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>
29 #include <linux/compat.h>
36 #include "btrfs_inode.h"
37 #include "transaction.h"
38 #include "compression.h"
41 * A fs_path is a helper to dynamically build path names with unknown size.
42 * It reallocates the internal buffer on demand.
43 * It allows fast adding of path elements on the right side (normal path) and
44 * fast adding to the left side (reversed path). A reversed path can also be
45 * unreversed if needed.
54 unsigned short buf_len
:15;
55 unsigned short reversed
:1;
59 * Average path length does not exceed 200 bytes, we'll have
60 * better packing in the slab and higher chance to satisfy
61 * a allocation later during send.
66 #define FS_PATH_INLINE_SIZE \
67 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
70 /* reused for each extent */
72 struct btrfs_root
*root
;
79 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
80 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
83 struct file
*send_filp
;
89 u64 cmd_send_size
[BTRFS_SEND_C_MAX
+ 1];
90 u64 flags
; /* 'flags' member of btrfs_ioctl_send_args is u64 */
92 struct btrfs_root
*send_root
;
93 struct btrfs_root
*parent_root
;
94 struct clone_root
*clone_roots
;
97 /* current state of the compare_tree call */
98 struct btrfs_path
*left_path
;
99 struct btrfs_path
*right_path
;
100 struct btrfs_key
*cmp_key
;
103 * infos of the currently processed inode. In case of deleted inodes,
104 * these are the values from the deleted inode.
109 int cur_inode_new_gen
;
110 int cur_inode_deleted
;
114 u64 cur_inode_last_extent
;
118 struct list_head new_refs
;
119 struct list_head deleted_refs
;
121 struct radix_tree_root name_cache
;
122 struct list_head name_cache_list
;
125 struct file_ra_state ra
;
130 * We process inodes by their increasing order, so if before an
131 * incremental send we reverse the parent/child relationship of
132 * directories such that a directory with a lower inode number was
133 * the parent of a directory with a higher inode number, and the one
134 * becoming the new parent got renamed too, we can't rename/move the
135 * directory with lower inode number when we finish processing it - we
136 * must process the directory with higher inode number first, then
137 * rename/move it and then rename/move the directory with lower inode
138 * number. Example follows.
140 * Tree state when the first send was performed:
152 * Tree state when the second (incremental) send is performed:
161 * The sequence of steps that lead to the second state was:
163 * mv /a/b/c/d /a/b/c2/d2
164 * mv /a/b/c /a/b/c2/d2/cc
166 * "c" has lower inode number, but we can't move it (2nd mv operation)
167 * before we move "d", which has higher inode number.
169 * So we just memorize which move/rename operations must be performed
170 * later when their respective parent is processed and moved/renamed.
173 /* Indexed by parent directory inode number. */
174 struct rb_root pending_dir_moves
;
177 * Reverse index, indexed by the inode number of a directory that
178 * is waiting for the move/rename of its immediate parent before its
179 * own move/rename can be performed.
181 struct rb_root waiting_dir_moves
;
184 * A directory that is going to be rm'ed might have a child directory
185 * which is in the pending directory moves index above. In this case,
186 * the directory can only be removed after the move/rename of its child
187 * is performed. Example:
207 * Sequence of steps that lead to the send snapshot:
208 * rm -f /a/b/c/foo.txt
210 * mv /a/b/c/x /a/b/YY
213 * When the child is processed, its move/rename is delayed until its
214 * parent is processed (as explained above), but all other operations
215 * like update utimes, chown, chgrp, etc, are performed and the paths
216 * that it uses for those operations must use the orphanized name of
217 * its parent (the directory we're going to rm later), so we need to
218 * memorize that name.
220 * Indexed by the inode number of the directory to be deleted.
222 struct rb_root orphan_dirs
;
225 struct pending_dir_move
{
227 struct list_head list
;
231 struct list_head update_refs
;
234 struct waiting_dir_move
{
238 * There might be some directory that could not be removed because it
239 * was waiting for this directory inode to be moved first. Therefore
240 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
246 struct orphan_dir_info
{
252 struct name_cache_entry
{
253 struct list_head list
;
255 * radix_tree has only 32bit entries but we need to handle 64bit inums.
256 * We use the lower 32bit of the 64bit inum to store it in the tree. If
257 * more then one inum would fall into the same entry, we use radix_list
258 * to store the additional entries. radix_list is also used to store
259 * entries where two entries have the same inum but different
262 struct list_head radix_list
;
268 int need_later_update
;
273 static void inconsistent_snapshot_error(struct send_ctx
*sctx
,
274 enum btrfs_compare_tree_result result
,
277 const char *result_string
;
280 case BTRFS_COMPARE_TREE_NEW
:
281 result_string
= "new";
283 case BTRFS_COMPARE_TREE_DELETED
:
284 result_string
= "deleted";
286 case BTRFS_COMPARE_TREE_CHANGED
:
287 result_string
= "updated";
289 case BTRFS_COMPARE_TREE_SAME
:
291 result_string
= "unchanged";
295 result_string
= "unexpected";
298 btrfs_err(sctx
->send_root
->fs_info
,
299 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
300 result_string
, what
, sctx
->cmp_key
->objectid
,
301 sctx
->send_root
->root_key
.objectid
,
303 sctx
->parent_root
->root_key
.objectid
: 0));
306 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
);
308 static struct waiting_dir_move
*
309 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
);
311 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
);
313 static int need_send_hole(struct send_ctx
*sctx
)
315 return (sctx
->parent_root
&& !sctx
->cur_inode_new
&&
316 !sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
&&
317 S_ISREG(sctx
->cur_inode_mode
));
320 static void fs_path_reset(struct fs_path
*p
)
323 p
->start
= p
->buf
+ p
->buf_len
- 1;
333 static struct fs_path
*fs_path_alloc(void)
337 p
= kmalloc(sizeof(*p
), GFP_KERNEL
);
341 p
->buf
= p
->inline_buf
;
342 p
->buf_len
= FS_PATH_INLINE_SIZE
;
347 static struct fs_path
*fs_path_alloc_reversed(void)
359 static void fs_path_free(struct fs_path
*p
)
363 if (p
->buf
!= p
->inline_buf
)
368 static int fs_path_len(struct fs_path
*p
)
370 return p
->end
- p
->start
;
373 static int fs_path_ensure_buf(struct fs_path
*p
, int len
)
381 if (p
->buf_len
>= len
)
384 if (len
> PATH_MAX
) {
389 path_len
= p
->end
- p
->start
;
390 old_buf_len
= p
->buf_len
;
393 * First time the inline_buf does not suffice
395 if (p
->buf
== p
->inline_buf
) {
396 tmp_buf
= kmalloc(len
, GFP_KERNEL
);
398 memcpy(tmp_buf
, p
->buf
, old_buf_len
);
400 tmp_buf
= krealloc(p
->buf
, len
, GFP_KERNEL
);
406 * The real size of the buffer is bigger, this will let the fast path
407 * happen most of the time
409 p
->buf_len
= ksize(p
->buf
);
412 tmp_buf
= p
->buf
+ old_buf_len
- path_len
- 1;
413 p
->end
= p
->buf
+ p
->buf_len
- 1;
414 p
->start
= p
->end
- path_len
;
415 memmove(p
->start
, tmp_buf
, path_len
+ 1);
418 p
->end
= p
->start
+ path_len
;
423 static int fs_path_prepare_for_add(struct fs_path
*p
, int name_len
,
429 new_len
= p
->end
- p
->start
+ name_len
;
430 if (p
->start
!= p
->end
)
432 ret
= fs_path_ensure_buf(p
, new_len
);
437 if (p
->start
!= p
->end
)
439 p
->start
-= name_len
;
440 *prepared
= p
->start
;
442 if (p
->start
!= p
->end
)
453 static int fs_path_add(struct fs_path
*p
, const char *name
, int name_len
)
458 ret
= fs_path_prepare_for_add(p
, name_len
, &prepared
);
461 memcpy(prepared
, name
, name_len
);
467 static int fs_path_add_path(struct fs_path
*p
, struct fs_path
*p2
)
472 ret
= fs_path_prepare_for_add(p
, p2
->end
- p2
->start
, &prepared
);
475 memcpy(prepared
, p2
->start
, p2
->end
- p2
->start
);
481 static int fs_path_add_from_extent_buffer(struct fs_path
*p
,
482 struct extent_buffer
*eb
,
483 unsigned long off
, int len
)
488 ret
= fs_path_prepare_for_add(p
, len
, &prepared
);
492 read_extent_buffer(eb
, prepared
, off
, len
);
498 static int fs_path_copy(struct fs_path
*p
, struct fs_path
*from
)
502 p
->reversed
= from
->reversed
;
505 ret
= fs_path_add_path(p
, from
);
511 static void fs_path_unreverse(struct fs_path
*p
)
520 len
= p
->end
- p
->start
;
522 p
->end
= p
->start
+ len
;
523 memmove(p
->start
, tmp
, len
+ 1);
527 static struct btrfs_path
*alloc_path_for_send(void)
529 struct btrfs_path
*path
;
531 path
= btrfs_alloc_path();
534 path
->search_commit_root
= 1;
535 path
->skip_locking
= 1;
536 path
->need_commit_sem
= 1;
540 static int write_buf(struct file
*filp
, const void *buf
, u32 len
, loff_t
*off
)
546 ret
= kernel_write(filp
, buf
+ pos
, len
- pos
, off
);
547 /* TODO handle that correctly */
548 /*if (ret == -ERESTARTSYS) {
562 static int tlv_put(struct send_ctx
*sctx
, u16 attr
, const void *data
, int len
)
564 struct btrfs_tlv_header
*hdr
;
565 int total_len
= sizeof(*hdr
) + len
;
566 int left
= sctx
->send_max_size
- sctx
->send_size
;
568 if (unlikely(left
< total_len
))
571 hdr
= (struct btrfs_tlv_header
*) (sctx
->send_buf
+ sctx
->send_size
);
572 hdr
->tlv_type
= cpu_to_le16(attr
);
573 hdr
->tlv_len
= cpu_to_le16(len
);
574 memcpy(hdr
+ 1, data
, len
);
575 sctx
->send_size
+= total_len
;
580 #define TLV_PUT_DEFINE_INT(bits) \
581 static int tlv_put_u##bits(struct send_ctx *sctx, \
582 u##bits attr, u##bits value) \
584 __le##bits __tmp = cpu_to_le##bits(value); \
585 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
588 TLV_PUT_DEFINE_INT(64)
590 static int tlv_put_string(struct send_ctx
*sctx
, u16 attr
,
591 const char *str
, int len
)
595 return tlv_put(sctx
, attr
, str
, len
);
598 static int tlv_put_uuid(struct send_ctx
*sctx
, u16 attr
,
601 return tlv_put(sctx
, attr
, uuid
, BTRFS_UUID_SIZE
);
604 static int tlv_put_btrfs_timespec(struct send_ctx
*sctx
, u16 attr
,
605 struct extent_buffer
*eb
,
606 struct btrfs_timespec
*ts
)
608 struct btrfs_timespec bts
;
609 read_extent_buffer(eb
, &bts
, (unsigned long)ts
, sizeof(bts
));
610 return tlv_put(sctx
, attr
, &bts
, sizeof(bts
));
614 #define TLV_PUT(sctx, attrtype, attrlen, data) \
616 ret = tlv_put(sctx, attrtype, attrlen, data); \
618 goto tlv_put_failure; \
621 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
623 ret = tlv_put_u##bits(sctx, attrtype, value); \
625 goto tlv_put_failure; \
628 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
629 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
630 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
631 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
632 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
634 ret = tlv_put_string(sctx, attrtype, str, len); \
636 goto tlv_put_failure; \
638 #define TLV_PUT_PATH(sctx, attrtype, p) \
640 ret = tlv_put_string(sctx, attrtype, p->start, \
641 p->end - p->start); \
643 goto tlv_put_failure; \
645 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
647 ret = tlv_put_uuid(sctx, attrtype, uuid); \
649 goto tlv_put_failure; \
651 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
653 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
655 goto tlv_put_failure; \
658 static int send_header(struct send_ctx
*sctx
)
660 struct btrfs_stream_header hdr
;
662 strcpy(hdr
.magic
, BTRFS_SEND_STREAM_MAGIC
);
663 hdr
.version
= cpu_to_le32(BTRFS_SEND_STREAM_VERSION
);
665 return write_buf(sctx
->send_filp
, &hdr
, sizeof(hdr
),
670 * For each command/item we want to send to userspace, we call this function.
672 static int begin_cmd(struct send_ctx
*sctx
, int cmd
)
674 struct btrfs_cmd_header
*hdr
;
676 if (WARN_ON(!sctx
->send_buf
))
679 BUG_ON(sctx
->send_size
);
681 sctx
->send_size
+= sizeof(*hdr
);
682 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
683 hdr
->cmd
= cpu_to_le16(cmd
);
688 static int send_cmd(struct send_ctx
*sctx
)
691 struct btrfs_cmd_header
*hdr
;
694 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
695 hdr
->len
= cpu_to_le32(sctx
->send_size
- sizeof(*hdr
));
698 crc
= btrfs_crc32c(0, (unsigned char *)sctx
->send_buf
, sctx
->send_size
);
699 hdr
->crc
= cpu_to_le32(crc
);
701 ret
= write_buf(sctx
->send_filp
, sctx
->send_buf
, sctx
->send_size
,
704 sctx
->total_send_size
+= sctx
->send_size
;
705 sctx
->cmd_send_size
[le16_to_cpu(hdr
->cmd
)] += sctx
->send_size
;
712 * Sends a move instruction to user space
714 static int send_rename(struct send_ctx
*sctx
,
715 struct fs_path
*from
, struct fs_path
*to
)
717 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
720 btrfs_debug(fs_info
, "send_rename %s -> %s", from
->start
, to
->start
);
722 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RENAME
);
726 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, from
);
727 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_TO
, to
);
729 ret
= send_cmd(sctx
);
737 * Sends a link instruction to user space
739 static int send_link(struct send_ctx
*sctx
,
740 struct fs_path
*path
, struct fs_path
*lnk
)
742 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
745 btrfs_debug(fs_info
, "send_link %s -> %s", path
->start
, lnk
->start
);
747 ret
= begin_cmd(sctx
, BTRFS_SEND_C_LINK
);
751 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
752 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, lnk
);
754 ret
= send_cmd(sctx
);
762 * Sends an unlink instruction to user space
764 static int send_unlink(struct send_ctx
*sctx
, struct fs_path
*path
)
766 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
769 btrfs_debug(fs_info
, "send_unlink %s", path
->start
);
771 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UNLINK
);
775 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
777 ret
= send_cmd(sctx
);
785 * Sends a rmdir instruction to user space
787 static int send_rmdir(struct send_ctx
*sctx
, struct fs_path
*path
)
789 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
792 btrfs_debug(fs_info
, "send_rmdir %s", path
->start
);
794 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RMDIR
);
798 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
800 ret
= send_cmd(sctx
);
808 * Helper function to retrieve some fields from an inode item.
810 static int __get_inode_info(struct btrfs_root
*root
, struct btrfs_path
*path
,
811 u64 ino
, u64
*size
, u64
*gen
, u64
*mode
, u64
*uid
,
815 struct btrfs_inode_item
*ii
;
816 struct btrfs_key key
;
819 key
.type
= BTRFS_INODE_ITEM_KEY
;
821 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
828 ii
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
829 struct btrfs_inode_item
);
831 *size
= btrfs_inode_size(path
->nodes
[0], ii
);
833 *gen
= btrfs_inode_generation(path
->nodes
[0], ii
);
835 *mode
= btrfs_inode_mode(path
->nodes
[0], ii
);
837 *uid
= btrfs_inode_uid(path
->nodes
[0], ii
);
839 *gid
= btrfs_inode_gid(path
->nodes
[0], ii
);
841 *rdev
= btrfs_inode_rdev(path
->nodes
[0], ii
);
846 static int get_inode_info(struct btrfs_root
*root
,
847 u64 ino
, u64
*size
, u64
*gen
,
848 u64
*mode
, u64
*uid
, u64
*gid
,
851 struct btrfs_path
*path
;
854 path
= alloc_path_for_send();
857 ret
= __get_inode_info(root
, path
, ino
, size
, gen
, mode
, uid
, gid
,
859 btrfs_free_path(path
);
863 typedef int (*iterate_inode_ref_t
)(int num
, u64 dir
, int index
,
868 * Helper function to iterate the entries in ONE btrfs_inode_ref or
869 * btrfs_inode_extref.
870 * The iterate callback may return a non zero value to stop iteration. This can
871 * be a negative value for error codes or 1 to simply stop it.
873 * path must point to the INODE_REF or INODE_EXTREF when called.
875 static int iterate_inode_ref(struct btrfs_root
*root
, struct btrfs_path
*path
,
876 struct btrfs_key
*found_key
, int resolve
,
877 iterate_inode_ref_t iterate
, void *ctx
)
879 struct extent_buffer
*eb
= path
->nodes
[0];
880 struct btrfs_item
*item
;
881 struct btrfs_inode_ref
*iref
;
882 struct btrfs_inode_extref
*extref
;
883 struct btrfs_path
*tmp_path
;
887 int slot
= path
->slots
[0];
894 unsigned long name_off
;
895 unsigned long elem_size
;
898 p
= fs_path_alloc_reversed();
902 tmp_path
= alloc_path_for_send();
909 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
910 ptr
= (unsigned long)btrfs_item_ptr(eb
, slot
,
911 struct btrfs_inode_ref
);
912 item
= btrfs_item_nr(slot
);
913 total
= btrfs_item_size(eb
, item
);
914 elem_size
= sizeof(*iref
);
916 ptr
= btrfs_item_ptr_offset(eb
, slot
);
917 total
= btrfs_item_size_nr(eb
, slot
);
918 elem_size
= sizeof(*extref
);
921 while (cur
< total
) {
924 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
925 iref
= (struct btrfs_inode_ref
*)(ptr
+ cur
);
926 name_len
= btrfs_inode_ref_name_len(eb
, iref
);
927 name_off
= (unsigned long)(iref
+ 1);
928 index
= btrfs_inode_ref_index(eb
, iref
);
929 dir
= found_key
->offset
;
931 extref
= (struct btrfs_inode_extref
*)(ptr
+ cur
);
932 name_len
= btrfs_inode_extref_name_len(eb
, extref
);
933 name_off
= (unsigned long)&extref
->name
;
934 index
= btrfs_inode_extref_index(eb
, extref
);
935 dir
= btrfs_inode_extref_parent(eb
, extref
);
939 start
= btrfs_ref_to_path(root
, tmp_path
, name_len
,
943 ret
= PTR_ERR(start
);
946 if (start
< p
->buf
) {
947 /* overflow , try again with larger buffer */
948 ret
= fs_path_ensure_buf(p
,
949 p
->buf_len
+ p
->buf
- start
);
952 start
= btrfs_ref_to_path(root
, tmp_path
,
957 ret
= PTR_ERR(start
);
960 BUG_ON(start
< p
->buf
);
964 ret
= fs_path_add_from_extent_buffer(p
, eb
, name_off
,
970 cur
+= elem_size
+ name_len
;
971 ret
= iterate(num
, dir
, index
, p
, ctx
);
978 btrfs_free_path(tmp_path
);
983 typedef int (*iterate_dir_item_t
)(int num
, struct btrfs_key
*di_key
,
984 const char *name
, int name_len
,
985 const char *data
, int data_len
,
989 * Helper function to iterate the entries in ONE btrfs_dir_item.
