2 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
35 #include "btrfs_inode.h"
36 #include "transaction.h"
37 #include "compression.h"
40 * A fs_path is a helper to dynamically build path names with unknown size.
41 * It reallocates the internal buffer on demand.
42 * It allows fast adding of path elements on the right side (normal path) and
43 * fast adding to the left side (reversed path). A reversed path can also be
44 * unreversed if needed.
53 unsigned short buf_len
:15;
54 unsigned short reversed
:1;
58 * Average path length does not exceed 200 bytes, we'll have
59 * better packing in the slab and higher chance to satisfy
60 * a allocation later during send.
65 #define FS_PATH_INLINE_SIZE \
66 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
69 /* reused for each extent */
71 struct btrfs_root
*root
;
78 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
79 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
82 struct file
*send_filp
;
88 u64 cmd_send_size
[BTRFS_SEND_C_MAX
+ 1];
89 u64 flags
; /* 'flags' member of btrfs_ioctl_send_args is u64 */
91 struct btrfs_root
*send_root
;
92 struct btrfs_root
*parent_root
;
93 struct clone_root
*clone_roots
;
96 /* current state of the compare_tree call */
97 struct btrfs_path
*left_path
;
98 struct btrfs_path
*right_path
;
99 struct btrfs_key
*cmp_key
;
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
108 int cur_inode_new_gen
;
109 int cur_inode_deleted
;
113 u64 cur_inode_last_extent
;
117 struct list_head new_refs
;
118 struct list_head deleted_refs
;
120 struct radix_tree_root name_cache
;
121 struct list_head name_cache_list
;
124 struct file_ra_state ra
;
129 * We process inodes by their increasing order, so if before an
130 * incremental send we reverse the parent/child relationship of
131 * directories such that a directory with a lower inode number was
132 * the parent of a directory with a higher inode number, and the one
133 * becoming the new parent got renamed too, we can't rename/move the
134 * directory with lower inode number when we finish processing it - we
135 * must process the directory with higher inode number first, then
136 * rename/move it and then rename/move the directory with lower inode
137 * number. Example follows.
139 * Tree state when the first send was performed:
151 * Tree state when the second (incremental) send is performed:
160 * The sequence of steps that lead to the second state was:
162 * mv /a/b/c/d /a/b/c2/d2
163 * mv /a/b/c /a/b/c2/d2/cc
165 * "c" has lower inode number, but we can't move it (2nd mv operation)
166 * before we move "d", which has higher inode number.
168 * So we just memorize which move/rename operations must be performed
169 * later when their respective parent is processed and moved/renamed.
172 /* Indexed by parent directory inode number. */
173 struct rb_root pending_dir_moves
;
176 * Reverse index, indexed by the inode number of a directory that
177 * is waiting for the move/rename of its immediate parent before its
178 * own move/rename can be performed.
180 struct rb_root waiting_dir_moves
;
183 * A directory that is going to be rm'ed might have a child directory
184 * which is in the pending directory moves index above. In this case,
185 * the directory can only be removed after the move/rename of its child
186 * is performed. Example:
206 * Sequence of steps that lead to the send snapshot:
207 * rm -f /a/b/c/foo.txt
209 * mv /a/b/c/x /a/b/YY
212 * When the child is processed, its move/rename is delayed until its
213 * parent is processed (as explained above), but all other operations
214 * like update utimes, chown, chgrp, etc, are performed and the paths
215 * that it uses for those operations must use the orphanized name of
216 * its parent (the directory we're going to rm later), so we need to
217 * memorize that name.
219 * Indexed by the inode number of the directory to be deleted.
221 struct rb_root orphan_dirs
;
224 struct pending_dir_move
{
226 struct list_head list
;
230 struct list_head update_refs
;
233 struct waiting_dir_move
{
237 * There might be some directory that could not be removed because it
238 * was waiting for this directory inode to be moved first. Therefore
239 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
245 struct orphan_dir_info
{
251 struct name_cache_entry
{
252 struct list_head list
;
254 * radix_tree has only 32bit entries but we need to handle 64bit inums.
255 * We use the lower 32bit of the 64bit inum to store it in the tree. If
256 * more then one inum would fall into the same entry, we use radix_list
257 * to store the additional entries. radix_list is also used to store
258 * entries where two entries have the same inum but different
261 struct list_head radix_list
;
267 int need_later_update
;
272 static void inconsistent_snapshot_error(struct send_ctx
*sctx
,
273 enum btrfs_compare_tree_result result
,
276 const char *result_string
;
279 case BTRFS_COMPARE_TREE_NEW
:
280 result_string
= "new";
282 case BTRFS_COMPARE_TREE_DELETED
:
283 result_string
= "deleted";
285 case BTRFS_COMPARE_TREE_CHANGED
:
286 result_string
= "updated";
288 case BTRFS_COMPARE_TREE_SAME
:
290 result_string
= "unchanged";
294 result_string
= "unexpected";
297 btrfs_err(sctx
->send_root
->fs_info
,
298 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
299 result_string
, what
, sctx
->cmp_key
->objectid
,
300 sctx
->send_root
->root_key
.objectid
,
302 sctx
->parent_root
->root_key
.objectid
: 0));
305 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
);
307 static struct waiting_dir_move
*
308 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
);
310 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
);
312 static int need_send_hole(struct send_ctx
*sctx
)
314 return (sctx
->parent_root
&& !sctx
->cur_inode_new
&&
315 !sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
&&
316 S_ISREG(sctx
->cur_inode_mode
));
319 static void fs_path_reset(struct fs_path
*p
)
322 p
->start
= p
->buf
+ p
->buf_len
- 1;
332 static struct fs_path
*fs_path_alloc(void)
336 p
= kmalloc(sizeof(*p
), GFP_KERNEL
);
340 p
->buf
= p
->inline_buf
;
341 p
->buf_len
= FS_PATH_INLINE_SIZE
;
346 static struct fs_path
*fs_path_alloc_reversed(void)
358 static void fs_path_free(struct fs_path
*p
)
362 if (p
->buf
!= p
->inline_buf
)
367 static int fs_path_len(struct fs_path
*p
)
369 return p
->end
- p
->start
;
372 static int fs_path_ensure_buf(struct fs_path
*p
, int len
)
380 if (p
->buf_len
>= len
)
383 if (len
> PATH_MAX
) {
388 path_len
= p
->end
- p
->start
;
389 old_buf_len
= p
->buf_len
;
392 * First time the inline_buf does not suffice
394 if (p
->buf
== p
->inline_buf
) {
395 tmp_buf
= kmalloc(len
, GFP_KERNEL
);
397 memcpy(tmp_buf
, p
->buf
, old_buf_len
);
399 tmp_buf
= krealloc(p
->buf
, len
, GFP_KERNEL
);
405 * The real size of the buffer is bigger, this will let the fast path
406 * happen most of the time
408 p
->buf_len
= ksize(p
->buf
);
411 tmp_buf
= p
->buf
+ old_buf_len
- path_len
- 1;
412 p
->end
= p
->buf
+ p
->buf_len
- 1;
413 p
->start
= p
->end
- path_len
;
414 memmove(p
->start
, tmp_buf
, path_len
+ 1);
417 p
->end
= p
->start
+ path_len
;
422 static int fs_path_prepare_for_add(struct fs_path
*p
, int name_len
,
428 new_len
= p
->end
- p
->start
+ name_len
;
429 if (p
->start
!= p
->end
)
431 ret
= fs_path_ensure_buf(p
, new_len
);
436 if (p
->start
!= p
->end
)
438 p
->start
-= name_len
;
439 *prepared
= p
->start
;
441 if (p
->start
!= p
->end
)
452 static int fs_path_add(struct fs_path
*p
, const char *name
, int name_len
)
457 ret
= fs_path_prepare_for_add(p
, name_len
, &prepared
);
460 memcpy(prepared
, name
, name_len
);
466 static int fs_path_add_path(struct fs_path
*p
, struct fs_path
*p2
)
471 ret
= fs_path_prepare_for_add(p
, p2
->end
- p2
->start
, &prepared
);
474 memcpy(prepared
, p2
->start
, p2
->end
- p2
->start
);
480 static int fs_path_add_from_extent_buffer(struct fs_path
*p
,
481 struct extent_buffer
*eb
,
482 unsigned long off
, int len
)
487 ret
= fs_path_prepare_for_add(p
, len
, &prepared
);
491 read_extent_buffer(eb
, prepared
, off
, len
);
497 static int fs_path_copy(struct fs_path
*p
, struct fs_path
*from
)
501 p
->reversed
= from
->reversed
;
504 ret
= fs_path_add_path(p
, from
);
510 static void fs_path_unreverse(struct fs_path
*p
)
519 len
= p
->end
- p
->start
;
521 p
->end
= p
->start
+ len
;
522 memmove(p
->start
, tmp
, len
+ 1);
526 static struct btrfs_path
*alloc_path_for_send(void)
528 struct btrfs_path
*path
;
530 path
= btrfs_alloc_path();
533 path
->search_commit_root
= 1;
534 path
->skip_locking
= 1;
535 path
->need_commit_sem
= 1;
539 static int write_buf(struct file
*filp
, const void *buf
, u32 len
, loff_t
*off
)
545 ret
= kernel_write(filp
, buf
+ pos
, len
- pos
, off
);
546 /* TODO handle that correctly */
547 /*if (ret == -ERESTARTSYS) {
561 static int tlv_put(struct send_ctx
*sctx
, u16 attr
, const void *data
, int len
)
563 struct btrfs_tlv_header
*hdr
;
564 int total_len
= sizeof(*hdr
) + len
;
565 int left
= sctx
->send_max_size
- sctx
->send_size
;
567 if (unlikely(left
< total_len
))
570 hdr
= (struct btrfs_tlv_header
*) (sctx
->send_buf
+ sctx
->send_size
);
571 hdr
->tlv_type
= cpu_to_le16(attr
);
572 hdr
->tlv_len
= cpu_to_le16(len
);
573 memcpy(hdr
+ 1, data
, len
);
574 sctx
->send_size
+= total_len
;
579 #define TLV_PUT_DEFINE_INT(bits) \
580 static int tlv_put_u##bits(struct send_ctx *sctx, \
581 u##bits attr, u##bits value) \
583 __le##bits __tmp = cpu_to_le##bits(value); \
584 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
587 TLV_PUT_DEFINE_INT(64)
589 static int tlv_put_string(struct send_ctx
*sctx
, u16 attr
,
590 const char *str
, int len
)
594 return tlv_put(sctx
, attr
, str
, len
);
597 static int tlv_put_uuid(struct send_ctx
*sctx
, u16 attr
,
600 return tlv_put(sctx
, attr
, uuid
, BTRFS_UUID_SIZE
);
603 static int tlv_put_btrfs_timespec(struct send_ctx
*sctx
, u16 attr
,
604 struct extent_buffer
*eb
,
605 struct btrfs_timespec
*ts
)
607 struct btrfs_timespec bts
;
608 read_extent_buffer(eb
, &bts
, (unsigned long)ts
, sizeof(bts
));
609 return tlv_put(sctx
, attr
, &bts
, sizeof(bts
));
613 #define TLV_PUT(sctx, attrtype, attrlen, data) \
615 ret = tlv_put(sctx, attrtype, attrlen, data); \
617 goto tlv_put_failure; \
620 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
622 ret = tlv_put_u##bits(sctx, attrtype, value); \
624 goto tlv_put_failure; \
627 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
628 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
629 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
630 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
631 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
633 ret = tlv_put_string(sctx, attrtype, str, len); \
635 goto tlv_put_failure; \
637 #define TLV_PUT_PATH(sctx, attrtype, p) \
639 ret = tlv_put_string(sctx, attrtype, p->start, \
640 p->end - p->start); \
642 goto tlv_put_failure; \
644 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
646 ret = tlv_put_uuid(sctx, attrtype, uuid); \
648 goto tlv_put_failure; \
650 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
652 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
654 goto tlv_put_failure; \
657 static int send_header(struct send_ctx
*sctx
)
659 struct btrfs_stream_header hdr
;
661 strcpy(hdr
.magic
, BTRFS_SEND_STREAM_MAGIC
);
662 hdr
.version
= cpu_to_le32(BTRFS_SEND_STREAM_VERSION
);
664 return write_buf(sctx
->send_filp
, &hdr
, sizeof(hdr
),
669 * For each command/item we want to send to userspace, we call this function.
