2 * Copyright (C) 2012 Alexander Block. All rights reserved.
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.
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.
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.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
35 #include "btrfs_inode.h"
36 #include "transaction.h"
38 static int g_verbose
= 0;
40 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
56 unsigned short buf_len
:15;
57 unsigned short reversed
:1;
61 * Average path length does not exceed 200 bytes, we'll have
62 * better packing in the slab and higher chance to satisfy
63 * a allocation later during send.
68 #define FS_PATH_INLINE_SIZE \
69 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
72 /* reused for each extent */
74 struct btrfs_root
*root
;
81 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
82 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
85 struct file
*send_filp
;
91 u64 cmd_send_size
[BTRFS_SEND_C_MAX
+ 1];
92 u64 flags
; /* 'flags' member of btrfs_ioctl_send_args is u64 */
94 struct btrfs_root
*send_root
;
95 struct btrfs_root
*parent_root
;
96 struct clone_root
*clone_roots
;
99 /* current state of the compare_tree call */
100 struct btrfs_path
*left_path
;
101 struct btrfs_path
*right_path
;
102 struct btrfs_key
*cmp_key
;
105 * infos of the currently processed inode. In case of deleted inodes,
106 * these are the values from the deleted inode.
111 int cur_inode_new_gen
;
112 int cur_inode_deleted
;
116 u64 cur_inode_last_extent
;
120 struct list_head new_refs
;
121 struct list_head deleted_refs
;
123 struct radix_tree_root name_cache
;
124 struct list_head name_cache_list
;
127 struct file_ra_state ra
;
132 * We process inodes by their increasing order, so if before an
133 * incremental send we reverse the parent/child relationship of
134 * directories such that a directory with a lower inode number was
135 * the parent of a directory with a higher inode number, and the one
136 * becoming the new parent got renamed too, we can't rename/move the
137 * directory with lower inode number when we finish processing it - we
138 * must process the directory with higher inode number first, then
139 * rename/move it and then rename/move the directory with lower inode
140 * number. Example follows.
142 * Tree state when the first send was performed:
154 * Tree state when the second (incremental) send is performed:
163 * The sequence of steps that lead to the second state was:
165 * mv /a/b/c/d /a/b/c2/d2
166 * mv /a/b/c /a/b/c2/d2/cc
168 * "c" has lower inode number, but we can't move it (2nd mv operation)
169 * before we move "d", which has higher inode number.
171 * So we just memorize which move/rename operations must be performed
172 * later when their respective parent is processed and moved/renamed.
175 /* Indexed by parent directory inode number. */
176 struct rb_root pending_dir_moves
;
179 * Reverse index, indexed by the inode number of a directory that
180 * is waiting for the move/rename of its immediate parent before its
181 * own move/rename can be performed.
183 struct rb_root waiting_dir_moves
;
186 * A directory that is going to be rm'ed might have a child directory
187 * which is in the pending directory moves index above. In this case,
188 * the directory can only be removed after the move/rename of its child
189 * is performed. Example:
209 * Sequence of steps that lead to the send snapshot:
210 * rm -f /a/b/c/foo.txt
212 * mv /a/b/c/x /a/b/YY
215 * When the child is processed, its move/rename is delayed until its
216 * parent is processed (as explained above), but all other operations
217 * like update utimes, chown, chgrp, etc, are performed and the paths
218 * that it uses for those operations must use the orphanized name of
219 * its parent (the directory we're going to rm later), so we need to
220 * memorize that name.
222 * Indexed by the inode number of the directory to be deleted.
224 struct rb_root orphan_dirs
;
227 struct pending_dir_move
{
229 struct list_head list
;
233 struct list_head update_refs
;
236 struct waiting_dir_move
{
240 * There might be some directory that could not be removed because it
241 * was waiting for this directory inode to be moved first. Therefore
242 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
247 struct orphan_dir_info
{
253 struct name_cache_entry
{
254 struct list_head list
;
256 * radix_tree has only 32bit entries but we need to handle 64bit inums.
257 * We use the lower 32bit of the 64bit inum to store it in the tree. If
258 * more then one inum would fall into the same entry, we use radix_list
259 * to store the additional entries. radix_list is also used to store
260 * entries where two entries have the same inum but different
263 struct list_head radix_list
;
269 int need_later_update
;
274 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
);
276 static struct waiting_dir_move
*
277 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
);
279 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
);
281 static int need_send_hole(struct send_ctx
*sctx
)
283 return (sctx
->parent_root
&& !sctx
->cur_inode_new
&&
284 !sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
&&
285 S_ISREG(sctx
->cur_inode_mode
));
288 static void fs_path_reset(struct fs_path
*p
)
291 p
->start
= p
->buf
+ p
->buf_len
- 1;
301 static struct fs_path
*fs_path_alloc(void)
305 p
= kmalloc(sizeof(*p
), GFP_NOFS
);
309 p
->buf
= p
->inline_buf
;
310 p
->buf_len
= FS_PATH_INLINE_SIZE
;
315 static struct fs_path
*fs_path_alloc_reversed(void)
327 static void fs_path_free(struct fs_path
*p
)
331 if (p
->buf
!= p
->inline_buf
)
336 static int fs_path_len(struct fs_path
*p
)
338 return p
->end
- p
->start
;
341 static int fs_path_ensure_buf(struct fs_path
*p
, int len
)
349 if (p
->buf_len
>= len
)
352 if (len
> PATH_MAX
) {
357 path_len
= p
->end
- p
->start
;
358 old_buf_len
= p
->buf_len
;
361 * First time the inline_buf does not suffice
363 if (p
->buf
== p
->inline_buf
) {
364 tmp_buf
= kmalloc(len
, GFP_NOFS
);
366 memcpy(tmp_buf
, p
->buf
, old_buf_len
);
368 tmp_buf
= krealloc(p
->buf
, len
, GFP_NOFS
);
374 * The real size of the buffer is bigger, this will let the fast path
375 * happen most of the time
377 p
->buf_len
= ksize(p
->buf
);
380 tmp_buf
= p
->buf
+ old_buf_len
- path_len
- 1;
381 p
->end
= p
->buf
+ p
->buf_len
- 1;
382 p
->start
= p
->end
- path_len
;
383 memmove(p
->start
, tmp_buf
, path_len
+ 1);
386 p
->end
= p
->start
+ path_len
;
391 static int fs_path_prepare_for_add(struct fs_path
*p
, int name_len
,
397 new_len
= p
->end
- p
->start
+ name_len
;
398 if (p
->start
!= p
->end
)
400 ret
= fs_path_ensure_buf(p
, new_len
);
405 if (p
->start
!= p
->end
)
407 p
->start
-= name_len
;
408 *prepared
= p
->start
;
410 if (p
->start
!= p
->end
)
421 static int fs_path_add(struct fs_path
*p
, const char *name
, int name_len
)
426 ret
= fs_path_prepare_for_add(p
, name_len
, &prepared
);
429 memcpy(prepared
, name
, name_len
);
435 static int fs_path_add_path(struct fs_path
*p
, struct fs_path
*p2
)
440 ret
= fs_path_prepare_for_add(p
, p2
->end
- p2
->start
, &prepared
);
443 memcpy(prepared
, p2
->start
, p2
->end
- p2
->start
);
449 static int fs_path_add_from_extent_buffer(struct fs_path
*p
,
450 struct extent_buffer
*eb
,
451 unsigned long off
, int len
)
456 ret
= fs_path_prepare_for_add(p
, len
, &prepared
);
460 read_extent_buffer(eb
, prepared
, off
, len
);
466 static int fs_path_copy(struct fs_path
*p
, struct fs_path
*from
)
470 p
->reversed
= from
->reversed
;
473 ret
= fs_path_add_path(p
, from
);
479 static void fs_path_unreverse(struct fs_path
*p
)
488 len
= p
->end
- p
->start
;
490 p
->end
= p
->start
+ len
;
491 memmove(p
->start
, tmp
, len
+ 1);
495 static struct btrfs_path
*alloc_path_for_send(void)
497 struct btrfs_path
*path
;
499 path
= btrfs_alloc_path();
502 path
->search_commit_root
= 1;
503 path
->skip_locking
= 1;
504 path
->need_commit_sem
= 1;
508 static int write_buf(struct file
*filp
, const void *buf
, u32 len
, loff_t
*off
)
518 ret
= vfs_write(filp
, (char *)buf
+ pos
, len
- pos
, off
);
519 /* TODO handle that correctly */
520 /*if (ret == -ERESTARTSYS) {
539 static int tlv_put(struct send_ctx
*sctx
, u16 attr
, const void *data
, int len
)
541 struct btrfs_tlv_header
*hdr
;
542 int total_len
= sizeof(*hdr
) + len
;
543 int left
= sctx
->send_max_size
- sctx
->send_size
;
545 if (unlikely(left
< total_len
))
548 hdr
= (struct btrfs_tlv_header
*) (sctx
->send_buf
+ sctx
->send_size
);
549 hdr
->tlv_type
= cpu_to_le16(attr
);
550 hdr
->tlv_len
= cpu_to_le16(len
);
551 memcpy(hdr
+ 1, data
, len
);
552 sctx
->send_size
+= total_len
;
557 #define TLV_PUT_DEFINE_INT(bits) \
558 static int tlv_put_u##bits(struct send_ctx *sctx, \
559 u##bits attr, u##bits value) \
561 __le##bits __tmp = cpu_to_le##bits(value); \
562 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
565 TLV_PUT_DEFINE_INT(64)
567 static int tlv_put_string(struct send_ctx
*sctx
, u16 attr
,
568 const char *str
, int len
)
572 return tlv_put(sctx
, attr
, str
, len
);
575 static int tlv_put_uuid(struct send_ctx
*sctx
, u16 attr
,
578 return tlv_put(sctx
, attr
, uuid
, BTRFS_UUID_SIZE
);
581 static int tlv_put_btrfs_timespec(struct send_ctx
*sctx
, u16 attr
,
582 struct extent_buffer
*eb
,
583 struct btrfs_timespec
*ts
)
585 struct btrfs_timespec bts
;
586 read_extent_buffer(eb
, &bts
, (unsigned long)ts
, sizeof(bts
));
587 return tlv_put(sctx
, attr
, &bts
, sizeof(bts
));
591 #define TLV_PUT(sctx, attrtype, attrlen, data) \
593 ret = tlv_put(sctx, attrtype, attrlen, data); \
595 goto tlv_put_failure; \
598 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
600 ret = tlv_put_u##bits(sctx, attrtype, value); \
602 goto tlv_put_failure; \
605 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
606 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
607 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
608 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
609 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
611 ret = tlv_put_string(sctx, attrtype, str, len); \
613 goto tlv_put_failure; \
615 #define TLV_PUT_PATH(sctx, attrtype, p) \
617 ret = tlv_put_string(sctx, attrtype, p->start, \
618 p->end - p->start); \
620 goto tlv_put_failure; \
622 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
624 ret = tlv_put_uuid(sctx, attrtype, uuid); \
626 goto tlv_put_failure; \
628 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
630 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
632 goto tlv_put_failure; \
635 static int send_header(struct send_ctx
*sctx
)
637 struct btrfs_stream_header hdr
;
639 strcpy(hdr
.magic
, BTRFS_SEND_STREAM_MAGIC
);
640 hdr
.version
= cpu_to_le32(BTRFS_SEND_STREAM_VERSION
);
642 return write_buf(sctx
->send_filp
, &hdr
, sizeof(hdr
),
647 * For each command/item we want to send to userspace, we call this function.
649 static int begin_cmd(struct send_ctx
*sctx
, int cmd
)
651 struct btrfs_cmd_header
*hdr
;
653 if (WARN_ON(!sctx
->send_buf
))
656 BUG_ON(sctx
->send_size
);
658 sctx
->send_size
+= sizeof(*hdr
);
659 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
660 hdr
->cmd
= cpu_to_le16(cmd
);
665 static int send_cmd(struct send_ctx
*sctx
)
668 struct btrfs_cmd_header
*hdr
;
671 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
672 hdr
->len
= cpu_to_le32(sctx
->send_size
- sizeof(*hdr
));
675 crc
= btrfs_crc32c(0, (unsigned char *)sctx
->send_buf
, sctx
->send_size
);
676 hdr
->crc
= cpu_to_le32(crc
);
678 ret
= write_buf(sctx
->send_filp
, sctx
->send_buf
, sctx
->send_size
,
681 sctx
->total_send_size
+= sctx
->send_size
;
682 sctx
->cmd_send_size
[le16_to_cpu(hdr
->cmd
)] += sctx
->send_size
;
689 * Sends a move instruction to user space
691 static int send_rename(struct send_ctx
*sctx
,
692 struct fs_path
*from
, struct fs_path
*to
)
696 verbose_printk("btrfs: send_rename %s -> %s\n", from
->start
, to
->start
);
698 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RENAME
);
702 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, from
);
703 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_TO
, to
);
705 ret
= send_cmd(sctx
);
713 * Sends a link instruction to user space
715 static int send_link(struct send_ctx
*sctx
,
716 struct fs_path
*path
, struct fs_path
*lnk
)
720 verbose_printk("btrfs: send_link %s -> %s\n", path
->start
, lnk
->start
);
722 ret
= begin_cmd(sctx
, BTRFS_SEND_C_LINK
);
726 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
727 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, lnk
);
729 ret
= send_cmd(sctx
);
737 * Sends an unlink instruction to user space
739 static int send_unlink(struct send_ctx
*sctx
, struct fs_path
*path
)
743 verbose_printk("btrfs: send_unlink %s\n", path
->start
);
745 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UNLINK
);
749 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
751 ret
= send_cmd(sctx
);
759 * Sends a rmdir instruction to user space
761 static int send_rmdir(struct send_ctx
*sctx
, struct fs_path
*path
)
765 verbose_printk("btrfs: send_rmdir %s\n", path
->start
);
767 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RMDIR
);
771 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
773 ret
= send_cmd(sctx
);
781 * Helper function to retrieve some fields from an inode item.
783 static int __get_inode_info(struct btrfs_root
*root
, struct btrfs_path
*path
,
784 u64 ino
, u64
*size
, u64
*gen
, u64
*mode
, u64
*uid
,
788 struct btrfs_inode_item
*ii
;
789 struct btrfs_key key
;
792 key
.type
= BTRFS_INODE_ITEM_KEY
;
794 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
801 ii
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
802 struct btrfs_inode_item
);
804 *size
= btrfs_inode_size(path
->nodes
[0], ii
);
806 *gen
= btrfs_inode_generation(path
->nodes
[0], ii
);
808 *mode
= btrfs_inode_mode(path
->nodes
[0], ii
);
810 *uid
= btrfs_inode_uid(path
->nodes
[0], ii
);
812 *gid
= btrfs_inode_gid(path
->nodes
[0], ii
);
814 *rdev
= btrfs_inode_rdev(path
->nodes
[0], ii
);
819 static int get_inode_info(struct btrfs_root
*root
,
820 u64 ino
, u64
*size
, u64
*gen
,
821 u64
*mode
, u64
*uid
, u64
*gid
,
824 struct btrfs_path
*path
;
827 path
= alloc_path_for_send();
830 ret
= __get_inode_info(root
, path
, ino
, size
, gen
, mode
, uid
, gid
,
832 btrfs_free_path(path
);
836 typedef int (*iterate_inode_ref_t
)(int num
, u64 dir
, int index
,
841 * Helper function to iterate the entries in ONE btrfs_inode_ref or
842 * btrfs_inode_extref.
843 * The iterate callback may return a non zero value to stop iteration. This can
844 * be a negative value for error codes or 1 to simply stop it.
846 * path must point to the INODE_REF or INODE_EXTREF when called.