990 * The iterate callback may return a non zero value to stop iteration. This can
991 * be a negative value for error codes or 1 to simply stop it.
993 * path must point to the dir item when called.
995 static int iterate_dir_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
996 iterate_dir_item_t iterate
, void *ctx
)
999 struct extent_buffer
*eb
;
1000 struct btrfs_item
*item
;
1001 struct btrfs_dir_item
*di
;
1002 struct btrfs_key di_key
;
1015 * Start with a small buffer (1 page). If later we end up needing more
1016 * space, which can happen for xattrs on a fs with a leaf size greater
1017 * then the page size, attempt to increase the buffer. Typically xattr
1021 buf
= kmalloc(buf_len
, GFP_KERNEL
);
1027 eb
= path
->nodes
[0];
1028 slot
= path
->slots
[0];
1029 item
= btrfs_item_nr(slot
);
1030 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
1033 total
= btrfs_item_size(eb
, item
);
1036 while (cur
< total
) {
1037 name_len
= btrfs_dir_name_len(eb
, di
);
1038 data_len
= btrfs_dir_data_len(eb
, di
);
1039 type
= btrfs_dir_type(eb
, di
);
1040 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
1042 if (type
== BTRFS_FT_XATTR
) {
1043 if (name_len
> XATTR_NAME_MAX
) {
1044 ret
= -ENAMETOOLONG
;
1047 if (name_len
+ data_len
>
1048 BTRFS_MAX_XATTR_SIZE(root
->fs_info
)) {
1056 if (name_len
+ data_len
> PATH_MAX
) {
1057 ret
= -ENAMETOOLONG
;
1062 ret
= btrfs_is_name_len_valid(eb
, path
->slots
[0],
1063 (unsigned long)(di
+ 1), name_len
+ data_len
);
1068 if (name_len
+ data_len
> buf_len
) {
1069 buf_len
= name_len
+ data_len
;
1070 if (is_vmalloc_addr(buf
)) {
1074 char *tmp
= krealloc(buf
, buf_len
,
1075 GFP_KERNEL
| __GFP_NOWARN
);
1082 buf
= kvmalloc(buf_len
, GFP_KERNEL
);
1090 read_extent_buffer(eb
, buf
, (unsigned long)(di
+ 1),
1091 name_len
+ data_len
);
1093 len
= sizeof(*di
) + name_len
+ data_len
;
1094 di
= (struct btrfs_dir_item
*)((char *)di
+ len
);
1097 ret
= iterate(num
, &di_key
, buf
, name_len
, buf
+ name_len
,
1098 data_len
, type
, ctx
);
1114 static int __copy_first_ref(int num
, u64 dir
, int index
,
1115 struct fs_path
*p
, void *ctx
)
1118 struct fs_path
*pt
= ctx
;
1120 ret
= fs_path_copy(pt
, p
);
1124 /* we want the first only */
1129 * Retrieve the first path of an inode. If an inode has more then one
1130 * ref/hardlink, this is ignored.
1132 static int get_inode_path(struct btrfs_root
*root
,
1133 u64 ino
, struct fs_path
*path
)
1136 struct btrfs_key key
, found_key
;
1137 struct btrfs_path
*p
;
1139 p
= alloc_path_for_send();
1143 fs_path_reset(path
);
1146 key
.type
= BTRFS_INODE_REF_KEY
;
1149 ret
= btrfs_search_slot_for_read(root
, &key
, p
, 1, 0);
1156 btrfs_item_key_to_cpu(p
->nodes
[0], &found_key
, p
->slots
[0]);
1157 if (found_key
.objectid
!= ino
||
1158 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1159 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1164 ret
= iterate_inode_ref(root
, p
, &found_key
, 1,
1165 __copy_first_ref
, path
);
1175 struct backref_ctx
{
1176 struct send_ctx
*sctx
;
1178 struct btrfs_path
*path
;
1179 /* number of total found references */
1183 * used for clones found in send_root. clones found behind cur_objectid
1184 * and cur_offset are not considered as allowed clones.
1189 /* may be truncated in case it's the last extent in a file */
1192 /* data offset in the file extent item */
1195 /* Just to check for bugs in backref resolving */
1199 static int __clone_root_cmp_bsearch(const void *key
, const void *elt
)
1201 u64 root
= (u64
)(uintptr_t)key
;
1202 struct clone_root
*cr
= (struct clone_root
*)elt
;
1204 if (root
< cr
->root
->objectid
)
1206 if (root
> cr
->root
->objectid
)
1211 static int __clone_root_cmp_sort(const void *e1
, const void *e2
)
1213 struct clone_root
*cr1
= (struct clone_root
*)e1
;
1214 struct clone_root
*cr2
= (struct clone_root
*)e2
;
1216 if (cr1
->root
->objectid
< cr2
->root
->objectid
)
1218 if (cr1
->root
->objectid
> cr2
->root
->objectid
)
1224 * Called for every backref that is found for the current extent.
1225 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1227 static int __iterate_backrefs(u64 ino
, u64 offset
, u64 root
, void *ctx_
)
1229 struct backref_ctx
*bctx
= ctx_
;
1230 struct clone_root
*found
;
1234 /* First check if the root is in the list of accepted clone sources */
1235 found
= bsearch((void *)(uintptr_t)root
, bctx
->sctx
->clone_roots
,
1236 bctx
->sctx
->clone_roots_cnt
,
1237 sizeof(struct clone_root
),
1238 __clone_root_cmp_bsearch
);
1242 if (found
->root
== bctx
->sctx
->send_root
&&
1243 ino
== bctx
->cur_objectid
&&
1244 offset
== bctx
->cur_offset
) {
1245 bctx
->found_itself
= 1;
1249 * There are inodes that have extents that lie behind its i_size. Don't
1250 * accept clones from these extents.
1252 ret
= __get_inode_info(found
->root
, bctx
->path
, ino
, &i_size
, NULL
, NULL
,
1254 btrfs_release_path(bctx
->path
);
1258 if (offset
+ bctx
->data_offset
+ bctx
->extent_len
> i_size
)
1262 * Make sure we don't consider clones from send_root that are
1263 * behind the current inode/offset.
1265 if (found
->root
== bctx
->sctx
->send_root
) {
1267 * TODO for the moment we don't accept clones from the inode
1268 * that is currently send. We may change this when
1269 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1272 if (ino
>= bctx
->cur_objectid
)
1277 found
->found_refs
++;
1278 if (ino
< found
->ino
) {
1280 found
->offset
= offset
;
1281 } else if (found
->ino
== ino
) {
1283 * same extent found more then once in the same file.
1285 if (found
->offset
> offset
+ bctx
->extent_len
)
1286 found
->offset
= offset
;
1293 * Given an inode, offset and extent item, it finds a good clone for a clone
1294 * instruction. Returns -ENOENT when none could be found. The function makes
1295 * sure that the returned clone is usable at the point where sending is at the
1296 * moment. This means, that no clones are accepted which lie behind the current
1299 * path must point to the extent item when called.
1301 static int find_extent_clone(struct send_ctx
*sctx
,
1302 struct btrfs_path
*path
,
1303 u64 ino
, u64 data_offset
,
1305 struct clone_root
**found
)
1307 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
1313 u64 extent_item_pos
;
1315 struct btrfs_file_extent_item
*fi
;
1316 struct extent_buffer
*eb
= path
->nodes
[0];
1317 struct backref_ctx
*backref_ctx
= NULL
;
1318 struct clone_root
*cur_clone_root
;
1319 struct btrfs_key found_key
;
1320 struct btrfs_path
*tmp_path
;
1324 tmp_path
= alloc_path_for_send();
1328 /* We only use this path under the commit sem */
1329 tmp_path
->need_commit_sem
= 0;
1331 backref_ctx
= kmalloc(sizeof(*backref_ctx
), GFP_KERNEL
);
1337 backref_ctx
->path
= tmp_path
;
1339 if (data_offset
>= ino_size
) {
1341 * There may be extents that lie behind the file's size.
1342 * I at least had this in combination with snapshotting while
1343 * writing large files.
1349 fi
= btrfs_item_ptr(eb
, path
->slots
[0],
1350 struct btrfs_file_extent_item
);
1351 extent_type
= btrfs_file_extent_type(eb
, fi
);
1352 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1356 compressed
= btrfs_file_extent_compression(eb
, fi
);
1358 num_bytes
= btrfs_file_extent_num_bytes(eb
, fi
);
1359 disk_byte
= btrfs_file_extent_disk_bytenr(eb
, fi
);
1360 if (disk_byte
== 0) {
1364 logical
= disk_byte
+ btrfs_file_extent_offset(eb
, fi
);
1366 down_read(&fs_info
->commit_root_sem
);
1367 ret
= extent_from_logical(fs_info
, disk_byte
, tmp_path
,
1368 &found_key
, &flags
);
1369 up_read(&fs_info
->commit_root_sem
);
1370 btrfs_release_path(tmp_path
);
1374 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1380 * Setup the clone roots.
1382 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1383 cur_clone_root
= sctx
->clone_roots
+ i
;
1384 cur_clone_root
->ino
= (u64
)-1;
1385 cur_clone_root
->offset
= 0;
1386 cur_clone_root
->found_refs
= 0;
1389 backref_ctx
->sctx
= sctx
;
1390 backref_ctx
->found
= 0;
1391 backref_ctx
->cur_objectid
= ino
;
1392 backref_ctx
->cur_offset
= data_offset
;
1393 backref_ctx
->found_itself
= 0;
1394 backref_ctx
->extent_len
= num_bytes
;
1396 * For non-compressed extents iterate_extent_inodes() gives us extent
1397 * offsets that already take into account the data offset, but not for
1398 * compressed extents, since the offset is logical and not relative to
1399 * the physical extent locations. We must take this into account to
1400 * avoid sending clone offsets that go beyond the source file's size,
1401 * which would result in the clone ioctl failing with -EINVAL on the
1404 if (compressed
== BTRFS_COMPRESS_NONE
)
1405 backref_ctx
->data_offset
= 0;
1407 backref_ctx
->data_offset
= btrfs_file_extent_offset(eb
, fi
);
1410 * The last extent of a file may be too large due to page alignment.
1411 * We need to adjust extent_len in this case so that the checks in
1412 * __iterate_backrefs work.
1414 if (data_offset
+ num_bytes
>= ino_size
)
1415 backref_ctx
->extent_len
= ino_size
- data_offset
;
1418 * Now collect all backrefs.
1420 if (compressed
== BTRFS_COMPRESS_NONE
)
1421 extent_item_pos
= logical
- found_key
.objectid
;
1423 extent_item_pos
= 0;
1424 ret
= iterate_extent_inodes(fs_info
, found_key
.objectid
,
1425 extent_item_pos
, 1, __iterate_backrefs
,
1426 backref_ctx
, false);
1431 if (!backref_ctx
->found_itself
) {
1432 /* found a bug in backref code? */
1435 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1436 ino
, data_offset
, disk_byte
, found_key
.objectid
);
1440 btrfs_debug(fs_info
,
1441 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1442 data_offset
, ino
, num_bytes
, logical
);
1444 if (!backref_ctx
->found
)
1445 btrfs_debug(fs_info
, "no clones found");
1447 cur_clone_root
= NULL
;
1448 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1449 if (sctx
->clone_roots
[i
].found_refs
) {
1450 if (!cur_clone_root
)
1451 cur_clone_root
= sctx
->clone_roots
+ i
;
1452 else if (sctx
->clone_roots
[i
].root
== sctx
->send_root
)
1453 /* prefer clones from send_root over others */
1454 cur_clone_root
= sctx
->clone_roots
+ i
;
1459 if (cur_clone_root
) {
1460 *found
= cur_clone_root
;
1467 btrfs_free_path(tmp_path
);
1472 static int read_symlink(struct btrfs_root
*root
,
1474 struct fs_path
*dest
)
1477 struct btrfs_path
*path
;
1478 struct btrfs_key key
;
1479 struct btrfs_file_extent_item
*ei
;
1485 path
= alloc_path_for_send();
1490 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1492 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1497 * An empty symlink inode. Can happen in rare error paths when
1498 * creating a symlink (transaction committed before the inode
1499 * eviction handler removed the symlink inode items and a crash
1500 * happened in between or the subvol was snapshoted in between).
1501 * Print an informative message to dmesg/syslog so that the user
1502 * can delete the symlink.
1504 btrfs_err(root
->fs_info
,
1505 "Found empty symlink inode %llu at root %llu",
1506 ino
, root
->root_key
.objectid
);
1511 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1512 struct btrfs_file_extent_item
);
1513 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
1514 compression
= btrfs_file_extent_compression(path
->nodes
[0], ei
);
1515 BUG_ON(type
!= BTRFS_FILE_EXTENT_INLINE
);
1516 BUG_ON(compression
);
1518 off
= btrfs_file_extent_inline_start(ei
);
1519 len
= btrfs_file_extent_inline_len(path
->nodes
[0], path
->slots
[0], ei
);
1521 ret
= fs_path_add_from_extent_buffer(dest
, path
->nodes
[0], off
, len
);
1524 btrfs_free_path(path
);
1529 * Helper function to generate a file name that is unique in the root of
1530 * send_root and parent_root. This is used to generate names for orphan inodes.
1532 static int gen_unique_name(struct send_ctx
*sctx
,
1534 struct fs_path
*dest
)
1537 struct btrfs_path
*path
;
1538 struct btrfs_dir_item
*di
;
1543 path
= alloc_path_for_send();
1548 len
= snprintf(tmp
, sizeof(tmp
), "o%llu-%llu-%llu",
1550 ASSERT(len
< sizeof(tmp
));
1552 di
= btrfs_lookup_dir_item(NULL
, sctx
->send_root
,
1553 path
, BTRFS_FIRST_FREE_OBJECTID
,
1554 tmp
, strlen(tmp
), 0);
1555 btrfs_release_path(path
);
1561 /* not unique, try again */
1566 if (!sctx
->parent_root
) {
1572 di
= btrfs_lookup_dir_item(NULL
, sctx
->parent_root
,
1573 path
, BTRFS_FIRST_FREE_OBJECTID
,
1574 tmp
, strlen(tmp
), 0);
1575 btrfs_release_path(path
);
1581 /* not unique, try again */
1589 ret
= fs_path_add(dest
, tmp
, strlen(tmp
));
1592 btrfs_free_path(path
);
1597 inode_state_no_change
,
1598 inode_state_will_create
,
1599 inode_state_did_create
,
1600 inode_state_will_delete
,
1601 inode_state_did_delete
,
1604 static int get_cur_inode_state(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1612 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &left_gen
, NULL
, NULL
,
1614 if (ret
< 0 && ret
!= -ENOENT
)
1618 if (!sctx
->parent_root
) {
1619 right_ret
= -ENOENT
;
1621 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
, &right_gen
,
1622 NULL
, NULL
, NULL
, NULL
);
1623 if (ret
< 0 && ret
!= -ENOENT
)
1628 if (!left_ret
&& !right_ret
) {
1629 if (left_gen
== gen
&& right_gen
== gen
) {
1630 ret
= inode_state_no_change
;
1631 } else if (left_gen
== gen
) {
1632 if (ino
< sctx
->send_progress
)
1633 ret
= inode_state_did_create
;
1635 ret
= inode_state_will_create
;
1636 } else if (right_gen
== gen
) {
1637 if (ino
< sctx
->send_progress
)
1638 ret
= inode_state_did_delete
;
1640 ret
= inode_state_will_delete
;
1644 } else if (!left_ret
) {
1645 if (left_gen
== gen
) {
1646 if (ino
< sctx
->send_progress
)
1647 ret
= inode_state_did_create
;
1649 ret
= inode_state_will_create
;
1653 } else if (!right_ret
) {
1654 if (right_gen
== gen
) {
1655 if (ino
< sctx
->send_progress
)
1656 ret
= inode_state_did_delete
;
1658 ret
= inode_state_will_delete
;
1670 static int is_inode_existent(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1674 if (ino
== BTRFS_FIRST_FREE_OBJECTID
)
1677 ret
= get_cur_inode_state(sctx
, ino
, gen
);
1681 if (ret
== inode_state_no_change
||
1682 ret
== inode_state_did_create
||
1683 ret
== inode_state_will_delete
)
1693 * Helper function to lookup a dir item in a dir.
1695 static int lookup_dir_item_inode(struct btrfs_root
*root
,
1696 u64 dir
, const char *name
, int name_len
,
1701 struct btrfs_dir_item
*di
;
1702 struct btrfs_key key
;
1703 struct btrfs_path
*path
;
1705 path
= alloc_path_for_send();
1709 di
= btrfs_lookup_dir_item(NULL
, root
, path
,
1710 dir
, name
, name_len
, 0);
1719 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
1720 if (key
.type
== BTRFS_ROOT_ITEM_KEY
) {
1724 *found_inode
= key
.objectid
;
1725 *found_type
= btrfs_dir_type(path
->nodes
[0], di
);
1728 btrfs_free_path(path
);
1733 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1734 * generation of the parent dir and the name of the dir entry.
1736 static int get_first_ref(struct btrfs_root
*root
, u64 ino
,
1737 u64
*dir
, u64
*dir_gen
, struct fs_path
*name
)
1740 struct btrfs_key key
;
1741 struct btrfs_key found_key
;
1742 struct btrfs_path
*path
;
1746 path
= alloc_path_for_send();
1751 key
.type
= BTRFS_INODE_REF_KEY
;
1754 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
1758 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1760 if (ret
|| found_key
.objectid
!= ino
||
1761 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1762 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1767 if (found_key
.type
== BTRFS_INODE_REF_KEY
) {
1768 struct btrfs_inode_ref
*iref
;
1769 iref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1770 struct btrfs_inode_ref
);
1771 len
= btrfs_inode_ref_name_len(path
->nodes
[0], iref
);
1772 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1773 (unsigned long)(iref
+ 1),
1775 parent_dir
= found_key
.offset
;
1777 struct btrfs_inode_extref
*extref
;
1778 extref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1779 struct btrfs_inode_extref
);
1780 len
= btrfs_inode_extref_name_len(path
->nodes
[0], extref
);
1781 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1782 (unsigned long)&extref
->name
, len
);
1783 parent_dir
= btrfs_inode_extref_parent(path
->nodes
[0], extref
);
1787 btrfs_release_path(path
);
1790 ret
= get_inode_info(root
, parent_dir
, NULL
, dir_gen
, NULL
,
1799 btrfs_free_path(path
);
1803 static int is_first_ref(struct btrfs_root
*root
,
1805 const char *name
, int name_len
)
1808 struct fs_path
*tmp_name
;
1811 tmp_name
= fs_path_alloc();
1815 ret
= get_first_ref(root
, ino
, &tmp_dir
, NULL
, tmp_name
);
1819 if (dir
!= tmp_dir
|| name_len
!= fs_path_len(tmp_name
)) {
1824 ret
= !memcmp(tmp_name
->start
, name
, name_len
);
1827 fs_path_free(tmp_name
);
1832 * Used by process_recorded_refs to determine if a new ref would overwrite an
1833 * already existing ref. In case it detects an overwrite, it returns the
1834 * inode/gen in who_ino/who_gen.
1835 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1836 * to make sure later references to the overwritten inode are possible.
1837 * Orphanizing is however only required for the first ref of an inode.
1838 * process_recorded_refs does an additional is_first_ref check to see if
1839 * orphanizing is really required.