671 static int begin_cmd(struct send_ctx
*sctx
, int cmd
)
673 struct btrfs_cmd_header
*hdr
;
675 if (WARN_ON(!sctx
->send_buf
))
678 BUG_ON(sctx
->send_size
);
680 sctx
->send_size
+= sizeof(*hdr
);
681 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
682 hdr
->cmd
= cpu_to_le16(cmd
);
687 static int send_cmd(struct send_ctx
*sctx
)
690 struct btrfs_cmd_header
*hdr
;
693 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
694 hdr
->len
= cpu_to_le32(sctx
->send_size
- sizeof(*hdr
));
697 crc
= btrfs_crc32c(0, (unsigned char *)sctx
->send_buf
, sctx
->send_size
);
698 hdr
->crc
= cpu_to_le32(crc
);
700 ret
= write_buf(sctx
->send_filp
, sctx
->send_buf
, sctx
->send_size
,
703 sctx
->total_send_size
+= sctx
->send_size
;
704 sctx
->cmd_send_size
[le16_to_cpu(hdr
->cmd
)] += sctx
->send_size
;
711 * Sends a move instruction to user space
713 static int send_rename(struct send_ctx
*sctx
,
714 struct fs_path
*from
, struct fs_path
*to
)
716 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
719 btrfs_debug(fs_info
, "send_rename %s -> %s", from
->start
, to
->start
);
721 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RENAME
);
725 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, from
);
726 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_TO
, to
);
728 ret
= send_cmd(sctx
);
736 * Sends a link instruction to user space
738 static int send_link(struct send_ctx
*sctx
,
739 struct fs_path
*path
, struct fs_path
*lnk
)
741 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
744 btrfs_debug(fs_info
, "send_link %s -> %s", path
->start
, lnk
->start
);
746 ret
= begin_cmd(sctx
, BTRFS_SEND_C_LINK
);
750 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
751 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, lnk
);
753 ret
= send_cmd(sctx
);
761 * Sends an unlink instruction to user space
763 static int send_unlink(struct send_ctx
*sctx
, struct fs_path
*path
)
765 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
768 btrfs_debug(fs_info
, "send_unlink %s", path
->start
);
770 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UNLINK
);
774 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
776 ret
= send_cmd(sctx
);
784 * Sends a rmdir instruction to user space
786 static int send_rmdir(struct send_ctx
*sctx
, struct fs_path
*path
)
788 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
791 btrfs_debug(fs_info
, "send_rmdir %s", path
->start
);
793 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RMDIR
);
797 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
799 ret
= send_cmd(sctx
);
807 * Helper function to retrieve some fields from an inode item.
809 static int __get_inode_info(struct btrfs_root
*root
, struct btrfs_path
*path
,
810 u64 ino
, u64
*size
, u64
*gen
, u64
*mode
, u64
*uid
,
814 struct btrfs_inode_item
*ii
;
815 struct btrfs_key key
;
818 key
.type
= BTRFS_INODE_ITEM_KEY
;
820 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
827 ii
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
828 struct btrfs_inode_item
);
830 *size
= btrfs_inode_size(path
->nodes
[0], ii
);
832 *gen
= btrfs_inode_generation(path
->nodes
[0], ii
);
834 *mode
= btrfs_inode_mode(path
->nodes
[0], ii
);
836 *uid
= btrfs_inode_uid(path
->nodes
[0], ii
);
838 *gid
= btrfs_inode_gid(path
->nodes
[0], ii
);
840 *rdev
= btrfs_inode_rdev(path
->nodes
[0], ii
);
845 static int get_inode_info(struct btrfs_root
*root
,
846 u64 ino
, u64
*size
, u64
*gen
,
847 u64
*mode
, u64
*uid
, u64
*gid
,
850 struct btrfs_path
*path
;
853 path
= alloc_path_for_send();
856 ret
= __get_inode_info(root
, path
, ino
, size
, gen
, mode
, uid
, gid
,
858 btrfs_free_path(path
);
862 typedef int (*iterate_inode_ref_t
)(int num
, u64 dir
, int index
,
867 * Helper function to iterate the entries in ONE btrfs_inode_ref or
868 * btrfs_inode_extref.
869 * The iterate callback may return a non zero value to stop iteration. This can
870 * be a negative value for error codes or 1 to simply stop it.
872 * path must point to the INODE_REF or INODE_EXTREF when called.
874 static int iterate_inode_ref(struct btrfs_root
*root
, struct btrfs_path
*path
,
875 struct btrfs_key
*found_key
, int resolve
,
876 iterate_inode_ref_t iterate
, void *ctx
)
878 struct extent_buffer
*eb
= path
->nodes
[0];
879 struct btrfs_item
*item
;
880 struct btrfs_inode_ref
*iref
;
881 struct btrfs_inode_extref
*extref
;
882 struct btrfs_path
*tmp_path
;
886 int slot
= path
->slots
[0];
893 unsigned long name_off
;
894 unsigned long elem_size
;
897 p
= fs_path_alloc_reversed();
901 tmp_path
= alloc_path_for_send();
908 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
909 ptr
= (unsigned long)btrfs_item_ptr(eb
, slot
,
910 struct btrfs_inode_ref
);
911 item
= btrfs_item_nr(slot
);
912 total
= btrfs_item_size(eb
, item
);
913 elem_size
= sizeof(*iref
);
915 ptr
= btrfs_item_ptr_offset(eb
, slot
);
916 total
= btrfs_item_size_nr(eb
, slot
);
917 elem_size
= sizeof(*extref
);
920 while (cur
< total
) {
923 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
924 iref
= (struct btrfs_inode_ref
*)(ptr
+ cur
);
925 name_len
= btrfs_inode_ref_name_len(eb
, iref
);
926 name_off
= (unsigned long)(iref
+ 1);
927 index
= btrfs_inode_ref_index(eb
, iref
);
928 dir
= found_key
->offset
;
930 extref
= (struct btrfs_inode_extref
*)(ptr
+ cur
);
931 name_len
= btrfs_inode_extref_name_len(eb
, extref
);
932 name_off
= (unsigned long)&extref
->name
;
933 index
= btrfs_inode_extref_index(eb
, extref
);
934 dir
= btrfs_inode_extref_parent(eb
, extref
);
938 start
= btrfs_ref_to_path(root
, tmp_path
, name_len
,
942 ret
= PTR_ERR(start
);
945 if (start
< p
->buf
) {
946 /* overflow , try again with larger buffer */
947 ret
= fs_path_ensure_buf(p
,
948 p
->buf_len
+ p
->buf
- start
);
951 start
= btrfs_ref_to_path(root
, tmp_path
,
956 ret
= PTR_ERR(start
);
959 BUG_ON(start
< p
->buf
);
963 ret
= fs_path_add_from_extent_buffer(p
, eb
, name_off
,
969 cur
+= elem_size
+ name_len
;
970 ret
= iterate(num
, dir
, index
, p
, ctx
);
977 btrfs_free_path(tmp_path
);
982 typedef int (*iterate_dir_item_t
)(int num
, struct btrfs_key
*di_key
,
983 const char *name
, int name_len
,
984 const char *data
, int data_len
,
988 * Helper function to iterate the entries in ONE btrfs_dir_item.
989 * The iterate callback may return a non zero value to stop iteration. This can
990 * be a negative value for error codes or 1 to simply stop it.
992 * path must point to the dir item when called.
994 static int iterate_dir_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
995 struct btrfs_key
*found_key
,
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
)
1275 if (ino
> bctx
->cur_objectid
)
1277 if (offset
+ bctx
->extent_len
> bctx
->cur_offset
)
1283 found
->found_refs
++;
1284 if (ino
< found
->ino
) {
1286 found
->offset
= offset
;
1287 } else if (found
->ino
== ino
) {
1289 * same extent found more then once in the same file.
1291 if (found
->offset
> offset
+ bctx
->extent_len
)
1292 found
->offset
= offset
;
1299 * Given an inode, offset and extent item, it finds a good clone for a clone
1300 * instruction. Returns -ENOENT when none could be found. The function makes
1301 * sure that the returned clone is usable at the point where sending is at the
1302 * moment. This means, that no clones are accepted which lie behind the current
1305 * path must point to the extent item when called.
