blob: 8b653967b16388ac6d8c432af8765b1736b9400b [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 Bo2131bcd32014-03-05 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) {
Robbie Ko443f9d22015-06-22 17:08:45 +08002950 struct orphan_dir_info *odi;
2951
2952 odi = get_orphan_dir_info(sctx, dir);
2953 free_orphan_dir_info(sctx, odi);
Alexander Block31db9f72012-07-25 23:19:24 +02002954 ret = 0;
2955 goto out;
2956 }
2957
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00002958 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02002959 }
2960
2961 ret = 1;
2962
2963out:
2964 btrfs_free_path(path);
2965 return ret;
2966}
2967
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002968static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2969{
Filipe Manana9dc44212014-02-19 14:31:44 +00002970 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002971
Filipe Manana9dc44212014-02-19 14:31:44 +00002972 return entry != NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002973}
2974
Filipe Manana8b191a62015-04-09 14:09:14 +01002975static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002976{
2977 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2978 struct rb_node *parent = NULL;
2979 struct waiting_dir_move *entry, *dm;
2980
David Sterbae780b0d2016-01-18 18:42:13 +01002981 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002982 if (!dm)
2983 return -ENOMEM;
2984 dm->ino = ino;
Filipe Manana9dc44212014-02-19 14:31:44 +00002985 dm->rmdir_ino = 0;
Filipe Manana8b191a62015-04-09 14:09:14 +01002986 dm->orphanized = orphanized;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00002987
2988 while (*p) {
2989 parent = *p;
2990 entry = rb_entry(parent, struct waiting_dir_move, node);
2991 if (ino < entry->ino) {
2992 p = &(*p)->rb_left;
2993 } else if (ino > entry->ino) {
2994 p = &(*p)->rb_right;
2995 } else {
2996 kfree(dm);
2997 return -EEXIST;
2998 }
2999 }
3000
3001 rb_link_node(&dm->node, parent, p);
3002 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3003 return 0;
3004}
3005
Filipe Manana9dc44212014-02-19 14:31:44 +00003006static struct waiting_dir_move *
3007get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003008{
3009 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3010 struct waiting_dir_move *entry;
3011
3012 while (n) {
3013 entry = rb_entry(n, struct waiting_dir_move, node);
Filipe Manana9dc44212014-02-19 14:31:44 +00003014 if (ino < entry->ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003015 n = n->rb_left;
Filipe Manana9dc44212014-02-19 14:31:44 +00003016 else if (ino > entry->ino)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003017 n = n->rb_right;
Filipe Manana9dc44212014-02-19 14:31:44 +00003018 else
3019 return entry;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003020 }
Filipe Manana9dc44212014-02-19 14:31:44 +00003021 return NULL;
3022}
3023
3024static void free_waiting_dir_move(struct send_ctx *sctx,
3025 struct waiting_dir_move *dm)
3026{
3027 if (!dm)
3028 return;
3029 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3030 kfree(dm);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003031}
3032
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003033static int add_pending_dir_move(struct send_ctx *sctx,
3034 u64 ino,
3035 u64 ino_gen,
Filipe Mananaf9594922014-03-27 20:14:01 +00003036 u64 parent_ino,
3037 struct list_head *new_refs,
Filipe Manana84471e22015-02-28 22:29:22 +00003038 struct list_head *deleted_refs,
3039 const bool is_orphan)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003040{
3041 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3042 struct rb_node *parent = NULL;
Chris Mason73b802f2014-03-21 15:30:44 -07003043 struct pending_dir_move *entry = NULL, *pm;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003044 struct recorded_ref *cur;
3045 int exists = 0;
3046 int ret;
3047
David Sterbae780b0d2016-01-18 18:42:13 +01003048 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003049 if (!pm)
3050 return -ENOMEM;
3051 pm->parent_ino = parent_ino;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003052 pm->ino = ino;
3053 pm->gen = ino_gen;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003054 INIT_LIST_HEAD(&pm->list);
3055 INIT_LIST_HEAD(&pm->update_refs);
3056 RB_CLEAR_NODE(&pm->node);
3057
3058 while (*p) {
3059 parent = *p;
3060 entry = rb_entry(parent, struct pending_dir_move, node);
3061 if (parent_ino < entry->parent_ino) {
3062 p = &(*p)->rb_left;
3063 } else if (parent_ino > entry->parent_ino) {
3064 p = &(*p)->rb_right;
3065 } else {
3066 exists = 1;
3067 break;
3068 }
3069 }
3070
Filipe Mananaf9594922014-03-27 20:14:01 +00003071 list_for_each_entry(cur, deleted_refs, list) {
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003072 ret = dup_ref(cur, &pm->update_refs);
3073 if (ret < 0)
3074 goto out;
3075 }
Filipe Mananaf9594922014-03-27 20:14:01 +00003076 list_for_each_entry(cur, new_refs, list) {
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003077 ret = dup_ref(cur, &pm->update_refs);
3078 if (ret < 0)
3079 goto out;
3080 }
3081
Filipe Manana8b191a62015-04-09 14:09:14 +01003082 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003083 if (ret)
3084 goto out;
3085
3086 if (exists) {
3087 list_add_tail(&pm->list, &entry->list);
3088 } else {
3089 rb_link_node(&pm->node, parent, p);
3090 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3091 }
3092 ret = 0;
3093out:
3094 if (ret) {
3095 __free_recorded_refs(&pm->update_refs);
3096 kfree(pm);
3097 }
3098 return ret;
3099}
3100
3101static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3102 u64 parent_ino)
3103{
3104 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3105 struct pending_dir_move *entry;
3106
3107 while (n) {
3108 entry = rb_entry(n, struct pending_dir_move, node);
3109 if (parent_ino < entry->parent_ino)
3110 n = n->rb_left;
3111 else if (parent_ino > entry->parent_ino)
3112 n = n->rb_right;
3113 else
3114 return entry;
3115 }
3116 return NULL;
3117}
3118
Robbie Ko801bec32015-06-23 18:39:46 +08003119static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3120 u64 ino, u64 gen, u64 *ancestor_ino)
3121{
3122 int ret = 0;
3123 u64 parent_inode = 0;
3124 u64 parent_gen = 0;
3125 u64 start_ino = ino;
3126
3127 *ancestor_ino = 0;
3128 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3129 fs_path_reset(name);
3130
3131 if (is_waiting_for_rm(sctx, ino))
3132 break;
3133 if (is_waiting_for_move(sctx, ino)) {
3134 if (*ancestor_ino == 0)
3135 *ancestor_ino = ino;
3136 ret = get_first_ref(sctx->parent_root, ino,
3137 &parent_inode, &parent_gen, name);
3138 } else {
3139 ret = __get_cur_name_and_parent(sctx, ino, gen,
3140 &parent_inode,
3141 &parent_gen, name);
3142 if (ret > 0) {
3143 ret = 0;
3144 break;
3145 }
3146 }
3147 if (ret < 0)
3148 break;
3149 if (parent_inode == start_ino) {
3150 ret = 1;
3151 if (*ancestor_ino == 0)
3152 *ancestor_ino = ino;
3153 break;
3154 }
3155 ino = parent_inode;
3156 gen = parent_gen;
3157 }
3158 return ret;
3159}
3160
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003161static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3162{
3163 struct fs_path *from_path = NULL;
3164 struct fs_path *to_path = NULL;
Filipe Manana2b863a12014-02-16 13:43:11 +00003165 struct fs_path *name = NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003166 u64 orig_progress = sctx->send_progress;
3167 struct recorded_ref *cur;
Filipe Manana2b863a12014-02-16 13:43:11 +00003168 u64 parent_ino, parent_gen;
Filipe Manana9dc44212014-02-19 14:31:44 +00003169 struct waiting_dir_move *dm = NULL;
3170 u64 rmdir_ino = 0;
Robbie Ko801bec32015-06-23 18:39:46 +08003171 u64 ancestor;
3172 bool is_orphan;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003173 int ret;
3174
Filipe Manana2b863a12014-02-16 13:43:11 +00003175 name = fs_path_alloc();
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003176 from_path = fs_path_alloc();
Filipe Manana2b863a12014-02-16 13:43:11 +00003177 if (!name || !from_path) {
3178 ret = -ENOMEM;
3179 goto out;
3180 }
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003181
Filipe Manana9dc44212014-02-19 14:31:44 +00003182 dm = get_waiting_dir_move(sctx, pm->ino);
3183 ASSERT(dm);
3184 rmdir_ino = dm->rmdir_ino;
Robbie Ko801bec32015-06-23 18:39:46 +08003185 is_orphan = dm->orphanized;
Filipe Manana9dc44212014-02-19 14:31:44 +00003186 free_waiting_dir_move(sctx, dm);
Filipe Manana2b863a12014-02-16 13:43:11 +00003187
Robbie Ko801bec32015-06-23 18:39:46 +08003188 if (is_orphan) {
Filipe Manana84471e22015-02-28 22:29:22 +00003189 ret = gen_unique_name(sctx, pm->ino,
3190 pm->gen, from_path);
3191 } else {
3192 ret = get_first_ref(sctx->parent_root, pm->ino,
3193 &parent_ino, &parent_gen, name);
3194 if (ret < 0)
3195 goto out;
3196 ret = get_cur_path(sctx, parent_ino, parent_gen,
3197 from_path);
3198 if (ret < 0)
3199 goto out;
3200 ret = fs_path_add_path(from_path, name);
3201 }
Filipe Mananac992ec92014-03-22 17:15:24 +00003202 if (ret < 0)
3203 goto out;
3204
Filipe Mananaf9594922014-03-27 20:14:01 +00003205 sctx->send_progress = sctx->cur_ino + 1;
Robbie Ko801bec32015-06-23 18:39:46 +08003206 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
Filipe Manana7969e772016-06-17 17:13:36 +01003207 if (ret < 0)
3208 goto out;
Robbie Ko801bec32015-06-23 18:39:46 +08003209 if (ret) {
3210 LIST_HEAD(deleted_refs);
3211 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3212 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3213 &pm->update_refs, &deleted_refs,
3214 is_orphan);
3215 if (ret < 0)
3216 goto out;
3217 if (rmdir_ino) {
3218 dm = get_waiting_dir_move(sctx, pm->ino);
3219 ASSERT(dm);
3220 dm->rmdir_ino = rmdir_ino;
3221 }
3222 goto out;
3223 }
Filipe Mananac992ec92014-03-22 17:15:24 +00003224 fs_path_reset(name);
3225 to_path = name;
3226 name = NULL;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003227 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3228 if (ret < 0)
3229 goto out;
3230
3231 ret = send_rename(sctx, from_path, to_path);
3232 if (ret < 0)
3233 goto out;
3234
Filipe Manana9dc44212014-02-19 14:31:44 +00003235 if (rmdir_ino) {
3236 struct orphan_dir_info *odi;
3237
3238 odi = get_orphan_dir_info(sctx, rmdir_ino);
3239 if (!odi) {
3240 /* already deleted */
3241 goto finish;
3242 }
Robbie Ko99ea42d2015-06-23 18:39:49 +08003243 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino);
Filipe Manana9dc44212014-02-19 14:31:44 +00003244 if (ret < 0)
3245 goto out;
3246 if (!ret)
3247 goto finish;
3248
3249 name = fs_path_alloc();
3250 if (!name) {
3251 ret = -ENOMEM;
3252 goto out;
3253 }
3254 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3255 if (ret < 0)
3256 goto out;
3257 ret = send_rmdir(sctx, name);
3258 if (ret < 0)
3259 goto out;
3260 free_orphan_dir_info(sctx, odi);
3261 }
3262
3263finish:
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003264 ret = send_utimes(sctx, pm->ino, pm->gen);
3265 if (ret < 0)
3266 goto out;
3267
3268 /*
3269 * After rename/move, need to update the utimes of both new parent(s)
3270 * and old parent(s).
3271 */
3272 list_for_each_entry(cur, &pm->update_refs, list) {
Robbie Ko764433a2015-06-23 18:39:50 +08003273 /*
3274 * The parent inode might have been deleted in the send snapshot
3275 */
3276 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3277 NULL, NULL, NULL, NULL, NULL);
3278 if (ret == -ENOENT) {
3279 ret = 0;
Filipe Manana9dc44212014-02-19 14:31:44 +00003280 continue;
Robbie Ko764433a2015-06-23 18:39:50 +08003281 }
3282 if (ret < 0)
3283 goto out;
3284
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003285 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3286 if (ret < 0)
3287 goto out;
3288 }
3289
3290out:
Filipe Manana2b863a12014-02-16 13:43:11 +00003291 fs_path_free(name);
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003292 fs_path_free(from_path);
3293 fs_path_free(to_path);
3294 sctx->send_progress = orig_progress;
3295
3296 return ret;
3297}
3298
3299static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3300{
3301 if (!list_empty(&m->list))
3302 list_del(&m->list);
3303 if (!RB_EMPTY_NODE(&m->node))
3304 rb_erase(&m->node, &sctx->pending_dir_moves);
3305 __free_recorded_refs(&m->update_refs);
3306 kfree(m);
3307}
3308
3309static void tail_append_pending_moves(struct pending_dir_move *moves,
3310 struct list_head *stack)
3311{
3312 if (list_empty(&moves->list)) {
3313 list_add_tail(&moves->list, stack);
3314 } else {
3315 LIST_HEAD(list);
3316 list_splice_init(&moves->list, &list);
3317 list_add_tail(&moves->list, stack);
3318 list_splice_tail(&list, stack);
3319 }
3320}
3321
3322static int apply_children_dir_moves(struct send_ctx *sctx)
3323{
3324 struct pending_dir_move *pm;
3325 struct list_head stack;
3326 u64 parent_ino = sctx->cur_ino;
3327 int ret = 0;
3328
3329 pm = get_pending_dir_moves(sctx, parent_ino);
3330 if (!pm)
3331 return 0;
3332
3333 INIT_LIST_HEAD(&stack);
3334 tail_append_pending_moves(pm, &stack);
3335
3336 while (!list_empty(&stack)) {
3337 pm = list_first_entry(&stack, struct pending_dir_move, list);
3338 parent_ino = pm->ino;
3339 ret = apply_dir_move(sctx, pm);
3340 free_pending_move(sctx, pm);
3341 if (ret)
3342 goto out;
3343 pm = get_pending_dir_moves(sctx, parent_ino);
3344 if (pm)
3345 tail_append_pending_moves(pm, &stack);
3346 }
3347 return 0;
3348
3349out:
3350 while (!list_empty(&stack)) {
3351 pm = list_first_entry(&stack, struct pending_dir_move, list);
3352 free_pending_move(sctx, pm);
3353 }
3354 return ret;
3355}
3356
Filipe Manana84471e22015-02-28 22:29:22 +00003357/*
3358 * We might need to delay a directory rename even when no ancestor directory
3359 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3360 * renamed. This happens when we rename a directory to the old name (the name
3361 * in the parent root) of some other unrelated directory that got its rename
3362 * delayed due to some ancestor with higher number that got renamed.
3363 *
3364 * Example:
3365 *
3366 * Parent snapshot:
3367 * . (ino 256)
3368 * |---- a/ (ino 257)
3369 * | |---- file (ino 260)
3370 * |
3371 * |---- b/ (ino 258)
3372 * |---- c/ (ino 259)
3373 *
3374 * Send snapshot:
3375 * . (ino 256)
3376 * |---- a/ (ino 258)
3377 * |---- x/ (ino 259)
3378 * |---- y/ (ino 257)
3379 * |----- file (ino 260)
3380 *
3381 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3382 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3383 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3384 * must issue is:
3385 *
3386 * 1 - rename 259 from 'c' to 'x'
3387 * 2 - rename 257 from 'a' to 'x/y'
3388 * 3 - rename 258 from 'b' to 'a'
3389 *
3390 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3391 * be done right away and < 0 on error.
