->cur_inode_last_extent = (u64)-1;
Alexander Blocke479d9b2012-07-28 16:09:35 +02005454
5455 /*
5456 * Set send_progress to current inode. This will tell all get_cur_xxx
5457 * functions that the current inode's refs are not updated yet. Later,
5458 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5459 */
Alexander Block31db9f72012-07-25 23:19:24 +02005460 sctx->send_progress = sctx->cur_ino;
5461
5462 if (result == BTRFS_COMPARE_TREE_NEW ||
5463 result == BTRFS_COMPARE_TREE_CHANGED) {
5464 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5465 sctx->left_path->slots[0],
5466 struct btrfs_inode_item);
5467 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5468 left_ii);
5469 } else {
5470 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5471 sctx->right_path->slots[0],
5472 struct btrfs_inode_item);
5473 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5474 right_ii);
5475 }
5476 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5477 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5478 sctx->right_path->slots[0],
5479 struct btrfs_inode_item);
5480
5481 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5482 right_ii);
Alexander Block6d85ed02012-08-01 14:48:59 +02005483
5484 /*
5485 * The cur_ino = root dir case is special here. We can't treat
5486 * the inode as deleted+reused because it would generate a
5487 * stream that tries to delete/mkdir the root dir.
5488 */
5489 if (left_gen != right_gen &&
5490 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
Alexander Block31db9f72012-07-25 23:19:24 +02005491 sctx->cur_inode_new_gen = 1;
5492 }
5493
5494 if (result == BTRFS_COMPARE_TREE_NEW) {
5495 sctx->cur_inode_gen = left_gen;
5496 sctx->cur_inode_new = 1;
5497 sctx->cur_inode_deleted = 0;
5498 sctx->cur_inode_size = btrfs_inode_size(
5499 sctx->left_path->nodes[0], left_ii);
5500 sctx->cur_inode_mode = btrfs_inode_mode(
5501 sctx->left_path->nodes[0], left_ii);
Liu Bo644d1942014-02-27 17:29:01 +08005502 sctx->cur_inode_rdev = btrfs_inode_rdev(
5503 sctx->left_path->nodes[0], left_ii);
Alexander Block31db9f72012-07-25 23:19:24 +02005504 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
Alexander Block1f4692d2012-07-28 10:42:24 +02005505 ret = send_create_inode_if_needed(sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02005506 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5507 sctx->cur_inode_gen = right_gen;
5508 sctx->cur_inode_new = 0;
5509 sctx->cur_inode_deleted = 1;
5510 sctx->cur_inode_size = btrfs_inode_size(
5511 sctx->right_path->nodes[0], right_ii);
5512 sctx->cur_inode_mode = btrfs_inode_mode(
5513 sctx->right_path->nodes[0], right_ii);
5514 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
Alexander Block766702e2012-07-28 14:11:31 +02005515 /*
5516 * We need to do some special handling in case the inode was
5517 * reported as changed with a changed generation number. This
5518 * means that the original inode was deleted and new inode
5519 * reused the same inum. So we have to treat the old inode as
5520 * deleted and the new one as new.
5521 */
Alexander Block31db9f72012-07-25 23:19:24 +02005522 if (sctx->cur_inode_new_gen) {
Alexander Block766702e2012-07-28 14:11:31 +02005523 /*
5524 * First, process the inode as if it was deleted.
5525 */
Alexander Block31db9f72012-07-25 23:19:24 +02005526 sctx->cur_inode_gen = right_gen;
5527 sctx->cur_inode_new = 0;
5528 sctx->cur_inode_deleted = 1;
5529 sctx->cur_inode_size = btrfs_inode_size(
5530 sctx->right_path->nodes[0], right_ii);
5531 sctx->cur_inode_mode = btrfs_inode_mode(
5532 sctx->right_path->nodes[0], right_ii);
5533 ret = process_all_refs(sctx,
5534 BTRFS_COMPARE_TREE_DELETED);
5535 if (ret < 0)
5536 goto out;
5537
Alexander Block766702e2012-07-28 14:11:31 +02005538 /*
5539 * Now process the inode as if it was new.