1307 static int find_extent_clone(struct send_ctx
*sctx
,
1308 struct btrfs_path
*path
,
1309 u64 ino
, u64 data_offset
,
1311 struct clone_root
**found
)
1313 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
1319 u64 extent_item_pos
;
1321 struct btrfs_file_extent_item
*fi
;
1322 struct extent_buffer
*eb
= path
->nodes
[0];
1323 struct backref_ctx
*backref_ctx
= NULL
;
1324 struct clone_root
*cur_clone_root
;
1325 struct btrfs_key found_key
;
1326 struct btrfs_path
*tmp_path
;
1330 tmp_path
= alloc_path_for_send();
1334 /* We only use this path under the commit sem */
1335 tmp_path
->need_commit_sem
= 0;
1337 backref_ctx
= kmalloc(sizeof(*backref_ctx
), GFP_KERNEL
);
1343 backref_ctx
->path
= tmp_path
;
1345 if (data_offset
>= ino_size
) {
1347 * There may be extents that lie behind the file's size.
1348 * I at least had this in combination with snapshotting while
1349 * writing large files.
1355 fi
= btrfs_item_ptr(eb
, path
->slots
[0],
1356 struct btrfs_file_extent_item
);
1357 extent_type
= btrfs_file_extent_type(eb
, fi
);
1358 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1362 compressed
= btrfs_file_extent_compression(eb
, fi
);
1364 num_bytes
= btrfs_file_extent_num_bytes(eb
, fi
);
1365 disk_byte
= btrfs_file_extent_disk_bytenr(eb
, fi
);
1366 if (disk_byte
== 0) {
1370 logical
= disk_byte
+ btrfs_file_extent_offset(eb
, fi
);
1372 down_read(&fs_info
->commit_root_sem
);
1373 ret
= extent_from_logical(fs_info
, disk_byte
, tmp_path
,
1374 &found_key
, &flags
);
1375 up_read(&fs_info
->commit_root_sem
);
1376 btrfs_release_path(tmp_path
);
1380 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1386 * Setup the clone roots.
1388 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1389 cur_clone_root
= sctx
->clone_roots
+ i
;
1390 cur_clone_root
->ino
= (u64
)-1;
1391 cur_clone_root
->offset
= 0;
1392 cur_clone_root
->found_refs
= 0;
1395 backref_ctx
->sctx
= sctx
;
1396 backref_ctx
->found
= 0;
1397 backref_ctx
->cur_objectid
= ino
;
1398 backref_ctx
->cur_offset
= data_offset
;
1399 backref_ctx
->found_itself
= 0;
1400 backref_ctx
->extent_len
= num_bytes
;
1402 * For non-compressed extents iterate_extent_inodes() gives us extent
1403 * offsets that already take into account the data offset, but not for
1404 * compressed extents, since the offset is logical and not relative to
1405 * the physical extent locations. We must take this into account to
1406 * avoid sending clone offsets that go beyond the source file's size,
1407 * which would result in the clone ioctl failing with -EINVAL on the
1410 if (compressed
== BTRFS_COMPRESS_NONE
)
1411 backref_ctx
->data_offset
= 0;
1413 backref_ctx
->data_offset
= btrfs_file_extent_offset(eb
, fi
);
1416 * The last extent of a file may be too large due to page alignment.
1417 * We need to adjust extent_len in this case so that the checks in
1418 * __iterate_backrefs work.
1420 if (data_offset
+ num_bytes
>= ino_size
)
1421 backref_ctx
->extent_len
= ino_size
- data_offset
;
1424 * Now collect all backrefs.
1426 if (compressed
== BTRFS_COMPRESS_NONE
)
1427 extent_item_pos
= logical
- found_key
.objectid
;
1429 extent_item_pos
= 0;
1430 ret
= iterate_extent_inodes(fs_info
, found_key
.objectid
,
1431 extent_item_pos
, 1, __iterate_backrefs
,
1437 if (!backref_ctx
->found_itself
) {
1438 /* found a bug in backref code? */
1441 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1442 ino
, data_offset
, disk_byte
, found_key
.objectid
);
1446 btrfs_debug(fs_info
,
1447 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1448 data_offset
, ino
, num_bytes
, logical
);
1450 if (!backref_ctx
->found
)
1451 btrfs_debug(fs_info
, "no clones found");
1453 cur_clone_root
= NULL
;
1454 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1455 if (sctx
->clone_roots
[i
].found_refs
) {
1456 if (!cur_clone_root
)
1457 cur_clone_root
= sctx
->clone_roots
+ i
;
1458 else if (sctx
->clone_roots
[i
].root
== sctx
->send_root
)
1459 /* prefer clones from send_root over others */
1460 cur_clone_root
= sctx
->clone_roots
+ i
;
1465 if (cur_clone_root
) {
1466 *found
= cur_clone_root
;
1473 btrfs_free_path(tmp_path
);
1478 static int read_symlink(struct btrfs_root
*root
,
1480 struct fs_path
*dest
)
1483 struct btrfs_path
*path
;
1484 struct btrfs_key key
;
1485 struct btrfs_file_extent_item
*ei
;
1491 path
= alloc_path_for_send();
1496 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1498 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1503 * An empty symlink inode. Can happen in rare error paths when
1504 * creating a symlink (transaction committed before the inode
1505 * eviction handler removed the symlink inode items and a crash
1506 * happened in between or the subvol was snapshoted in between).
1507 * Print an informative message to dmesg/syslog so that the user
1508 * can delete the symlink.
1510 btrfs_err(root
->fs_info
,
1511 "Found empty symlink inode %llu at root %llu",
1512 ino
, root
->root_key
.objectid
);
1517 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1518 struct btrfs_file_extent_item
);
1519 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
1520 compression
= btrfs_file_extent_compression(path
->nodes
[0], ei
);
1521 BUG_ON(type
!= BTRFS_FILE_EXTENT_INLINE
);
1522 BUG_ON(compression
);
1524 off
= btrfs_file_extent_inline_start(ei
);
1525 len
= btrfs_file_extent_inline_len(path
->nodes
[0], path
->slots
[0], ei
);
1527 ret
= fs_path_add_from_extent_buffer(dest
, path
->nodes
[0], off
, len
);
1530 btrfs_free_path(path
);
1535 * Helper function to generate a file name that is unique in the root of
1536 * send_root and parent_root. This is used to generate names for orphan inodes.
1538 static int gen_unique_name(struct send_ctx
*sctx
,
1540 struct fs_path
*dest
)
1543 struct btrfs_path
*path
;
1544 struct btrfs_dir_item
*di
;
1549 path
= alloc_path_for_send();
1554 len
= snprintf(tmp
, sizeof(tmp
), "o%llu-%llu-%llu",
1556 ASSERT(len
< sizeof(tmp
));
1558 di
= btrfs_lookup_dir_item(NULL
, sctx
->send_root
,
1559 path
, BTRFS_FIRST_FREE_OBJECTID
,
1560 tmp
, strlen(tmp
), 0);
1561 btrfs_release_path(path
);
1567 /* not unique, try again */
1572 if (!sctx
->parent_root
) {
1578 di
= btrfs_lookup_dir_item(NULL
, sctx
->parent_root
,
1579 path
, BTRFS_FIRST_FREE_OBJECTID
,
1580 tmp
, strlen(tmp
), 0);
1581 btrfs_release_path(path
);
1587 /* not unique, try again */
1595 ret
= fs_path_add(dest
, tmp
, strlen(tmp
));
1598 btrfs_free_path(path
);
1603 inode_state_no_change
,
1604 inode_state_will_create
,
1605 inode_state_did_create
,
1606 inode_state_will_delete
,
1607 inode_state_did_delete
,
1610 static int get_cur_inode_state(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1618 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &left_gen
, NULL
, NULL
,
1620 if (ret
< 0 && ret
!= -ENOENT
)
1624 if (!sctx
->parent_root
) {
1625 right_ret
= -ENOENT
;
1627 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
, &right_gen
,
1628 NULL
, NULL
, NULL
, NULL
);
1629 if (ret
< 0 && ret
!= -ENOENT
)
1634 if (!left_ret
&& !right_ret
) {
1635 if (left_gen
== gen
&& right_gen
== gen
) {
1636 ret
= inode_state_no_change
;
1637 } else if (left_gen
== gen
) {
1638 if (ino
< sctx
->send_progress
)
1639 ret
= inode_state_did_create
;
1641 ret
= inode_state_will_create
;
1642 } else if (right_gen
== gen
) {
1643 if (ino
< sctx
->send_progress
)
1644 ret
= inode_state_did_delete
;
1646 ret
= inode_state_will_delete
;
1650 } else if (!left_ret
) {
1651 if (left_gen
== gen
) {
1652 if (ino
< sctx
->send_progress
)
1653 ret
= inode_state_did_create
;
1655 ret
= inode_state_will_create
;
1659 } else if (!right_ret
) {
1660 if (right_gen
== gen
) {
1661 if (ino
< sctx
->send_progress
)
1662 ret
= inode_state_did_delete
;
1664 ret
= inode_state_will_delete
;
1676 static int is_inode_existent(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1680 if (ino
== BTRFS_FIRST_FREE_OBJECTID
)
1683 ret
= get_cur_inode_state(sctx
, ino
, gen
);
1687 if (ret
== inode_state_no_change
||
1688 ret
== inode_state_did_create
||
1689 ret
== inode_state_will_delete
)
1699 * Helper function to lookup a dir item in a dir.
1701 static int lookup_dir_item_inode(struct btrfs_root
*root
,
1702 u64 dir
, const char *name
, int name_len
,
1707 struct btrfs_dir_item
*di
;
1708 struct btrfs_key key
;
1709 struct btrfs_path
*path
;
1711 path
= alloc_path_for_send();
1715 di
= btrfs_lookup_dir_item(NULL
, root
, path
,
1716 dir
, name
, name_len
, 0);
1725 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
1726 if (key
.type
== BTRFS_ROOT_ITEM_KEY
) {
1730 *found_inode
= key
.objectid
;
1731 *found_type
= btrfs_dir_type(path
->nodes
[0], di
);
1734 btrfs_free_path(path
);
1739 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1740 * generation of the parent dir and the name of the dir entry.
