blob: 3d6a4dde8fe4625defea8fec0f05a4e4a1031344 [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"
Anand Jainebb87652016-03-10 17:26:59 +080037#include "compression.h"
Alexander Block31db9f72012-07-25 23:19:24 +020038
39static int g_verbose = 0;
40
41#define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
42
43/*
44 * A fs_path is a helper to dynamically build path names with unknown size.
45 * It reallocates the internal buffer on demand.
46 * It allows fast adding of path elements on the right side (normal path) and
47 * fast adding to the left side (reversed path). A reversed path can also be
48 * unreversed if needed.
49 */
50struct fs_path {
51 union {
52 struct {
53 char *start;
54 char *end;
Alexander Block31db9f72012-07-25 23:19:24 +020055
56 char *buf;
David Sterba1f5a7ff2014-02-03 19:23:47 +010057 unsigned short buf_len:15;
58 unsigned short reversed:1;
Alexander Block31db9f72012-07-25 23:19:24 +020059 char inline_buf[];
60 };
David Sterbaace01052014-02-05 16:17:34 +010061 /*
62 * Average path length does not exceed 200 bytes, we'll have
63 * better packing in the slab and higher chance to satisfy
64 * a allocation later during send.
65 */
66 char pad[256];
Alexander Block31db9f72012-07-25 23:19:24 +020067 };
68};
69#define FS_PATH_INLINE_SIZE \
70 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
71
72
73/* reused for each extent */
74struct clone_root {
75 struct btrfs_root *root;
76 u64 ino;
77 u64 offset;
78
79 u64 found_refs;
80};
81
82#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
83#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
84
85struct send_ctx {
86 struct file *send_filp;
87 loff_t send_off;
88 char *send_buf;
89 u32 send_size;
90 u32 send_max_size;
91 u64 total_send_size;
92 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
Mark Fashehcb95e7b2013-02-04 20:54:57 +000093 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
Alexander Block31db9f72012-07-25 23:19:24 +020094
Alexander Block31db9f72012-07-25 23:19:24 +020095 struct btrfs_root *send_root;
96 struct btrfs_root *parent_root;
97 struct clone_root *clone_roots;
98 int clone_roots_cnt;
99
100 /* current state of the compare_tree call */
101 struct btrfs_path *left_path;
102 struct btrfs_path *right_path;
103 struct btrfs_key *cmp_key;
104
105 /*
106 * infos of the currently processed inode. In case of deleted inodes,
107 * these are the values from the deleted inode.
108 */
109 u64 cur_ino;
110 u64 cur_inode_gen;
111 int cur_inode_new;
112 int cur_inode_new_gen;
113 int cur_inode_deleted;
Alexander Block31db9f72012-07-25 23:19:24 +0200114 u64 cur_inode_size;
115 u64 cur_inode_mode;
Liu Bo644d1942014-02-27 17:29:01 +0800116 u64 cur_inode_rdev;
Josef Bacik16e75492013-10-22 12:18:51 -0400117 u64 cur_inode_last_extent;
Alexander Block31db9f72012-07-25 23:19:24 +0200118
119 u64 send_progress;
120
121 struct list_head new_refs;
122 struct list_head deleted_refs;
123
124 struct radix_tree_root name_cache;
125 struct list_head name_cache_list;
126 int name_cache_size;
127
Liu Bo2131bcd2014-03-05 10:07:35 +0800128 struct file_ra_state ra;
129
Alexander Block31db9f72012-07-25 23:19:24 +0200130 char *read_buf;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +0000131
132 /*
133 * We process inodes by their increasing order, so if before an
134 * incremental send we reverse the parent/child relationship of
135 * directories such that a directory with a lower inode number was
136 * the parent of a directory with a higher inode number, and the one
137 * becoming the new parent got renamed too, we can't rename/move the
138 * directory with lower inode number when we finish processing it - we
139 * must process the directory with higher inode number first, then
140 * rename/move it and then rename/move the directory with lower inode
141 * number. Example follows.
142 *
143 * Tree state when the first send was performed:
144 *
145 * .
146 * |-- a (ino 257)
147 * |-- b (ino 258)
148 * |
149 * |
150 * |-- c (ino 259)
151 * | |-- d (ino 260)
152 * |
153 * |-- c2 (ino 261)
154 *
155 * Tree state when the second (incremental) send is performed:
156 *
157 * .
158 * |-- a (ino 257)
159 * |-- b (ino 258)
160 * |-- c2 (ino 261)
161 * |-- d2 (ino 260)
162 * |-- cc (ino 259)
163 *
164 * The sequence of steps that lead to the second state was:
165 *
166 * mv /a/b/c/d /a/b/c2/d2
167 * mv /a/b/c /a/b/c2/d2/cc
168 *
169 * "c" has lower inode number, but we can't move it (2nd mv operation)
170 * before we move "d", which has higher inode number.
171 *
172 * So we just memorize which move/rename operations must be performed
173 * later when their respective parent is processed and moved/renamed.
174 */
175
176 /* Indexed by parent directory inode number. */
177 struct rb_root pending_dir_moves;
178
179 /*
180 * Reverse index, indexed by the inode number of a directory that
181 * is waiting for the move/rename of its immediate parent before its
182 * own move/rename can be performed.
183 */
184 struct rb_root waiting_dir_moves;
Filipe Manana9dc44212014-02-19 14:31:44 +0000185
186 /*
187 * A directory that is going to be rm'ed might have a child directory
188 * which is in the pending directory moves index above. In this case,
189 * the directory can only be removed after the move/rename of its child
190 * is performed. Example:
191 *
192 * Parent snapshot:
193 *
194 * . (ino 256)
195 * |-- a/ (ino 257)
196 * |-- b/ (ino 258)
197 * |-- c/ (ino 259)
198 * | |-- x/ (ino 260)
199 * |
200 * |-- y/ (ino 261)
201 *
202 * Send snapshot:
203 *
204 * . (ino 256)
205 * |-- a/ (ino 257)
206 * |-- b/ (ino 258)
207 * |-- YY/ (ino 261)
208 * |-- x/ (ino 260)
209 *
210 * Sequence of steps that lead to the send snapshot:
211 * rm -f /a/b/c/foo.txt
212 * mv /a/b/y /a/b/YY
213 * mv /a/b/c/x /a/b/YY
214 * rmdir /a/b/c
215 *
216 * When the child is processed, its move/rename is delayed until its
217 * parent is processed (as explained above), but all other operations
218 * like update utimes, chown, chgrp, etc, are performed and the paths
219 * that it uses for those operations must use the orphanized name of
220 * its parent (the directory we're going to rm later), so we need to
221 * memorize that name.
222 *
223 * Indexed by the inode number of the directory to be deleted.
224 */
225 struct rb_root orphan_dirs;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +0000226};
227
228struct pending_dir_move {
229 struct rb_node node;
230 struct list_head list;
231 u64 parent_ino;
232 u64 ino;
233 u64 gen;
234 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
David Sterbae780b0d2016-01-18 18:42:13 +0100307 p = kmalloc(sizeof(*p), GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +0200308 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 Sterbae780b0d2016-01-18 18:42:13 +0100366 tmp_buf = kmalloc(len, GFP_KERNEL);
Filipe Manana01a9a8a2014-05-21 17:38:13 +0100367 if (tmp_buf)
368 memcpy(tmp_buf, p->buf, old_buf_len);
369 } else {
David Sterbae780b0d2016-01-18 18:42:13 +0100370 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
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;
David Sterbae780b0d2016-01-18 18:42:13 +0100998 buf = kmalloc(buf_len, GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +0200999 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,
David Sterbae780b0d2016-01-18 18:42:13 +01001045 GFP_KERNEL | __GFP_NOWARN);
Filipe Manana4395e0c2014-08-20 10:45:45 +01001046
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
David Sterbae780b0d2016-01-18 18:42:13 +01001306 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
Alexander Block35075bb2012-07-28 12:44:34 +02001307 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
Nicholas D Steeves01327612016-05-19 21:18:45 -04001833 * not deleted and then re-created, if it was then we have no overwrite
Josef Bacikebdad912013-08-06 16:47:48 -04001834 * 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 */
Robbie Ko801bec32015-06-23 18:39:46 +08001863 if (other_inode > sctx->send_progress ||
1864 is_waiting_for_move(sctx, other_inode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02001865 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02001866 who_gen, NULL, NULL, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02001867 if (ret < 0)
1868 goto out;
1869
1870 ret = 1;
1871 *who_ino = other_inode;
1872 } else {
1873 ret = 0;
1874 }
1875
1876out:
1877 return ret;
1878}
1879
Alexander Block766702e2012-07-28 14:11:31 +02001880/*
1881 * Checks if the ref was overwritten by an already processed inode. This is
1882 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1883 * thus the orphan name needs be used.
1884 * process_recorded_refs also uses it to avoid unlinking of refs that were
1885 * overwritten.
1886 */
Alexander Block31db9f72012-07-25 23:19:24 +02001887static int did_overwrite_ref(struct send_ctx *sctx,
1888 u64 dir, u64 dir_gen,
1889 u64 ino, u64 ino_gen,
1890 const char *name, int name_len)
1891{
1892 int ret = 0;
1893 u64 gen;
1894 u64 ow_inode;
1895 u8 other_type;
1896
1897 if (!sctx->parent_root)
1898 goto out;
1899
1900 ret = is_inode_existent(sctx, dir, dir_gen);
1901 if (ret <= 0)
1902 goto out;
1903
1904 /* check if the ref was overwritten by another ref */
1905 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1906 &ow_inode, &other_type);
1907 if (ret < 0 && ret != -ENOENT)
1908 goto out;
1909 if (ret) {
1910 /* was never and will never be overwritten */
1911 ret = 0;
1912 goto out;
1913 }
1914
1915 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02001916 NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02001917 if (ret < 0)
1918 goto out;
1919
1920 if (ow_inode == ino && gen == ino_gen) {
1921 ret = 0;
1922 goto out;
1923 }
1924
Filipe Manana8b191a62015-04-09 14:09:14 +01001925 /*
1926 * We know that it is or will be overwritten. Check this now.
1927 * The current inode being processed might have been the one that caused
Filipe Mananab786f162015-09-26 15:30:23 +01001928 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1929 * the current inode being processed.
Filipe Manana8b191a62015-04-09 14:09:14 +01001930 */
Filipe Mananab786f162015-09-26 15:30:23 +01001931 if ((ow_inode < sctx->send_progress) ||
1932 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1933 gen == sctx->cur_inode_gen))
Alexander Block31db9f72012-07-25 23:19:24 +02001934 ret = 1;
1935 else
1936 ret = 0;
1937
1938out:
1939 return ret;
1940}
1941
Alexander Block766702e2012-07-28 14:11:31 +02001942/*
1943 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1944 * that got overwritten. This is used by process_recorded_refs to determine
1945 * if it has to use the path as returned by get_cur_path or the orphan name.
1946 */
Alexander Block31db9f72012-07-25 23:19:24 +02001947static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1948{
1949 int ret = 0;
1950 struct fs_path *name = NULL;
1951 u64 dir;
1952 u64 dir_gen;
1953
1954 if (!sctx->parent_root)
1955 goto out;
1956
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001957 name = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02001958 if (!name)
1959 return -ENOMEM;
1960
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001961 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02001962 if (ret < 0)
1963 goto out;
1964
1965 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1966 name->start, fs_path_len(name));
Alexander Block31db9f72012-07-25 23:19:24 +02001967
1968out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00001969 fs_path_free(name);
Alexander Block31db9f72012-07-25 23:19:24 +02001970 return ret;
1971}
1972
Alexander Block766702e2012-07-28 14:11:31 +02001973/*
1974 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1975 * so we need to do some special handling in case we have clashes. This function
1976 * takes care of this with the help of name_cache_entry::radix_list.
Alexander Block5dc67d02012-08-01 12:07:43 +02001977 * In case of error, nce is kfreed.
Alexander Block766702e2012-07-28 14:11:31 +02001978 */
Alexander Block31db9f72012-07-25 23:19:24 +02001979static int name_cache_insert(struct send_ctx *sctx,
1980 struct name_cache_entry *nce)
1981{
1982 int ret = 0;
Alexander Block7e0926f2012-07-28 14:20:58 +02001983 struct list_head *nce_head;
Alexander Block31db9f72012-07-25 23:19:24 +02001984
Alexander Block7e0926f2012-07-28 14:20:58 +02001985 nce_head = radix_tree_lookup(&sctx->name_cache,
1986 (unsigned long)nce->ino);
1987 if (!nce_head) {
David Sterbae780b0d2016-01-18 18:42:13 +01001988 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
Tsutomu Itohcfa7a9c2012-12-17 06:38:51 +00001989 if (!nce_head) {
1990 kfree(nce);
Alexander Block31db9f72012-07-25 23:19:24 +02001991 return -ENOMEM;
Tsutomu Itohcfa7a9c2012-12-17 06:38:51 +00001992 }
Alexander Block7e0926f2012-07-28 14:20:58 +02001993 INIT_LIST_HEAD(nce_head);
Alexander Block31db9f72012-07-25 23:19:24 +02001994
Alexander Block7e0926f2012-07-28 14:20:58 +02001995 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
Alexander Block5dc67d02012-08-01 12:07:43 +02001996 if (ret < 0) {
1997 kfree(nce_head);
1998 kfree(nce);
Alexander Block31db9f72012-07-25 23:19:24 +02001999 return ret;
Alexander Block5dc67d02012-08-01 12:07:43 +02002000 }
Alexander Block31db9f72012-07-25 23:19:24 +02002001 }
Alexander Block7e0926f2012-07-28 14:20:58 +02002002 list_add_tail(&nce->radix_list, nce_head);
Alexander Block31db9f72012-07-25 23:19:24 +02002003 list_add_tail(&nce->list, &sctx->name_cache_list);
2004 sctx->name_cache_size++;
2005
2006 return ret;
2007}
2008
2009static void name_cache_delete(struct send_ctx *sctx,
2010 struct name_cache_entry *nce)
2011{
Alexander Block7e0926f2012-07-28 14:20:58 +02002012 struct list_head *nce_head;
Alexander Block31db9f72012-07-25 23:19:24 +02002013
Alexander Block7e0926f2012-07-28 14:20:58 +02002014 nce_head = radix_tree_lookup(&sctx->name_cache,
2015 (unsigned long)nce->ino);
David Sterba57fb8912014-02-03 19:24:40 +01002016 if (!nce_head) {
2017 btrfs_err(sctx->send_root->fs_info,
2018 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2019 nce->ino, sctx->name_cache_size);
2020 }
Alexander Block31db9f72012-07-25 23:19:24 +02002021
Alexander Block7e0926f2012-07-28 14:20:58 +02002022 list_del(&nce->radix_list);
Alexander Block31db9f72012-07-25 23:19:24 +02002023 list_del(&nce->list);
Alexander Block31db9f72012-07-25 23:19:24 +02002024 sctx->name_cache_size--;
Alexander Block7e0926f2012-07-28 14:20:58 +02002025
David Sterba57fb8912014-02-03 19:24:40 +01002026 /*
2027 * We may not get to the final release of nce_head if the lookup fails
2028 */
2029 if (nce_head && list_empty(nce_head)) {
Alexander Block7e0926f2012-07-28 14:20:58 +02002030 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2031 kfree(nce_head);
2032 }
Alexander Block31db9f72012-07-25 23:19:24 +02002033}
2034
2035static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2036 u64 ino, u64 gen)
2037{
Alexander Block7e0926f2012-07-28 14:20:58 +02002038 struct list_head *nce_head;
2039 struct name_cache_entry *cur;
Alexander Block31db9f72012-07-25 23:19:24 +02002040
Alexander Block7e0926f2012-07-28 14:20:58 +02002041 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2042 if (!nce_head)
Alexander Block31db9f72012-07-25 23:19:24 +02002043 return NULL;
2044
Alexander Block7e0926f2012-07-28 14:20:58 +02002045 list_for_each_entry(cur, nce_head, radix_list) {
2046 if (cur->ino == ino && cur->gen == gen)
2047 return cur;
2048 }
Alexander Block31db9f72012-07-25 23:19:24 +02002049 return NULL;
2050}
2051
Alexander Block766702e2012-07-28 14:11:31 +02002052/*
2053 * Removes the entry from the list and adds it back to the end. This marks the
2054 * entry as recently used so that name_cache_clean_unused does not remove it.
2055 */
Alexander Block31db9f72012-07-25 23:19:24 +02002056static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2057{
2058 list_del(&nce->list);
2059 list_add_tail(&nce->list, &sctx->name_cache_list);
2060}
2061
Alexander Block766702e2012-07-28 14:11:31 +02002062/*
2063 * Remove some entries from the beginning of name_cache_list.
2064 */
Alexander Block31db9f72012-07-25 23:19:24 +02002065static void name_cache_clean_unused(struct send_ctx *sctx)
2066{
2067 struct name_cache_entry *nce;
2068
2069 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2070 return;
2071
2072 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2073 nce = list_entry(sctx->name_cache_list.next,
2074 struct name_cache_entry, list);
2075 name_cache_delete(sctx, nce);
2076 kfree(nce);
2077 }
2078}
2079
2080static void name_cache_free(struct send_ctx *sctx)
2081{
2082 struct name_cache_entry *nce;
Alexander Block31db9f72012-07-25 23:19:24 +02002083
Alexander Blocke938c8a2012-07-28 16:33:49 +02002084 while (!list_empty(&sctx->name_cache_list)) {
2085 nce = list_entry(sctx->name_cache_list.next,
2086 struct name_cache_entry, list);
Alexander Block31db9f72012-07-25 23:19:24 +02002087 name_cache_delete(sctx, nce);
Alexander Block17589bd2012-07-28 14:13:35 +02002088 kfree(nce);
Alexander Block31db9f72012-07-25 23:19:24 +02002089 }
2090}
2091
Alexander Block766702e2012-07-28 14:11:31 +02002092/*
2093 * Used by get_cur_path for each ref up to the root.
2094 * Returns 0 if it succeeded.
2095 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2096 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2097 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2098 * Returns <0 in case of error.
2099 */
Alexander Block31db9f72012-07-25 23:19:24 +02002100static int __get_cur_name_and_parent(struct send_ctx *sctx,
2101 u64 ino, u64 gen,
2102 u64 *parent_ino,
2103 u64 *parent_gen,
2104 struct fs_path *dest)
2105{
2106 int ret;
2107 int nce_ret;
Alexander Block31db9f72012-07-25 23:19:24 +02002108 struct name_cache_entry *nce = NULL;
2109
Alexander Block766702e2012-07-28 14:11:31 +02002110 /*
2111 * First check if we already did a call to this function with the same
2112 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2113 * return the cached result.
2114 */
Alexander Block31db9f72012-07-25 23:19:24 +02002115 nce = name_cache_search(sctx, ino, gen);
2116 if (nce) {
2117 if (ino < sctx->send_progress && nce->need_later_update) {
2118 name_cache_delete(sctx, nce);
2119 kfree(nce);
2120 nce = NULL;
2121 } else {
2122 name_cache_used(sctx, nce);
2123 *parent_ino = nce->parent_ino;
2124 *parent_gen = nce->parent_gen;
2125 ret = fs_path_add(dest, nce->name, nce->name_len);
2126 if (ret < 0)
2127 goto out;
2128 ret = nce->ret;
2129 goto out;
2130 }
2131 }
2132
Alexander Block766702e2012-07-28 14:11:31 +02002133 /*
2134 * If the inode is not existent yet, add the orphan name and return 1.
2135 * This should only happen for the parent dir that we determine in
2136 * __record_new_ref
2137 */
Alexander Block31db9f72012-07-25 23:19:24 +02002138 ret = is_inode_existent(sctx, ino, gen);
2139 if (ret < 0)
2140 goto out;
2141
2142 if (!ret) {
2143 ret = gen_unique_name(sctx, ino, gen, dest);
2144 if (ret < 0)
2145 goto out;
2146 ret = 1;
2147 goto out_cache;
2148 }
2149
Alexander Block766702e2012-07-28 14:11:31 +02002150 /*
2151 * Depending on whether the inode was already processed or not, use
2152 * send_root or parent_root for ref lookup.
2153 */
Filipe Mananabf0d1f42014-02-21 00:01:32 +00002154 if (ino < sctx->send_progress)
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002155 ret = get_first_ref(sctx->send_root, ino,
2156 parent_ino, parent_gen, dest);
Alexander Block31db9f72012-07-25 23:19:24 +02002157 else
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002158 ret = get_first_ref(sctx->parent_root, ino,
2159 parent_ino, parent_gen, dest);
Alexander Block31db9f72012-07-25 23:19:24 +02002160 if (ret < 0)
2161 goto out;
2162
Alexander Block766702e2012-07-28 14:11:31 +02002163 /*
2164 * Check if the ref was overwritten by an inode's ref that was processed
2165 * earlier. If yes, treat as orphan and return 1.