5540 */
Alexander Block31db9f72012-07-25 23:19:24 +02005541 sctx->cur_inode_gen = left_gen;
5542 sctx->cur_inode_new = 1;
5543 sctx->cur_inode_deleted = 0;
5544 sctx->cur_inode_size = btrfs_inode_size(
5545 sctx->left_path->nodes[0], left_ii);
5546 sctx->cur_inode_mode = btrfs_inode_mode(
5547 sctx->left_path->nodes[0], left_ii);
Liu Bo644d1942014-02-27 17:29:01 +08005548 sctx->cur_inode_rdev = btrfs_inode_rdev(
5549 sctx->left_path->nodes[0], left_ii);
Alexander Block1f4692d2012-07-28 10:42:24 +02005550 ret = send_create_inode_if_needed(sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02005551 if (ret < 0)
5552 goto out;
5553
5554 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5555 if (ret < 0)
5556 goto out;
Alexander Blocke479d9b2012-07-28 16:09:35 +02005557 /*
5558 * Advance send_progress now as we did not get into
5559 * process_recorded_refs_if_needed in the new_gen case.
5560 */
5561 sctx->send_progress = sctx->cur_ino + 1;
Alexander Block766702e2012-07-28 14:11:31 +02005562
5563 /*
5564 * Now process all extents and xattrs of the inode as if
5565 * they were all new.
5566 */
Alexander Block31db9f72012-07-25 23:19:24 +02005567 ret = process_all_extents(sctx);
5568 if (ret < 0)
5569 goto out;
5570 ret = process_all_new_xattrs(sctx);
5571 if (ret < 0)
5572 goto out;
5573 } else {
5574 sctx->cur_inode_gen = left_gen;
5575 sctx->cur_inode_new = 0;
5576 sctx->cur_inode_new_gen = 0;
5577 sctx->cur_inode_deleted = 0;
5578 sctx->cur_inode_size = btrfs_inode_size(
5579 sctx->left_path->nodes[0], left_ii);
5580 sctx->cur_inode_mode = btrfs_inode_mode(
5581 sctx->left_path->nodes[0], left_ii);
5582 }
5583 }
5584
5585out:
5586 return ret;
5587}
5588
Alexander Block766702e2012-07-28 14:11:31 +02005589/*
5590 * We have to process new refs before deleted refs, but compare_trees gives us
5591 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5592 * first and later process them in process_recorded_refs.
5593 * For the cur_inode_new_gen case, we skip recording completely because
5594 * changed_inode did already initiate processing of refs. The reason for this is
5595 * that in this case, compare_tree actually compares the refs of 2 different
5596 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5597 * refs of the right tree as deleted and all refs of the left tree as new.
5598 */
Alexander Block31db9f72012-07-25 23:19:24 +02005599static int changed_ref(struct send_ctx *sctx,
5600 enum btrfs_compare_tree_result result)
5601{
5602 int ret = 0;
5603
5604 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5605
5606 if (!sctx->cur_inode_new_gen &&
5607 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5608 if (result == BTRFS_COMPARE_TREE_NEW)
5609 ret = record_new_ref(sctx);
5610 else if (result == BTRFS_COMPARE_TREE_DELETED)
5611 ret = record_deleted_ref(sctx);
5612 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5613 ret = record_changed_ref(sctx);
5614 }
5615
5616 return ret;
5617}
5618
Alexander Block766702e2012-07-28 14:11:31 +02005619/*
5620 * Process new/deleted/changed xattrs. We skip processing in the
5621 * cur_inode_new_gen case because changed_inode did already initiate processing