1742 static int get_first_ref(struct btrfs_root
*root
, u64 ino
,
1743 u64
*dir
, u64
*dir_gen
, struct fs_path
*name
)
1746 struct btrfs_key key
;
1747 struct btrfs_key found_key
;
1748 struct btrfs_path
*path
;
1752 path
= alloc_path_for_send();
1757 key
.type
= BTRFS_INODE_REF_KEY
;
1760 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
1764 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1766 if (ret
|| found_key
.objectid
!= ino
||
1767 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1768 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1773 if (found_key
.type
== BTRFS_INODE_REF_KEY
) {
1774 struct btrfs_inode_ref
*iref
;
1775 iref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1776 struct btrfs_inode_ref
);
1777 len
= btrfs_inode_ref_name_len(path
->nodes
[0], iref
);
1778 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1779 (unsigned long)(iref
+ 1),
1781 parent_dir
= found_key
.offset
;
1783 struct btrfs_inode_extref
*extref
;
1784 extref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1785 struct btrfs_inode_extref
);
1786 len
= btrfs_inode_extref_name_len(path
->nodes
[0], extref
);
1787 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1788 (unsigned long)&extref
->name
, len
);
1789 parent_dir
= btrfs_inode_extref_parent(path
->nodes
[0], extref
);
1793 btrfs_release_path(path
);
1796 ret
= get_inode_info(root
, parent_dir
, NULL
, dir_gen
, NULL
,
1805 btrfs_free_path(path
);
1809 static int is_first_ref(struct btrfs_root
*root
,
1811 const char *name
, int name_len
)
1814 struct fs_path
*tmp_name
;
1817 tmp_name
= fs_path_alloc();
1821 ret
= get_first_ref(root
, ino
, &tmp_dir
, NULL
, tmp_name
);
1825 if (dir
!= tmp_dir
|| name_len
!= fs_path_len(tmp_name
)) {
1830 ret
= !memcmp(tmp_name
->start
, name
, name_len
);
1833 fs_path_free(tmp_name
);
1838 * Used by process_recorded_refs to determine if a new ref would overwrite an
1839 * already existing ref. In case it detects an overwrite, it returns the
1840 * inode/gen in who_ino/who_gen.
1841 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1842 * to make sure later references to the overwritten inode are possible.
1843 * Orphanizing is however only required for the first ref of an inode.
1844 * process_recorded_refs does an additional is_first_ref check to see if
1845 * orphanizing is really required.
1847 static int will_overwrite_ref(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
1848 const char *name
, int name_len
,
1849 u64
*who_ino
, u64
*who_gen
, u64
*who_mode
)
1853 u64 other_inode
= 0;
1856 if (!sctx
->parent_root
)
1859 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1864 * If we have a parent root we need to verify that the parent dir was
1865 * not deleted and then re-created, if it was then we have no overwrite
1866 * and we can just unlink this entry.
1868 if (sctx
->parent_root
&& dir
!= BTRFS_FIRST_FREE_OBJECTID
) {
1869 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &gen
, NULL
,
1871 if (ret
< 0 && ret
!= -ENOENT
)
1881 ret
= lookup_dir_item_inode(sctx
->parent_root
, dir
, name
, name_len
,
1882 &other_inode
, &other_type
);
1883 if (ret
< 0 && ret
!= -ENOENT
)
1891 * Check if the overwritten ref was already processed. If yes, the ref
1892 * was already unlinked/moved, so we can safely assume that we will not
1893 * overwrite anything at this point in time.
1895 if (other_inode
> sctx
->send_progress
||
1896 is_waiting_for_move(sctx
, other_inode
)) {
1897 ret
= get_inode_info(sctx
->parent_root
, other_inode
, NULL
,
1898 who_gen
, who_mode
, NULL
, NULL
, NULL
);
1903 *who_ino
= other_inode
;
1913 * Checks if the ref was overwritten by an already processed inode. This is
1914 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1915 * thus the orphan name needs be used.
1916 * process_recorded_refs also uses it to avoid unlinking of refs that were
1919 static int did_overwrite_ref(struct send_ctx
*sctx
,
1920 u64 dir
, u64 dir_gen
,
1921 u64 ino
, u64 ino_gen
,
1922 const char *name
, int name_len
)
1929 if (!sctx
->parent_root
)
1932 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1936 if (dir
!= BTRFS_FIRST_FREE_OBJECTID
) {
1937 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &gen
, NULL
,
1939 if (ret
< 0 && ret
!= -ENOENT
)
1949 /* check if the ref was overwritten by another ref */
1950 ret
= lookup_dir_item_inode(sctx
->send_root
, dir
, name
, name_len
,
1951 &ow_inode
, &other_type
);
1952 if (ret
< 0 && ret
!= -ENOENT
)
1955 /* was never and will never be overwritten */
1960 ret
= get_inode_info(sctx
->send_root
, ow_inode
, NULL
, &gen
, NULL
, NULL
,
1965 if (ow_inode
== ino
&& gen
== ino_gen
) {
1971 * We know that it is or will be overwritten. Check this now.
1972 * The current inode being processed might have been the one that caused
1973 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1974 * the current inode being processed.
1976 if ((ow_inode
< sctx
->send_progress
) ||
1977 (ino
!= sctx
->cur_ino
&& ow_inode
== sctx
->cur_ino
&&
1978 gen
== sctx
->cur_inode_gen
))
1988 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1989 * that got overwritten. This is used by process_recorded_refs to determine
1990 * if it has to use the path as returned by get_cur_path or the orphan name.
1992 static int did_overwrite_first_ref(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1995 struct fs_path
*name
= NULL
;
1999 if (!sctx
->parent_root
)
2002 name
= fs_path_alloc();
2006 ret
= get_first_ref(sctx
->parent_root
, ino
, &dir
, &dir_gen
, name
);
2010 ret
= did_overwrite_ref(sctx
, dir
, dir_gen
, ino
, gen
,
2011 name
->start
, fs_path_len(name
));
2019 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2020 * so we need to do some special handling in case we have clashes. This function
2021 * takes care of this with the help of name_cache_entry::radix_list.
2022 * In case of error, nce is kfreed.
2024 static int name_cache_insert(struct send_ctx
*sctx
,
2025 struct name_cache_entry
*nce
)
2028 struct list_head
*nce_head
;
2030 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
2031 (unsigned long)nce
->ino
);
2033 nce_head
= kmalloc(sizeof(*nce_head
), GFP_KERNEL
);
2038 INIT_LIST_HEAD(nce_head
);
2040 ret
= radix_tree_insert(&sctx
->name_cache
, nce
->ino
, nce_head
);
2047 list_add_tail(&nce
->radix_list
, nce_head
);
2048 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
2049 sctx
->name_cache_size
++;
2054 static void name_cache_delete(struct send_ctx
*sctx
,
2055 struct name_cache_entry
*nce
)
2057 struct list_head
*nce_head
;
2059 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
2060 (unsigned long)nce
->ino
);
2062 btrfs_err(sctx
->send_root
->fs_info
,
2063 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2064 nce
->ino
, sctx
->name_cache_size
);
2067 list_del(&nce
->radix_list
);
2068 list_del(&nce
->list
);
2069 sctx
->name_cache_size
--;
2072 * We may not get to the final release of nce_head if the lookup fails
2074 if (nce_head
&& list_empty(nce_head
)) {
2075 radix_tree_delete(&sctx
->name_cache
, (unsigned long)nce
->ino
);
2080 static struct name_cache_entry
*name_cache_search(struct send_ctx
*sctx
,
2083 struct list_head
*nce_head
;
2084 struct name_cache_entry
*cur
;
2086 nce_head
= radix_tree_lookup(&sctx
->name_cache
, (unsigned long)ino
);
2090 list_for_each_entry(cur
, nce_head
, radix_list
) {
2091 if (cur
->ino
== ino
&& cur
->gen
== gen
)
2098 * Removes the entry from the list and adds it back to the end. This marks the
2099 * entry as recently used so that name_cache_clean_unused does not remove it.
2101 static void name_cache_used(struct send_ctx
*sctx
, struct name_cache_entry
*nce
)
2103 list_del(&nce
->list
);
2104 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
2108 * Remove some entries from the beginning of name_cache_list.
2110 static void name_cache_clean_unused(struct send_ctx
*sctx
)
2112 struct name_cache_entry
*nce
;
2114 if (sctx
->name_cache_size
< SEND_CTX_NAME_CACHE_CLEAN_SIZE
)
2117 while (sctx
->name_cache_size
> SEND_CTX_MAX_NAME_CACHE_SIZE
) {
2118 nce
= list_entry(sctx
->name_cache_list
.next
,
2119 struct name_cache_entry
, list
);
2120 name_cache_delete(sctx
, nce
);
2125 static void name_cache_free(struct send_ctx
*sctx
)
2127 struct name_cache_entry
*nce
;
2129 while (!list_empty(&sctx
->name_cache_list
)) {
2130 nce
= list_entry(sctx
->name_cache_list
.next
,
2131 struct name_cache_entry
, list
);
2132 name_cache_delete(sctx
, nce
);
2138 * Used by get_cur_path for each ref up to the root.
2139 * Returns 0 if it succeeded.
2140 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2141 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2142 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2143 * Returns <0 in case of error.
2145 static int __get_cur_name_and_parent(struct send_ctx
*sctx
,
2149 struct fs_path
*dest
)
2153 struct name_cache_entry
*nce
= NULL
;
2156 * First check if we already did a call to this function with the same
2157 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2158 * return the cached result.
2160 nce
= name_cache_search(sctx
, ino
, gen
);
2162 if (ino
< sctx
->send_progress
&& nce
->need_later_update
) {
2163 name_cache_delete(sctx
, nce
);
2167 name_cache_used(sctx
, nce
);
2168 *parent_ino
= nce
->parent_ino
;
2169 *parent_gen
= nce
->parent_gen
;
2170 ret
= fs_path_add(dest
, nce
->name
, nce
->name_len
);
2179 * If the inode is not existent yet, add the orphan name and return 1.
2180 * This should only happen for the parent dir that we determine in
2183 ret
= is_inode_existent(sctx
, ino
, gen
);
2188 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2196 * Depending on whether the inode was already processed or not, use
2197 * send_root or parent_root for ref lookup.