848 static int iterate_inode_ref(struct btrfs_root
*root
, struct btrfs_path
*path
,
849 struct btrfs_key
*found_key
, int resolve
,
850 iterate_inode_ref_t iterate
, void *ctx
)
852 struct extent_buffer
*eb
= path
->nodes
[0];
853 struct btrfs_item
*item
;
854 struct btrfs_inode_ref
*iref
;
855 struct btrfs_inode_extref
*extref
;
856 struct btrfs_path
*tmp_path
;
860 int slot
= path
->slots
[0];
867 unsigned long name_off
;
868 unsigned long elem_size
;
871 p
= fs_path_alloc_reversed();
875 tmp_path
= alloc_path_for_send();
882 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
883 ptr
= (unsigned long)btrfs_item_ptr(eb
, slot
,
884 struct btrfs_inode_ref
);
885 item
= btrfs_item_nr(slot
);
886 total
= btrfs_item_size(eb
, item
);
887 elem_size
= sizeof(*iref
);
889 ptr
= btrfs_item_ptr_offset(eb
, slot
);
890 total
= btrfs_item_size_nr(eb
, slot
);
891 elem_size
= sizeof(*extref
);
894 while (cur
< total
) {
897 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
898 iref
= (struct btrfs_inode_ref
*)(ptr
+ cur
);
899 name_len
= btrfs_inode_ref_name_len(eb
, iref
);
900 name_off
= (unsigned long)(iref
+ 1);
901 index
= btrfs_inode_ref_index(eb
, iref
);
902 dir
= found_key
->offset
;
904 extref
= (struct btrfs_inode_extref
*)(ptr
+ cur
);
905 name_len
= btrfs_inode_extref_name_len(eb
, extref
);
906 name_off
= (unsigned long)&extref
->name
;
907 index
= btrfs_inode_extref_index(eb
, extref
);
908 dir
= btrfs_inode_extref_parent(eb
, extref
);
912 start
= btrfs_ref_to_path(root
, tmp_path
, name_len
,
916 ret
= PTR_ERR(start
);
919 if (start
< p
->buf
) {
920 /* overflow , try again with larger buffer */
921 ret
= fs_path_ensure_buf(p
,
922 p
->buf_len
+ p
->buf
- start
);
925 start
= btrfs_ref_to_path(root
, tmp_path
,
930 ret
= PTR_ERR(start
);
933 BUG_ON(start
< p
->buf
);
937 ret
= fs_path_add_from_extent_buffer(p
, eb
, name_off
,
943 cur
+= elem_size
+ name_len
;
944 ret
= iterate(num
, dir
, index
, p
, ctx
);
951 btrfs_free_path(tmp_path
);
956 typedef int (*iterate_dir_item_t
)(int num
, struct btrfs_key
*di_key
,
957 const char *name
, int name_len
,
958 const char *data
, int data_len
,
962 * Helper function to iterate the entries in ONE btrfs_dir_item.
963 * The iterate callback may return a non zero value to stop iteration. This can
964 * be a negative value for error codes or 1 to simply stop it.
966 * path must point to the dir item when called.
968 static int iterate_dir_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
969 struct btrfs_key
*found_key
,
970 iterate_dir_item_t iterate
, void *ctx
)
973 struct extent_buffer
*eb
;
974 struct btrfs_item
*item
;
975 struct btrfs_dir_item
*di
;
976 struct btrfs_key di_key
;
978 const int buf_len
= PATH_MAX
;
988 buf
= kmalloc(buf_len
, GFP_NOFS
);
995 slot
= path
->slots
[0];
996 item
= btrfs_item_nr(slot
);
997 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
1000 total
= btrfs_item_size(eb
, item
);
1003 while (cur
< total
) {
1004 name_len
= btrfs_dir_name_len(eb
, di
);
1005 data_len
= btrfs_dir_data_len(eb
, di
);
1006 type
= btrfs_dir_type(eb
, di
);
1007 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
1012 if (name_len
+ data_len
> buf_len
) {
1013 ret
= -ENAMETOOLONG
;
1017 read_extent_buffer(eb
, buf
, (unsigned long)(di
+ 1),
1018 name_len
+ data_len
);
1020 len
= sizeof(*di
) + name_len
+ data_len
;
1021 di
= (struct btrfs_dir_item
*)((char *)di
+ len
);
1024 ret
= iterate(num
, &di_key
, buf
, name_len
, buf
+ name_len
,
1025 data_len
, type
, ctx
);
1041 static int __copy_first_ref(int num
, u64 dir
, int index
,
1042 struct fs_path
*p
, void *ctx
)
1045 struct fs_path
*pt
= ctx
;
1047 ret
= fs_path_copy(pt
, p
);
1051 /* we want the first only */
1056 * Retrieve the first path of an inode. If an inode has more then one
1057 * ref/hardlink, this is ignored.
1059 static int get_inode_path(struct btrfs_root
*root
,
1060 u64 ino
, struct fs_path
*path
)
1063 struct btrfs_key key
, found_key
;
1064 struct btrfs_path
*p
;
1066 p
= alloc_path_for_send();
1070 fs_path_reset(path
);
1073 key
.type
= BTRFS_INODE_REF_KEY
;
1076 ret
= btrfs_search_slot_for_read(root
, &key
, p
, 1, 0);
1083 btrfs_item_key_to_cpu(p
->nodes
[0], &found_key
, p
->slots
[0]);
1084 if (found_key
.objectid
!= ino
||
1085 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1086 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1091 ret
= iterate_inode_ref(root
, p
, &found_key
, 1,
1092 __copy_first_ref
, path
);
1102 struct backref_ctx
{
1103 struct send_ctx
*sctx
;
1105 struct btrfs_path
*path
;
1106 /* number of total found references */
1110 * used for clones found in send_root. clones found behind cur_objectid
1111 * and cur_offset are not considered as allowed clones.
1116 /* may be truncated in case it's the last extent in a file */
1119 /* Just to check for bugs in backref resolving */
1123 static int __clone_root_cmp_bsearch(const void *key
, const void *elt
)
1125 u64 root
= (u64
)(uintptr_t)key
;
1126 struct clone_root
*cr
= (struct clone_root
*)elt
;
1128 if (root
< cr
->root
->objectid
)
1130 if (root
> cr
->root
->objectid
)
1135 static int __clone_root_cmp_sort(const void *e1
, const void *e2
)
1137 struct clone_root
*cr1
= (struct clone_root
*)e1
;
1138 struct clone_root
*cr2
= (struct clone_root
*)e2
;
1140 if (cr1
->root
->objectid
< cr2
->root
->objectid
)
1142 if (cr1
->root
->objectid
> cr2
->root
->objectid
)
1148 * Called for every backref that is found for the current extent.
1149 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1151 static int __iterate_backrefs(u64 ino
, u64 offset
, u64 root
, void *ctx_
)
1153 struct backref_ctx
*bctx
= ctx_
;
1154 struct clone_root
*found
;
1158 /* First check if the root is in the list of accepted clone sources */
1159 found
= bsearch((void *)(uintptr_t)root
, bctx
->sctx
->clone_roots
,
1160 bctx
->sctx
->clone_roots_cnt
,
1161 sizeof(struct clone_root
),
1162 __clone_root_cmp_bsearch
);
1166 if (found
->root
== bctx
->sctx
->send_root
&&
1167 ino
== bctx
->cur_objectid
&&
1168 offset
== bctx
->cur_offset
) {
1169 bctx
->found_itself
= 1;
1173 * There are inodes that have extents that lie behind its i_size. Don't
1174 * accept clones from these extents.
1176 ret
= __get_inode_info(found
->root
, bctx
->path
, ino
, &i_size
, NULL
, NULL
,
1178 btrfs_release_path(bctx
->path
);
1182 if (offset
+ bctx
->extent_len
> i_size
)
1186 * Make sure we don't consider clones from send_root that are
1187 * behind the current inode/offset.
1189 if (found
->root
== bctx
->sctx
->send_root
) {
1191 * TODO for the moment we don't accept clones from the inode
1192 * that is currently send. We may change this when
1193 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1196 if (ino
>= bctx
->cur_objectid
)
1199 if (ino
> bctx
->cur_objectid
)
1201 if (offset
+ bctx
->extent_len
> bctx
->cur_offset
)
1207 found
->found_refs
++;
1208 if (ino
< found
->ino
) {
1210 found
->offset
= offset
;
1211 } else if (found
->ino
== ino
) {
1213 * same extent found more then once in the same file.
1215 if (found
->offset
> offset
+ bctx
->extent_len
)
1216 found
->offset
= offset
;
1223 * Given an inode, offset and extent item, it finds a good clone for a clone
1224 * instruction. Returns -ENOENT when none could be found. The function makes
1225 * sure that the returned clone is usable at the point where sending is at the
1226 * moment. This means, that no clones are accepted which lie behind the current
1229 * path must point to the extent item when called.
1231 static int find_extent_clone(struct send_ctx
*sctx
,
1232 struct btrfs_path
*path
,
1233 u64 ino
, u64 data_offset
,
1235 struct clone_root
**found
)
1242 u64 extent_item_pos
;
1244 struct btrfs_file_extent_item
*fi
;
1245 struct extent_buffer
*eb
= path
->nodes
[0];
1246 struct backref_ctx
*backref_ctx
= NULL
;
1247 struct clone_root
*cur_clone_root
;
1248 struct btrfs_key found_key
;
1249 struct btrfs_path
*tmp_path
;
1253 tmp_path
= alloc_path_for_send();
1257 /* We only use this path under the commit sem */
1258 tmp_path
->need_commit_sem
= 0;
1260 backref_ctx
= kmalloc(sizeof(*backref_ctx
), GFP_NOFS
);
1266 backref_ctx
->path
= tmp_path
;
1268 if (data_offset
>= ino_size
) {
1270 * There may be extents that lie behind the file's size.
1271 * I at least had this in combination with snapshotting while
1272 * writing large files.
1278 fi
= btrfs_item_ptr(eb
, path
->slots
[0],
1279 struct btrfs_file_extent_item
);
1280 extent_type
= btrfs_file_extent_type(eb
, fi
);
1281 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1285 compressed
= btrfs_file_extent_compression(eb
, fi
);
1287 num_bytes
= btrfs_file_extent_num_bytes(eb
, fi
);
1288 disk_byte
= btrfs_file_extent_disk_bytenr(eb
, fi
);
1289 if (disk_byte
== 0) {
1293 logical
= disk_byte
+ btrfs_file_extent_offset(eb
, fi
);
1295 down_read(&sctx
->send_root
->fs_info
->commit_root_sem
);
1296 ret
= extent_from_logical(sctx
->send_root
->fs_info
, disk_byte
, tmp_path
,
1297 &found_key
, &flags
);
1298 up_read(&sctx
->send_root
->fs_info
->commit_root_sem
);
1299 btrfs_release_path(tmp_path
);
1303 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1309 * Setup the clone roots.
1311 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1312 cur_clone_root
= sctx
->clone_roots
+ i
;
1313 cur_clone_root
->ino
= (u64
)-1;
1314 cur_clone_root
->offset
= 0;
1315 cur_clone_root
->found_refs
= 0;
1318 backref_ctx
->sctx
= sctx
;
1319 backref_ctx
->found
= 0;
1320 backref_ctx
->cur_objectid
= ino
;
1321 backref_ctx
->cur_offset
= data_offset
;
1322 backref_ctx
->found_itself
= 0;
1323 backref_ctx
->extent_len
= num_bytes
;
1326 * The last extent of a file may be too large due to page alignment.
1327 * We need to adjust extent_len in this case so that the checks in
1328 * __iterate_backrefs work.
1330 if (data_offset
+ num_bytes
>= ino_size
)
1331 backref_ctx
->extent_len
= ino_size
- data_offset
;
1334 * Now collect all backrefs.
1336 if (compressed
== BTRFS_COMPRESS_NONE
)
1337 extent_item_pos
= logical
- found_key
.objectid
;
1339 extent_item_pos
= 0;
1340 ret
= iterate_extent_inodes(sctx
->send_root
->fs_info
,
1341 found_key
.objectid
, extent_item_pos
, 1,
1342 __iterate_backrefs
, backref_ctx
);
1347 if (!backref_ctx
->found_itself
) {
1348 /* found a bug in backref code? */
1350 btrfs_err(sctx
->send_root
->fs_info
, "did not find backref in "
1351 "send_root. inode=%llu, offset=%llu, "
1352 "disk_byte=%llu found extent=%llu\n",
1353 ino
, data_offset
, disk_byte
, found_key
.objectid
);
1357 verbose_printk(KERN_DEBUG
"btrfs: find_extent_clone: data_offset=%llu, "
1359 "num_bytes=%llu, logical=%llu\n",
1360 data_offset
, ino
, num_bytes
, logical
);
1362 if (!backref_ctx
->found
)
1363 verbose_printk("btrfs: no clones found\n");
1365 cur_clone_root
= NULL
;
1366 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1367 if (sctx
->clone_roots
[i
].found_refs
) {
1368 if (!cur_clone_root
)
1369 cur_clone_root
= sctx
->clone_roots
+ i
;
1370 else if (sctx
->clone_roots
[i
].root
== sctx
->send_root
)
1371 /* prefer clones from send_root over others */
1372 cur_clone_root
= sctx
->clone_roots
+ i
;
1377 if (cur_clone_root
) {
1378 if (compressed
!= BTRFS_COMPRESS_NONE
) {
1380 * Offsets given by iterate_extent_inodes() are relative
1381 * to the start of the extent, we need to add logical
1382 * offset from the file extent item.
1383 * (See why at backref.c:check_extent_in_eb())
1385 cur_clone_root
->offset
+= btrfs_file_extent_offset(eb
,
1388 *found
= cur_clone_root
;
1395 btrfs_free_path(tmp_path
);
1400 static int read_symlink(struct btrfs_root
*root
,
1402 struct fs_path
*dest
)
1405 struct btrfs_path
*path
;
1406 struct btrfs_key key
;
1407 struct btrfs_file_extent_item
*ei
;
1413 path
= alloc_path_for_send();
1418 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1420 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1425 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1426 struct btrfs_file_extent_item
);
1427 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
1428 compression
= btrfs_file_extent_compression(path
->nodes
[0], ei
);
1429 BUG_ON(type
!= BTRFS_FILE_EXTENT_INLINE
);
1430 BUG_ON(compression
);
1432 off
= btrfs_file_extent_inline_start(ei
);
1433 len
= btrfs_file_extent_inline_len(path
->nodes
[0], path
->slots
[0], ei
);
1435 ret
= fs_path_add_from_extent_buffer(dest
, path
->nodes
[0], off
, len
);
1438 btrfs_free_path(path
);
1443 * Helper function to generate a file name that is unique in the root of
1444 * send_root and parent_root. This is used to generate names for orphan inodes.
1446 static int gen_unique_name(struct send_ctx
*sctx
,
1448 struct fs_path
*dest
)
1451 struct btrfs_path
*path
;
1452 struct btrfs_dir_item
*di
;
1457 path
= alloc_path_for_send();
1462 len
= snprintf(tmp
, sizeof(tmp
), "o%llu-%llu-%llu",
1464 ASSERT(len
< sizeof(tmp
));
1466 di
= btrfs_lookup_dir_item(NULL
, sctx
->send_root
,
1467 path
, BTRFS_FIRST_FREE_OBJECTID
,
1468 tmp
, strlen(tmp
), 0);
1469 btrfs_release_path(path
);
1475 /* not unique, try again */
1480 if (!sctx
->parent_root
) {
1486 di
= btrfs_lookup_dir_item(NULL
, sctx
->parent_root
,
1487 path
, BTRFS_FIRST_FREE_OBJECTID
,
1488 tmp
, strlen(tmp
), 0);
1489 btrfs_release_path(path
);
1495 /* not unique, try again */
1503 ret
= fs_path_add(dest
, tmp
, strlen(tmp
));
1506 btrfs_free_path(path
);
1511 inode_state_no_change
,
1512 inode_state_will_create
,
1513 inode_state_did_create
,
1514 inode_state_will_delete
,
1515 inode_state_did_delete
,
1518 static int get_cur_inode_state(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1526 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &left_gen
, NULL
, NULL
,
1528 if (ret
< 0 && ret
!= -ENOENT
)
1532 if (!sctx
->parent_root
) {
1533 right_ret
= -ENOENT
;
1535 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
, &right_gen
,
1536 NULL
, NULL
, NULL
, NULL
);
1537 if (ret
< 0 && ret
!= -ENOENT
)
1542 if (!left_ret
&& !right_ret
) {
1543 if (left_gen
== gen
&& right_gen
== gen
) {
1544 ret
= inode_state_no_change
;
1545 } else if (left_gen
== gen
) {
1546 if (ino
< sctx
->send_progress
)
1547 ret
= inode_state_did_create
;
1549 ret
= inode_state_will_create
;
1550 } else if (right_gen
== gen
) {
1551 if (ino
< sctx
->send_progress
)
1552 ret
= inode_state_did_delete
;
1554 ret
= inode_state_will_delete
;
1558 } else if (!left_ret
) {
1559 if (left_gen
== gen
) {
1560 if (ino
< sctx
->send_progress
)
1561 ret
= inode_state_did_create
;
1563 ret
= inode_state_will_create
;
1567 } else if (!right_ret
) {
1568 if (right_gen
== gen
) {
1569 if (ino
< sctx
->send_progress
)
1570 ret
= inode_state_did_delete
;
1572 ret
= inode_state_will_delete
;
1584 static int is_inode_existent(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1588 ret
= get_cur_inode_state(sctx
, ino
, gen
);
1592 if (ret
== inode_state_no_change
||
1593 ret
== inode_state_did_create
||
1594 ret
== inode_state_will_delete
)
1604 * Helper function to lookup a dir item in a dir.