5622 * of xattrs. The reason is the same as in changed_ref
5623 */
Alexander Block31db9f72012-07-25 23:19:24 +02005624static int changed_xattr(struct send_ctx *sctx,
5625 enum btrfs_compare_tree_result result)
5626{
5627 int ret = 0;
5628
5629 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5630
5631 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5632 if (result == BTRFS_COMPARE_TREE_NEW)
5633 ret = process_new_xattr(sctx);
5634 else if (result == BTRFS_COMPARE_TREE_DELETED)
5635 ret = process_deleted_xattr(sctx);
5636 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5637 ret = process_changed_xattr(sctx);
5638 }
5639
5640 return ret;
5641}
5642
Alexander Block766702e2012-07-28 14:11:31 +02005643/*
5644 * Process new/deleted/changed extents. We skip processing in the
5645 * cur_inode_new_gen case because changed_inode did already initiate processing
5646 * of extents. The reason is the same as in changed_ref
5647 */
Alexander Block31db9f72012-07-25 23:19:24 +02005648static int changed_extent(struct send_ctx *sctx,
5649 enum btrfs_compare_tree_result result)
5650{
5651 int ret = 0;
5652
5653 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5654
5655 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5656 if (result != BTRFS_COMPARE_TREE_DELETED)
5657 ret = process_extent(sctx, sctx->left_path,
5658 sctx->cmp_key);
5659 }
5660
5661 return ret;
5662}
5663
Josef Bacikba5e8f22013-08-16 16:52:55 -04005664static int dir_changed(struct send_ctx *sctx, u64 dir)
5665{
5666 u64 orig_gen, new_gen;
5667 int ret;
5668
5669 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5670 NULL, NULL);
5671 if (ret)
5672 return ret;
5673
5674 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5675 NULL, NULL, NULL);
5676 if (ret)
5677 return ret;
5678
5679 return (orig_gen != new_gen) ? 1 : 0;
5680}
5681
5682static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5683 struct btrfs_key *key)
5684{
5685 struct btrfs_inode_extref *extref;
5686 struct extent_buffer *leaf;
5687 u64 dirid = 0, last_dirid = 0;
5688 unsigned long ptr;
5689 u32 item_size;
5690 u32 cur_offset = 0;
5691 int ref_name_len;
5692 int ret = 0;
5693
5694 /* Easy case, just check this one dirid */
5695 if (key->type == BTRFS_INODE_REF_KEY) {
5696 dirid = key->offset;
5697
5698 ret = dir_changed(sctx, dirid);
5699 goto out;
5700 }
5701
5702 leaf = path->nodes[0];
5703 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5704 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5705 while (cur_offset < item_size) {
5706 extref = (struct btrfs_inode_extref *)(ptr +
5707 cur_offset);
5708 dirid = btrfs_inode_extref_parent(leaf, extref);
5709 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5710 cur_offset += ref_name_len + sizeof(*extref);
5711 if (dirid == last_dirid)
5712 continue;
5713 ret = dir_changed(sctx, dirid);
5714 if (ret)
5715 break;
5716 last_dirid = dirid;
5717 }
5718out:
5719 return ret;
5720}
5721
Alexander Block766702e2012-07-28 14:11:31 +02005722/*
5723 * Updates compare related fields in sctx and simply forwards to the actual
5724 * changed_xxx functions.