2199 if (ino
< sctx
->send_progress
)
2200 ret
= get_first_ref(sctx
->send_root
, ino
,
2201 parent_ino
, parent_gen
, dest
);
2203 ret
= get_first_ref(sctx
->parent_root
, ino
,
2204 parent_ino
, parent_gen
, dest
);
2209 * Check if the ref was overwritten by an inode's ref that was processed
2210 * earlier. If yes, treat as orphan and return 1.
2212 ret
= did_overwrite_ref(sctx
, *parent_ino
, *parent_gen
, ino
, gen
,
2213 dest
->start
, dest
->end
- dest
->start
);
2217 fs_path_reset(dest
);
2218 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2226 * Store the result of the lookup in the name cache.
2228 nce
= kmalloc(sizeof(*nce
) + fs_path_len(dest
) + 1, GFP_KERNEL
);
2236 nce
->parent_ino
= *parent_ino
;
2237 nce
->parent_gen
= *parent_gen
;
2238 nce
->name_len
= fs_path_len(dest
);
2240 strcpy(nce
->name
, dest
->start
);
2242 if (ino
< sctx
->send_progress
)
2243 nce
->need_later_update
= 0;
2245 nce
->need_later_update
= 1;
2247 nce_ret
= name_cache_insert(sctx
, nce
);
2250 name_cache_clean_unused(sctx
);
2257 * Magic happens here. This function returns the first ref to an inode as it
2258 * would look like while receiving the stream at this point in time.
2259 * We walk the path up to the root. For every inode in between, we check if it
2260 * was already processed/sent. If yes, we continue with the parent as found
2261 * in send_root. If not, we continue with the parent as found in parent_root.
2262 * If we encounter an inode that was deleted at this point in time, we use the
2263 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2264 * that were not created yet and overwritten inodes/refs.
2266 * When do we have have orphan inodes:
2267 * 1. When an inode is freshly created and thus no valid refs are available yet
2268 * 2. When a directory lost all it's refs (deleted) but still has dir items
2269 * inside which were not processed yet (pending for move/delete). If anyone
2270 * tried to get the path to the dir items, it would get a path inside that
2272 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2273 * of an unprocessed inode. If in that case the first ref would be
2274 * overwritten, the overwritten inode gets "orphanized". Later when we
2275 * process this overwritten inode, it is restored at a new place by moving
2278 * sctx->send_progress tells this function at which point in time receiving
2281 static int get_cur_path(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2282 struct fs_path
*dest
)
2285 struct fs_path
*name
= NULL
;
2286 u64 parent_inode
= 0;
2290 name
= fs_path_alloc();
2297 fs_path_reset(dest
);
2299 while (!stop
&& ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
2300 struct waiting_dir_move
*wdm
;
2302 fs_path_reset(name
);
2304 if (is_waiting_for_rm(sctx
, ino
)) {
2305 ret
= gen_unique_name(sctx
, ino
, gen
, name
);
2308 ret
= fs_path_add_path(dest
, name
);
2312 wdm
= get_waiting_dir_move(sctx
, ino
);
2313 if (wdm
&& wdm
->orphanized
) {
2314 ret
= gen_unique_name(sctx
, ino
, gen
, name
);
2317 ret
= get_first_ref(sctx
->parent_root
, ino
,
2318 &parent_inode
, &parent_gen
, name
);
2320 ret
= __get_cur_name_and_parent(sctx
, ino
, gen
,
2330 ret
= fs_path_add_path(dest
, name
);
2341 fs_path_unreverse(dest
);
2346 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2348 static int send_subvol_begin(struct send_ctx
*sctx
)
2351 struct btrfs_root
*send_root
= sctx
->send_root
;
2352 struct btrfs_root
*parent_root
= sctx
->parent_root
;
2353 struct btrfs_path
*path
;
2354 struct btrfs_key key
;
2355 struct btrfs_root_ref
*ref
;
2356 struct extent_buffer
*leaf
;
2360 path
= btrfs_alloc_path();
2364 name
= kmalloc(BTRFS_PATH_NAME_MAX
, GFP_KERNEL
);
2366 btrfs_free_path(path
);
2370 key
.objectid
= send_root
->objectid
;
2371 key
.type
= BTRFS_ROOT_BACKREF_KEY
;
2374 ret
= btrfs_search_slot_for_read(send_root
->fs_info
->tree_root
,
2383 leaf
= path
->nodes
[0];
2384 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2385 if (key
.type
!= BTRFS_ROOT_BACKREF_KEY
||
2386 key
.objectid
!= send_root
->objectid
) {
2390 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
2391 namelen
= btrfs_root_ref_name_len(leaf
, ref
);
2392 read_extent_buffer(leaf
, name
, (unsigned long)(ref
+ 1), namelen
);
2393 btrfs_release_path(path
);
2396 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SNAPSHOT
);
2400 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SUBVOL
);
2405 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_PATH
, name
, namelen
);
2407 if (!btrfs_is_empty_uuid(sctx
->send_root
->root_item
.received_uuid
))
2408 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_UUID
,
2409 sctx
->send_root
->root_item
.received_uuid
);
2411 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_UUID
,
2412 sctx
->send_root
->root_item
.uuid
);
2414 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CTRANSID
,
2415 le64_to_cpu(sctx
->send_root
->root_item
.ctransid
));
2417 if (!btrfs_is_empty_uuid(parent_root
->root_item
.received_uuid
))
2418 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2419 parent_root
->root_item
.received_uuid
);
2421 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2422 parent_root
->root_item
.uuid
);
2423 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
2424 le64_to_cpu(sctx
->parent_root
->root_item
.ctransid
));
2427 ret
= send_cmd(sctx
);
2431 btrfs_free_path(path
);
2436 static int send_truncate(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 size
)
2438 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2442 btrfs_debug(fs_info
, "send_truncate %llu size=%llu", ino
, size
);
2444 p
= fs_path_alloc();
2448 ret
= begin_cmd(sctx
, BTRFS_SEND_C_TRUNCATE
);
2452 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2455 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2456 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, size
);
2458 ret
= send_cmd(sctx
);
2466 static int send_chmod(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 mode
)
2468 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2472 btrfs_debug(fs_info
, "send_chmod %llu mode=%llu", ino
, mode
);
2474 p
= fs_path_alloc();
2478 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHMOD
);
2482 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2485 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2486 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
& 07777);
2488 ret
= send_cmd(sctx
);
2496 static int send_chown(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 uid
, u64 gid
)
2498 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2502 btrfs_debug(fs_info
, "send_chown %llu uid=%llu, gid=%llu",
2505 p
= fs_path_alloc();
2509 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHOWN
);
2513 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2516 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2517 TLV_PUT_U64(sctx
, BTRFS_SEND_A_UID
, uid
);
2518 TLV_PUT_U64(sctx
, BTRFS_SEND_A_GID
, gid
);
2520 ret
= send_cmd(sctx
);
2528 static int send_utimes(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
2530 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2532 struct fs_path
*p
= NULL
;
2533 struct btrfs_inode_item
*ii
;
2534 struct btrfs_path
*path
= NULL
;
2535 struct extent_buffer
*eb
;
2536 struct btrfs_key key
;
2539 btrfs_debug(fs_info
, "send_utimes %llu", ino
);
2541 p
= fs_path_alloc();
2545 path
= alloc_path_for_send();
2552 key
.type
= BTRFS_INODE_ITEM_KEY
;
2554 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2560 eb
= path
->nodes
[0];
2561 slot
= path
->slots
[0];
2562 ii
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
2564 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UTIMES
);
2568 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2571 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2572 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_ATIME
, eb
, &ii
->atime
);
2573 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_MTIME
, eb
, &ii
->mtime
);
2574 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_CTIME
, eb
, &ii
->ctime
);
2575 /* TODO Add otime support when the otime patches get into upstream */
2577 ret
= send_cmd(sctx
);
2582 btrfs_free_path(path
);
2587 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2588 * a valid path yet because we did not process the refs yet. So, the inode
2589 * is created as orphan.
2591 static int send_create_inode(struct send_ctx
*sctx
, u64 ino
)
2593 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
2601 btrfs_debug(fs_info
, "send_create_inode %llu", ino
);
2603 p
= fs_path_alloc();
2607 if (ino
!= sctx
->cur_ino
) {
2608 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &gen
, &mode
,
2613 gen
= sctx
->cur_inode_gen
;
2614 mode
= sctx
->cur_inode_mode
;
2615 rdev
= sctx
->cur_inode_rdev
;
2618 if (S_ISREG(mode
)) {
2619 cmd
= BTRFS_SEND_C_MKFILE
;
2620 } else if (S_ISDIR(mode
)) {
2621 cmd
= BTRFS_SEND_C_MKDIR
;
2622 } else if (S_ISLNK(mode
)) {
2623 cmd
= BTRFS_SEND_C_SYMLINK
;
2624 } else if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
2625 cmd
= BTRFS_SEND_C_MKNOD
;
2626 } else if (S_ISFIFO(mode
)) {
2627 cmd
= BTRFS_SEND_C_MKFIFO
;
2628 } else if (S_ISSOCK(mode
)) {
2629 cmd
= BTRFS_SEND_C_MKSOCK
;
2631 btrfs_warn(sctx
->send_root
->fs_info
, "unexpected inode type %o",
2632 (int)(mode
& S_IFMT
));
2637 ret
= begin_cmd(sctx
, cmd
);
2641 ret
= gen_unique_name(sctx
, ino
, gen
, p
);
2645 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2646 TLV_PUT_U64(sctx
, BTRFS_SEND_A_INO
, ino
);
2648 if (S_ISLNK(mode
)) {
2650 ret
= read_symlink(sctx
->send_root
, ino
, p
);
2653 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, p
);
2654 } else if (S_ISCHR(mode
) || S_ISBLK(mode
) ||
2655 S_ISFIFO(mode
) || S_ISSOCK(mode
)) {
2656 TLV_PUT_U64(sctx
, BTRFS_SEND_A_RDEV
, new_encode_dev(rdev
));
2657 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
);
2660 ret
= send_cmd(sctx
);
2672 * We need some special handling for inodes that get processed before the parent
2673 * directory got created. See process_recorded_refs for details.
2674 * This function does the check if we already created the dir out of order.