1841 static int will_overwrite_ref(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
1842 const char *name
, int name_len
,
1843 u64
*who_ino
, u64
*who_gen
, u64
*who_mode
)
1847 u64 other_inode
= 0;
1850 if (!sctx
->parent_root
)
1853 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1858 * If we have a parent root we need to verify that the parent dir was
1859 * not deleted and then re-created, if it was then we have no overwrite
1860 * and we can just unlink this entry.
1862 if (sctx
->parent_root
&& dir
!= BTRFS_FIRST_FREE_OBJECTID
) {
1863 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &gen
, NULL
,
1865 if (ret
< 0 && ret
!= -ENOENT
)
1875 ret
= lookup_dir_item_inode(sctx
->parent_root
, dir
, name
, name_len
,
1876 &other_inode
, &other_type
);
1877 if (ret
< 0 && ret
!= -ENOENT
)
1885 * Check if the overwritten ref was already processed. If yes, the ref
1886 * was already unlinked/moved, so we can safely assume that we will not
1887 * overwrite anything at this point in time.
1889 if (other_inode
> sctx
->send_progress
||
1890 is_waiting_for_move(sctx
, other_inode
)) {
1891 ret
= get_inode_info(sctx
->parent_root
, other_inode
, NULL
,
1892 who_gen
, who_mode
, NULL
, NULL
, NULL
);
1897 *who_ino
= other_inode
;
1907 * Checks if the ref was overwritten by an already processed inode. This is
1908 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1909 * thus the orphan name needs be used.
1910 * process_recorded_refs also uses it to avoid unlinking of refs that were
1913 static int did_overwrite_ref(struct send_ctx
*sctx
,
1914 u64 dir
, u64 dir_gen
,
1915 u64 ino
, u64 ino_gen
,
1916 const char *name
, int name_len
)
1923 if (!sctx
->parent_root
)
1926 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1930 if (dir
!= BTRFS_FIRST_FREE_OBJECTID
) {
1931 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &gen
, NULL
,
1933 if (ret
< 0 && ret
!= -ENOENT
)
1943 /* check if the ref was overwritten by another ref */
1944 ret
= lookup_dir_item_inode(sctx
->send_root
, dir
, name
, name_len
,
1945 &ow_inode
, &other_type
);
1946 if (ret
< 0 && ret
!= -ENOENT
)
1949 /* was never and will never be overwritten */
1954 ret
= get_inode_info(sctx
->send_root
, ow_inode
, NULL
, &gen
, NULL
, NULL
,
1959 if (ow_inode
== ino
&& gen
== ino_gen
) {
1965 * We know that it is or will be overwritten. Check this now.
1966 * The current inode being processed might have been the one that caused
1967 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1968 * the current inode being processed.
1970 if ((ow_inode
< sctx
->send_progress
) ||
1971 (ino
!= sctx
->cur_ino
&& ow_inode
== sctx
->cur_ino
&&
1972 gen
== sctx
->cur_inode_gen
))
1982 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1983 * that got overwritten. This is used by process_recorded_refs to determine
1984 * if it has to use the path as returned by get_cur_path or the orphan name.
1986 static int did_overwrite_first_ref(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1989 struct fs_path
*name
= NULL
;
1993 if (!sctx
->parent_root
)
1996 name
= fs_path_alloc();
2000 ret
= get_first_ref(sctx
->parent_root
, ino
, &dir
, &dir_gen
, name
);
2004 ret
= did_overwrite_ref(sctx
, dir
, dir_gen
, ino
, gen
,
2005 name
->start
, fs_path_len(name
));
2013 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2014 * so we need to do some special handling in case we have clashes. This function
2015 * takes care of this with the help of name_cache_entry::radix_list.
2016 * In case of error, nce is kfreed.
2018 static int name_cache_insert(struct send_ctx
*sctx
,
2019 struct name_cache_entry
*nce
)
2022 struct list_head
*nce_head
;
2024 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
2025 (unsigned long)nce
->ino
);
2027 nce_head
= kmalloc(sizeof(*nce_head
), GFP_KERNEL
);
2032 INIT_LIST_HEAD(nce_head
);
2034 ret
= radix_tree_insert(&sctx
->name_cache
, nce
->ino
, nce_head
);
2041 list_add_tail(&nce
->radix_list
, nce_head
);
2042 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
2043 sctx
->name_cache_size
++;
2048 static void name_cache_delete(struct send_ctx
*sctx
,
2049 struct name_cache_entry
*nce
)
2051 struct list_head
*nce_head
;
2053 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
2054 (unsigned long)nce
->ino
);
2056 btrfs_err(sctx
->send_root
->fs_info
,
2057 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2058 nce
->ino
, sctx
->name_cache_size
);
2061 list_del(&nce
->radix_list
);
2062 list_del(&nce
->list
);
2063 sctx
->name_cache_size
--;
2066 * We may not get to the final release of nce_head if the lookup fails
2068 if (nce_head
&& list_empty(nce_head
)) {
2069 radix_tree_delete(&sctx
->name_cache
, (unsigned long)nce
->ino
);
2074 static struct name_cache_entry
*name_cache_search(struct send_ctx
*sctx
,
2077 struct list_head
*nce_head
;
2078 struct name_cache_entry
*cur
;
2080 nce_head
= radix_tree_lookup(&sctx
->name_cache
, (unsigned long)ino
);
2084 list_for_each_entry(cur
, nce_head
, radix_list
) {
2085 if (cur
->ino
== ino
&& cur
->gen
== gen
)
2092 * Removes the entry from the list and adds it back to the end. This marks the
2093 * entry as recently used so that name_cache_clean_unused does not remove it.
2095 static void name_cache_used(struct send_ctx
*sctx
, struct name_cache_entry
*nce
)
2097 list_del(&nce
->list
);
2098 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
2102 * Remove some entries from the beginning of name_cache_list.
2104 static void name_cache_clean_unused(struct send_ctx
*sctx
)
2106 struct name_cache_entry
*nce
;
2108 if (sctx
->name_cache_size
< SEND_CTX_NAME_CACHE_CLEAN_SIZE
)
2111 while (sctx
->name_cache_size
> SEND_CTX_MAX_NAME_CACHE_SIZE
) {
2112 nce
= list_entry(sctx
->name_cache_list
.next
,
2113 struct name_cache_entry
, list
);
2114 name_cache_delete(sctx
, nce
);
2119 static void name_cache_free(struct send_ctx
*sctx
)
2121 struct name_cache_entry
*nce
;
2123 while (!list_empty(&sctx
->name_cache_list
)) {
2124 nce
= list_entry(sctx
->name_cache_list
.next
,
2125 struct name_cache_entry
, list
);
2126 name_cache_delete(sctx
, nce
);
2132 * Used by get_cur_path for each ref up to the root.
2133 * Returns 0 if it succeeded.
2134 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2135 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2136 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2137 * Returns <0 in case of error.
2139 static int __get_cur_name_and_parent(struct send_ctx
*sctx
,
2143 struct fs_path
*dest
)
2147 struct name_cache_entry
*nce
= NULL
;
2150 * First check if we already did a call to this function with the same
2151 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2152 * return the cached result.
2154 nce
= name_cache_search(sctx
, ino
, gen
);
2156 if (ino
< sctx
->send_progress
&& nce
->need_later_update
) {
2157 name_cache_delete(sctx
, nce
);
2161 name_cache_used(sctx
, nce
);
2162 *parent_ino
= nce
->parent_ino
;
2163 *parent_gen
= nce
->parent_gen
;
2164 ret
= fs_path_add(dest
, nce
->name
, nce
->name_len
);
2173 * If the inode is not existent yet, add the orphan name and return 1.
2174 * This should only happen for the parent dir that we determine in
2177 ret
= is_inode_existent(sctx
, ino
, gen
);
2182 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2190 * Depending on whether the inode was already processed or not, use
2191 * send_root or parent_root for ref lookup.
2193 if (ino
< sctx
->send_progress
)
2194 ret
= get_first_ref(sctx
->send_root
, ino
,
2195 parent_ino
, parent_gen
, dest
);
2197 ret
= get_first_ref(sctx
->parent_root
, ino
,
2198 parent_ino
, parent_gen
, dest
);
2203 * Check if the ref was overwritten by an inode's ref that was processed
2204 * earlier. If yes, treat as orphan and return 1.
2206 ret
= did_overwrite_ref(sctx
, *parent_ino
, *parent_gen
, ino
, gen
,
2207 dest
->start
, dest
->end
- dest
->start
);
2211 fs_path_reset(dest
);
2212 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2220 * Store the result of the lookup in the name cache.
2222 nce
= kmalloc(sizeof(*nce
) + fs_path_len(dest
) + 1, GFP_KERNEL
);
2230 nce
->parent_ino
= *parent_ino
;
2231 nce
->parent_gen
= *parent_gen
;
2232 nce
->name_len
= fs_path_len(dest
);
2234 strcpy(nce
->name
, dest
->start
);
2236 if (ino
< sctx
->send_progress
)
2237 nce
->need_later_update
= 0;
2239 nce
->need_later_update
= 1;
2241 nce_ret
= name_cache_insert(sctx
, nce
);
2244 name_cache_clean_unused(sctx
);
2251 * Magic happens here. This function returns the first ref to an inode as it
2252 * would look like while receiving the stream at this point in time.
2253 * We walk the path up to the root. For every inode in between, we check if it
2254 * was already processed/sent. If yes, we continue with the parent as found
2255 * in send_root. If not, we continue with the parent as found in parent_root.
2256 * If we encounter an inode that was deleted at this point in time, we use the
2257 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2258 * that were not created yet and overwritten inodes/refs.
2260 * When do we have have orphan inodes:
2261 * 1. When an inode is freshly created and thus no valid refs are available yet
2262 * 2. When a directory lost all it's refs (deleted) but still has dir items
2263 * inside which were not processed yet (pending for move/delete). If anyone
2264 * tried to get the path to the dir items, it would get a path inside that
2266 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2267 * of an unprocessed inode. If in that case the first ref would be
2268 * overwritten, the overwritten inode gets "orphanized". Later when we
2269 * process this overwritten inode, it is restored at a new place by moving
2272 * sctx->send_progress tells this function at which point in time receiving
2275 static int get_cur_path(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2276 struct fs_path
*dest
)
2279 struct fs_path
*name
= NULL
;
2280 u64 parent_inode
= 0;
2284 name
= fs_path_alloc();
2291 fs_path_reset(dest
);
2293 while (!stop
&& ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
2294 struct waiting_dir_move
*wdm
;
2296 fs_path_reset(name
);
2298 if (is_waiting_for_rm(sctx
, ino
)) {
2299 ret
= gen_unique_name(sctx
, ino
, gen
, name
);
2302 ret
= fs_path_add_path(dest
, name
);
2306 wdm
= get_waiting_dir_move(sctx
, ino
);
2307 if (wdm
&& wdm
->orphanized
) {
2308 ret
= gen_unique_name(sctx
, ino
, gen
, name
);
2311 ret
= get_first_ref(sctx
->parent_root
, ino
,
2312 &parent_inode
, &parent_gen
, name
);
2314 ret
= __get_cur_name_and_parent(sctx
, ino
, gen
,
2324 ret
= fs_path_add_path(dest
, name
);
2335 fs_path_unreverse(dest
);
2340 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2342 static int send_subvol_begin(struct send_ctx
*sctx
)
2345 struct btrfs_root
*send_root
= sctx
->send_root
;
2346 struct btrfs_root
*parent_root
= sctx
->parent_root
;
2347 struct btrfs_path
*path
;
2348 struct btrfs_key key
;
2349 struct btrfs_root_ref
*ref
;
2350 struct extent_buffer
*leaf
;
2354 path
= btrfs_alloc_path();
2358 name
= kmalloc(BTRFS_PATH_NAME_MAX
, GFP_KERNEL
);
2360 btrfs_free_path(path
);
2364 key
.objectid
= send_root
->objectid
;
2365 key
.type
= BTRFS_ROOT_BACKREF_KEY
;
2368 ret
= btrfs_search_slot_for_read(send_root
->fs_info
->tree_root
,
2377 leaf
= path
->nodes
[0];
2378 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2379 if (key
.type
!= BTRFS_ROOT_BACKREF_KEY
||
2380 key
.objectid
!= send_root
->objectid
) {
2384 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
2385 namelen
= btrfs_root_ref_name_len(leaf
, ref
);
2386 read_extent_buffer(leaf
, name
, (unsigned long)(ref
+ 1), namelen
);
2387 btrfs_release_path(path
);
2390 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SNAPSHOT
);
2394 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SUBVOL
);
2399 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_PATH
, name
, namelen
);
2401 if (!btrfs_is_empty_uuid(sctx
->send_root
->root_item
.received_uuid
))
2402 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_UUID
,
2403 sctx
->send_root
->root_item
.received_uuid
);
2405 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_UUID
,
2406 sctx
->send_root
->root_item
.uuid
);
2408 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CTRANSID
,
2409 le64_to_cpu(sctx
->send_root
->root_item
.ctransid
));
2411 if (!btrfs_is_empty_uuid(parent_root
->root_item
.received_uuid
))
2412 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2413 parent_root
->root_item
.received_uuid
);
2415 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2416 parent_root
->root_item
.uuid
);
2417 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
2418 le64_to_cpu(sctx
->parent_root
->root_item
.ctransid
));
2421 ret
= send_cmd(sctx
);
2425 btrfs_free_path(path
);
2430 static int send_truncate(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 size
)
2432 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2436 btrfs_debug(fs_info
, "send_truncate %llu size=%llu", ino
, size
);
2438 p
= fs_path_alloc();
2442 ret
= begin_cmd(sctx
, BTRFS_SEND_C_TRUNCATE
);
2446 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2449 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2450 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, size
);
2452 ret
= send_cmd(sctx
);
2460 static int send_chmod(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 mode
)
2462 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2466 btrfs_debug(fs_info
, "send_chmod %llu mode=%llu", ino
, mode
);
2468 p
= fs_path_alloc();
2472 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHMOD
);
2476 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2479 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2480 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
& 07777);
2482 ret
= send_cmd(sctx
);
2490 static int send_chown(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 uid
, u64 gid
)
2492 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2496 btrfs_debug(fs_info
, "send_chown %llu uid=%llu, gid=%llu",
2499 p
= fs_path_alloc();
2503 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHOWN
);
2507 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2510 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2511 TLV_PUT_U64(sctx
, BTRFS_SEND_A_UID
, uid
);
2512 TLV_PUT_U64(sctx
, BTRFS_SEND_A_GID
, gid
);
2514 ret
= send_cmd(sctx
);
2522 static int send_utimes(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
2524 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2526 struct fs_path
*p
= NULL
;
2527 struct btrfs_inode_item
*ii
;
2528 struct btrfs_path
*path
= NULL
;
2529 struct extent_buffer
*eb
;
2530 struct btrfs_key key
;
2533 btrfs_debug(fs_info
, "send_utimes %llu", ino
);
2535 p
= fs_path_alloc();
2539 path
= alloc_path_for_send();
2546 key
.type
= BTRFS_INODE_ITEM_KEY
;
2548 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2554 eb
= path
->nodes
[0];
2555 slot
= path
->slots
[0];
2556 ii
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
2558 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UTIMES
);
2562 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2565 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2566 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_ATIME
, eb
, &ii
->atime
);
2567 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_MTIME
, eb
, &ii
->mtime
);
2568 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_CTIME
, eb
, &ii
->ctime
);
2569 /* TODO Add otime support when the otime patches get into upstream */
2571 ret
= send_cmd(sctx
);
2576 btrfs_free_path(path
);
2581 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2582 * a valid path yet because we did not process the refs yet. So, the inode
2583 * is created as orphan.
2585 static int send_create_inode(struct send_ctx
*sctx
, u64 ino
)
2587 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2595 btrfs_debug(fs_info
, "send_create_inode %llu", ino
);
2597 p
= fs_path_alloc();
2601 if (ino
!= sctx
->cur_ino
) {
2602 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &gen
, &mode
,
2607 gen
= sctx
->cur_inode_gen
;
2608 mode
= sctx
->cur_inode_mode
;
2609 rdev
= sctx
->cur_inode_rdev
;
2612 if (S_ISREG(mode
)) {
2613 cmd
= BTRFS_SEND_C_MKFILE
;
2614 } else if (S_ISDIR(mode
)) {
2615 cmd
= BTRFS_SEND_C_MKDIR
;
2616 } else if (S_ISLNK(mode
)) {
2617 cmd
= BTRFS_SEND_C_SYMLINK
;
2618 } else if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
2619 cmd
= BTRFS_SEND_C_MKNOD
;
2620 } else if (S_ISFIFO(mode
)) {
2621 cmd
= BTRFS_SEND_C_MKFIFO
;
2622 } else if (S_ISSOCK(mode
)) {
2623 cmd
= BTRFS_SEND_C_MKSOCK
;
2625 btrfs_warn(sctx
->send_root
->fs_info
, "unexpected inode type %o",
2626 (int)(mode
& S_IFMT
));
2631 ret
= begin_cmd(sctx
, cmd
);
2635 ret
= gen_unique_name(sctx
, ino
, gen
, p
);
2639 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2640 TLV_PUT_U64(sctx
, BTRFS_SEND_A_INO
, ino
);
2642 if (S_ISLNK(mode
)) {
2644 ret
= read_symlink(sctx
->send_root
, ino
, p
);
2647 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, p
);
2648 } else if (S_ISCHR(mode
) || S_ISBLK(mode
) ||
2649 S_ISFIFO(mode
) || S_ISSOCK(mode
)) {
2650 TLV_PUT_U64(sctx
, BTRFS_SEND_A_RDEV
, new_encode_dev(rdev
));
2651 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
);
2654 ret
= send_cmd(sctx
);
2666 * We need some special handling for inodes that get processed before the parent
2667 * directory got created. See process_recorded_refs for details.
2668 * This function does the check if we already created the dir out of order.
2670 static int did_create_dir(struct send_ctx
*sctx
, u64 dir
)
2673 struct btrfs_path
*path
= NULL
;
2674 struct btrfs_key key
;
2675 struct btrfs_key found_key
;
2676 struct btrfs_key di_key
;
2677 struct extent_buffer
*eb
;
2678 struct btrfs_dir_item
*di
;
2681 path
= alloc_path_for_send();
2688 key
.type
= BTRFS_DIR_INDEX_KEY
;
2690 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2695 eb
= path
->nodes
[0];
2696 slot
= path
->slots
[0];
2697 if (slot
>= btrfs_header_nritems(eb
)) {
2698 ret
= btrfs_next_leaf(sctx
->send_root
, path
);
2701 } else if (ret
> 0) {
2708 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
2709 if (found_key
.objectid
!= key
.objectid
||
2710 found_key
.type
!= key
.type
) {
2715 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
2716 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
2718 if (di_key
.type
!= BTRFS_ROOT_ITEM_KEY
&&
2719 di_key
.objectid
< sctx
->send_progress
) {
2728 btrfs_free_path(path
);
2733 * Only creates the inode if it is:
2734 * 1. Not a directory
2735 * 2. Or a directory which was not created already due to out of order
2736 * directories. See did_create_dir and process_recorded_refs for details.
2738 static int send_create_inode_if_needed(struct send_ctx
*sctx
)
2742 if (S_ISDIR(sctx
->cur_inode_mode
)) {
2743 ret
= did_create_dir(sctx
, sctx
->cur_ino
);
2752 ret
= send_create_inode(sctx
, sctx
->cur_ino
);
2760 struct recorded_ref
{
2761 struct list_head list
;
2763 struct fs_path
*full_path
;
2769 static void set_ref_path(struct recorded_ref
*ref
, struct fs_path
*path
)
2771 ref
->full_path
= path
;
2772 ref
->name
= (char *)kbasename(ref
->full_path
->start
);
2773 ref
->name_len
= ref
->full_path
->end
- ref
->name
;
2777 * We need to process new refs before deleted refs, but compare_tree gives us
2778 * everything mixed. So we first record all refs and later process them.