5725 */
Alexander Block31db9f72012-07-25 23:19:24 +02005726static int changed_cb(struct btrfs_root *left_root,
5727 struct btrfs_root *right_root,
5728 struct btrfs_path *left_path,
5729 struct btrfs_path *right_path,
5730 struct btrfs_key *key,
5731 enum btrfs_compare_tree_result result,
5732 void *ctx)
5733{
5734 int ret = 0;
5735 struct send_ctx *sctx = ctx;
5736
Josef Bacikba5e8f22013-08-16 16:52:55 -04005737 if (result == BTRFS_COMPARE_TREE_SAME) {
Josef Bacik16e75492013-10-22 12:18:51 -04005738 if (key->type == BTRFS_INODE_REF_KEY ||
5739 key->type == BTRFS_INODE_EXTREF_KEY) {
5740 ret = compare_refs(sctx, left_path, key);
5741 if (!ret)
5742 return 0;
5743 if (ret < 0)
5744 return ret;
5745 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5746 return maybe_send_hole(sctx, left_path, key);
5747 } else {
Josef Bacikba5e8f22013-08-16 16:52:55 -04005748 return 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005749 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04005750 result = BTRFS_COMPARE_TREE_CHANGED;
5751 ret = 0;
5752 }
5753
Alexander Block31db9f72012-07-25 23:19:24 +02005754 sctx->left_path = left_path;
5755 sctx->right_path = right_path;
5756 sctx->cmp_key = key;
5757
5758 ret = finish_inode_if_needed(sctx, 0);
5759 if (ret < 0)
5760 goto out;
5761
Alexander Block2981e222012-08-01 14:47:03 +02005762 /* Ignore non-FS objects */
5763 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5764 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5765 goto out;
5766
Alexander Block31db9f72012-07-25 23:19:24 +02005767 if (key->type == BTRFS_INODE_ITEM_KEY)
5768 ret = changed_inode(sctx, result);
Jan Schmidt96b5bd72012-10-15 08:30:45 +00005769 else if (key->type == BTRFS_INODE_REF_KEY ||
5770 key->type == BTRFS_INODE_EXTREF_KEY)
Alexander Block31db9f72012-07-25 23:19:24 +02005771 ret = changed_ref(sctx, result);
5772 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5773 ret = changed_xattr(sctx, result);
5774 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5775 ret = changed_extent(sctx, result);
5776
5777out:
5778 return ret;
5779}
5780
5781static int full_send_tree(struct send_ctx *sctx)
5782{
5783 int ret;
Alexander Block31db9f72012-07-25 23:19:24 +02005784 struct btrfs_root *send_root = sctx->send_root;
5785 struct btrfs_key key;
5786 struct btrfs_key found_key;
5787 struct btrfs_path *path;
5788 struct extent_buffer *eb;
5789 int slot;
Alexander Block31db9f72012-07-25 23:19:24 +02005790
5791 path = alloc_path_for_send();
5792 if (!path)
5793 return -ENOMEM;
5794
Alexander Block31db9f72012-07-25 23:19:24 +02005795 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5796 key.type = BTRFS_INODE_ITEM_KEY;
5797 key.offset = 0;
5798
Alexander Block31db9f72012-07-25 23:19:24 +02005799 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5800 if (ret < 0)
5801 goto out;
5802 if (ret)
5803 goto out_finish;
5804
5805 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02005806 eb = path->nodes[0];
5807 slot = path->slots[0];
5808 btrfs_item_key_to_cpu(eb, &found_key, slot);
5809
5810 ret = changed_cb(send_root, NULL, path, NULL,
5811 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5812 if (ret < 0)
5813 goto out;
5814
5815 key.objectid = found_key.objectid;
5816 key.type = found_key.type;
5817 key.offset = found_key.offset + 1;
5818
5819 ret = btrfs_next_item(send_root, path);
5820 if (ret < 0)
5821 goto out;
5822 if (ret) {
5823 ret = 0;
5824 break;
5825 }
5826 }
5827
5828out_finish:
5829 ret = finish_inode_if_needed(sctx, 1);
5830
5831out:
5832 btrfs_free_path(path);
Alexander Block31db9f72012-07-25 23:19:24 +02005833 return ret;
5834}
5835
5836static int send_subvol(struct send_ctx *sctx)
5837{
5838 int ret;
5839
Stefan Behrensc2c71322013-04-10 17:10:52 +00005840 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5841 ret = send_header(sctx);
5842 if (ret < 0)
5843 goto out;
5844 }
Alexander Block31db9f72012-07-25 23:19:24 +02005845
5846 ret = send_subvol_begin(sctx);
5847 if (ret < 0)
5848 goto out;
5849
5850 if (sctx->parent_root) {
5851 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5852 changed_cb, sctx);
5853 if (ret < 0)
5854 goto out;
5855 ret = finish_inode_if_needed(sctx, 1);
5856 if (ret < 0)
5857 goto out;
5858 } else {
5859 ret = full_send_tree(sctx);
5860 if (ret < 0)
5861 goto out;
5862 }
5863
5864out:
Alexander Block31db9f72012-07-25 23:19:24 +02005865 free_recorded_refs(sctx);
5866 return ret;
5867}
5868
Filipe Mananae5fa8f82014-10-21 11:11:41 +01005869/*
5870 * If orphan cleanup did remove any orphans from a root, it means the tree
5871 * was modified and therefore the commit root is not the same as the current
5872 * root anymore. This is a problem, because send uses the commit root and
5873 * therefore can see inode items that don't exist in the current root anymore,
5874 * and for example make calls to btrfs_iget, which will do tree lookups based
5875 * on the current root and not on the commit root. Those lookups will fail,
5876 * returning a -ESTALE error, and making send fail with that error. So make
5877 * sure a send does not see any orphans we have just removed, and that it will
5878 * see the same inodes regardless of whether a transaction commit happened
5879 * before it started (meaning that the commit root will be the same as the
5880 * current root) or not.