2676 static int did_create_dir(struct send_ctx
*sctx
, u64 dir
)
2679 struct btrfs_path
*path
= NULL
;
2680 struct btrfs_key key
;
2681 struct btrfs_key found_key
;
2682 struct btrfs_key di_key
;
2683 struct extent_buffer
*eb
;
2684 struct btrfs_dir_item
*di
;
2687 path
= alloc_path_for_send();
2694 key
.type
= BTRFS_DIR_INDEX_KEY
;
2696 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2701 eb
= path
->nodes
[0];
2702 slot
= path
->slots
[0];
2703 if (slot
>= btrfs_header_nritems(eb
)) {
2704 ret
= btrfs_next_leaf(sctx
->send_root
, path
);
2707 } else if (ret
> 0) {
2714 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
2715 if (found_key
.objectid
!= key
.objectid
||
2716 found_key
.type
!= key
.type
) {
2721 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
2722 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
2724 if (di_key
.type
!= BTRFS_ROOT_ITEM_KEY
&&
2725 di_key
.objectid
< sctx
->send_progress
) {
2734 btrfs_free_path(path
);
2739 * Only creates the inode if it is:
2740 * 1. Not a directory
2741 * 2. Or a directory which was not created already due to out of order
2742 * directories. See did_create_dir and process_recorded_refs for details.
2744 static int send_create_inode_if_needed(struct send_ctx
*sctx
)
2748 if (S_ISDIR(sctx
->cur_inode_mode
)) {
2749 ret
= did_create_dir(sctx
, sctx
->cur_ino
);
2758 ret
= send_create_inode(sctx
, sctx
->cur_ino
);
2766 struct recorded_ref
{
2767 struct list_head list
;
2769 struct fs_path
*full_path
;
2775 static void set_ref_path(struct recorded_ref
*ref
, struct fs_path
*path
)
2777 ref
->full_path
= path
;
2778 ref
->name
= (char *)kbasename(ref
->full_path
->start
);
2779 ref
->name_len
= ref
->full_path
->end
- ref
->name
;
2783 * We need to process new refs before deleted refs, but compare_tree gives us
2784 * everything mixed. So we first record all refs and later process them.
2785 * This function is a helper to record one ref.
2787 static int __record_ref(struct list_head
*head
, u64 dir
,
2788 u64 dir_gen
, struct fs_path
*path
)
2790 struct recorded_ref
*ref
;
2792 ref
= kmalloc(sizeof(*ref
), GFP_KERNEL
);
2797 ref
->dir_gen
= dir_gen
;
2798 set_ref_path(ref
, path
);
2799 list_add_tail(&ref
->list
, head
);
2803 static int dup_ref(struct recorded_ref
*ref
, struct list_head
*list
)
2805 struct recorded_ref
*new;
2807 new = kmalloc(sizeof(*ref
), GFP_KERNEL
);
2811 new->dir
= ref
->dir
;
2812 new->dir_gen
= ref
->dir_gen
;
2813 new->full_path
= NULL
;
2814 INIT_LIST_HEAD(&new->list
);
2815 list_add_tail(&new->list
, list
);
2819 static void __free_recorded_refs(struct list_head
*head
)
2821 struct recorded_ref
*cur
;
2823 while (!list_empty(head
)) {
2824 cur
= list_entry(head
->next
, struct recorded_ref
, list
);
2825 fs_path_free(cur
->full_path
);
2826 list_del(&cur
->list
);
2831 static void free_recorded_refs(struct send_ctx
*sctx
)
2833 __free_recorded_refs(&sctx
->new_refs
);
2834 __free_recorded_refs(&sctx
->deleted_refs
);
2838 * Renames/moves a file/dir to its orphan name. Used when the first
2839 * ref of an unprocessed inode gets overwritten and for all non empty
2842 static int orphanize_inode(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2843 struct fs_path
*path
)
2846 struct fs_path
*orphan
;
2848 orphan
= fs_path_alloc();
2852 ret
= gen_unique_name(sctx
, ino
, gen
, orphan
);
2856 ret
= send_rename(sctx
, path
, orphan
);
2859 fs_path_free(orphan
);
2863 static struct orphan_dir_info
*
2864 add_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2866 struct rb_node
**p
= &sctx
->orphan_dirs
.rb_node
;
2867 struct rb_node
*parent
= NULL
;
2868 struct orphan_dir_info
*entry
, *odi
;
2870 odi
= kmalloc(sizeof(*odi
), GFP_KERNEL
);
2872 return ERR_PTR(-ENOMEM
);
2878 entry
= rb_entry(parent
, struct orphan_dir_info
, node
);
2879 if (dir_ino
< entry
->ino
) {
2881 } else if (dir_ino
> entry
->ino
) {
2882 p
= &(*p
)->rb_right
;
2889 rb_link_node(&odi
->node
, parent
, p
);
2890 rb_insert_color(&odi
->node
, &sctx
->orphan_dirs
);
2894 static struct orphan_dir_info
*
2895 get_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2897 struct rb_node
*n
= sctx
->orphan_dirs
.rb_node
;
2898 struct orphan_dir_info
*entry
;
2901 entry
= rb_entry(n
, struct orphan_dir_info
, node
);
2902 if (dir_ino
< entry
->ino
)
2904 else if (dir_ino
> entry
->ino
)
2912 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
)
2914 struct orphan_dir_info
*odi
= get_orphan_dir_info(sctx
, dir_ino
);
2919 static void free_orphan_dir_info(struct send_ctx
*sctx
,
2920 struct orphan_dir_info
*odi
)
2924 rb_erase(&odi
->node
, &sctx
->orphan_dirs
);
2929 * Returns 1 if a directory can be removed at this point in time.
2930 * We check this by iterating all dir items and checking if the inode behind
2931 * the dir item was already processed.
2933 static int can_rmdir(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
2937 struct btrfs_root
*root
= sctx
->parent_root
;
2938 struct btrfs_path
*path
;
2939 struct btrfs_key key
;
2940 struct btrfs_key found_key
;
2941 struct btrfs_key loc
;
2942 struct btrfs_dir_item
*di
;
2945 * Don't try to rmdir the top/root subvolume dir.
2947 if (dir
== BTRFS_FIRST_FREE_OBJECTID
)
2950 path
= alloc_path_for_send();
2955 key
.type
= BTRFS_DIR_INDEX_KEY
;
2957 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2962 struct waiting_dir_move
*dm
;
2964 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
2965 ret
= btrfs_next_leaf(root
, path
);
2972 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2974 if (found_key
.objectid
!= key
.objectid
||
2975 found_key
.type
!= key
.type
)
2978 di
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2979 struct btrfs_dir_item
);
2980 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &loc
);
2982 dm
= get_waiting_dir_move(sctx
, loc
.objectid
);
2984 struct orphan_dir_info
*odi
;
2986 odi
= add_orphan_dir_info(sctx
, dir
);
2992 dm
->rmdir_ino
= dir
;
2997 if (loc
.objectid
> send_progress
) {
2998 struct orphan_dir_info
*odi
;
3000 odi
= get_orphan_dir_info(sctx
, dir
);
3001 free_orphan_dir_info(sctx
, odi
);
3012 btrfs_free_path(path
);
3016 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
)
3018 struct waiting_dir_move
*entry
= get_waiting_dir_move(sctx
, ino
);
3020 return entry
!= NULL
;
3023 static int add_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
, bool orphanized
)
3025 struct rb_node
**p
= &sctx
->waiting_dir_moves
.rb_node
;
3026 struct rb_node
*parent
= NULL
;
3027 struct waiting_dir_move
*entry
, *dm
;
3029 dm
= kmalloc(sizeof(*dm
), GFP_KERNEL
);
3034 dm
->orphanized
= orphanized
;
3038 entry
= rb_entry(parent
, struct waiting_dir_move
, node
);
3039 if (ino
< entry
->ino
) {
3041 } else if (ino
> entry
->ino
) {
3042 p
= &(*p
)->rb_right
;
3049 rb_link_node(&dm
->node
, parent
, p
);
3050 rb_insert_color(&dm
->node
, &sctx
->waiting_dir_moves
);
3054 static struct waiting_dir_move
*
3055 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
)
3057 struct rb_node
*n
= sctx
->waiting_dir_moves
.rb_node
;
3058 struct waiting_dir_move
*entry
;
3061 entry
= rb_entry(n
, struct waiting_dir_move
, node
);
3062 if (ino
< entry
->ino
)
3064 else if (ino
> entry
->ino
)
3072 static void free_waiting_dir_move(struct send_ctx
*sctx
,
3073 struct waiting_dir_move
*dm
)
3077 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
3081 static int add_pending_dir_move(struct send_ctx
*sctx
,
3085 struct list_head
*new_refs
,
3086 struct list_head
*deleted_refs
,
3087 const bool is_orphan
)
3089 struct rb_node
**p
= &sctx
->pending_dir_moves
.rb_node
;
3090 struct rb_node
*parent
= NULL
;
3091 struct pending_dir_move
*entry
= NULL
, *pm
;
3092 struct recorded_ref
*cur
;
3096 pm
= kmalloc(sizeof(*pm
), GFP_KERNEL
);
3099 pm
->parent_ino
= parent_ino
;
3102 INIT_LIST_HEAD(&pm
->list
);
3103 INIT_LIST_HEAD(&pm
->update_refs
);
3104 RB_CLEAR_NODE(&pm
->node
);
3108 entry
= rb_entry(parent
, struct pending_dir_move
, node
);
3109 if (parent_ino
< entry
->parent_ino
) {
3111 } else if (parent_ino
> entry
->parent_ino
) {
3112 p
= &(*p
)->rb_right
;
3119 list_for_each_entry(cur
, deleted_refs
, list
) {
3120 ret
= dup_ref(cur
, &pm
->update_refs
);
3124 list_for_each_entry(cur
, new_refs
, list
) {
3125 ret
= dup_ref(cur
, &pm
->update_refs
);
3130 ret
= add_waiting_dir_move(sctx
, pm
->ino
, is_orphan
);
3135 list_add_tail(&pm
->list
, &entry
->list
);
3137 rb_link_node(&pm
->node
, parent
, p
);
3138 rb_insert_color(&pm
->node
, &sctx
->pending_dir_moves
);
3143 __free_recorded_refs(&pm
->update_refs
);
3149 static struct pending_dir_move
*get_pending_dir_moves(struct send_ctx
*sctx
,
3152 struct rb_node
*n
= sctx
->pending_dir_moves
.rb_node
;
3153 struct pending_dir_move
*entry
;