3392 */
3393static int wait_for_dest_dir_move(struct send_ctx *sctx,
3394 struct recorded_ref *parent_ref,
3395 const bool is_orphan)
3396{
3397 struct btrfs_path *path;
3398 struct btrfs_key key;
3399 struct btrfs_key di_key;
3400 struct btrfs_dir_item *di;
3401 u64 left_gen;
3402 u64 right_gen;
3403 int ret = 0;
Robbie Ko801bec32015-06-23 18:39:46 +08003404 struct waiting_dir_move *wdm;
Filipe Manana84471e22015-02-28 22:29:22 +00003405
3406 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3407 return 0;
3408
3409 path = alloc_path_for_send();
3410 if (!path)
3411 return -ENOMEM;
3412
3413 key.objectid = parent_ref->dir;
3414 key.type = BTRFS_DIR_ITEM_KEY;
3415 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3416
3417 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3418 if (ret < 0) {
3419 goto out;
3420 } else if (ret > 0) {
3421 ret = 0;
3422 goto out;
3423 }
3424
3425 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3426 parent_ref->name, parent_ref->name_len);
3427 if (!di) {
3428 ret = 0;
3429 goto out;
3430 }
3431 /*
3432 * di_key.objectid has the number of the inode that has a dentry in the
3433 * parent directory with the same name that sctx->cur_ino is being
3434 * renamed to. We need to check if that inode is in the send root as
3435 * well and if it is currently marked as an inode with a pending rename,
3436 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3437 * that it happens after that other inode is renamed.
3438 */
3439 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3440 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3441 ret = 0;
3442 goto out;
3443 }
3444
3445 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3446 &left_gen, NULL, NULL, NULL, NULL);
3447 if (ret < 0)
3448 goto out;
3449 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3450 &right_gen, NULL, NULL, NULL, NULL);
3451 if (ret < 0) {
3452 if (ret == -ENOENT)
3453 ret = 0;
3454 goto out;
3455 }
3456
3457 /* Different inode, no need to delay the rename of sctx->cur_ino */
3458 if (right_gen != left_gen) {
3459 ret = 0;
3460 goto out;
3461 }
3462
Robbie Ko801bec32015-06-23 18:39:46 +08003463 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3464 if (wdm && !wdm->orphanized) {
Filipe Manana84471e22015-02-28 22:29:22 +00003465 ret = add_pending_dir_move(sctx,
3466 sctx->cur_ino,
3467 sctx->cur_inode_gen,
3468 di_key.objectid,
3469 &sctx->new_refs,
3470 &sctx->deleted_refs,
3471 is_orphan);
3472 if (!ret)
3473 ret = 1;
3474 }
3475out:
3476 btrfs_free_path(path);
3477 return ret;
3478}
3479
Filipe Manana80aa6022015-03-27 17:50:45 +00003480/*
3481 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3482 * Return 1 if true, 0 if false and < 0 on error.
3483 */
3484static int is_ancestor(struct btrfs_root *root,
3485 const u64 ino1,
3486 const u64 ino1_gen,
3487 const u64 ino2,
3488 struct fs_path *fs_path)
3489{
3490 u64 ino = ino2;
3491
3492 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3493 int ret;
3494 u64 parent;
3495 u64 parent_gen;
3496
3497 fs_path_reset(fs_path);
3498 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3499 if (ret < 0) {
3500 if (ret == -ENOENT && ino == ino2)
3501 ret = 0;
3502 return ret;
3503 }
3504 if (parent == ino1)
3505 return parent_gen == ino1_gen ? 1 : 0;
3506 ino = parent;
3507 }
3508 return 0;
3509}
3510
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003511static int wait_for_parent_move(struct send_ctx *sctx,
Filipe Manana8b191a62015-04-09 14:09:14 +01003512 struct recorded_ref *parent_ref,
3513 const bool is_orphan)
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003514{
Filipe Mananaf9594922014-03-27 20:14:01 +00003515 int ret = 0;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003516 u64 ino = parent_ref->dir;
3517 u64 parent_ino_before, parent_ino_after;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003518 struct fs_path *path_before = NULL;
3519 struct fs_path *path_after = NULL;
3520 int len1, len2;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003521
3522 path_after = fs_path_alloc();
Filipe Mananaf9594922014-03-27 20:14:01 +00003523 path_before = fs_path_alloc();
3524 if (!path_after || !path_before) {
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003525 ret = -ENOMEM;
3526 goto out;
3527 }
3528
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003529 /*
Filipe Mananaf9594922014-03-27 20:14:01 +00003530 * Our current directory inode may not yet be renamed/moved because some
3531 * ancestor (immediate or not) has to be renamed/moved first. So find if
3532 * such ancestor exists and make sure our own rename/move happens after
Filipe Manana80aa6022015-03-27 17:50:45 +00003533 * that ancestor is processed to avoid path build infinite loops (done
3534 * at get_cur_path()).
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003535 */
Filipe Mananaf9594922014-03-27 20:14:01 +00003536 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3537 if (is_waiting_for_move(sctx, ino)) {
Filipe Manana80aa6022015-03-27 17:50:45 +00003538 /*
3539 * If the current inode is an ancestor of ino in the
3540 * parent root, we need to delay the rename of the
3541 * current inode, otherwise don't delayed the rename
3542 * because we can end up with a circular dependency
3543 * of renames, resulting in some directories never
3544 * getting the respective rename operations issued in
3545 * the send stream or getting into infinite path build
3546 * loops.
3547 */
3548 ret = is_ancestor(sctx->parent_root,
3549 sctx->cur_ino, sctx->cur_inode_gen,
3550 ino, path_before);
Filipe Manana4122ea62016-06-27 16:54:44 +01003551 if (ret)
3552 break;
Filipe Mananaf9594922014-03-27 20:14:01 +00003553 }
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003554
3555 fs_path_reset(path_before);
3556 fs_path_reset(path_after);
3557
3558 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
Filipe Mananaf9594922014-03-27 20:14:01 +00003559 NULL, path_after);
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003560 if (ret < 0)
3561 goto out;
3562 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3563 NULL, path_before);
Filipe Mananaf9594922014-03-27 20:14:01 +00003564 if (ret < 0 && ret != -ENOENT) {
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003565 goto out;
Filipe Mananaf9594922014-03-27 20:14:01 +00003566 } else if (ret == -ENOENT) {
Filipe Mananabf8e8ca2014-10-02 19:17:32 +01003567 ret = 0;
Filipe Mananaf9594922014-03-27 20:14:01 +00003568 break;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003569 }
3570
3571 len1 = fs_path_len(path_before);
3572 len2 = fs_path_len(path_after);
Filipe Mananaf9594922014-03-27 20:14:01 +00003573 if (ino > sctx->cur_ino &&
3574 (parent_ino_before != parent_ino_after || len1 != len2 ||
3575 memcmp(path_before->start, path_after->start, len1))) {
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003576 ret = 1;
Filipe Mananaf9594922014-03-27 20:14:01 +00003577 break;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003578 }
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003579 ino = parent_ino_after;
Filipe Mananabfa7e1f2014-03-19 14:20:54 +00003580 }
3581
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003582out:
3583 fs_path_free(path_before);
3584 fs_path_free(path_after);
3585
Filipe Mananaf9594922014-03-27 20:14:01 +00003586 if (ret == 1) {
3587 ret = add_pending_dir_move(sctx,
3588 sctx->cur_ino,
3589 sctx->cur_inode_gen,
3590 ino,
3591 &sctx->new_refs,
Filipe Manana84471e22015-02-28 22:29:22 +00003592 &sctx->deleted_refs,
Filipe Manana8b191a62015-04-09 14:09:14 +01003593 is_orphan);
Filipe Mananaf9594922014-03-27 20:14:01 +00003594 if (!ret)
3595 ret = 1;
3596 }
3597
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003598 return ret;
3599}
3600
Alexander Block31db9f72012-07-25 23:19:24 +02003601/*
3602 * This does all the move/link/unlink/rmdir magic.
3603 */
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00003604static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
Alexander Block31db9f72012-07-25 23:19:24 +02003605{
3606 int ret = 0;
3607 struct recorded_ref *cur;
Alexander Block1f4692d2012-07-28 10:42:24 +02003608 struct recorded_ref *cur2;
Josef Bacikba5e8f22013-08-16 16:52:55 -04003609 struct list_head check_dirs;
Alexander Block31db9f72012-07-25 23:19:24 +02003610 struct fs_path *valid_path = NULL;
Chris Masonb24baf62012-07-25 19:21:10 -04003611 u64 ow_inode = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02003612 u64 ow_gen;
3613 int did_overwrite = 0;
3614 int is_orphan = 0;
Filipe Manana29d6d302014-02-16 21:01:39 +00003615 u64 last_dir_ino_rm = 0;
Filipe Manana84471e22015-02-28 22:29:22 +00003616 bool can_rename = true;
Alexander Block31db9f72012-07-25 23:19:24 +02003617
3618verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3619
Alexander Block6d85ed02012-08-01 14:48:59 +02003620 /*
3621 * This should never happen as the root dir always has the same ref
3622 * which is always '..'
3623 */
3624 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
Josef Bacikba5e8f22013-08-16 16:52:55 -04003625 INIT_LIST_HEAD(&check_dirs);
Alexander Block6d85ed02012-08-01 14:48:59 +02003626
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003627 valid_path = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02003628 if (!valid_path) {
3629 ret = -ENOMEM;
3630 goto out;
3631 }
3632
Alexander Block31db9f72012-07-25 23:19:24 +02003633 /*
3634 * First, check if the first ref of the current inode was overwritten
3635 * before. If yes, we know that the current inode was already orphanized
3636 * and thus use the orphan name. If not, we can use get_cur_path to
3637 * get the path of the first ref as it would like while receiving at
3638 * this point in time.
3639 * New inodes are always orphan at the beginning, so force to use the
3640 * orphan name in this case.
3641 * The first ref is stored in valid_path and will be updated if it
3642 * gets moved around.
3643 */
3644 if (!sctx->cur_inode_new) {
3645 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3646 sctx->cur_inode_gen);
3647 if (ret < 0)
3648 goto out;
3649 if (ret)
3650 did_overwrite = 1;
3651 }
3652 if (sctx->cur_inode_new || did_overwrite) {
3653 ret = gen_unique_name(sctx, sctx->cur_ino,
3654 sctx->cur_inode_gen, valid_path);
3655 if (ret < 0)
3656 goto out;
3657 is_orphan = 1;
3658 } else {
3659 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3660 valid_path);
3661 if (ret < 0)
3662 goto out;
3663 }
3664
3665 list_for_each_entry(cur, &sctx->new_refs, list) {
3666 /*
Alexander Block1f4692d2012-07-28 10:42:24 +02003667 * We may have refs where the parent directory does not exist
3668 * yet. This happens if the parent directories inum is higher
3669 * the the current inum. To handle this case, we create the
3670 * parent directory out of order. But we need to check if this
3671 * did already happen before due to other refs in the same dir.
3672 */
3673 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3674 if (ret < 0)
3675 goto out;
3676 if (ret == inode_state_will_create) {
3677 ret = 0;
3678 /*
3679 * First check if any of the current inodes refs did
3680 * already create the dir.
3681 */
3682 list_for_each_entry(cur2, &sctx->new_refs, list) {
3683 if (cur == cur2)
3684 break;
3685 if (cur2->dir == cur->dir) {
3686 ret = 1;
3687 break;
3688 }
3689 }
3690
3691 /*
3692 * If that did not happen, check if a previous inode
3693 * did already create the dir.
3694 */
3695 if (!ret)
3696 ret = did_create_dir(sctx, cur->dir);
3697 if (ret < 0)
3698 goto out;
3699 if (!ret) {
3700 ret = send_create_inode(sctx, cur->dir);
3701 if (ret < 0)
3702 goto out;
3703 }
3704 }
3705
3706 /*
Alexander Block31db9f72012-07-25 23:19:24 +02003707 * Check if this new ref would overwrite the first ref of
3708 * another unprocessed inode. If yes, orphanize the
3709 * overwritten inode. If we find an overwritten ref that is
3710 * not the first ref, simply unlink it.
3711 */
3712 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3713 cur->name, cur->name_len,
3714 &ow_inode, &ow_gen);
3715 if (ret < 0)
3716 goto out;
3717 if (ret) {
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003718 ret = is_first_ref(sctx->parent_root,
3719 ow_inode, cur->dir, cur->name,
3720 cur->name_len);
Alexander Block31db9f72012-07-25 23:19:24 +02003721 if (ret < 0)
3722 goto out;
3723 if (ret) {
Filipe Manana8996a482015-03-12 15:16:20 +00003724 struct name_cache_entry *nce;
Robbie Ko801bec32015-06-23 18:39:46 +08003725 struct waiting_dir_move *wdm;
Filipe Manana8996a482015-03-12 15:16:20 +00003726
Alexander Block31db9f72012-07-25 23:19:24 +02003727 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3728 cur->full_path);
3729 if (ret < 0)
3730 goto out;
Robbie Ko801bec32015-06-23 18:39:46 +08003731
3732 /*
3733 * If ow_inode has its rename operation delayed
3734 * make sure that its orphanized name is used in
3735 * the source path when performing its rename
3736 * operation.
3737 */
3738 if (is_waiting_for_move(sctx, ow_inode)) {
3739 wdm = get_waiting_dir_move(sctx,
3740 ow_inode);
3741 ASSERT(wdm);
3742 wdm->orphanized = true;
3743 }
3744
Filipe Manana8996a482015-03-12 15:16:20 +00003745 /*
3746 * Make sure we clear our orphanized inode's
3747 * name from the name cache. This is because the
3748 * inode ow_inode might be an ancestor of some
3749 * other inode that will be orphanized as well
3750 * later and has an inode number greater than
3751 * sctx->send_progress. We need to prevent
3752 * future name lookups from using the old name
3753 * and get instead the orphan name.
3754 */
3755 nce = name_cache_search(sctx, ow_inode, ow_gen);
3756 if (nce) {
3757 name_cache_delete(sctx, nce);
3758 kfree(nce);
3759 }
Robbie Ko801bec32015-06-23 18:39:46 +08003760
3761 /*
3762 * ow_inode might currently be an ancestor of
3763 * cur_ino, therefore compute valid_path (the
3764 * current path of cur_ino) again because it
3765 * might contain the pre-orphanization name of
3766 * ow_inode, which is no longer valid.
3767 */
3768 fs_path_reset(valid_path);
3769 ret = get_cur_path(sctx, sctx->cur_ino,
3770 sctx->cur_inode_gen, valid_path);
3771 if (ret < 0)
3772 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003773 } else {
3774 ret = send_unlink(sctx, cur->full_path);
3775 if (ret < 0)
3776 goto out;
3777 }
3778 }
3779
Filipe Manana84471e22015-02-28 22:29:22 +00003780 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3781 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3782 if (ret < 0)
3783 goto out;
3784 if (ret == 1) {
3785 can_rename = false;
3786 *pending_move = 1;
3787 }
3788 }
3789
Filipe Manana8b191a62015-04-09 14:09:14 +01003790 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3791 can_rename) {
3792 ret = wait_for_parent_move(sctx, cur, is_orphan);
3793 if (ret < 0)
3794 goto out;
3795 if (ret == 1) {
3796 can_rename = false;
3797 *pending_move = 1;
3798 }
3799 }
3800
Alexander Block31db9f72012-07-25 23:19:24 +02003801 /*
3802 * link/move the ref to the new place. If we have an orphan
3803 * inode, move it and update valid_path. If not, link or move
3804 * it depending on the inode mode.