1606 static int lookup_dir_item_inode(struct btrfs_root
*root
,
1607 u64 dir
, const char *name
, int name_len
,
1612 struct btrfs_dir_item
*di
;
1613 struct btrfs_key key
;
1614 struct btrfs_path
*path
;
1616 path
= alloc_path_for_send();
1620 di
= btrfs_lookup_dir_item(NULL
, root
, path
,
1621 dir
, name
, name_len
, 0);
1630 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
1631 *found_inode
= key
.objectid
;
1632 *found_type
= btrfs_dir_type(path
->nodes
[0], di
);
1635 btrfs_free_path(path
);
1640 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1641 * generation of the parent dir and the name of the dir entry.
1643 static int get_first_ref(struct btrfs_root
*root
, u64 ino
,
1644 u64
*dir
, u64
*dir_gen
, struct fs_path
*name
)
1647 struct btrfs_key key
;
1648 struct btrfs_key found_key
;
1649 struct btrfs_path
*path
;
1653 path
= alloc_path_for_send();
1658 key
.type
= BTRFS_INODE_REF_KEY
;
1661 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
1665 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1667 if (ret
|| found_key
.objectid
!= ino
||
1668 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1669 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1674 if (found_key
.type
== BTRFS_INODE_REF_KEY
) {
1675 struct btrfs_inode_ref
*iref
;
1676 iref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1677 struct btrfs_inode_ref
);
1678 len
= btrfs_inode_ref_name_len(path
->nodes
[0], iref
);
1679 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1680 (unsigned long)(iref
+ 1),
1682 parent_dir
= found_key
.offset
;
1684 struct btrfs_inode_extref
*extref
;
1685 extref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1686 struct btrfs_inode_extref
);
1687 len
= btrfs_inode_extref_name_len(path
->nodes
[0], extref
);
1688 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1689 (unsigned long)&extref
->name
, len
);
1690 parent_dir
= btrfs_inode_extref_parent(path
->nodes
[0], extref
);
1694 btrfs_release_path(path
);
1697 ret
= get_inode_info(root
, parent_dir
, NULL
, dir_gen
, NULL
,
1706 btrfs_free_path(path
);
1710 static int is_first_ref(struct btrfs_root
*root
,
1712 const char *name
, int name_len
)
1715 struct fs_path
*tmp_name
;
1718 tmp_name
= fs_path_alloc();
1722 ret
= get_first_ref(root
, ino
, &tmp_dir
, NULL
, tmp_name
);
1726 if (dir
!= tmp_dir
|| name_len
!= fs_path_len(tmp_name
)) {
1731 ret
= !memcmp(tmp_name
->start
, name
, name_len
);
1734 fs_path_free(tmp_name
);
1739 * Used by process_recorded_refs to determine if a new ref would overwrite an
1740 * already existing ref. In case it detects an overwrite, it returns the
1741 * inode/gen in who_ino/who_gen.
1742 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1743 * to make sure later references to the overwritten inode are possible.
1744 * Orphanizing is however only required for the first ref of an inode.
1745 * process_recorded_refs does an additional is_first_ref check to see if
1746 * orphanizing is really required.
1748 static int will_overwrite_ref(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
1749 const char *name
, int name_len
,
1750 u64
*who_ino
, u64
*who_gen
)
1754 u64 other_inode
= 0;
1757 if (!sctx
->parent_root
)
1760 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1765 * If we have a parent root we need to verify that the parent dir was
1766 * not delted and then re-created, if it was then we have no overwrite
1767 * and we can just unlink this entry.
1769 if (sctx
->parent_root
) {
1770 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &gen
, NULL
,
1772 if (ret
< 0 && ret
!= -ENOENT
)
1782 ret
= lookup_dir_item_inode(sctx
->parent_root
, dir
, name
, name_len
,
1783 &other_inode
, &other_type
);
1784 if (ret
< 0 && ret
!= -ENOENT
)
1792 * Check if the overwritten ref was already processed. If yes, the ref
1793 * was already unlinked/moved, so we can safely assume that we will not
1794 * overwrite anything at this point in time.
1796 if (other_inode
> sctx
->send_progress
) {
1797 ret
= get_inode_info(sctx
->parent_root
, other_inode
, NULL
,
1798 who_gen
, NULL
, NULL
, NULL
, NULL
);
1803 *who_ino
= other_inode
;
1813 * Checks if the ref was overwritten by an already processed inode. This is
1814 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1815 * thus the orphan name needs be used.
1816 * process_recorded_refs also uses it to avoid unlinking of refs that were
1819 static int did_overwrite_ref(struct send_ctx
*sctx
,
1820 u64 dir
, u64 dir_gen
,
1821 u64 ino
, u64 ino_gen
,
1822 const char *name
, int name_len
)
1829 if (!sctx
->parent_root
)
1832 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1836 /* check if the ref was overwritten by another ref */
1837 ret
= lookup_dir_item_inode(sctx
->send_root
, dir
, name
, name_len
,
1838 &ow_inode
, &other_type
);
1839 if (ret
< 0 && ret
!= -ENOENT
)
1842 /* was never and will never be overwritten */
1847 ret
= get_inode_info(sctx
->send_root
, ow_inode
, NULL
, &gen
, NULL
, NULL
,
1852 if (ow_inode
== ino
&& gen
== ino_gen
) {
1857 /* we know that it is or will be overwritten. check this now */
1858 if (ow_inode
< sctx
->send_progress
)
1868 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1869 * that got overwritten. This is used by process_recorded_refs to determine
1870 * if it has to use the path as returned by get_cur_path or the orphan name.
1872 static int did_overwrite_first_ref(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1875 struct fs_path
*name
= NULL
;
1879 if (!sctx
->parent_root
)
1882 name
= fs_path_alloc();
1886 ret
= get_first_ref(sctx
->parent_root
, ino
, &dir
, &dir_gen
, name
);
1890 ret
= did_overwrite_ref(sctx
, dir
, dir_gen
, ino
, gen
,
1891 name
->start
, fs_path_len(name
));
1899 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1900 * so we need to do some special handling in case we have clashes. This function
1901 * takes care of this with the help of name_cache_entry::radix_list.
1902 * In case of error, nce is kfreed.
1904 static int name_cache_insert(struct send_ctx
*sctx
,
1905 struct name_cache_entry
*nce
)
1908 struct list_head
*nce_head
;
1910 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
1911 (unsigned long)nce
->ino
);
1913 nce_head
= kmalloc(sizeof(*nce_head
), GFP_NOFS
);
1918 INIT_LIST_HEAD(nce_head
);
1920 ret
= radix_tree_insert(&sctx
->name_cache
, nce
->ino
, nce_head
);
1927 list_add_tail(&nce
->radix_list
, nce_head
);
1928 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
1929 sctx
->name_cache_size
++;
1934 static void name_cache_delete(struct send_ctx
*sctx
,
1935 struct name_cache_entry
*nce
)
1937 struct list_head
*nce_head
;
1939 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
1940 (unsigned long)nce
->ino
);
1942 btrfs_err(sctx
->send_root
->fs_info
,
1943 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
1944 nce
->ino
, sctx
->name_cache_size
);
1947 list_del(&nce
->radix_list
);
1948 list_del(&nce
->list
);
1949 sctx
->name_cache_size
--;
1952 * We may not get to the final release of nce_head if the lookup fails
1954 if (nce_head
&& list_empty(nce_head
)) {
1955 radix_tree_delete(&sctx
->name_cache
, (unsigned long)nce
->ino
);
1960 static struct name_cache_entry
*name_cache_search(struct send_ctx
*sctx
,
1963 struct list_head
*nce_head
;
1964 struct name_cache_entry
*cur
;
1966 nce_head
= radix_tree_lookup(&sctx
->name_cache
, (unsigned long)ino
);
1970 list_for_each_entry(cur
, nce_head
, radix_list
) {
1971 if (cur
->ino
== ino
&& cur
->gen
== gen
)
1978 * Removes the entry from the list and adds it back to the end. This marks the
1979 * entry as recently used so that name_cache_clean_unused does not remove it.
1981 static void name_cache_used(struct send_ctx
*sctx
, struct name_cache_entry
*nce
)
1983 list_del(&nce
->list
);
1984 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
1988 * Remove some entries from the beginning of name_cache_list.
1990 static void name_cache_clean_unused(struct send_ctx
*sctx
)
1992 struct name_cache_entry
*nce
;
1994 if (sctx
->name_cache_size
< SEND_CTX_NAME_CACHE_CLEAN_SIZE
)
1997 while (sctx
->name_cache_size
> SEND_CTX_MAX_NAME_CACHE_SIZE
) {
1998 nce
= list_entry(sctx
->name_cache_list
.next
,
1999 struct name_cache_entry
, list
);
2000 name_cache_delete(sctx
, nce
);
2005 static void name_cache_free(struct send_ctx
*sctx
)
2007 struct name_cache_entry
*nce
;
2009 while (!list_empty(&sctx
->name_cache_list
)) {
2010 nce
= list_entry(sctx
->name_cache_list
.next
,
2011 struct name_cache_entry
, list
);
2012 name_cache_delete(sctx
, nce
);
2018 * Used by get_cur_path for each ref up to the root.
2019 * Returns 0 if it succeeded.
2020 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2021 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2022 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2023 * Returns <0 in case of error.
2025 static int __get_cur_name_and_parent(struct send_ctx
*sctx
,
2029 struct fs_path
*dest
)
2033 struct name_cache_entry
*nce
= NULL
;
2036 * First check if we already did a call to this function with the same
2037 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2038 * return the cached result.
2040 nce
= name_cache_search(sctx
, ino
, gen
);
2042 if (ino
< sctx
->send_progress
&& nce
->need_later_update
) {
2043 name_cache_delete(sctx
, nce
);
2047 name_cache_used(sctx
, nce
);
2048 *parent_ino
= nce
->parent_ino
;
2049 *parent_gen
= nce
->parent_gen
;
2050 ret
= fs_path_add(dest
, nce
->name
, nce
->name_len
);
2059 * If the inode is not existent yet, add the orphan name and return 1.
2060 * This should only happen for the parent dir that we determine in
2063 ret
= is_inode_existent(sctx
, ino
, gen
);
2068 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2076 * Depending on whether the inode was already processed or not, use
2077 * send_root or parent_root for ref lookup.
2079 if (ino
< sctx
->send_progress
)
2080 ret
= get_first_ref(sctx
->send_root
, ino
,
2081 parent_ino
, parent_gen
, dest
);
2083 ret
= get_first_ref(sctx
->parent_root
, ino
,
2084 parent_ino
, parent_gen
, dest
);
2089 * Check if the ref was overwritten by an inode's ref that was processed
2090 * earlier. If yes, treat as orphan and return 1.
2092 ret
= did_overwrite_ref(sctx
, *parent_ino
, *parent_gen
, ino
, gen
,
2093 dest
->start
, dest
->end
- dest
->start
);
2097 fs_path_reset(dest
);
2098 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2106 * Store the result of the lookup in the name cache.
2108 nce
= kmalloc(sizeof(*nce
) + fs_path_len(dest
) + 1, GFP_NOFS
);
2116 nce
->parent_ino
= *parent_ino
;
2117 nce
->parent_gen
= *parent_gen
;
2118 nce
->name_len
= fs_path_len(dest
);
2120 strcpy(nce
->name
, dest
->start
);
2122 if (ino
< sctx
->send_progress
)
2123 nce
->need_later_update
= 0;
2125 nce
->need_later_update
= 1;
2127 nce_ret
= name_cache_insert(sctx
, nce
);
2130 name_cache_clean_unused(sctx
);
2137 * Magic happens here. This function returns the first ref to an inode as it
2138 * would look like while receiving the stream at this point in time.
2139 * We walk the path up to the root. For every inode in between, we check if it
2140 * was already processed/sent. If yes, we continue with the parent as found
2141 * in send_root. If not, we continue with the parent as found in parent_root.
2142 * If we encounter an inode that was deleted at this point in time, we use the
2143 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2144 * that were not created yet and overwritten inodes/refs.
2146 * When do we have have orphan inodes:
2147 * 1. When an inode is freshly created and thus no valid refs are available yet
2148 * 2. When a directory lost all it's refs (deleted) but still has dir items
2149 * inside which were not processed yet (pending for move/delete). If anyone
2150 * tried to get the path to the dir items, it would get a path inside that
2152 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2153 * of an unprocessed inode. If in that case the first ref would be
2154 * overwritten, the overwritten inode gets "orphanized". Later when we
2155 * process this overwritten inode, it is restored at a new place by moving
2158 * sctx->send_progress tells this function at which point in time receiving
2161 static int get_cur_path(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2162 struct fs_path
*dest
)
2165 struct fs_path
*name
= NULL
;
2166 u64 parent_inode
= 0;
2170 name
= fs_path_alloc();
2177 fs_path_reset(dest
);
2179 while (!stop
&& ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
2180 fs_path_reset(name
);
2182 if (is_waiting_for_rm(sctx
, ino
)) {
2183 ret
= gen_unique_name(sctx
, ino
, gen
, name
);
2186 ret
= fs_path_add_path(dest
, name
);
2190 if (is_waiting_for_move(sctx
, ino
)) {
2191 ret
= get_first_ref(sctx
->parent_root
, ino
,
2192 &parent_inode
, &parent_gen
, name
);
2194 ret
= __get_cur_name_and_parent(sctx
, ino
, gen
,
2204 ret
= fs_path_add_path(dest
, name
);
2215 fs_path_unreverse(dest
);
2220 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2222 static int send_subvol_begin(struct send_ctx
*sctx
)
2225 struct btrfs_root
*send_root
= sctx
->send_root
;
2226 struct btrfs_root
*parent_root
= sctx
->parent_root
;
2227 struct btrfs_path
*path
;
2228 struct btrfs_key key
;
2229 struct btrfs_root_ref
*ref
;
2230 struct extent_buffer
*leaf
;
2234 path
= btrfs_alloc_path();
2238 name
= kmalloc(BTRFS_PATH_NAME_MAX
, GFP_NOFS
);
2240 btrfs_free_path(path
);
2244 key
.objectid
= send_root
->objectid
;
2245 key
.type
= BTRFS_ROOT_BACKREF_KEY
;
2248 ret
= btrfs_search_slot_for_read(send_root
->fs_info
->tree_root
,
2257 leaf
= path
->nodes
[0];
2258 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2259 if (key
.type
!= BTRFS_ROOT_BACKREF_KEY
||
2260 key
.objectid
!= send_root
->objectid
) {
2264 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
2265 namelen
= btrfs_root_ref_name_len(leaf
, ref
);
2266 read_extent_buffer(leaf
, name
, (unsigned long)(ref
+ 1), namelen
);
2267 btrfs_release_path(path
);
2270 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SNAPSHOT
);
2274 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SUBVOL
);
2279 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_PATH
, name
, namelen
);
2280 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_UUID
,
2281 sctx
->send_root
->root_item
.uuid
);
2282 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CTRANSID
,
2283 le64_to_cpu(sctx
->send_root
->root_item
.ctransid
));
2285 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2286 sctx
->parent_root
->root_item
.uuid
);
2287 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
2288 le64_to_cpu(sctx
->parent_root
->root_item
.ctransid
));
2291 ret
= send_cmd(sctx
);
2295 btrfs_free_path(path
);
2300 static int send_truncate(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 size
)
2305 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino
, size
);
2307 p
= fs_path_alloc();
2311 ret
= begin_cmd(sctx
, BTRFS_SEND_C_TRUNCATE
);
2315 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2318 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2319 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, size
);
2321 ret
= send_cmd(sctx
);
2329 static int send_chmod(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 mode
)
2334 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino
, mode
);
2336 p
= fs_path_alloc();
2340 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHMOD
);
2344 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2347 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2348 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
& 07777);
2350 ret
= send_cmd(sctx
);
2358 static int send_chown(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 uid
, u64 gid
)
2363 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino
, uid
, gid
);
2365 p
= fs_path_alloc();
2369 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHOWN
);
2373 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2376 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2377 TLV_PUT_U64(sctx
, BTRFS_SEND_A_UID
, uid
);
2378 TLV_PUT_U64(sctx
, BTRFS_SEND_A_GID
, gid
);
2380 ret
= send_cmd(sctx
);
2388 static int send_utimes(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
2391 struct fs_path
*p
= NULL
;
2392 struct btrfs_inode_item
*ii
;
2393 struct btrfs_path
*path
= NULL
;
2394 struct extent_buffer
*eb
;
2395 struct btrfs_key key
;
2398 verbose_printk("btrfs: send_utimes %llu\n", ino
);
2400 p
= fs_path_alloc();
2404 path
= alloc_path_for_send();
2411 key
.type
= BTRFS_INODE_ITEM_KEY
;
2413 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2417 eb
= path
->nodes
[0];
2418 slot
= path
->slots
[0];
2419 ii
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
2421 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UTIMES
);
2425 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2428 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2429 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_ATIME
, eb
,
2430 btrfs_inode_atime(ii
));
2431 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_MTIME
, eb
,
2432 btrfs_inode_mtime(ii
));
2433 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_CTIME
, eb
,
2434 btrfs_inode_ctime(ii
));
2435 /* TODO Add otime support when the otime patches get into upstream */
2437 ret
= send_cmd(sctx
);
2442 btrfs_free_path(path
);
2447 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2448 * a valid path yet because we did not process the refs yet. So, the inode
2449 * is created as orphan.