2779 * This function is a helper to record one ref.
2781 static int __record_ref(struct list_head
*head
, u64 dir
,
2782 u64 dir_gen
, struct fs_path
*path
)
2784 struct recorded_ref
*ref
;
2786 ref
= kmalloc(sizeof(*ref
), GFP_KERNEL
);
2791 ref
->dir_gen
= dir_gen
;
2792 set_ref_path(ref
, path
);
2793 list_add_tail(&ref
->list
, head
);
2797 static int dup_ref(struct recorded_ref
*ref
, struct list_head
*list
)
2799 struct recorded_ref
*new;
2801 new = kmalloc(sizeof(*ref
), GFP_KERNEL
);
2805 new->dir
= ref
->dir
;
2806 new->dir_gen
= ref
->dir_gen
;
2807 new->full_path
= NULL
;
2808 INIT_LIST_HEAD(&new->list
);
2809 list_add_tail(&new->list
, list
);
2813 static void __free_recorded_refs(struct list_head
*head
)
2815 struct recorded_ref
*cur
;
2817 while (!list_empty(head
)) {
2818 cur
= list_entry(head
->next
, struct recorded_ref
, list
);
2819 fs_path_free(cur
->full_path
);
2820 list_del(&cur
->list
);
2825 static void free_recorded_refs(struct send_ctx
*sctx
)
2827 __free_recorded_refs(&sctx
->new_refs
);
2828 __free_recorded_refs(&sctx
->deleted_refs
);
2832 * Renames/moves a file/dir to its orphan name. Used when the first
2833 * ref of an unprocessed inode gets overwritten and for all non empty
2836 static int orphanize_inode(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2837 struct fs_path
*path
)
2840 struct fs_path
*orphan
;
2842 orphan
= fs_path_alloc();
2846 ret
= gen_unique_name(sctx
, ino
, gen
, orphan
);
2850 ret
= send_rename(sctx
, path
, orphan
);
2853 fs_path_free(orphan
);
2857 static struct orphan_dir_info
*
2858 add_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2860 struct rb_node
**p
= &sctx
->orphan_dirs
.rb_node
;
2861 struct rb_node
*parent
= NULL
;
2862 struct orphan_dir_info
*entry
, *odi
;
2864 odi
= kmalloc(sizeof(*odi
), GFP_KERNEL
);
2866 return ERR_PTR(-ENOMEM
);
2872 entry
= rb_entry(parent
, struct orphan_dir_info
, node
);
2873 if (dir_ino
< entry
->ino
) {
2875 } else if (dir_ino
> entry
->ino
) {
2876 p
= &(*p
)->rb_right
;
2883 rb_link_node(&odi
->node
, parent
, p
);
2884 rb_insert_color(&odi
->node
, &sctx
->orphan_dirs
);
2888 static struct orphan_dir_info
*
2889 get_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2891 struct rb_node
*n
= sctx
->orphan_dirs
.rb_node
;
2892 struct orphan_dir_info
*entry
;
2895 entry
= rb_entry(n
, struct orphan_dir_info
, node
);
2896 if (dir_ino
< entry
->ino
)
2898 else if (dir_ino
> entry
->ino
)
2906 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
)
2908 struct orphan_dir_info
*odi
= get_orphan_dir_info(sctx
, dir_ino
);
2913 static void free_orphan_dir_info(struct send_ctx
*sctx
,
2914 struct orphan_dir_info
*odi
)
2918 rb_erase(&odi
->node
, &sctx
->orphan_dirs
);
2923 * Returns 1 if a directory can be removed at this point in time.
2924 * We check this by iterating all dir items and checking if the inode behind
2925 * the dir item was already processed.
2927 static int can_rmdir(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
2931 struct btrfs_root
*root
= sctx
->parent_root
;
2932 struct btrfs_path
*path
;
2933 struct btrfs_key key
;
2934 struct btrfs_key found_key
;
2935 struct btrfs_key loc
;
2936 struct btrfs_dir_item
*di
;
2939 * Don't try to rmdir the top/root subvolume dir.
2941 if (dir
== BTRFS_FIRST_FREE_OBJECTID
)
2944 path
= alloc_path_for_send();
2949 key
.type
= BTRFS_DIR_INDEX_KEY
;
2951 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2956 struct waiting_dir_move
*dm
;
2958 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
2959 ret
= btrfs_next_leaf(root
, path
);
2966 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2968 if (found_key
.objectid
!= key
.objectid
||
2969 found_key
.type
!= key
.type
)
2972 di
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2973 struct btrfs_dir_item
);
2974 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &loc
);
2976 dm
= get_waiting_dir_move(sctx
, loc
.objectid
);
2978 struct orphan_dir_info
*odi
;
2980 odi
= add_orphan_dir_info(sctx
, dir
);
2986 dm
->rmdir_ino
= dir
;
2991 if (loc
.objectid
> send_progress
) {
2992 struct orphan_dir_info
*odi
;
2994 odi
= get_orphan_dir_info(sctx
, dir
);
2995 free_orphan_dir_info(sctx
, odi
);
3006 btrfs_free_path(path
);
3010 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
)
3012 struct waiting_dir_move
*entry
= get_waiting_dir_move(sctx
, ino
);
3014 return entry
!= NULL
;
3017 static int add_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
, bool orphanized
)
3019 struct rb_node
**p
= &sctx
->waiting_dir_moves
.rb_node
;
3020 struct rb_node
*parent
= NULL
;
3021 struct waiting_dir_move
*entry
, *dm
;
3023 dm
= kmalloc(sizeof(*dm
), GFP_KERNEL
);
3028 dm
->orphanized
= orphanized
;
3032 entry
= rb_entry(parent
, struct waiting_dir_move
, node
);
3033 if (ino
< entry
->ino
) {
3035 } else if (ino
> entry
->ino
) {
3036 p
= &(*p
)->rb_right
;
3043 rb_link_node(&dm
->node
, parent
, p
);
3044 rb_insert_color(&dm
->node
, &sctx
->waiting_dir_moves
);
3048 static struct waiting_dir_move
*
3049 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
)
3051 struct rb_node
*n
= sctx
->waiting_dir_moves
.rb_node
;
3052 struct waiting_dir_move
*entry
;
3055 entry
= rb_entry(n
, struct waiting_dir_move
, node
);
3056 if (ino
< entry
->ino
)
3058 else if (ino
> entry
->ino
)
3066 static void free_waiting_dir_move(struct send_ctx
*sctx
,
3067 struct waiting_dir_move
*dm
)
3071 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
3075 static int add_pending_dir_move(struct send_ctx
*sctx
,
3079 struct list_head
*new_refs
,
3080 struct list_head
*deleted_refs
,
3081 const bool is_orphan
)
3083 struct rb_node
**p
= &sctx
->pending_dir_moves
.rb_node
;
3084 struct rb_node
*parent
= NULL
;
3085 struct pending_dir_move
*entry
= NULL
, *pm
;
3086 struct recorded_ref
*cur
;
3090 pm
= kmalloc(sizeof(*pm
), GFP_KERNEL
);
3093 pm
->parent_ino
= parent_ino
;
3096 INIT_LIST_HEAD(&pm
->list
);
3097 INIT_LIST_HEAD(&pm
->update_refs
);
3098 RB_CLEAR_NODE(&pm
->node
);
3102 entry
= rb_entry(parent
, struct pending_dir_move
, node
);
3103 if (parent_ino
< entry
->parent_ino
) {
3105 } else if (parent_ino
> entry
->parent_ino
) {
3106 p
= &(*p
)->rb_right
;
3113 list_for_each_entry(cur
, deleted_refs
, list
) {
3114 ret
= dup_ref(cur
, &pm
->update_refs
);
3118 list_for_each_entry(cur
, new_refs
, list
) {
3119 ret
= dup_ref(cur
, &pm
->update_refs
);
3124 ret
= add_waiting_dir_move(sctx
, pm
->ino
, is_orphan
);
3129 list_add_tail(&pm
->list
, &entry
->list
);
3131 rb_link_node(&pm
->node
, parent
, p
);
3132 rb_insert_color(&pm
->node
, &sctx
->pending_dir_moves
);
3137 __free_recorded_refs(&pm
->update_refs
);
3143 static struct pending_dir_move
*get_pending_dir_moves(struct send_ctx
*sctx
,
3146 struct rb_node
*n
= sctx
->pending_dir_moves
.rb_node
;
3147 struct pending_dir_move
*entry
;
3150 entry
= rb_entry(n
, struct pending_dir_move
, node
);
3151 if (parent_ino
< entry
->parent_ino
)
3153 else if (parent_ino
> entry
->parent_ino
)
3161 static int path_loop(struct send_ctx
*sctx
, struct fs_path
*name
,
3162 u64 ino
, u64 gen
, u64
*ancestor_ino
)
3165 u64 parent_inode
= 0;
3167 u64 start_ino
= ino
;
3170 while (ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
3171 fs_path_reset(name
);
3173 if (is_waiting_for_rm(sctx
, ino
))
3175 if (is_waiting_for_move(sctx
, ino
)) {
3176 if (*ancestor_ino
== 0)
3177 *ancestor_ino
= ino
;
3178 ret
= get_first_ref(sctx
->parent_root
, ino
,
3179 &parent_inode
, &parent_gen
, name
);
3181 ret
= __get_cur_name_and_parent(sctx
, ino
, gen
,
3191 if (parent_inode
== start_ino
) {
3193 if (*ancestor_ino
== 0)
3194 *ancestor_ino
= ino
;
3203 static int apply_dir_move(struct send_ctx
*sctx
, struct pending_dir_move
*pm
)
3205 struct fs_path
*from_path
= NULL
;
3206 struct fs_path
*to_path
= NULL
;
3207 struct fs_path
*name
= NULL
;
3208 u64 orig_progress
= sctx
->send_progress
;
3209 struct recorded_ref
*cur
;
3210 u64 parent_ino
, parent_gen
;
3211 struct waiting_dir_move
*dm
= NULL
;
3217 name
= fs_path_alloc();
3218 from_path
= fs_path_alloc();
3219 if (!name
|| !from_path
) {
3224 dm
= get_waiting_dir_move(sctx
, pm
->ino
);
3226 rmdir_ino
= dm
->rmdir_ino
;
3227 is_orphan
= dm
->orphanized
;
3228 free_waiting_dir_move(sctx
, dm
);
3231 ret
= gen_unique_name(sctx
, pm
->ino
,
3232 pm
->gen
, from_path
);
3234 ret
= get_first_ref(sctx
->parent_root
, pm
->ino
,
3235 &parent_ino
, &parent_gen
, name
);
3238 ret
= get_cur_path(sctx
, parent_ino
, parent_gen
,
3242 ret
= fs_path_add_path(from_path
, name
);
3247 sctx
->send_progress
= sctx
->cur_ino
+ 1;
3248 ret
= path_loop(sctx
, name
, pm
->ino
, pm
->gen
, &ancestor
);
3252 LIST_HEAD(deleted_refs
);
3253 ASSERT(ancestor
> BTRFS_FIRST_FREE_OBJECTID
);
3254 ret
= add_pending_dir_move(sctx
, pm
->ino
, pm
->gen
, ancestor
,
3255 &pm
->update_refs
, &deleted_refs
,
3260 dm
= get_waiting_dir_move(sctx
, pm
->ino
);
3262 dm
->rmdir_ino
= rmdir_ino
;
3266 fs_path_reset(name
);
3269 ret
= get_cur_path(sctx
, pm
->ino
, pm
->gen
, to_path
);
3273 ret
= send_rename(sctx
, from_path
, to_path
);
3278 struct orphan_dir_info
*odi
;
3280 odi
= get_orphan_dir_info(sctx
, rmdir_ino
);
3282 /* already deleted */
3285 ret
= can_rmdir(sctx
, rmdir_ino
, odi
->gen
, sctx
->cur_ino
);
3291 name
= fs_path_alloc();
3296 ret
= get_cur_path(sctx
, rmdir_ino
, odi
->gen
, name
);
3299 ret
= send_rmdir(sctx
, name
);
3302 free_orphan_dir_info(sctx
, odi
);
3306 ret
= send_utimes(sctx
, pm
->ino
, pm
->gen
);
3311 * After rename/move, need to update the utimes of both new parent(s)
3312 * and old parent(s).
3314 list_for_each_entry(cur
, &pm
->update_refs
, list
) {
3316 * The parent inode might have been deleted in the send snapshot
3318 ret
= get_inode_info(sctx
->send_root
, cur
->dir
, NULL
,
3319 NULL
, NULL
, NULL
, NULL
, NULL
);
3320 if (ret
== -ENOENT
) {
3327 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
3334 fs_path_free(from_path
);
3335 fs_path_free(to_path
);
3336 sctx
->send_progress
= orig_progress
;
3341 static void free_pending_move(struct send_ctx
*sctx
, struct pending_dir_move
*m
)
3343 if (!list_empty(&m
->list
))
3345 if (!RB_EMPTY_NODE(&m
->node
))
3346 rb_erase(&m
->node
, &sctx
->pending_dir_moves
);
3347 __free_recorded_refs(&m
->update_refs
);
3351 static void tail_append_pending_moves(struct pending_dir_move
*moves
,
3352 struct list_head
*stack
)
3354 if (list_empty(&moves
->list
)) {
3355 list_add_tail(&moves
->list
, stack
);
3358 list_splice_init(&moves
->list
, &list
);
3359 list_add_tail(&moves
->list
, stack
);
3360 list_splice_tail(&list
, stack
);
3364 static int apply_children_dir_moves(struct send_ctx
*sctx
)
3366 struct pending_dir_move
*pm
;
3367 struct list_head stack
;
3368 u64 parent_ino
= sctx
->cur_ino
;
3371 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3375 INIT_LIST_HEAD(&stack
);
3376 tail_append_pending_moves(pm
, &stack
);
3378 while (!list_empty(&stack
)) {
3379 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3380 parent_ino
= pm
->ino
;
3381 ret
= apply_dir_move(sctx
, pm
);
3382 free_pending_move(sctx
, pm
);
3385 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3387 tail_append_pending_moves(pm
, &stack
);
3392 while (!list_empty(&stack
)) {
3393 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3394 free_pending_move(sctx
, pm
);
3400 * We might need to delay a directory rename even when no ancestor directory
3401 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3402 * renamed. This happens when we rename a directory to the old name (the name
3403 * in the parent root) of some other unrelated directory that got its rename
3404 * delayed due to some ancestor with higher number that got renamed.
3410 * |---- a/ (ino 257)
3411 * | |---- file (ino 260)
3413 * |---- b/ (ino 258)
3414 * |---- c/ (ino 259)
3418 * |---- a/ (ino 258)
3419 * |---- x/ (ino 259)
3420 * |---- y/ (ino 257)
3421 * |----- file (ino 260)
3423 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3424 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3425 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3428 * 1 - rename 259 from 'c' to 'x'
3429 * 2 - rename 257 from 'a' to 'x/y'
3430 * 3 - rename 258 from 'b' to 'a'
3432 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3433 * be done right away and < 0 on error.
3435 static int wait_for_dest_dir_move(struct send_ctx
*sctx
,
3436 struct recorded_ref
*parent_ref
,
3437 const bool is_orphan
)
3439 struct btrfs_fs_info
*fs_info
= sctx
->parent_root
->fs_info
;
3440 struct btrfs_path
*path
;
3441 struct btrfs_key key
;
3442 struct btrfs_key di_key
;
3443 struct btrfs_dir_item
*di
;
3447 struct waiting_dir_move
*wdm
;
3449 if (RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
))
3452 path
= alloc_path_for_send();
3456 key
.objectid
= parent_ref
->dir
;
3457 key
.type
= BTRFS_DIR_ITEM_KEY
;
3458 key
.offset
= btrfs_name_hash(parent_ref
->name
, parent_ref
->name_len
);
3460 ret
= btrfs_search_slot(NULL
, sctx
->parent_root
, &key
, path
, 0, 0);
3463 } else if (ret
> 0) {
3468 di
= btrfs_match_dir_item_name(fs_info
, path
, parent_ref
->name
,
3469 parent_ref
->name_len
);
3475 * di_key.objectid has the number of the inode that has a dentry in the
3476 * parent directory with the same name that sctx->cur_ino is being
3477 * renamed to. We need to check if that inode is in the send root as
3478 * well and if it is currently marked as an inode with a pending rename,
3479 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3480 * that it happens after that other inode is renamed.
3482 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &di_key
);
3483 if (di_key
.type
!= BTRFS_INODE_ITEM_KEY
) {
3488 ret
= get_inode_info(sctx
->parent_root
, di_key
.objectid
, NULL
,
3489 &left_gen
, NULL
, NULL
, NULL
, NULL
);
3492 ret
= get_inode_info(sctx
->send_root
, di_key
.objectid
, NULL
,
3493 &right_gen
, NULL
, NULL
, NULL
, NULL
);
3500 /* Different inode, no need to delay the rename of sctx->cur_ino */
3501 if (right_gen
!= left_gen
) {
3506 wdm
= get_waiting_dir_move(sctx
, di_key
.objectid
);
3507 if (wdm
&& !wdm
->orphanized
) {
3508 ret
= add_pending_dir_move(sctx
,
3510 sctx
->cur_inode_gen
,
3513 &sctx
->deleted_refs
,
3519 btrfs_free_path(path
);
3524 * Check if inode ino2, or any of its ancestors, is inode ino1.
3525 * Return 1 if true, 0 if false and < 0 on error.
3527 static int check_ino_in_path(struct btrfs_root
*root
,
3532 struct fs_path
*fs_path
)
3537 return ino1_gen
== ino2_gen
;
3539 while (ino
> BTRFS_FIRST_FREE_OBJECTID
) {
3544 fs_path_reset(fs_path
);
3545 ret
= get_first_ref(root
, ino
, &parent
, &parent_gen
, fs_path
);
3549 return parent_gen
== ino1_gen
;
3556 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3557 * possible path (in case ino2 is not a directory and has multiple hard links).
3558 * Return 1 if true, 0 if false and < 0 on error.