5881 */
5882static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
5883{
5884 int i;
5885 struct btrfs_trans_handle *trans = NULL;
5886
5887again:
5888 if (sctx->parent_root &&
5889 sctx->parent_root->node != sctx->parent_root->commit_root)
5890 goto commit_trans;
5891
5892 for (i = 0; i < sctx->clone_roots_cnt; i++)
5893 if (sctx->clone_roots[i].root->node !=
5894 sctx->clone_roots[i].root->commit_root)
5895 goto commit_trans;
5896
5897 if (trans)
5898 return btrfs_end_transaction(trans, sctx->send_root);
5899
5900 return 0;
5901
5902commit_trans:
5903 /* Use any root, all fs roots will get their commit roots updated. */
5904 if (!trans) {
5905 trans = btrfs_join_transaction(sctx->send_root);
5906 if (IS_ERR(trans))
5907 return PTR_ERR(trans);
5908 goto again;
5909 }
5910
5911 return btrfs_commit_transaction(trans, sctx->send_root);
5912}
5913
David Sterba66ef7d62013-12-17 15:07:20 +01005914static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5915{
5916 spin_lock(&root->root_item_lock);
5917 root->send_in_progress--;
5918 /*
5919 * Not much left to do, we don't know why it's unbalanced and
5920 * can't blindly reset it to 0.
5921 */
5922 if (root->send_in_progress < 0)
5923 btrfs_err(root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02005924 "send_in_progres unbalanced %d root %llu",
David Sterba66ef7d62013-12-17 15:07:20 +01005925 root->send_in_progress, root->root_key.objectid);
5926 spin_unlock(&root->root_item_lock);
5927}
5928
Alexander Block31db9f72012-07-25 23:19:24 +02005929long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5930{
5931 int ret = 0;
5932 struct btrfs_root *send_root;
5933 struct btrfs_root *clone_root;
5934 struct btrfs_fs_info *fs_info;
5935 struct btrfs_ioctl_send_args *arg = NULL;
5936 struct btrfs_key key;
Alexander Block31db9f72012-07-25 23:19:24 +02005937 struct send_ctx *sctx = NULL;
5938 u32 i;
5939 u64 *clone_sources_tmp = NULL;
David Sterba2c686532013-12-16 17:34:17 +01005940 int clone_sources_to_rollback = 0;
Wang Shilong896c14f2014-01-07 17:25:18 +08005941 int sort_clone_roots = 0;
Wang Shilong18f687d2014-01-07 17:25:19 +08005942 int index;
Alexander Block31db9f72012-07-25 23:19:24 +02005943
5944 if (!capable(CAP_SYS_ADMIN))
5945 return -EPERM;
5946
Al Viro496ad9a2013-01-23 17:07:38 -05005947 send_root = BTRFS_I(file_inode(mnt_file))->root;
Alexander Block31db9f72012-07-25 23:19:24 +02005948 fs_info = send_root->fs_info;
5949
Josef Bacik139f8072013-05-20 11:26:50 -04005950 /*
David Sterba2c686532013-12-16 17:34:17 +01005951 * The subvolume must remain read-only during send, protect against
David Sterba521e0542014-04-15 16:41:44 +02005952 * making it RW. This also protects against deletion.