3805 */
Filipe Manana84471e22015-02-28 22:29:22 +00003806 if (is_orphan && can_rename) {
Alexander Block31db9f72012-07-25 23:19:24 +02003807 ret = send_rename(sctx, valid_path, cur->full_path);
3808 if (ret < 0)
3809 goto out;
3810 is_orphan = 0;
3811 ret = fs_path_copy(valid_path, cur->full_path);
3812 if (ret < 0)
3813 goto out;
Filipe Manana84471e22015-02-28 22:29:22 +00003814 } else if (can_rename) {
Alexander Block31db9f72012-07-25 23:19:24 +02003815 if (S_ISDIR(sctx->cur_inode_mode)) {
3816 /*
3817 * Dirs can't be linked, so move it. For moved
3818 * dirs, we always have one new and one deleted
3819 * ref. The deleted ref is ignored later.
3820 */
Filipe Manana8b191a62015-04-09 14:09:14 +01003821 ret = send_rename(sctx, valid_path,
3822 cur->full_path);
3823 if (!ret)
3824 ret = fs_path_copy(valid_path,
3825 cur->full_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003826 if (ret < 0)
3827 goto out;
3828 } else {
3829 ret = send_link(sctx, cur->full_path,
3830 valid_path);
3831 if (ret < 0)
3832 goto out;
3833 }
3834 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04003835 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003836 if (ret < 0)
3837 goto out;
3838 }
3839
3840 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3841 /*
3842 * Check if we can already rmdir the directory. If not,
3843 * orphanize it. For every dir item inside that gets deleted
3844 * later, we do this check again and rmdir it then if possible.
3845 * See the use of check_dirs for more details.
3846 */
Filipe Manana9dc44212014-02-19 14:31:44 +00003847 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3848 sctx->cur_ino);
Alexander Block31db9f72012-07-25 23:19:24 +02003849 if (ret < 0)
3850 goto out;
3851 if (ret) {
3852 ret = send_rmdir(sctx, valid_path);
3853 if (ret < 0)
3854 goto out;
3855 } else if (!is_orphan) {
3856 ret = orphanize_inode(sctx, sctx->cur_ino,
3857 sctx->cur_inode_gen, valid_path);
3858 if (ret < 0)
3859 goto out;
3860 is_orphan = 1;
3861 }
3862
3863 list_for_each_entry(cur, &sctx->deleted_refs, list) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04003864 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003865 if (ret < 0)
3866 goto out;
3867 }
Alexander Blockccf16262012-07-28 11:46:29 +02003868 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3869 !list_empty(&sctx->deleted_refs)) {
3870 /*
3871 * We have a moved dir. Add the old parent to check_dirs
3872 */
3873 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3874 list);
Josef Bacikba5e8f22013-08-16 16:52:55 -04003875 ret = dup_ref(cur, &check_dirs);
Alexander Blockccf16262012-07-28 11:46:29 +02003876 if (ret < 0)
3877 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003878 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3879 /*
3880 * We have a non dir inode. Go through all deleted refs and
3881 * unlink them if they were not already overwritten by other
3882 * inodes.
3883 */
3884 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3885 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3886 sctx->cur_ino, sctx->cur_inode_gen,
3887 cur->name, cur->name_len);
3888 if (ret < 0)
3889 goto out;
3890 if (!ret) {
Alexander Block1f4692d2012-07-28 10:42:24 +02003891 ret = send_unlink(sctx, cur->full_path);
3892 if (ret < 0)
3893 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02003894 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04003895 ret = dup_ref(cur, &check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003896 if (ret < 0)
3897 goto out;
3898 }
Alexander Block31db9f72012-07-25 23:19:24 +02003899 /*
3900 * If the inode is still orphan, unlink the orphan. This may
3901 * happen when a previous inode did overwrite the first ref
3902 * of this inode and no new refs were added for the current
Alexander Block766702e2012-07-28 14:11:31 +02003903 * inode. Unlinking does not mean that the inode is deleted in
3904 * all cases. There may still be links to this inode in other
3905 * places.
Alexander Block31db9f72012-07-25 23:19:24 +02003906 */
Alexander Block1f4692d2012-07-28 10:42:24 +02003907 if (is_orphan) {
Alexander Block31db9f72012-07-25 23:19:24 +02003908 ret = send_unlink(sctx, valid_path);
3909 if (ret < 0)
3910 goto out;
3911 }
3912 }
3913
3914 /*
3915 * We did collect all parent dirs where cur_inode was once located. We
3916 * now go through all these dirs and check if they are pending for
3917 * deletion and if it's finally possible to perform the rmdir now.
3918 * We also update the inode stats of the parent dirs here.
3919 */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003920 list_for_each_entry(cur, &check_dirs, list) {
Alexander Block766702e2012-07-28 14:11:31 +02003921 /*
3922 * In case we had refs into dirs that were not processed yet,
3923 * we don't need to do the utime and rmdir logic for these dirs.
3924 * The dir will be processed later.
3925 */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003926 if (cur->dir > sctx->cur_ino)
Alexander Block31db9f72012-07-25 23:19:24 +02003927 continue;
3928
Josef Bacikba5e8f22013-08-16 16:52:55 -04003929 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
Alexander Block31db9f72012-07-25 23:19:24 +02003930 if (ret < 0)
3931 goto out;
3932
3933 if (ret == inode_state_did_create ||
3934 ret == inode_state_no_change) {
3935 /* TODO delayed utimes */
Josef Bacikba5e8f22013-08-16 16:52:55 -04003936 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
Alexander Block31db9f72012-07-25 23:19:24 +02003937 if (ret < 0)
3938 goto out;
Filipe Manana29d6d302014-02-16 21:01:39 +00003939 } else if (ret == inode_state_did_delete &&
3940 cur->dir != last_dir_ino_rm) {
Filipe Manana9dc44212014-02-19 14:31:44 +00003941 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3942 sctx->cur_ino);
Alexander Block31db9f72012-07-25 23:19:24 +02003943 if (ret < 0)
3944 goto out;
3945 if (ret) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04003946 ret = get_cur_path(sctx, cur->dir,
3947 cur->dir_gen, valid_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003948 if (ret < 0)
3949 goto out;
3950 ret = send_rmdir(sctx, valid_path);
3951 if (ret < 0)
3952 goto out;
Filipe Manana29d6d302014-02-16 21:01:39 +00003953 last_dir_ino_rm = cur->dir;
Alexander Block31db9f72012-07-25 23:19:24 +02003954 }
3955 }
3956 }
3957
Alexander Block31db9f72012-07-25 23:19:24 +02003958 ret = 0;
3959
3960out:
Josef Bacikba5e8f22013-08-16 16:52:55 -04003961 __free_recorded_refs(&check_dirs);
Alexander Block31db9f72012-07-25 23:19:24 +02003962 free_recorded_refs(sctx);
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003963 fs_path_free(valid_path);
Alexander Block31db9f72012-07-25 23:19:24 +02003964 return ret;
3965}
3966
Liu Boa4d96d62014-03-03 21:31:03 +08003967static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3968 struct fs_path *name, void *ctx, struct list_head *refs)
Alexander Block31db9f72012-07-25 23:19:24 +02003969{
3970 int ret = 0;
3971 struct send_ctx *sctx = ctx;
3972 struct fs_path *p;
3973 u64 gen;
3974
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003975 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02003976 if (!p)
3977 return -ENOMEM;
3978
Liu Boa4d96d62014-03-03 21:31:03 +08003979 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02003980 NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02003981 if (ret < 0)
3982 goto out;
3983
Alexander Block31db9f72012-07-25 23:19:24 +02003984 ret = get_cur_path(sctx, dir, gen, p);
3985 if (ret < 0)
3986 goto out;
3987 ret = fs_path_add_path(p, name);
3988 if (ret < 0)
3989 goto out;
3990
Liu Boa4d96d62014-03-03 21:31:03 +08003991 ret = __record_ref(refs, dir, gen, p);
Alexander Block31db9f72012-07-25 23:19:24 +02003992
3993out:
3994 if (ret)
Tsutomu Itoh924794c2013-05-08 07:51:52 +00003995 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02003996 return ret;
3997}
3998
Liu Boa4d96d62014-03-03 21:31:03 +08003999static int __record_new_ref(int num, u64 dir, int index,
4000 struct fs_path *name,
4001 void *ctx)
4002{
4003 struct send_ctx *sctx = ctx;
4004 return record_ref(sctx->send_root, num, dir, index, name,
4005 ctx, &sctx->new_refs);
4006}
4007
4008
Alexander Block31db9f72012-07-25 23:19:24 +02004009static int __record_deleted_ref(int num, u64 dir, int index,
4010 struct fs_path *name,
4011 void *ctx)
4012{
Alexander Block31db9f72012-07-25 23:19:24 +02004013 struct send_ctx *sctx = ctx;
Liu Boa4d96d62014-03-03 21:31:03 +08004014 return record_ref(sctx->parent_root, num, dir, index, name,
4015 ctx, &sctx->deleted_refs);
Alexander Block31db9f72012-07-25 23:19:24 +02004016}
4017
4018static int record_new_ref(struct send_ctx *sctx)
4019{
4020 int ret;
4021
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004022 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4023 sctx->cmp_key, 0, __record_new_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
4032static int record_deleted_ref(struct send_ctx *sctx)
4033{
4034 int ret;
4035
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004036 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4037 sctx->cmp_key, 0, __record_deleted_ref, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004038 if (ret < 0)
4039 goto out;
4040 ret = 0;
4041
4042out:
4043 return ret;
4044}
4045
4046struct find_ref_ctx {
4047 u64 dir;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004048 u64 dir_gen;
4049 struct btrfs_root *root;
Alexander Block31db9f72012-07-25 23:19:24 +02004050 struct fs_path *name;
4051 int found_idx;
4052};
4053
4054static int __find_iref(int num, u64 dir, int index,
4055 struct fs_path *name,
4056 void *ctx_)
4057{
4058 struct find_ref_ctx *ctx = ctx_;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004059 u64 dir_gen;
4060 int ret;
Alexander Block31db9f72012-07-25 23:19:24 +02004061
4062 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4063 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
Josef Bacikba5e8f22013-08-16 16:52:55 -04004064 /*
4065 * To avoid doing extra lookups we'll only do this if everything
4066 * else matches.
4067 */
4068 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4069 NULL, NULL, NULL);
4070 if (ret)
4071 return ret;
4072 if (dir_gen != ctx->dir_gen)
4073 return 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004074 ctx->found_idx = num;
4075 return 1;
4076 }
4077 return 0;
4078}
4079
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004080static int find_iref(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02004081 struct btrfs_path *path,
4082 struct btrfs_key *key,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004083 u64 dir, u64 dir_gen, struct fs_path *name)
Alexander Block31db9f72012-07-25 23:19:24 +02004084{
4085 int ret;
4086 struct find_ref_ctx ctx;
4087
4088 ctx.dir = dir;
4089 ctx.name = name;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004090 ctx.dir_gen = dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004091 ctx.found_idx = -1;
Josef Bacikba5e8f22013-08-16 16:52:55 -04004092 ctx.root = root;
Alexander Block31db9f72012-07-25 23:19:24 +02004093
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004094 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004095 if (ret < 0)
4096 return ret;
4097
4098 if (ctx.found_idx == -1)
4099 return -ENOENT;
4100
4101 return ctx.found_idx;
4102}
4103
4104static int __record_changed_new_ref(int num, u64 dir, int index,
4105 struct fs_path *name,
4106 void *ctx)
4107{
Josef Bacikba5e8f22013-08-16 16:52:55 -04004108 u64 dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004109 int ret;
4110 struct send_ctx *sctx = ctx;
4111
Josef Bacikba5e8f22013-08-16 16:52:55 -04004112 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4113 NULL, NULL, NULL);
4114 if (ret)
4115 return ret;
4116
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004117 ret = find_iref(sctx->parent_root, sctx->right_path,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004118 sctx->cmp_key, dir, dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02004119 if (ret == -ENOENT)
4120 ret = __record_new_ref(num, dir, index, name, sctx);
4121 else if (ret > 0)
4122 ret = 0;
4123
4124 return ret;
4125}
4126
4127static int __record_changed_deleted_ref(int num, u64 dir, int index,
4128 struct fs_path *name,
4129 void *ctx)
4130{
Josef Bacikba5e8f22013-08-16 16:52:55 -04004131 u64 dir_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02004132 int ret;
4133 struct send_ctx *sctx = ctx;
4134
Josef Bacikba5e8f22013-08-16 16:52:55 -04004135 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4136 NULL, NULL, NULL);
4137 if (ret)
4138 return ret;
4139
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004140 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
Josef Bacikba5e8f22013-08-16 16:52:55 -04004141 dir, dir_gen, name);
Alexander Block31db9f72012-07-25 23:19:24 +02004142 if (ret == -ENOENT)
4143 ret = __record_deleted_ref(num, dir, index, name, sctx);
4144 else if (ret > 0)
4145 ret = 0;
4146
4147 return ret;
4148}
4149
4150static int record_changed_ref(struct send_ctx *sctx)
4151{
4152 int ret = 0;
4153
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004154 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004155 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4156 if (ret < 0)
4157 goto out;
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004158 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004159 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4160 if (ret < 0)
4161 goto out;
4162 ret = 0;
4163
4164out:
4165 return ret;
4166}
4167
4168/*
4169 * Record and process all refs at once. Needed when an inode changes the
4170 * generation number, which means that it was deleted and recreated.
4171 */
4172static int process_all_refs(struct send_ctx *sctx,
4173 enum btrfs_compare_tree_result cmd)
4174{
4175 int ret;
4176 struct btrfs_root *root;
4177 struct btrfs_path *path;
4178 struct btrfs_key key;
4179 struct btrfs_key found_key;
4180 struct extent_buffer *eb;
4181 int slot;
4182 iterate_inode_ref_t cb;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00004183 int pending_move = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004184
4185 path = alloc_path_for_send();
4186 if (!path)
4187 return -ENOMEM;
4188
4189 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4190 root = sctx->send_root;
4191 cb = __record_new_ref;
4192 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4193 root = sctx->parent_root;
4194 cb = __record_deleted_ref;
4195 } else {
David Sterba4d1a63b2014-02-03 19:24:19 +01004196 btrfs_err(sctx->send_root->fs_info,
4197 "Wrong command %d in process_all_refs", cmd);
4198 ret = -EINVAL;
4199 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004200 }
4201
4202 key.objectid = sctx->cmp_key->objectid;
4203 key.type = BTRFS_INODE_REF_KEY;
4204 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004205 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4206 if (ret < 0)
4207 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004208
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004209 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02004210 eb = path->nodes[0];
4211 slot = path->slots[0];
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004212 if (slot >= btrfs_header_nritems(eb)) {
4213 ret = btrfs_next_leaf(root, path);
4214 if (ret < 0)
4215 goto out;
4216 else if (ret > 0)
4217 break;
4218 continue;
4219 }
4220
Alexander Block31db9f72012-07-25 23:19:24 +02004221 btrfs_item_key_to_cpu(eb, &found_key, slot);
4222
4223 if (found_key.objectid != key.objectid ||
Jan Schmidt96b5bd72012-10-15 08:30:45 +00004224 (found_key.type != BTRFS_INODE_REF_KEY &&
4225 found_key.type != BTRFS_INODE_EXTREF_KEY))
Alexander Block31db9f72012-07-25 23:19:24 +02004226 break;
Alexander Block31db9f72012-07-25 23:19:24 +02004227
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004228 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004229 if (ret < 0)
4230 goto out;
4231
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004232 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02004233 }
Alexander Blocke938c8a2012-07-28 16:33:49 +02004234 btrfs_release_path(path);
Alexander Block31db9f72012-07-25 23:19:24 +02004235
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00004236 ret = process_recorded_refs(sctx, &pending_move);
4237 /* Only applicable to an incremental send. */
4238 ASSERT(pending_move == 0);
Alexander Block31db9f72012-07-25 23:19:24 +02004239
4240out:
4241 btrfs_free_path(path);
4242 return ret;
4243}
4244
4245static int send_set_xattr(struct send_ctx *sctx,
4246 struct fs_path *path,
4247 const char *name, int name_len,
4248 const char *data, int data_len)
4249{
4250 int ret = 0;
4251
4252 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4253 if (ret < 0)
4254 goto out;
4255
4256 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4257 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4258 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4259
4260 ret = send_cmd(sctx);
4261
4262tlv_put_failure:
4263out:
4264 return ret;
4265}
4266
4267static int send_remove_xattr(struct send_ctx *sctx,
4268 struct fs_path *path,
4269 const char *name, int name_len)
4270{
4271 int ret = 0;
4272
4273 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4274 if (ret < 0)
4275 goto out;
4276
4277 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4278 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4279
4280 ret = send_cmd(sctx);
4281
4282tlv_put_failure:
4283out:
4284 return ret;
4285}
4286
4287static int __process_new_xattr(int num, struct btrfs_key *di_key,
4288 const char *name, int name_len,
4289 const char *data, int data_len,
4290 u8 type, void *ctx)
4291{
4292 int ret;
4293 struct send_ctx *sctx = ctx;
4294 struct fs_path *p;
4295 posix_acl_xattr_header dummy_acl;
4296
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004297 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004298 if (!p)
4299 return -ENOMEM;
4300
4301 /*
Nicholas D Steeves01327612016-05-19 21:18:45 -04004302 * This hack is needed because empty acls are stored as zero byte
Alexander Block31db9f72012-07-25 23:19:24 +02004303 * data in xattrs. Problem with that is, that receiving these zero byte
Nicholas D Steeves01327612016-05-19 21:18:45 -04004304 * acls will fail later. To fix this, we send a dummy acl list that
Alexander Block31db9f72012-07-25 23:19:24 +02004305 * only contains the version number and no entries.