3560 static int is_ancestor(struct btrfs_root
*root
,
3564 struct fs_path
*fs_path
)
3566 bool free_fs_path
= false;
3568 struct btrfs_path
*path
= NULL
;
3569 struct btrfs_key key
;
3572 fs_path
= fs_path_alloc();
3575 free_fs_path
= true;
3578 path
= alloc_path_for_send();
3584 key
.objectid
= ino2
;
3585 key
.type
= BTRFS_INODE_REF_KEY
;
3588 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3593 struct extent_buffer
*leaf
= path
->nodes
[0];
3594 int slot
= path
->slots
[0];
3598 if (slot
>= btrfs_header_nritems(leaf
)) {
3599 ret
= btrfs_next_leaf(root
, path
);
3607 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
3608 if (key
.objectid
!= ino2
)
3610 if (key
.type
!= BTRFS_INODE_REF_KEY
&&
3611 key
.type
!= BTRFS_INODE_EXTREF_KEY
)
3614 item_size
= btrfs_item_size_nr(leaf
, slot
);
3615 while (cur_offset
< item_size
) {
3619 if (key
.type
== BTRFS_INODE_EXTREF_KEY
) {
3621 struct btrfs_inode_extref
*extref
;
3623 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
3624 extref
= (struct btrfs_inode_extref
*)
3626 parent
= btrfs_inode_extref_parent(leaf
,
3628 cur_offset
+= sizeof(*extref
);
3629 cur_offset
+= btrfs_inode_extref_name_len(leaf
,
3632 parent
= key
.offset
;
3633 cur_offset
= item_size
;
3636 ret
= get_inode_info(root
, parent
, NULL
, &parent_gen
,
3637 NULL
, NULL
, NULL
, NULL
);
3640 ret
= check_ino_in_path(root
, ino1
, ino1_gen
,
3641 parent
, parent_gen
, fs_path
);
3649 btrfs_free_path(path
);
3651 fs_path_free(fs_path
);
3655 static int wait_for_parent_move(struct send_ctx
*sctx
,
3656 struct recorded_ref
*parent_ref
,
3657 const bool is_orphan
)
3660 u64 ino
= parent_ref
->dir
;
3661 u64 ino_gen
= parent_ref
->dir_gen
;
3662 u64 parent_ino_before
, parent_ino_after
;
3663 struct fs_path
*path_before
= NULL
;
3664 struct fs_path
*path_after
= NULL
;
3667 path_after
= fs_path_alloc();
3668 path_before
= fs_path_alloc();
3669 if (!path_after
|| !path_before
) {
3675 * Our current directory inode may not yet be renamed/moved because some
3676 * ancestor (immediate or not) has to be renamed/moved first. So find if
3677 * such ancestor exists and make sure our own rename/move happens after
3678 * that ancestor is processed to avoid path build infinite loops (done
3679 * at get_cur_path()).
3681 while (ino
> BTRFS_FIRST_FREE_OBJECTID
) {
3682 u64 parent_ino_after_gen
;
3684 if (is_waiting_for_move(sctx
, ino
)) {
3686 * If the current inode is an ancestor of ino in the
3687 * parent root, we need to delay the rename of the
3688 * current inode, otherwise don't delayed the rename
3689 * because we can end up with a circular dependency
3690 * of renames, resulting in some directories never
3691 * getting the respective rename operations issued in
3692 * the send stream or getting into infinite path build
3695 ret
= is_ancestor(sctx
->parent_root
,
3696 sctx
->cur_ino
, sctx
->cur_inode_gen
,
3702 fs_path_reset(path_before
);
3703 fs_path_reset(path_after
);
3705 ret
= get_first_ref(sctx
->send_root
, ino
, &parent_ino_after
,
3706 &parent_ino_after_gen
, path_after
);
3709 ret
= get_first_ref(sctx
->parent_root
, ino
, &parent_ino_before
,
3711 if (ret
< 0 && ret
!= -ENOENT
) {
3713 } else if (ret
== -ENOENT
) {
3718 len1
= fs_path_len(path_before
);
3719 len2
= fs_path_len(path_after
);
3720 if (ino
> sctx
->cur_ino
&&
3721 (parent_ino_before
!= parent_ino_after
|| len1
!= len2
||
3722 memcmp(path_before
->start
, path_after
->start
, len1
))) {
3725 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
,
3726 &parent_ino_gen
, NULL
, NULL
, NULL
,
3730 if (ino_gen
== parent_ino_gen
) {
3735 ino
= parent_ino_after
;
3736 ino_gen
= parent_ino_after_gen
;
3740 fs_path_free(path_before
);
3741 fs_path_free(path_after
);
3744 ret
= add_pending_dir_move(sctx
,
3746 sctx
->cur_inode_gen
,
3749 &sctx
->deleted_refs
,
3758 static int update_ref_path(struct send_ctx
*sctx
, struct recorded_ref
*ref
)
3761 struct fs_path
*new_path
;
3764 * Our reference's name member points to its full_path member string, so
3765 * we use here a new path.
3767 new_path
= fs_path_alloc();
3771 ret
= get_cur_path(sctx
, ref
->dir
, ref
->dir_gen
, new_path
);
3773 fs_path_free(new_path
);
3776 ret
= fs_path_add(new_path
, ref
->name
, ref
->name_len
);
3778 fs_path_free(new_path
);
3782 fs_path_free(ref
->full_path
);
3783 set_ref_path(ref
, new_path
);
3789 * This does all the move/link/unlink/rmdir magic.
3791 static int process_recorded_refs(struct send_ctx
*sctx
, int *pending_move
)
3793 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
3795 struct recorded_ref
*cur
;
3796 struct recorded_ref
*cur2
;
3797 struct list_head check_dirs
;
3798 struct fs_path
*valid_path
= NULL
;
3802 int did_overwrite
= 0;
3804 u64 last_dir_ino_rm
= 0;
3805 bool can_rename
= true;
3806 bool orphanized_dir
= false;
3807 bool orphanized_ancestor
= false;
3809 btrfs_debug(fs_info
, "process_recorded_refs %llu", sctx
->cur_ino
);
3812 * This should never happen as the root dir always has the same ref
3813 * which is always '..'
3815 BUG_ON(sctx
->cur_ino
<= BTRFS_FIRST_FREE_OBJECTID
);
3816 INIT_LIST_HEAD(&check_dirs
);
3818 valid_path
= fs_path_alloc();
3825 * First, check if the first ref of the current inode was overwritten
3826 * before. If yes, we know that the current inode was already orphanized
3827 * and thus use the orphan name. If not, we can use get_cur_path to
3828 * get the path of the first ref as it would like while receiving at
3829 * this point in time.
3830 * New inodes are always orphan at the beginning, so force to use the
3831 * orphan name in this case.
3832 * The first ref is stored in valid_path and will be updated if it
3833 * gets moved around.
3835 if (!sctx
->cur_inode_new
) {
3836 ret
= did_overwrite_first_ref(sctx
, sctx
->cur_ino
,
3837 sctx
->cur_inode_gen
);
3843 if (sctx
->cur_inode_new
|| did_overwrite
) {
3844 ret
= gen_unique_name(sctx
, sctx
->cur_ino
,
3845 sctx
->cur_inode_gen
, valid_path
);
3850 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3856 list_for_each_entry(cur
, &sctx
->new_refs
, list
) {
3858 * We may have refs where the parent directory does not exist
3859 * yet. This happens if the parent directories inum is higher
3860 * the the current inum. To handle this case, we create the
3861 * parent directory out of order. But we need to check if this
3862 * did already happen before due to other refs in the same dir.
3864 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3867 if (ret
== inode_state_will_create
) {
3870 * First check if any of the current inodes refs did
3871 * already create the dir.
3873 list_for_each_entry(cur2
, &sctx
->new_refs
, list
) {
3876 if (cur2
->dir
== cur
->dir
) {
3883 * If that did not happen, check if a previous inode
3884 * did already create the dir.
3887 ret
= did_create_dir(sctx
, cur
->dir
);
3891 ret
= send_create_inode(sctx
, cur
->dir
);
3898 * Check if this new ref would overwrite the first ref of
3899 * another unprocessed inode. If yes, orphanize the
3900 * overwritten inode. If we find an overwritten ref that is
3901 * not the first ref, simply unlink it.
3903 ret
= will_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3904 cur
->name
, cur
->name_len
,
3905 &ow_inode
, &ow_gen
, &ow_mode
);
3909 ret
= is_first_ref(sctx
->parent_root
,
3910 ow_inode
, cur
->dir
, cur
->name
,
3915 struct name_cache_entry
*nce
;
3916 struct waiting_dir_move
*wdm
;
3918 ret
= orphanize_inode(sctx
, ow_inode
, ow_gen
,
3922 if (S_ISDIR(ow_mode
))
3923 orphanized_dir
= true;
3926 * If ow_inode has its rename operation delayed
3927 * make sure that its orphanized name is used in
3928 * the source path when performing its rename
3931 if (is_waiting_for_move(sctx
, ow_inode
)) {
3932 wdm
= get_waiting_dir_move(sctx
,
3935 wdm
->orphanized
= true;
3939 * Make sure we clear our orphanized inode's
3940 * name from the name cache. This is because the
3941 * inode ow_inode might be an ancestor of some
3942 * other inode that will be orphanized as well
3943 * later and has an inode number greater than
3944 * sctx->send_progress. We need to prevent
3945 * future name lookups from using the old name
3946 * and get instead the orphan name.
3948 nce
= name_cache_search(sctx
, ow_inode
, ow_gen
);
3950 name_cache_delete(sctx
, nce
);
3955 * ow_inode might currently be an ancestor of
3956 * cur_ino, therefore compute valid_path (the
3957 * current path of cur_ino) again because it
3958 * might contain the pre-orphanization name of
3959 * ow_inode, which is no longer valid.
3961 ret
= is_ancestor(sctx
->parent_root
,
3963 sctx
->cur_ino
, NULL
);
3965 orphanized_ancestor
= true;
3966 fs_path_reset(valid_path
);
3967 ret
= get_cur_path(sctx
, sctx
->cur_ino
,
3968 sctx
->cur_inode_gen
,
3974 ret
= send_unlink(sctx
, cur
->full_path
);
3980 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->parent_root
) {
3981 ret
= wait_for_dest_dir_move(sctx
, cur
, is_orphan
);
3990 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->parent_root
&&
3992 ret
= wait_for_parent_move(sctx
, cur
, is_orphan
);
4002 * link/move the ref to the new place. If we have an orphan
4003 * inode, move it and update valid_path. If not, link or move
4004 * it depending on the inode mode.
4006 if (is_orphan
&& can_rename
) {
4007 ret
= send_rename(sctx
, valid_path
, cur
->full_path
);
4011 ret
= fs_path_copy(valid_path
, cur
->full_path
);
4014 } else if (can_rename
) {
4015 if (S_ISDIR(sctx
->cur_inode_mode
)) {
4017 * Dirs can't be linked, so move it. For moved
4018 * dirs, we always have one new and one deleted
4019 * ref. The deleted ref is ignored later.
4021 ret
= send_rename(sctx
, valid_path
,
4024 ret
= fs_path_copy(valid_path
,
4030 * We might have previously orphanized an inode
4031 * which is an ancestor of our current inode,
4032 * so our reference's full path, which was
4033 * computed before any such orphanizations, must
4036 if (orphanized_dir
) {
4037 ret
= update_ref_path(sctx
, cur
);
4041 ret
= send_link(sctx
, cur
->full_path
,
4047 ret
= dup_ref(cur
, &check_dirs
);
4052 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->cur_inode_deleted
) {
4054 * Check if we can already rmdir the directory. If not,
4055 * orphanize it. For every dir item inside that gets deleted
4056 * later, we do this check again and rmdir it then if possible.
4057 * See the use of check_dirs for more details.
4059 ret
= can_rmdir(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
4064 ret
= send_rmdir(sctx
, valid_path
);
4067 } else if (!is_orphan
) {
4068 ret
= orphanize_inode(sctx
, sctx
->cur_ino
,
4069 sctx
->cur_inode_gen
, valid_path
);
4075 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
4076 ret
= dup_ref(cur
, &check_dirs
);
4080 } else if (S_ISDIR(sctx
->cur_inode_mode
) &&
4081 !list_empty(&sctx
->deleted_refs
)) {
4083 * We have a moved dir. Add the old parent to check_dirs
4085 cur
= list_entry(sctx
->deleted_refs
.next
, struct recorded_ref
,
4087 ret
= dup_ref(cur
, &check_dirs
);
4090 } else if (!S_ISDIR(sctx
->cur_inode_mode
)) {
4092 * We have a non dir inode. Go through all deleted refs and
4093 * unlink them if they were not already overwritten by other
4096 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
4097 ret
= did_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
4098 sctx
->cur_ino
, sctx
->cur_inode_gen
,
4099 cur
->name
, cur
->name_len
);
4104 * If we orphanized any ancestor before, we need
4105 * to recompute the full path for deleted names,
4106 * since any such path was computed before we
4107 * processed any references and orphanized any
4110 if (orphanized_ancestor
) {
4111 ret
= update_ref_path(sctx
, cur
);
4115 ret
= send_unlink(sctx
, cur
->full_path
);
4119 ret
= dup_ref(cur
, &check_dirs
);
4124 * If the inode is still orphan, unlink the orphan. This may
4125 * happen when a previous inode did overwrite the first ref
4126 * of this inode and no new refs were added for the current
4127 * inode. Unlinking does not mean that the inode is deleted in
4128 * all cases. There may still be links to this inode in other
4132 ret
= send_unlink(sctx
, valid_path
);
4139 * We did collect all parent dirs where cur_inode was once located. We
4140 * now go through all these dirs and check if they are pending for
4141 * deletion and if it's finally possible to perform the rmdir now.
4142 * We also update the inode stats of the parent dirs here.
4144 list_for_each_entry(cur
, &check_dirs
, list
) {
4146 * In case we had refs into dirs that were not processed yet,
4147 * we don't need to do the utime and rmdir logic for these dirs.
4148 * The dir will be processed later.
4150 if (cur
->dir
> sctx
->cur_ino
)
4153 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
4157 if (ret
== inode_state_did_create
||
4158 ret
== inode_state_no_change
) {
4159 /* TODO delayed utimes */
4160 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
4163 } else if (ret
== inode_state_did_delete
&&
4164 cur
->dir
!= last_dir_ino_rm
) {
4165 ret
= can_rmdir(sctx
, cur
->dir
, cur
->dir_gen
,
4170 ret
= get_cur_path(sctx
, cur
->dir
,
4171 cur
->dir_gen
, valid_path
);
4174 ret
= send_rmdir(sctx
, valid_path
);
4177 last_dir_ino_rm
= cur
->dir
;
4185 __free_recorded_refs(&check_dirs
);
4186 free_recorded_refs(sctx
);
4187 fs_path_free(valid_path
);
4191 static int record_ref(struct btrfs_root
*root
, u64 dir
, struct fs_path
*name
,
4192 void *ctx
, struct list_head
*refs
)
4195 struct send_ctx
*sctx
= ctx
;
4199 p
= fs_path_alloc();
4203 ret
= get_inode_info(root
, dir
, NULL
, &gen
, NULL
, NULL
,
4208 ret
= get_cur_path(sctx
, dir
, gen
, p
);
4211 ret
= fs_path_add_path(p
, name
);
4215 ret
= __record_ref(refs
, dir
, gen
, p
);
4223 static int __record_new_ref(int num
, u64 dir
, int index
,
4224 struct fs_path
*name
,
4227 struct send_ctx
*sctx
= ctx
;
4228 return record_ref(sctx
->send_root
, dir
, name
, ctx
, &sctx
->new_refs
);
4232 static int __record_deleted_ref(int num
, u64 dir
, int index
,
4233 struct fs_path
*name
,
4236 struct send_ctx
*sctx
= ctx
;
4237 return record_ref(sctx
->parent_root
, dir
, name
, ctx
,
4238 &sctx
->deleted_refs
);
4241 static int record_new_ref(struct send_ctx
*sctx
)
4245 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
4246 sctx
->cmp_key
, 0, __record_new_ref
, sctx
);
4255 static int record_deleted_ref(struct send_ctx
*sctx
)
4259 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
4260 sctx
->cmp_key
, 0, __record_deleted_ref
, sctx
);
4269 struct find_ref_ctx
{
4272 struct btrfs_root
*root
;
4273 struct fs_path
*name
;
4277 static int __find_iref(int num
, u64 dir
, int index
,
4278 struct fs_path
*name
,
4281 struct find_ref_ctx
*ctx
= ctx_
;
4285 if (dir
== ctx
->dir
&& fs_path_len(name
) == fs_path_len(ctx
->name
) &&
4286 strncmp(name
->start
, ctx
->name
->start
, fs_path_len(name
)) == 0) {
4288 * To avoid doing extra lookups we'll only do this if everything
4291 ret
= get_inode_info(ctx
->root
, dir
, NULL
, &dir_gen
, NULL
,
4295 if (dir_gen
!= ctx
->dir_gen
)
4297 ctx
->found_idx
= num
;
4303 static int find_iref(struct btrfs_root
*root
,
4304 struct btrfs_path
*path
,
4305 struct btrfs_key
*key
,
4306 u64 dir
, u64 dir_gen
, struct fs_path
*name
)
4309 struct find_ref_ctx ctx
;
4313 ctx
.dir_gen
= dir_gen
;
4317 ret
= iterate_inode_ref(root
, path
, key
, 0, __find_iref
, &ctx
);
4321 if (ctx
.found_idx
== -1)
4324 return ctx
.found_idx
;
4327 static int __record_changed_new_ref(int num
, u64 dir
, int index
,
4328 struct fs_path
*name
,
4333 struct send_ctx
*sctx
= ctx
;
4335 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &dir_gen
, NULL
,
4340 ret
= find_iref(sctx
->parent_root
, sctx
->right_path
,
4341 sctx
->cmp_key
, dir
, dir_gen
, name
);
4343 ret
= __record_new_ref(num
, dir
, index
, name
, sctx
);
4350 static int __record_changed_deleted_ref(int num
, u64 dir
, int index
,
4351 struct fs_path
*name
,
4356 struct send_ctx
*sctx
= ctx
;
4358 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &dir_gen
, NULL
,
4363 ret
= find_iref(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
4364 dir
, dir_gen
, name
);
4366 ret
= __record_deleted_ref(num
, dir
, index
, name
, sctx
);
4373 static int record_changed_ref(struct send_ctx
*sctx
)
4377 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
4378 sctx
->cmp_key
, 0, __record_changed_new_ref
, sctx
);
4381 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
4382 sctx
->cmp_key
, 0, __record_changed_deleted_ref
, sctx
);
4392 * Record and process all refs at once. Needed when an inode changes the
4393 * generation number, which means that it was deleted and recreated.
4395 static int process_all_refs(struct send_ctx
*sctx
,
4396 enum btrfs_compare_tree_result cmd
)
4399 struct btrfs_root
*root
;
4400 struct btrfs_path
*path
;
4401 struct btrfs_key key
;
4402 struct btrfs_key found_key
;
4403 struct extent_buffer
*eb
;
4405 iterate_inode_ref_t cb
;
4406 int pending_move
= 0;
4408 path
= alloc_path_for_send();
4412 if (cmd
== BTRFS_COMPARE_TREE_NEW
) {
4413 root
= sctx
->send_root
;
4414 cb
= __record_new_ref
;
4415 } else if (cmd
== BTRFS_COMPARE_TREE_DELETED
) {
4416 root
= sctx
->parent_root
;
4417 cb
= __record_deleted_ref
;
4419 btrfs_err(sctx
->send_root
->fs_info
,
4420 "Wrong command %d in process_all_refs", cmd
);
4425 key
.objectid
= sctx
->cmp_key
->objectid
;
4426 key
.type
= BTRFS_INODE_REF_KEY
;
4428 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4433 eb
= path
->nodes
[0];
4434 slot
= path
->slots
[0];
4435 if (slot
>= btrfs_header_nritems(eb
)) {
4436 ret
= btrfs_next_leaf(root
, path
);
4444 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4446 if (found_key
.objectid
!= key
.objectid
||
4447 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
4448 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
))
4451 ret
= iterate_inode_ref(root
, path
, &found_key
, 0, cb
, sctx
);
4457 btrfs_release_path(path
);
4460 * We don't actually care about pending_move as we are simply
4461 * re-creating this inode and will be rename'ing it into place once we
4462 * rename the parent directory.