2451 static int send_create_inode(struct send_ctx
*sctx
, u64 ino
)
2460 verbose_printk("btrfs: send_create_inode %llu\n", ino
);
2462 p
= fs_path_alloc();
2466 if (ino
!= sctx
->cur_ino
) {
2467 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &gen
, &mode
,
2472 gen
= sctx
->cur_inode_gen
;
2473 mode
= sctx
->cur_inode_mode
;
2474 rdev
= sctx
->cur_inode_rdev
;
2477 if (S_ISREG(mode
)) {
2478 cmd
= BTRFS_SEND_C_MKFILE
;
2479 } else if (S_ISDIR(mode
)) {
2480 cmd
= BTRFS_SEND_C_MKDIR
;
2481 } else if (S_ISLNK(mode
)) {
2482 cmd
= BTRFS_SEND_C_SYMLINK
;
2483 } else if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
2484 cmd
= BTRFS_SEND_C_MKNOD
;
2485 } else if (S_ISFIFO(mode
)) {
2486 cmd
= BTRFS_SEND_C_MKFIFO
;
2487 } else if (S_ISSOCK(mode
)) {
2488 cmd
= BTRFS_SEND_C_MKSOCK
;
2490 printk(KERN_WARNING
"btrfs: unexpected inode type %o",
2491 (int)(mode
& S_IFMT
));
2496 ret
= begin_cmd(sctx
, cmd
);
2500 ret
= gen_unique_name(sctx
, ino
, gen
, p
);
2504 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2505 TLV_PUT_U64(sctx
, BTRFS_SEND_A_INO
, ino
);
2507 if (S_ISLNK(mode
)) {
2509 ret
= read_symlink(sctx
->send_root
, ino
, p
);
2512 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, p
);
2513 } else if (S_ISCHR(mode
) || S_ISBLK(mode
) ||
2514 S_ISFIFO(mode
) || S_ISSOCK(mode
)) {
2515 TLV_PUT_U64(sctx
, BTRFS_SEND_A_RDEV
, new_encode_dev(rdev
));
2516 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
);
2519 ret
= send_cmd(sctx
);
2531 * We need some special handling for inodes that get processed before the parent
2532 * directory got created. See process_recorded_refs for details.
2533 * This function does the check if we already created the dir out of order.
2535 static int did_create_dir(struct send_ctx
*sctx
, u64 dir
)
2538 struct btrfs_path
*path
= NULL
;
2539 struct btrfs_key key
;
2540 struct btrfs_key found_key
;
2541 struct btrfs_key di_key
;
2542 struct extent_buffer
*eb
;
2543 struct btrfs_dir_item
*di
;
2546 path
= alloc_path_for_send();
2553 key
.type
= BTRFS_DIR_INDEX_KEY
;
2555 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2560 eb
= path
->nodes
[0];
2561 slot
= path
->slots
[0];
2562 if (slot
>= btrfs_header_nritems(eb
)) {
2563 ret
= btrfs_next_leaf(sctx
->send_root
, path
);
2566 } else if (ret
> 0) {
2573 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
2574 if (found_key
.objectid
!= key
.objectid
||
2575 found_key
.type
!= key
.type
) {
2580 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
2581 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
2583 if (di_key
.type
!= BTRFS_ROOT_ITEM_KEY
&&
2584 di_key
.objectid
< sctx
->send_progress
) {
2593 btrfs_free_path(path
);
2598 * Only creates the inode if it is:
2599 * 1. Not a directory
2600 * 2. Or a directory which was not created already due to out of order
2601 * directories. See did_create_dir and process_recorded_refs for details.
2603 static int send_create_inode_if_needed(struct send_ctx
*sctx
)
2607 if (S_ISDIR(sctx
->cur_inode_mode
)) {
2608 ret
= did_create_dir(sctx
, sctx
->cur_ino
);
2617 ret
= send_create_inode(sctx
, sctx
->cur_ino
);
2625 struct recorded_ref
{
2626 struct list_head list
;
2629 struct fs_path
*full_path
;
2637 * We need to process new refs before deleted refs, but compare_tree gives us
2638 * everything mixed. So we first record all refs and later process them.
2639 * This function is a helper to record one ref.
2641 static int __record_ref(struct list_head
*head
, u64 dir
,
2642 u64 dir_gen
, struct fs_path
*path
)
2644 struct recorded_ref
*ref
;
2646 ref
= kmalloc(sizeof(*ref
), GFP_NOFS
);
2651 ref
->dir_gen
= dir_gen
;
2652 ref
->full_path
= path
;
2654 ref
->name
= (char *)kbasename(ref
->full_path
->start
);
2655 ref
->name_len
= ref
->full_path
->end
- ref
->name
;
2656 ref
->dir_path
= ref
->full_path
->start
;
2657 if (ref
->name
== ref
->full_path
->start
)
2658 ref
->dir_path_len
= 0;
2660 ref
->dir_path_len
= ref
->full_path
->end
-
2661 ref
->full_path
->start
- 1 - ref
->name_len
;
2663 list_add_tail(&ref
->list
, head
);
2667 static int dup_ref(struct recorded_ref
*ref
, struct list_head
*list
)
2669 struct recorded_ref
*new;
2671 new = kmalloc(sizeof(*ref
), GFP_NOFS
);
2675 new->dir
= ref
->dir
;
2676 new->dir_gen
= ref
->dir_gen
;
2677 new->full_path
= NULL
;
2678 INIT_LIST_HEAD(&new->list
);
2679 list_add_tail(&new->list
, list
);
2683 static void __free_recorded_refs(struct list_head
*head
)
2685 struct recorded_ref
*cur
;
2687 while (!list_empty(head
)) {
2688 cur
= list_entry(head
->next
, struct recorded_ref
, list
);
2689 fs_path_free(cur
->full_path
);
2690 list_del(&cur
->list
);
2695 static void free_recorded_refs(struct send_ctx
*sctx
)
2697 __free_recorded_refs(&sctx
->new_refs
);
2698 __free_recorded_refs(&sctx
->deleted_refs
);
2702 * Renames/moves a file/dir to its orphan name. Used when the first
2703 * ref of an unprocessed inode gets overwritten and for all non empty
2706 static int orphanize_inode(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2707 struct fs_path
*path
)
2710 struct fs_path
*orphan
;
2712 orphan
= fs_path_alloc();
2716 ret
= gen_unique_name(sctx
, ino
, gen
, orphan
);
2720 ret
= send_rename(sctx
, path
, orphan
);
2723 fs_path_free(orphan
);
2727 static struct orphan_dir_info
*
2728 add_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2730 struct rb_node
**p
= &sctx
->orphan_dirs
.rb_node
;
2731 struct rb_node
*parent
= NULL
;
2732 struct orphan_dir_info
*entry
, *odi
;
2734 odi
= kmalloc(sizeof(*odi
), GFP_NOFS
);
2736 return ERR_PTR(-ENOMEM
);
2742 entry
= rb_entry(parent
, struct orphan_dir_info
, node
);
2743 if (dir_ino
< entry
->ino
) {
2745 } else if (dir_ino
> entry
->ino
) {
2746 p
= &(*p
)->rb_right
;
2753 rb_link_node(&odi
->node
, parent
, p
);
2754 rb_insert_color(&odi
->node
, &sctx
->orphan_dirs
);
2758 static struct orphan_dir_info
*
2759 get_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2761 struct rb_node
*n
= sctx
->orphan_dirs
.rb_node
;
2762 struct orphan_dir_info
*entry
;
2765 entry
= rb_entry(n
, struct orphan_dir_info
, node
);
2766 if (dir_ino
< entry
->ino
)
2768 else if (dir_ino
> entry
->ino
)
2776 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
)
2778 struct orphan_dir_info
*odi
= get_orphan_dir_info(sctx
, dir_ino
);
2783 static void free_orphan_dir_info(struct send_ctx
*sctx
,
2784 struct orphan_dir_info
*odi
)
2788 rb_erase(&odi
->node
, &sctx
->orphan_dirs
);
2793 * Returns 1 if a directory can be removed at this point in time.
2794 * We check this by iterating all dir items and checking if the inode behind
2795 * the dir item was already processed.
2797 static int can_rmdir(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
2801 struct btrfs_root
*root
= sctx
->parent_root
;
2802 struct btrfs_path
*path
;
2803 struct btrfs_key key
;
2804 struct btrfs_key found_key
;
2805 struct btrfs_key loc
;
2806 struct btrfs_dir_item
*di
;
2809 * Don't try to rmdir the top/root subvolume dir.
2811 if (dir
== BTRFS_FIRST_FREE_OBJECTID
)
2814 path
= alloc_path_for_send();
2819 key
.type
= BTRFS_DIR_INDEX_KEY
;
2821 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2826 struct waiting_dir_move
*dm
;
2828 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
2829 ret
= btrfs_next_leaf(root
, path
);
2836 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2838 if (found_key
.objectid
!= key
.objectid
||
2839 found_key
.type
!= key
.type
)
2842 di
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2843 struct btrfs_dir_item
);
2844 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &loc
);
2846 dm
= get_waiting_dir_move(sctx
, loc
.objectid
);
2848 struct orphan_dir_info
*odi
;
2850 odi
= add_orphan_dir_info(sctx
, dir
);
2856 dm
->rmdir_ino
= dir
;
2861 if (loc
.objectid
> send_progress
) {
2872 btrfs_free_path(path
);
2876 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
)
2878 struct waiting_dir_move
*entry
= get_waiting_dir_move(sctx
, ino
);
2880 return entry
!= NULL
;
2883 static int add_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
)
2885 struct rb_node
**p
= &sctx
->waiting_dir_moves
.rb_node
;
2886 struct rb_node
*parent
= NULL
;
2887 struct waiting_dir_move
*entry
, *dm
;
2889 dm
= kmalloc(sizeof(*dm
), GFP_NOFS
);
2897 entry
= rb_entry(parent
, struct waiting_dir_move
, node
);
2898 if (ino
< entry
->ino
) {
2900 } else if (ino
> entry
->ino
) {
2901 p
= &(*p
)->rb_right
;
2908 rb_link_node(&dm
->node
, parent
, p
);
2909 rb_insert_color(&dm
->node
, &sctx
->waiting_dir_moves
);
2913 static struct waiting_dir_move
*
2914 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
)
2916 struct rb_node
*n
= sctx
->waiting_dir_moves
.rb_node
;
2917 struct waiting_dir_move
*entry
;
2920 entry
= rb_entry(n
, struct waiting_dir_move
, node
);
2921 if (ino
< entry
->ino
)
2923 else if (ino
> entry
->ino
)
2931 static void free_waiting_dir_move(struct send_ctx
*sctx
,
2932 struct waiting_dir_move
*dm
)
2936 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
2940 static int add_pending_dir_move(struct send_ctx
*sctx
,
2945 struct rb_node
**p
= &sctx
->pending_dir_moves
.rb_node
;
2946 struct rb_node
*parent
= NULL
;
2947 struct pending_dir_move
*entry
= NULL
, *pm
;
2948 struct recorded_ref
*cur
;
2952 pm
= kmalloc(sizeof(*pm
), GFP_NOFS
);
2955 pm
->parent_ino
= parent_ino
;
2958 INIT_LIST_HEAD(&pm
->list
);
2959 INIT_LIST_HEAD(&pm
->update_refs
);
2960 RB_CLEAR_NODE(&pm
->node
);
2964 entry
= rb_entry(parent
, struct pending_dir_move
, node
);
2965 if (parent_ino
< entry
->parent_ino
) {
2967 } else if (parent_ino
> entry
->parent_ino
) {
2968 p
= &(*p
)->rb_right
;
2975 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
2976 ret
= dup_ref(cur
, &pm
->update_refs
);
2980 list_for_each_entry(cur
, &sctx
->new_refs
, list
) {
2981 ret
= dup_ref(cur
, &pm
->update_refs
);
2986 ret
= add_waiting_dir_move(sctx
, pm
->ino
);
2991 list_add_tail(&pm
->list
, &entry
->list
);
2993 rb_link_node(&pm
->node
, parent
, p
);
2994 rb_insert_color(&pm
->node
, &sctx
->pending_dir_moves
);
2999 __free_recorded_refs(&pm
->update_refs
);
3005 static struct pending_dir_move
*get_pending_dir_moves(struct send_ctx
*sctx
,
3008 struct rb_node
*n
= sctx
->pending_dir_moves
.rb_node
;
3009 struct pending_dir_move
*entry
;
3012 entry
= rb_entry(n
, struct pending_dir_move
, node
);
3013 if (parent_ino
< entry
->parent_ino
)
3015 else if (parent_ino
> entry
->parent_ino
)
3023 static int apply_dir_move(struct send_ctx
*sctx
, struct pending_dir_move
*pm
)
3025 struct fs_path
*from_path
= NULL
;
3026 struct fs_path
*to_path
= NULL
;
3027 struct fs_path
*name
= NULL
;
3028 u64 orig_progress
= sctx
->send_progress
;
3029 struct recorded_ref
*cur
;
3030 u64 parent_ino
, parent_gen
;
3031 struct waiting_dir_move
*dm
= NULL
;
3035 name
= fs_path_alloc();
3036 from_path
= fs_path_alloc();
3037 if (!name
|| !from_path
) {
3042 dm
= get_waiting_dir_move(sctx
, pm
->ino
);
3044 rmdir_ino
= dm
->rmdir_ino
;
3045 free_waiting_dir_move(sctx
, dm
);
3047 ret
= get_first_ref(sctx
->parent_root
, pm
->ino
,
3048 &parent_ino
, &parent_gen
, name
);
3052 ret
= get_cur_path(sctx
, parent_ino
, parent_gen
,
3056 ret
= fs_path_add_path(from_path
, name
);
3060 fs_path_reset(name
);
3064 sctx
->send_progress
= sctx
->cur_ino
+ 1;
3065 ret
= get_cur_path(sctx
, pm
->ino
, pm
->gen
, to_path
);
3069 ret
= send_rename(sctx
, from_path
, to_path
);
3074 struct orphan_dir_info
*odi
;
3076 odi
= get_orphan_dir_info(sctx
, rmdir_ino
);
3078 /* already deleted */
3081 ret
= can_rmdir(sctx
, rmdir_ino
, odi
->gen
, sctx
->cur_ino
+ 1);
3087 name
= fs_path_alloc();
3092 ret
= get_cur_path(sctx
, rmdir_ino
, odi
->gen
, name
);
3095 ret
= send_rmdir(sctx
, name
);
3098 free_orphan_dir_info(sctx
, odi
);
3102 ret
= send_utimes(sctx
, pm
->ino
, pm
->gen
);
3107 * After rename/move, need to update the utimes of both new parent(s)
3108 * and old parent(s).