4306 */
4307 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4308 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4309 if (data_len == 0) {
4310 dummy_acl.a_version =
4311 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4312 data = (char *)&dummy_acl;
4313 data_len = sizeof(dummy_acl);
4314 }
4315 }
4316
4317 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4318 if (ret < 0)
4319 goto out;
4320
4321 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4322
4323out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004324 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004325 return ret;
4326}
4327
4328static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4329 const char *name, int name_len,
4330 const char *data, int data_len,
4331 u8 type, void *ctx)
4332{
4333 int ret;
4334 struct send_ctx *sctx = ctx;
4335 struct fs_path *p;
4336
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004337 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004338 if (!p)
4339 return -ENOMEM;
4340
4341 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4342 if (ret < 0)
4343 goto out;
4344
4345 ret = send_remove_xattr(sctx, p, name, name_len);
4346
4347out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004348 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004349 return ret;
4350}
4351
4352static int process_new_xattr(struct send_ctx *sctx)
4353{
4354 int ret = 0;
4355
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004356 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4357 sctx->cmp_key, __process_new_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004358
4359 return ret;
4360}
4361
4362static int process_deleted_xattr(struct send_ctx *sctx)
4363{
4364 int ret;
4365
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004366 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4367 sctx->cmp_key, __process_deleted_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004368
4369 return ret;
4370}
4371
4372struct find_xattr_ctx {
4373 const char *name;
4374 int name_len;
4375 int found_idx;
4376 char *found_data;
4377 int found_data_len;
4378};
4379
4380static int __find_xattr(int num, struct btrfs_key *di_key,
4381 const char *name, int name_len,
4382 const char *data, int data_len,
4383 u8 type, void *vctx)
4384{
4385 struct find_xattr_ctx *ctx = vctx;
4386
4387 if (name_len == ctx->name_len &&
4388 strncmp(name, ctx->name, name_len) == 0) {
4389 ctx->found_idx = num;
4390 ctx->found_data_len = data_len;
David Sterbae780b0d2016-01-18 18:42:13 +01004391 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02004392 if (!ctx->found_data)
4393 return -ENOMEM;
Alexander Block31db9f72012-07-25 23:19:24 +02004394 return 1;
4395 }
4396 return 0;
4397}
4398
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004399static int find_xattr(struct btrfs_root *root,
Alexander Block31db9f72012-07-25 23:19:24 +02004400 struct btrfs_path *path,
4401 struct btrfs_key *key,
4402 const char *name, int name_len,
4403 char **data, int *data_len)
4404{
4405 int ret;
4406 struct find_xattr_ctx ctx;
4407
4408 ctx.name = name;
4409 ctx.name_len = name_len;
4410 ctx.found_idx = -1;
4411 ctx.found_data = NULL;
4412 ctx.found_data_len = 0;
4413
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004414 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004415 if (ret < 0)
4416 return ret;
4417
4418 if (ctx.found_idx == -1)
4419 return -ENOENT;
4420 if (data) {
4421 *data = ctx.found_data;
4422 *data_len = ctx.found_data_len;
4423 } else {
4424 kfree(ctx.found_data);
4425 }
4426 return ctx.found_idx;
4427}
4428
4429
4430static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4431 const char *name, int name_len,
4432 const char *data, int data_len,
4433 u8 type, void *ctx)
4434{
4435 int ret;
4436 struct send_ctx *sctx = ctx;
4437 char *found_data = NULL;
4438 int found_data_len = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004439
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004440 ret = find_xattr(sctx->parent_root, sctx->right_path,
4441 sctx->cmp_key, name, name_len, &found_data,
4442 &found_data_len);
Alexander Block31db9f72012-07-25 23:19:24 +02004443 if (ret == -ENOENT) {
4444 ret = __process_new_xattr(num, di_key, name, name_len, data,
4445 data_len, type, ctx);
4446 } else if (ret >= 0) {
4447 if (data_len != found_data_len ||
4448 memcmp(data, found_data, data_len)) {
4449 ret = __process_new_xattr(num, di_key, name, name_len,
4450 data, data_len, type, ctx);
4451 } else {
4452 ret = 0;
4453 }
4454 }
4455
4456 kfree(found_data);
Alexander Block31db9f72012-07-25 23:19:24 +02004457 return ret;
4458}
4459
4460static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4461 const char *name, int name_len,
4462 const char *data, int data_len,
4463 u8 type, void *ctx)
4464{
4465 int ret;
4466 struct send_ctx *sctx = ctx;
4467
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004468 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4469 name, name_len, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02004470 if (ret == -ENOENT)
4471 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4472 data_len, type, ctx);
4473 else if (ret >= 0)
4474 ret = 0;
4475
4476 return ret;
4477}
4478
4479static int process_changed_xattr(struct send_ctx *sctx)
4480{
4481 int ret = 0;
4482
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004483 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004484 sctx->cmp_key, __process_changed_new_xattr, sctx);
4485 if (ret < 0)
4486 goto out;
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004487 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
Alexander Block31db9f72012-07-25 23:19:24 +02004488 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4489
4490out:
4491 return ret;
4492}
4493
4494static int process_all_new_xattrs(struct send_ctx *sctx)
4495{
4496 int ret;
4497 struct btrfs_root *root;
4498 struct btrfs_path *path;
4499 struct btrfs_key key;
4500 struct btrfs_key found_key;
4501 struct extent_buffer *eb;
4502 int slot;
4503
4504 path = alloc_path_for_send();
4505 if (!path)
4506 return -ENOMEM;
4507
4508 root = sctx->send_root;
4509
4510 key.objectid = sctx->cmp_key->objectid;
4511 key.type = BTRFS_XATTR_ITEM_KEY;
4512 key.offset = 0;
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004513 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4514 if (ret < 0)
4515 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02004516
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004517 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02004518 eb = path->nodes[0];
4519 slot = path->slots[0];
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004520 if (slot >= btrfs_header_nritems(eb)) {
4521 ret = btrfs_next_leaf(root, path);
4522 if (ret < 0) {
4523 goto out;
4524 } else if (ret > 0) {
4525 ret = 0;
4526 break;
4527 }
4528 continue;
4529 }
Alexander Block31db9f72012-07-25 23:19:24 +02004530
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004531 btrfs_item_key_to_cpu(eb, &found_key, slot);
Alexander Block31db9f72012-07-25 23:19:24 +02004532 if (found_key.objectid != key.objectid ||
4533 found_key.type != key.type) {
4534 ret = 0;
4535 goto out;
4536 }
4537
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004538 ret = iterate_dir_item(root, path, &found_key,
4539 __process_new_xattr, sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02004540 if (ret < 0)
4541 goto out;
4542
Filipe David Borba Mananadff6d0a2014-02-05 16:48:56 +00004543 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02004544 }
4545
4546out:
4547 btrfs_free_path(path);
4548 return ret;
4549}
4550
Josef Baciked259092013-10-25 11:36:01 -04004551static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4552{
4553 struct btrfs_root *root = sctx->send_root;
4554 struct btrfs_fs_info *fs_info = root->fs_info;
4555 struct inode *inode;
4556 struct page *page;
4557 char *addr;
4558 struct btrfs_key key;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004559 pgoff_t index = offset >> PAGE_SHIFT;
Josef Baciked259092013-10-25 11:36:01 -04004560 pgoff_t last_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004561 unsigned pg_offset = offset & ~PAGE_MASK;
Josef Baciked259092013-10-25 11:36:01 -04004562 ssize_t ret = 0;
4563
4564 key.objectid = sctx->cur_ino;
4565 key.type = BTRFS_INODE_ITEM_KEY;
4566 key.offset = 0;
4567
4568 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4569 if (IS_ERR(inode))
4570 return PTR_ERR(inode);
4571
4572 if (offset + len > i_size_read(inode)) {
4573 if (offset > i_size_read(inode))
4574 len = 0;
4575 else
4576 len = offset - i_size_read(inode);
4577 }
4578 if (len == 0)
4579 goto out;
4580
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004581 last_index = (offset + len - 1) >> PAGE_SHIFT;
Liu Bo2131bcd32014-03-05 10:07:35 +08004582
4583 /* initial readahead */
4584 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4585 file_ra_state_init(&sctx->ra, inode->i_mapping);
4586 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4587 last_index - index + 1);
4588
Josef Baciked259092013-10-25 11:36:01 -04004589 while (index <= last_index) {
4590 unsigned cur_len = min_t(unsigned, len,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004591 PAGE_SIZE - pg_offset);
David Sterbae780b0d2016-01-18 18:42:13 +01004592 page = find_or_create_page(inode->i_mapping, index, GFP_KERNEL);
Josef Baciked259092013-10-25 11:36:01 -04004593 if (!page) {
4594 ret = -ENOMEM;
4595 break;
4596 }
4597
4598 if (!PageUptodate(page)) {
4599 btrfs_readpage(NULL, page);
4600 lock_page(page);
4601 if (!PageUptodate(page)) {
4602 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004603 put_page(page);
Josef Baciked259092013-10-25 11:36:01 -04004604 ret = -EIO;
4605 break;
4606 }
4607 }
4608
4609 addr = kmap(page);
4610 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4611 kunmap(page);
4612 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004613 put_page(page);
Josef Baciked259092013-10-25 11:36:01 -04004614 index++;
4615 pg_offset = 0;
4616 len -= cur_len;
4617 ret += cur_len;
4618 }
4619out:
4620 iput(inode);
4621 return ret;
4622}
4623
Alexander Block31db9f72012-07-25 23:19:24 +02004624/*
4625 * Read some bytes from the current inode/file and send a write command to
4626 * user space.
4627 */
4628static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4629{
4630 int ret = 0;
4631 struct fs_path *p;
Josef Baciked259092013-10-25 11:36:01 -04004632 ssize_t num_read = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004633
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004634 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004635 if (!p)
4636 return -ENOMEM;
4637
Alexander Block31db9f72012-07-25 23:19:24 +02004638verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4639
Josef Baciked259092013-10-25 11:36:01 -04004640 num_read = fill_read_buf(sctx, offset, len);
4641 if (num_read <= 0) {
4642 if (num_read < 0)
4643 ret = num_read;
Alexander Block31db9f72012-07-25 23:19:24 +02004644 goto out;
Josef Baciked259092013-10-25 11:36:01 -04004645 }
Alexander Block31db9f72012-07-25 23:19:24 +02004646
4647 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4648 if (ret < 0)
4649 goto out;
4650
4651 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4652 if (ret < 0)
4653 goto out;
4654
4655 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4656 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
Alexander Blocke938c8a2012-07-28 16:33:49 +02004657 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
Alexander Block31db9f72012-07-25 23:19:24 +02004658
4659 ret = send_cmd(sctx);
4660
4661tlv_put_failure:
4662out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004663 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004664 if (ret < 0)
4665 return ret;
Alexander Blocke938c8a2012-07-28 16:33:49 +02004666 return num_read;
Alexander Block31db9f72012-07-25 23:19:24 +02004667}
4668
4669/*
4670 * Send a clone command to user space.
4671 */
4672static int send_clone(struct send_ctx *sctx,
4673 u64 offset, u32 len,
4674 struct clone_root *clone_root)
4675{
4676 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02004677 struct fs_path *p;
4678 u64 gen;
4679
4680verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4681 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4682 clone_root->root->objectid, clone_root->ino,
4683 clone_root->offset);
4684
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004685 p = fs_path_alloc();
Alexander Block31db9f72012-07-25 23:19:24 +02004686 if (!p)
4687 return -ENOMEM;
4688
4689 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4690 if (ret < 0)
4691 goto out;
4692
4693 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4694 if (ret < 0)
4695 goto out;
4696
4697 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4698 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4699 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4700
Alexander Blocke938c8a2012-07-28 16:33:49 +02004701 if (clone_root->root == sctx->send_root) {
Alexander Block31db9f72012-07-25 23:19:24 +02004702 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02004703 &gen, NULL, NULL, NULL, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02004704 if (ret < 0)
4705 goto out;
4706 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4707 } else {
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004708 ret = get_inode_path(clone_root->root, clone_root->ino, p);
Alexander Block31db9f72012-07-25 23:19:24 +02004709 }
4710 if (ret < 0)
4711 goto out;
4712
Josef Bacik37b8d272015-06-04 17:17:25 -04004713 /*
4714 * If the parent we're using has a received_uuid set then use that as
4715 * our clone source as that is what we will look for when doing a
4716 * receive.
4717 *
4718 * This covers the case that we create a snapshot off of a received
4719 * subvolume and then use that as the parent and try to receive on a
4720 * different host.
4721 */
4722 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4723 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4724 clone_root->root->root_item.received_uuid);
4725 else
4726 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4727 clone_root->root->root_item.uuid);
Alexander Block31db9f72012-07-25 23:19:24 +02004728 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
Filipe David Borba Manana5a0f4e22013-12-03 15:55:48 +00004729 le64_to_cpu(clone_root->root->root_item.ctransid));
Alexander Block31db9f72012-07-25 23:19:24 +02004730 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4731 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4732 clone_root->offset);
4733
4734 ret = send_cmd(sctx);
4735
4736tlv_put_failure:
4737out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004738 fs_path_free(p);
Alexander Block31db9f72012-07-25 23:19:24 +02004739 return ret;
4740}
4741
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004742/*
4743 * Send an update extent command to user space.