2166 */
Alexander Block31db9f72012-07-25 23:19:24 +02002167 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2168 dest->start, dest->end - dest->start);
2169 if (ret < 0)
2170 goto out;
2171 if (ret) {
2172 fs_path_reset(dest);
2173 ret = gen_unique_name(sctx, ino, gen, dest);
2174 if (ret < 0)
2175 goto out;
2176 ret = 1;
2177 }
2178
2179out_cache:
Alexander Block766702e2012-07-28 14:11:31 +02002180 /*
2181 * Store the result of the lookup in the name cache.
2182 */
David Sterbae780b0d2016-01-18 18:42:13 +01002183 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02002184 if (!nce) {
2185 ret = -ENOMEM;
2186 goto out;
2187 }
2188
2189 nce->ino = ino;
2190 nce->gen = gen;
2191 nce->parent_ino = *parent_ino;
2192 nce->parent_gen = *parent_gen;
2193 nce->name_len = fs_path_len(dest);
2194 nce->ret = ret;
2195 strcpy(nce->name, dest->start);
Alexander Block31db9f72012-07-25 23:19:24 +02002196
2197 if (ino < sctx->send_progress)
2198 nce->need_later_update = 0;
2199 else
2200 nce->need_later_update = 1;
2201
2202 nce_ret = name_cache_insert(sctx, nce);
2203 if (nce_ret < 0)
2204 ret = nce_ret;
2205 name_cache_clean_unused(sctx);
2206
2207out:
Alexander Block31db9f72012-07-25 23:19:24 +02002208 return ret;
2209}
2210
2211/*
2212 * Magic happens here. This function returns the first ref to an inode as it
2213 * would look like while receiving the stream at this point in time.
2214 * We walk the path up to the root. For every inode in between, we check if it
2215 * was already processed/sent. If yes, we continue with the parent as found
2216 * in send_root. If not, we continue with the parent as found in parent_root.
2217 * If we encounter an inode that was deleted at this point in time, we use the
2218 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2219 * that were not created yet and overwritten inodes/refs.
2220 *
2221 * When do we have have orphan inodes:
2222 * 1. When an inode is freshly created and thus no valid refs are available yet
2223 * 2. When a directory lost all it's refs (deleted) but still has dir items
2224 * inside which were not processed yet (pending for move/delete). If anyone
2225 * tried to get the path to the dir items, it would get a path inside that
2226 * orphan directory.
2227 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2228 * of an unprocessed inode. If in that case the first ref would be
2229 * overwritten, the overwritten inode gets "orphanized". Later when we
2230 * process this overwritten inode, it is restored at a new place by moving
2231 * the orphan inode.
2232 *
2233 * sctx->send_progress tells this function at which point in time receiving
2234 * would be.
2235 */
2236static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2237 struct fs_path *dest)
2238{
2239 int ret = 0;
2240 struct fs_path *name = NULL;
2241 u64 parent_inode = 0;
2242 u64 parent_gen = 0;
2243 int stop = 0;
2244
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002245 name = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002246 if (!name) {
2247 ret = -ENOMEM;
2248 goto out;
2249 }
2250
2251 dest->reversed = 1;
2252 fs_path_reset(dest);
2253
2254 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
Filipe Manana8b191a62015-04-09 14:09:14 +01002255 struct waiting_dir_move *wdm;
2256
Alexander Block31db9f72012-07-25 23:19:24 +02002257 fs_path_reset(name);
2258
Filipe Manana9dc44212014-02-19 14:31:44 +00002259 if (is_waiting_for_rm(sctx, ino)) {
2260 ret = gen_unique_name(sctx, ino, gen, name);
2261 if (ret < 0)
2262 goto out;
2263 ret = fs_path_add_path(dest, name);
2264 break;
2265 }
2266
Filipe Manana8b191a62015-04-09 14:09:14 +01002267 wdm = get_waiting_dir_move(sctx, ino);
2268 if (wdm && wdm->orphanized) {
2269 ret = gen_unique_name(sctx, ino, gen, name);
2270 stop = 1;
2271 } else if (wdm) {
Filipe Mananabf0d1f42014-02-21 00:01:32 +00002272 ret = get_first_ref(sctx->parent_root, ino,
2273 &parent_inode, &parent_gen, name);
2274 } else {
2275 ret = __get_cur_name_and_parent(sctx, ino, gen,
2276 &parent_inode,
2277 &parent_gen, name);
2278 if (ret)
2279 stop = 1;
2280 }
2281
Alexander Block31db9f72012-07-25 23:19:24 +02002282 if (ret < 0)
2283 goto out;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002284
Alexander Block31db9f72012-07-25 23:19:24 +02002285 ret = fs_path_add_path(dest, name);
2286 if (ret < 0)
2287 goto out;
2288
2289 ino = parent_inode;
2290 gen = parent_gen;
2291 }
2292
2293out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002294 fs_path_free(name);
Alexander Block31db9f72012-07-25 23:19:24 +02002295 if (!ret)
2296 fs_path_unreverse(dest);
2297 return ret;
2298}
2299
2300/*
Alexander Block31db9f72012-07-25 23:19:24 +02002301 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2302 */
2303static int send_subvol_begin(struct send_ctx *sctx)
2304{
2305 int ret;
2306 struct btrfs_root *send_root = sctx->send_root;
2307 struct btrfs_root *parent_root = sctx->parent_root;
2308 struct btrfs_path *path;
2309 struct btrfs_key key;
2310 struct btrfs_root_ref *ref;
2311 struct extent_buffer *leaf;
2312 char *name = NULL;
2313 int namelen;
2314
Wang Shilongffcfaf82014-01-15 00:26:43 +08002315 path = btrfs_alloc_path();
Alexander Block31db9f72012-07-25 23:19:24 +02002316 if (!path)
2317 return -ENOMEM;
2318
David Sterbae780b0d2016-01-18 18:42:13 +01002319 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02002320 if (!name) {
2321 btrfs_free_path(path);
2322 return -ENOMEM;
2323 }
2324
2325 key.objectid = send_root->objectid;
2326 key.type = BTRFS_ROOT_BACKREF_KEY;
2327 key.offset = 0;
2328
2329 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2330 &key, path, 1, 0);
2331 if (ret < 0)
2332 goto out;
2333 if (ret) {
2334 ret = -ENOENT;
2335 goto out;
2336 }
2337
2338 leaf = path->nodes[0];
2339 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2340 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2341 key.objectid != send_root->objectid) {
2342 ret = -ENOENT;
2343 goto out;
2344 }
2345 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2346 namelen = btrfs_root_ref_name_len(leaf, ref);
2347 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2348 btrfs_release_path(path);
2349
Alexander Block31db9f72012-07-25 23:19:24 +02002350 if (parent_root) {
2351 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2352 if (ret < 0)
2353 goto out;
2354 } else {
2355 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2356 if (ret < 0)
2357 goto out;
2358 }
2359
2360 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
Robin Ruedeb96b1db2015-09-30 21:23:33 +02002361
2362 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2363 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2364 sctx->send_root->root_item.received_uuid);
2365 else
2366 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2367 sctx->send_root->root_item.uuid);
2368
Alexander Block31db9f72012-07-25 23:19:24 +02002369 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
Filipe David Borba Manana5a0f4e22013-12-03 15:55:48 +00002370 le64_to_cpu(sctx->send_root->root_item.ctransid));
Alexander Block31db9f72012-07-25 23:19:24 +02002371 if (parent_root) {
Josef Bacik37b8d272015-06-04 17:17:25 -04002372 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2373 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2374 parent_root->root_item.received_uuid);
2375 else
2376 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2377 parent_root->root_item.uuid);
Alexander Block31db9f72012-07-25 23:19:24 +02002378 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
Filipe David Borba Manana5a0f4e22013-12-03 15:55:48 +00002379 le64_to_cpu(sctx->parent_root->root_item.ctransid));
Alexander Block31db9f72012-07-25 23:19:24 +02002380 }
2381
2382 ret = send_cmd(sctx);
2383
2384tlv_put_failure:
2385out:
2386 btrfs_free_path(path);
2387 kfree(name);
2388 return ret;
2389}
2390
2391static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2392{
2393 int ret = 0;
2394 struct fs_path *p;
2395
2396verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2397
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002398 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002399 if (!p)
2400 return -ENOMEM;
2401
2402 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2403 if (ret < 0)
2404 goto out;
2405
2406 ret = get_cur_path(sctx, ino, gen, p);
2407 if (ret < 0)
2408 goto out;
2409 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2410 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2411
2412 ret = send_cmd(sctx);
2413
2414tlv_put_failure:
2415out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002416 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002417 return ret;
2418}
2419
2420static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2421{
2422 int ret = 0;
2423 struct fs_path *p;
2424
2425verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2426
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002427 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002428 if (!p)
2429 return -ENOMEM;
2430
2431 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2432 if (ret < 0)
2433 goto out;
2434
2435 ret = get_cur_path(sctx, ino, gen, p);
2436 if (ret < 0)
2437 goto out;
2438 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2439 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2440
2441 ret = send_cmd(sctx);
2442
2443tlv_put_failure:
2444out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002445 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002446 return ret;
2447}
2448
2449static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2450{
2451 int ret = 0;
2452 struct fs_path *p;
2453
2454verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2455
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002456 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002457 if (!p)
2458 return -ENOMEM;
2459
2460 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2461 if (ret < 0)
2462 goto out;
2463
2464 ret = get_cur_path(sctx, ino, gen, p);
2465 if (ret < 0)
2466 goto out;
2467 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2468 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2469 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2470
2471 ret = send_cmd(sctx);
2472
2473tlv_put_failure:
2474out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002475 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002476 return ret;
2477}
2478
2479static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2480{
2481 int ret = 0;
2482 struct fs_path *p = NULL;
2483 struct btrfs_inode_item *ii;
2484 struct btrfs_path *path = NULL;
2485 struct extent_buffer *eb;
2486 struct btrfs_key key;
2487 int slot;
2488
2489verbose_printk("btrfs: send_utimes %llu\n", ino);
2490
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002491 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002492 if (!p)
2493 return -ENOMEM;
2494
2495 path = alloc_path_for_send();
2496 if (!path) {
2497 ret = -ENOMEM;
2498 goto out;
2499 }
2500
2501 key.objectid = ino;
2502 key.type = BTRFS_INODE_ITEM_KEY;
2503 key.offset = 0;
2504 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2505 if (ret < 0)
2506 goto out;
2507
2508 eb = path->nodes[0];
2509 slot = path->slots[0];
2510 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2511
2512 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2513 if (ret < 0)
2514 goto out;
2515
2516 ret = get_cur_path(sctx, ino, gen, p);
2517 if (ret < 0)
2518 goto out;
2519 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
David Sterbaa937b972014-12-12 17:39:12 +01002520 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2521 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2522 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
Alexander Block766702e2012-07-28 14:11:31 +02002523 /* TODO Add otime support when the otime patches get into upstream */
Alexander Block31db9f72012-07-25 23:19:24 +02002524
2525 ret = send_cmd(sctx);
2526
2527tlv_put_failure:
2528out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002529 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002530 btrfs_free_path(path);
2531 return ret;
2532}
2533
2534/*
2535 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2536 * a valid path yet because we did not process the refs yet. So, the inode
2537 * is created as orphan.
2538 */
Alexander Block1f4692d2012-07-28 10:42:24 +02002539static int send_create_inode(struct send_ctx *sctx, u64 ino)
Alexander Block31db9f72012-07-25 23:19:24 +02002540{
2541 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02002542 struct fs_path *p;
Alexander Block31db9f72012-07-25 23:19:24 +02002543 int cmd;
Alexander Block1f4692d2012-07-28 10:42:24 +02002544 u64 gen;
Alexander Block31db9f72012-07-25 23:19:24 +02002545 u64 mode;
Alexander Block1f4692d2012-07-28 10:42:24 +02002546 u64 rdev;
Alexander Block31db9f72012-07-25 23:19:24 +02002547
Alexander Block1f4692d2012-07-28 10:42:24 +02002548verbose_printk("btrfs: send_create_inode %llu\n", ino);
Alexander Block31db9f72012-07-25 23:19:24 +02002549
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002550 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002551 if (!p)
2552 return -ENOMEM;
2553
Liu Bo644d1942014-02-27 17:29:01 +08002554 if (ino != sctx->cur_ino) {
2555 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2556 NULL, NULL, &rdev);
2557 if (ret < 0)
2558 goto out;
2559 } else {
2560 gen = sctx->cur_inode_gen;
2561 mode = sctx->cur_inode_mode;
2562 rdev = sctx->cur_inode_rdev;
2563 }
Alexander Block31db9f72012-07-25 23:19:24 +02002564
Alexander Blocke938c8a2012-07-28 16:33:49 +02002565 if (S_ISREG(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002566 cmd = BTRFS_SEND_C_MKFILE;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002567 } else if (S_ISDIR(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002568 cmd = BTRFS_SEND_C_MKDIR;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002569 } else if (S_ISLNK(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002570 cmd = BTRFS_SEND_C_SYMLINK;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002571 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002572 cmd = BTRFS_SEND_C_MKNOD;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002573 } else if (S_ISFIFO(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002574 cmd = BTRFS_SEND_C_MKFIFO;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002575 } else if (S_ISSOCK(mode)) {
Alexander Block31db9f72012-07-25 23:19:24 +02002576 cmd = BTRFS_SEND_C_MKSOCK;
Alexander Blocke938c8a2012-07-28 16:33:49 +02002577 } else {
David Sterbaf14d1042015-10-08 11:37:06 +02002578 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
Alexander Block31db9f72012-07-25 23:19:24 +02002579 (int)(mode & S_IFMT));
2580 ret = -ENOTSUPP;
2581 goto out;
2582 }
2583
2584 ret = begin_cmd(sctx, cmd);
2585 if (ret < 0)
2586 goto out;
2587
Alexander Block1f4692d2012-07-28 10:42:24 +02002588 ret = gen_unique_name(sctx, ino, gen, p);
Alexander Block31db9f72012-07-25 23:19:24 +02002589 if (ret < 0)
2590 goto out;
2591
2592 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
Alexander Block1f4692d2012-07-28 10:42:24 +02002593 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
Alexander Block31db9f72012-07-25 23:19:24 +02002594
2595 if (S_ISLNK(mode)) {
2596 fs_path_reset(p);
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002597 ret = read_symlink(sctx->send_root, ino, p);
Alexander Block31db9f72012-07-25 23:19:24 +02002598 if (ret < 0)
2599 goto out;
2600 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2601 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2602 S_ISFIFO(mode) || S_ISSOCK(mode)) {
Arne Jansend79e5042012-10-15 18:28:46 +00002603 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2604 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
Alexander Block31db9f72012-07-25 23:19:24 +02002605 }
2606
2607 ret = send_cmd(sctx);
2608 if (ret < 0)
2609 goto out;
2610
2611
2612tlv_put_failure:
2613out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002614 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02002615 return ret;
2616}
2617
Alexander Block1f4692d2012-07-28 10:42:24 +02002618/*
2619 * We need some special handling for inodes that get processed before the parent
2620 * directory got created. See process_recorded_refs for details.
2621 * This function does the check if we already created the dir out of order.
2622 */
2623static int did_create_dir(struct send_ctx *sctx, u64 dir)
2624{
2625 int ret = 0;
2626 struct btrfs_path *path = NULL;
2627 struct btrfs_key key;
2628 struct btrfs_key found_key;
2629 struct btrfs_key di_key;
2630 struct extent_buffer *eb;
2631 struct btrfs_dir_item *di;
2632 int slot;
2633
2634 path = alloc_path_for_send();
2635 if (!path) {
2636 ret = -ENOMEM;
2637 goto out;
2638 }
2639
2640 key.objectid = dir;
2641 key.type = BTRFS_DIR_INDEX_KEY;
2642 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002643 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2644 if (ret < 0)
2645 goto out;
2646
Alexander Block1f4692d2012-07-28 10:42:24 +02002647 while (1) {
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002648 eb = path->nodes[0];
2649 slot = path->slots[0];
2650 if (slot >= btrfs_header_nritems(eb)) {
2651 ret = btrfs_next_leaf(sctx->send_root, path);
2652 if (ret < 0) {
2653 goto out;
2654 } else if (ret > 0) {
2655 ret = 0;
2656 break;
2657 }
2658 continue;
Alexander Block1f4692d2012-07-28 10:42:24 +02002659 }
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002660
2661 btrfs_item_key_to_cpu(eb, &found_key, slot);
2662 if (found_key.objectid != key.objectid ||
Alexander Block1f4692d2012-07-28 10:42:24 +02002663 found_key.type != key.type) {
2664 ret = 0;
2665 goto out;
2666 }
2667
2668 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2669 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2670
Josef Bacika0525412013-08-12 10:56:14 -04002671 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2672 di_key.objectid < sctx->send_progress) {
Alexander Block1f4692d2012-07-28 10:42:24 +02002673 ret = 1;
2674 goto out;
2675 }
2676
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002677 path->slots[0]++;
Alexander Block1f4692d2012-07-28 10:42:24 +02002678 }
2679
2680out:
2681 btrfs_free_path(path);
2682 return ret;
2683}
2684
2685/*
2686 * Only creates the inode if it is:
2687 * 1. Not a directory
2688 * 2. Or a directory which was not created already due to out of order
2689 * directories. See did_create_dir and process_recorded_refs for details.
2690 */
2691static int send_create_inode_if_needed(struct send_ctx *sctx)
2692{
2693 int ret;
2694
2695 if (S_ISDIR(sctx->cur_inode_mode)) {
2696 ret = did_create_dir(sctx, sctx->cur_ino);
2697 if (ret < 0)
2698 goto out;
2699 if (ret) {
2700 ret = 0;
2701 goto out;
2702 }
2703 }
2704
2705 ret = send_create_inode(sctx, sctx->cur_ino);
2706 if (ret < 0)
2707 goto out;
2708
2709out:
2710 return ret;
2711}
2712
Alexander Block31db9f72012-07-25 23:19:24 +02002713struct recorded_ref {
2714 struct list_head list;
2715 char *dir_path;
2716 char *name;
2717 struct fs_path *full_path;
2718 u64 dir;
2719 u64 dir_gen;
2720 int dir_path_len;
2721 int name_len;
2722};
2723
2724/*
2725 * We need to process new refs before deleted refs, but compare_tree gives us
2726 * everything mixed. So we first record all refs and later process them.
2727 * This function is a helper to record one ref.
2728 */
Liu Boa4d96d62014-03-03 21:31:03 +08002729static int __record_ref(struct list_head *head, u64 dir,
Alexander Block31db9f72012-07-25 23:19:24 +02002730 u64 dir_gen, struct fs_path *path)
2731{
2732 struct recorded_ref *ref;
Alexander Block31db9f72012-07-25 23:19:24 +02002733
David Sterbae780b0d2016-01-18 18:42:13 +01002734 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02002735 if (!ref)
2736 return -ENOMEM;
2737
2738 ref->dir = dir;
2739 ref->dir_gen = dir_gen;
2740 ref->full_path = path;
2741
Andy Shevchenkoed848852013-08-21 10:32:13 +03002742 ref->name = (char *)kbasename(ref->full_path->start);
2743 ref->name_len = ref->full_path->end - ref->name;
2744 ref->dir_path = ref->full_path->start;
2745 if (ref->name == ref->full_path->start)
Alexander Block31db9f72012-07-25 23:19:24 +02002746 ref->dir_path_len = 0;
Andy Shevchenkoed848852013-08-21 10:32:13 +03002747 else
Alexander Block31db9f72012-07-25 23:19:24 +02002748 ref->dir_path_len = ref->full_path->end -
2749 ref->full_path->start - 1 - ref->name_len;
Alexander Block31db9f72012-07-25 23:19:24 +02002750
2751 list_add_tail(&ref->list, head);
2752 return 0;
2753}
2754
Josef Bacikba5e8f22013-08-16 16:52:55 -04002755static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2756{
2757 struct recorded_ref *new;
2758
David Sterbae780b0d2016-01-18 18:42:13 +01002759 new = kmalloc(sizeof(*ref), GFP_KERNEL);
Josef Bacikba5e8f22013-08-16 16:52:55 -04002760 if (!new)
2761 return -ENOMEM;
2762
2763 new->dir = ref->dir;
2764 new->dir_gen = ref->dir_gen;
2765 new->full_path = NULL;
2766 INIT_LIST_HEAD(&new->list);
2767 list_add_tail(&new->list, list);
2768 return 0;
2769}
2770
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002771static void __free_recorded_refs(struct list_head *head)
Alexander Block31db9f72012-07-25 23:19:24 +02002772{
2773 struct recorded_ref *cur;
Alexander Block31db9f72012-07-25 23:19:24 +02002774
Alexander Blocke938c8a2012-07-28 16:33:49 +02002775 while (!list_empty(head)) {
2776 cur = list_entry(head->next, struct recorded_ref, list);
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002777 fs_path_free(cur->full_path);
Alexander Blocke938c8a2012-07-28 16:33:49 +02002778 list_del(&cur->list);
Alexander Block31db9f72012-07-25 23:19:24 +02002779 kfree(cur);
2780 }
Alexander Block31db9f72012-07-25 23:19:24 +02002781}
2782
2783static void free_recorded_refs(struct send_ctx *sctx)
2784{
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002785 __free_recorded_refs(&sctx->new_refs);
2786 __free_recorded_refs(&sctx->deleted_refs);
Alexander Block31db9f72012-07-25 23:19:24 +02002787}
2788
2789/*
Alexander Block766702e2012-07-28 14:11:31 +02002790 * Renames/moves a file/dir to its orphan name. Used when the first
Alexander Block31db9f72012-07-25 23:19:24 +02002791 * ref of an unprocessed inode gets overwritten and for all non empty
2792 * directories.