4464 ret
= process_recorded_refs(sctx
, &pending_move
);
4466 btrfs_free_path(path
);
4470 static int send_set_xattr(struct send_ctx
*sctx
,
4471 struct fs_path
*path
,
4472 const char *name
, int name_len
,
4473 const char *data
, int data_len
)
4477 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SET_XATTR
);
4481 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
4482 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
4483 TLV_PUT(sctx
, BTRFS_SEND_A_XATTR_DATA
, data
, data_len
);
4485 ret
= send_cmd(sctx
);
4492 static int send_remove_xattr(struct send_ctx
*sctx
,
4493 struct fs_path
*path
,
4494 const char *name
, int name_len
)
4498 ret
= begin_cmd(sctx
, BTRFS_SEND_C_REMOVE_XATTR
);
4502 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
4503 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
4505 ret
= send_cmd(sctx
);
4512 static int __process_new_xattr(int num
, struct btrfs_key
*di_key
,
4513 const char *name
, int name_len
,
4514 const char *data
, int data_len
,
4518 struct send_ctx
*sctx
= ctx
;
4520 struct posix_acl_xattr_header dummy_acl
;
4522 p
= fs_path_alloc();
4527 * This hack is needed because empty acls are stored as zero byte
4528 * data in xattrs. Problem with that is, that receiving these zero byte
4529 * acls will fail later. To fix this, we send a dummy acl list that
4530 * only contains the version number and no entries.
4532 if (!strncmp(name
, XATTR_NAME_POSIX_ACL_ACCESS
, name_len
) ||
4533 !strncmp(name
, XATTR_NAME_POSIX_ACL_DEFAULT
, name_len
)) {
4534 if (data_len
== 0) {
4535 dummy_acl
.a_version
=
4536 cpu_to_le32(POSIX_ACL_XATTR_VERSION
);
4537 data
= (char *)&dummy_acl
;
4538 data_len
= sizeof(dummy_acl
);
4542 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4546 ret
= send_set_xattr(sctx
, p
, name
, name_len
, data
, data_len
);
4553 static int __process_deleted_xattr(int num
, struct btrfs_key
*di_key
,
4554 const char *name
, int name_len
,
4555 const char *data
, int data_len
,
4559 struct send_ctx
*sctx
= ctx
;
4562 p
= fs_path_alloc();
4566 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4570 ret
= send_remove_xattr(sctx
, p
, name
, name_len
);
4577 static int process_new_xattr(struct send_ctx
*sctx
)
4581 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4582 __process_new_xattr
, sctx
);
4587 static int process_deleted_xattr(struct send_ctx
*sctx
)
4589 return iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4590 __process_deleted_xattr
, sctx
);
4593 struct find_xattr_ctx
{
4601 static int __find_xattr(int num
, struct btrfs_key
*di_key
,
4602 const char *name
, int name_len
,
4603 const char *data
, int data_len
,
4604 u8 type
, void *vctx
)
4606 struct find_xattr_ctx
*ctx
= vctx
;
4608 if (name_len
== ctx
->name_len
&&
4609 strncmp(name
, ctx
->name
, name_len
) == 0) {
4610 ctx
->found_idx
= num
;
4611 ctx
->found_data_len
= data_len
;
4612 ctx
->found_data
= kmemdup(data
, data_len
, GFP_KERNEL
);
4613 if (!ctx
->found_data
)
4620 static int find_xattr(struct btrfs_root
*root
,
4621 struct btrfs_path
*path
,
4622 struct btrfs_key
*key
,
4623 const char *name
, int name_len
,
4624 char **data
, int *data_len
)
4627 struct find_xattr_ctx ctx
;
4630 ctx
.name_len
= name_len
;
4632 ctx
.found_data
= NULL
;
4633 ctx
.found_data_len
= 0;
4635 ret
= iterate_dir_item(root
, path
, __find_xattr
, &ctx
);
4639 if (ctx
.found_idx
== -1)
4642 *data
= ctx
.found_data
;
4643 *data_len
= ctx
.found_data_len
;
4645 kfree(ctx
.found_data
);
4647 return ctx
.found_idx
;
4651 static int __process_changed_new_xattr(int num
, struct btrfs_key
*di_key
,
4652 const char *name
, int name_len
,
4653 const char *data
, int data_len
,
4657 struct send_ctx
*sctx
= ctx
;
4658 char *found_data
= NULL
;
4659 int found_data_len
= 0;
4661 ret
= find_xattr(sctx
->parent_root
, sctx
->right_path
,
4662 sctx
->cmp_key
, name
, name_len
, &found_data
,
4664 if (ret
== -ENOENT
) {
4665 ret
= __process_new_xattr(num
, di_key
, name
, name_len
, data
,
4666 data_len
, type
, ctx
);
4667 } else if (ret
>= 0) {
4668 if (data_len
!= found_data_len
||
4669 memcmp(data
, found_data
, data_len
)) {
4670 ret
= __process_new_xattr(num
, di_key
, name
, name_len
,
4671 data
, data_len
, type
, ctx
);
4681 static int __process_changed_deleted_xattr(int num
, struct btrfs_key
*di_key
,
4682 const char *name
, int name_len
,
4683 const char *data
, int data_len
,
4687 struct send_ctx
*sctx
= ctx
;
4689 ret
= find_xattr(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
4690 name
, name_len
, NULL
, NULL
);
4692 ret
= __process_deleted_xattr(num
, di_key
, name
, name_len
, data
,
4693 data_len
, type
, ctx
);
4700 static int process_changed_xattr(struct send_ctx
*sctx
)
4704 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4705 __process_changed_new_xattr
, sctx
);
4708 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4709 __process_changed_deleted_xattr
, sctx
);
4715 static int process_all_new_xattrs(struct send_ctx
*sctx
)
4718 struct btrfs_root
*root
;
4719 struct btrfs_path
*path
;
4720 struct btrfs_key key
;
4721 struct btrfs_key found_key
;
4722 struct extent_buffer
*eb
;
4725 path
= alloc_path_for_send();
4729 root
= sctx
->send_root
;
4731 key
.objectid
= sctx
->cmp_key
->objectid
;
4732 key
.type
= BTRFS_XATTR_ITEM_KEY
;
4734 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4739 eb
= path
->nodes
[0];
4740 slot
= path
->slots
[0];
4741 if (slot
>= btrfs_header_nritems(eb
)) {
4742 ret
= btrfs_next_leaf(root
, path
);
4745 } else if (ret
> 0) {
4752 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4753 if (found_key
.objectid
!= key
.objectid
||
4754 found_key
.type
!= key
.type
) {
4759 ret
= iterate_dir_item(root
, path
, __process_new_xattr
, sctx
);
4767 btrfs_free_path(path
);
4771 static ssize_t
fill_read_buf(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4773 struct btrfs_root
*root
= sctx
->send_root
;
4774 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4775 struct inode
*inode
;
4778 struct btrfs_key key
;
4779 pgoff_t index
= offset
>> PAGE_SHIFT
;
4781 unsigned pg_offset
= offset
& ~PAGE_MASK
;
4784 key
.objectid
= sctx
->cur_ino
;
4785 key
.type
= BTRFS_INODE_ITEM_KEY
;
4788 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
4790 return PTR_ERR(inode
);
4792 if (offset
+ len
> i_size_read(inode
)) {
4793 if (offset
> i_size_read(inode
))
4796 len
= offset
- i_size_read(inode
);
4801 last_index
= (offset
+ len
- 1) >> PAGE_SHIFT
;
4803 /* initial readahead */
4804 memset(&sctx
->ra
, 0, sizeof(struct file_ra_state
));
4805 file_ra_state_init(&sctx
->ra
, inode
->i_mapping
);
4807 while (index
<= last_index
) {
4808 unsigned cur_len
= min_t(unsigned, len
,
4809 PAGE_SIZE
- pg_offset
);
4811 page
= find_lock_page(inode
->i_mapping
, index
);
4813 page_cache_sync_readahead(inode
->i_mapping
, &sctx
->ra
,
4814 NULL
, index
, last_index
+ 1 - index
);
4816 page
= find_or_create_page(inode
->i_mapping
, index
,
4824 if (PageReadahead(page
)) {
4825 page_cache_async_readahead(inode
->i_mapping
, &sctx
->ra
,
4826 NULL
, page
, index
, last_index
+ 1 - index
);
4829 if (!PageUptodate(page
)) {
4830 btrfs_readpage(NULL
, page
);
4832 if (!PageUptodate(page
)) {
4841 memcpy(sctx
->read_buf
+ ret
, addr
+ pg_offset
, cur_len
);
4856 * Read some bytes from the current inode/file and send a write command to
4859 static int send_write(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4861 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
4864 ssize_t num_read
= 0;
4866 p
= fs_path_alloc();
4870 btrfs_debug(fs_info
, "send_write offset=%llu, len=%d", offset
, len
);
4872 num_read
= fill_read_buf(sctx
, offset
, len
);
4873 if (num_read
<= 0) {
4879 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4883 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4887 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4888 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4889 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, num_read
);
4891 ret
= send_cmd(sctx
);
4902 * Send a clone command to user space.
4904 static int send_clone(struct send_ctx
*sctx
,
4905 u64 offset
, u32 len
,
4906 struct clone_root
*clone_root
)
4912 btrfs_debug(sctx
->send_root
->fs_info
,
4913 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4914 offset
, len
, clone_root
->root
->objectid
, clone_root
->ino
,
4915 clone_root
->offset
);
4917 p
= fs_path_alloc();
4921 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CLONE
);
4925 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4929 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4930 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_LEN
, len
);
4931 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4933 if (clone_root
->root
== sctx
->send_root
) {
4934 ret
= get_inode_info(sctx
->send_root
, clone_root
->ino
, NULL
,
4935 &gen
, NULL
, NULL
, NULL
, NULL
);
4938 ret
= get_cur_path(sctx
, clone_root
->ino
, gen
, p
);
4940 ret
= get_inode_path(clone_root
->root
, clone_root
->ino
, p
);
4946 * If the parent we're using has a received_uuid set then use that as
4947 * our clone source as that is what we will look for when doing a
4950 * This covers the case that we create a snapshot off of a received
4951 * subvolume and then use that as the parent and try to receive on a
4954 if (!btrfs_is_empty_uuid(clone_root
->root
->root_item
.received_uuid
))
4955 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4956 clone_root
->root
->root_item
.received_uuid
);
4958 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4959 clone_root
->root
->root_item
.uuid
);
4960 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
4961 le64_to_cpu(clone_root
->root
->root_item
.ctransid
));
4962 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_CLONE_PATH
, p
);
4963 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_OFFSET
,
4964 clone_root
->offset
);
4966 ret
= send_cmd(sctx
);
4975 * Send an update extent command to user space.
4977 static int send_update_extent(struct send_ctx
*sctx
,
4978 u64 offset
, u32 len
)
4983 p
= fs_path_alloc();
4987 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UPDATE_EXTENT
);
4991 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4995 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4996 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4997 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, len
);
4999 ret
= send_cmd(sctx
);
5007 static int send_hole(struct send_ctx
*sctx
, u64 end
)
5009 struct fs_path
*p
= NULL
;
5010 u64 offset
= sctx
->cur_inode_last_extent
;
5014 p
= fs_path_alloc();
5017 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
5019 goto tlv_put_failure
;
5020 memset(sctx
->read_buf
, 0, BTRFS_SEND_READ_SIZE
);
5021 while (offset
< end
) {
5022 len
= min_t(u64
, end
- offset
, BTRFS_SEND_READ_SIZE
);
5024 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
5027 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
5028 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
5029 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, len
);
5030 ret
= send_cmd(sctx
);
5040 static int send_extent_data(struct send_ctx
*sctx
,
5046 if (sctx
->flags
& BTRFS_SEND_FLAG_NO_FILE_DATA
)
5047 return send_update_extent(sctx
, offset
, len
);
5049 while (sent
< len
) {
5050 u64 size
= len
- sent
;
5053 if (size
> BTRFS_SEND_READ_SIZE
)
5054 size
= BTRFS_SEND_READ_SIZE
;
5055 ret
= send_write(sctx
, offset
+ sent
, size
);
5065 static int clone_range(struct send_ctx
*sctx
,
5066 struct clone_root
*clone_root
,
5067 const u64 disk_byte
,
5072 struct btrfs_path
*path
;
5073 struct btrfs_key key
;
5077 * Prevent cloning from a zero offset with a length matching the sector
5078 * size because in some scenarios this will make the receiver fail.
5080 * For example, if in the source filesystem the extent at offset 0
5081 * has a length of sectorsize and it was written using direct IO, then
5082 * it can never be an inline extent (even if compression is enabled).
5083 * Then this extent can be cloned in the original filesystem to a non
5084 * zero file offset, but it may not be possible to clone in the
5085 * destination filesystem because it can be inlined due to compression
5086 * on the destination filesystem (as the receiver's write operations are
5087 * always done using buffered IO). The same happens when the original
5088 * filesystem does not have compression enabled but the destination
5091 if (clone_root
->offset
== 0 &&
5092 len
== sctx
->send_root
->fs_info
->sectorsize
)
5093 return send_extent_data(sctx
, offset
, len
);
5095 path
= alloc_path_for_send();
5100 * We can't send a clone operation for the entire range if we find
5101 * extent items in the respective range in the source file that
5102 * refer to different extents or if we find holes.
5103 * So check for that and do a mix of clone and regular write/copy
5104 * operations if needed.
5108 * mkfs.btrfs -f /dev/sda
5109 * mount /dev/sda /mnt
5110 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5111 * cp --reflink=always /mnt/foo /mnt/bar
5112 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5113 * btrfs subvolume snapshot -r /mnt /mnt/snap
5115 * If when we send the snapshot and we are processing file bar (which
5116 * has a higher inode number than foo) we blindly send a clone operation
5117 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5118 * a file bar that matches the content of file foo - iow, doesn't match
5119 * the content from bar in the original filesystem.
5121 key
.objectid
= clone_root
->ino
;
5122 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5123 key
.offset
= clone_root
->offset
;
5124 ret
= btrfs_search_slot(NULL
, clone_root
->root
, &key
, path
, 0, 0);
5127 if (ret
> 0 && path
->slots
[0] > 0) {
5128 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0] - 1);
5129 if (key
.objectid
== clone_root
->ino
&&
5130 key
.type
== BTRFS_EXTENT_DATA_KEY
)
5135 struct extent_buffer
*leaf
= path
->nodes
[0];
5136 int slot
= path
->slots
[0];
5137 struct btrfs_file_extent_item
*ei
;
5142 if (slot
>= btrfs_header_nritems(leaf
)) {
5143 ret
= btrfs_next_leaf(clone_root
->root
, path
);
5151 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5154 * We might have an implicit trailing hole (NO_HOLES feature
5155 * enabled). We deal with it after leaving this loop.
5157 if (key
.objectid
!= clone_root
->ino
||
5158 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5161 ei
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5162 type
= btrfs_file_extent_type(leaf
, ei
);
5163 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5164 ext_len
= btrfs_file_extent_inline_len(leaf
, slot
, ei
);
5165 ext_len
= PAGE_ALIGN(ext_len
);
5167 ext_len
= btrfs_file_extent_num_bytes(leaf
, ei
);
5170 if (key
.offset
+ ext_len
<= clone_root
->offset
)
5173 if (key
.offset
> clone_root
->offset
) {
5174 /* Implicit hole, NO_HOLES feature enabled. */
5175 u64 hole_len
= key
.offset
- clone_root
->offset
;
5179 ret
= send_extent_data(sctx
, offset
, hole_len
);
5187 clone_root
->offset
+= hole_len
;
5188 data_offset
+= hole_len
;
5191 if (key
.offset
>= clone_root
->offset
+ len
)
5194 clone_len
= min_t(u64
, ext_len
, len
);
5196 if (btrfs_file_extent_disk_bytenr(leaf
, ei
) == disk_byte
&&
5197 btrfs_file_extent_offset(leaf
, ei
) == data_offset
)
5198 ret
= send_clone(sctx
, offset
, clone_len
, clone_root
);
5200 ret
= send_extent_data(sctx
, offset
, clone_len
);
5208 offset
+= clone_len
;
5209 clone_root
->offset
+= clone_len
;
5210 data_offset
+= clone_len
;
5216 ret
= send_extent_data(sctx
, offset
, len
);
5220 btrfs_free_path(path
);
5224 static int send_write_or_clone(struct send_ctx
*sctx
,
5225 struct btrfs_path
*path
,
5226 struct btrfs_key
*key
,
5227 struct clone_root
*clone_root
)
5230 struct btrfs_file_extent_item
*ei
;
5231 u64 offset
= key
->offset
;
5234 u64 bs
= sctx
->send_root
->fs_info
->sb
->s_blocksize
;
5236 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5237 struct btrfs_file_extent_item
);
5238 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
5239 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5240 len
= btrfs_file_extent_inline_len(path
->nodes
[0],
5241 path
->slots
[0], ei
);
5243 * it is possible the inline item won't cover the whole page,
5244 * but there may be items after this page. Make
5245 * sure to send the whole thing
5247 len
= PAGE_ALIGN(len
);
5249 len
= btrfs_file_extent_num_bytes(path
->nodes
[0], ei
);
5252 if (offset
+ len
> sctx
->cur_inode_size
)
5253 len
= sctx
->cur_inode_size
- offset
;
5259 if (clone_root
&& IS_ALIGNED(offset
+ len
, bs
)) {
5263 disk_byte
= btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
);
5264 data_offset
= btrfs_file_extent_offset(path
->nodes
[0], ei
);
5265 ret
= clone_range(sctx
, clone_root
, disk_byte
, data_offset
,
5268 ret
= send_extent_data(sctx
, offset
, len
);
5274 static int is_extent_unchanged(struct send_ctx
*sctx
,
5275 struct btrfs_path
*left_path
,
5276 struct btrfs_key
*ekey
)
5279 struct btrfs_key key
;
5280 struct btrfs_path
*path
= NULL
;
5281 struct extent_buffer
*eb
;
5283 struct btrfs_key found_key
;
5284 struct btrfs_file_extent_item
*ei
;
5289 u64 left_offset_fixed
;
5297 path
= alloc_path_for_send();
5301 eb
= left_path
->nodes
[0];
5302 slot
= left_path
->slots
[0];
5303 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
5304 left_type
= btrfs_file_extent_type(eb
, ei
);
5306 if (left_type
!= BTRFS_FILE_EXTENT_REG
) {
5310 left_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
5311 left_len
= btrfs_file_extent_num_bytes(eb
, ei
);
5312 left_offset
= btrfs_file_extent_offset(eb
, ei
);
5313 left_gen
= btrfs_file_extent_generation(eb
, ei
);
5316 * Following comments will refer to these graphics. L is the left
5317 * extents which we are checking at the moment. 1-8 are the right
5318 * extents that we iterate.
5321 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5324 * |--1--|-2b-|...(same as above)
5326 * Alternative situation. Happens on files where extents got split.
5328 * |-----------7-----------|-6-|
5330 * Alternative situation. Happens on files which got larger.
5333 * Nothing follows after 8.
5336 key
.objectid
= ekey
->objectid
;
5337 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5338 key
.offset
= ekey
->offset
;
5339 ret
= btrfs_search_slot_for_read(sctx
->parent_root
, &key
, path
, 0, 0);
5348 * Handle special case where the right side has no extents at all.
5350 eb
= path
->nodes
[0];
5351 slot
= path
->slots
[0];
5352 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5353 if (found_key
.objectid
!= key
.objectid
||
5354 found_key
.type
!= key
.type
) {
5355 /* If we're a hole then just pretend nothing changed */
5356 ret
= (left_disknr
) ? 0 : 1;
5361 * We're now on 2a, 2b or 7.