4744 */
4745static int send_update_extent(struct send_ctx *sctx,
4746 u64 offset, u32 len)
4747{
4748 int ret = 0;
4749 struct fs_path *p;
4750
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004751 p = fs_path_alloc();
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004752 if (!p)
4753 return -ENOMEM;
4754
4755 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4756 if (ret < 0)
4757 goto out;
4758
4759 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4760 if (ret < 0)
4761 goto out;
4762
4763 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4764 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4765 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4766
4767 ret = send_cmd(sctx);
4768
4769tlv_put_failure:
4770out:
Tsutomu Itoh924794c2013-05-08 07:51:52 +00004771 fs_path_free(p);
Mark Fashehcb95e7b2013-02-04 20:54:57 +00004772 return ret;
4773}
4774
Josef Bacik16e75492013-10-22 12:18:51 -04004775static int send_hole(struct send_ctx *sctx, u64 end)
4776{
4777 struct fs_path *p = NULL;
4778 u64 offset = sctx->cur_inode_last_extent;
4779 u64 len;
4780 int ret = 0;
4781
4782 p = fs_path_alloc();
4783 if (!p)
4784 return -ENOMEM;
Filipe Mananac715e152014-03-31 14:52:14 +01004785 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4786 if (ret < 0)
4787 goto tlv_put_failure;
Josef Bacik16e75492013-10-22 12:18:51 -04004788 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4789 while (offset < end) {
4790 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4791
4792 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4793 if (ret < 0)
4794 break;
Josef Bacik16e75492013-10-22 12:18:51 -04004795 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4796 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4797 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4798 ret = send_cmd(sctx);
4799 if (ret < 0)
4800 break;
4801 offset += len;
4802 }
4803tlv_put_failure:
4804 fs_path_free(p);
4805 return ret;
4806}
4807
Filipe Mananad906d492015-10-02 10:47:34 +01004808static int send_extent_data(struct send_ctx *sctx,
4809 const u64 offset,
4810 const u64 len)
4811{
4812 u64 sent = 0;
4813
4814 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4815 return send_update_extent(sctx, offset, len);
4816
4817 while (sent < len) {
4818 u64 size = len - sent;
4819 int ret;
4820
4821 if (size > BTRFS_SEND_READ_SIZE)
4822 size = BTRFS_SEND_READ_SIZE;
4823 ret = send_write(sctx, offset + sent, size);
4824 if (ret < 0)
4825 return ret;
4826 if (!ret)
4827 break;
4828 sent += ret;
4829 }
4830 return 0;
4831}
4832
4833static int clone_range(struct send_ctx *sctx,
4834 struct clone_root *clone_root,
4835 const u64 disk_byte,
4836 u64 data_offset,
4837 u64 offset,
4838 u64 len)
4839{
4840 struct btrfs_path *path;
4841 struct btrfs_key key;
4842 int ret;
4843
4844 path = alloc_path_for_send();
4845 if (!path)
4846 return -ENOMEM;
4847
4848 /*
4849 * We can't send a clone operation for the entire range if we find
4850 * extent items in the respective range in the source file that
4851 * refer to different extents or if we find holes.
4852 * So check for that and do a mix of clone and regular write/copy
4853 * operations if needed.
4854 *
4855 * Example:
4856 *
4857 * mkfs.btrfs -f /dev/sda
4858 * mount /dev/sda /mnt
4859 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
4860 * cp --reflink=always /mnt/foo /mnt/bar
4861 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
4862 * btrfs subvolume snapshot -r /mnt /mnt/snap
4863 *
4864 * If when we send the snapshot and we are processing file bar (which
4865 * has a higher inode number than foo) we blindly send a clone operation
4866 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
4867 * a file bar that matches the content of file foo - iow, doesn't match
4868 * the content from bar in the original filesystem.
4869 */
4870 key.objectid = clone_root->ino;
4871 key.type = BTRFS_EXTENT_DATA_KEY;
4872 key.offset = clone_root->offset;
4873 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
4874 if (ret < 0)
4875 goto out;
4876 if (ret > 0 && path->slots[0] > 0) {
4877 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4878 if (key.objectid == clone_root->ino &&
4879 key.type == BTRFS_EXTENT_DATA_KEY)
4880 path->slots[0]--;
4881 }
4882
4883 while (true) {
4884 struct extent_buffer *leaf = path->nodes[0];
4885 int slot = path->slots[0];
4886 struct btrfs_file_extent_item *ei;
4887 u8 type;
4888 u64 ext_len;
4889 u64 clone_len;
4890
4891 if (slot >= btrfs_header_nritems(leaf)) {
4892 ret = btrfs_next_leaf(clone_root->root, path);
4893 if (ret < 0)
4894 goto out;
4895 else if (ret > 0)
4896 break;
4897 continue;
4898 }
4899
4900 btrfs_item_key_to_cpu(leaf, &key, slot);
4901
4902 /*
4903 * We might have an implicit trailing hole (NO_HOLES feature
4904 * enabled). We deal with it after leaving this loop.
4905 */
4906 if (key.objectid != clone_root->ino ||
4907 key.type != BTRFS_EXTENT_DATA_KEY)
4908 break;
4909
4910 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4911 type = btrfs_file_extent_type(leaf, ei);
4912 if (type == BTRFS_FILE_EXTENT_INLINE) {
4913 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004914 ext_len = PAGE_ALIGN(ext_len);
Filipe Mananad906d492015-10-02 10:47:34 +01004915 } else {
4916 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
4917 }
4918
4919 if (key.offset + ext_len <= clone_root->offset)
4920 goto next;
4921
4922 if (key.offset > clone_root->offset) {
4923 /* Implicit hole, NO_HOLES feature enabled. */
4924 u64 hole_len = key.offset - clone_root->offset;
4925
4926 if (hole_len > len)
4927 hole_len = len;
4928 ret = send_extent_data(sctx, offset, hole_len);
4929 if (ret < 0)
4930 goto out;
4931
4932 len -= hole_len;
4933 if (len == 0)
4934 break;
4935 offset += hole_len;
4936 clone_root->offset += hole_len;
4937 data_offset += hole_len;
4938 }
4939
4940 if (key.offset >= clone_root->offset + len)
4941 break;
4942
4943 clone_len = min_t(u64, ext_len, len);
4944
4945 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
4946 btrfs_file_extent_offset(leaf, ei) == data_offset)
4947 ret = send_clone(sctx, offset, clone_len, clone_root);
4948 else
4949 ret = send_extent_data(sctx, offset, clone_len);
4950
4951 if (ret < 0)
4952 goto out;
4953
4954 len -= clone_len;
4955 if (len == 0)
4956 break;
4957 offset += clone_len;
4958 clone_root->offset += clone_len;
4959 data_offset += clone_len;
4960next:
4961 path->slots[0]++;
4962 }
4963
4964 if (len > 0)
4965 ret = send_extent_data(sctx, offset, len);
4966 else
4967 ret = 0;
4968out:
4969 btrfs_free_path(path);
4970 return ret;
4971}
4972
Alexander Block31db9f72012-07-25 23:19:24 +02004973static int send_write_or_clone(struct send_ctx *sctx,
4974 struct btrfs_path *path,
4975 struct btrfs_key *key,
4976 struct clone_root *clone_root)
4977{
4978 int ret = 0;
4979 struct btrfs_file_extent_item *ei;
4980 u64 offset = key->offset;
Alexander Block31db9f72012-07-25 23:19:24 +02004981 u64 len;
Alexander Block31db9f72012-07-25 23:19:24 +02004982 u8 type;
Filipe David Borba Manana28e5dd82014-01-12 02:26:28 +00004983 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
Alexander Block31db9f72012-07-25 23:19:24 +02004984
4985 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4986 struct btrfs_file_extent_item);
4987 type = btrfs_file_extent_type(path->nodes[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004988 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08004989 len = btrfs_file_extent_inline_len(path->nodes[0],
4990 path->slots[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004991 /*
4992 * it is possible the inline item won't cover the whole page,
4993 * but there may be items after this page. Make
4994 * sure to send the whole thing
4995 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004996 len = PAGE_ALIGN(len);
Chris Mason74dd17f2012-08-07 16:25:13 -04004997 } else {
Alexander Block31db9f72012-07-25 23:19:24 +02004998 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
Chris Mason74dd17f2012-08-07 16:25:13 -04004999 }
Alexander Block31db9f72012-07-25 23:19:24 +02005000
5001 if (offset + len > sctx->cur_inode_size)
5002 len = sctx->cur_inode_size - offset;
5003 if (len == 0) {
5004 ret = 0;
5005 goto out;
5006 }
5007
Filipe David Borba Manana28e5dd82014-01-12 02:26:28 +00005008 if (clone_root && IS_ALIGNED(offset + len, bs)) {
Filipe Mananad906d492015-10-02 10:47:34 +01005009 u64 disk_byte;
5010 u64 data_offset;
5011
5012 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5013 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5014 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5015 offset, len);
Mark Fashehcb95e7b2013-02-04 20:54:57 +00005016 } else {
Filipe Mananad906d492015-10-02 10:47:34 +01005017 ret = send_extent_data(sctx, offset, len);
Alexander Block31db9f72012-07-25 23:19:24 +02005018 }
Alexander Block31db9f72012-07-25 23:19:24 +02005019out:
5020 return ret;
5021}
5022
5023static int is_extent_unchanged(struct send_ctx *sctx,
5024 struct btrfs_path *left_path,
5025 struct btrfs_key *ekey)
5026{
5027 int ret = 0;
5028 struct btrfs_key key;
5029 struct btrfs_path *path = NULL;
5030 struct extent_buffer *eb;
5031 int slot;
5032 struct btrfs_key found_key;
5033 struct btrfs_file_extent_item *ei;
5034 u64 left_disknr;
5035 u64 right_disknr;
5036 u64 left_offset;
5037 u64 right_offset;
5038 u64 left_offset_fixed;
5039 u64 left_len;
5040 u64 right_len;
Chris Mason74dd17f2012-08-07 16:25:13 -04005041 u64 left_gen;
5042 u64 right_gen;
Alexander Block31db9f72012-07-25 23:19:24 +02005043 u8 left_type;
5044 u8 right_type;
5045
5046 path = alloc_path_for_send();
5047 if (!path)
5048 return -ENOMEM;
5049
5050 eb = left_path->nodes[0];
5051 slot = left_path->slots[0];
Alexander Block31db9f72012-07-25 23:19:24 +02005052 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5053 left_type = btrfs_file_extent_type(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02005054
5055 if (left_type != BTRFS_FILE_EXTENT_REG) {
5056 ret = 0;
5057 goto out;
5058 }
Chris Mason74dd17f2012-08-07 16:25:13 -04005059 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5060 left_len = btrfs_file_extent_num_bytes(eb, ei);
5061 left_offset = btrfs_file_extent_offset(eb, ei);
5062 left_gen = btrfs_file_extent_generation(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02005063
5064 /*
5065 * Following comments will refer to these graphics. L is the left
5066 * extents which we are checking at the moment. 1-8 are the right
5067 * extents that we iterate.
5068 *
5069 * |-----L-----|
5070 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5071 *
5072 * |-----L-----|
5073 * |--1--|-2b-|...(same as above)
5074 *
5075 * Alternative situation. Happens on files where extents got split.
5076 * |-----L-----|
5077 * |-----------7-----------|-6-|
5078 *
5079 * Alternative situation. Happens on files which got larger.
5080 * |-----L-----|
5081 * |-8-|
5082 * Nothing follows after 8.
5083 */
5084
5085 key.objectid = ekey->objectid;
5086 key.type = BTRFS_EXTENT_DATA_KEY;
5087 key.offset = ekey->offset;
5088 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5089 if (ret < 0)
5090 goto out;
5091 if (ret) {
5092 ret = 0;
5093 goto out;
5094 }
5095
5096 /*
5097 * Handle special case where the right side has no extents at all.
5098 */
5099 eb = path->nodes[0];
5100 slot = path->slots[0];
5101 btrfs_item_key_to_cpu(eb, &found_key, slot);
5102 if (found_key.objectid != key.objectid ||
5103 found_key.type != key.type) {
Josef Bacik57cfd462013-08-20 15:55:39 -04005104 /* If we're a hole then just pretend nothing changed */
5105 ret = (left_disknr) ? 0 : 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005106 goto out;
5107 }
5108
5109 /*
5110 * We're now on 2a, 2b or 7.
5111 */
5112 key = found_key;
5113 while (key.offset < ekey->offset + left_len) {
5114 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5115 right_type = btrfs_file_extent_type(eb, ei);
Alexander Block31db9f72012-07-25 23:19:24 +02005116 if (right_type != BTRFS_FILE_EXTENT_REG) {
5117 ret = 0;
5118 goto out;
5119 }
5120
Josef Bacik007d31f2013-10-31 16:49:02 -04005121 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5122 right_len = btrfs_file_extent_num_bytes(eb, ei);
5123 right_offset = btrfs_file_extent_offset(eb, ei);
5124 right_gen = btrfs_file_extent_generation(eb, ei);
5125
Alexander Block31db9f72012-07-25 23:19:24 +02005126 /*
5127 * Are we at extent 8? If yes, we know the extent is changed.
5128 * This may only happen on the first iteration.
5129 */
Alexander Blockd8347fa2012-08-01 12:49:15 +02005130 if (found_key.offset + right_len <= ekey->offset) {
Josef Bacik57cfd462013-08-20 15:55:39 -04005131 /* If we're a hole just pretend nothing changed */
5132 ret = (left_disknr) ? 0 : 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005133 goto out;
5134 }
5135
5136 left_offset_fixed = left_offset;
5137 if (key.offset < ekey->offset) {
5138 /* Fix the right offset for 2a and 7. */
5139 right_offset += ekey->offset - key.offset;
5140 } else {
5141 /* Fix the left offset for all behind 2a and 2b */
5142 left_offset_fixed += key.offset - ekey->offset;
5143 }
5144
5145 /*
5146 * Check if we have the same extent.
5147 */
Alexander Block39540962012-08-01 12:46:05 +02005148 if (left_disknr != right_disknr ||
Chris Mason74dd17f2012-08-07 16:25:13 -04005149 left_offset_fixed != right_offset ||
5150 left_gen != right_gen) {
Alexander Block31db9f72012-07-25 23:19:24 +02005151 ret = 0;
5152 goto out;
5153 }
5154
5155 /*
5156 * Go to the next extent.
5157 */
5158 ret = btrfs_next_item(sctx->parent_root, path);
5159 if (ret < 0)
5160 goto out;
5161 if (!ret) {
5162 eb = path->nodes[0];
5163 slot = path->slots[0];
5164 btrfs_item_key_to_cpu(eb, &found_key, slot);
5165 }
5166 if (ret || found_key.objectid != key.objectid ||
5167 found_key.type != key.type) {
5168 key.offset += right_len;
5169 break;
Jan Schmidtadaa4b82013-03-21 14:30:23 +00005170 }
5171 if (found_key.offset != key.offset + right_len) {
5172 ret = 0;
5173 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005174 }
5175 key = found_key;
5176 }
5177
5178 /*
5179 * We're now behind the left extent (treat as unchanged) or at the end
5180 * of the right side (treat as changed).