2793 */
2794static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2795 struct fs_path *path)
2796{
2797 int ret;
2798 struct fs_path *orphan;
2799
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002800 orphan = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02002801 if (!orphan)
2802 return -ENOMEM;
2803
2804 ret = gen_unique_name(sctx, ino, gen, orphan);
2805 if (ret < 0)
2806 goto out;
2807
2808 ret = send_rename(sctx, path, orphan);
2809
2810out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00002811 fs_path_free(orphan);
Alexander Block31db9f72012-07-25 23:19:24 +02002812 return ret;
2813}
2814
Filipe Manana9dc44212014-02-19 14:31:44 +00002815static struct orphan_dir_info *
2816add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2817{
2818 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2819 struct rb_node *parent = NULL;
2820 struct orphan_dir_info *entry, *odi;
2821
David Sterbae780b0d2016-01-18 18:42:13 +01002822 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
Filipe Manana9dc44212014-02-19 14:31:44 +00002823 if (!odi)
2824 return ERR_PTR(-ENOMEM);
2825 odi->ino = dir_ino;
2826 odi->gen = 0;
2827
2828 while (*p) {
2829 parent = *p;
2830 entry = rb_entry(parent, struct orphan_dir_info, node);
2831 if (dir_ino < entry->ino) {
2832 p = &(*p)->rb_left;
2833 } else if (dir_ino > entry->ino) {
2834 p = &(*p)->rb_right;
2835 } else {
2836 kfree(odi);
2837 return entry;
2838 }
2839 }
2840
2841 rb_link_node(&odi->node, parent, p);
2842 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2843 return odi;
2844}
2845
2846static struct orphan_dir_info *
2847get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2848{
2849 struct rb_node *n = sctx->orphan_dirs.rb_node;
2850 struct orphan_dir_info *entry;
2851
2852 while (n) {
2853 entry = rb_entry(n, struct orphan_dir_info, node);
2854 if (dir_ino < entry->ino)
2855 n = n->rb_left;
2856 else if (dir_ino > entry->ino)
2857 n = n->rb_right;
2858 else
2859 return entry;
2860 }
2861 return NULL;
2862}
2863
2864static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2865{
2866 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2867
2868 return odi != NULL;
2869}
2870
2871static void free_orphan_dir_info(struct send_ctx *sctx,
2872 struct orphan_dir_info *odi)
2873{
2874 if (!odi)
2875 return;
2876 rb_erase(&odi->node, &sctx->orphan_dirs);
2877 kfree(odi);
2878}
2879
Alexander Block31db9f72012-07-25 23:19:24 +02002880/*
2881 * Returns 1 if a directory can be removed at this point in time.
2882 * We check this by iterating all dir items and checking if the inode behind
2883 * the dir item was already processed.
2884 */
Filipe Manana9dc44212014-02-19 14:31:44 +00002885static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2886 u64 send_progress)
Alexander Block31db9f72012-07-25 23:19:24 +02002887{
2888 int ret = 0;
2889 struct btrfs_root *root = sctx->parent_root;
2890 struct btrfs_path *path;
2891 struct btrfs_key key;
2892 struct btrfs_key found_key;
2893 struct btrfs_key loc;
2894 struct btrfs_dir_item *di;
2895
Alexander Block6d85ed02012-08-01 14:48:59 +02002896 /*
2897 * Don't try to rmdir the top/root subvolume dir.
2898 */
2899 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2900 return 0;
2901
Alexander Block31db9f72012-07-25 23:19:24 +02002902 path = alloc_path_for_send();
2903 if (!path)
2904 return -ENOMEM;
2905
2906 key.objectid = dir;
2907 key.type = BTRFS_DIR_INDEX_KEY;
2908 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002909 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2910 if (ret < 0)
2911 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02002912
2913 while (1) {
Filipe Manana9dc44212014-02-19 14:31:44 +00002914 struct waiting_dir_move *dm;
2915
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002916 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2917 ret = btrfs_next_leaf(root, path);
2918 if (ret < 0)
2919 goto out;
2920 else if (ret > 0)
2921 break;
2922 continue;
Alexander Block31db9f72012-07-25 23:19:24 +02002923 }
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002924 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2925 path->slots[0]);
2926 if (found_key.objectid != key.objectid ||
2927 found_key.type != key.type)
Alexander Block31db9f72012-07-25 23:19:24 +02002928 break;
Alexander Block31db9f72012-07-25 23:19:24 +02002929
2930 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2931 struct btrfs_dir_item);
2932 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2933
Filipe Manana9dc44212014-02-19 14:31:44 +00002934 dm = get_waiting_dir_move(sctx, loc.objectid);
2935 if (dm) {
2936 struct orphan_dir_info *odi;
2937
2938 odi = add_orphan_dir_info(sctx, dir);
2939 if (IS_ERR(odi)) {
2940 ret = PTR_ERR(odi);
2941 goto out;
2942 }
2943 odi->gen = dir_gen;
2944 dm->rmdir_ino = dir;
2945 ret = 0;
2946 goto out;
2947 }
2948
Alexander Block31db9f72012-07-25 23:19:24 +02002949 if (loc.objectid > send_progress) {
2950 ret = 0;
2951 goto out;
2952 }
2953
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002954 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02002955 }
2956
2957 ret = 1;
2958
2959out:
2960 btrfs_free_path(path);
2961 return ret;
2962}
2963
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002964static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2965{
Filipe Manana9dc44212014-02-19 14:31:44 +00002966 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002967
Filipe Manana9dc44212014-02-19 14:31:44 +00002968 return entry != NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002969}
2970
Filipe Manana8b191a62015-04-09 14:09:14 +01002971static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002972{
2973 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2974 struct rb_node *parent = NULL;
2975 struct waiting_dir_move *entry, *dm;
2976
David Sterbae780b0d2016-01-18 18:42:13 +01002977 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002978 if (!dm)
2979 return -ENOMEM;
2980 dm->ino = ino;
Filipe Manana9dc44212014-02-19 14:31:44 +00002981 dm->rmdir_ino = 0;
Filipe Manana8b191a62015-04-09 14:09:14 +01002982 dm->orphanized = orphanized;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002983
2984 while (*p) {
2985 parent = *p;
2986 entry = rb_entry(parent, struct waiting_dir_move, node);
2987 if (ino < entry->ino) {
2988 p = &(*p)->rb_left;
2989 } else if (ino > entry->ino) {
2990 p = &(*p)->rb_right;
2991 } else {
2992 kfree(dm);
2993 return -EEXIST;
2994 }
2995 }
2996
2997 rb_link_node(&dm->node, parent, p);
2998 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2999 return 0;
3000}
3001
Filipe Manana9dc44212014-02-19 14:31:44 +00003002static struct waiting_dir_move *
3003get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003004{
3005 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3006 struct waiting_dir_move *entry;
3007
3008 while (n) {
3009 entry = rb_entry(n, struct waiting_dir_move, node);
Filipe Manana9dc44212014-02-19 14:31:44 +00003010 if (ino < entry->ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003011 n = n->rb_left;
Filipe Manana9dc44212014-02-19 14:31:44 +00003012 else if (ino > entry->ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003013 n = n->rb_right;
Filipe Manana9dc44212014-02-19 14:31:44 +00003014 else
3015 return entry;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003016 }
Filipe Manana9dc44212014-02-19 14:31:44 +00003017 return NULL;
3018}
3019
3020static void free_waiting_dir_move(struct send_ctx *sctx,
3021 struct waiting_dir_move *dm)
3022{
3023 if (!dm)
3024 return;
3025 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3026 kfree(dm);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003027}
3028
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003029static int add_pending_dir_move(struct send_ctx *sctx,
3030 u64 ino,
3031 u64 ino_gen,
Filipe Mananaf9594922014-03-27 20:14:01 +00003032 u64 parent_ino,
3033 struct list_head *new_refs,
Filipe Manana84471e22015-02-28 22:29:22 +00003034 struct list_head *deleted_refs,
3035 const bool is_orphan)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003036{
3037 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3038 struct rb_node *parent = NULL;
Chris Mason73b802f2014-03-21 15:30:44 -07003039 struct pending_dir_move *entry = NULL, *pm;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003040 struct recorded_ref *cur;
3041 int exists = 0;
3042 int ret;
3043
David Sterbae780b0d2016-01-18 18:42:13 +01003044 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003045 if (!pm)
3046 return -ENOMEM;
3047 pm->parent_ino = parent_ino;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003048 pm->ino = ino;
3049 pm->gen = ino_gen;
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
Robbie Ko801bec32015-06-23 18:39:46 +08003115static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3116 u64 ino, u64 gen, u64 *ancestor_ino)
3117{
3118 int ret = 0;
3119 u64 parent_inode = 0;
3120 u64 parent_gen = 0;
3121 u64 start_ino = ino;
3122
3123 *ancestor_ino = 0;
3124 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3125 fs_path_reset(name);
3126
3127 if (is_waiting_for_rm(sctx, ino))
3128 break;
3129 if (is_waiting_for_move(sctx, ino)) {
3130 if (*ancestor_ino == 0)
3131 *ancestor_ino = ino;
3132 ret = get_first_ref(sctx->parent_root, ino,
3133 &parent_inode, &parent_gen, name);
3134 } else {
3135 ret = __get_cur_name_and_parent(sctx, ino, gen,
3136 &parent_inode,
3137 &parent_gen, name);
3138 if (ret > 0) {
3139 ret = 0;
3140 break;
3141 }
3142 }
3143 if (ret < 0)
3144 break;
3145 if (parent_inode == start_ino) {
3146 ret = 1;
3147 if (*ancestor_ino == 0)
3148 *ancestor_ino = ino;
3149 break;
3150 }
3151 ino = parent_inode;
3152 gen = parent_gen;
3153 }
3154 return ret;
3155}
3156
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003157static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3158{
3159 struct fs_path *from_path = NULL;
3160 struct fs_path *to_path = NULL;
Filipe Manana2b863a12014-02-16 13:43:11 +00003161 struct fs_path *name = NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003162 u64 orig_progress = sctx->send_progress;
3163 struct recorded_ref *cur;
Filipe Manana2b863a12014-02-16 13:43:11 +00003164 u64 parent_ino, parent_gen;
Filipe Manana9dc44212014-02-19 14:31:44 +00003165 struct waiting_dir_move *dm = NULL;
3166 u64 rmdir_ino = 0;
Robbie Ko801bec32015-06-23 18:39:46 +08003167 u64 ancestor;
3168 bool is_orphan;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003169 int ret;
3170
Filipe Manana2b863a12014-02-16 13:43:11 +00003171 name = fs_path_alloc();
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003172 from_path = fs_path_alloc();
Filipe Manana2b863a12014-02-16 13:43:11 +00003173 if (!name || !from_path) {
3174 ret = -ENOMEM;
3175 goto out;
3176 }
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003177
Filipe Manana9dc44212014-02-19 14:31:44 +00003178 dm = get_waiting_dir_move(sctx, pm->ino);
3179 ASSERT(dm);
3180 rmdir_ino = dm->rmdir_ino;
Robbie Ko801bec32015-06-23 18:39:46 +08003181 is_orphan = dm->orphanized;
Filipe Manana9dc44212014-02-19 14:31:44 +00003182 free_waiting_dir_move(sctx, dm);
Filipe Manana2b863a12014-02-16 13:43:11 +00003183
Robbie Ko801bec32015-06-23 18:39:46 +08003184 if (is_orphan) {
Filipe Manana84471e22015-02-28 22:29:22 +00003185 ret = gen_unique_name(sctx, pm->ino,
3186 pm->gen, from_path);
3187 } else {
3188 ret = get_first_ref(sctx->parent_root, pm->ino,
3189 &parent_ino, &parent_gen, name);
3190 if (ret < 0)
3191 goto out;
3192 ret = get_cur_path(sctx, parent_ino, parent_gen,
3193 from_path);
3194 if (ret < 0)
3195 goto out;
3196 ret = fs_path_add_path(from_path, name);
3197 }
Filipe Mananac992ec92014-03-22 17:15:24 +00003198 if (ret < 0)
3199 goto out;
3200
Filipe Mananaf9594922014-03-27 20:14:01 +00003201 sctx->send_progress = sctx->cur_ino + 1;
Robbie Ko801bec32015-06-23 18:39:46 +08003202 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
Filipe Manana7969e772016-06-17 17:13:36 +01003203 if (ret < 0)
3204 goto out;
Robbie Ko801bec32015-06-23 18:39:46 +08003205 if (ret) {
3206 LIST_HEAD(deleted_refs);
3207 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3208 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3209 &pm->update_refs, &deleted_refs,
3210 is_orphan);
3211 if (ret < 0)
3212 goto out;
3213 if (rmdir_ino) {
3214 dm = get_waiting_dir_move(sctx, pm->ino);
3215 ASSERT(dm);
3216 dm->rmdir_ino = rmdir_ino;
3217 }
3218 goto out;
3219 }
Filipe Mananac992ec92014-03-22 17:15:24 +00003220 fs_path_reset(name);
3221 to_path = name;
3222 name = NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003223 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3224 if (ret < 0)
3225 goto out;
3226
3227 ret = send_rename(sctx, from_path, to_path);
3228 if (ret < 0)
3229 goto out;
3230
Filipe Manana9dc44212014-02-19 14:31:44 +00003231 if (rmdir_ino) {
3232 struct orphan_dir_info *odi;
3233
3234 odi = get_orphan_dir_info(sctx, rmdir_ino);
3235 if (!odi) {
3236 /* already deleted */
3237 goto finish;
3238 }
3239 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1);
3240 if (ret < 0)
3241 goto out;
3242 if (!ret)
3243 goto finish;
3244
3245 name = fs_path_alloc();
3246 if (!name) {
3247 ret = -ENOMEM;
3248 goto out;
3249 }
3250 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3251 if (ret < 0)
3252 goto out;
3253 ret = send_rmdir(sctx, name);
3254 if (ret < 0)
3255 goto out;
3256 free_orphan_dir_info(sctx, odi);
3257 }
3258
3259finish:
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003260 ret = send_utimes(sctx, pm->ino, pm->gen);
3261 if (ret < 0)
3262 goto out;
3263
3264 /*
3265 * After rename/move, need to update the utimes of both new parent(s)
3266 * and old parent(s).
3267 */
3268 list_for_each_entry(cur, &pm->update_refs, list) {
Filipe Manana9dc44212014-02-19 14:31:44 +00003269 if (cur->dir == rmdir_ino)
3270 continue;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003271 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3272 if (ret < 0)
3273 goto out;
3274 }
3275
3276out:
Filipe Manana2b863a12014-02-16 13:43:11 +00003277 fs_path_free(name);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003278 fs_path_free(from_path);
3279 fs_path_free(to_path);
3280 sctx->send_progress = orig_progress;
3281
3282 return ret;
3283}
3284
3285static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3286{
3287 if (!list_empty(&m->list))
3288 list_del(&m->list);
3289 if (!RB_EMPTY_NODE(&m->node))
3290 rb_erase(&m->node, &sctx->pending_dir_moves);
3291 __free_recorded_refs(&m->update_refs);
3292 kfree(m);
3293}
3294
3295static void tail_append_pending_moves(struct pending_dir_move *moves,
3296 struct list_head *stack)
3297{
3298 if (list_empty(&moves->list)) {
3299 list_add_tail(&moves->list, stack);
3300 } else {
3301 LIST_HEAD(list);
3302 list_splice_init(&moves->list, &list);
3303 list_add_tail(&moves->list, stack);
3304 list_splice_tail(&list, stack);
3305 }
3306}
3307
3308static int apply_children_dir_moves(struct send_ctx *sctx)
3309{
3310 struct pending_dir_move *pm;
3311 struct list_head stack;
3312 u64 parent_ino = sctx->cur_ino;
3313 int ret = 0;
3314
3315 pm = get_pending_dir_moves(sctx, parent_ino);
3316 if (!pm)
3317 return 0;
3318
3319 INIT_LIST_HEAD(&stack);
3320 tail_append_pending_moves(pm, &stack);
3321
3322 while (!list_empty(&stack)) {
3323 pm = list_first_entry(&stack, struct pending_dir_move, list);
3324 parent_ino = pm->ino;
3325 ret = apply_dir_move(sctx, pm);
3326 free_pending_move(sctx, pm);
3327 if (ret)
3328 goto out;
3329 pm = get_pending_dir_moves(sctx, parent_ino);
3330 if (pm)
3331 tail_append_pending_moves(pm, &stack);
3332 }
3333 return 0;
3334
3335out:
3336 while (!list_empty(&stack)) {
3337 pm = list_first_entry(&stack, struct pending_dir_move, list);
3338 free_pending_move(sctx, pm);
3339 }
3340 return ret;
3341}
3342
Filipe Manana84471e22015-02-28 22:29:22 +00003343/*
3344 * We might need to delay a directory rename even when no ancestor directory
3345 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3346 * renamed. This happens when we rename a directory to the old name (the name
3347 * in the parent root) of some other unrelated directory that got its rename
3348 * delayed due to some ancestor with higher number that got renamed.
3349 *
3350 * Example:
3351 *
3352 * Parent snapshot:
3353 * . (ino 256)
3354 * |---- a/ (ino 257)
3355 * | |---- file (ino 260)
3356 * |
3357 * |---- b/ (ino 258)
3358 * |---- c/ (ino 259)
3359 *
3360 * Send snapshot:
3361 * . (ino 256)
3362 * |---- a/ (ino 258)
3363 * |---- x/ (ino 259)
3364 * |---- y/ (ino 257)
3365 * |----- file (ino 260)
3366 *
3367 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3368 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3369 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3370 * must issue is:
3371 *
3372 * 1 - rename 259 from 'c' to 'x'
3373 * 2 - rename 257 from 'a' to 'x/y'
3374 * 3 - rename 258 from 'b' to 'a'
3375 *
3376 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3377 * be done right away and < 0 on error.
3378 */
3379static int wait_for_dest_dir_move(struct send_ctx *sctx,
3380 struct recorded_ref *parent_ref,
3381 const bool is_orphan)
3382{
3383 struct btrfs_path *path;
3384 struct btrfs_key key;
3385 struct btrfs_key di_key;
3386 struct btrfs_dir_item *di;
3387 u64 left_gen;
3388 u64 right_gen;
3389 int ret = 0;
Robbie Ko801bec32015-06-23 18:39:46 +08003390 struct waiting_dir_move *wdm;
Filipe Manana84471e22015-02-28 22:29:22 +00003391
3392 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3393 return 0;
3394
3395 path = alloc_path_for_send();
3396 if (!path)
3397 return -ENOMEM;
3398
3399 key.objectid = parent_ref->dir;
3400 key.type = BTRFS_DIR_ITEM_KEY;
3401 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3402
3403 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3404 if (ret < 0) {
3405 goto out;
3406 } else if (ret > 0) {
3407 ret = 0;
3408 goto out;
3409 }
3410
3411 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3412 parent_ref->name, parent_ref->name_len);
3413 if (!di) {
3414 ret = 0;
3415 goto out;
3416 }
3417 /*
3418 * di_key.objectid has the number of the inode that has a dentry in the
3419 * parent directory with the same name that sctx->cur_ino is being
3420 * renamed to. We need to check if that inode is in the send root as
3421 * well and if it is currently marked as an inode with a pending rename,
3422 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3423 * that it happens after that other inode is renamed.
3424 */
3425 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3426 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3427 ret = 0;
3428 goto out;
3429 }
3430
3431 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3432 &left_gen, NULL, NULL, NULL, NULL);
3433 if (ret < 0)
3434 goto out;
3435 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3436 &right_gen, NULL, NULL, NULL, NULL);
3437 if (ret < 0) {
3438 if (ret == -ENOENT)
3439 ret = 0;
3440 goto out;
3441 }
3442
3443 /* Different inode, no need to delay the rename of sctx->cur_ino */
3444 if (right_gen != left_gen) {
3445 ret = 0;
3446 goto out;
3447 }
3448
Robbie Ko801bec32015-06-23 18:39:46 +08003449 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3450 if (wdm && !wdm->orphanized) {
Filipe Manana84471e22015-02-28 22:29:22 +00003451 ret = add_pending_dir_move(sctx,
3452 sctx->cur_ino,
3453 sctx->cur_inode_gen,
3454 di_key.objectid,
3455 &sctx->new_refs,
3456 &sctx->deleted_refs,
3457 is_orphan);
3458 if (!ret)
3459 ret = 1;
3460 }
3461out:
3462 btrfs_free_path(path);
3463 return ret;
3464}
3465
Filipe Manana80aa6022015-03-27 17:50:45 +00003466/*
3467 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3468 * Return 1 if true, 0 if false and < 0 on error.