3110 list_for_each_entry(cur
, &pm
->update_refs
, list
) {
3111 if (cur
->dir
== rmdir_ino
)
3113 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
3120 fs_path_free(from_path
);
3121 fs_path_free(to_path
);
3122 sctx
->send_progress
= orig_progress
;
3127 static void free_pending_move(struct send_ctx
*sctx
, struct pending_dir_move
*m
)
3129 if (!list_empty(&m
->list
))
3131 if (!RB_EMPTY_NODE(&m
->node
))
3132 rb_erase(&m
->node
, &sctx
->pending_dir_moves
);
3133 __free_recorded_refs(&m
->update_refs
);
3137 static void tail_append_pending_moves(struct pending_dir_move
*moves
,
3138 struct list_head
*stack
)
3140 if (list_empty(&moves
->list
)) {
3141 list_add_tail(&moves
->list
, stack
);
3144 list_splice_init(&moves
->list
, &list
);
3145 list_add_tail(&moves
->list
, stack
);
3146 list_splice_tail(&list
, stack
);
3150 static int apply_children_dir_moves(struct send_ctx
*sctx
)
3152 struct pending_dir_move
*pm
;
3153 struct list_head stack
;
3154 u64 parent_ino
= sctx
->cur_ino
;
3157 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3161 INIT_LIST_HEAD(&stack
);
3162 tail_append_pending_moves(pm
, &stack
);
3164 while (!list_empty(&stack
)) {
3165 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3166 parent_ino
= pm
->ino
;
3167 ret
= apply_dir_move(sctx
, pm
);
3168 free_pending_move(sctx
, pm
);
3171 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3173 tail_append_pending_moves(pm
, &stack
);
3178 while (!list_empty(&stack
)) {
3179 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3180 free_pending_move(sctx
, pm
);
3185 static int wait_for_parent_move(struct send_ctx
*sctx
,
3186 struct recorded_ref
*parent_ref
)
3189 u64 ino
= parent_ref
->dir
;
3190 u64 parent_ino_before
, parent_ino_after
;
3192 struct fs_path
*path_before
= NULL
;
3193 struct fs_path
*path_after
= NULL
;
3195 int register_upper_dirs
;
3198 if (is_waiting_for_move(sctx
, ino
))
3201 if (parent_ref
->dir
<= sctx
->cur_ino
)
3204 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
, &old_gen
,
3205 NULL
, NULL
, NULL
, NULL
);
3211 if (parent_ref
->dir_gen
!= old_gen
)
3214 path_before
= fs_path_alloc();
3218 ret
= get_first_ref(sctx
->parent_root
, ino
, &parent_ino_before
,
3220 if (ret
== -ENOENT
) {
3223 } else if (ret
< 0) {
3227 path_after
= fs_path_alloc();
3233 ret
= get_first_ref(sctx
->send_root
, ino
, &parent_ino_after
,
3235 if (ret
== -ENOENT
) {
3238 } else if (ret
< 0) {
3242 len1
= fs_path_len(path_before
);
3243 len2
= fs_path_len(path_after
);
3244 if (parent_ino_before
!= parent_ino_after
|| len1
!= len2
||
3245 memcmp(path_before
->start
, path_after
->start
, len1
)) {
3252 * Ok, our new most direct ancestor has a higher inode number but
3253 * wasn't moved/renamed. So maybe some of the new ancestors higher in
3254 * the hierarchy have an higher inode number too *and* were renamed
3255 * or moved - in this case we need to wait for the ancestor's rename
3256 * or move operation before we can do the move/rename for the current
3259 register_upper_dirs
= 0;
3260 ino
= parent_ino_after
;
3262 while ((ret
== 0 || register_upper_dirs
) && ino
> sctx
->cur_ino
) {
3265 fs_path_reset(path_before
);
3266 fs_path_reset(path_after
);
3268 ret
= get_first_ref(sctx
->send_root
, ino
, &parent_ino_after
,
3269 &parent_gen
, path_after
);
3272 ret
= get_first_ref(sctx
->parent_root
, ino
, &parent_ino_before
,
3274 if (ret
== -ENOENT
) {
3277 } else if (ret
< 0) {
3281 len1
= fs_path_len(path_before
);
3282 len2
= fs_path_len(path_after
);
3283 if (parent_ino_before
!= parent_ino_after
|| len1
!= len2
||
3284 memcmp(path_before
->start
, path_after
->start
, len1
)) {
3286 if (register_upper_dirs
) {
3289 register_upper_dirs
= 1;
3290 ino
= parent_ref
->dir
;
3291 gen
= parent_ref
->dir_gen
;
3294 } else if (register_upper_dirs
) {
3295 ret
= add_pending_dir_move(sctx
, ino
, gen
,
3297 if (ret
< 0 && ret
!= -EEXIST
)
3301 ino
= parent_ino_after
;
3306 fs_path_free(path_before
);
3307 fs_path_free(path_after
);
3313 * This does all the move/link/unlink/rmdir magic.
3315 static int process_recorded_refs(struct send_ctx
*sctx
, int *pending_move
)
3318 struct recorded_ref
*cur
;
3319 struct recorded_ref
*cur2
;
3320 struct list_head check_dirs
;
3321 struct fs_path
*valid_path
= NULL
;
3324 int did_overwrite
= 0;
3326 u64 last_dir_ino_rm
= 0;
3328 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx
->cur_ino
);
3331 * This should never happen as the root dir always has the same ref
3332 * which is always '..'
3334 BUG_ON(sctx
->cur_ino
<= BTRFS_FIRST_FREE_OBJECTID
);
3335 INIT_LIST_HEAD(&check_dirs
);
3337 valid_path
= fs_path_alloc();
3344 * First, check if the first ref of the current inode was overwritten
3345 * before. If yes, we know that the current inode was already orphanized
3346 * and thus use the orphan name. If not, we can use get_cur_path to
3347 * get the path of the first ref as it would like while receiving at
3348 * this point in time.
3349 * New inodes are always orphan at the beginning, so force to use the
3350 * orphan name in this case.
3351 * The first ref is stored in valid_path and will be updated if it
3352 * gets moved around.
3354 if (!sctx
->cur_inode_new
) {
3355 ret
= did_overwrite_first_ref(sctx
, sctx
->cur_ino
,
3356 sctx
->cur_inode_gen
);
3362 if (sctx
->cur_inode_new
|| did_overwrite
) {
3363 ret
= gen_unique_name(sctx
, sctx
->cur_ino
,
3364 sctx
->cur_inode_gen
, valid_path
);
3369 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3375 list_for_each_entry(cur
, &sctx
->new_refs
, list
) {
3377 * We may have refs where the parent directory does not exist
3378 * yet. This happens if the parent directories inum is higher
3379 * the the current inum. To handle this case, we create the
3380 * parent directory out of order. But we need to check if this
3381 * did already happen before due to other refs in the same dir.
3383 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3386 if (ret
== inode_state_will_create
) {
3389 * First check if any of the current inodes refs did
3390 * already create the dir.
3392 list_for_each_entry(cur2
, &sctx
->new_refs
, list
) {
3395 if (cur2
->dir
== cur
->dir
) {
3402 * If that did not happen, check if a previous inode
3403 * did already create the dir.
3406 ret
= did_create_dir(sctx
, cur
->dir
);
3410 ret
= send_create_inode(sctx
, cur
->dir
);
3417 * Check if this new ref would overwrite the first ref of
3418 * another unprocessed inode. If yes, orphanize the
3419 * overwritten inode. If we find an overwritten ref that is
3420 * not the first ref, simply unlink it.
3422 ret
= will_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3423 cur
->name
, cur
->name_len
,
3424 &ow_inode
, &ow_gen
);
3428 ret
= is_first_ref(sctx
->parent_root
,
3429 ow_inode
, cur
->dir
, cur
->name
,
3434 ret
= orphanize_inode(sctx
, ow_inode
, ow_gen
,
3439 ret
= send_unlink(sctx
, cur
->full_path
);
3446 * link/move the ref to the new place. If we have an orphan
3447 * inode, move it and update valid_path. If not, link or move
3448 * it depending on the inode mode.
3451 ret
= send_rename(sctx
, valid_path
, cur
->full_path
);
3455 ret
= fs_path_copy(valid_path
, cur
->full_path
);
3459 if (S_ISDIR(sctx
->cur_inode_mode
)) {
3461 * Dirs can't be linked, so move it. For moved
3462 * dirs, we always have one new and one deleted
3463 * ref. The deleted ref is ignored later.
3465 ret
= wait_for_parent_move(sctx
, cur
);
3469 ret
= add_pending_dir_move(sctx
,
3471 sctx
->cur_inode_gen
,
3475 ret
= send_rename(sctx
, valid_path
,
3478 ret
= fs_path_copy(valid_path
,
3484 ret
= send_link(sctx
, cur
->full_path
,
3490 ret
= dup_ref(cur
, &check_dirs
);
3495 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->cur_inode_deleted
) {
3497 * Check if we can already rmdir the directory. If not,
3498 * orphanize it. For every dir item inside that gets deleted
3499 * later, we do this check again and rmdir it then if possible.
3500 * See the use of check_dirs for more details.
3502 ret
= can_rmdir(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3507 ret
= send_rmdir(sctx
, valid_path
);
3510 } else if (!is_orphan
) {
3511 ret
= orphanize_inode(sctx
, sctx
->cur_ino
,
3512 sctx
->cur_inode_gen
, valid_path
);
3518 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
3519 ret
= dup_ref(cur
, &check_dirs
);
3523 } else if (S_ISDIR(sctx
->cur_inode_mode
) &&
3524 !list_empty(&sctx
->deleted_refs
)) {
3526 * We have a moved dir. Add the old parent to check_dirs
3528 cur
= list_entry(sctx
->deleted_refs
.next
, struct recorded_ref
,
3530 ret
= dup_ref(cur
, &check_dirs
);
3533 } else if (!S_ISDIR(sctx
->cur_inode_mode
)) {
3535 * We have a non dir inode. Go through all deleted refs and
3536 * unlink them if they were not already overwritten by other
3539 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
3540 ret
= did_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3541 sctx
->cur_ino
, sctx
->cur_inode_gen
,
3542 cur
->name
, cur
->name_len
);
3546 ret
= send_unlink(sctx
, cur
->full_path
);
3550 ret
= dup_ref(cur
, &check_dirs
);
3555 * If the inode is still orphan, unlink the orphan. This may
3556 * happen when a previous inode did overwrite the first ref
3557 * of this inode and no new refs were added for the current
3558 * inode. Unlinking does not mean that the inode is deleted in
3559 * all cases. There may still be links to this inode in other
3563 ret
= send_unlink(sctx
, valid_path
);
3570 * We did collect all parent dirs where cur_inode was once located. We
3571 * now go through all these dirs and check if they are pending for
3572 * deletion and if it's finally possible to perform the rmdir now.
3573 * We also update the inode stats of the parent dirs here.
3575 list_for_each_entry(cur
, &check_dirs
, list
) {
3577 * In case we had refs into dirs that were not processed yet,
3578 * we don't need to do the utime and rmdir logic for these dirs.
3579 * The dir will be processed later.
3581 if (cur
->dir
> sctx
->cur_ino
)
3584 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3588 if (ret
== inode_state_did_create
||
3589 ret
== inode_state_no_change
) {
3590 /* TODO delayed utimes */
3591 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
3594 } else if (ret
== inode_state_did_delete
&&
3595 cur
->dir
!= last_dir_ino_rm
) {
3596 ret
= can_rmdir(sctx
, cur
->dir
, cur
->dir_gen
,
3601 ret
= get_cur_path(sctx
, cur
->dir
,
3602 cur
->dir_gen
, valid_path
);
3605 ret
= send_rmdir(sctx
, valid_path
);
3608 last_dir_ino_rm
= cur
->dir
;
3616 __free_recorded_refs(&check_dirs
);
3617 free_recorded_refs(sctx
);
3618 fs_path_free(valid_path
);
3622 static int record_ref(struct btrfs_root
*root
, int num
, u64 dir
, int index
,
3623 struct fs_path
*name
, void *ctx
, struct list_head
*refs
)
3626 struct send_ctx
*sctx
= ctx
;
3630 p
= fs_path_alloc();
3634 ret
= get_inode_info(root
, dir
, NULL
, &gen
, NULL
, NULL
,
3639 ret
= get_cur_path(sctx
, dir
, gen
, p
);
3642 ret
= fs_path_add_path(p
, name
);
3646 ret
= __record_ref(refs
, dir
, gen
, p
);
3654 static int __record_new_ref(int num
, u64 dir
, int index
,
3655 struct fs_path
*name
,
3658 struct send_ctx
*sctx
= ctx
;
3659 return record_ref(sctx
->send_root
, num
, dir
, index
, name
,
3660 ctx
, &sctx
->new_refs
);
3664 static int __record_deleted_ref(int num
, u64 dir
, int index
,
3665 struct fs_path
*name
,
3668 struct send_ctx
*sctx
= ctx
;
3669 return record_ref(sctx
->parent_root
, num
, dir
, index
, name
,
3670 ctx
, &sctx
->deleted_refs
);
3673 static int record_new_ref(struct send_ctx
*sctx
)
3677 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
3678 sctx
->cmp_key
, 0, __record_new_ref
, sctx
);
3687 static int record_deleted_ref(struct send_ctx
*sctx
)
3691 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
3692 sctx
->cmp_key
, 0, __record_deleted_ref
, sctx
);
3701 struct find_ref_ctx
{
3704 struct btrfs_root
*root
;
3705 struct fs_path
*name
;
3709 static int __find_iref(int num
, u64 dir
, int index
,
3710 struct fs_path
*name
,
3713 struct find_ref_ctx
*ctx
= ctx_
;
3717 if (dir
== ctx
->dir
&& fs_path_len(name
) == fs_path_len(ctx
->name
) &&
3718 strncmp(name
->start
, ctx
->name
->start
, fs_path_len(name
)) == 0) {
3720 * To avoid doing extra lookups we'll only do this if everything
3723 ret
= get_inode_info(ctx
->root
, dir
, NULL
, &dir_gen
, NULL
,
3727 if (dir_gen
!= ctx
->dir_gen
)
3729 ctx
->found_idx
= num
;
3735 static int find_iref(struct btrfs_root
*root
,
3736 struct btrfs_path
*path
,
3737 struct btrfs_key
*key
,
3738 u64 dir
, u64 dir_gen
, struct fs_path
*name
)
3741 struct find_ref_ctx ctx
;
3745 ctx
.dir_gen
= dir_gen
;
3749 ret
= iterate_inode_ref(root
, path
, key
, 0, __find_iref
, &ctx
);
3753 if (ctx
.found_idx
== -1)
3756 return ctx
.found_idx
;
3759 static int __record_changed_new_ref(int num
, u64 dir
, int index
,
3760 struct fs_path
*name
,
3765 struct send_ctx
*sctx
= ctx
;
3767 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &dir_gen
, NULL
,
3772 ret
= find_iref(sctx
->parent_root
, sctx
->right_path
,
3773 sctx
->cmp_key
, dir
, dir_gen
, name
);
3775 ret
= __record_new_ref(num
, dir
, index
, name
, sctx
);
3782 static int __record_changed_deleted_ref(int num
, u64 dir
, int index
,
3783 struct fs_path
*name
,
3788 struct send_ctx
*sctx
= ctx
;
3790 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &dir_gen
, NULL
,
3795 ret
= find_iref(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
3796 dir
, dir_gen
, name
);
3798 ret
= __record_deleted_ref(num
, dir
, index
, name
, sctx
);
3805 static int record_changed_ref(struct send_ctx
*sctx
)
3809 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
3810 sctx
->cmp_key
, 0, __record_changed_new_ref
, sctx
);
3813 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
3814 sctx
->cmp_key
, 0, __record_changed_deleted_ref
, sctx
);
3824 * Record and process all refs at once. Needed when an inode changes the
3825 * generation number, which means that it was deleted and recreated.