5181 */
5182 if (key.offset >= ekey->offset + left_len)
5183 ret = 1;
5184 else
5185 ret = 0;
5186
5187
5188out:
5189 btrfs_free_path(path);
5190 return ret;
5191}
5192
Josef Bacik16e75492013-10-22 12:18:51 -04005193static int get_last_extent(struct send_ctx *sctx, u64 offset)
5194{
5195 struct btrfs_path *path;
5196 struct btrfs_root *root = sctx->send_root;
5197 struct btrfs_file_extent_item *fi;
5198 struct btrfs_key key;
5199 u64 extent_end;
5200 u8 type;
5201 int ret;
5202
5203 path = alloc_path_for_send();
5204 if (!path)
5205 return -ENOMEM;
5206
5207 sctx->cur_inode_last_extent = 0;
5208
5209 key.objectid = sctx->cur_ino;
5210 key.type = BTRFS_EXTENT_DATA_KEY;
5211 key.offset = offset;
5212 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5213 if (ret < 0)
5214 goto out;
5215 ret = 0;
5216 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5217 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5218 goto out;
5219
5220 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5221 struct btrfs_file_extent_item);
5222 type = btrfs_file_extent_type(path->nodes[0], fi);
5223 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08005224 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5225 path->slots[0], fi);
Josef Bacik16e75492013-10-22 12:18:51 -04005226 extent_end = ALIGN(key.offset + size,
5227 sctx->send_root->sectorsize);
5228 } else {
5229 extent_end = key.offset +
5230 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5231 }
5232 sctx->cur_inode_last_extent = extent_end;
5233out:
5234 btrfs_free_path(path);
5235 return ret;
5236}
5237
5238static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5239 struct btrfs_key *key)
5240{
5241 struct btrfs_file_extent_item *fi;
5242 u64 extent_end;
5243 u8 type;
5244 int ret = 0;
5245
5246 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5247 return 0;
5248
5249 if (sctx->cur_inode_last_extent == (u64)-1) {
5250 ret = get_last_extent(sctx, key->offset - 1);
5251 if (ret)
5252 return ret;
5253 }
5254
5255 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5256 struct btrfs_file_extent_item);
5257 type = btrfs_file_extent_type(path->nodes[0], fi);
5258 if (type == BTRFS_FILE_EXTENT_INLINE) {
Chris Mason514ac8a2014-01-03 21:07:00 -08005259 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5260 path->slots[0], fi);
Josef Bacik16e75492013-10-22 12:18:51 -04005261 extent_end = ALIGN(key->offset + size,
5262 sctx->send_root->sectorsize);
5263 } else {
5264 extent_end = key->offset +
5265 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5266 }
Filipe David Borba Mananabf54f412014-01-28 01:38:06 +00005267
5268 if (path->slots[0] == 0 &&
5269 sctx->cur_inode_last_extent < key->offset) {
5270 /*
5271 * We might have skipped entire leafs that contained only
5272 * file extent items for our current inode. These leafs have
5273 * a generation number smaller (older) than the one in the
5274 * current leaf and the leaf our last extent came from, and
5275 * are located between these 2 leafs.
5276 */
5277 ret = get_last_extent(sctx, key->offset - 1);
5278 if (ret)
5279 return ret;
5280 }
5281
Josef Bacik16e75492013-10-22 12:18:51 -04005282 if (sctx->cur_inode_last_extent < key->offset)
5283 ret = send_hole(sctx, key->offset);
5284 sctx->cur_inode_last_extent = extent_end;
5285 return ret;
5286}
5287
Alexander Block31db9f72012-07-25 23:19:24 +02005288static int process_extent(struct send_ctx *sctx,
5289 struct btrfs_path *path,
5290 struct btrfs_key *key)
5291{
Alexander Block31db9f72012-07-25 23:19:24 +02005292 struct clone_root *found_clone = NULL;
Josef Bacik57cfd462013-08-20 15:55:39 -04005293 int ret = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02005294
5295 if (S_ISLNK(sctx->cur_inode_mode))
5296 return 0;
5297
5298 if (sctx->parent_root && !sctx->cur_inode_new) {
5299 ret = is_extent_unchanged(sctx, path, key);
5300 if (ret < 0)
5301 goto out;
5302 if (ret) {
5303 ret = 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005304 goto out_hole;
Alexander Block31db9f72012-07-25 23:19:24 +02005305 }
Josef Bacik57cfd462013-08-20 15:55:39 -04005306 } else {
5307 struct btrfs_file_extent_item *ei;
5308 u8 type;
5309
5310 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5311 struct btrfs_file_extent_item);
5312 type = btrfs_file_extent_type(path->nodes[0], ei);
5313 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5314 type == BTRFS_FILE_EXTENT_REG) {
5315 /*
5316 * The send spec does not have a prealloc command yet,
5317 * so just leave a hole for prealloc'ed extents until
5318 * we have enough commands queued up to justify rev'ing
5319 * the send spec.
5320 */
5321 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5322 ret = 0;
5323 goto out;
5324 }
5325
5326 /* Have a hole, just skip it. */
5327 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5328 ret = 0;
5329 goto out;
5330 }
5331 }
Alexander Block31db9f72012-07-25 23:19:24 +02005332 }
5333
5334 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5335 sctx->cur_inode_size, &found_clone);
5336 if (ret != -ENOENT && ret < 0)
5337 goto out;
5338
5339 ret = send_write_or_clone(sctx, path, key, found_clone);
Josef Bacik16e75492013-10-22 12:18:51 -04005340 if (ret)
5341 goto out;
5342out_hole:
5343 ret = maybe_send_hole(sctx, path, key);
Alexander Block31db9f72012-07-25 23:19:24 +02005344out:
5345 return ret;
5346}
5347
5348static int process_all_extents(struct send_ctx *sctx)
5349{
5350 int ret;
5351 struct btrfs_root *root;
5352 struct btrfs_path *path;
5353 struct btrfs_key key;
5354 struct btrfs_key found_key;
5355 struct extent_buffer *eb;
5356 int slot;
5357
5358 root = sctx->send_root;
5359 path = alloc_path_for_send();
5360 if (!path)
5361 return -ENOMEM;
5362
5363 key.objectid = sctx->cmp_key->objectid;
5364 key.type = BTRFS_EXTENT_DATA_KEY;
5365 key.offset = 0;
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005366 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5367 if (ret < 0)
5368 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005369
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005370 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02005371 eb = path->nodes[0];
5372 slot = path->slots[0];
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005373
5374 if (slot >= btrfs_header_nritems(eb)) {
5375 ret = btrfs_next_leaf(root, path);
5376 if (ret < 0) {
5377 goto out;
5378 } else if (ret > 0) {
5379 ret = 0;
5380 break;
5381 }
5382 continue;
5383 }
5384
Alexander Block31db9f72012-07-25 23:19:24 +02005385 btrfs_item_key_to_cpu(eb, &found_key, slot);
5386
5387 if (found_key.objectid != key.objectid ||
5388 found_key.type != key.type) {
5389 ret = 0;
5390 goto out;
5391 }
5392
5393 ret = process_extent(sctx, path, &found_key);
5394 if (ret < 0)
5395 goto out;
5396
Filipe David Borba Manana7fdd29d2014-01-24 17:42:09 +00005397 path->slots[0]++;
Alexander Block31db9f72012-07-25 23:19:24 +02005398 }
5399
5400out:
5401 btrfs_free_path(path);
5402 return ret;
5403}
5404
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005405static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5406 int *pending_move,
5407 int *refs_processed)
Alexander Block31db9f72012-07-25 23:19:24 +02005408{
5409 int ret = 0;
5410
5411 if (sctx->cur_ino == 0)
5412 goto out;
5413 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
Jan Schmidt96b5bd72012-10-15 08:30:45 +00005414 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
Alexander Block31db9f72012-07-25 23:19:24 +02005415 goto out;
5416 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5417 goto out;
5418
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005419 ret = process_recorded_refs(sctx, pending_move);
Alexander Blocke479d9b2012-07-28 16:09:35 +02005420 if (ret < 0)
5421 goto out;
5422
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005423 *refs_processed = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005424out:
5425 return ret;
5426}
5427
5428static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5429{
5430 int ret = 0;
5431 u64 left_mode;
5432 u64 left_uid;
5433 u64 left_gid;
5434 u64 right_mode;
5435 u64 right_uid;
5436 u64 right_gid;
5437 int need_chmod = 0;
5438 int need_chown = 0;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005439 int pending_move = 0;
5440 int refs_processed = 0;
Alexander Block31db9f72012-07-25 23:19:24 +02005441
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005442 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5443 &refs_processed);
Alexander Block31db9f72012-07-25 23:19:24 +02005444 if (ret < 0)
5445 goto out;
5446
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005447 /*
5448 * We have processed the refs and thus need to advance send_progress.
5449 * Now, calls to get_cur_xxx will take the updated refs of the current
5450 * inode into account.
5451 *
5452 * On the other hand, if our current inode is a directory and couldn't
5453 * be moved/renamed because its parent was renamed/moved too and it has
5454 * a higher inode number, we can only move/rename our current inode
5455 * after we moved/renamed its parent. Therefore in this case operate on
5456 * the old path (pre move/rename) of our current inode, and the
5457 * move/rename will be performed later.
5458 */
5459 if (refs_processed && !pending_move)
5460 sctx->send_progress = sctx->cur_ino + 1;
5461
Alexander Block31db9f72012-07-25 23:19:24 +02005462 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5463 goto out;
5464 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5465 goto out;
5466
5467 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
Alexander Block85a7b332012-07-26 23:39:10 +02005468 &left_mode, &left_uid, &left_gid, NULL);
Alexander Block31db9f72012-07-25 23:19:24 +02005469 if (ret < 0)
5470 goto out;
5471
Alex Lyakase2d044f2012-10-17 13:52:47 +00005472 if (!sctx->parent_root || sctx->cur_inode_new) {
5473 need_chown = 1;
5474 if (!S_ISLNK(sctx->cur_inode_mode))
Alexander Block31db9f72012-07-25 23:19:24 +02005475 need_chmod = 1;
Alex Lyakase2d044f2012-10-17 13:52:47 +00005476 } else {
5477 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5478 NULL, NULL, &right_mode, &right_uid,
5479 &right_gid, NULL);
5480 if (ret < 0)
5481 goto out;
Alexander Block31db9f72012-07-25 23:19:24 +02005482
Alex Lyakase2d044f2012-10-17 13:52:47 +00005483 if (left_uid != right_uid || left_gid != right_gid)
5484 need_chown = 1;
5485 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5486 need_chmod = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02005487 }
5488
5489 if (S_ISREG(sctx->cur_inode_mode)) {
Josef Bacik16e75492013-10-22 12:18:51 -04005490 if (need_send_hole(sctx)) {
Filipe Manana766b5e52014-03-30 23:02:53 +01005491 if (sctx->cur_inode_last_extent == (u64)-1 ||
5492 sctx->cur_inode_last_extent <
5493 sctx->cur_inode_size) {
Josef Bacik16e75492013-10-22 12:18:51 -04005494 ret = get_last_extent(sctx, (u64)-1);
5495 if (ret)
5496 goto out;
5497 }
5498 if (sctx->cur_inode_last_extent <
5499 sctx->cur_inode_size) {
5500 ret = send_hole(sctx, sctx->cur_inode_size);
5501 if (ret)
5502 goto out;
5503 }
5504 }
Alexander Block31db9f72012-07-25 23:19:24 +02005505 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5506 sctx->cur_inode_size);
5507 if (ret < 0)
5508 goto out;
5509 }
5510
5511 if (need_chown) {
5512 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5513 left_uid, left_gid);
5514 if (ret < 0)
5515 goto out;
5516 }
5517 if (need_chmod) {
5518 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5519 left_mode);
5520 if (ret < 0)
5521 goto out;
5522 }
5523
5524 /*
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005525 * If other directory inodes depended on our current directory
5526 * inode's move/rename, now do their move/rename operations.
Alexander Block31db9f72012-07-25 23:19:24 +02005527 */
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005528 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5529 ret = apply_children_dir_moves(sctx);
5530 if (ret)
5531 goto out;
Filipe Mananafcbd2152014-03-03 12:28:40 +00005532 /*
5533 * Need to send that every time, no matter if it actually
5534 * changed between the two trees as we have done changes to
5535 * the inode before. If our inode is a directory and it's
5536 * waiting to be moved/renamed, we will send its utimes when
5537 * it's moved/renamed, therefore we don't need to do it here.
5538 */
5539 sctx->send_progress = sctx->cur_ino + 1;
5540 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5541 if (ret < 0)
5542 goto out;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00005543 }
5544
Alexander Block31db9f72012-07-25 23:19:24 +02005545out:
5546 return ret;
5547}
5548
5549static int changed_inode(struct send_ctx *sctx,
5550 enum btrfs_compare_tree_result result)
5551{
5552 int ret = 0;
5553 struct btrfs_key *key = sctx->cmp_key;
5554 struct btrfs_inode_item *left_ii = NULL;
5555 struct btrfs_inode_item *right_ii = NULL;
5556 u64 left_gen = 0;
5557 u64 right_gen = 0;
5558
Alexander Block31db9f72012-07-25 23:19:24 +02005559 sctx->cur_ino = key->objectid;
5560 sctx->cur_inode_new_gen = 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005561 sctx->cur_inode_last_extent = (u64)-1;
Alexander Blocke479d9b2012-07-28 16:09:35 +02005562
5563 /*
5564 * Set send_progress to current inode. This will tell all get_cur_xxx
5565 * functions that the current inode's refs are not updated yet. Later,
5566 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5567 */
Alexander Block31db9f72012-07-25 23:19:24 +02005568 sctx->send_progress = sctx->cur_ino;
5569
5570 if (result == BTRFS_COMPARE_TREE_NEW ||
5571 result == BTRFS_COMPARE_TREE_CHANGED) {
5572 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5573 sctx->left_path->slots[0],
5574 struct btrfs_inode_item);
5575 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5576 left_ii);
5577 } else {
5578 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5579 sctx->right_path->slots[0],
5580 struct btrfs_inode_item);
5581 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5582 right_ii);
5583 }
5584 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5585 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5586 sctx->right_path->slots[0],
5587 struct btrfs_inode_item);
5588
5589 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5590 right_ii);
Alexander Block6d85ed02012-08-01 14:48:59 +02005591
5592 /*
5593 * The cur_ino = root dir case is special here. We can't treat
5594 * the inode as deleted+reused because it would generate a
5595 * stream that tries to delete/mkdir the root dir.
5596 */
5597 if (left_gen != right_gen &&
5598 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
Alexander Block31db9f72012-07-25 23:19:24 +02005599 sctx->cur_inode_new_gen = 1;
5600 }
5601
5602 if (result == BTRFS_COMPARE_TREE_NEW) {
5603 sctx->cur_inode_gen = left_gen;
5604 sctx->cur_inode_new = 1;
5605 sctx->cur_inode_deleted = 0;
5606 sctx->cur_inode_size = btrfs_inode_size(
5607 sctx->left_path->nodes[0], left_ii);
5608 sctx->cur_inode_mode = btrfs_inode_mode(
5609 sctx->left_path->nodes[0], left_ii);
Liu Bo644d1942014-02-27 17:29:01 +08005610 sctx->cur_inode_rdev = btrfs_inode_rdev(
5611 sctx->left_path->nodes[0], left_ii);
Alexander Block31db9f72012-07-25 23:19:24 +02005612 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
Alexander Block1f4692d2012-07-28 10:42:24 +02005613 ret = send_create_inode_if_needed(sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02005614 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5615 sctx->cur_inode_gen = right_gen;
5616 sctx->cur_inode_new = 0;
5617 sctx->cur_inode_deleted = 1;
5618 sctx->cur_inode_size = btrfs_inode_size(
5619 sctx->right_path->nodes[0], right_ii);
5620 sctx->cur_inode_mode = btrfs_inode_mode(
5621 sctx->right_path->nodes[0], right_ii);
5622 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
Alexander Block766702e2012-07-28 14:11:31 +02005623 /*
5624 * We need to do some special handling in case the inode was
5625 * reported as changed with a changed generation number. This
5626 * means that the original inode was deleted and new inode
5627 * reused the same inum. So we have to treat the old inode as
5628 * deleted and the new one as new.
5629 */
Alexander Block31db9f72012-07-25 23:19:24 +02005630 if (sctx->cur_inode_new_gen) {
Alexander Block766702e2012-07-28 14:11:31 +02005631 /*
5632 * First, process the inode as if it was deleted.