3469 */
3470static int is_ancestor(struct btrfs_root *root,
3471 const u64 ino1,
3472 const u64 ino1_gen,
3473 const u64 ino2,
3474 struct fs_path *fs_path)
3475{
3476 u64 ino = ino2;
3477
3478 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3479 int ret;
3480 u64 parent;
3481 u64 parent_gen;
3482
3483 fs_path_reset(fs_path);
3484 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3485 if (ret < 0) {
3486 if (ret == -ENOENT && ino == ino2)
3487 ret = 0;
3488 return ret;
3489 }
3490 if (parent == ino1)
3491 return parent_gen == ino1_gen ? 1 : 0;
3492 ino = parent;
3493 }
3494 return 0;
3495}
3496
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003497static int wait_for_parent_move(struct send_ctx *sctx,
Filipe Manana8b191a62015-04-09 14:09:14 +01003498 struct recorded_ref *parent_ref,
3499 const bool is_orphan)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003500{
Filipe Mananaf9594922014-03-27 20:14:01 +00003501 int ret = 0;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003502 u64 ino = parent_ref->dir;
3503 u64 parent_ino_before, parent_ino_after;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003504 struct fs_path *path_before = NULL;
3505 struct fs_path *path_after = NULL;
3506 int len1, len2;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003507
3508 path_after = fs_path_alloc();
Filipe Mananaf9594922014-03-27 20:14:01 +00003509 path_before = fs_path_alloc();
3510 if (!path_after || !path_before) {
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003511 ret = -ENOMEM;
3512 goto out;
3513 }
3514
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003515 /*
Filipe Mananaf9594922014-03-27 20:14:01 +00003516 * Our current directory inode may not yet be renamed/moved because some
3517 * ancestor (immediate or not) has to be renamed/moved first. So find if
3518 * such ancestor exists and make sure our own rename/move happens after
Filipe Manana80aa6022015-03-27 17:50:45 +00003519 * that ancestor is processed to avoid path build infinite loops (done
3520 * at get_cur_path()).
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003521 */
Filipe Mananaf9594922014-03-27 20:14:01 +00003522 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3523 if (is_waiting_for_move(sctx, ino)) {
Filipe Manana80aa6022015-03-27 17:50:45 +00003524 /*
3525 * If the current inode is an ancestor of ino in the
3526 * parent root, we need to delay the rename of the
3527 * current inode, otherwise don't delayed the rename
3528 * because we can end up with a circular dependency
3529 * of renames, resulting in some directories never
3530 * getting the respective rename operations issued in
3531 * the send stream or getting into infinite path build
3532 * loops.
3533 */
3534 ret = is_ancestor(sctx->parent_root,
3535 sctx->cur_ino, sctx->cur_inode_gen,
3536 ino, path_before);
Filipe Manana4122ea62016-06-27 16:54:44 +01003537 if (ret)
3538 break;
Filipe Mananaf9594922014-03-27 20:14:01 +00003539 }
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003540
3541 fs_path_reset(path_before);
3542 fs_path_reset(path_after);
3543
3544 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
Filipe Mananaf9594922014-03-27 20:14:01 +00003545 NULL, path_after);
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003546 if (ret < 0)
3547 goto out;
3548 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3549 NULL, path_before);
Filipe Mananaf9594922014-03-27 20:14:01 +00003550 if (ret < 0 && ret != -ENOENT) {
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003551 goto out;
Filipe Mananaf9594922014-03-27 20:14:01 +00003552 } else if (ret == -ENOENT) {
Filipe Mananabf8e8ca2014-10-02 19:17:32 +01003553 ret = 0;
Filipe Mananaf9594922014-03-27 20:14:01 +00003554 break;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003555 }
3556
3557 len1 = fs_path_len(path_before);
3558 len2 = fs_path_len(path_after);
Filipe Mananaf9594922014-03-27 20:14:01 +00003559 if (ino > sctx->cur_ino &&
3560 (parent_ino_before != parent_ino_after || len1 != len2 ||
3561 memcmp(path_before->start, path_after->start, len1))) {
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003562 ret = 1;
Filipe Mananaf9594922014-03-27 20:14:01 +00003563 break;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003564 }
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003565 ino = parent_ino_after;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003566 }
3567
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003568out:
3569 fs_path_free(path_before);
3570 fs_path_free(path_after);
3571
Filipe Mananaf9594922014-03-27 20:14:01 +00003572 if (ret == 1) {
3573 ret = add_pending_dir_move(sctx,
3574 sctx->cur_ino,
3575 sctx->cur_inode_gen,
3576 ino,
3577 &sctx->new_refs,
Filipe Manana84471e22015-02-28 22:29:22 +00003578 &sctx->deleted_refs,
Filipe Manana8b191a62015-04-09 14:09:14 +01003579 is_orphan);
Filipe Mananaf9594922014-03-27 20:14:01 +00003580 if (!ret)
3581 ret = 1;
3582 }
3583
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003584 return ret;
3585}
3586
Alexander Block31db9f72012-07-25 23:19:24 +02003587/*
3588 * This does all the move/link/unlink/rmdir magic.
3589 */
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003590static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
Alexander Block31db9f72012-07-25 23:19:24 +02003591{
3592 int ret = 0;
3593 struct recorded_ref *cur;
Alexander Block1f4692d2012-07-28 10:42:24 +02003594 struct recorded_ref *cur2;
Josef Bacikba5e8f22013-08-16 16:52:55 -04003595 struct list_head check_dirs;
Alexander Block31db9f72012-07-25 23:19:24 +02003596 struct fs_path *valid_path = NULL;
Chris Masonb24baf62012-07-25 19:21:10 -04003597 u64 ow_inode = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02003598 u64 ow_gen;
3599 int did_overwrite = 0;
3600 int is_orphan = 0;
Filipe Manana29d6d302014-02-16 21:01:39 +00003601 u64 last_dir_ino_rm = 0;
Filipe Manana84471e22015-02-28 22:29:22 +00003602 bool can_rename = true;
Alexander Block31db9f72012-07-25 23:19:24 +02003603
3604verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3605
Alexander Block6d85ed02012-08-01 14:48:59 +02003606 /*
3607 * This should never happen as the root dir always has the same ref
3608 * which is always '..'
3609 */
3610 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
Josef Bacikba5e8f22013-08-16 16:52:55 -04003611 INIT_LIST_HEAD(&check_dirs);
Alexander Block6d85ed02012-08-01 14:48:59 +02003612
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003613 valid_path = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02003614 if (!valid_path) {
3615 ret = -ENOMEM;
3616 goto out;
3617 }
3618
Alexander Block31db9f72012-07-25 23:19:24 +02003619 /*
3620 * First, check if the first ref of the current inode was overwritten
3621 * before. If yes, we know that the current inode was already orphanized
3622 * and thus use the orphan name. If not, we can use get_cur_path to
3623 * get the path of the first ref as it would like while receiving at
3624 * this point in time.
3625 * New inodes are always orphan at the beginning, so force to use the
3626 * orphan name in this case.
3627 * The first ref is stored in valid_path and will be updated if it
3628 * gets moved around.
3629 */
3630 if (!sctx->cur_inode_new) {
3631 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3632 sctx->cur_inode_gen);
3633 if (ret < 0)
3634 goto out;
3635 if (ret)
3636 did_overwrite = 1;
3637 }
3638 if (sctx->cur_inode_new || did_overwrite) {
3639 ret = gen_unique_name(sctx, sctx->cur_ino,
3640 sctx->cur_inode_gen, valid_path);
3641 if (ret < 0)
3642 goto out;
3643 is_orphan = 1;
3644 } else {
3645 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3646 valid_path);
3647 if (ret < 0)
3648 goto out;
3649 }
3650
3651 list_for_each_entry(cur, &sctx->new_refs, list) {
3652 /*
Alexander Block1f4692d2012-07-28 10:42:24 +02003653 * We may have refs where the parent directory does not exist
3654 * yet. This happens if the parent directories inum is higher
3655 * the the current inum. To handle this case, we create the
3656 * parent directory out of order. But we need to check if this
3657 * did already happen before due to other refs in the same dir.
3658 */
3659 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3660 if (ret < 0)
3661 goto out;
3662 if (ret == inode_state_will_create) {
3663 ret = 0;
3664 /*
3665 * First check if any of the current inodes refs did
3666 * already create the dir.
3667 */
3668 list_for_each_entry(cur2, &sctx->new_refs, list) {
3669 if (cur == cur2)
3670 break;
3671 if (cur2->dir == cur->dir) {
3672 ret = 1;
3673 break;
3674 }
3675 }
3676
3677 /*
3678 * If that did not happen, check if a previous inode
3679 * did already create the dir.
3680 */
3681 if (!ret)
3682 ret = did_create_dir(sctx, cur->dir);
3683 if (ret < 0)
3684 goto out;
3685 if (!ret) {
3686 ret = send_create_inode(sctx, cur->dir);
3687 if (ret < 0)
3688 goto out;
3689 }
3690 }
3691
3692 /*
Alexander Block31db9f72012-07-25 23:19:24 +02003693 * Check if this new ref would overwrite the first ref of
3694 * another unprocessed inode. If yes, orphanize the
3695 * overwritten inode. If we find an overwritten ref that is
3696 * not the first ref, simply unlink it.
3697 */
3698 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3699 cur->name, cur->name_len,
3700 &ow_inode, &ow_gen);
3701 if (ret < 0)
3702 goto out;
3703 if (ret) {
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003704 ret = is_first_ref(sctx->parent_root,
3705 ow_inode, cur->dir, cur->name,
3706 cur->name_len);
Alexander Block31db9f72012-07-25 23:19:24 +02003707 if (ret < 0)
3708 goto out;
3709 if (ret) {
Filipe Manana8996a482015-03-12 15:16:20 +00003710 struct name_cache_entry *nce;
Robbie Ko801bec32015-06-23 18:39:46 +08003711 struct waiting_dir_move *wdm;
Filipe Manana8996a482015-03-12 15:16:20 +00003712
Alexander Block31db9f72012-07-25 23:19:24 +02003713 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3714 cur->full_path);
3715 if (ret < 0)
3716 goto out;
Robbie Ko801bec32015-06-23 18:39:46 +08003717
3718 /*
3719 * If ow_inode has its rename operation delayed
3720 * make sure that its orphanized name is used in
3721 * the source path when performing its rename
3722 * operation.
3723 */
3724 if (is_waiting_for_move(sctx, ow_inode)) {
3725 wdm = get_waiting_dir_move(sctx,
3726 ow_inode);
3727 ASSERT(wdm);
3728 wdm->orphanized = true;
3729 }
3730
Filipe Manana8996a482015-03-12 15:16:20 +00003731 /*
3732 * Make sure we clear our orphanized inode's
3733 * name from the name cache. This is because the
3734 * inode ow_inode might be an ancestor of some
3735 * other inode that will be orphanized as well
3736 * later and has an inode number greater than
3737 * sctx->send_progress. We need to prevent
3738 * future name lookups from using the old name
3739 * and get instead the orphan name.
3740 */
3741 nce = name_cache_search(sctx, ow_inode, ow_gen);
3742 if (nce) {
3743 name_cache_delete(sctx, nce);
3744 kfree(nce);
3745 }
Robbie Ko801bec32015-06-23 18:39:46 +08003746
3747 /*
3748 * ow_inode might currently be an ancestor of
3749 * cur_ino, therefore compute valid_path (the
3750 * current path of cur_ino) again because it
3751 * might contain the pre-orphanization name of
3752 * ow_inode, which is no longer valid.
3753 */
3754 fs_path_reset(valid_path);
3755 ret = get_cur_path(sctx, sctx->cur_ino,
3756 sctx->cur_inode_gen, valid_path);
3757 if (ret < 0)
3758 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003759 } else {
3760 ret = send_unlink(sctx, cur->full_path);
3761 if (ret < 0)
3762 goto out;
3763 }
3764 }
3765
Filipe Manana84471e22015-02-28 22:29:22 +00003766 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3767 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3768 if (ret < 0)
3769 goto out;
3770 if (ret == 1) {
3771 can_rename = false;
3772 *pending_move = 1;
3773 }
3774 }
3775
Filipe Manana8b191a62015-04-09 14:09:14 +01003776 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3777 can_rename) {
3778 ret = wait_for_parent_move(sctx, cur, is_orphan);
3779 if (ret < 0)
3780 goto out;
3781 if (ret == 1) {
3782 can_rename = false;
3783 *pending_move = 1;
3784 }
3785 }
3786
Alexander Block31db9f72012-07-25 23:19:24 +02003787 /*
3788 * link/move the ref to the new place. If we have an orphan
3789 * inode, move it and update valid_path. If not, link or move
3790 * it depending on the inode mode.
3791 */
Filipe Manana84471e22015-02-28 22:29:22 +00003792 if (is_orphan && can_rename) {
Alexander Block31db9f72012-07-25 23:19:24 +02003793 ret = send_rename(sctx, valid_path, cur->full_path);
3794 if (ret < 0)
3795 goto out;
3796 is_orphan = 0;
3797 ret = fs_path_copy(valid_path, cur->full_path);
3798 if (ret < 0)
3799 goto out;
Filipe Manana84471e22015-02-28 22:29:22 +00003800 } else if (can_rename) {
Alexander Block31db9f72012-07-25 23:19:24 +02003801 if (S_ISDIR(sctx->cur_inode_mode)) {
3802 /*
3803 * Dirs can't be linked, so move it. For moved
3804 * dirs, we always have one new and one deleted
3805 * ref. The deleted ref is ignored later.
3806 */
Filipe Manana8b191a62015-04-09 14:09:14 +01003807 ret = send_rename(sctx, valid_path,
3808 cur->full_path);
3809 if (!ret)
3810 ret = fs_path_copy(valid_path,
3811 cur->full_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003812 if (ret < 0)
3813 goto out;
3814 } else {
3815 ret = send_link(sctx, cur->full_path,
3816 valid_path);
3817 if (ret < 0)
3818 goto out;
3819 }
3820 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04003821 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003822 if (ret < 0)
3823 goto out;
3824 }
3825
3826 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3827 /*
3828 * Check if we can already rmdir the directory. If not,
3829 * orphanize it. For every dir item inside that gets deleted
3830 * later, we do this check again and rmdir it then if possible.
3831 * See the use of check_dirs for more details.
3832 */
Filipe Manana9dc44212014-02-19 14:31:44 +00003833 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3834 sctx->cur_ino);
Alexander Block31db9f72012-07-25 23:19:24 +02003835 if (ret < 0)
3836 goto out;
3837 if (ret) {
3838 ret = send_rmdir(sctx, valid_path);
3839 if (ret < 0)
3840 goto out;
3841 } else if (!is_orphan) {
3842 ret = orphanize_inode(sctx, sctx->cur_ino,
3843 sctx->cur_inode_gen, valid_path);
3844 if (ret < 0)
3845 goto out;
3846 is_orphan = 1;
3847 }
3848
3849 list_for_each_entry(cur, &sctx->deleted_refs, list) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04003850 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003851 if (ret < 0)
3852 goto out;
3853 }
Alexander Blockccf16262012-07-28 11:46:29 +02003854 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3855 !list_empty(&sctx->deleted_refs)) {
3856 /*
3857 * We have a moved dir. Add the old parent to check_dirs
3858 */
3859 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3860 list);
Josef Bacikba5e8f22013-08-16 16:52:55 -04003861 ret = dup_ref(cur, &check_dirs);
Alexander Blockccf16262012-07-28 11:46:29 +02003862 if (ret < 0)
3863 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003864 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3865 /*
3866 * We have a non dir inode. Go through all deleted refs and
3867 * unlink them if they were not already overwritten by other
3868 * inodes.
3869 */
3870 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3871 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3872 sctx->cur_ino, sctx->cur_inode_gen,
3873 cur->name, cur->name_len);
3874 if (ret < 0)
3875 goto out;
3876 if (!ret) {
Alexander Block1f4692d2012-07-28 10:42:24 +02003877 ret = send_unlink(sctx, cur->full_path);
3878 if (ret < 0)
3879 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003880 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04003881 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003882 if (ret < 0)
3883 goto out;
3884 }
Alexander Block31db9f72012-07-25 23:19:24 +02003885 /*
3886 * If the inode is still orphan, unlink the orphan. This may
3887 * happen when a previous inode did overwrite the first ref
3888 * of this inode and no new refs were added for the current
Alexander Block766702e2012-07-28 14:11:31 +02003889 * inode. Unlinking does not mean that the inode is deleted in
3890 * all cases. There may still be links to this inode in other
3891 * places.
Alexander Block31db9f72012-07-25 23:19:24 +02003892 */
Alexander Block1f4692d2012-07-28 10:42:24 +02003893 if (is_orphan) {
Alexander Block31db9f72012-07-25 23:19:24 +02003894 ret = send_unlink(sctx, valid_path);
3895 if (ret < 0)
3896 goto out;
3897 }
3898 }
3899
3900 /*
3901 * We did collect all parent dirs where cur_inode was once located. We
3902 * now go through all these dirs and check if they are pending for
3903 * deletion and if it's finally possible to perform the rmdir now.
3904 * We also update the inode stats of the parent dirs here.
3905 */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003906 list_for_each_entry(cur, &check_dirs, list) {
Alexander Block766702e2012-07-28 14:11:31 +02003907 /*
3908 * In case we had refs into dirs that were not processed yet,
3909 * we don't need to do the utime and rmdir logic for these dirs.
3910 * The dir will be processed later.
3911 */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003912 if (cur->dir > sctx->cur_ino)
Alexander Block31db9f72012-07-25 23:19:24 +02003913 continue;
3914
Josef Bacikba5e8f22013-08-16 16:52:55 -04003915 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
Alexander Block31db9f72012-07-25 23:19:24 +02003916 if (ret < 0)
3917 goto out;
3918
3919 if (ret == inode_state_did_create ||
3920 ret == inode_state_no_change) {
3921 /* TODO delayed utimes */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003922 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
Alexander Block31db9f72012-07-25 23:19:24 +02003923 if (ret < 0)
3924 goto out;
Filipe Manana29d6d302014-02-16 21:01:39 +00003925 } else if (ret == inode_state_did_delete &&
3926 cur->dir != last_dir_ino_rm) {
Filipe Manana9dc44212014-02-19 14:31:44 +00003927 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3928 sctx->cur_ino);
Alexander Block31db9f72012-07-25 23:19:24 +02003929 if (ret < 0)
3930 goto out;
3931 if (ret) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04003932 ret = get_cur_path(sctx, cur->dir,
3933 cur->dir_gen, valid_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003934 if (ret < 0)
3935 goto out;
3936 ret = send_rmdir(sctx, valid_path);
3937 if (ret < 0)
3938 goto out;
Filipe Manana29d6d302014-02-16 21:01:39 +00003939 last_dir_ino_rm = cur->dir;
Alexander Block31db9f72012-07-25 23:19:24 +02003940 }
3941 }
3942 }
3943
Alexander Block31db9f72012-07-25 23:19:24 +02003944 ret = 0;
3945
3946out:
Josef Bacikba5e8f22013-08-16 16:52:55 -04003947 __free_recorded_refs(&check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003948 free_recorded_refs(sctx);
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003949 fs_path_free(valid_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003950 return ret;
3951}
3952
Liu Boa4d96d62014-03-03 21:31:03 +08003953static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3954 struct fs_path *name, void *ctx, struct list_head *refs)
Alexander Block31db9f72012-07-25 23:19:24 +02003955{
3956 int ret = 0;
3957 struct send_ctx *sctx = ctx;
3958 struct fs_path *p;
3959 u64 gen;
3960
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003961 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02003962 if (!p)
3963 return -ENOMEM;
3964
Liu Boa4d96d62014-03-03 21:31:03 +08003965 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02003966 NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02003967 if (ret < 0)
3968 goto out;
3969
Alexander Block31db9f72012-07-25 23:19:24 +02003970 ret = get_cur_path(sctx, dir, gen, p);
3971 if (ret < 0)
3972 goto out;
3973 ret = fs_path_add_path(p, name);
3974 if (ret < 0)
3975 goto out;
3976
Liu Boa4d96d62014-03-03 21:31:03 +08003977 ret = __record_ref(refs, dir, gen, p);
Alexander Block31db9f72012-07-25 23:19:24 +02003978
3979out:
3980 if (ret)
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003981 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02003982 return ret;
3983}
3984
Liu Boa4d96d62014-03-03 21:31:03 +08003985static int __record_new_ref(int num, u64 dir, int index,
3986 struct fs_path *name,
3987 void *ctx)
3988{
3989 struct send_ctx *sctx = ctx;
3990 return record_ref(sctx->send_root, num, dir, index, name,
3991 ctx, &sctx->new_refs);
3992}
3993
3994
Alexander Block31db9f72012-07-25 23:19:24 +02003995static int __record_deleted_ref(int num, u64 dir, int index,
3996 struct fs_path *name,
3997 void *ctx)
3998{
Alexander Block31db9f72012-07-25 23:19:24 +02003999 struct send_ctx *sctx = ctx;
Liu Boa4d96d62014-03-03 21:31:03 +08004000 return record_ref(sctx->parent_root, num, dir, index, name,
4001 ctx, &sctx->deleted_refs);
Alexander Block31db9f72012-07-25 23:19:24 +02004002}
4003
4004static int record_new_ref(struct send_ctx *sctx)
4005{
4006 int ret;
4007
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004008 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4009 sctx->cmp_key, 0, __record_new_ref, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004010 if (ret < 0)
4011 goto out;
4012 ret = 0;
4013
4014out:
4015 return ret;
4016}
4017
4018static int record_deleted_ref(struct send_ctx *sctx)
4019{
4020 int ret;
4021
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004022 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4023 sctx->cmp_key, 0, __record_deleted_ref, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004024 if (ret < 0)
4025 goto out;
4026 ret = 0;
4027
4028out:
4029 return ret;
4030}
4031
4032struct find_ref_ctx {
4033 u64 dir;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004034 u64 dir_gen;
4035 struct btrfs_root *root;
Alexander Block31db9f72012-07-25 23:19:24 +02004036 struct fs_path *name;
4037 int found_idx;
4038};
4039
4040static int __find_iref(int num, u64 dir, int index,
4041 struct fs_path *name,
4042 void *ctx_)
4043{
4044 struct find_ref_ctx *ctx = ctx_;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004045 u64 dir_gen;
4046 int ret;
Alexander Block31db9f72012-07-25 23:19:24 +02004047
4048 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4049 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04004050 /*
4051 * To avoid doing extra lookups we'll only do this if everything
4052 * else matches.