David Sterba2c686532013-12-16 17:34:17 +01005953 */
5954 spin_lock(&send_root->root_item_lock);
5955 send_root->send_in_progress++;
5956 spin_unlock(&send_root->root_item_lock);
5957
5958 /*
Josef Bacik139f8072013-05-20 11:26:50 -04005959 * This is done when we lookup the root, it should already be complete
5960 * by the time we get here.
5961 */
5962 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5963
5964 /*
David Sterba2c686532013-12-16 17:34:17 +01005965 * Userspace tools do the checks and warn the user if it's
5966 * not RO.
5967 */
5968 if (!btrfs_root_readonly(send_root)) {
5969 ret = -EPERM;
5970 goto out;
5971 }
5972
Alexander Block31db9f72012-07-25 23:19:24 +02005973 arg = memdup_user(arg_, sizeof(*arg));
5974 if (IS_ERR(arg)) {
5975 ret = PTR_ERR(arg);
5976 arg = NULL;
5977 goto out;
5978 }
5979
5980 if (!access_ok(VERIFY_READ, arg->clone_sources,
Dan Carpenter700ff4f2013-01-10 03:57:25 -05005981 sizeof(*arg->clone_sources) *
5982 arg->clone_sources_count)) {
Alexander Block31db9f72012-07-25 23:19:24 +02005983 ret = -EFAULT;
5984 goto out;
5985 }
5986
Stefan Behrensc2c71322013-04-10 17:10:52 +00005987 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
Mark Fashehcb95e7b2013-02-04 20:54:57 +00005988 ret = -EINVAL;
5989 goto out;
5990 }
5991
Alexander Block31db9f72012-07-25 23:19:24 +02005992 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5993 if (!sctx) {
5994 ret = -ENOMEM;
5995 goto out;
5996 }
5997
5998 INIT_LIST_HEAD(&sctx->new_refs);
5999 INIT_LIST_HEAD(&sctx->deleted_refs);
6000 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
6001 INIT_LIST_HEAD(&sctx->name_cache_list);
6002
Mark Fashehcb95e7b2013-02-04 20:54:57 +00006003 sctx->flags = arg->flags;
6004
Alexander Block31db9f72012-07-25 23:19:24 +02006005 sctx->send_filp = fget(arg->send_fd);
Tsutomu Itohecc7ada2013-04-19 01:04:46 +00006006 if (!sctx->send_filp) {
6007 ret = -EBADF;
Alexander Block31db9f72012-07-25 23:19:24 +02006008 goto out;
6009 }
6010
Alexander Block31db9f72012-07-25 23:19:24 +02006011 sctx->send_root = send_root;
David Sterba521e0542014-04-15 16:41:44 +02006012 /*
6013 * Unlikely but possible, if the subvolume is marked for deletion but
6014 * is slow to remove the directory entry, send can still be started
6015 */
6016 if (btrfs_root_dead(sctx->send_root)) {
6017 ret = -EPERM;
6018 goto out;
6019 }
6020
Alexander Block31db9f72012-07-25 23:19:24 +02006021 sctx->clone_roots_cnt = arg->clone_sources_count;
6022
6023 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6024 sctx->send_buf = vmalloc(sctx->send_max_size);
6025 if (!sctx->send_buf) {
6026 ret = -ENOMEM;
6027 goto out;
6028 }
6029
6030 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
6031 if (!sctx->read_buf) {
6032 ret = -ENOMEM;
6033 goto out;
6034 }
6035
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006036 sctx->pending_dir_moves = RB_ROOT;
6037 sctx->waiting_dir_moves = RB_ROOT;
Filipe Manana9dc44212014-02-19 14:31:44 +00006038 sctx->orphan_dirs = RB_ROOT;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006039
Alexander Block31db9f72012-07-25 23:19:24 +02006040 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
6041 (arg->clone_sources_count + 1));
6042 if (!sctx->clone_roots) {
6043 ret = -ENOMEM;
6044 goto out;
6045 }
6046
6047 if (arg->clone_sources_count) {
6048 clone_sources_tmp = vmalloc(arg->clone_sources_count *
6049 sizeof(*arg->clone_sources));
6050 if (!clone_sources_tmp) {
6051 ret = -ENOMEM;
6052 goto out;
6053 }
6054
6055 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