3156 entry
= rb_entry(n
, struct pending_dir_move
, node
);
3157 if (parent_ino
< entry
->parent_ino
)
3159 else if (parent_ino
> entry
->parent_ino
)
3167 static int path_loop(struct send_ctx
*sctx
, struct fs_path
*name
,
3168 u64 ino
, u64 gen
, u64
*ancestor_ino
)
3171 u64 parent_inode
= 0;
3173 u64 start_ino
= ino
;
3176 while (ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
3177 fs_path_reset(name
);
3179 if (is_waiting_for_rm(sctx
, ino
))
3181 if (is_waiting_for_move(sctx
, ino
)) {
3182 if (*ancestor_ino
== 0)
3183 *ancestor_ino
= ino
;
3184 ret
= get_first_ref(sctx
->parent_root
, ino
,
3185 &parent_inode
, &parent_gen
, name
);
3187 ret
= __get_cur_name_and_parent(sctx
, ino
, gen
,
3197 if (parent_inode
== start_ino
) {
3199 if (*ancestor_ino
== 0)
3200 *ancestor_ino
= ino
;
3209 static int apply_dir_move(struct send_ctx
*sctx
, struct pending_dir_move
*pm
)
3211 struct fs_path
*from_path
= NULL
;
3212 struct fs_path
*to_path
= NULL
;
3213 struct fs_path
*name
= NULL
;
3214 u64 orig_progress
= sctx
->send_progress
;
3215 struct recorded_ref
*cur
;
3216 u64 parent_ino
, parent_gen
;
3217 struct waiting_dir_move
*dm
= NULL
;
3223 name
= fs_path_alloc();
3224 from_path
= fs_path_alloc();
3225 if (!name
|| !from_path
) {
3230 dm
= get_waiting_dir_move(sctx
, pm
->ino
);
3232 rmdir_ino
= dm
->rmdir_ino
;
3233 is_orphan
= dm
->orphanized
;
3234 free_waiting_dir_move(sctx
, dm
);
3237 ret
= gen_unique_name(sctx
, pm
->ino
,
3238 pm
->gen
, from_path
);
3240 ret
= get_first_ref(sctx
->parent_root
, pm
->ino
,
3241 &parent_ino
, &parent_gen
, name
);
3244 ret
= get_cur_path(sctx
, parent_ino
, parent_gen
,
3248 ret
= fs_path_add_path(from_path
, name
);
3253 sctx
->send_progress
= sctx
->cur_ino
+ 1;
3254 ret
= path_loop(sctx
, name
, pm
->ino
, pm
->gen
, &ancestor
);
3258 LIST_HEAD(deleted_refs
);
3259 ASSERT(ancestor
> BTRFS_FIRST_FREE_OBJECTID
);
3260 ret
= add_pending_dir_move(sctx
, pm
->ino
, pm
->gen
, ancestor
,
3261 &pm
->update_refs
, &deleted_refs
,
3266 dm
= get_waiting_dir_move(sctx
, pm
->ino
);
3268 dm
->rmdir_ino
= rmdir_ino
;
3272 fs_path_reset(name
);
3275 ret
= get_cur_path(sctx
, pm
->ino
, pm
->gen
, to_path
);
3279 ret
= send_rename(sctx
, from_path
, to_path
);
3284 struct orphan_dir_info
*odi
;
3286 odi
= get_orphan_dir_info(sctx
, rmdir_ino
);
3288 /* already deleted */
3291 ret
= can_rmdir(sctx
, rmdir_ino
, odi
->gen
, sctx
->cur_ino
);
3297 name
= fs_path_alloc();
3302 ret
= get_cur_path(sctx
, rmdir_ino
, odi
->gen
, name
);
3305 ret
= send_rmdir(sctx
, name
);
3308 free_orphan_dir_info(sctx
, odi
);
3312 ret
= send_utimes(sctx
, pm
->ino
, pm
->gen
);
3317 * After rename/move, need to update the utimes of both new parent(s)
3318 * and old parent(s).
3320 list_for_each_entry(cur
, &pm
->update_refs
, list
) {
3322 * The parent inode might have been deleted in the send snapshot
3324 ret
= get_inode_info(sctx
->send_root
, cur
->dir
, NULL
,
3325 NULL
, NULL
, NULL
, NULL
, NULL
);
3326 if (ret
== -ENOENT
) {
3333 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
3340 fs_path_free(from_path
);
3341 fs_path_free(to_path
);
3342 sctx
->send_progress
= orig_progress
;
3347 static void free_pending_move(struct send_ctx
*sctx
, struct pending_dir_move
*m
)
3349 if (!list_empty(&m
->list
))
3351 if (!RB_EMPTY_NODE(&m
->node
))
3352 rb_erase(&m
->node
, &sctx
->pending_dir_moves
);
3353 __free_recorded_refs(&m
->update_refs
);
3357 static void tail_append_pending_moves(struct pending_dir_move
*moves
,
3358 struct list_head
*stack
)
3360 if (list_empty(&moves
->list
)) {
3361 list_add_tail(&moves
->list
, stack
);
3364 list_splice_init(&moves
->list
, &list
);
3365 list_add_tail(&moves
->list
, stack
);
3366 list_splice_tail(&list
, stack
);
3370 static int apply_children_dir_moves(struct send_ctx
*sctx
)
3372 struct pending_dir_move
*pm
;
3373 struct list_head stack
;
3374 u64 parent_ino
= sctx
->cur_ino
;
3377 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3381 INIT_LIST_HEAD(&stack
);
3382 tail_append_pending_moves(pm
, &stack
);
3384 while (!list_empty(&stack
)) {
3385 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3386 parent_ino
= pm
->ino
;
3387 ret
= apply_dir_move(sctx
, pm
);
3388 free_pending_move(sctx
, pm
);
3391 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3393 tail_append_pending_moves(pm
, &stack
);
3398 while (!list_empty(&stack
)) {
3399 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3400 free_pending_move(sctx
, pm
);
3406 * We might need to delay a directory rename even when no ancestor directory
3407 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3408 * renamed. This happens when we rename a directory to the old name (the name
3409 * in the parent root) of some other unrelated directory that got its rename
3410 * delayed due to some ancestor with higher number that got renamed.
3416 * |---- a/ (ino 257)
3417 * | |---- file (ino 260)
3419 * |---- b/ (ino 258)
3420 * |---- c/ (ino 259)
3424 * |---- a/ (ino 258)
3425 * |---- x/ (ino 259)
3426 * |---- y/ (ino 257)
3427 * |----- file (ino 260)
3429 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3430 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3431 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3434 * 1 - rename 259 from 'c' to 'x'
3435 * 2 - rename 257 from 'a' to 'x/y'
3436 * 3 - rename 258 from 'b' to 'a'
3438 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3439 * be done right away and < 0 on error.
3441 static int wait_for_dest_dir_move(struct send_ctx
*sctx
,
3442 struct recorded_ref
*parent_ref
,
3443 const bool is_orphan
)
3445 struct btrfs_fs_info
*fs_info
= sctx
->parent_root
->fs_info
;
3446 struct btrfs_path
*path
;
3447 struct btrfs_key key
;
3448 struct btrfs_key di_key
;
3449 struct btrfs_dir_item
*di
;
3453 struct waiting_dir_move
*wdm
;
3455 if (RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
))
3458 path
= alloc_path_for_send();
3462 key
.objectid
= parent_ref
->dir
;
3463 key
.type
= BTRFS_DIR_ITEM_KEY
;
3464 key
.offset
= btrfs_name_hash(parent_ref
->name
, parent_ref
->name_len
);
3466 ret
= btrfs_search_slot(NULL
, sctx
->parent_root
, &key
, path
, 0, 0);
3469 } else if (ret
> 0) {
3474 di
= btrfs_match_dir_item_name(fs_info
, path
, parent_ref
->name
,
3475 parent_ref
->name_len
);
3481 * di_key.objectid has the number of the inode that has a dentry in the
3482 * parent directory with the same name that sctx->cur_ino is being
3483 * renamed to. We need to check if that inode is in the send root as
3484 * well and if it is currently marked as an inode with a pending rename,
3485 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3486 * that it happens after that other inode is renamed.
3488 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &di_key
);
3489 if (di_key
.type
!= BTRFS_INODE_ITEM_KEY
) {
3494 ret
= get_inode_info(sctx
->parent_root
, di_key
.objectid
, NULL
,
3495 &left_gen
, NULL
, NULL
, NULL
, NULL
);
3498 ret
= get_inode_info(sctx
->send_root
, di_key
.objectid
, NULL
,
3499 &right_gen
, NULL
, NULL
, NULL
, NULL
);
3506 /* Different inode, no need to delay the rename of sctx->cur_ino */
3507 if (right_gen
!= left_gen
) {
3512 wdm
= get_waiting_dir_move(sctx
, di_key
.objectid
);
3513 if (wdm
&& !wdm
->orphanized
) {
3514 ret
= add_pending_dir_move(sctx
,
3516 sctx
->cur_inode_gen
,
3519 &sctx
->deleted_refs
,
3525 btrfs_free_path(path
);
3530 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3531 * Return 1 if true, 0 if false and < 0 on error.
3533 static int is_ancestor(struct btrfs_root
*root
,
3537 struct fs_path
*fs_path
)
3540 bool free_path
= false;
3544 fs_path
= fs_path_alloc();
3550 while (ino
> BTRFS_FIRST_FREE_OBJECTID
) {
3554 fs_path_reset(fs_path
);
3555 ret
= get_first_ref(root
, ino
, &parent
, &parent_gen
, fs_path
);
3557 if (ret
== -ENOENT
&& ino
== ino2
)
3561 if (parent
== ino1
) {
3562 ret
= parent_gen
== ino1_gen
? 1 : 0;
3569 fs_path_free(fs_path
);
3573 static int wait_for_parent_move(struct send_ctx
*sctx
,
3574 struct recorded_ref
*parent_ref
,
3575 const bool is_orphan
)
3578 u64 ino
= parent_ref
->dir
;
3579 u64 ino_gen
= parent_ref
->dir_gen
;
3580 u64 parent_ino_before
, parent_ino_after
;
3581 struct fs_path
*path_before
= NULL
;
3582 struct fs_path
*path_after
= NULL
;
3585 path_after
= fs_path_alloc();
3586 path_before
= fs_path_alloc();
3587 if (!path_after
|| !path_before
) {
3593 * Our current directory inode may not yet be renamed/moved because some
3594 * ancestor (immediate or not) has to be renamed/moved first. So find if
3595 * such ancestor exists and make sure our own rename/move happens after
3596 * that ancestor is processed to avoid path build infinite loops (done
3597 * at get_cur_path()).