4053 */
4054 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4055 NULL, NULL, NULL);
4056 if (ret)
4057 return ret;
4058 if (dir_gen != ctx->dir_gen)
4059 return 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004060 ctx->found_idx = num;
4061 return 1;
4062 }
4063 return 0;
4064}
4065
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004066static int find_iref(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02004067 struct btrfs_path *path,
4068 struct btrfs_key *key,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004069 u64 dir, u64 dir_gen, struct fs_path *name)
Alexander Block31db9f72012-07-25 23:19:24 +02004070{
4071 int ret;
4072 struct find_ref_ctx ctx;
4073
4074 ctx.dir = dir;
4075 ctx.name = name;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004076 ctx.dir_gen = dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004077 ctx.found_idx = -1;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004078 ctx.root = root;
Alexander Block31db9f72012-07-25 23:19:24 +02004079
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004080 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004081 if (ret < 0)
4082 return ret;
4083
4084 if (ctx.found_idx == -1)
4085 return -ENOENT;
4086
4087 return ctx.found_idx;
4088}
4089
4090static int __record_changed_new_ref(int num, u64 dir, int index,
4091 struct fs_path *name,
4092 void *ctx)
4093{
Josef Bacikba5e8f22013-08-16 16:52:55 -04004094 u64 dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004095 int ret;
4096 struct send_ctx *sctx = ctx;
4097
Josef Bacikba5e8f22013-08-16 16:52:55 -04004098 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4099 NULL, NULL, NULL);
4100 if (ret)
4101 return ret;
4102
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004103 ret = find_iref(sctx->parent_root, sctx->right_path,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004104 sctx->cmp_key, dir, dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02004105 if (ret == -ENOENT)
4106 ret = __record_new_ref(num, dir, index, name, sctx);
4107 else if (ret > 0)
4108 ret = 0;
4109
4110 return ret;
4111}
4112
4113static int __record_changed_deleted_ref(int num, u64 dir, int index,
4114 struct fs_path *name,
4115 void *ctx)
4116{
Josef Bacikba5e8f22013-08-16 16:52:55 -04004117 u64 dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004118 int ret;
4119 struct send_ctx *sctx = ctx;
4120
Josef Bacikba5e8f22013-08-16 16:52:55 -04004121 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4122 NULL, NULL, NULL);
4123 if (ret)
4124 return ret;
4125
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004126 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004127 dir, dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02004128 if (ret == -ENOENT)
4129 ret = __record_deleted_ref(num, dir, index, name, sctx);
4130 else if (ret > 0)
4131 ret = 0;
4132
4133 return ret;
4134}
4135
4136static int record_changed_ref(struct send_ctx *sctx)
4137{
4138 int ret = 0;
4139
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004140 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004141 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4142 if (ret < 0)
4143 goto out;
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004144 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004145 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4146 if (ret < 0)
4147 goto out;
4148 ret = 0;
4149
4150out:
4151 return ret;
4152}
4153
4154/*
4155 * Record and process all refs at once. Needed when an inode changes the
4156 * generation number, which means that it was deleted and recreated.
4157 */
4158static int process_all_refs(struct send_ctx *sctx,
4159 enum btrfs_compare_tree_result cmd)
4160{
4161 int ret;
4162 struct btrfs_root *root;
4163 struct btrfs_path *path;
4164 struct btrfs_key key;
4165 struct btrfs_key found_key;
4166 struct extent_buffer *eb;
4167 int slot;
4168 iterate_inode_ref_t cb;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00004169 int pending_move = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004170
4171 path = alloc_path_for_send();
4172 if (!path)
4173 return -ENOMEM;
4174
4175 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4176 root = sctx->send_root;
4177 cb = __record_new_ref;
4178 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4179 root = sctx->parent_root;
4180 cb = __record_deleted_ref;
4181 } else {
David Sterba4d1a63b2014-02-03 19:24:19 +01004182 btrfs_err(sctx->send_root->fs_info,
4183 "Wrong command %d in process_all_refs", cmd);
4184 ret = -EINVAL;
4185 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004186 }
4187
4188 key.objectid = sctx->cmp_key->objectid;
4189 key.type = BTRFS_INODE_REF_KEY;
4190 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004191 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4192 if (ret < 0)
4193 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004194
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004195 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02004196 eb = path->nodes[0];
4197 slot = path->slots[0];
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004198 if (slot >= btrfs_header_nritems(eb)) {
4199 ret = btrfs_next_leaf(root, path);
4200 if (ret < 0)
4201 goto out;
4202 else if (ret > 0)
4203 break;
4204 continue;
4205 }
4206
Alexander Block31db9f72012-07-25 23:19:24 +02004207 btrfs_item_key_to_cpu(eb, &found_key, slot);
4208
4209 if (found_key.objectid != key.objectid ||
Jan Schmidt96b5bd72012-10-15 08:30:45 +00004210 (found_key.type != BTRFS_INODE_REF_KEY &&
4211 found_key.type != BTRFS_INODE_EXTREF_KEY))
Alexander Block31db9f72012-07-25 23:19:24 +02004212 break;
Alexander Block31db9f72012-07-25 23:19:24 +02004213
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004214 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004215 if (ret < 0)
4216 goto out;
4217
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004218 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02004219 }
Alexander Blocke938c8a2012-07-28 16:33:49 +02004220 btrfs_release_path(path);
Alexander Block31db9f72012-07-25 23:19:24 +02004221
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00004222 ret = process_recorded_refs(sctx, &pending_move);
4223 /* Only applicable to an incremental send. */
4224 ASSERT(pending_move == 0);
Alexander Block31db9f72012-07-25 23:19:24 +02004225
4226out:
4227 btrfs_free_path(path);
4228 return ret;
4229}
4230
4231static int send_set_xattr(struct send_ctx *sctx,
4232 struct fs_path *path,
4233 const char *name, int name_len,
4234 const char *data, int data_len)
4235{
4236 int ret = 0;
4237
4238 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4239 if (ret < 0)
4240 goto out;
4241
4242 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4243 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4244 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4245
4246 ret = send_cmd(sctx);
4247
4248tlv_put_failure:
4249out:
4250 return ret;
4251}
4252
4253static int send_remove_xattr(struct send_ctx *sctx,
4254 struct fs_path *path,
4255 const char *name, int name_len)
4256{
4257 int ret = 0;
4258
4259 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4260 if (ret < 0)
4261 goto out;
4262
4263 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4264 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4265
4266 ret = send_cmd(sctx);
4267
4268tlv_put_failure:
4269out:
4270 return ret;
4271}
4272
4273static int __process_new_xattr(int num, struct btrfs_key *di_key,
4274 const char *name, int name_len,
4275 const char *data, int data_len,
4276 u8 type, void *ctx)
4277{
4278 int ret;
4279 struct send_ctx *sctx = ctx;
4280 struct fs_path *p;
4281 posix_acl_xattr_header dummy_acl;
4282
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004283 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004284 if (!p)
4285 return -ENOMEM;
4286
4287 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04004288 * This hack is needed because empty acls are stored as zero byte
Alexander Block31db9f72012-07-25 23:19:24 +02004289 * data in xattrs. Problem with that is, that receiving these zero byte
Nicholas D Steeves01327612016-05-19 21:18:45 -04004290 * acls will fail later. To fix this, we send a dummy acl list that
Alexander Block31db9f72012-07-25 23:19:24 +02004291 * only contains the version number and no entries.
4292 */
4293 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4294 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4295 if (data_len == 0) {
4296 dummy_acl.a_version =
4297 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4298 data = (char *)&dummy_acl;
4299 data_len = sizeof(dummy_acl);
4300 }
4301 }
4302
4303 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4304 if (ret < 0)
4305 goto out;
4306
4307 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4308
4309out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004310 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004311 return ret;
4312}
4313
4314static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4315 const char *name, int name_len,
4316 const char *data, int data_len,
4317 u8 type, void *ctx)
4318{
4319 int ret;
4320 struct send_ctx *sctx = ctx;
4321 struct fs_path *p;
4322
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004323 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004324 if (!p)
4325 return -ENOMEM;
4326
4327 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4328 if (ret < 0)
4329 goto out;
4330
4331 ret = send_remove_xattr(sctx, p, name, name_len);
4332
4333out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004334 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004335 return ret;
4336}
4337
4338static int process_new_xattr(struct send_ctx *sctx)
4339{
4340 int ret = 0;
4341
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004342 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4343 sctx->cmp_key, __process_new_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004344
4345 return ret;
4346}
4347
4348static int process_deleted_xattr(struct send_ctx *sctx)
4349{
4350 int ret;
4351
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004352 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4353 sctx->cmp_key, __process_deleted_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004354
4355 return ret;
4356}
4357
4358struct find_xattr_ctx {
4359 const char *name;
4360 int name_len;
4361 int found_idx;
4362 char *found_data;
4363 int found_data_len;
4364};
4365
4366static int __find_xattr(int num, struct btrfs_key *di_key,
4367 const char *name, int name_len,
4368 const char *data, int data_len,
4369 u8 type, void *vctx)
4370{
4371 struct find_xattr_ctx *ctx = vctx;
4372
4373 if (name_len == ctx->name_len &&
4374 strncmp(name, ctx->name, name_len) == 0) {
4375 ctx->found_idx = num;
4376 ctx->found_data_len = data_len;
David Sterbae780b0d2016-01-18 18:42:13 +01004377 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02004378 if (!ctx->found_data)
4379 return -ENOMEM;
Alexander Block31db9f72012-07-25 23:19:24 +02004380 return 1;
4381 }
4382 return 0;
4383}
4384
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004385static int find_xattr(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02004386 struct btrfs_path *path,
4387 struct btrfs_key *key,
4388 const char *name, int name_len,
4389 char **data, int *data_len)
4390{
4391 int ret;
4392 struct find_xattr_ctx ctx;
4393
4394 ctx.name = name;
4395 ctx.name_len = name_len;
4396 ctx.found_idx = -1;
4397 ctx.found_data = NULL;
4398 ctx.found_data_len = 0;
4399
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004400 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004401 if (ret < 0)
4402 return ret;
4403
4404 if (ctx.found_idx == -1)
4405 return -ENOENT;
4406 if (data) {
4407 *data = ctx.found_data;
4408 *data_len = ctx.found_data_len;
4409 } else {
4410 kfree(ctx.found_data);
4411 }
4412 return ctx.found_idx;
4413}
4414
4415
4416static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4417 const char *name, int name_len,
4418 const char *data, int data_len,
4419 u8 type, void *ctx)
4420{
4421 int ret;
4422 struct send_ctx *sctx = ctx;
4423 char *found_data = NULL;
4424 int found_data_len = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004425
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004426 ret = find_xattr(sctx->parent_root, sctx->right_path,
4427 sctx->cmp_key, name, name_len, &found_data,
4428 &found_data_len);
Alexander Block31db9f72012-07-25 23:19:24 +02004429 if (ret == -ENOENT) {
4430 ret = __process_new_xattr(num, di_key, name, name_len, data,
4431 data_len, type, ctx);
4432 } else if (ret >= 0) {
4433 if (data_len != found_data_len ||
4434 memcmp(data, found_data, data_len)) {
4435 ret = __process_new_xattr(num, di_key, name, name_len,
4436 data, data_len, type, ctx);
4437 } else {
4438 ret = 0;
4439 }
4440 }
4441
4442 kfree(found_data);
Alexander Block31db9f72012-07-25 23:19:24 +02004443 return ret;
4444}
4445
4446static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4447 const char *name, int name_len,
4448 const char *data, int data_len,
4449 u8 type, void *ctx)
4450{
4451 int ret;
4452 struct send_ctx *sctx = ctx;
4453
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004454 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4455 name, name_len, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02004456 if (ret == -ENOENT)
4457 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4458 data_len, type, ctx);
4459 else if (ret >= 0)
4460 ret = 0;
4461
4462 return ret;
4463}
4464
4465static int process_changed_xattr(struct send_ctx *sctx)
4466{
4467 int ret = 0;
4468
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004469 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004470 sctx->cmp_key, __process_changed_new_xattr, sctx);
4471 if (ret < 0)
4472 goto out;
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004473 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004474 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4475
4476out:
4477 return ret;
4478}
4479
4480static int process_all_new_xattrs(struct send_ctx *sctx)
4481{
4482 int ret;
4483 struct btrfs_root *root;
4484 struct btrfs_path *path;
4485 struct btrfs_key key;
4486 struct btrfs_key found_key;
4487 struct extent_buffer *eb;
4488 int slot;
4489
4490 path = alloc_path_for_send();
4491 if (!path)
4492 return -ENOMEM;
4493
4494 root = sctx->send_root;
4495
4496 key.objectid = sctx->cmp_key->objectid;
4497 key.type = BTRFS_XATTR_ITEM_KEY;
4498 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004499 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4500 if (ret < 0)
4501 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004502
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004503 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02004504 eb = path->nodes[0];
4505 slot = path->slots[0];
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004506 if (slot >= btrfs_header_nritems(eb)) {
4507 ret = btrfs_next_leaf(root, path);
4508 if (ret < 0) {
4509 goto out;
4510 } else if (ret > 0) {
4511 ret = 0;
4512 break;
4513 }
4514 continue;
4515 }
Alexander Block31db9f72012-07-25 23:19:24 +02004516
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004517 btrfs_item_key_to_cpu(eb, &found_key, slot);
Alexander Block31db9f72012-07-25 23:19:24 +02004518 if (found_key.objectid != key.objectid ||
4519 found_key.type != key.type) {
4520 ret = 0;
4521 goto out;
4522 }
4523
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004524 ret = iterate_dir_item(root, path, &found_key,
4525 __process_new_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004526 if (ret < 0)
4527 goto out;
4528
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004529 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02004530 }
4531
4532out:
4533 btrfs_free_path(path);
4534 return ret;
4535}
4536
Josef Baciked259092013-10-25 11:36:01 -04004537static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4538{
4539 struct btrfs_root *root = sctx->send_root;
4540 struct btrfs_fs_info *fs_info = root->fs_info;
4541 struct inode *inode;
4542 struct page *page;
4543 char *addr;
4544 struct btrfs_key key;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004545 pgoff_t index = offset >> PAGE_SHIFT;
Josef Baciked259092013-10-25 11:36:01 -04004546 pgoff_t last_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004547 unsigned pg_offset = offset & ~PAGE_MASK;
Josef Baciked259092013-10-25 11:36:01 -04004548 ssize_t ret = 0;
4549
4550 key.objectid = sctx->cur_ino;
4551 key.type = BTRFS_INODE_ITEM_KEY;
4552 key.offset = 0;
4553
4554 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4555 if (IS_ERR(inode))
4556 return PTR_ERR(inode);
4557
4558 if (offset + len > i_size_read(inode)) {
4559 if (offset > i_size_read(inode))
4560 len = 0;
4561 else
4562 len = offset - i_size_read(inode);
4563 }
4564 if (len == 0)
4565 goto out;
4566
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004567 last_index = (offset + len - 1) >> PAGE_SHIFT;
Liu Bo2131bcd2014-03-05 10:07:35 +08004568
4569 /* initial readahead */
4570 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4571 file_ra_state_init(&sctx->ra, inode->i_mapping);
4572 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4573 last_index - index + 1);
4574
Josef Baciked259092013-10-25 11:36:01 -04004575 while (index <= last_index) {
4576 unsigned cur_len = min_t(unsigned, len,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004577 PAGE_SIZE - pg_offset);
David Sterbae780b0d2016-01-18 18:42:13 +01004578 page = find_or_create_page(inode->i_mapping, index, GFP_KERNEL);
Josef Baciked259092013-10-25 11:36:01 -04004579 if (!page) {
4580 ret = -ENOMEM;
4581 break;
4582 }
4583
4584 if (!PageUptodate(page)) {
4585 btrfs_readpage(NULL, page);
4586 lock_page(page);
4587 if (!PageUptodate(page)) {
4588 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004589 put_page(page);
Josef Baciked259092013-10-25 11:36:01 -04004590 ret = -EIO;
4591 break;
4592 }
4593 }
4594
4595 addr = kmap(page);
4596 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4597 kunmap(page);
4598 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004599 put_page(page);
Josef Baciked259092013-10-25 11:36:01 -04004600 index++;
4601 pg_offset = 0;
4602 len -= cur_len;
4603 ret += cur_len;
4604 }
4605out:
4606 iput(inode);
4607 return ret;
4608}
4609
Alexander Block31db9f72012-07-25 23:19:24 +02004610/*
4611 * Read some bytes from the current inode/file and send a write command to
4612 * user space.
4613 */
4614static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4615{
4616 int ret = 0;
4617 struct fs_path *p;
Josef Baciked259092013-10-25 11:36:01 -04004618 ssize_t num_read = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004619
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004620 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004621 if (!p)
4622 return -ENOMEM;
4623
Alexander Block31db9f72012-07-25 23:19:24 +02004624verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4625
Josef Baciked259092013-10-25 11:36:01 -04004626 num_read = fill_read_buf(sctx, offset, len);
4627 if (num_read <= 0) {
4628 if (num_read < 0)
4629 ret = num_read;
Alexander Block31db9f72012-07-25 23:19:24 +02004630 goto out;
Josef Baciked259092013-10-25 11:36:01 -04004631 }
Alexander Block31db9f72012-07-25 23:19:24 +02004632
4633 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4634 if (ret < 0)
4635 goto out;
4636
4637 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4638 if (ret < 0)
4639 goto out;
4640
4641 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4642 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
Alexander Blocke938c8a2012-07-28 16:33:49 +02004643 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
Alexander Block31db9f72012-07-25 23:19:24 +02004644
4645 ret = send_cmd(sctx);
4646
4647tlv_put_failure:
4648out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004649 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004650 if (ret < 0)
4651 return ret;
Alexander Blocke938c8a2012-07-28 16:33:49 +02004652 return num_read;
Alexander Block31db9f72012-07-25 23:19:24 +02004653}
4654
4655/*
4656 * Send a clone command to user space.
4657 */
4658static int send_clone(struct send_ctx *sctx,
4659 u64 offset, u32 len,
4660 struct clone_root *clone_root)
4661{
4662 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004663 struct fs_path *p;
4664 u64 gen;
4665
4666verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4667 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4668 clone_root->root->objectid, clone_root->ino,
4669 clone_root->offset);
4670
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004671 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004672 if (!p)
4673 return -ENOMEM;
4674
4675 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4676 if (ret < 0)
4677 goto out;
4678
4679 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4680 if (ret < 0)
4681 goto out;
4682
4683 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4684 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4685 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4686
Alexander Blocke938c8a2012-07-28 16:33:49 +02004687 if (clone_root->root == sctx->send_root) {
Alexander Block31db9f72012-07-25 23:19:24 +02004688 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02004689 &gen, NULL, NULL, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02004690 if (ret < 0)
4691 goto out;
4692 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4693 } else {
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004694 ret = get_inode_path(clone_root->root, clone_root->ino, p);
Alexander Block31db9f72012-07-25 23:19:24 +02004695 }
4696 if (ret < 0)
4697 goto out;
4698
Josef Bacik37b8d272015-06-04 17:17:25 -04004699 /*
4700 * If the parent we're using has a received_uuid set then use that as
4701 * our clone source as that is what we will look for when doing a
4702 * receive.