3827 static int process_all_refs(struct send_ctx
*sctx
,
3828 enum btrfs_compare_tree_result cmd
)
3831 struct btrfs_root
*root
;
3832 struct btrfs_path
*path
;
3833 struct btrfs_key key
;
3834 struct btrfs_key found_key
;
3835 struct extent_buffer
*eb
;
3837 iterate_inode_ref_t cb
;
3838 int pending_move
= 0;
3840 path
= alloc_path_for_send();
3844 if (cmd
== BTRFS_COMPARE_TREE_NEW
) {
3845 root
= sctx
->send_root
;
3846 cb
= __record_new_ref
;
3847 } else if (cmd
== BTRFS_COMPARE_TREE_DELETED
) {
3848 root
= sctx
->parent_root
;
3849 cb
= __record_deleted_ref
;
3851 btrfs_err(sctx
->send_root
->fs_info
,
3852 "Wrong command %d in process_all_refs", cmd
);
3857 key
.objectid
= sctx
->cmp_key
->objectid
;
3858 key
.type
= BTRFS_INODE_REF_KEY
;
3860 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3865 eb
= path
->nodes
[0];
3866 slot
= path
->slots
[0];
3867 if (slot
>= btrfs_header_nritems(eb
)) {
3868 ret
= btrfs_next_leaf(root
, path
);
3876 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
3878 if (found_key
.objectid
!= key
.objectid
||
3879 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
3880 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
))
3883 ret
= iterate_inode_ref(root
, path
, &found_key
, 0, cb
, sctx
);
3889 btrfs_release_path(path
);
3891 ret
= process_recorded_refs(sctx
, &pending_move
);
3892 /* Only applicable to an incremental send. */
3893 ASSERT(pending_move
== 0);
3896 btrfs_free_path(path
);
3900 static int send_set_xattr(struct send_ctx
*sctx
,
3901 struct fs_path
*path
,
3902 const char *name
, int name_len
,
3903 const char *data
, int data_len
)
3907 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SET_XATTR
);
3911 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
3912 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
3913 TLV_PUT(sctx
, BTRFS_SEND_A_XATTR_DATA
, data
, data_len
);
3915 ret
= send_cmd(sctx
);
3922 static int send_remove_xattr(struct send_ctx
*sctx
,
3923 struct fs_path
*path
,
3924 const char *name
, int name_len
)
3928 ret
= begin_cmd(sctx
, BTRFS_SEND_C_REMOVE_XATTR
);
3932 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
3933 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
3935 ret
= send_cmd(sctx
);
3942 static int __process_new_xattr(int num
, struct btrfs_key
*di_key
,
3943 const char *name
, int name_len
,
3944 const char *data
, int data_len
,
3948 struct send_ctx
*sctx
= ctx
;
3950 posix_acl_xattr_header dummy_acl
;
3952 p
= fs_path_alloc();
3957 * This hack is needed because empty acl's are stored as zero byte
3958 * data in xattrs. Problem with that is, that receiving these zero byte
3959 * acl's will fail later. To fix this, we send a dummy acl list that
3960 * only contains the version number and no entries.
3962 if (!strncmp(name
, XATTR_NAME_POSIX_ACL_ACCESS
, name_len
) ||
3963 !strncmp(name
, XATTR_NAME_POSIX_ACL_DEFAULT
, name_len
)) {
3964 if (data_len
== 0) {
3965 dummy_acl
.a_version
=
3966 cpu_to_le32(POSIX_ACL_XATTR_VERSION
);
3967 data
= (char *)&dummy_acl
;
3968 data_len
= sizeof(dummy_acl
);
3972 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
3976 ret
= send_set_xattr(sctx
, p
, name
, name_len
, data
, data_len
);
3983 static int __process_deleted_xattr(int num
, struct btrfs_key
*di_key
,
3984 const char *name
, int name_len
,
3985 const char *data
, int data_len
,
3989 struct send_ctx
*sctx
= ctx
;
3992 p
= fs_path_alloc();
3996 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4000 ret
= send_remove_xattr(sctx
, p
, name
, name_len
);
4007 static int process_new_xattr(struct send_ctx
*sctx
)
4011 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4012 sctx
->cmp_key
, __process_new_xattr
, sctx
);
4017 static int process_deleted_xattr(struct send_ctx
*sctx
)
4021 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4022 sctx
->cmp_key
, __process_deleted_xattr
, sctx
);
4027 struct find_xattr_ctx
{
4035 static int __find_xattr(int num
, struct btrfs_key
*di_key
,
4036 const char *name
, int name_len
,
4037 const char *data
, int data_len
,
4038 u8 type
, void *vctx
)
4040 struct find_xattr_ctx
*ctx
= vctx
;
4042 if (name_len
== ctx
->name_len
&&
4043 strncmp(name
, ctx
->name
, name_len
) == 0) {
4044 ctx
->found_idx
= num
;
4045 ctx
->found_data_len
= data_len
;
4046 ctx
->found_data
= kmemdup(data
, data_len
, GFP_NOFS
);
4047 if (!ctx
->found_data
)
4054 static int find_xattr(struct btrfs_root
*root
,
4055 struct btrfs_path
*path
,
4056 struct btrfs_key
*key
,
4057 const char *name
, int name_len
,
4058 char **data
, int *data_len
)
4061 struct find_xattr_ctx ctx
;
4064 ctx
.name_len
= name_len
;
4066 ctx
.found_data
= NULL
;
4067 ctx
.found_data_len
= 0;
4069 ret
= iterate_dir_item(root
, path
, key
, __find_xattr
, &ctx
);
4073 if (ctx
.found_idx
== -1)
4076 *data
= ctx
.found_data
;
4077 *data_len
= ctx
.found_data_len
;
4079 kfree(ctx
.found_data
);
4081 return ctx
.found_idx
;
4085 static int __process_changed_new_xattr(int num
, struct btrfs_key
*di_key
,
4086 const char *name
, int name_len
,
4087 const char *data
, int data_len
,
4091 struct send_ctx
*sctx
= ctx
;
4092 char *found_data
= NULL
;
4093 int found_data_len
= 0;
4095 ret
= find_xattr(sctx
->parent_root
, sctx
->right_path
,
4096 sctx
->cmp_key
, name
, name_len
, &found_data
,
4098 if (ret
== -ENOENT
) {
4099 ret
= __process_new_xattr(num
, di_key
, name
, name_len
, data
,
4100 data_len
, type
, ctx
);
4101 } else if (ret
>= 0) {
4102 if (data_len
!= found_data_len
||
4103 memcmp(data
, found_data
, data_len
)) {
4104 ret
= __process_new_xattr(num
, di_key
, name
, name_len
,
4105 data
, data_len
, type
, ctx
);
4115 static int __process_changed_deleted_xattr(int num
, struct btrfs_key
*di_key
,
4116 const char *name
, int name_len
,
4117 const char *data
, int data_len
,
4121 struct send_ctx
*sctx
= ctx
;
4123 ret
= find_xattr(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
4124 name
, name_len
, NULL
, NULL
);
4126 ret
= __process_deleted_xattr(num
, di_key
, name
, name_len
, data
,
4127 data_len
, type
, ctx
);
4134 static int process_changed_xattr(struct send_ctx
*sctx
)
4138 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4139 sctx
->cmp_key
, __process_changed_new_xattr
, sctx
);
4142 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4143 sctx
->cmp_key
, __process_changed_deleted_xattr
, sctx
);
4149 static int process_all_new_xattrs(struct send_ctx
*sctx
)
4152 struct btrfs_root
*root
;
4153 struct btrfs_path
*path
;
4154 struct btrfs_key key
;
4155 struct btrfs_key found_key
;
4156 struct extent_buffer
*eb
;
4159 path
= alloc_path_for_send();
4163 root
= sctx
->send_root
;
4165 key
.objectid
= sctx
->cmp_key
->objectid
;
4166 key
.type
= BTRFS_XATTR_ITEM_KEY
;
4168 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4173 eb
= path
->nodes
[0];
4174 slot
= path
->slots
[0];
4175 if (slot
>= btrfs_header_nritems(eb
)) {
4176 ret
= btrfs_next_leaf(root
, path
);
4179 } else if (ret
> 0) {
4186 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4187 if (found_key
.objectid
!= key
.objectid
||
4188 found_key
.type
!= key
.type
) {
4193 ret
= iterate_dir_item(root
, path
, &found_key
,
4194 __process_new_xattr
, sctx
);
4202 btrfs_free_path(path
);
4206 static ssize_t
fill_read_buf(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4208 struct btrfs_root
*root
= sctx
->send_root
;
4209 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4210 struct inode
*inode
;
4213 struct btrfs_key key
;
4214 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
4216 unsigned pg_offset
= offset
& ~PAGE_CACHE_MASK
;
4219 key
.objectid
= sctx
->cur_ino
;
4220 key
.type
= BTRFS_INODE_ITEM_KEY
;
4223 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
4225 return PTR_ERR(inode
);
4227 if (offset
+ len
> i_size_read(inode
)) {
4228 if (offset
> i_size_read(inode
))
4231 len
= offset
- i_size_read(inode
);
4236 last_index
= (offset
+ len
- 1) >> PAGE_CACHE_SHIFT
;
4238 /* initial readahead */
4239 memset(&sctx
->ra
, 0, sizeof(struct file_ra_state
));
4240 file_ra_state_init(&sctx
->ra
, inode
->i_mapping
);
4241 btrfs_force_ra(inode
->i_mapping
, &sctx
->ra
, NULL
, index
,
4242 last_index
- index
+ 1);
4244 while (index
<= last_index
) {
4245 unsigned cur_len
= min_t(unsigned, len
,
4246 PAGE_CACHE_SIZE
- pg_offset
);
4247 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
4253 if (!PageUptodate(page
)) {
4254 btrfs_readpage(NULL
, page
);
4256 if (!PageUptodate(page
)) {
4258 page_cache_release(page
);
4265 memcpy(sctx
->read_buf
+ ret
, addr
+ pg_offset
, cur_len
);
4268 page_cache_release(page
);
4280 * Read some bytes from the current inode/file and send a write command to
4283 static int send_write(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4287 ssize_t num_read
= 0;
4289 p
= fs_path_alloc();
4293 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset
, len
);
4295 num_read
= fill_read_buf(sctx
, offset
, len
);
4296 if (num_read
<= 0) {
4302 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4306 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4310 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4311 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4312 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, num_read
);
4314 ret
= send_cmd(sctx
);
4325 * Send a clone command to user space.
4327 static int send_clone(struct send_ctx
*sctx
,
4328 u64 offset
, u32 len
,
4329 struct clone_root
*clone_root
)
4335 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4336 "clone_inode=%llu, clone_offset=%llu\n", offset
, len
,
4337 clone_root
->root
->objectid
, clone_root
->ino
,
4338 clone_root
->offset
);
4340 p
= fs_path_alloc();
4344 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CLONE
);
4348 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4352 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4353 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_LEN
, len
);
4354 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4356 if (clone_root
->root
== sctx
->send_root
) {
4357 ret
= get_inode_info(sctx
->send_root
, clone_root
->ino
, NULL
,
4358 &gen
, NULL
, NULL
, NULL
, NULL
);
4361 ret
= get_cur_path(sctx
, clone_root
->ino
, gen
, p
);
4363 ret
= get_inode_path(clone_root
->root
, clone_root
->ino
, p
);
4368 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4369 clone_root
->root
->root_item
.uuid
);
4370 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
4371 le64_to_cpu(clone_root
->root
->root_item
.ctransid
));
4372 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_CLONE_PATH
, p
);
4373 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_OFFSET
,
4374 clone_root
->offset
);
4376 ret
= send_cmd(sctx
);
4385 * Send an update extent command to user space.
4387 static int send_update_extent(struct send_ctx
*sctx
,
4388 u64 offset
, u32 len
)
4393 p
= fs_path_alloc();
4397 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UPDATE_EXTENT
);
4401 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4405 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4406 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4407 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, len
);
4409 ret
= send_cmd(sctx
);
4417 static int send_hole(struct send_ctx
*sctx
, u64 end
)
4419 struct fs_path
*p
= NULL
;
4420 u64 offset
= sctx
->cur_inode_last_extent
;
4424 p
= fs_path_alloc();
4427 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4429 goto tlv_put_failure
;
4430 memset(sctx
->read_buf
, 0, BTRFS_SEND_READ_SIZE
);
4431 while (offset
< end
) {
4432 len
= min_t(u64
, end
- offset
, BTRFS_SEND_READ_SIZE
);
4434 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4437 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4438 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4439 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, len
);
4440 ret
= send_cmd(sctx
);
4450 static int send_write_or_clone(struct send_ctx
*sctx
,
4451 struct btrfs_path
*path
,
4452 struct btrfs_key
*key
,
4453 struct clone_root
*clone_root
)
4456 struct btrfs_file_extent_item
*ei
;
4457 u64 offset
= key
->offset
;
4462 u64 bs
= sctx
->send_root
->fs_info
->sb
->s_blocksize
;
4464 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4465 struct btrfs_file_extent_item
);
4466 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
4467 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4468 len
= btrfs_file_extent_inline_len(path
->nodes
[0],
4469 path
->slots
[0], ei
);
4471 * it is possible the inline item won't cover the whole page,
4472 * but there may be items after this page. Make
4473 * sure to send the whole thing
4475 len
= PAGE_CACHE_ALIGN(len
);
4477 len
= btrfs_file_extent_num_bytes(path
->nodes
[0], ei
);
4480 if (offset
+ len
> sctx
->cur_inode_size
)
4481 len
= sctx
->cur_inode_size
- offset
;
4487 if (clone_root
&& IS_ALIGNED(offset
+ len
, bs
)) {
4488 ret
= send_clone(sctx
, offset
, len
, clone_root
);
4489 } else if (sctx
->flags
& BTRFS_SEND_FLAG_NO_FILE_DATA
) {
4490 ret
= send_update_extent(sctx
, offset
, len
);
4494 if (l
> BTRFS_SEND_READ_SIZE
)
4495 l
= BTRFS_SEND_READ_SIZE
;
4496 ret
= send_write(sctx
, pos
+ offset
, l
);
4509 static int is_extent_unchanged(struct send_ctx
*sctx
,
4510 struct btrfs_path
*left_path
,
4511 struct btrfs_key
*ekey
)
4514 struct btrfs_key key
;
4515 struct btrfs_path
*path
= NULL
;
4516 struct extent_buffer
*eb
;
4518 struct btrfs_key found_key
;
4519 struct btrfs_file_extent_item
*ei
;
4524 u64 left_offset_fixed
;
4532 path
= alloc_path_for_send();
4536 eb
= left_path
->nodes
[0];
4537 slot
= left_path
->slots
[0];
4538 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
4539 left_type
= btrfs_file_extent_type(eb
, ei
);
4541 if (left_type
!= BTRFS_FILE_EXTENT_REG
) {
4545 left_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
4546 left_len
= btrfs_file_extent_num_bytes(eb
, ei
);
4547 left_offset
= btrfs_file_extent_offset(eb
, ei
);
4548 left_gen
= btrfs_file_extent_generation(eb
, ei
);
4551 * Following comments will refer to these graphics. L is the left
4552 * extents which we are checking at the moment. 1-8 are the right
4553 * extents that we iterate.
4556 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4559 * |--1--|-2b-|...(same as above)
4561 * Alternative situation. Happens on files where extents got split.