5364 while (key
.offset
< ekey
->offset
+ left_len
) {
5365 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
5366 right_type
= btrfs_file_extent_type(eb
, ei
);
5367 if (right_type
!= BTRFS_FILE_EXTENT_REG
&&
5368 right_type
!= BTRFS_FILE_EXTENT_INLINE
) {
5373 if (right_type
== BTRFS_FILE_EXTENT_INLINE
) {
5374 right_len
= btrfs_file_extent_inline_len(eb
, slot
, ei
);
5375 right_len
= PAGE_ALIGN(right_len
);
5377 right_len
= btrfs_file_extent_num_bytes(eb
, ei
);
5381 * Are we at extent 8? If yes, we know the extent is changed.
5382 * This may only happen on the first iteration.
5384 if (found_key
.offset
+ right_len
<= ekey
->offset
) {
5385 /* If we're a hole just pretend nothing changed */
5386 ret
= (left_disknr
) ? 0 : 1;
5391 * We just wanted to see if when we have an inline extent, what
5392 * follows it is a regular extent (wanted to check the above
5393 * condition for inline extents too). This should normally not
5394 * happen but it's possible for example when we have an inline
5395 * compressed extent representing data with a size matching
5396 * the page size (currently the same as sector size).
5398 if (right_type
== BTRFS_FILE_EXTENT_INLINE
) {
5403 right_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
5404 right_offset
= btrfs_file_extent_offset(eb
, ei
);
5405 right_gen
= btrfs_file_extent_generation(eb
, ei
);
5407 left_offset_fixed
= left_offset
;
5408 if (key
.offset
< ekey
->offset
) {
5409 /* Fix the right offset for 2a and 7. */
5410 right_offset
+= ekey
->offset
- key
.offset
;
5412 /* Fix the left offset for all behind 2a and 2b */
5413 left_offset_fixed
+= key
.offset
- ekey
->offset
;
5417 * Check if we have the same extent.
5419 if (left_disknr
!= right_disknr
||
5420 left_offset_fixed
!= right_offset
||
5421 left_gen
!= right_gen
) {
5427 * Go to the next extent.
5429 ret
= btrfs_next_item(sctx
->parent_root
, path
);
5433 eb
= path
->nodes
[0];
5434 slot
= path
->slots
[0];
5435 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5437 if (ret
|| found_key
.objectid
!= key
.objectid
||
5438 found_key
.type
!= key
.type
) {
5439 key
.offset
+= right_len
;
5442 if (found_key
.offset
!= key
.offset
+ right_len
) {
5450 * We're now behind the left extent (treat as unchanged) or at the end
5451 * of the right side (treat as changed).
5453 if (key
.offset
>= ekey
->offset
+ left_len
)
5460 btrfs_free_path(path
);
5464 static int get_last_extent(struct send_ctx
*sctx
, u64 offset
)
5466 struct btrfs_path
*path
;
5467 struct btrfs_root
*root
= sctx
->send_root
;
5468 struct btrfs_file_extent_item
*fi
;
5469 struct btrfs_key key
;
5474 path
= alloc_path_for_send();
5478 sctx
->cur_inode_last_extent
= 0;
5480 key
.objectid
= sctx
->cur_ino
;
5481 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5482 key
.offset
= offset
;
5483 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 0, 1);
5487 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
5488 if (key
.objectid
!= sctx
->cur_ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5491 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5492 struct btrfs_file_extent_item
);
5493 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
5494 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5495 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
5496 path
->slots
[0], fi
);
5497 extent_end
= ALIGN(key
.offset
+ size
,
5498 sctx
->send_root
->fs_info
->sectorsize
);
5500 extent_end
= key
.offset
+
5501 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
5503 sctx
->cur_inode_last_extent
= extent_end
;
5505 btrfs_free_path(path
);
5509 static int range_is_hole_in_parent(struct send_ctx
*sctx
,
5513 struct btrfs_path
*path
;
5514 struct btrfs_key key
;
5515 struct btrfs_root
*root
= sctx
->parent_root
;
5516 u64 search_start
= start
;
5519 path
= alloc_path_for_send();
5523 key
.objectid
= sctx
->cur_ino
;
5524 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5525 key
.offset
= search_start
;
5526 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5529 if (ret
> 0 && path
->slots
[0] > 0)
5532 while (search_start
< end
) {
5533 struct extent_buffer
*leaf
= path
->nodes
[0];
5534 int slot
= path
->slots
[0];
5535 struct btrfs_file_extent_item
*fi
;
5538 if (slot
>= btrfs_header_nritems(leaf
)) {
5539 ret
= btrfs_next_leaf(root
, path
);
5547 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5548 if (key
.objectid
< sctx
->cur_ino
||
5549 key
.type
< BTRFS_EXTENT_DATA_KEY
)
5551 if (key
.objectid
> sctx
->cur_ino
||
5552 key
.type
> BTRFS_EXTENT_DATA_KEY
||
5556 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5557 if (btrfs_file_extent_type(leaf
, fi
) ==
5558 BTRFS_FILE_EXTENT_INLINE
) {
5559 u64 size
= btrfs_file_extent_inline_len(leaf
, slot
, fi
);
5561 extent_end
= ALIGN(key
.offset
+ size
,
5562 root
->fs_info
->sectorsize
);
5564 extent_end
= key
.offset
+
5565 btrfs_file_extent_num_bytes(leaf
, fi
);
5567 if (extent_end
<= start
)
5569 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0) {
5570 search_start
= extent_end
;
5580 btrfs_free_path(path
);
5584 static int maybe_send_hole(struct send_ctx
*sctx
, struct btrfs_path
*path
,
5585 struct btrfs_key
*key
)
5587 struct btrfs_file_extent_item
*fi
;
5592 if (sctx
->cur_ino
!= key
->objectid
|| !need_send_hole(sctx
))
5595 if (sctx
->cur_inode_last_extent
== (u64
)-1) {
5596 ret
= get_last_extent(sctx
, key
->offset
- 1);
5601 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5602 struct btrfs_file_extent_item
);
5603 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
5604 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5605 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
5606 path
->slots
[0], fi
);
5607 extent_end
= ALIGN(key
->offset
+ size
,
5608 sctx
->send_root
->fs_info
->sectorsize
);
5610 extent_end
= key
->offset
+
5611 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
5614 if (path
->slots
[0] == 0 &&
5615 sctx
->cur_inode_last_extent
< key
->offset
) {
5617 * We might have skipped entire leafs that contained only
5618 * file extent items for our current inode. These leafs have
5619 * a generation number smaller (older) than the one in the
5620 * current leaf and the leaf our last extent came from, and
5621 * are located between these 2 leafs.
5623 ret
= get_last_extent(sctx
, key
->offset
- 1);
5628 if (sctx
->cur_inode_last_extent
< key
->offset
) {
5629 ret
= range_is_hole_in_parent(sctx
,
5630 sctx
->cur_inode_last_extent
,
5635 ret
= send_hole(sctx
, key
->offset
);
5639 sctx
->cur_inode_last_extent
= extent_end
;
5643 static int process_extent(struct send_ctx
*sctx
,
5644 struct btrfs_path
*path
,
5645 struct btrfs_key
*key
)
5647 struct clone_root
*found_clone
= NULL
;
5650 if (S_ISLNK(sctx
->cur_inode_mode
))
5653 if (sctx
->parent_root
&& !sctx
->cur_inode_new
) {
5654 ret
= is_extent_unchanged(sctx
, path
, key
);
5662 struct btrfs_file_extent_item
*ei
;
5665 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5666 struct btrfs_file_extent_item
);
5667 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
5668 if (type
== BTRFS_FILE_EXTENT_PREALLOC
||
5669 type
== BTRFS_FILE_EXTENT_REG
) {
5671 * The send spec does not have a prealloc command yet,
5672 * so just leave a hole for prealloc'ed extents until
5673 * we have enough commands queued up to justify rev'ing
5676 if (type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5681 /* Have a hole, just skip it. */
5682 if (btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
) == 0) {
5689 ret
= find_extent_clone(sctx
, path
, key
->objectid
, key
->offset
,
5690 sctx
->cur_inode_size
, &found_clone
);
5691 if (ret
!= -ENOENT
&& ret
< 0)
5694 ret
= send_write_or_clone(sctx
, path
, key
, found_clone
);
5698 ret
= maybe_send_hole(sctx
, path
, key
);
5703 static int process_all_extents(struct send_ctx
*sctx
)
5706 struct btrfs_root
*root
;
5707 struct btrfs_path
*path
;
5708 struct btrfs_key key
;
5709 struct btrfs_key found_key
;
5710 struct extent_buffer
*eb
;
5713 root
= sctx
->send_root
;
5714 path
= alloc_path_for_send();
5718 key
.objectid
= sctx
->cmp_key
->objectid
;
5719 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5721 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5726 eb
= path
->nodes
[0];
5727 slot
= path
->slots
[0];
5729 if (slot
>= btrfs_header_nritems(eb
)) {
5730 ret
= btrfs_next_leaf(root
, path
);
5733 } else if (ret
> 0) {
5740 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5742 if (found_key
.objectid
!= key
.objectid
||
5743 found_key
.type
!= key
.type
) {
5748 ret
= process_extent(sctx
, path
, &found_key
);
5756 btrfs_free_path(path
);
5760 static int process_recorded_refs_if_needed(struct send_ctx
*sctx
, int at_end
,
5762 int *refs_processed
)
5766 if (sctx
->cur_ino
== 0)
5768 if (!at_end
&& sctx
->cur_ino
== sctx
->cmp_key
->objectid
&&
5769 sctx
->cmp_key
->type
<= BTRFS_INODE_EXTREF_KEY
)
5771 if (list_empty(&sctx
->new_refs
) && list_empty(&sctx
->deleted_refs
))
5774 ret
= process_recorded_refs(sctx
, pending_move
);
5778 *refs_processed
= 1;
5783 static int finish_inode_if_needed(struct send_ctx
*sctx
, int at_end
)
5794 int pending_move
= 0;
5795 int refs_processed
= 0;
5797 ret
= process_recorded_refs_if_needed(sctx
, at_end
, &pending_move
,
5803 * We have processed the refs and thus need to advance send_progress.
5804 * Now, calls to get_cur_xxx will take the updated refs of the current
5805 * inode into account.
5807 * On the other hand, if our current inode is a directory and couldn't
5808 * be moved/renamed because its parent was renamed/moved too and it has
5809 * a higher inode number, we can only move/rename our current inode
5810 * after we moved/renamed its parent. Therefore in this case operate on
5811 * the old path (pre move/rename) of our current inode, and the
5812 * move/rename will be performed later.
5814 if (refs_processed
&& !pending_move
)
5815 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5817 if (sctx
->cur_ino
== 0 || sctx
->cur_inode_deleted
)
5819 if (!at_end
&& sctx
->cmp_key
->objectid
== sctx
->cur_ino
)
5822 ret
= get_inode_info(sctx
->send_root
, sctx
->cur_ino
, NULL
, NULL
,
5823 &left_mode
, &left_uid
, &left_gid
, NULL
);
5827 if (!sctx
->parent_root
|| sctx
->cur_inode_new
) {
5829 if (!S_ISLNK(sctx
->cur_inode_mode
))
5832 ret
= get_inode_info(sctx
->parent_root
, sctx
->cur_ino
,
5833 NULL
, NULL
, &right_mode
, &right_uid
,
5838 if (left_uid
!= right_uid
|| left_gid
!= right_gid
)
5840 if (!S_ISLNK(sctx
->cur_inode_mode
) && left_mode
!= right_mode
)
5844 if (S_ISREG(sctx
->cur_inode_mode
)) {
5845 if (need_send_hole(sctx
)) {
5846 if (sctx
->cur_inode_last_extent
== (u64
)-1 ||
5847 sctx
->cur_inode_last_extent
<
5848 sctx
->cur_inode_size
) {
5849 ret
= get_last_extent(sctx
, (u64
)-1);
5853 if (sctx
->cur_inode_last_extent
<
5854 sctx
->cur_inode_size
) {
5855 ret
= send_hole(sctx
, sctx
->cur_inode_size
);
5860 ret
= send_truncate(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5861 sctx
->cur_inode_size
);
5867 ret
= send_chown(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5868 left_uid
, left_gid
);
5873 ret
= send_chmod(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5880 * If other directory inodes depended on our current directory
5881 * inode's move/rename, now do their move/rename operations.
5883 if (!is_waiting_for_move(sctx
, sctx
->cur_ino
)) {
5884 ret
= apply_children_dir_moves(sctx
);
5888 * Need to send that every time, no matter if it actually
5889 * changed between the two trees as we have done changes to
5890 * the inode before. If our inode is a directory and it's
5891 * waiting to be moved/renamed, we will send its utimes when
5892 * it's moved/renamed, therefore we don't need to do it here.
5894 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5895 ret
= send_utimes(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
);
5904 static int changed_inode(struct send_ctx
*sctx
,
5905 enum btrfs_compare_tree_result result
)
5908 struct btrfs_key
*key
= sctx
->cmp_key
;
5909 struct btrfs_inode_item
*left_ii
= NULL
;
5910 struct btrfs_inode_item
*right_ii
= NULL
;
5914 sctx
->cur_ino
= key
->objectid
;
5915 sctx
->cur_inode_new_gen
= 0;
5916 sctx
->cur_inode_last_extent
= (u64
)-1;
5919 * Set send_progress to current inode. This will tell all get_cur_xxx
5920 * functions that the current inode's refs are not updated yet. Later,
5921 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5923 sctx
->send_progress
= sctx
->cur_ino
;
5925 if (result
== BTRFS_COMPARE_TREE_NEW
||
5926 result
== BTRFS_COMPARE_TREE_CHANGED
) {
5927 left_ii
= btrfs_item_ptr(sctx
->left_path
->nodes
[0],
5928 sctx
->left_path
->slots
[0],
5929 struct btrfs_inode_item
);
5930 left_gen
= btrfs_inode_generation(sctx
->left_path
->nodes
[0],
5933 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5934 sctx
->right_path
->slots
[0],
5935 struct btrfs_inode_item
);
5936 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5939 if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5940 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5941 sctx
->right_path
->slots
[0],
5942 struct btrfs_inode_item
);
5944 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5948 * The cur_ino = root dir case is special here. We can't treat
5949 * the inode as deleted+reused because it would generate a
5950 * stream that tries to delete/mkdir the root dir.
5952 if (left_gen
!= right_gen
&&
5953 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5954 sctx
->cur_inode_new_gen
= 1;
5957 if (result
== BTRFS_COMPARE_TREE_NEW
) {
5958 sctx
->cur_inode_gen
= left_gen
;
5959 sctx
->cur_inode_new
= 1;
5960 sctx
->cur_inode_deleted
= 0;
5961 sctx
->cur_inode_size
= btrfs_inode_size(
5962 sctx
->left_path
->nodes
[0], left_ii
);
5963 sctx
->cur_inode_mode
= btrfs_inode_mode(
5964 sctx
->left_path
->nodes
[0], left_ii
);
5965 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
5966 sctx
->left_path
->nodes
[0], left_ii
);
5967 if (sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5968 ret
= send_create_inode_if_needed(sctx
);
5969 } else if (result
== BTRFS_COMPARE_TREE_DELETED
) {
5970 sctx
->cur_inode_gen
= right_gen
;
5971 sctx
->cur_inode_new
= 0;
5972 sctx
->cur_inode_deleted
= 1;
5973 sctx
->cur_inode_size
= btrfs_inode_size(
5974 sctx
->right_path
->nodes
[0], right_ii
);
5975 sctx
->cur_inode_mode
= btrfs_inode_mode(
5976 sctx
->right_path
->nodes
[0], right_ii
);
5977 } else if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5979 * We need to do some special handling in case the inode was
5980 * reported as changed with a changed generation number. This
5981 * means that the original inode was deleted and new inode
5982 * reused the same inum. So we have to treat the old inode as
5983 * deleted and the new one as new.
5985 if (sctx
->cur_inode_new_gen
) {
5987 * First, process the inode as if it was deleted.
5989 sctx
->cur_inode_gen
= right_gen
;
5990 sctx
->cur_inode_new
= 0;
5991 sctx
->cur_inode_deleted
= 1;
5992 sctx
->cur_inode_size
= btrfs_inode_size(
5993 sctx
->right_path
->nodes
[0], right_ii
);
5994 sctx
->cur_inode_mode
= btrfs_inode_mode(
5995 sctx
->right_path
->nodes
[0], right_ii
);
5996 ret
= process_all_refs(sctx
,
5997 BTRFS_COMPARE_TREE_DELETED
);
6002 * Now process the inode as if it was new.
6004 sctx
->cur_inode_gen
= left_gen
;
6005 sctx
->cur_inode_new
= 1;
6006 sctx
->cur_inode_deleted
= 0;
6007 sctx
->cur_inode_size
= btrfs_inode_size(
6008 sctx
->left_path
->nodes
[0], left_ii
);
6009 sctx
->cur_inode_mode
= btrfs_inode_mode(
6010 sctx
->left_path
->nodes
[0], left_ii
);
6011 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
6012 sctx
->left_path
->nodes
[0], left_ii
);
6013 ret
= send_create_inode_if_needed(sctx
);
6017 ret
= process_all_refs(sctx
, BTRFS_COMPARE_TREE_NEW
);
6021 * Advance send_progress now as we did not get into
6022 * process_recorded_refs_if_needed in the new_gen case.
6024 sctx
->send_progress
= sctx
->cur_ino
+ 1;
6027 * Now process all extents and xattrs of the inode as if
6028 * they were all new.
6030 ret
= process_all_extents(sctx
);
6033 ret
= process_all_new_xattrs(sctx
);
6037 sctx
->cur_inode_gen
= left_gen
;
6038 sctx
->cur_inode_new
= 0;
6039 sctx
->cur_inode_new_gen
= 0;
6040 sctx
->cur_inode_deleted
= 0;
6041 sctx
->cur_inode_size
= btrfs_inode_size(
6042 sctx
->left_path
->nodes
[0], left_ii
);
6043 sctx
->cur_inode_mode
= btrfs_inode_mode(
6044 sctx
->left_path
->nodes
[0], left_ii
);
6053 * We have to process new refs before deleted refs, but compare_trees gives us
6054 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6055 * first and later process them in process_recorded_refs.
6056 * For the cur_inode_new_gen case, we skip recording completely because
6057 * changed_inode did already initiate processing of refs. The reason for this is
6058 * that in this case, compare_tree actually compares the refs of 2 different
6059 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6060 * refs of the right tree as deleted and all refs of the left tree as new.