4703 *
4704 * This covers the case that we create a snapshot off of a received
4705 * subvolume and then use that as the parent and try to receive on a
4706 * different host.
4707 */
4708 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4709 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4710 clone_root->root->root_item.received_uuid);
4711 else
4712 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4713 clone_root->root->root_item.uuid);
Alexander Block31db9f72012-07-25 23:19:24 +02004714 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
Filipe David Borba Manana5a0f4e22013-12-03 15:55:48 +00004715 le64_to_cpu(clone_root->root->root_item.ctransid));
Alexander Block31db9f72012-07-25 23:19:24 +02004716 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4717 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4718 clone_root->offset);
4719
4720 ret = send_cmd(sctx);
4721
4722tlv_put_failure:
4723out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004724 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004725 return ret;
4726}
4727
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004728/*
4729 * Send an update extent command to user space.
4730 */
4731static int send_update_extent(struct send_ctx *sctx,
4732 u64 offset, u32 len)
4733{
4734 int ret = 0;
4735 struct fs_path *p;
4736
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004737 p = fs_path_alloc();
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004738 if (!p)
4739 return -ENOMEM;
4740
4741 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4742 if (ret < 0)
4743 goto out;
4744
4745 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4746 if (ret < 0)
4747 goto out;
4748
4749 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4750 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4751 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4752
4753 ret = send_cmd(sctx);
4754
4755tlv_put_failure:
4756out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004757 fs_path_free(p);
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004758 return ret;
4759}
4760
Josef Bacik16e75492013-10-22 12:18:51 -04004761static int send_hole(struct send_ctx *sctx, u64 end)
4762{
4763 struct fs_path *p = NULL;
4764 u64 offset = sctx->cur_inode_last_extent;
4765 u64 len;
4766 int ret = 0;
4767
4768 p = fs_path_alloc();
4769 if (!p)
4770 return -ENOMEM;
Filipe Mananac715e152014-03-31 14:52:14 +01004771 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4772 if (ret < 0)
4773 goto tlv_put_failure;
Josef Bacik16e75492013-10-22 12:18:51 -04004774 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4775 while (offset < end) {
4776 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4777
4778 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4779 if (ret < 0)
4780 break;
Josef Bacik16e75492013-10-22 12:18:51 -04004781 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4782 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4783 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4784 ret = send_cmd(sctx);
4785 if (ret < 0)
4786 break;
4787 offset += len;
4788 }
4789tlv_put_failure:
4790 fs_path_free(p);
4791 return ret;
4792}
4793
Filipe Mananad906d492015-10-02 10:47:34 +01004794static int send_extent_data(struct send_ctx *sctx,
4795 const u64 offset,
4796 const u64 len)
4797{
4798 u64 sent = 0;
4799
4800 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4801 return send_update_extent(sctx, offset, len);
4802
4803 while (sent < len) {
4804 u64 size = len - sent;
4805 int ret;
4806
4807 if (size > BTRFS_SEND_READ_SIZE)
4808 size = BTRFS_SEND_READ_SIZE;
4809 ret = send_write(sctx, offset + sent, size);
4810 if (ret < 0)
4811 return ret;
4812 if (!ret)
4813 break;
4814 sent += ret;
4815 }
4816 return 0;
4817}
4818
4819static int clone_range(struct send_ctx *sctx,
4820 struct clone_root *clone_root,
4821 const u64 disk_byte,
4822 u64 data_offset,
4823 u64 offset,
4824 u64 len)
4825{
4826 struct btrfs_path *path;
4827 struct btrfs_key key;
4828 int ret;
4829
4830 path = alloc_path_for_send();
4831 if (!path)
4832 return -ENOMEM;
4833
4834 /*
4835 * We can't send a clone operation for the entire range if we find
4836 * extent items in the respective range in the source file that
4837 * refer to different extents or if we find holes.
4838 * So check for that and do a mix of clone and regular write/copy
4839 * operations if needed.
4840 *
4841 * Example:
4842 *
4843 * mkfs.btrfs -f /dev/sda
4844 * mount /dev/sda /mnt
4845 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
4846 * cp --reflink=always /mnt/foo /mnt/bar
4847 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
4848 * btrfs subvolume snapshot -r /mnt /mnt/snap
4849 *
4850 * If when we send the snapshot and we are processing file bar (which
4851 * has a higher inode number than foo) we blindly send a clone operation
4852 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
4853 * a file bar that matches the content of file foo - iow, doesn't match
4854 * the content from bar in the original filesystem.
4855 */
4856 key.objectid = clone_root->ino;
4857 key.type = BTRFS_EXTENT_DATA_KEY;
4858 key.offset = clone_root->offset;
4859 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
4860 if (ret < 0)
4861 goto out;
4862 if (ret > 0 && path->slots[0] > 0) {
4863 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4864 if (key.objectid == clone_root->ino &&
4865 key.type == BTRFS_EXTENT_DATA_KEY)
4866 path->slots[0]--;
4867 }
4868
4869 while (true) {
4870 struct extent_buffer *leaf = path->nodes[0];
4871 int slot = path->slots[0];
4872 struct btrfs_file_extent_item *ei;
4873 u8 type;
4874 u64 ext_len;
4875 u64 clone_len;
4876
4877 if (slot >= btrfs_header_nritems(leaf)) {
4878 ret = btrfs_next_leaf(clone_root->root, path);
4879 if (ret < 0)
4880 goto out;
4881 else if (ret > 0)
4882 break;
4883 continue;
4884 }
4885
4886 btrfs_item_key_to_cpu(leaf, &key, slot);
4887
4888 /*
4889 * We might have an implicit trailing hole (NO_HOLES feature
4890 * enabled). We deal with it after leaving this loop.
4891 */
4892 if (key.objectid != clone_root->ino ||
4893 key.type != BTRFS_EXTENT_DATA_KEY)
4894 break;
4895
4896 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4897 type = btrfs_file_extent_type(leaf, ei);
4898 if (type == BTRFS_FILE_EXTENT_INLINE) {
4899 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004900 ext_len = PAGE_ALIGN(ext_len);
Filipe Mananad906d492015-10-02 10:47:34 +01004901 } else {
4902 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
4903 }
4904
4905 if (key.offset + ext_len <= clone_root->offset)
4906 goto next;
4907
4908 if (key.offset > clone_root->offset) {
4909 /* Implicit hole, NO_HOLES feature enabled. */
4910 u64 hole_len = key.offset - clone_root->offset;
4911
4912 if (hole_len > len)
4913 hole_len = len;
4914 ret = send_extent_data(sctx, offset, hole_len);
4915 if (ret < 0)
4916 goto out;
4917
4918 len -= hole_len;
4919 if (len == 0)
4920 break;
4921 offset += hole_len;
4922 clone_root->offset += hole_len;
4923 data_offset += hole_len;
4924 }
4925
4926 if (key.offset >= clone_root->offset + len)
4927 break;
4928
4929 clone_len = min_t(u64, ext_len, len);
4930
4931 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
4932 btrfs_file_extent_offset(leaf, ei) == data_offset)
4933 ret = send_clone(sctx, offset, clone_len, clone_root);
4934 else
4935 ret = send_extent_data(sctx, offset, clone_len);
4936
4937 if (ret < 0)
4938 goto out;
4939
4940 len -= clone_len;
4941 if (len == 0)
4942 break;
4943 offset += clone_len;
4944 clone_root->offset += clone_len;
4945 data_offset += clone_len;
4946next:
4947 path->slots[0]++;
4948 }
4949
4950 if (len > 0)
4951 ret = send_extent_data(sctx, offset, len);
4952 else
4953 ret = 0;
4954out:
4955 btrfs_free_path(path);
4956 return ret;
4957}
4958
Alexander Block31db9f72012-07-25 23:19:24 +02004959static int send_write_or_clone(struct send_ctx *sctx,
4960 struct btrfs_path *path,
4961 struct btrfs_key *key,
4962 struct clone_root *clone_root)
4963{
4964 int ret = 0;
4965 struct btrfs_file_extent_item *ei;
4966 u64 offset = key->offset;
Alexander Block31db9f72012-07-25 23:19:24 +02004967 u64 len;
Alexander Block31db9f72012-07-25 23:19:24 +02004968 u8 type;
Filipe David Borba Manana28e5dd82014-01-12 02:26:28 +00004969 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
Alexander Block31db9f72012-07-25 23:19:24 +02004970
4971 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4972 struct btrfs_file_extent_item);
4973 type = btrfs_file_extent_type(path->nodes[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004974 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08004975 len = btrfs_file_extent_inline_len(path->nodes[0],
4976 path->slots[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004977 /*
4978 * it is possible the inline item won't cover the whole page,
4979 * but there may be items after this page. Make
4980 * sure to send the whole thing
4981 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004982 len = PAGE_ALIGN(len);
Chris Mason74dd17f2012-08-07 16:25:13 -04004983 } else {
Alexander Block31db9f72012-07-25 23:19:24 +02004984 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004985 }
Alexander Block31db9f72012-07-25 23:19:24 +02004986
4987 if (offset + len > sctx->cur_inode_size)
4988 len = sctx->cur_inode_size - offset;
4989 if (len == 0) {
4990 ret = 0;
4991 goto out;
4992 }
4993
Filipe David Borba Manana28e5dd82014-01-12 02:26:28 +00004994 if (clone_root && IS_ALIGNED(offset + len, bs)) {
Filipe Mananad906d492015-10-02 10:47:34 +01004995 u64 disk_byte;
4996 u64 data_offset;
4997
4998 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
4999 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5000 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5001 offset, len);
Mark Fashehcb95e7b2013-02-04 20:54:57 +00005002 } else {
Filipe Mananad906d492015-10-02 10:47:34 +01005003 ret = send_extent_data(sctx, offset, len);
Alexander Block31db9f72012-07-25 23:19:24 +02005004 }
Alexander Block31db9f72012-07-25 23:19:24 +02005005out:
5006 return ret;
5007}
5008
5009static int is_extent_unchanged(struct send_ctx *sctx,
5010 struct btrfs_path *left_path,
5011 struct btrfs_key *ekey)
5012{
5013 int ret = 0;
5014 struct btrfs_key key;
5015 struct btrfs_path *path = NULL;
5016 struct extent_buffer *eb;
5017 int slot;
5018 struct btrfs_key found_key;
5019 struct btrfs_file_extent_item *ei;
5020 u64 left_disknr;
5021 u64 right_disknr;
5022 u64 left_offset;
5023 u64 right_offset;
5024 u64 left_offset_fixed;
5025 u64 left_len;
5026 u64 right_len;
Chris Mason74dd17f2012-08-07 16:25:13 -04005027 u64 left_gen;
5028 u64 right_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02005029 u8 left_type;
5030 u8 right_type;
5031
5032 path = alloc_path_for_send();
5033 if (!path)
5034 return -ENOMEM;
5035
5036 eb = left_path->nodes[0];
5037 slot = left_path->slots[0];
Alexander Block31db9f72012-07-25 23:19:24 +02005038 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5039 left_type = btrfs_file_extent_type(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02005040
5041 if (left_type != BTRFS_FILE_EXTENT_REG) {
5042 ret = 0;
5043 goto out;
5044 }
Chris Mason74dd17f2012-08-07 16:25:13 -04005045 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5046 left_len = btrfs_file_extent_num_bytes(eb, ei);
5047 left_offset = btrfs_file_extent_offset(eb, ei);
5048 left_gen = btrfs_file_extent_generation(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02005049
5050 /*
5051 * Following comments will refer to these graphics. L is the left
5052 * extents which we are checking at the moment. 1-8 are the right
5053 * extents that we iterate.
5054 *
5055 * |-----L-----|
5056 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5057 *
5058 * |-----L-----|
5059 * |--1--|-2b-|...(same as above)
5060 *
5061 * Alternative situation. Happens on files where extents got split.
5062 * |-----L-----|
5063 * |-----------7-----------|-6-|
5064 *
5065 * Alternative situation. Happens on files which got larger.
5066 * |-----L-----|
5067 * |-8-|
5068 * Nothing follows after 8.
5069 */
5070
5071 key.objectid = ekey->objectid;
5072 key.type = BTRFS_EXTENT_DATA_KEY;
5073 key.offset = ekey->offset;
5074 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5075 if (ret < 0)
5076 goto out;
5077 if (ret) {
5078 ret = 0;
5079 goto out;
5080 }
5081
5082 /*
5083 * Handle special case where the right side has no extents at all.
5084 */
5085 eb = path->nodes[0];
5086 slot = path->slots[0];
5087 btrfs_item_key_to_cpu(eb, &found_key, slot);
5088 if (found_key.objectid != key.objectid ||
5089 found_key.type != key.type) {
Josef Bacik57cfd462013-08-20 15:55:39 -04005090 /* If we're a hole then just pretend nothing changed */
5091 ret = (left_disknr) ? 0 : 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005092 goto out;
5093 }
5094
5095 /*
5096 * We're now on 2a, 2b or 7.
5097 */
5098 key = found_key;
5099 while (key.offset < ekey->offset + left_len) {
5100 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5101 right_type = btrfs_file_extent_type(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02005102 if (right_type != BTRFS_FILE_EXTENT_REG) {
5103 ret = 0;
5104 goto out;
5105 }
5106
Josef Bacik007d31f2013-10-31 16:49:02 -04005107 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5108 right_len = btrfs_file_extent_num_bytes(eb, ei);
5109 right_offset = btrfs_file_extent_offset(eb, ei);
5110 right_gen = btrfs_file_extent_generation(eb, ei);
5111
Alexander Block31db9f72012-07-25 23:19:24 +02005112 /*
5113 * Are we at extent 8? If yes, we know the extent is changed.
5114 * This may only happen on the first iteration.
5115 */
Alexander Blockd8347fa2012-08-01 12:49:15 +02005116 if (found_key.offset + right_len <= ekey->offset) {
Josef Bacik57cfd462013-08-20 15:55:39 -04005117 /* If we're a hole just pretend nothing changed */
5118 ret = (left_disknr) ? 0 : 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005119 goto out;
5120 }
5121
5122 left_offset_fixed = left_offset;
5123 if (key.offset < ekey->offset) {
5124 /* Fix the right offset for 2a and 7. */
5125 right_offset += ekey->offset - key.offset;
5126 } else {
5127 /* Fix the left offset for all behind 2a and 2b */
5128 left_offset_fixed += key.offset - ekey->offset;
5129 }
5130
5131 /*
5132 * Check if we have the same extent.
5133 */
Alexander Block39540962012-08-01 12:46:05 +02005134 if (left_disknr != right_disknr ||
Chris Mason74dd17f2012-08-07 16:25:13 -04005135 left_offset_fixed != right_offset ||
5136 left_gen != right_gen) {
Alexander Block31db9f72012-07-25 23:19:24 +02005137 ret = 0;
5138 goto out;
5139 }
5140
5141 /*
5142 * Go to the next extent.
5143 */
5144 ret = btrfs_next_item(sctx->parent_root, path);
5145 if (ret < 0)
5146 goto out;
5147 if (!ret) {
5148 eb = path->nodes[0];
5149 slot = path->slots[0];
5150 btrfs_item_key_to_cpu(eb, &found_key, slot);
5151 }
5152 if (ret || found_key.objectid != key.objectid ||
5153 found_key.type != key.type) {
5154 key.offset += right_len;
5155 break;
Jan Schmidtadaa4b82013-03-21 14:30:23 +00005156 }
5157 if (found_key.offset != key.offset + right_len) {
5158 ret = 0;
5159 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005160 }
5161 key = found_key;
5162 }
5163
5164 /*
5165 * We're now behind the left extent (treat as unchanged) or at the end
5166 * of the right side (treat as changed).
5167 */
5168 if (key.offset >= ekey->offset + left_len)
5169 ret = 1;
5170 else
5171 ret = 0;
5172
5173
5174out:
5175 btrfs_free_path(path);
5176 return ret;
5177}
5178
Josef Bacik16e75492013-10-22 12:18:51 -04005179static int get_last_extent(struct send_ctx *sctx, u64 offset)
5180{
5181 struct btrfs_path *path;
5182 struct btrfs_root *root = sctx->send_root;
5183 struct btrfs_file_extent_item *fi;
5184 struct btrfs_key key;
5185 u64 extent_end;
5186 u8 type;
5187 int ret;
5188
5189 path = alloc_path_for_send();
5190 if (!path)
5191 return -ENOMEM;
5192
5193 sctx->cur_inode_last_extent = 0;
5194
5195 key.objectid = sctx->cur_ino;
5196 key.type = BTRFS_EXTENT_DATA_KEY;
5197 key.offset = offset;
5198 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5199 if (ret < 0)
5200 goto out;
5201 ret = 0;
5202 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5203 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5204 goto out;
5205
5206 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5207 struct btrfs_file_extent_item);
5208 type = btrfs_file_extent_type(path->nodes[0], fi);
5209 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08005210 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5211 path->slots[0], fi);
Josef Bacik16e75492013-10-22 12:18:51 -04005212 extent_end = ALIGN(key.offset + size,
5213 sctx->send_root->sectorsize);
5214 } else {
5215 extent_end = key.offset +
5216 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5217 }
5218 sctx->cur_inode_last_extent = extent_end;
5219out:
5220 btrfs_free_path(path);
5221 return ret;
5222}
5223
5224static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5225 struct btrfs_key *key)
5226{
5227 struct btrfs_file_extent_item *fi;
5228 u64 extent_end;
5229 u8 type;
5230 int ret = 0;
5231
5232 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5233 return 0;
5234
5235 if (sctx->cur_inode_last_extent == (u64)-1) {
5236 ret = get_last_extent(sctx, key->offset - 1);
5237 if (ret)
5238 return ret;
5239 }
5240
5241 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5242 struct btrfs_file_extent_item);
5243 type = btrfs_file_extent_type(path->nodes[0], fi);
5244 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08005245 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5246 path->slots[0], fi);
Josef Bacik16e75492013-10-22 12:18:51 -04005247 extent_end = ALIGN(key->offset + size,
5248 sctx->send_root->sectorsize);
5249 } else {
5250 extent_end = key->offset +
5251 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5252 }
Filipe David Borba Mananabf54f412014-01-28 01:38:06 +00005253
5254 if (path->slots[0] == 0 &&
5255 sctx->cur_inode_last_extent < key->offset) {
5256 /*
5257 * We might have skipped entire leafs that contained only
5258 * file extent items for our current inode. These leafs have
5259 * a generation number smaller (older) than the one in the
5260 * current leaf and the leaf our last extent came from, and
5261 * are located between these 2 leafs.
5262 */
5263 ret = get_last_extent(sctx, key->offset - 1);
5264 if (ret)
5265 return ret;
5266 }
5267
Josef Bacik16e75492013-10-22 12:18:51 -04005268 if (sctx->cur_inode_last_extent < key->offset)
5269 ret = send_hole(sctx, key->offset);
5270 sctx->cur_inode_last_extent = extent_end;
5271 return ret;
5272}
5273
Alexander Block31db9f72012-07-25 23:19:24 +02005274static int process_extent(struct send_ctx *sctx,
5275 struct btrfs_path *path,
5276 struct btrfs_key *key)
5277{
Alexander Block31db9f72012-07-25 23:19:24 +02005278 struct clone_root *found_clone = NULL;
Josef Bacik57cfd462013-08-20 15:55:39 -04005279 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02005280
5281 if (S_ISLNK(sctx->cur_inode_mode))
5282 return 0;
5283
5284 if (sctx->parent_root && !sctx->cur_inode_new) {
5285 ret = is_extent_unchanged(sctx, path, key);
5286 if (ret < 0)
5287 goto out;
5288 if (ret) {
5289 ret = 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005290 goto out_hole;
Alexander Block31db9f72012-07-25 23:19:24 +02005291 }
Josef Bacik57cfd462013-08-20 15:55:39 -04005292 } else {
5293 struct btrfs_file_extent_item *ei;
5294 u8 type;
5295
5296 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5297 struct btrfs_file_extent_item);
5298 type = btrfs_file_extent_type(path->nodes[0], ei);
5299 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5300 type == BTRFS_FILE_EXTENT_REG) {
5301 /*
5302 * The send spec does not have a prealloc command yet,
5303 * so just leave a hole for prealloc'ed extents until
5304 * we have enough commands queued up to justify rev'ing
5305 * the send spec.