6056 arg->clone_sources_count *
6057 sizeof(*arg->clone_sources));
6058 if (ret) {
6059 ret = -EFAULT;
6060 goto out;
6061 }
6062
6063 for (i = 0; i < arg->clone_sources_count; i++) {
6064 key.objectid = clone_sources_tmp[i];
6065 key.type = BTRFS_ROOT_ITEM_KEY;
6066 key.offset = (u64)-1;
Wang Shilong18f687d2014-01-07 17:25:19 +08006067
6068 index = srcu_read_lock(&fs_info->subvol_srcu);
6069
Alexander Block31db9f72012-07-25 23:19:24 +02006070 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
Alexander Block31db9f72012-07-25 23:19:24 +02006071 if (IS_ERR(clone_root)) {
Wang Shilong18f687d2014-01-07 17:25:19 +08006072 srcu_read_unlock(&fs_info->subvol_srcu, index);
Alexander Block31db9f72012-07-25 23:19:24 +02006073 ret = PTR_ERR(clone_root);
6074 goto out;
6075 }
David Sterba2c686532013-12-16 17:34:17 +01006076 spin_lock(&clone_root->root_item_lock);
Filipe Manana5cc2b172015-03-02 20:53:52 +00006077 if (!btrfs_root_readonly(clone_root) ||
6078 btrfs_root_dead(clone_root)) {
David Sterba2c686532013-12-16 17:34:17 +01006079 spin_unlock(&clone_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006080 srcu_read_unlock(&fs_info->subvol_srcu, index);
David Sterba2c686532013-12-16 17:34:17 +01006081 ret = -EPERM;
6082 goto out;
6083 }
Filipe Manana2f1f4652015-03-02 20:53:53 +00006084 clone_root->send_in_progress++;
David Sterba2c686532013-12-16 17:34:17 +01006085 spin_unlock(&clone_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006086 srcu_read_unlock(&fs_info->subvol_srcu, index);
6087
Alexander Block31db9f72012-07-25 23:19:24 +02006088 sctx->clone_roots[i].root = clone_root;
Filipe Manana2f1f4652015-03-02 20:53:53 +00006089 clone_sources_to_rollback = i + 1;
Alexander Block31db9f72012-07-25 23:19:24 +02006090 }
6091 vfree(clone_sources_tmp);
6092 clone_sources_tmp = NULL;
6093 }
6094
6095 if (arg->parent_root) {
6096 key.objectid = arg->parent_root;
6097 key.type = BTRFS_ROOT_ITEM_KEY;
6098 key.offset = (u64)-1;
Wang Shilong18f687d2014-01-07 17:25:19 +08006099
6100 index = srcu_read_lock(&fs_info->subvol_srcu);
6101
Alexander Block31db9f72012-07-25 23:19:24 +02006102 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
Stefan Behrensb1b19592013-05-13 14:42:57 +00006103 if (IS_ERR(sctx->parent_root)) {
Wang Shilong18f687d2014-01-07 17:25:19 +08006104 srcu_read_unlock(&fs_info->subvol_srcu, index);
Stefan Behrensb1b19592013-05-13 14:42:57 +00006105 ret = PTR_ERR(sctx->parent_root);
Alexander Block31db9f72012-07-25 23:19:24 +02006106 goto out;
6107 }
Wang Shilong18f687d2014-01-07 17:25:19 +08006108
David Sterba2c686532013-12-16 17:34:17 +01006109 spin_lock(&sctx->parent_root->root_item_lock);
6110 sctx->parent_root->send_in_progress++;
David Sterba521e0542014-04-15 16:41:44 +02006111 if (!btrfs_root_readonly(sctx->parent_root) ||
6112 btrfs_root_dead(sctx->parent_root)) {
David Sterba2c686532013-12-16 17:34:17 +01006113 spin_unlock(&sctx->parent_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006114 srcu_read_unlock(&fs_info->subvol_srcu, index);
David Sterba2c686532013-12-16 17:34:17 +01006115 ret = -EPERM;
6116 goto out;
6117 }
6118 spin_unlock(&sctx->parent_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006119
6120 srcu_read_unlock(&fs_info->subvol_srcu, index);
Alexander Block31db9f72012-07-25 23:19:24 +02006121 }
6122
6123 /*
6124 * Clones from send_root are allowed, but only if the clone source
6125 * is behind the current send position. This is checked while searching
6126 * for possible clone sources.