6062 static int changed_ref(struct send_ctx
*sctx
,
6063 enum btrfs_compare_tree_result result
)
6067 if (sctx
->cur_ino
!= sctx
->cmp_key
->objectid
) {
6068 inconsistent_snapshot_error(sctx
, result
, "reference");
6072 if (!sctx
->cur_inode_new_gen
&&
6073 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
6074 if (result
== BTRFS_COMPARE_TREE_NEW
)
6075 ret
= record_new_ref(sctx
);
6076 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
6077 ret
= record_deleted_ref(sctx
);
6078 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
6079 ret
= record_changed_ref(sctx
);
6086 * Process new/deleted/changed xattrs. We skip processing in the
6087 * cur_inode_new_gen case because changed_inode did already initiate processing
6088 * of xattrs. The reason is the same as in changed_ref
6090 static int changed_xattr(struct send_ctx
*sctx
,
6091 enum btrfs_compare_tree_result result
)
6095 if (sctx
->cur_ino
!= sctx
->cmp_key
->objectid
) {
6096 inconsistent_snapshot_error(sctx
, result
, "xattr");
6100 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
6101 if (result
== BTRFS_COMPARE_TREE_NEW
)
6102 ret
= process_new_xattr(sctx
);
6103 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
6104 ret
= process_deleted_xattr(sctx
);
6105 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
6106 ret
= process_changed_xattr(sctx
);
6113 * Process new/deleted/changed extents. We skip processing in the
6114 * cur_inode_new_gen case because changed_inode did already initiate processing
6115 * of extents. The reason is the same as in changed_ref
6117 static int changed_extent(struct send_ctx
*sctx
,
6118 enum btrfs_compare_tree_result result
)
6122 if (sctx
->cur_ino
!= sctx
->cmp_key
->objectid
) {
6124 if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
6125 struct extent_buffer
*leaf_l
;
6126 struct extent_buffer
*leaf_r
;
6127 struct btrfs_file_extent_item
*ei_l
;
6128 struct btrfs_file_extent_item
*ei_r
;
6130 leaf_l
= sctx
->left_path
->nodes
[0];
6131 leaf_r
= sctx
->right_path
->nodes
[0];
6132 ei_l
= btrfs_item_ptr(leaf_l
,
6133 sctx
->left_path
->slots
[0],
6134 struct btrfs_file_extent_item
);
6135 ei_r
= btrfs_item_ptr(leaf_r
,
6136 sctx
->right_path
->slots
[0],
6137 struct btrfs_file_extent_item
);
6140 * We may have found an extent item that has changed
6141 * only its disk_bytenr field and the corresponding
6142 * inode item was not updated. This case happens due to
6143 * very specific timings during relocation when a leaf
6144 * that contains file extent items is COWed while
6145 * relocation is ongoing and its in the stage where it
6146 * updates data pointers. So when this happens we can
6147 * safely ignore it since we know it's the same extent,
6148 * but just at different logical and physical locations
6149 * (when an extent is fully replaced with a new one, we
6150 * know the generation number must have changed too,
6151 * since snapshot creation implies committing the current
6152 * transaction, and the inode item must have been updated
6154 * This replacement of the disk_bytenr happens at
6155 * relocation.c:replace_file_extents() through
6156 * relocation.c:btrfs_reloc_cow_block().
6158 if (btrfs_file_extent_generation(leaf_l
, ei_l
) ==
6159 btrfs_file_extent_generation(leaf_r
, ei_r
) &&
6160 btrfs_file_extent_ram_bytes(leaf_l
, ei_l
) ==
6161 btrfs_file_extent_ram_bytes(leaf_r
, ei_r
) &&
6162 btrfs_file_extent_compression(leaf_l
, ei_l
) ==
6163 btrfs_file_extent_compression(leaf_r
, ei_r
) &&
6164 btrfs_file_extent_encryption(leaf_l
, ei_l
) ==
6165 btrfs_file_extent_encryption(leaf_r
, ei_r
) &&
6166 btrfs_file_extent_other_encoding(leaf_l
, ei_l
) ==
6167 btrfs_file_extent_other_encoding(leaf_r
, ei_r
) &&
6168 btrfs_file_extent_type(leaf_l
, ei_l
) ==
6169 btrfs_file_extent_type(leaf_r
, ei_r
) &&
6170 btrfs_file_extent_disk_bytenr(leaf_l
, ei_l
) !=
6171 btrfs_file_extent_disk_bytenr(leaf_r
, ei_r
) &&
6172 btrfs_file_extent_disk_num_bytes(leaf_l
, ei_l
) ==
6173 btrfs_file_extent_disk_num_bytes(leaf_r
, ei_r
) &&
6174 btrfs_file_extent_offset(leaf_l
, ei_l
) ==
6175 btrfs_file_extent_offset(leaf_r
, ei_r
) &&
6176 btrfs_file_extent_num_bytes(leaf_l
, ei_l
) ==
6177 btrfs_file_extent_num_bytes(leaf_r
, ei_r
))
6181 inconsistent_snapshot_error(sctx
, result
, "extent");
6185 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
6186 if (result
!= BTRFS_COMPARE_TREE_DELETED
)
6187 ret
= process_extent(sctx
, sctx
->left_path
,
6194 static int dir_changed(struct send_ctx
*sctx
, u64 dir
)
6196 u64 orig_gen
, new_gen
;
6199 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &new_gen
, NULL
, NULL
,
6204 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &orig_gen
, NULL
,
6209 return (orig_gen
!= new_gen
) ? 1 : 0;
6212 static int compare_refs(struct send_ctx
*sctx
, struct btrfs_path
*path
,
6213 struct btrfs_key
*key
)
6215 struct btrfs_inode_extref
*extref
;
6216 struct extent_buffer
*leaf
;
6217 u64 dirid
= 0, last_dirid
= 0;
6224 /* Easy case, just check this one dirid */
6225 if (key
->type
== BTRFS_INODE_REF_KEY
) {
6226 dirid
= key
->offset
;
6228 ret
= dir_changed(sctx
, dirid
);
6232 leaf
= path
->nodes
[0];
6233 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
6234 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
6235 while (cur_offset
< item_size
) {
6236 extref
= (struct btrfs_inode_extref
*)(ptr
+
6238 dirid
= btrfs_inode_extref_parent(leaf
, extref
);
6239 ref_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
6240 cur_offset
+= ref_name_len
+ sizeof(*extref
);
6241 if (dirid
== last_dirid
)
6243 ret
= dir_changed(sctx
, dirid
);
6253 * Updates compare related fields in sctx and simply forwards to the actual
6254 * changed_xxx functions.
6256 static int changed_cb(struct btrfs_path
*left_path
,
6257 struct btrfs_path
*right_path
,
6258 struct btrfs_key
*key
,
6259 enum btrfs_compare_tree_result result
,
6263 struct send_ctx
*sctx
= ctx
;
6265 if (result
== BTRFS_COMPARE_TREE_SAME
) {
6266 if (key
->type
== BTRFS_INODE_REF_KEY
||
6267 key
->type
== BTRFS_INODE_EXTREF_KEY
) {
6268 ret
= compare_refs(sctx
, left_path
, key
);
6273 } else if (key
->type
== BTRFS_EXTENT_DATA_KEY
) {
6274 return maybe_send_hole(sctx
, left_path
, key
);
6278 result
= BTRFS_COMPARE_TREE_CHANGED
;
6282 sctx
->left_path
= left_path
;
6283 sctx
->right_path
= right_path
;
6284 sctx
->cmp_key
= key
;
6286 ret
= finish_inode_if_needed(sctx
, 0);
6290 /* Ignore non-FS objects */
6291 if (key
->objectid
== BTRFS_FREE_INO_OBJECTID
||
6292 key
->objectid
== BTRFS_FREE_SPACE_OBJECTID
)
6295 if (key
->type
== BTRFS_INODE_ITEM_KEY
)
6296 ret
= changed_inode(sctx
, result
);
6297 else if (key
->type
== BTRFS_INODE_REF_KEY
||
6298 key
->type
== BTRFS_INODE_EXTREF_KEY
)
6299 ret
= changed_ref(sctx
, result
);
6300 else if (key
->type
== BTRFS_XATTR_ITEM_KEY
)
6301 ret
= changed_xattr(sctx
, result
);
6302 else if (key
->type
== BTRFS_EXTENT_DATA_KEY
)
6303 ret
= changed_extent(sctx
, result
);
6309 static int full_send_tree(struct send_ctx
*sctx
)
6312 struct btrfs_root
*send_root
= sctx
->send_root
;
6313 struct btrfs_key key
;
6314 struct btrfs_key found_key
;
6315 struct btrfs_path
*path
;
6316 struct extent_buffer
*eb
;
6319 path
= alloc_path_for_send();
6323 key
.objectid
= BTRFS_FIRST_FREE_OBJECTID
;
6324 key
.type
= BTRFS_INODE_ITEM_KEY
;
6327 ret
= btrfs_search_slot_for_read(send_root
, &key
, path
, 1, 0);
6334 eb
= path
->nodes
[0];
6335 slot
= path
->slots
[0];
6336 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6338 ret
= changed_cb(path
, NULL
, &found_key
,
6339 BTRFS_COMPARE_TREE_NEW
, sctx
);
6343 key
.objectid
= found_key
.objectid
;
6344 key
.type
= found_key
.type
;
6345 key
.offset
= found_key
.offset
+ 1;
6347 ret
= btrfs_next_item(send_root
, path
);
6357 ret
= finish_inode_if_needed(sctx
, 1);
6360 btrfs_free_path(path
);
6364 static int send_subvol(struct send_ctx
*sctx
)
6368 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_STREAM_HEADER
)) {
6369 ret
= send_header(sctx
);
6374 ret
= send_subvol_begin(sctx
);
6378 if (sctx
->parent_root
) {
6379 ret
= btrfs_compare_trees(sctx
->send_root
, sctx
->parent_root
,
6383 ret
= finish_inode_if_needed(sctx
, 1);
6387 ret
= full_send_tree(sctx
);
6393 free_recorded_refs(sctx
);
6398 * If orphan cleanup did remove any orphans from a root, it means the tree
6399 * was modified and therefore the commit root is not the same as the current
6400 * root anymore. This is a problem, because send uses the commit root and
6401 * therefore can see inode items that don't exist in the current root anymore,
6402 * and for example make calls to btrfs_iget, which will do tree lookups based
6403 * on the current root and not on the commit root. Those lookups will fail,
6404 * returning a -ESTALE error, and making send fail with that error. So make
6405 * sure a send does not see any orphans we have just removed, and that it will
6406 * see the same inodes regardless of whether a transaction commit happened
6407 * before it started (meaning that the commit root will be the same as the
6408 * current root) or not.
6410 static int ensure_commit_roots_uptodate(struct send_ctx
*sctx
)
6413 struct btrfs_trans_handle
*trans
= NULL
;
6416 if (sctx
->parent_root
&&
6417 sctx
->parent_root
->node
!= sctx
->parent_root
->commit_root
)
6420 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
6421 if (sctx
->clone_roots
[i
].root
->node
!=
6422 sctx
->clone_roots
[i
].root
->commit_root
)
6426 return btrfs_end_transaction(trans
);
6431 /* Use any root, all fs roots will get their commit roots updated. */
6433 trans
= btrfs_join_transaction(sctx
->send_root
);
6435 return PTR_ERR(trans
);
6439 return btrfs_commit_transaction(trans
);
6442 static void btrfs_root_dec_send_in_progress(struct btrfs_root
* root
)
6444 spin_lock(&root
->root_item_lock
);
6445 root
->send_in_progress
--;
6447 * Not much left to do, we don't know why it's unbalanced and
6448 * can't blindly reset it to 0.
6450 if (root
->send_in_progress
< 0)
6451 btrfs_err(root
->fs_info
,
6452 "send_in_progres unbalanced %d root %llu",
6453 root
->send_in_progress
, root
->root_key
.objectid
);
6454 spin_unlock(&root
->root_item_lock
);
6457 long btrfs_ioctl_send(struct file
*mnt_file
, struct btrfs_ioctl_send_args
*arg
)
6460 struct btrfs_root
*send_root
= BTRFS_I(file_inode(mnt_file
))->root
;
6461 struct btrfs_fs_info
*fs_info
= send_root
->fs_info
;
6462 struct btrfs_root
*clone_root
;
6463 struct btrfs_key key
;
6464 struct send_ctx
*sctx
= NULL
;
6466 u64
*clone_sources_tmp
= NULL
;
6467 int clone_sources_to_rollback
= 0;
6468 unsigned alloc_size
;
6469 int sort_clone_roots
= 0;
6472 if (!capable(CAP_SYS_ADMIN
))
6476 * The subvolume must remain read-only during send, protect against
6477 * making it RW. This also protects against deletion.
6479 spin_lock(&send_root
->root_item_lock
);
6480 send_root
->send_in_progress
++;
6481 spin_unlock(&send_root
->root_item_lock
);
6484 * This is done when we lookup the root, it should already be complete
6485 * by the time we get here.
6487 WARN_ON(send_root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
);
6490 * Userspace tools do the checks and warn the user if it's
6493 if (!btrfs_root_readonly(send_root
)) {
6499 * Check that we don't overflow at later allocations, we request
6500 * clone_sources_count + 1 items, and compare to unsigned long inside
6503 if (arg
->clone_sources_count
>
6504 ULONG_MAX
/ sizeof(struct clone_root
) - 1) {
6509 if (!access_ok(VERIFY_READ
, arg
->clone_sources
,
6510 sizeof(*arg
->clone_sources
) *
6511 arg
->clone_sources_count
)) {
6516 if (arg
->flags
& ~BTRFS_SEND_FLAG_MASK
) {
6521 sctx
= kzalloc(sizeof(struct send_ctx
), GFP_KERNEL
);
6527 INIT_LIST_HEAD(&sctx
->new_refs
);
6528 INIT_LIST_HEAD(&sctx
->deleted_refs
);
6529 INIT_RADIX_TREE(&sctx
->name_cache
, GFP_KERNEL
);
6530 INIT_LIST_HEAD(&sctx
->name_cache_list
);
6532 sctx
->flags
= arg
->flags
;
6534 sctx
->send_filp
= fget(arg
->send_fd
);
6535 if (!sctx
->send_filp
) {
6540 sctx
->send_root
= send_root
;
6542 * Unlikely but possible, if the subvolume is marked for deletion but
6543 * is slow to remove the directory entry, send can still be started
6545 if (btrfs_root_dead(sctx
->send_root
)) {
6550 sctx
->clone_roots_cnt
= arg
->clone_sources_count
;
6552 sctx
->send_max_size
= BTRFS_SEND_BUF_SIZE
;
6553 sctx
->send_buf
= kvmalloc(sctx
->send_max_size
, GFP_KERNEL
);
6554 if (!sctx
->send_buf
) {
6559 sctx
->read_buf
= kvmalloc(BTRFS_SEND_READ_SIZE
, GFP_KERNEL
);
6560 if (!sctx
->read_buf
) {
6565 sctx
->pending_dir_moves
= RB_ROOT
;
6566 sctx
->waiting_dir_moves
= RB_ROOT
;
6567 sctx
->orphan_dirs
= RB_ROOT
;
6569 alloc_size
= sizeof(struct clone_root
) * (arg
->clone_sources_count
+ 1);
6571 sctx
->clone_roots
= kzalloc(alloc_size
, GFP_KERNEL
);
6572 if (!sctx
->clone_roots
) {
6577 alloc_size
= arg
->clone_sources_count
* sizeof(*arg
->clone_sources
);
6579 if (arg
->clone_sources_count
) {
6580 clone_sources_tmp
= kvmalloc(alloc_size
, GFP_KERNEL
);
6581 if (!clone_sources_tmp
) {
6586 ret
= copy_from_user(clone_sources_tmp
, arg
->clone_sources
,
6593 for (i
= 0; i
< arg
->clone_sources_count
; i
++) {
6594 key
.objectid
= clone_sources_tmp
[i
];
6595 key
.type
= BTRFS_ROOT_ITEM_KEY
;
6596 key
.offset
= (u64
)-1;
6598 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
6600 clone_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
6601 if (IS_ERR(clone_root
)) {
6602 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6603 ret
= PTR_ERR(clone_root
);
6606 spin_lock(&clone_root
->root_item_lock
);
6607 if (!btrfs_root_readonly(clone_root
) ||
6608 btrfs_root_dead(clone_root
)) {
6609 spin_unlock(&clone_root
->root_item_lock
);
6610 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6614 clone_root
->send_in_progress
++;
6615 spin_unlock(&clone_root
->root_item_lock
);
6616 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6618 sctx
->clone_roots
[i
].root
= clone_root
;
6619 clone_sources_to_rollback
= i
+ 1;
6621 kvfree(clone_sources_tmp
);
6622 clone_sources_tmp
= NULL
;
6625 if (arg
->parent_root
) {
6626 key
.objectid
= arg
->parent_root
;
6627 key
.type
= BTRFS_ROOT_ITEM_KEY
;
6628 key
.offset
= (u64
)-1;
6630 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
6632 sctx
->parent_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
6633 if (IS_ERR(sctx
->parent_root
)) {
6634 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6635 ret
= PTR_ERR(sctx
->parent_root
);
6639 spin_lock(&sctx
->parent_root
->root_item_lock
);
6640 sctx
->parent_root
->send_in_progress
++;
6641 if (!btrfs_root_readonly(sctx
->parent_root
) ||
6642 btrfs_root_dead(sctx
->parent_root
)) {
6643 spin_unlock(&sctx
->parent_root
->root_item_lock
);
6644 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6648 spin_unlock(&sctx
->parent_root
->root_item_lock
);
6650 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6654 * Clones from send_root are allowed, but only if the clone source
6655 * is behind the current send position. This is checked while searching
6656 * for possible clone sources.
6658 sctx
->clone_roots
[sctx
->clone_roots_cnt
++].root
= sctx
->send_root
;
6660 /* We do a bsearch later */
6661 sort(sctx
->clone_roots
, sctx
->clone_roots_cnt
,
6662 sizeof(*sctx
->clone_roots
), __clone_root_cmp_sort
,
6664 sort_clone_roots
= 1;
6666 ret
= ensure_commit_roots_uptodate(sctx
);
6670 current
->journal_info
= BTRFS_SEND_TRANS_STUB
;
6671 ret
= send_subvol(sctx
);
6672 current
->journal_info
= NULL
;
6676 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_END_CMD
)) {
6677 ret
= begin_cmd(sctx
, BTRFS_SEND_C_END
);
6680 ret
= send_cmd(sctx
);
6686 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
));
6687 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
)) {
6689 struct pending_dir_move
*pm
;
6691 n
= rb_first(&sctx
->pending_dir_moves
);
6692 pm
= rb_entry(n
, struct pending_dir_move
, node
);
6693 while (!list_empty(&pm
->list
)) {
6694 struct pending_dir_move
*pm2
;
6696 pm2
= list_first_entry(&pm
->list
,
6697 struct pending_dir_move
, list
);
6698 free_pending_move(sctx
, pm2
);
6700 free_pending_move(sctx
, pm
);
6703 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
));
6704 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
)) {
6706 struct waiting_dir_move
*dm
;
6708 n
= rb_first(&sctx
->waiting_dir_moves
);
6709 dm
= rb_entry(n
, struct waiting_dir_move
, node
);
6710 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
6714 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
));
6715 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
)) {
6717 struct orphan_dir_info
*odi
;
6719 n
= rb_first(&sctx
->orphan_dirs
);
6720 odi
= rb_entry(n
, struct orphan_dir_info
, node
);
6721 free_orphan_dir_info(sctx
, odi
);
6724 if (sort_clone_roots
) {
6725 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
6726 btrfs_root_dec_send_in_progress(
6727 sctx
->clone_roots
[i
].root
);
6729 for (i
= 0; sctx
&& i
< clone_sources_to_rollback
; i
++)
6730 btrfs_root_dec_send_in_progress(
6731 sctx
->clone_roots
[i
].root
);
6733 btrfs_root_dec_send_in_progress(send_root
);
6735 if (sctx
&& !IS_ERR_OR_NULL(sctx
->parent_root
))
6736 btrfs_root_dec_send_in_progress(sctx
->parent_root
);
6738 kvfree(clone_sources_tmp
);
6741 if (sctx
->send_filp
)
6742 fput(sctx
->send_filp
);
6744 kvfree(sctx
->clone_roots
);
6745 kvfree(sctx
->send_buf
);
6746 kvfree(sctx
->read_buf
);
6748 name_cache_free(sctx
);