4563 * |-----------7-----------|-6-|
4565 * Alternative situation. Happens on files which got larger.
4568 * Nothing follows after 8.
4571 key
.objectid
= ekey
->objectid
;
4572 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4573 key
.offset
= ekey
->offset
;
4574 ret
= btrfs_search_slot_for_read(sctx
->parent_root
, &key
, path
, 0, 0);
4583 * Handle special case where the right side has no extents at all.
4585 eb
= path
->nodes
[0];
4586 slot
= path
->slots
[0];
4587 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4588 if (found_key
.objectid
!= key
.objectid
||
4589 found_key
.type
!= key
.type
) {
4590 /* If we're a hole then just pretend nothing changed */
4591 ret
= (left_disknr
) ? 0 : 1;
4596 * We're now on 2a, 2b or 7.
4599 while (key
.offset
< ekey
->offset
+ left_len
) {
4600 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
4601 right_type
= btrfs_file_extent_type(eb
, ei
);
4602 if (right_type
!= BTRFS_FILE_EXTENT_REG
) {
4607 right_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
4608 right_len
= btrfs_file_extent_num_bytes(eb
, ei
);
4609 right_offset
= btrfs_file_extent_offset(eb
, ei
);
4610 right_gen
= btrfs_file_extent_generation(eb
, ei
);
4613 * Are we at extent 8? If yes, we know the extent is changed.
4614 * This may only happen on the first iteration.
4616 if (found_key
.offset
+ right_len
<= ekey
->offset
) {
4617 /* If we're a hole just pretend nothing changed */
4618 ret
= (left_disknr
) ? 0 : 1;
4622 left_offset_fixed
= left_offset
;
4623 if (key
.offset
< ekey
->offset
) {
4624 /* Fix the right offset for 2a and 7. */
4625 right_offset
+= ekey
->offset
- key
.offset
;
4627 /* Fix the left offset for all behind 2a and 2b */
4628 left_offset_fixed
+= key
.offset
- ekey
->offset
;
4632 * Check if we have the same extent.
4634 if (left_disknr
!= right_disknr
||
4635 left_offset_fixed
!= right_offset
||
4636 left_gen
!= right_gen
) {
4642 * Go to the next extent.
4644 ret
= btrfs_next_item(sctx
->parent_root
, path
);
4648 eb
= path
->nodes
[0];
4649 slot
= path
->slots
[0];
4650 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4652 if (ret
|| found_key
.objectid
!= key
.objectid
||
4653 found_key
.type
!= key
.type
) {
4654 key
.offset
+= right_len
;
4657 if (found_key
.offset
!= key
.offset
+ right_len
) {
4665 * We're now behind the left extent (treat as unchanged) or at the end
4666 * of the right side (treat as changed).
4668 if (key
.offset
>= ekey
->offset
+ left_len
)
4675 btrfs_free_path(path
);
4679 static int get_last_extent(struct send_ctx
*sctx
, u64 offset
)
4681 struct btrfs_path
*path
;
4682 struct btrfs_root
*root
= sctx
->send_root
;
4683 struct btrfs_file_extent_item
*fi
;
4684 struct btrfs_key key
;
4689 path
= alloc_path_for_send();
4693 sctx
->cur_inode_last_extent
= 0;
4695 key
.objectid
= sctx
->cur_ino
;
4696 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4697 key
.offset
= offset
;
4698 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 0, 1);
4702 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
4703 if (key
.objectid
!= sctx
->cur_ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4706 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4707 struct btrfs_file_extent_item
);
4708 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
4709 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4710 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
4711 path
->slots
[0], fi
);
4712 extent_end
= ALIGN(key
.offset
+ size
,
4713 sctx
->send_root
->sectorsize
);
4715 extent_end
= key
.offset
+
4716 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
4718 sctx
->cur_inode_last_extent
= extent_end
;
4720 btrfs_free_path(path
);
4724 static int maybe_send_hole(struct send_ctx
*sctx
, struct btrfs_path
*path
,
4725 struct btrfs_key
*key
)
4727 struct btrfs_file_extent_item
*fi
;
4732 if (sctx
->cur_ino
!= key
->objectid
|| !need_send_hole(sctx
))
4735 if (sctx
->cur_inode_last_extent
== (u64
)-1) {
4736 ret
= get_last_extent(sctx
, key
->offset
- 1);
4741 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4742 struct btrfs_file_extent_item
);
4743 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
4744 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4745 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
4746 path
->slots
[0], fi
);
4747 extent_end
= ALIGN(key
->offset
+ size
,
4748 sctx
->send_root
->sectorsize
);
4750 extent_end
= key
->offset
+
4751 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
4754 if (path
->slots
[0] == 0 &&
4755 sctx
->cur_inode_last_extent
< key
->offset
) {
4757 * We might have skipped entire leafs that contained only
4758 * file extent items for our current inode. These leafs have
4759 * a generation number smaller (older) than the one in the
4760 * current leaf and the leaf our last extent came from, and
4761 * are located between these 2 leafs.
4763 ret
= get_last_extent(sctx
, key
->offset
- 1);
4768 if (sctx
->cur_inode_last_extent
< key
->offset
)
4769 ret
= send_hole(sctx
, key
->offset
);
4770 sctx
->cur_inode_last_extent
= extent_end
;
4774 static int process_extent(struct send_ctx
*sctx
,
4775 struct btrfs_path
*path
,
4776 struct btrfs_key
*key
)
4778 struct clone_root
*found_clone
= NULL
;
4781 if (S_ISLNK(sctx
->cur_inode_mode
))
4784 if (sctx
->parent_root
&& !sctx
->cur_inode_new
) {
4785 ret
= is_extent_unchanged(sctx
, path
, key
);
4793 struct btrfs_file_extent_item
*ei
;
4796 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4797 struct btrfs_file_extent_item
);
4798 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
4799 if (type
== BTRFS_FILE_EXTENT_PREALLOC
||
4800 type
== BTRFS_FILE_EXTENT_REG
) {
4802 * The send spec does not have a prealloc command yet,
4803 * so just leave a hole for prealloc'ed extents until
4804 * we have enough commands queued up to justify rev'ing
4807 if (type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4812 /* Have a hole, just skip it. */
4813 if (btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
) == 0) {
4820 ret
= find_extent_clone(sctx
, path
, key
->objectid
, key
->offset
,
4821 sctx
->cur_inode_size
, &found_clone
);
4822 if (ret
!= -ENOENT
&& ret
< 0)
4825 ret
= send_write_or_clone(sctx
, path
, key
, found_clone
);
4829 ret
= maybe_send_hole(sctx
, path
, key
);
4834 static int process_all_extents(struct send_ctx
*sctx
)
4837 struct btrfs_root
*root
;
4838 struct btrfs_path
*path
;
4839 struct btrfs_key key
;
4840 struct btrfs_key found_key
;
4841 struct extent_buffer
*eb
;
4844 root
= sctx
->send_root
;
4845 path
= alloc_path_for_send();
4849 key
.objectid
= sctx
->cmp_key
->objectid
;
4850 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4852 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4857 eb
= path
->nodes
[0];
4858 slot
= path
->slots
[0];
4860 if (slot
>= btrfs_header_nritems(eb
)) {
4861 ret
= btrfs_next_leaf(root
, path
);
4864 } else if (ret
> 0) {
4871 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4873 if (found_key
.objectid
!= key
.objectid
||
4874 found_key
.type
!= key
.type
) {
4879 ret
= process_extent(sctx
, path
, &found_key
);
4887 btrfs_free_path(path
);
4891 static int process_recorded_refs_if_needed(struct send_ctx
*sctx
, int at_end
,
4893 int *refs_processed
)
4897 if (sctx
->cur_ino
== 0)
4899 if (!at_end
&& sctx
->cur_ino
== sctx
->cmp_key
->objectid
&&
4900 sctx
->cmp_key
->type
<= BTRFS_INODE_EXTREF_KEY
)
4902 if (list_empty(&sctx
->new_refs
) && list_empty(&sctx
->deleted_refs
))
4905 ret
= process_recorded_refs(sctx
, pending_move
);
4909 *refs_processed
= 1;
4914 static int finish_inode_if_needed(struct send_ctx
*sctx
, int at_end
)
4925 int pending_move
= 0;
4926 int refs_processed
= 0;
4928 ret
= process_recorded_refs_if_needed(sctx
, at_end
, &pending_move
,
4934 * We have processed the refs and thus need to advance send_progress.
4935 * Now, calls to get_cur_xxx will take the updated refs of the current
4936 * inode into account.
4938 * On the other hand, if our current inode is a directory and couldn't
4939 * be moved/renamed because its parent was renamed/moved too and it has
4940 * a higher inode number, we can only move/rename our current inode
4941 * after we moved/renamed its parent. Therefore in this case operate on
4942 * the old path (pre move/rename) of our current inode, and the
4943 * move/rename will be performed later.
4945 if (refs_processed
&& !pending_move
)
4946 sctx
->send_progress
= sctx
->cur_ino
+ 1;
4948 if (sctx
->cur_ino
== 0 || sctx
->cur_inode_deleted
)
4950 if (!at_end
&& sctx
->cmp_key
->objectid
== sctx
->cur_ino
)
4953 ret
= get_inode_info(sctx
->send_root
, sctx
->cur_ino
, NULL
, NULL
,
4954 &left_mode
, &left_uid
, &left_gid
, NULL
);
4958 if (!sctx
->parent_root
|| sctx
->cur_inode_new
) {
4960 if (!S_ISLNK(sctx
->cur_inode_mode
))
4963 ret
= get_inode_info(sctx
->parent_root
, sctx
->cur_ino
,
4964 NULL
, NULL
, &right_mode
, &right_uid
,
4969 if (left_uid
!= right_uid
|| left_gid
!= right_gid
)
4971 if (!S_ISLNK(sctx
->cur_inode_mode
) && left_mode
!= right_mode
)
4975 if (S_ISREG(sctx
->cur_inode_mode
)) {
4976 if (need_send_hole(sctx
)) {
4977 if (sctx
->cur_inode_last_extent
== (u64
)-1 ||
4978 sctx
->cur_inode_last_extent
<
4979 sctx
->cur_inode_size
) {
4980 ret
= get_last_extent(sctx
, (u64
)-1);
4984 if (sctx
->cur_inode_last_extent
<
4985 sctx
->cur_inode_size
) {
4986 ret
= send_hole(sctx
, sctx
->cur_inode_size
);
4991 ret
= send_truncate(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
4992 sctx
->cur_inode_size
);
4998 ret
= send_chown(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
4999 left_uid
, left_gid
);
5004 ret
= send_chmod(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5011 * If other directory inodes depended on our current directory
5012 * inode's move/rename, now do their move/rename operations.
5014 if (!is_waiting_for_move(sctx
, sctx
->cur_ino
)) {
5015 ret
= apply_children_dir_moves(sctx
);
5019 * Need to send that every time, no matter if it actually
5020 * changed between the two trees as we have done changes to
5021 * the inode before. If our inode is a directory and it's
5022 * waiting to be moved/renamed, we will send its utimes when
5023 * it's moved/renamed, therefore we don't need to do it here.
5025 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5026 ret
= send_utimes(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
);
5035 static int changed_inode(struct send_ctx
*sctx
,
5036 enum btrfs_compare_tree_result result
)
5039 struct btrfs_key
*key
= sctx
->cmp_key
;
5040 struct btrfs_inode_item
*left_ii
= NULL
;
5041 struct btrfs_inode_item
*right_ii
= NULL
;
5045 sctx
->cur_ino
= key
->objectid
;
5046 sctx
->cur_inode_new_gen
= 0;
5047 sctx
->cur_inode_last_extent
= (u64
)-1;
5050 * Set send_progress to current inode. This will tell all get_cur_xxx
5051 * functions that the current inode's refs are not updated yet. Later,
5052 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5054 sctx
->send_progress
= sctx
->cur_ino
;
5056 if (result
== BTRFS_COMPARE_TREE_NEW
||
5057 result
== BTRFS_COMPARE_TREE_CHANGED
) {
5058 left_ii
= btrfs_item_ptr(sctx
->left_path
->nodes
[0],
5059 sctx
->left_path
->slots
[0],
5060 struct btrfs_inode_item
);
5061 left_gen
= btrfs_inode_generation(sctx
->left_path
->nodes
[0],
5064 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5065 sctx
->right_path
->slots
[0],
5066 struct btrfs_inode_item
);
5067 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5070 if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5071 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5072 sctx
->right_path
->slots
[0],
5073 struct btrfs_inode_item
);
5075 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5079 * The cur_ino = root dir case is special here. We can't treat
5080 * the inode as deleted+reused because it would generate a
5081 * stream that tries to delete/mkdir the root dir.
5083 if (left_gen
!= right_gen
&&
5084 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5085 sctx
->cur_inode_new_gen
= 1;
5088 if (result
== BTRFS_COMPARE_TREE_NEW
) {
5089 sctx
->cur_inode_gen
= left_gen
;
5090 sctx
->cur_inode_new
= 1;
5091 sctx
->cur_inode_deleted
= 0;
5092 sctx
->cur_inode_size
= btrfs_inode_size(
5093 sctx
->left_path
->nodes
[0], left_ii
);
5094 sctx
->cur_inode_mode
= btrfs_inode_mode(
5095 sctx
->left_path
->nodes
[0], left_ii
);
5096 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
5097 sctx
->left_path
->nodes
[0], left_ii
);
5098 if (sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5099 ret
= send_create_inode_if_needed(sctx
);
5100 } else if (result
== BTRFS_COMPARE_TREE_DELETED
) {
5101 sctx
->cur_inode_gen
= right_gen
;
5102 sctx
->cur_inode_new
= 0;
5103 sctx
->cur_inode_deleted
= 1;
5104 sctx
->cur_inode_size
= btrfs_inode_size(
5105 sctx
->right_path
->nodes
[0], right_ii
);
5106 sctx
->cur_inode_mode
= btrfs_inode_mode(
5107 sctx
->right_path
->nodes
[0], right_ii
);
5108 } else if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5110 * We need to do some special handling in case the inode was
5111 * reported as changed with a changed generation number. This
5112 * means that the original inode was deleted and new inode
5113 * reused the same inum. So we have to treat the old inode as
5114 * deleted and the new one as new.
5116 if (sctx
->cur_inode_new_gen
) {
5118 * First, process the inode as if it was deleted.
5120 sctx
->cur_inode_gen
= right_gen
;
5121 sctx
->cur_inode_new
= 0;
5122 sctx
->cur_inode_deleted
= 1;
5123 sctx
->cur_inode_size
= btrfs_inode_size(
5124 sctx
->right_path
->nodes
[0], right_ii
);
5125 sctx
->cur_inode_mode
= btrfs_inode_mode(
5126 sctx
->right_path
->nodes
[0], right_ii
);
5127 ret
= process_all_refs(sctx
,
5128 BTRFS_COMPARE_TREE_DELETED
);
5133 * Now process the inode as if it was new.
5135 sctx
->cur_inode_gen
= left_gen
;
5136 sctx
->cur_inode_new
= 1;
5137 sctx
->cur_inode_deleted
= 0;
5138 sctx
->cur_inode_size
= btrfs_inode_size(
5139 sctx
->left_path
->nodes
[0], left_ii
);
5140 sctx
->cur_inode_mode
= btrfs_inode_mode(
5141 sctx
->left_path
->nodes
[0], left_ii
);
5142 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
5143 sctx
->left_path
->nodes
[0], left_ii
);
5144 ret
= send_create_inode_if_needed(sctx
);
5148 ret
= process_all_refs(sctx
, BTRFS_COMPARE_TREE_NEW
);
5152 * Advance send_progress now as we did not get into
5153 * process_recorded_refs_if_needed in the new_gen case.
5155 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5158 * Now process all extents and xattrs of the inode as if
5159 * they were all new.