6127 */
6128 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6129
6130 /* We do a bsearch later */
6131 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6132 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6133 NULL);
Wang Shilong896c14f2014-01-07 17:25:18 +08006134 sort_clone_roots = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02006135
Filipe Mananae5fa8f82014-10-21 11:11:41 +01006136 ret = ensure_commit_roots_uptodate(sctx);
6137 if (ret)
6138 goto out;
6139
David Sterba2755a0d2014-07-31 00:43:18 +02006140 current->journal_info = BTRFS_SEND_TRANS_STUB;
Alexander Block31db9f72012-07-25 23:19:24 +02006141 ret = send_subvol(sctx);
Josef Bacika26e8c92014-03-28 17:07:27 -04006142 current->journal_info = NULL;
Alexander Block31db9f72012-07-25 23:19:24 +02006143 if (ret < 0)
6144 goto out;
6145
Stefan Behrensc2c71322013-04-10 17:10:52 +00006146 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6147 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6148 if (ret < 0)
6149 goto out;
6150 ret = send_cmd(sctx);
6151 if (ret < 0)
6152 goto out;
6153 }
Alexander Block31db9f72012-07-25 23:19:24 +02006154
6155out:
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006156 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6157 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6158 struct rb_node *n;
6159 struct pending_dir_move *pm;
6160
6161 n = rb_first(&sctx->pending_dir_moves);
6162 pm = rb_entry(n, struct pending_dir_move, node);
6163 while (!list_empty(&pm->list)) {
6164 struct pending_dir_move *pm2;
6165
6166 pm2 = list_first_entry(&pm->list,
6167 struct pending_dir_move, list);
6168 free_pending_move(sctx, pm2);
6169 }
6170 free_pending_move(sctx, pm);
6171 }
6172
6173 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6174 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6175 struct rb_node *n;
6176 struct waiting_dir_move *dm;
6177
6178 n = rb_first(&sctx->waiting_dir_moves);
6179 dm = rb_entry(n, struct waiting_dir_move, node);
6180 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6181 kfree(dm);
6182 }
6183
Filipe Manana9dc44212014-02-19 14:31:44 +00006184 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6185 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6186 struct rb_node *n;
6187 struct orphan_dir_info *odi;
6188
6189 n = rb_first(&sctx->orphan_dirs);
6190 odi = rb_entry(n, struct orphan_dir_info, node);
6191 free_orphan_dir_info(sctx, odi);
6192 }
6193
Wang Shilong896c14f2014-01-07 17:25:18 +08006194 if (sort_clone_roots) {
6195 for (i = 0; i < sctx->clone_roots_cnt; i++)
6196 btrfs_root_dec_send_in_progress(
6197 sctx->clone_roots[i].root);
6198 } else {
6199 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6200 btrfs_root_dec_send_in_progress(
6201 sctx->clone_roots[i].root);
6202
6203 btrfs_root_dec_send_in_progress(send_root);
6204 }
David Sterba66ef7d62013-12-17 15:07:20 +01006205 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6206 btrfs_root_dec_send_in_progress(sctx->parent_root);
David Sterba2c686532013-12-16 17:34:17 +01006207
Alexander Block31db9f72012-07-25 23:19:24 +02006208 kfree(arg);
6209 vfree(clone_sources_tmp);
6210
6211 if (sctx) {
6212 if (sctx->send_filp)
6213 fput(sctx->send_filp);
6214
6215 vfree(sctx->clone_roots);
6216 vfree(sctx->send_buf);
6217 vfree(sctx->read_buf);
6218
6219 name_cache_free(sctx);
6220
6221 kfree(sctx);
6222 }
6223
6224 return ret;
6225}