5306 */
5307 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5308 ret = 0;
5309 goto out;
5310 }
5311
5312 /* Have a hole, just skip it. */
5313 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5314 ret = 0;
5315 goto out;
5316 }
5317 }
Alexander Block31db9f72012-07-25 23:19:24 +02005318 }
5319
5320 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5321 sctx->cur_inode_size, &found_clone);
5322 if (ret != -ENOENT && ret < 0)
5323 goto out;
5324
5325 ret = send_write_or_clone(sctx, path, key, found_clone);
Josef Bacik16e75492013-10-22 12:18:51 -04005326 if (ret)
5327 goto out;
5328out_hole:
5329 ret = maybe_send_hole(sctx, path, key);
Alexander Block31db9f72012-07-25 23:19:24 +02005330out:
5331 return ret;
5332}
5333
5334static int process_all_extents(struct send_ctx *sctx)
5335{
5336 int ret;
5337 struct btrfs_root *root;
5338 struct btrfs_path *path;
5339 struct btrfs_key key;
5340 struct btrfs_key found_key;
5341 struct extent_buffer *eb;
5342 int slot;
5343
5344 root = sctx->send_root;
5345 path = alloc_path_for_send();
5346 if (!path)
5347 return -ENOMEM;
5348
5349 key.objectid = sctx->cmp_key->objectid;
5350 key.type = BTRFS_EXTENT_DATA_KEY;
5351 key.offset = 0;
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005352 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5353 if (ret < 0)
5354 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005355
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005356 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02005357 eb = path->nodes[0];
5358 slot = path->slots[0];
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005359
5360 if (slot >= btrfs_header_nritems(eb)) {
5361 ret = btrfs_next_leaf(root, path);
5362 if (ret < 0) {
5363 goto out;
5364 } else if (ret > 0) {
5365 ret = 0;
5366 break;
5367 }
5368 continue;
5369 }
5370
Alexander Block31db9f72012-07-25 23:19:24 +02005371 btrfs_item_key_to_cpu(eb, &found_key, slot);
5372
5373 if (found_key.objectid != key.objectid ||
5374 found_key.type != key.type) {
5375 ret = 0;
5376 goto out;
5377 }
5378
5379 ret = process_extent(sctx, path, &found_key);
5380 if (ret < 0)
5381 goto out;
5382
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005383 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02005384 }
5385
5386out:
5387 btrfs_free_path(path);
5388 return ret;
5389}
5390
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005391static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5392 int *pending_move,
5393 int *refs_processed)
Alexander Block31db9f72012-07-25 23:19:24 +02005394{
5395 int ret = 0;
5396
5397 if (sctx->cur_ino == 0)
5398 goto out;
5399 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
Jan Schmidt96b5bd72012-10-15 08:30:45 +00005400 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
Alexander Block31db9f72012-07-25 23:19:24 +02005401 goto out;
5402 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5403 goto out;
5404
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005405 ret = process_recorded_refs(sctx, pending_move);
Alexander Blocke479d9b2012-07-28 16:09:35 +02005406 if (ret < 0)
5407 goto out;
5408
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005409 *refs_processed = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005410out:
5411 return ret;
5412}
5413
5414static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5415{
5416 int ret = 0;
5417 u64 left_mode;
5418 u64 left_uid;
5419 u64 left_gid;
5420 u64 right_mode;
5421 u64 right_uid;
5422 u64 right_gid;
5423 int need_chmod = 0;
5424 int need_chown = 0;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005425 int pending_move = 0;
5426 int refs_processed = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02005427
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005428 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5429 &refs_processed);
Alexander Block31db9f72012-07-25 23:19:24 +02005430 if (ret < 0)
5431 goto out;
5432
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005433 /*
5434 * We have processed the refs and thus need to advance send_progress.
5435 * Now, calls to get_cur_xxx will take the updated refs of the current
5436 * inode into account.
5437 *
5438 * On the other hand, if our current inode is a directory and couldn't
5439 * be moved/renamed because its parent was renamed/moved too and it has
5440 * a higher inode number, we can only move/rename our current inode
5441 * after we moved/renamed its parent. Therefore in this case operate on
5442 * the old path (pre move/rename) of our current inode, and the
5443 * move/rename will be performed later.
5444 */
5445 if (refs_processed && !pending_move)
5446 sctx->send_progress = sctx->cur_ino + 1;
5447
Alexander Block31db9f72012-07-25 23:19:24 +02005448 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5449 goto out;
5450 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5451 goto out;
5452
5453 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02005454 &left_mode, &left_uid, &left_gid, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02005455 if (ret < 0)
5456 goto out;
5457
Alex Lyakase2d044f2012-10-17 13:52:47 +00005458 if (!sctx->parent_root || sctx->cur_inode_new) {
5459 need_chown = 1;
5460 if (!S_ISLNK(sctx->cur_inode_mode))
Alexander Block31db9f72012-07-25 23:19:24 +02005461 need_chmod = 1;
Alex Lyakase2d044f2012-10-17 13:52:47 +00005462 } else {
5463 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5464 NULL, NULL, &right_mode, &right_uid,
5465 &right_gid, NULL);
5466 if (ret < 0)
5467 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005468
Alex Lyakase2d044f2012-10-17 13:52:47 +00005469 if (left_uid != right_uid || left_gid != right_gid)
5470 need_chown = 1;
5471 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5472 need_chmod = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005473 }
5474
5475 if (S_ISREG(sctx->cur_inode_mode)) {
Josef Bacik16e75492013-10-22 12:18:51 -04005476 if (need_send_hole(sctx)) {
Filipe Manana766b5e52014-03-30 23:02:53 +01005477 if (sctx->cur_inode_last_extent == (u64)-1 ||
5478 sctx->cur_inode_last_extent <
5479 sctx->cur_inode_size) {
Josef Bacik16e75492013-10-22 12:18:51 -04005480 ret = get_last_extent(sctx, (u64)-1);
5481 if (ret)
5482 goto out;
5483 }
5484 if (sctx->cur_inode_last_extent <
5485 sctx->cur_inode_size) {
5486 ret = send_hole(sctx, sctx->cur_inode_size);
5487 if (ret)
5488 goto out;
5489 }
5490 }
Alexander Block31db9f72012-07-25 23:19:24 +02005491 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5492 sctx->cur_inode_size);
5493 if (ret < 0)
5494 goto out;
5495 }
5496
5497 if (need_chown) {
5498 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5499 left_uid, left_gid);
5500 if (ret < 0)
5501 goto out;
5502 }
5503 if (need_chmod) {
5504 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5505 left_mode);
5506 if (ret < 0)
5507 goto out;
5508 }
5509
5510 /*
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005511 * If other directory inodes depended on our current directory
5512 * inode's move/rename, now do their move/rename operations.
Alexander Block31db9f72012-07-25 23:19:24 +02005513 */
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005514 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5515 ret = apply_children_dir_moves(sctx);
5516 if (ret)
5517 goto out;
Filipe Mananafcbd2152014-03-03 12:28:40 +00005518 /*
5519 * Need to send that every time, no matter if it actually
5520 * changed between the two trees as we have done changes to
5521 * the inode before. If our inode is a directory and it's
5522 * waiting to be moved/renamed, we will send its utimes when
5523 * it's moved/renamed, therefore we don't need to do it here.
5524 */
5525 sctx->send_progress = sctx->cur_ino + 1;
5526 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5527 if (ret < 0)
5528 goto out;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005529 }
5530
Alexander Block31db9f72012-07-25 23:19:24 +02005531out:
5532 return ret;
5533}
5534
5535static int changed_inode(struct send_ctx *sctx,
5536 enum btrfs_compare_tree_result result)
5537{
5538 int ret = 0;
5539 struct btrfs_key *key = sctx->cmp_key;
5540 struct btrfs_inode_item *left_ii = NULL;
5541 struct btrfs_inode_item *right_ii = NULL;
5542 u64 left_gen = 0;
5543 u64 right_gen = 0;
5544
Alexander Block31db9f72012-07-25 23:19:24 +02005545 sctx->cur_ino = key->objectid;
5546 sctx->cur_inode_new_gen = 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005547 sctx->cur_inode_last_extent = (u64)-1;
Alexander Blocke479d9b2012-07-28 16:09:35 +02005548
5549 /*
5550 * Set send_progress to current inode. This will tell all get_cur_xxx
5551 * functions that the current inode's refs are not updated yet. Later,
5552 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5553 */
Alexander Block31db9f72012-07-25 23:19:24 +02005554 sctx->send_progress = sctx->cur_ino;
5555
5556 if (result == BTRFS_COMPARE_TREE_NEW ||
5557 result == BTRFS_COMPARE_TREE_CHANGED) {
5558 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5559 sctx->left_path->slots[0],
5560 struct btrfs_inode_item);
5561 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5562 left_ii);
5563 } else {
5564 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5565 sctx->right_path->slots[0],
5566 struct btrfs_inode_item);
5567 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5568 right_ii);
5569 }
5570 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5571 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5572 sctx->right_path->slots[0],
5573 struct btrfs_inode_item);
5574
5575 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5576 right_ii);
Alexander Block6d85ed02012-08-01 14:48:59 +02005577
5578 /*
5579 * The cur_ino = root dir case is special here. We can't treat
5580 * the inode as deleted+reused because it would generate a
5581 * stream that tries to delete/mkdir the root dir.
5582 */
5583 if (left_gen != right_gen &&
5584 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
Alexander Block31db9f72012-07-25 23:19:24 +02005585 sctx->cur_inode_new_gen = 1;
5586 }
5587
5588 if (result == BTRFS_COMPARE_TREE_NEW) {
5589 sctx->cur_inode_gen = left_gen;
5590 sctx->cur_inode_new = 1;
5591 sctx->cur_inode_deleted = 0;
5592 sctx->cur_inode_size = btrfs_inode_size(
5593 sctx->left_path->nodes[0], left_ii);
5594 sctx->cur_inode_mode = btrfs_inode_mode(
5595 sctx->left_path->nodes[0], left_ii);
Liu Bo644d1942014-02-27 17:29:01 +08005596 sctx->cur_inode_rdev = btrfs_inode_rdev(
5597 sctx->left_path->nodes[0], left_ii);
Alexander Block31db9f72012-07-25 23:19:24 +02005598 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
Alexander Block1f4692d2012-07-28 10:42:24 +02005599 ret = send_create_inode_if_needed(sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02005600 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5601 sctx->cur_inode_gen = right_gen;
5602 sctx->cur_inode_new = 0;
5603 sctx->cur_inode_deleted = 1;
5604 sctx->cur_inode_size = btrfs_inode_size(
5605 sctx->right_path->nodes[0], right_ii);
5606 sctx->cur_inode_mode = btrfs_inode_mode(
5607 sctx->right_path->nodes[0], right_ii);
5608 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
Alexander Block766702e2012-07-28 14:11:31 +02005609 /*
5610 * We need to do some special handling in case the inode was
5611 * reported as changed with a changed generation number. This
5612 * means that the original inode was deleted and new inode
5613 * reused the same inum. So we have to treat the old inode as
5614 * deleted and the new one as new.
5615 */
Alexander Block31db9f72012-07-25 23:19:24 +02005616 if (sctx->cur_inode_new_gen) {
Alexander Block766702e2012-07-28 14:11:31 +02005617 /*
5618 * First, process the inode as if it was deleted.
5619 */
Alexander Block31db9f72012-07-25 23:19:24 +02005620 sctx->cur_inode_gen = right_gen;
5621 sctx->cur_inode_new = 0;
5622 sctx->cur_inode_deleted = 1;
5623 sctx->cur_inode_size = btrfs_inode_size(
5624 sctx->right_path->nodes[0], right_ii);
5625 sctx->cur_inode_mode = btrfs_inode_mode(
5626 sctx->right_path->nodes[0], right_ii);
5627 ret = process_all_refs(sctx,
5628 BTRFS_COMPARE_TREE_DELETED);
5629 if (ret < 0)
5630 goto out;
5631
Alexander Block766702e2012-07-28 14:11:31 +02005632 /*
5633 * Now process the inode as if it was new.
5634 */
Alexander Block31db9f72012-07-25 23:19:24 +02005635 sctx->cur_inode_gen = left_gen;
5636 sctx->cur_inode_new = 1;
5637 sctx->cur_inode_deleted = 0;
5638 sctx->cur_inode_size = btrfs_inode_size(
5639 sctx->left_path->nodes[0], left_ii);
5640 sctx->cur_inode_mode = btrfs_inode_mode(
5641 sctx->left_path->nodes[0], left_ii);
Liu Bo644d1942014-02-27 17:29:01 +08005642 sctx->cur_inode_rdev = btrfs_inode_rdev(
5643 sctx->left_path->nodes[0], left_ii);
Alexander Block1f4692d2012-07-28 10:42:24 +02005644 ret = send_create_inode_if_needed(sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02005645 if (ret < 0)
5646 goto out;
5647
5648 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5649 if (ret < 0)
5650 goto out;
Alexander Blocke479d9b2012-07-28 16:09:35 +02005651 /*
5652 * Advance send_progress now as we did not get into
5653 * process_recorded_refs_if_needed in the new_gen case.
5654 */
5655 sctx->send_progress = sctx->cur_ino + 1;
Alexander Block766702e2012-07-28 14:11:31 +02005656
5657 /*
5658 * Now process all extents and xattrs of the inode as if
5659 * they were all new.
5660 */
Alexander Block31db9f72012-07-25 23:19:24 +02005661 ret = process_all_extents(sctx);
5662 if (ret < 0)
5663 goto out;
5664 ret = process_all_new_xattrs(sctx);
5665 if (ret < 0)
5666 goto out;
5667 } else {
5668 sctx->cur_inode_gen = left_gen;
5669 sctx->cur_inode_new = 0;
5670 sctx->cur_inode_new_gen = 0;
5671 sctx->cur_inode_deleted = 0;
5672 sctx->cur_inode_size = btrfs_inode_size(
5673 sctx->left_path->nodes[0], left_ii);
5674 sctx->cur_inode_mode = btrfs_inode_mode(
5675 sctx->left_path->nodes[0], left_ii);
5676 }
5677 }
5678
5679out:
5680 return ret;
5681}
5682
Alexander Block766702e2012-07-28 14:11:31 +02005683/*
5684 * We have to process new refs before deleted refs, but compare_trees gives us
5685 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5686 * first and later process them in process_recorded_refs.
5687 * For the cur_inode_new_gen case, we skip recording completely because
5688 * changed_inode did already initiate processing of refs. The reason for this is
5689 * that in this case, compare_tree actually compares the refs of 2 different
5690 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5691 * refs of the right tree as deleted and all refs of the left tree as new.
5692 */
Alexander Block31db9f72012-07-25 23:19:24 +02005693static int changed_ref(struct send_ctx *sctx,
5694 enum btrfs_compare_tree_result result)
5695{
5696 int ret = 0;
5697
5698 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5699
5700 if (!sctx->cur_inode_new_gen &&
5701 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5702 if (result == BTRFS_COMPARE_TREE_NEW)
5703 ret = record_new_ref(sctx);
5704 else if (result == BTRFS_COMPARE_TREE_DELETED)
5705 ret = record_deleted_ref(sctx);
5706 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5707 ret = record_changed_ref(sctx);
5708 }
5709
5710 return ret;
5711}
5712
Alexander Block766702e2012-07-28 14:11:31 +02005713/*
5714 * Process new/deleted/changed xattrs. We skip processing in the
5715 * cur_inode_new_gen case because changed_inode did already initiate processing
5716 * of xattrs. The reason is the same as in changed_ref
5717 */
Alexander Block31db9f72012-07-25 23:19:24 +02005718static int changed_xattr(struct send_ctx *sctx,
5719 enum btrfs_compare_tree_result result)
5720{
5721 int ret = 0;
5722
5723 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5724
5725 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5726 if (result == BTRFS_COMPARE_TREE_NEW)
5727 ret = process_new_xattr(sctx);
5728 else if (result == BTRFS_COMPARE_TREE_DELETED)
5729 ret = process_deleted_xattr(sctx);
5730 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5731 ret = process_changed_xattr(sctx);
5732 }
5733
5734 return ret;
5735}
5736
Alexander Block766702e2012-07-28 14:11:31 +02005737/*
5738 * Process new/deleted/changed extents. We skip processing in the
5739 * cur_inode_new_gen case because changed_inode did already initiate processing
5740 * of extents. The reason is the same as in changed_ref
5741 */
Alexander Block31db9f72012-07-25 23:19:24 +02005742static int changed_extent(struct send_ctx *sctx,
5743 enum btrfs_compare_tree_result result)
5744{
5745 int ret = 0;
5746
5747 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5748
5749 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5750 if (result != BTRFS_COMPARE_TREE_DELETED)
5751 ret = process_extent(sctx, sctx->left_path,
5752 sctx->cmp_key);
5753 }
5754
5755 return ret;
5756}
5757
Josef Bacikba5e8f22013-08-16 16:52:55 -04005758static int dir_changed(struct send_ctx *sctx, u64 dir)
5759{
5760 u64 orig_gen, new_gen;
5761 int ret;
5762
5763 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5764 NULL, NULL);
5765 if (ret)
5766 return ret;
5767
5768 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5769 NULL, NULL, NULL);
5770 if (ret)
5771 return ret;
5772
5773 return (orig_gen != new_gen) ? 1 : 0;
5774}
5775
5776static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5777 struct btrfs_key *key)
5778{
5779 struct btrfs_inode_extref *extref;
5780 struct extent_buffer *leaf;
5781 u64 dirid = 0, last_dirid = 0;
5782 unsigned long ptr;
5783 u32 item_size;
5784 u32 cur_offset = 0;
5785 int ref_name_len;
5786 int ret = 0;
5787
5788 /* Easy case, just check this one dirid */
5789 if (key->type == BTRFS_INODE_REF_KEY) {
5790 dirid = key->offset;
5791
5792 ret = dir_changed(sctx, dirid);
5793 goto out;
5794 }
5795
5796 leaf = path->nodes[0];
5797 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5798 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5799 while (cur_offset < item_size) {
5800 extref = (struct btrfs_inode_extref *)(ptr +
5801 cur_offset);
5802 dirid = btrfs_inode_extref_parent(leaf, extref);
5803 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5804 cur_offset += ref_name_len + sizeof(*extref);
5805 if (dirid == last_dirid)
5806 continue;
5807 ret = dir_changed(sctx, dirid);
5808 if (ret)
5809 break;
5810 last_dirid = dirid;
5811 }
5812out:
5813 return ret;
5814}
5815
Alexander Block766702e2012-07-28 14:11:31 +02005816/*
5817 * Updates compare related fields in sctx and simply forwards to the actual
5818 * changed_xxx functions.