5161 ret
= process_all_extents(sctx
);
5164 ret
= process_all_new_xattrs(sctx
);
5168 sctx
->cur_inode_gen
= left_gen
;
5169 sctx
->cur_inode_new
= 0;
5170 sctx
->cur_inode_new_gen
= 0;
5171 sctx
->cur_inode_deleted
= 0;
5172 sctx
->cur_inode_size
= btrfs_inode_size(
5173 sctx
->left_path
->nodes
[0], left_ii
);
5174 sctx
->cur_inode_mode
= btrfs_inode_mode(
5175 sctx
->left_path
->nodes
[0], left_ii
);
5184 * We have to process new refs before deleted refs, but compare_trees gives us
5185 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5186 * first and later process them in process_recorded_refs.
5187 * For the cur_inode_new_gen case, we skip recording completely because
5188 * changed_inode did already initiate processing of refs. The reason for this is
5189 * that in this case, compare_tree actually compares the refs of 2 different
5190 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5191 * refs of the right tree as deleted and all refs of the left tree as new.
5193 static int changed_ref(struct send_ctx
*sctx
,
5194 enum btrfs_compare_tree_result result
)
5198 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
5200 if (!sctx
->cur_inode_new_gen
&&
5201 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
5202 if (result
== BTRFS_COMPARE_TREE_NEW
)
5203 ret
= record_new_ref(sctx
);
5204 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
5205 ret
= record_deleted_ref(sctx
);
5206 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
5207 ret
= record_changed_ref(sctx
);
5214 * Process new/deleted/changed xattrs. We skip processing in the
5215 * cur_inode_new_gen case because changed_inode did already initiate processing
5216 * of xattrs. The reason is the same as in changed_ref
5218 static int changed_xattr(struct send_ctx
*sctx
,
5219 enum btrfs_compare_tree_result result
)
5223 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
5225 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
5226 if (result
== BTRFS_COMPARE_TREE_NEW
)
5227 ret
= process_new_xattr(sctx
);
5228 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
5229 ret
= process_deleted_xattr(sctx
);
5230 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
5231 ret
= process_changed_xattr(sctx
);
5238 * Process new/deleted/changed extents. We skip processing in the
5239 * cur_inode_new_gen case because changed_inode did already initiate processing
5240 * of extents. The reason is the same as in changed_ref
5242 static int changed_extent(struct send_ctx
*sctx
,
5243 enum btrfs_compare_tree_result result
)
5247 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
5249 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
5250 if (result
!= BTRFS_COMPARE_TREE_DELETED
)
5251 ret
= process_extent(sctx
, sctx
->left_path
,
5258 static int dir_changed(struct send_ctx
*sctx
, u64 dir
)
5260 u64 orig_gen
, new_gen
;
5263 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &new_gen
, NULL
, NULL
,
5268 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &orig_gen
, NULL
,
5273 return (orig_gen
!= new_gen
) ? 1 : 0;
5276 static int compare_refs(struct send_ctx
*sctx
, struct btrfs_path
*path
,
5277 struct btrfs_key
*key
)
5279 struct btrfs_inode_extref
*extref
;
5280 struct extent_buffer
*leaf
;
5281 u64 dirid
= 0, last_dirid
= 0;
5288 /* Easy case, just check this one dirid */
5289 if (key
->type
== BTRFS_INODE_REF_KEY
) {
5290 dirid
= key
->offset
;
5292 ret
= dir_changed(sctx
, dirid
);
5296 leaf
= path
->nodes
[0];
5297 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
5298 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
5299 while (cur_offset
< item_size
) {
5300 extref
= (struct btrfs_inode_extref
*)(ptr
+
5302 dirid
= btrfs_inode_extref_parent(leaf
, extref
);
5303 ref_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
5304 cur_offset
+= ref_name_len
+ sizeof(*extref
);
5305 if (dirid
== last_dirid
)
5307 ret
= dir_changed(sctx
, dirid
);
5317 * Updates compare related fields in sctx and simply forwards to the actual
5318 * changed_xxx functions.
5320 static int changed_cb(struct btrfs_root
*left_root
,
5321 struct btrfs_root
*right_root
,
5322 struct btrfs_path
*left_path
,
5323 struct btrfs_path
*right_path
,
5324 struct btrfs_key
*key
,
5325 enum btrfs_compare_tree_result result
,
5329 struct send_ctx
*sctx
= ctx
;
5331 if (result
== BTRFS_COMPARE_TREE_SAME
) {
5332 if (key
->type
== BTRFS_INODE_REF_KEY
||
5333 key
->type
== BTRFS_INODE_EXTREF_KEY
) {
5334 ret
= compare_refs(sctx
, left_path
, key
);
5339 } else if (key
->type
== BTRFS_EXTENT_DATA_KEY
) {
5340 return maybe_send_hole(sctx
, left_path
, key
);
5344 result
= BTRFS_COMPARE_TREE_CHANGED
;
5348 sctx
->left_path
= left_path
;
5349 sctx
->right_path
= right_path
;
5350 sctx
->cmp_key
= key
;
5352 ret
= finish_inode_if_needed(sctx
, 0);
5356 /* Ignore non-FS objects */
5357 if (key
->objectid
== BTRFS_FREE_INO_OBJECTID
||
5358 key
->objectid
== BTRFS_FREE_SPACE_OBJECTID
)
5361 if (key
->type
== BTRFS_INODE_ITEM_KEY
)
5362 ret
= changed_inode(sctx
, result
);
5363 else if (key
->type
== BTRFS_INODE_REF_KEY
||
5364 key
->type
== BTRFS_INODE_EXTREF_KEY
)
5365 ret
= changed_ref(sctx
, result
);
5366 else if (key
->type
== BTRFS_XATTR_ITEM_KEY
)
5367 ret
= changed_xattr(sctx
, result
);
5368 else if (key
->type
== BTRFS_EXTENT_DATA_KEY
)
5369 ret
= changed_extent(sctx
, result
);
5375 static int full_send_tree(struct send_ctx
*sctx
)
5378 struct btrfs_root
*send_root
= sctx
->send_root
;
5379 struct btrfs_key key
;
5380 struct btrfs_key found_key
;
5381 struct btrfs_path
*path
;
5382 struct extent_buffer
*eb
;
5385 path
= alloc_path_for_send();
5389 key
.objectid
= BTRFS_FIRST_FREE_OBJECTID
;
5390 key
.type
= BTRFS_INODE_ITEM_KEY
;
5393 ret
= btrfs_search_slot_for_read(send_root
, &key
, path
, 1, 0);
5400 eb
= path
->nodes
[0];
5401 slot
= path
->slots
[0];
5402 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5404 ret
= changed_cb(send_root
, NULL
, path
, NULL
,
5405 &found_key
, BTRFS_COMPARE_TREE_NEW
, sctx
);
5409 key
.objectid
= found_key
.objectid
;
5410 key
.type
= found_key
.type
;
5411 key
.offset
= found_key
.offset
+ 1;
5413 ret
= btrfs_next_item(send_root
, path
);
5423 ret
= finish_inode_if_needed(sctx
, 1);
5426 btrfs_free_path(path
);
5430 static int send_subvol(struct send_ctx
*sctx
)
5434 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_STREAM_HEADER
)) {
5435 ret
= send_header(sctx
);
5440 ret
= send_subvol_begin(sctx
);
5444 if (sctx
->parent_root
) {
5445 ret
= btrfs_compare_trees(sctx
->send_root
, sctx
->parent_root
,
5449 ret
= finish_inode_if_needed(sctx
, 1);
5453 ret
= full_send_tree(sctx
);
5459 free_recorded_refs(sctx
);
5463 static void btrfs_root_dec_send_in_progress(struct btrfs_root
* root
)
5465 spin_lock(&root
->root_item_lock
);
5466 root
->send_in_progress
--;
5468 * Not much left to do, we don't know why it's unbalanced and
5469 * can't blindly reset it to 0.
5471 if (root
->send_in_progress
< 0)
5472 btrfs_err(root
->fs_info
,
5473 "send_in_progres unbalanced %d root %llu\n",
5474 root
->send_in_progress
, root
->root_key
.objectid
);
5475 spin_unlock(&root
->root_item_lock
);
5478 long btrfs_ioctl_send(struct file
*mnt_file
, void __user
*arg_
)
5481 struct btrfs_root
*send_root
;
5482 struct btrfs_root
*clone_root
;
5483 struct btrfs_fs_info
*fs_info
;
5484 struct btrfs_ioctl_send_args
*arg
= NULL
;
5485 struct btrfs_key key
;
5486 struct send_ctx
*sctx
= NULL
;
5488 u64
*clone_sources_tmp
= NULL
;
5489 int clone_sources_to_rollback
= 0;
5490 int sort_clone_roots
= 0;
5493 if (!capable(CAP_SYS_ADMIN
))
5496 send_root
= BTRFS_I(file_inode(mnt_file
))->root
;
5497 fs_info
= send_root
->fs_info
;
5500 * The subvolume must remain read-only during send, protect against
5501 * making it RW. This also protects against deletion.
5503 spin_lock(&send_root
->root_item_lock
);
5504 send_root
->send_in_progress
++;
5505 spin_unlock(&send_root
->root_item_lock
);
5508 * This is done when we lookup the root, it should already be complete
5509 * by the time we get here.
5511 WARN_ON(send_root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
);
5514 * Userspace tools do the checks and warn the user if it's
5517 if (!btrfs_root_readonly(send_root
)) {
5522 arg
= memdup_user(arg_
, sizeof(*arg
));
5529 if (!access_ok(VERIFY_READ
, arg
->clone_sources
,
5530 sizeof(*arg
->clone_sources
) *
5531 arg
->clone_sources_count
)) {
5536 if (arg
->flags
& ~BTRFS_SEND_FLAG_MASK
) {
5541 sctx
= kzalloc(sizeof(struct send_ctx
), GFP_NOFS
);
5547 INIT_LIST_HEAD(&sctx
->new_refs
);
5548 INIT_LIST_HEAD(&sctx
->deleted_refs
);
5549 INIT_RADIX_TREE(&sctx
->name_cache
, GFP_NOFS
);
5550 INIT_LIST_HEAD(&sctx
->name_cache_list
);
5552 sctx
->flags
= arg
->flags
;
5554 sctx
->send_filp
= fget(arg
->send_fd
);
5555 if (!sctx
->send_filp
) {
5560 sctx
->send_root
= send_root
;
5562 * Unlikely but possible, if the subvolume is marked for deletion but
5563 * is slow to remove the directory entry, send can still be started
5565 if (btrfs_root_dead(sctx
->send_root
)) {
5570 sctx
->clone_roots_cnt
= arg
->clone_sources_count
;
5572 sctx
->send_max_size
= BTRFS_SEND_BUF_SIZE
;
5573 sctx
->send_buf
= vmalloc(sctx
->send_max_size
);
5574 if (!sctx
->send_buf
) {
5579 sctx
->read_buf
= vmalloc(BTRFS_SEND_READ_SIZE
);
5580 if (!sctx
->read_buf
) {
5585 sctx
->pending_dir_moves
= RB_ROOT
;
5586 sctx
->waiting_dir_moves
= RB_ROOT
;
5587 sctx
->orphan_dirs
= RB_ROOT
;
5589 sctx
->clone_roots
= vzalloc(sizeof(struct clone_root
) *
5590 (arg
->clone_sources_count
+ 1));
5591 if (!sctx
->clone_roots
) {
5596 if (arg
->clone_sources_count
) {
5597 clone_sources_tmp
= vmalloc(arg
->clone_sources_count
*
5598 sizeof(*arg
->clone_sources
));
5599 if (!clone_sources_tmp
) {
5604 ret
= copy_from_user(clone_sources_tmp
, arg
->clone_sources
,
5605 arg
->clone_sources_count
*
5606 sizeof(*arg
->clone_sources
));
5612 for (i
= 0; i
< arg
->clone_sources_count
; i
++) {
5613 key
.objectid
= clone_sources_tmp
[i
];
5614 key
.type
= BTRFS_ROOT_ITEM_KEY
;
5615 key
.offset
= (u64
)-1;
5617 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5619 clone_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
5620 if (IS_ERR(clone_root
)) {
5621 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5622 ret
= PTR_ERR(clone_root
);
5625 clone_sources_to_rollback
= i
+ 1;
5626 spin_lock(&clone_root
->root_item_lock
);
5627 clone_root
->send_in_progress
++;
5628 if (!btrfs_root_readonly(clone_root
)) {
5629 spin_unlock(&clone_root
->root_item_lock
);
5630 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5634 spin_unlock(&clone_root
->root_item_lock
);
5635 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5637 sctx
->clone_roots
[i
].root
= clone_root
;
5639 vfree(clone_sources_tmp
);
5640 clone_sources_tmp
= NULL
;
5643 if (arg
->parent_root
) {
5644 key
.objectid
= arg
->parent_root
;
5645 key
.type
= BTRFS_ROOT_ITEM_KEY
;
5646 key
.offset
= (u64
)-1;
5648 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5650 sctx
->parent_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
5651 if (IS_ERR(sctx
->parent_root
)) {
5652 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5653 ret
= PTR_ERR(sctx
->parent_root
);
5657 spin_lock(&sctx
->parent_root
->root_item_lock
);
5658 sctx
->parent_root
->send_in_progress
++;
5659 if (!btrfs_root_readonly(sctx
->parent_root
) ||
5660 btrfs_root_dead(sctx
->parent_root
)) {
5661 spin_unlock(&sctx
->parent_root
->root_item_lock
);
5662 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5666 spin_unlock(&sctx
->parent_root
->root_item_lock
);
5668 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5672 * Clones from send_root are allowed, but only if the clone source
5673 * is behind the current send position. This is checked while searching
5674 * for possible clone sources.
5676 sctx
->clone_roots
[sctx
->clone_roots_cnt
++].root
= sctx
->send_root
;
5678 /* We do a bsearch later */
5679 sort(sctx
->clone_roots
, sctx
->clone_roots_cnt
,
5680 sizeof(*sctx
->clone_roots
), __clone_root_cmp_sort
,
5682 sort_clone_roots
= 1;
5684 current
->journal_info
= (void *)BTRFS_SEND_TRANS_STUB
;
5685 ret
= send_subvol(sctx
);
5686 current
->journal_info
= NULL
;
5690 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_END_CMD
)) {
5691 ret
= begin_cmd(sctx
, BTRFS_SEND_C_END
);
5694 ret
= send_cmd(sctx
);
5700 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
));
5701 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
)) {
5703 struct pending_dir_move
*pm
;
5705 n
= rb_first(&sctx
->pending_dir_moves
);
5706 pm
= rb_entry(n
, struct pending_dir_move
, node
);
5707 while (!list_empty(&pm
->list
)) {
5708 struct pending_dir_move
*pm2
;
5710 pm2
= list_first_entry(&pm
->list
,
5711 struct pending_dir_move
, list
);
5712 free_pending_move(sctx
, pm2
);
5714 free_pending_move(sctx
, pm
);
5717 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
));
5718 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
)) {
5720 struct waiting_dir_move
*dm
;
5722 n
= rb_first(&sctx
->waiting_dir_moves
);
5723 dm
= rb_entry(n
, struct waiting_dir_move
, node
);
5724 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
5728 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
));
5729 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
)) {
5731 struct orphan_dir_info
*odi
;
5733 n
= rb_first(&sctx
->orphan_dirs
);
5734 odi
= rb_entry(n
, struct orphan_dir_info
, node
);
5735 free_orphan_dir_info(sctx
, odi
);
5738 if (sort_clone_roots
) {
5739 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
5740 btrfs_root_dec_send_in_progress(
5741 sctx
->clone_roots
[i
].root
);
5743 for (i
= 0; sctx
&& i
< clone_sources_to_rollback
; i
++)
5744 btrfs_root_dec_send_in_progress(
5745 sctx
->clone_roots
[i
].root
);
5747 btrfs_root_dec_send_in_progress(send_root
);
5749 if (sctx
&& !IS_ERR_OR_NULL(sctx
->parent_root
))
5750 btrfs_root_dec_send_in_progress(sctx
->parent_root
);
5753 vfree(clone_sources_tmp
);
5756 if (sctx
->send_filp
)
5757 fput(sctx
->send_filp
);
5759 vfree(sctx
->clone_roots
);
5760 vfree(sctx
->send_buf
);
5761 vfree(sctx
->read_buf
);
5763 name_cache_free(sctx
);