5633 */
Alexander Block31db9f72012-07-25 23:19:24 +02005634 sctx->cur_inode_gen = right_gen;
5635 sctx->cur_inode_new = 0;
5636 sctx->cur_inode_deleted = 1;
5637 sctx->cur_inode_size = btrfs_inode_size(
5638 sctx->right_path->nodes[0], right_ii);
5639 sctx->cur_inode_mode = btrfs_inode_mode(
5640 sctx->right_path->nodes[0], right_ii);
5641 ret = process_all_refs(sctx,
5642 BTRFS_COMPARE_TREE_DELETED);
5643 if (ret < 0)
5644 goto out;
5645
Alexander Block766702e2012-07-28 14:11:31 +02005646 /*
5647 * Now process the inode as if it was new.
5648 */
Alexander Block31db9f72012-07-25 23:19:24 +02005649 sctx->cur_inode_gen = left_gen;
5650 sctx->cur_inode_new = 1;
5651 sctx->cur_inode_deleted = 0;
5652 sctx->cur_inode_size = btrfs_inode_size(
5653 sctx->left_path->nodes[0], left_ii);
5654 sctx->cur_inode_mode = btrfs_inode_mode(
5655 sctx->left_path->nodes[0], left_ii);
Liu Bo644d1942014-02-27 17:29:01 +08005656 sctx->cur_inode_rdev = btrfs_inode_rdev(
5657 sctx->left_path->nodes[0], left_ii);
Alexander Block1f4692d2012-07-28 10:42:24 +02005658 ret = send_create_inode_if_needed(sctx);
Alexander Block31db9f72012-07-25 23:19:24 +02005659 if (ret < 0)
5660 goto out;
5661
5662 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5663 if (ret < 0)
5664 goto out;
Alexander Blocke479d9b2012-07-28 16:09:35 +02005665 /*
5666 * Advance send_progress now as we did not get into
5667 * process_recorded_refs_if_needed in the new_gen case.
5668 */
5669 sctx->send_progress = sctx->cur_ino + 1;
Alexander Block766702e2012-07-28 14:11:31 +02005670
5671 /*
5672 * Now process all extents and xattrs of the inode as if
5673 * they were all new.
5674 */
Alexander Block31db9f72012-07-25 23:19:24 +02005675 ret = process_all_extents(sctx);
5676 if (ret < 0)
5677 goto out;
5678 ret = process_all_new_xattrs(sctx);
5679 if (ret < 0)
5680 goto out;
5681 } else {
5682 sctx->cur_inode_gen = left_gen;
5683 sctx->cur_inode_new = 0;
5684 sctx->cur_inode_new_gen = 0;
5685 sctx->cur_inode_deleted = 0;
5686 sctx->cur_inode_size = btrfs_inode_size(
5687 sctx->left_path->nodes[0], left_ii);
5688 sctx->cur_inode_mode = btrfs_inode_mode(
5689 sctx->left_path->nodes[0], left_ii);
5690 }
5691 }
5692
5693out:
5694 return ret;
5695}
5696
Alexander Block766702e2012-07-28 14:11:31 +02005697/*
5698 * We have to process new refs before deleted refs, but compare_trees gives us
5699 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5700 * first and later process them in process_recorded_refs.
5701 * For the cur_inode_new_gen case, we skip recording completely because
5702 * changed_inode did already initiate processing of refs. The reason for this is
5703 * that in this case, compare_tree actually compares the refs of 2 different
5704 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5705 * refs of the right tree as deleted and all refs of the left tree as new.
5706 */
Alexander Block31db9f72012-07-25 23:19:24 +02005707static int changed_ref(struct send_ctx *sctx,
5708 enum btrfs_compare_tree_result result)
5709{
5710 int ret = 0;
5711
5712 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5713
5714 if (!sctx->cur_inode_new_gen &&
5715 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5716 if (result == BTRFS_COMPARE_TREE_NEW)
5717 ret = record_new_ref(sctx);
5718 else if (result == BTRFS_COMPARE_TREE_DELETED)
5719 ret = record_deleted_ref(sctx);
5720 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5721 ret = record_changed_ref(sctx);
5722 }
5723
5724 return ret;
5725}
5726
Alexander Block766702e2012-07-28 14:11:31 +02005727/*
5728 * Process new/deleted/changed xattrs. We skip processing in the
5729 * cur_inode_new_gen case because changed_inode did already initiate processing
5730 * of xattrs. The reason is the same as in changed_ref
5731 */
Alexander Block31db9f72012-07-25 23:19:24 +02005732static int changed_xattr(struct send_ctx *sctx,
5733 enum btrfs_compare_tree_result result)
5734{
5735 int ret = 0;
5736
5737 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5738
5739 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5740 if (result == BTRFS_COMPARE_TREE_NEW)
5741 ret = process_new_xattr(sctx);
5742 else if (result == BTRFS_COMPARE_TREE_DELETED)
5743 ret = process_deleted_xattr(sctx);
5744 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5745 ret = process_changed_xattr(sctx);
5746 }
5747
5748 return ret;
5749}
5750
Alexander Block766702e2012-07-28 14:11:31 +02005751/*
5752 * Process new/deleted/changed extents. We skip processing in the
5753 * cur_inode_new_gen case because changed_inode did already initiate processing
5754 * of extents. The reason is the same as in changed_ref
5755 */
Alexander Block31db9f72012-07-25 23:19:24 +02005756static int changed_extent(struct send_ctx *sctx,
5757 enum btrfs_compare_tree_result result)
5758{
5759 int ret = 0;
5760
5761 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5762
5763 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5764 if (result != BTRFS_COMPARE_TREE_DELETED)
5765 ret = process_extent(sctx, sctx->left_path,
5766 sctx->cmp_key);
5767 }
5768
5769 return ret;
5770}
5771
Josef Bacikba5e8f22013-08-16 16:52:55 -04005772static int dir_changed(struct send_ctx *sctx, u64 dir)
5773{
5774 u64 orig_gen, new_gen;
5775 int ret;
5776
5777 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5778 NULL, NULL);
5779 if (ret)
5780 return ret;
5781
5782 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5783 NULL, NULL, NULL);
5784 if (ret)
5785 return ret;
5786
5787 return (orig_gen != new_gen) ? 1 : 0;
5788}
5789
5790static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5791 struct btrfs_key *key)
5792{
5793 struct btrfs_inode_extref *extref;
5794 struct extent_buffer *leaf;
5795 u64 dirid = 0, last_dirid = 0;
5796 unsigned long ptr;
5797 u32 item_size;
5798 u32 cur_offset = 0;
5799 int ref_name_len;
5800 int ret = 0;
5801
5802 /* Easy case, just check this one dirid */
5803 if (key->type == BTRFS_INODE_REF_KEY) {
5804 dirid = key->offset;
5805
5806 ret = dir_changed(sctx, dirid);
5807 goto out;
5808 }
5809
5810 leaf = path->nodes[0];
5811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5812 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5813 while (cur_offset < item_size) {
5814 extref = (struct btrfs_inode_extref *)(ptr +
5815 cur_offset);
5816 dirid = btrfs_inode_extref_parent(leaf, extref);
5817 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5818 cur_offset += ref_name_len + sizeof(*extref);
5819 if (dirid == last_dirid)
5820 continue;
5821 ret = dir_changed(sctx, dirid);
5822 if (ret)
5823 break;
5824 last_dirid = dirid;
5825 }
5826out:
5827 return ret;
5828}
5829
Alexander Block766702e2012-07-28 14:11:31 +02005830/*
5831 * Updates compare related fields in sctx and simply forwards to the actual
5832 * changed_xxx functions.
5833 */
Alexander Block31db9f72012-07-25 23:19:24 +02005834static int changed_cb(struct btrfs_root *left_root,
5835 struct btrfs_root *right_root,
5836 struct btrfs_path *left_path,
5837 struct btrfs_path *right_path,
5838 struct btrfs_key *key,
5839 enum btrfs_compare_tree_result result,
5840 void *ctx)
5841{
5842 int ret = 0;
5843 struct send_ctx *sctx = ctx;
5844
Josef Bacikba5e8f22013-08-16 16:52:55 -04005845 if (result == BTRFS_COMPARE_TREE_SAME) {
Josef Bacik16e75492013-10-22 12:18:51 -04005846 if (key->type == BTRFS_INODE_REF_KEY ||
5847 key->type == BTRFS_INODE_EXTREF_KEY) {
5848 ret = compare_refs(sctx, left_path, key);
5849 if (!ret)
5850 return 0;
5851 if (ret < 0)
5852 return ret;
5853 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5854 return maybe_send_hole(sctx, left_path, key);
5855 } else {
Josef Bacikba5e8f22013-08-16 16:52:55 -04005856 return 0;
Josef Bacik16e75492013-10-22 12:18:51 -04005857 }
Josef Bacikba5e8f22013-08-16 16:52:55 -04005858 result = BTRFS_COMPARE_TREE_CHANGED;
5859 ret = 0;
5860 }
5861
Alexander Block31db9f72012-07-25 23:19:24 +02005862 sctx->left_path = left_path;
5863 sctx->right_path = right_path;
5864 sctx->cmp_key = key;
5865
5866 ret = finish_inode_if_needed(sctx, 0);
5867 if (ret < 0)
5868 goto out;
5869
Alexander Block2981e222012-08-01 14:47:03 +02005870 /* Ignore non-FS objects */
5871 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5872 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5873 goto out;
5874
Alexander Block31db9f72012-07-25 23:19:24 +02005875 if (key->type == BTRFS_INODE_ITEM_KEY)
5876 ret = changed_inode(sctx, result);
Jan Schmidt96b5bd72012-10-15 08:30:45 +00005877 else if (key->type == BTRFS_INODE_REF_KEY ||
5878 key->type == BTRFS_INODE_EXTREF_KEY)
Alexander Block31db9f72012-07-25 23:19:24 +02005879 ret = changed_ref(sctx, result);
5880 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5881 ret = changed_xattr(sctx, result);
5882 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5883 ret = changed_extent(sctx, result);
5884
5885out:
5886 return ret;
5887}
5888
5889static int full_send_tree(struct send_ctx *sctx)
5890{
5891 int ret;
Alexander Block31db9f72012-07-25 23:19:24 +02005892 struct btrfs_root *send_root = sctx->send_root;
5893 struct btrfs_key key;
5894 struct btrfs_key found_key;
5895 struct btrfs_path *path;
5896 struct extent_buffer *eb;
5897 int slot;
Alexander Block31db9f72012-07-25 23:19:24 +02005898
5899 path = alloc_path_for_send();
5900 if (!path)
5901 return -ENOMEM;
5902
Alexander Block31db9f72012-07-25 23:19:24 +02005903 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5904 key.type = BTRFS_INODE_ITEM_KEY;
5905 key.offset = 0;
5906
Alexander Block31db9f72012-07-25 23:19:24 +02005907 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5908 if (ret < 0)
5909 goto out;
5910 if (ret)
5911 goto out_finish;
5912
5913 while (1) {
Alexander Block31db9f72012-07-25 23:19:24 +02005914 eb = path->nodes[0];
5915 slot = path->slots[0];
5916 btrfs_item_key_to_cpu(eb, &found_key, slot);
5917
5918 ret = changed_cb(send_root, NULL, path, NULL,
5919 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5920 if (ret < 0)
5921 goto out;
5922
5923 key.objectid = found_key.objectid;
5924 key.type = found_key.type;
5925 key.offset = found_key.offset + 1;
5926
5927 ret = btrfs_next_item(send_root, path);
5928 if (ret < 0)
5929 goto out;
5930 if (ret) {
5931 ret = 0;
5932 break;
5933 }
5934 }
5935
5936out_finish:
5937 ret = finish_inode_if_needed(sctx, 1);
5938
5939out:
5940 btrfs_free_path(path);
Alexander Block31db9f72012-07-25 23:19:24 +02005941 return ret;
5942}
5943
5944static int send_subvol(struct send_ctx *sctx)
5945{
5946 int ret;
5947
Stefan Behrensc2c71322013-04-10 17:10:52 +00005948 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5949 ret = send_header(sctx);
5950 if (ret < 0)
5951 goto out;
5952 }
Alexander Block31db9f72012-07-25 23:19:24 +02005953
5954 ret = send_subvol_begin(sctx);
5955 if (ret < 0)
5956 goto out;
5957
5958 if (sctx->parent_root) {
5959 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5960 changed_cb, sctx);
5961 if (ret < 0)
5962 goto out;
5963 ret = finish_inode_if_needed(sctx, 1);
5964 if (ret < 0)
5965 goto out;
5966 } else {
5967 ret = full_send_tree(sctx);
5968 if (ret < 0)
5969 goto out;
5970 }
5971
5972out:
Alexander Block31db9f72012-07-25 23:19:24 +02005973 free_recorded_refs(sctx);
5974 return ret;
5975}
5976
Filipe Mananae5fa8f82014-10-21 11:11:41 +01005977/*
5978 * If orphan cleanup did remove any orphans from a root, it means the tree
5979 * was modified and therefore the commit root is not the same as the current
5980 * root anymore. This is a problem, because send uses the commit root and
5981 * therefore can see inode items that don't exist in the current root anymore,
5982 * and for example make calls to btrfs_iget, which will do tree lookups based
5983 * on the current root and not on the commit root. Those lookups will fail,
5984 * returning a -ESTALE error, and making send fail with that error. So make
5985 * sure a send does not see any orphans we have just removed, and that it will
5986 * see the same inodes regardless of whether a transaction commit happened
5987 * before it started (meaning that the commit root will be the same as the
5988 * current root) or not.
5989 */
5990static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
5991{
5992 int i;
5993 struct btrfs_trans_handle *trans = NULL;
5994
5995again:
5996 if (sctx->parent_root &&
5997 sctx->parent_root->node != sctx->parent_root->commit_root)
5998 goto commit_trans;
5999
6000 for (i = 0; i < sctx->clone_roots_cnt; i++)
6001 if (sctx->clone_roots[i].root->node !=
6002 sctx->clone_roots[i].root->commit_root)
6003 goto commit_trans;
6004
6005 if (trans)
6006 return btrfs_end_transaction(trans, sctx->send_root);
6007
6008 return 0;
6009
6010commit_trans:
6011 /* Use any root, all fs roots will get their commit roots updated. */
6012 if (!trans) {
6013 trans = btrfs_join_transaction(sctx->send_root);
6014 if (IS_ERR(trans))
6015 return PTR_ERR(trans);
6016 goto again;
6017 }
6018
6019 return btrfs_commit_transaction(trans, sctx->send_root);
6020}
6021
David Sterba66ef7d62013-12-17 15:07:20 +01006022static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6023{
6024 spin_lock(&root->root_item_lock);
6025 root->send_in_progress--;
6026 /*
6027 * Not much left to do, we don't know why it's unbalanced and
6028 * can't blindly reset it to 0.