5819 */
Alexander Block31db9f72012-07-25 23:19:24 +02005820static int changed_cb(struct btrfs_root *left_root,
5821 struct btrfs_root *right_root,
5822 struct btrfs_path *left_path,
5823 struct btrfs_path *right_path,
5824 struct btrfs_key *key,
5825 enum btrfs_compare_tree_result result,
5826 void *ctx)
5827{
5828 int ret = 0;
5829 struct send_ctx *sctx = ctx;
5830
Josef Bacikba5e8f22013-08-16 16:52:55 -04005831 if (result == BTRFS_COMPARE_TREE_SAME) {
Josef Bacik16e75492013-10-22 12:18:51 -04005832 if (key->type == BTRFS_INODE_REF_KEY ||
5833 key->type == BTRFS_INODE_EXTREF_KEY) {
5834 ret = compare_refs(sctx, left_path, key);
5835 if (!ret)
5836 return 0;
5837 if (ret < 0)
5838 return ret;
5839 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5840 return maybe_send_hole(sctx, left_path, key);
5841 } else {
Josef Bacikba5e8f22013-08-16 16:52:55 -04005842 return 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005843 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04005844 result = BTRFS_COMPARE_TREE_CHANGED;
5845 ret = 0;
5846 }
5847
Alexander Block31db9f72012-07-25 23:19:24 +02005848 sctx->left_path = left_path;
5849 sctx->right_path = right_path;
5850 sctx->cmp_key = key;
5851
5852 ret = finish_inode_if_needed(sctx, 0);
5853 if (ret < 0)
5854 goto out;
5855
Alexander Block2981e222012-08-01 14:47:03 +02005856 /* Ignore non-FS objects */
5857 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5858 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5859 goto out;
5860
Alexander Block31db9f72012-07-25 23:19:24 +02005861 if (key->type == BTRFS_INODE_ITEM_KEY)
5862 ret = changed_inode(sctx, result);
Jan Schmidt96b5bd72012-10-15 08:30:45 +00005863 else if (key->type == BTRFS_INODE_REF_KEY ||
5864 key->type == BTRFS_INODE_EXTREF_KEY)
Alexander Block31db9f72012-07-25 23:19:24 +02005865 ret = changed_ref(sctx, result);
5866 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5867 ret = changed_xattr(sctx, result);
5868 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5869 ret = changed_extent(sctx, result);
5870
5871out:
5872 return ret;
5873}
5874
5875static int full_send_tree(struct send_ctx *sctx)
5876{
5877 int ret;
Alexander Block31db9f72012-07-25 23:19:24 +02005878 struct btrfs_root *send_root = sctx->send_root;
5879 struct btrfs_key key;
5880 struct btrfs_key found_key;
5881 struct btrfs_path *path;
5882 struct extent_buffer *eb;
5883 int slot;
Alexander Block31db9f72012-07-25 23:19:24 +02005884
5885 path = alloc_path_for_send();
5886 if (!path)
5887 return -ENOMEM;
5888
Alexander Block31db9f72012-07-25 23:19:24 +02005889 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5890 key.type = BTRFS_INODE_ITEM_KEY;
5891 key.offset = 0;
5892
Alexander Block31db9f72012-07-25 23:19:24 +02005893 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5894 if (ret < 0)
5895 goto out;
5896 if (ret)
5897 goto out_finish;
5898
5899 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02005900 eb = path->nodes[0];
5901 slot = path->slots[0];
5902 btrfs_item_key_to_cpu(eb, &found_key, slot);
5903
5904 ret = changed_cb(send_root, NULL, path, NULL,
5905 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5906 if (ret < 0)
5907 goto out;
5908
5909 key.objectid = found_key.objectid;
5910 key.type = found_key.type;
5911 key.offset = found_key.offset + 1;
5912
5913 ret = btrfs_next_item(send_root, path);
5914 if (ret < 0)
5915 goto out;
5916 if (ret) {
5917 ret = 0;
5918 break;
5919 }
5920 }
5921
5922out_finish:
5923 ret = finish_inode_if_needed(sctx, 1);
5924
5925out:
5926 btrfs_free_path(path);
Alexander Block31db9f72012-07-25 23:19:24 +02005927 return ret;
5928}
5929
5930static int send_subvol(struct send_ctx *sctx)
5931{
5932 int ret;
5933
Stefan Behrensc2c71322013-04-10 17:10:52 +00005934 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5935 ret = send_header(sctx);
5936 if (ret < 0)
5937 goto out;
5938 }
Alexander Block31db9f72012-07-25 23:19:24 +02005939
5940 ret = send_subvol_begin(sctx);
5941 if (ret < 0)
5942 goto out;
5943
5944 if (sctx->parent_root) {
5945 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5946 changed_cb, sctx);
5947 if (ret < 0)
5948 goto out;
5949 ret = finish_inode_if_needed(sctx, 1);
5950 if (ret < 0)
5951 goto out;
5952 } else {
5953 ret = full_send_tree(sctx);
5954 if (ret < 0)
5955 goto out;
5956 }
5957
5958out:
Alexander Block31db9f72012-07-25 23:19:24 +02005959 free_recorded_refs(sctx);
5960 return ret;
5961}
5962
Filipe Mananae5fa8f82014-10-21 11:11:41 +01005963/*
5964 * If orphan cleanup did remove any orphans from a root, it means the tree
5965 * was modified and therefore the commit root is not the same as the current
5966 * root anymore. This is a problem, because send uses the commit root and
5967 * therefore can see inode items that don't exist in the current root anymore,
5968 * and for example make calls to btrfs_iget, which will do tree lookups based
5969 * on the current root and not on the commit root. Those lookups will fail,
5970 * returning a -ESTALE error, and making send fail with that error. So make
5971 * sure a send does not see any orphans we have just removed, and that it will
5972 * see the same inodes regardless of whether a transaction commit happened
5973 * before it started (meaning that the commit root will be the same as the
5974 * current root) or not.
5975 */
5976static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
5977{
5978 int i;
5979 struct btrfs_trans_handle *trans = NULL;
5980
5981again:
5982 if (sctx->parent_root &&
5983 sctx->parent_root->node != sctx->parent_root->commit_root)
5984 goto commit_trans;
5985
5986 for (i = 0; i < sctx->clone_roots_cnt; i++)
5987 if (sctx->clone_roots[i].root->node !=
5988 sctx->clone_roots[i].root->commit_root)
5989 goto commit_trans;
5990
5991 if (trans)
5992 return btrfs_end_transaction(trans, sctx->send_root);
5993
5994 return 0;
5995
5996commit_trans:
5997 /* Use any root, all fs roots will get their commit roots updated. */
5998 if (!trans) {
5999 trans = btrfs_join_transaction(sctx->send_root);
6000 if (IS_ERR(trans))
6001 return PTR_ERR(trans);
6002 goto again;
6003 }
6004
6005 return btrfs_commit_transaction(trans, sctx->send_root);
6006}
6007
David Sterba66ef7d62013-12-17 15:07:20 +01006008static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6009{
6010 spin_lock(&root->root_item_lock);
6011 root->send_in_progress--;
6012 /*
6013 * Not much left to do, we don't know why it's unbalanced and
6014 * can't blindly reset it to 0.
6015 */
6016 if (root->send_in_progress < 0)
6017 btrfs_err(root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02006018 "send_in_progres unbalanced %d root %llu",
David Sterba66ef7d62013-12-17 15:07:20 +01006019 root->send_in_progress, root->root_key.objectid);
6020 spin_unlock(&root->root_item_lock);
6021}
6022
Alexander Block31db9f72012-07-25 23:19:24 +02006023long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
6024{
6025 int ret = 0;
6026 struct btrfs_root *send_root;
6027 struct btrfs_root *clone_root;
6028 struct btrfs_fs_info *fs_info;
6029 struct btrfs_ioctl_send_args *arg = NULL;
6030 struct btrfs_key key;
Alexander Block31db9f72012-07-25 23:19:24 +02006031 struct send_ctx *sctx = NULL;
6032 u32 i;
6033 u64 *clone_sources_tmp = NULL;
David Sterba2c686532013-12-16 17:34:17 +01006034 int clone_sources_to_rollback = 0;
David Sterbae55d1152016-04-11 18:52:02 +02006035 unsigned alloc_size;
Wang Shilong896c14f2014-01-07 17:25:18 +08006036 int sort_clone_roots = 0;
Wang Shilong18f687d2014-01-07 17:25:19 +08006037 int index;
Alexander Block31db9f72012-07-25 23:19:24 +02006038
6039 if (!capable(CAP_SYS_ADMIN))
6040 return -EPERM;
6041
Al Viro496ad9a2013-01-23 17:07:38 -05006042 send_root = BTRFS_I(file_inode(mnt_file))->root;
Alexander Block31db9f72012-07-25 23:19:24 +02006043 fs_info = send_root->fs_info;
6044
Josef Bacik139f8072013-05-20 11:26:50 -04006045 /*
David Sterba2c686532013-12-16 17:34:17 +01006046 * The subvolume must remain read-only during send, protect against
David Sterba521e0542014-04-15 16:41:44 +02006047 * making it RW. This also protects against deletion.
David Sterba2c686532013-12-16 17:34:17 +01006048 */
6049 spin_lock(&send_root->root_item_lock);
6050 send_root->send_in_progress++;
6051 spin_unlock(&send_root->root_item_lock);
6052
6053 /*
Josef Bacik139f8072013-05-20 11:26:50 -04006054 * This is done when we lookup the root, it should already be complete
6055 * by the time we get here.
6056 */
6057 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6058
6059 /*
David Sterba2c686532013-12-16 17:34:17 +01006060 * Userspace tools do the checks and warn the user if it's
6061 * not RO.
6062 */
6063 if (!btrfs_root_readonly(send_root)) {
6064 ret = -EPERM;
6065 goto out;
6066 }
6067
Alexander Block31db9f72012-07-25 23:19:24 +02006068 arg = memdup_user(arg_, sizeof(*arg));
6069 if (IS_ERR(arg)) {
6070 ret = PTR_ERR(arg);
6071 arg = NULL;
6072 goto out;
6073 }
6074
Dan Carpenterf5ecec32016-04-13 09:40:59 +03006075 if (arg->clone_sources_count >
6076 ULLONG_MAX / sizeof(*arg->clone_sources)) {
6077 ret = -EINVAL;
6078 goto out;
6079 }
6080
Alexander Block31db9f72012-07-25 23:19:24 +02006081 if (!access_ok(VERIFY_READ, arg->clone_sources,
Dan Carpenter700ff4f2013-01-10 03:57:25 -05006082 sizeof(*arg->clone_sources) *
6083 arg->clone_sources_count)) {
Alexander Block31db9f72012-07-25 23:19:24 +02006084 ret = -EFAULT;
6085 goto out;
6086 }
6087
Stefan Behrensc2c71322013-04-10 17:10:52 +00006088 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
Mark Fashehcb95e7b2013-02-04 20:54:57 +00006089 ret = -EINVAL;
6090 goto out;
6091 }
6092
David Sterbae780b0d2016-01-18 18:42:13 +01006093 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02006094 if (!sctx) {
6095 ret = -ENOMEM;
6096 goto out;
6097 }
6098
6099 INIT_LIST_HEAD(&sctx->new_refs);
6100 INIT_LIST_HEAD(&sctx->deleted_refs);
David Sterbae780b0d2016-01-18 18:42:13 +01006101 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02006102 INIT_LIST_HEAD(&sctx->name_cache_list);
6103
Mark Fashehcb95e7b2013-02-04 20:54:57 +00006104 sctx->flags = arg->flags;
6105
Alexander Block31db9f72012-07-25 23:19:24 +02006106 sctx->send_filp = fget(arg->send_fd);
Tsutomu Itohecc7ada2013-04-19 01:04:46 +00006107 if (!sctx->send_filp) {
6108 ret = -EBADF;
Alexander Block31db9f72012-07-25 23:19:24 +02006109 goto out;
6110 }
6111
Alexander Block31db9f72012-07-25 23:19:24 +02006112 sctx->send_root = send_root;
David Sterba521e0542014-04-15 16:41:44 +02006113 /*
6114 * Unlikely but possible, if the subvolume is marked for deletion but
6115 * is slow to remove the directory entry, send can still be started
6116 */
6117 if (btrfs_root_dead(sctx->send_root)) {
6118 ret = -EPERM;
6119 goto out;
6120 }
6121
Alexander Block31db9f72012-07-25 23:19:24 +02006122 sctx->clone_roots_cnt = arg->clone_sources_count;
6123
6124 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
David Sterba6ff48ce2016-04-11 18:40:08 +02006125 sctx->send_buf = kmalloc(sctx->send_max_size, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006126 if (!sctx->send_buf) {
David Sterba6ff48ce2016-04-11 18:40:08 +02006127 sctx->send_buf = vmalloc(sctx->send_max_size);
6128 if (!sctx->send_buf) {
6129 ret = -ENOMEM;
6130 goto out;
6131 }
Alexander Block31db9f72012-07-25 23:19:24 +02006132 }
6133
David Sterbaeb5b75f2016-04-11 18:40:08 +02006134 sctx->read_buf = kmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006135 if (!sctx->read_buf) {
David Sterbaeb5b75f2016-04-11 18:40:08 +02006136 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
6137 if (!sctx->read_buf) {
6138 ret = -ENOMEM;
6139 goto out;
6140 }
Alexander Block31db9f72012-07-25 23:19:24 +02006141 }
6142
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006143 sctx->pending_dir_moves = RB_ROOT;
6144 sctx->waiting_dir_moves = RB_ROOT;
Filipe Manana9dc44212014-02-19 14:31:44 +00006145 sctx->orphan_dirs = RB_ROOT;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006146
David Sterbae55d1152016-04-11 18:52:02 +02006147 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6148
David Sterbac03d01f2016-04-11 18:40:08 +02006149 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006150 if (!sctx->clone_roots) {
David Sterbac03d01f2016-04-11 18:40:08 +02006151 sctx->clone_roots = vzalloc(alloc_size);
6152 if (!sctx->clone_roots) {
6153 ret = -ENOMEM;
6154 goto out;
6155 }
Alexander Block31db9f72012-07-25 23:19:24 +02006156 }
6157
David Sterbae55d1152016-04-11 18:52:02 +02006158 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6159
Alexander Block31db9f72012-07-25 23:19:24 +02006160 if (arg->clone_sources_count) {
David Sterba2f913062016-04-11 18:40:08 +02006161 clone_sources_tmp = kmalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006162 if (!clone_sources_tmp) {
David Sterba2f913062016-04-11 18:40:08 +02006163 clone_sources_tmp = vmalloc(alloc_size);
6164 if (!clone_sources_tmp) {
6165 ret = -ENOMEM;
6166 goto out;
6167 }
Alexander Block31db9f72012-07-25 23:19:24 +02006168 }
6169
6170 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
David Sterbae55d1152016-04-11 18:52:02 +02006171 alloc_size);
Alexander Block31db9f72012-07-25 23:19:24 +02006172 if (ret) {
6173 ret = -EFAULT;
6174 goto out;
6175 }
6176
6177 for (i = 0; i < arg->clone_sources_count; i++) {
6178 key.objectid = clone_sources_tmp[i];
6179 key.type = BTRFS_ROOT_ITEM_KEY;
6180 key.offset = (u64)-1;
Wang Shilong18f687d2014-01-07 17:25:19 +08006181
6182 index = srcu_read_lock(&fs_info->subvol_srcu);
6183
Alexander Block31db9f72012-07-25 23:19:24 +02006184 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
Alexander Block31db9f72012-07-25 23:19:24 +02006185 if (IS_ERR(clone_root)) {
Wang Shilong18f687d2014-01-07 17:25:19 +08006186 srcu_read_unlock(&fs_info->subvol_srcu, index);
Alexander Block31db9f72012-07-25 23:19:24 +02006187 ret = PTR_ERR(clone_root);
6188 goto out;
6189 }
David Sterba2c686532013-12-16 17:34:17 +01006190 spin_lock(&clone_root->root_item_lock);
Filipe Manana5cc2b172015-03-02 20:53:52 +00006191 if (!btrfs_root_readonly(clone_root) ||
6192 btrfs_root_dead(clone_root)) {
David Sterba2c686532013-12-16 17:34:17 +01006193 spin_unlock(&clone_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006194 srcu_read_unlock(&fs_info->subvol_srcu, index);
David Sterba2c686532013-12-16 17:34:17 +01006195 ret = -EPERM;
6196 goto out;
6197 }
Filipe Manana2f1f4652015-03-02 20:53:53 +00006198 clone_root->send_in_progress++;
David Sterba2c686532013-12-16 17:34:17 +01006199 spin_unlock(&clone_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006200 srcu_read_unlock(&fs_info->subvol_srcu, index);
6201
Alexander Block31db9f72012-07-25 23:19:24 +02006202 sctx->clone_roots[i].root = clone_root;
Filipe Manana2f1f4652015-03-02 20:53:53 +00006203 clone_sources_to_rollback = i + 1;
Alexander Block31db9f72012-07-25 23:19:24 +02006204 }
David Sterba2f913062016-04-11 18:40:08 +02006205 kvfree(clone_sources_tmp);
Alexander Block31db9f72012-07-25 23:19:24 +02006206 clone_sources_tmp = NULL;
6207 }
6208
6209 if (arg->parent_root) {
6210 key.objectid = arg->parent_root;
6211 key.type = BTRFS_ROOT_ITEM_KEY;
6212 key.offset = (u64)-1;
Wang Shilong18f687d2014-01-07 17:25:19 +08006213
6214 index = srcu_read_lock(&fs_info->subvol_srcu);
6215
Alexander Block31db9f72012-07-25 23:19:24 +02006216 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
Stefan Behrensb1b19592013-05-13 14:42:57 +00006217 if (IS_ERR(sctx->parent_root)) {
Wang Shilong18f687d2014-01-07 17:25:19 +08006218 srcu_read_unlock(&fs_info->subvol_srcu, index);
Stefan Behrensb1b19592013-05-13 14:42:57 +00006219 ret = PTR_ERR(sctx->parent_root);
Alexander Block31db9f72012-07-25 23:19:24 +02006220 goto out;
6221 }
Wang Shilong18f687d2014-01-07 17:25:19 +08006222
David Sterba2c686532013-12-16 17:34:17 +01006223 spin_lock(&sctx->parent_root->root_item_lock);
6224 sctx->parent_root->send_in_progress++;
David Sterba521e0542014-04-15 16:41:44 +02006225 if (!btrfs_root_readonly(sctx->parent_root) ||
6226 btrfs_root_dead(sctx->parent_root)) {
David Sterba2c686532013-12-16 17:34:17 +01006227 spin_unlock(&sctx->parent_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006228 srcu_read_unlock(&fs_info->subvol_srcu, index);
David Sterba2c686532013-12-16 17:34:17 +01006229 ret = -EPERM;
6230 goto out;
6231 }
6232 spin_unlock(&sctx->parent_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006233
6234 srcu_read_unlock(&fs_info->subvol_srcu, index);
Alexander Block31db9f72012-07-25 23:19:24 +02006235 }
6236
6237 /*
6238 * Clones from send_root are allowed, but only if the clone source
6239 * is behind the current send position. This is checked while searching
6240 * for possible clone sources.
6241 */
6242 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6243
6244 /* We do a bsearch later */
6245 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6246 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6247 NULL);
Wang Shilong896c14f2014-01-07 17:25:18 +08006248 sort_clone_roots = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02006249
Filipe Mananae5fa8f82014-10-21 11:11:41 +01006250 ret = ensure_commit_roots_uptodate(sctx);
6251 if (ret)
6252 goto out;
6253
David Sterba2755a0d2014-07-31 00:43:18 +02006254 current->journal_info = BTRFS_SEND_TRANS_STUB;
Alexander Block31db9f72012-07-25 23:19:24 +02006255 ret = send_subvol(sctx);
Josef Bacika26e8c92014-03-28 17:07:27 -04006256 current->journal_info = NULL;
Alexander Block31db9f72012-07-25 23:19:24 +02006257 if (ret < 0)
6258 goto out;
6259
Stefan Behrensc2c71322013-04-10 17:10:52 +00006260 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6261 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6262 if (ret < 0)
6263 goto out;
6264 ret = send_cmd(sctx);
6265 if (ret < 0)
6266 goto out;
6267 }
Alexander Block31db9f72012-07-25 23:19:24 +02006268
6269out:
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006270 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6271 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6272 struct rb_node *n;
6273 struct pending_dir_move *pm;
6274
6275 n = rb_first(&sctx->pending_dir_moves);
6276 pm = rb_entry(n, struct pending_dir_move, node);
6277 while (!list_empty(&pm->list)) {
6278 struct pending_dir_move *pm2;
6279
6280 pm2 = list_first_entry(&pm->list,
6281 struct pending_dir_move, list);
6282 free_pending_move(sctx, pm2);
6283 }
6284 free_pending_move(sctx, pm);
6285 }
6286
6287 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6288 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6289 struct rb_node *n;
6290 struct waiting_dir_move *dm;
6291
6292 n = rb_first(&sctx->waiting_dir_moves);
6293 dm = rb_entry(n, struct waiting_dir_move, node);
6294 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6295 kfree(dm);
6296 }
6297
Filipe Manana9dc44212014-02-19 14:31:44 +00006298 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6299 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6300 struct rb_node *n;
6301 struct orphan_dir_info *odi;
6302
6303 n = rb_first(&sctx->orphan_dirs);
6304 odi = rb_entry(n, struct orphan_dir_info, node);
6305 free_orphan_dir_info(sctx, odi);
6306 }
6307
Wang Shilong896c14f2014-01-07 17:25:18 +08006308 if (sort_clone_roots) {
6309 for (i = 0; i < sctx->clone_roots_cnt; i++)
6310 btrfs_root_dec_send_in_progress(
6311 sctx->clone_roots[i].root);
6312 } else {
6313 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6314 btrfs_root_dec_send_in_progress(
6315 sctx->clone_roots[i].root);
6316
6317 btrfs_root_dec_send_in_progress(send_root);
6318 }
David Sterba66ef7d62013-12-17 15:07:20 +01006319 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6320 btrfs_root_dec_send_in_progress(sctx->parent_root);
David Sterba2c686532013-12-16 17:34:17 +01006321
Alexander Block31db9f72012-07-25 23:19:24 +02006322 kfree(arg);
David Sterba2f913062016-04-11 18:40:08 +02006323 kvfree(clone_sources_tmp);
Alexander Block31db9f72012-07-25 23:19:24 +02006324
6325 if (sctx) {
6326 if (sctx->send_filp)
6327 fput(sctx->send_filp);
6328
David Sterbac03d01f2016-04-11 18:40:08 +02006329 kvfree(sctx->clone_roots);
David Sterba6ff48ce2016-04-11 18:40:08 +02006330 kvfree(sctx->send_buf);
David Sterbaeb5b75f2016-04-11 18:40:08 +02006331 kvfree(sctx->read_buf);
Alexander Block31db9f72012-07-25 23:19:24 +02006332
6333 name_cache_free(sctx);
6334
6335 kfree(sctx);
6336 }
6337
6338 return ret;
6339}