6029 */
6030 if (root->send_in_progress < 0)
6031 btrfs_err(root->fs_info,
David Sterba351fd352014-05-15 16:48:20 +02006032 "send_in_progres unbalanced %d root %llu",
David Sterba66ef7d62013-12-17 15:07:20 +01006033 root->send_in_progress, root->root_key.objectid);
6034 spin_unlock(&root->root_item_lock);
6035}
6036
Alexander Block31db9f72012-07-25 23:19:24 +02006037long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
6038{
6039 int ret = 0;
6040 struct btrfs_root *send_root;
6041 struct btrfs_root *clone_root;
6042 struct btrfs_fs_info *fs_info;
6043 struct btrfs_ioctl_send_args *arg = NULL;
6044 struct btrfs_key key;
Alexander Block31db9f72012-07-25 23:19:24 +02006045 struct send_ctx *sctx = NULL;
6046 u32 i;
6047 u64 *clone_sources_tmp = NULL;
David Sterba2c686532013-12-16 17:34:17 +01006048 int clone_sources_to_rollback = 0;
David Sterbae55d1152016-04-11 18:52:02 +02006049 unsigned alloc_size;
Wang Shilong896c14f2014-01-07 17:25:18 +08006050 int sort_clone_roots = 0;
Wang Shilong18f687d2014-01-07 17:25:19 +08006051 int index;
Alexander Block31db9f72012-07-25 23:19:24 +02006052
6053 if (!capable(CAP_SYS_ADMIN))
6054 return -EPERM;
6055
Al Viro496ad9a2013-01-23 17:07:38 -05006056 send_root = BTRFS_I(file_inode(mnt_file))->root;
Alexander Block31db9f72012-07-25 23:19:24 +02006057 fs_info = send_root->fs_info;
6058
Josef Bacik139f8072013-05-20 11:26:50 -04006059 /*
David Sterba2c686532013-12-16 17:34:17 +01006060 * The subvolume must remain read-only during send, protect against
David Sterba521e0542014-04-15 16:41:44 +02006061 * making it RW. This also protects against deletion.
David Sterba2c686532013-12-16 17:34:17 +01006062 */
6063 spin_lock(&send_root->root_item_lock);
6064 send_root->send_in_progress++;
6065 spin_unlock(&send_root->root_item_lock);
6066
6067 /*
Josef Bacik139f8072013-05-20 11:26:50 -04006068 * This is done when we lookup the root, it should already be complete
6069 * by the time we get here.
6070 */
6071 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6072
6073 /*
David Sterba2c686532013-12-16 17:34:17 +01006074 * Userspace tools do the checks and warn the user if it's
6075 * not RO.
6076 */
6077 if (!btrfs_root_readonly(send_root)) {
6078 ret = -EPERM;
6079 goto out;
6080 }
6081
Alexander Block31db9f72012-07-25 23:19:24 +02006082 arg = memdup_user(arg_, sizeof(*arg));
6083 if (IS_ERR(arg)) {
6084 ret = PTR_ERR(arg);
6085 arg = NULL;
6086 goto out;
6087 }
6088
Dan Carpenterf5ecec32016-04-13 09:40:59 +03006089 if (arg->clone_sources_count >
6090 ULLONG_MAX / sizeof(*arg->clone_sources)) {
6091 ret = -EINVAL;
6092 goto out;
6093 }
6094
Alexander Block31db9f72012-07-25 23:19:24 +02006095 if (!access_ok(VERIFY_READ, arg->clone_sources,
Dan Carpenter700ff4f2013-01-10 03:57:25 -05006096 sizeof(*arg->clone_sources) *
6097 arg->clone_sources_count)) {
Alexander Block31db9f72012-07-25 23:19:24 +02006098 ret = -EFAULT;
6099 goto out;
6100 }
6101
Stefan Behrensc2c71322013-04-10 17:10:52 +00006102 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
Mark Fashehcb95e7b2013-02-04 20:54:57 +00006103 ret = -EINVAL;
6104 goto out;
6105 }
6106
David Sterbae780b0d2016-01-18 18:42:13 +01006107 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02006108 if (!sctx) {
6109 ret = -ENOMEM;
6110 goto out;
6111 }
6112
6113 INIT_LIST_HEAD(&sctx->new_refs);
6114 INIT_LIST_HEAD(&sctx->deleted_refs);
David Sterbae780b0d2016-01-18 18:42:13 +01006115 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
Alexander Block31db9f72012-07-25 23:19:24 +02006116 INIT_LIST_HEAD(&sctx->name_cache_list);
6117
Mark Fashehcb95e7b2013-02-04 20:54:57 +00006118 sctx->flags = arg->flags;
6119
Alexander Block31db9f72012-07-25 23:19:24 +02006120 sctx->send_filp = fget(arg->send_fd);
Tsutomu Itohecc7ada2013-04-19 01:04:46 +00006121 if (!sctx->send_filp) {
6122 ret = -EBADF;
Alexander Block31db9f72012-07-25 23:19:24 +02006123 goto out;
6124 }
6125
Alexander Block31db9f72012-07-25 23:19:24 +02006126 sctx->send_root = send_root;
David Sterba521e0542014-04-15 16:41:44 +02006127 /*
6128 * Unlikely but possible, if the subvolume is marked for deletion but
6129 * is slow to remove the directory entry, send can still be started
6130 */
6131 if (btrfs_root_dead(sctx->send_root)) {
6132 ret = -EPERM;
6133 goto out;
6134 }
6135
Alexander Block31db9f72012-07-25 23:19:24 +02006136 sctx->clone_roots_cnt = arg->clone_sources_count;
6137
6138 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
David Sterba6ff48ce2016-04-11 18:40:08 +02006139 sctx->send_buf = kmalloc(sctx->send_max_size, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006140 if (!sctx->send_buf) {
David Sterba6ff48ce2016-04-11 18:40:08 +02006141 sctx->send_buf = vmalloc(sctx->send_max_size);
6142 if (!sctx->send_buf) {
6143 ret = -ENOMEM;
6144 goto out;
6145 }
Alexander Block31db9f72012-07-25 23:19:24 +02006146 }
6147
David Sterbaeb5b75f2016-04-11 18:40:08 +02006148 sctx->read_buf = kmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006149 if (!sctx->read_buf) {
David Sterbaeb5b75f2016-04-11 18:40:08 +02006150 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
6151 if (!sctx->read_buf) {
6152 ret = -ENOMEM;
6153 goto out;
6154 }
Alexander Block31db9f72012-07-25 23:19:24 +02006155 }
6156
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006157 sctx->pending_dir_moves = RB_ROOT;
6158 sctx->waiting_dir_moves = RB_ROOT;
Filipe Manana9dc44212014-02-19 14:31:44 +00006159 sctx->orphan_dirs = RB_ROOT;
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006160
David Sterbae55d1152016-04-11 18:52:02 +02006161 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6162
David Sterbac03d01f2016-04-11 18:40:08 +02006163 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006164 if (!sctx->clone_roots) {
David Sterbac03d01f2016-04-11 18:40:08 +02006165 sctx->clone_roots = vzalloc(alloc_size);
6166 if (!sctx->clone_roots) {
6167 ret = -ENOMEM;
6168 goto out;
6169 }
Alexander Block31db9f72012-07-25 23:19:24 +02006170 }
6171
David Sterbae55d1152016-04-11 18:52:02 +02006172 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6173
Alexander Block31db9f72012-07-25 23:19:24 +02006174 if (arg->clone_sources_count) {
David Sterba2f913062016-04-11 18:40:08 +02006175 clone_sources_tmp = kmalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
Alexander Block31db9f72012-07-25 23:19:24 +02006176 if (!clone_sources_tmp) {
David Sterba2f913062016-04-11 18:40:08 +02006177 clone_sources_tmp = vmalloc(alloc_size);
6178 if (!clone_sources_tmp) {
6179 ret = -ENOMEM;
6180 goto out;
6181 }
Alexander Block31db9f72012-07-25 23:19:24 +02006182 }
6183
6184 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
David Sterbae55d1152016-04-11 18:52:02 +02006185 alloc_size);
Alexander Block31db9f72012-07-25 23:19:24 +02006186 if (ret) {
6187 ret = -EFAULT;
6188 goto out;
6189 }
6190
6191 for (i = 0; i < arg->clone_sources_count; i++) {
6192 key.objectid = clone_sources_tmp[i];
6193 key.type = BTRFS_ROOT_ITEM_KEY;
6194 key.offset = (u64)-1;
Wang Shilong18f687d2014-01-07 17:25:19 +08006195
6196 index = srcu_read_lock(&fs_info->subvol_srcu);
6197
Alexander Block31db9f72012-07-25 23:19:24 +02006198 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
Alexander Block31db9f72012-07-25 23:19:24 +02006199 if (IS_ERR(clone_root)) {
Wang Shilong18f687d2014-01-07 17:25:19 +08006200 srcu_read_unlock(&fs_info->subvol_srcu, index);
Alexander Block31db9f72012-07-25 23:19:24 +02006201 ret = PTR_ERR(clone_root);
6202 goto out;
6203 }
David Sterba2c686532013-12-16 17:34:17 +01006204 spin_lock(&clone_root->root_item_lock);
Filipe Manana5cc2b172015-03-02 20:53:52 +00006205 if (!btrfs_root_readonly(clone_root) ||
6206 btrfs_root_dead(clone_root)) {
David Sterba2c686532013-12-16 17:34:17 +01006207 spin_unlock(&clone_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006208 srcu_read_unlock(&fs_info->subvol_srcu, index);
David Sterba2c686532013-12-16 17:34:17 +01006209 ret = -EPERM;
6210 goto out;
6211 }
Filipe Manana2f1f4652015-03-02 20:53:53 +00006212 clone_root->send_in_progress++;
David Sterba2c686532013-12-16 17:34:17 +01006213 spin_unlock(&clone_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006214 srcu_read_unlock(&fs_info->subvol_srcu, index);
6215
Alexander Block31db9f72012-07-25 23:19:24 +02006216 sctx->clone_roots[i].root = clone_root;
Filipe Manana2f1f4652015-03-02 20:53:53 +00006217 clone_sources_to_rollback = i + 1;
Alexander Block31db9f72012-07-25 23:19:24 +02006218 }
David Sterba2f913062016-04-11 18:40:08 +02006219 kvfree(clone_sources_tmp);
Alexander Block31db9f72012-07-25 23:19:24 +02006220 clone_sources_tmp = NULL;
6221 }
6222
6223 if (arg->parent_root) {
6224 key.objectid = arg->parent_root;
6225 key.type = BTRFS_ROOT_ITEM_KEY;
6226 key.offset = (u64)-1;
Wang Shilong18f687d2014-01-07 17:25:19 +08006227
6228 index = srcu_read_lock(&fs_info->subvol_srcu);
6229
Alexander Block31db9f72012-07-25 23:19:24 +02006230 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
Stefan Behrensb1b19592013-05-13 14:42:57 +00006231 if (IS_ERR(sctx->parent_root)) {
Wang Shilong18f687d2014-01-07 17:25:19 +08006232 srcu_read_unlock(&fs_info->subvol_srcu, index);
Stefan Behrensb1b19592013-05-13 14:42:57 +00006233 ret = PTR_ERR(sctx->parent_root);
Alexander Block31db9f72012-07-25 23:19:24 +02006234 goto out;
6235 }
Wang Shilong18f687d2014-01-07 17:25:19 +08006236
David Sterba2c686532013-12-16 17:34:17 +01006237 spin_lock(&sctx->parent_root->root_item_lock);
6238 sctx->parent_root->send_in_progress++;
David Sterba521e0542014-04-15 16:41:44 +02006239 if (!btrfs_root_readonly(sctx->parent_root) ||
6240 btrfs_root_dead(sctx->parent_root)) {
David Sterba2c686532013-12-16 17:34:17 +01006241 spin_unlock(&sctx->parent_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006242 srcu_read_unlock(&fs_info->subvol_srcu, index);
David Sterba2c686532013-12-16 17:34:17 +01006243 ret = -EPERM;
6244 goto out;
6245 }
6246 spin_unlock(&sctx->parent_root->root_item_lock);
Wang Shilong18f687d2014-01-07 17:25:19 +08006247
6248 srcu_read_unlock(&fs_info->subvol_srcu, index);
Alexander Block31db9f72012-07-25 23:19:24 +02006249 }
6250
6251 /*
6252 * Clones from send_root are allowed, but only if the clone source
6253 * is behind the current send position. This is checked while searching
6254 * for possible clone sources.
6255 */
6256 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6257
6258 /* We do a bsearch later */
6259 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6260 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6261 NULL);
Wang Shilong896c14f2014-01-07 17:25:18 +08006262 sort_clone_roots = 1;
Alexander Block31db9f72012-07-25 23:19:24 +02006263
Filipe Mananae5fa8f82014-10-21 11:11:41 +01006264 ret = ensure_commit_roots_uptodate(sctx);
6265 if (ret)
6266 goto out;
6267
David Sterba2755a0d2014-07-31 00:43:18 +02006268 current->journal_info = BTRFS_SEND_TRANS_STUB;
Alexander Block31db9f72012-07-25 23:19:24 +02006269 ret = send_subvol(sctx);
Josef Bacika26e8c92014-03-28 17:07:27 -04006270 current->journal_info = NULL;
Alexander Block31db9f72012-07-25 23:19:24 +02006271 if (ret < 0)
6272 goto out;
6273
Stefan Behrensc2c71322013-04-10 17:10:52 +00006274 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6275 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6276 if (ret < 0)
6277 goto out;
6278 ret = send_cmd(sctx);
6279 if (ret < 0)
6280 goto out;
6281 }
Alexander Block31db9f72012-07-25 23:19:24 +02006282
6283out:
Filipe David Borba Manana9f037402014-01-22 10:00:53 +00006284 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6285 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6286 struct rb_node *n;
6287 struct pending_dir_move *pm;
6288
6289 n = rb_first(&sctx->pending_dir_moves);
6290 pm = rb_entry(n, struct pending_dir_move, node);
6291 while (!list_empty(&pm->list)) {
6292 struct pending_dir_move *pm2;
6293
6294 pm2 = list_first_entry(&pm->list,
6295 struct pending_dir_move, list);
6296 free_pending_move(sctx, pm2);
6297 }
6298 free_pending_move(sctx, pm);
6299 }
6300
6301 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6302 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6303 struct rb_node *n;
6304 struct waiting_dir_move *dm;
6305
6306 n = rb_first(&sctx->waiting_dir_moves);
6307 dm = rb_entry(n, struct waiting_dir_move, node);
6308 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6309 kfree(dm);
6310 }
6311
Filipe Manana9dc44212014-02-19 14:31:44 +00006312 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6313 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6314 struct rb_node *n;
6315 struct orphan_dir_info *odi;
6316
6317 n = rb_first(&sctx->orphan_dirs);
6318 odi = rb_entry(n, struct orphan_dir_info, node);
6319 free_orphan_dir_info(sctx, odi);
6320 }
6321
Wang Shilong896c14f2014-01-07 17:25:18 +08006322 if (sort_clone_roots) {
6323 for (i = 0; i < sctx->clone_roots_cnt; i++)
6324 btrfs_root_dec_send_in_progress(
6325 sctx->clone_roots[i].root);
6326 } else {
6327 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6328 btrfs_root_dec_send_in_progress(
6329 sctx->clone_roots[i].root);
6330
6331 btrfs_root_dec_send_in_progress(send_root);
6332 }
David Sterba66ef7d62013-12-17 15:07:20 +01006333 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6334 btrfs_root_dec_send_in_progress(sctx->parent_root);
David Sterba2c686532013-12-16 17:34:17 +01006335
Alexander Block31db9f72012-07-25 23:19:24 +02006336 kfree(arg);
David Sterba2f913062016-04-11 18:40:08 +02006337 kvfree(clone_sources_tmp);
Alexander Block31db9f72012-07-25 23:19:24 +02006338
6339 if (sctx) {
6340 if (sctx->send_filp)
6341 fput(sctx->send_filp);
6342
David Sterbac03d01f2016-04-11 18:40:08 +02006343 kvfree(sctx->clone_roots);
David Sterba6ff48ce2016-04-11 18:40:08 +02006344 kvfree(sctx->send_buf);
David Sterbaeb5b75f2016-04-11 18:40:08 +02006345 kvfree(sctx->read_buf);
Alexander Block31db9f72012-07-25 23:19:24 +02006346
6347 name_cache_free(sctx);
6348
6349 kfree(sctx);
6350 }
6351
6352 return ret;
6353}