3599 while (ino
> BTRFS_FIRST_FREE_OBJECTID
) {
3600 u64 parent_ino_after_gen
;
3602 if (is_waiting_for_move(sctx
, ino
)) {
3604 * If the current inode is an ancestor of ino in the
3605 * parent root, we need to delay the rename of the
3606 * current inode, otherwise don't delayed the rename
3607 * because we can end up with a circular dependency
3608 * of renames, resulting in some directories never
3609 * getting the respective rename operations issued in
3610 * the send stream or getting into infinite path build
3613 ret
= is_ancestor(sctx
->parent_root
,
3614 sctx
->cur_ino
, sctx
->cur_inode_gen
,
3620 fs_path_reset(path_before
);
3621 fs_path_reset(path_after
);
3623 ret
= get_first_ref(sctx
->send_root
, ino
, &parent_ino_after
,
3624 &parent_ino_after_gen
, path_after
);
3627 ret
= get_first_ref(sctx
->parent_root
, ino
, &parent_ino_before
,
3629 if (ret
< 0 && ret
!= -ENOENT
) {
3631 } else if (ret
== -ENOENT
) {
3636 len1
= fs_path_len(path_before
);
3637 len2
= fs_path_len(path_after
);
3638 if (ino
> sctx
->cur_ino
&&
3639 (parent_ino_before
!= parent_ino_after
|| len1
!= len2
||
3640 memcmp(path_before
->start
, path_after
->start
, len1
))) {
3643 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
,
3644 &parent_ino_gen
, NULL
, NULL
, NULL
,
3648 if (ino_gen
== parent_ino_gen
) {
3653 ino
= parent_ino_after
;
3654 ino_gen
= parent_ino_after_gen
;
3658 fs_path_free(path_before
);
3659 fs_path_free(path_after
);
3662 ret
= add_pending_dir_move(sctx
,
3664 sctx
->cur_inode_gen
,
3667 &sctx
->deleted_refs
,
3676 static int update_ref_path(struct send_ctx
*sctx
, struct recorded_ref
*ref
)
3679 struct fs_path
*new_path
;
3682 * Our reference's name member points to its full_path member string, so
3683 * we use here a new path.
3685 new_path
= fs_path_alloc();
3689 ret
= get_cur_path(sctx
, ref
->dir
, ref
->dir_gen
, new_path
);
3691 fs_path_free(new_path
);
3694 ret
= fs_path_add(new_path
, ref
->name
, ref
->name_len
);
3696 fs_path_free(new_path
);
3700 fs_path_free(ref
->full_path
);
3701 set_ref_path(ref
, new_path
);
3707 * This does all the move/link/unlink/rmdir magic.
3709 static int process_recorded_refs(struct send_ctx
*sctx
, int *pending_move
)
3711 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
3713 struct recorded_ref
*cur
;
3714 struct recorded_ref
*cur2
;
3715 struct list_head check_dirs
;
3716 struct fs_path
*valid_path
= NULL
;
3720 int did_overwrite
= 0;
3722 u64 last_dir_ino_rm
= 0;
3723 bool can_rename
= true;
3724 bool orphanized_dir
= false;
3725 bool orphanized_ancestor
= false;
3727 btrfs_debug(fs_info
, "process_recorded_refs %llu", sctx
->cur_ino
);
3730 * This should never happen as the root dir always has the same ref
3731 * which is always '..'
3733 BUG_ON(sctx
->cur_ino
<= BTRFS_FIRST_FREE_OBJECTID
);
3734 INIT_LIST_HEAD(&check_dirs
);
3736 valid_path
= fs_path_alloc();
3743 * First, check if the first ref of the current inode was overwritten
3744 * before. If yes, we know that the current inode was already orphanized
3745 * and thus use the orphan name. If not, we can use get_cur_path to
3746 * get the path of the first ref as it would like while receiving at
3747 * this point in time.
3748 * New inodes are always orphan at the beginning, so force to use the
3749 * orphan name in this case.
3750 * The first ref is stored in valid_path and will be updated if it
3751 * gets moved around.
3753 if (!sctx
->cur_inode_new
) {
3754 ret
= did_overwrite_first_ref(sctx
, sctx
->cur_ino
,
3755 sctx
->cur_inode_gen
);
3761 if (sctx
->cur_inode_new
|| did_overwrite
) {
3762 ret
= gen_unique_name(sctx
, sctx
->cur_ino
,
3763 sctx
->cur_inode_gen
, valid_path
);
3768 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3774 list_for_each_entry(cur
, &sctx
->new_refs
, list
) {
3776 * We may have refs where the parent directory does not exist
3777 * yet. This happens if the parent directories inum is higher
3778 * the the current inum. To handle this case, we create the
3779 * parent directory out of order. But we need to check if this
3780 * did already happen before due to other refs in the same dir.
3782 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3785 if (ret
== inode_state_will_create
) {
3788 * First check if any of the current inodes refs did
3789 * already create the dir.
3791 list_for_each_entry(cur2
, &sctx
->new_refs
, list
) {
3794 if (cur2
->dir
== cur
->dir
) {
3801 * If that did not happen, check if a previous inode
3802 * did already create the dir.
3805 ret
= did_create_dir(sctx
, cur
->dir
);
3809 ret
= send_create_inode(sctx
, cur
->dir
);
3816 * Check if this new ref would overwrite the first ref of
3817 * another unprocessed inode. If yes, orphanize the
3818 * overwritten inode. If we find an overwritten ref that is
3819 * not the first ref, simply unlink it.
3821 ret
= will_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3822 cur
->name
, cur
->name_len
,
3823 &ow_inode
, &ow_gen
, &ow_mode
);
3827 ret
= is_first_ref(sctx
->parent_root
,
3828 ow_inode
, cur
->dir
, cur
->name
,
3833 struct name_cache_entry
*nce
;
3834 struct waiting_dir_move
*wdm
;
3836 ret
= orphanize_inode(sctx
, ow_inode
, ow_gen
,
3840 if (S_ISDIR(ow_mode
))
3841 orphanized_dir
= true;
3844 * If ow_inode has its rename operation delayed
3845 * make sure that its orphanized name is used in
3846 * the source path when performing its rename
3849 if (is_waiting_for_move(sctx
, ow_inode
)) {
3850 wdm
= get_waiting_dir_move(sctx
,
3853 wdm
->orphanized
= true;
3857 * Make sure we clear our orphanized inode's
3858 * name from the name cache. This is because the
3859 * inode ow_inode might be an ancestor of some
3860 * other inode that will be orphanized as well
3861 * later and has an inode number greater than
3862 * sctx->send_progress. We need to prevent
3863 * future name lookups from using the old name
3864 * and get instead the orphan name.
3866 nce
= name_cache_search(sctx
, ow_inode
, ow_gen
);
3868 name_cache_delete(sctx
, nce
);
3873 * ow_inode might currently be an ancestor of
3874 * cur_ino, therefore compute valid_path (the
3875 * current path of cur_ino) again because it
3876 * might contain the pre-orphanization name of
3877 * ow_inode, which is no longer valid.
3879 ret
= is_ancestor(sctx
->parent_root
,
3881 sctx
->cur_ino
, NULL
);
3883 orphanized_ancestor
= true;
3884 fs_path_reset(valid_path
);
3885 ret
= get_cur_path(sctx
, sctx
->cur_ino
,
3886 sctx
->cur_inode_gen
,
3892 ret
= send_unlink(sctx
, cur
->full_path
);
3898 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->parent_root
) {
3899 ret
= wait_for_dest_dir_move(sctx
, cur
, is_orphan
);
3908 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->parent_root
&&
3910 ret
= wait_for_parent_move(sctx
, cur
, is_orphan
);
3920 * link/move the ref to the new place. If we have an orphan
3921 * inode, move it and update valid_path. If not, link or move
3922 * it depending on the inode mode.
3924 if (is_orphan
&& can_rename
) {
3925 ret
= send_rename(sctx
, valid_path
, cur
->full_path
);
3929 ret
= fs_path_copy(valid_path
, cur
->full_path
);
3932 } else if (can_rename
) {
3933 if (S_ISDIR(sctx
->cur_inode_mode
)) {
3935 * Dirs can't be linked, so move it. For moved
3936 * dirs, we always have one new and one deleted
3937 * ref. The deleted ref is ignored later.
3939 ret
= send_rename(sctx
, valid_path
,
3942 ret
= fs_path_copy(valid_path
,
3948 * We might have previously orphanized an inode
3949 * which is an ancestor of our current inode,
3950 * so our reference's full path, which was
3951 * computed before any such orphanizations, must
3954 if (orphanized_dir
) {
3955 ret
= update_ref_path(sctx
, cur
);
3959 ret
= send_link(sctx
, cur
->full_path
,
3965 ret
= dup_ref(cur
, &check_dirs
);
3970 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->cur_inode_deleted
) {
3972 * Check if we can already rmdir the directory. If not,
3973 * orphanize it. For every dir item inside that gets deleted
3974 * later, we do this check again and rmdir it then if possible.
3975 * See the use of check_dirs for more details.
3977 ret
= can_rmdir(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3982 ret
= send_rmdir(sctx
, valid_path
);
3985 } else if (!is_orphan
) {
3986 ret
= orphanize_inode(sctx
, sctx
->cur_ino
,
3987 sctx
->cur_inode_gen
, valid_path
);
3993 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
3994 ret
= dup_ref(cur
, &check_dirs
);
3998 } else if (S_ISDIR(sctx
->cur_inode_mode
) &&
3999 !list_empty(&sctx
->deleted_refs
)) {
4001 * We have a moved dir. Add the old parent to check_dirs
4003 cur
= list_entry(sctx
->deleted_refs
.next
, struct recorded_ref
,
4005 ret
= dup_ref(cur
, &check_dirs
);
4008 } else if (!S_ISDIR(sctx
->cur_inode_mode
)) {
4010 * We have a non dir inode. Go through all deleted refs and
4011 * unlink them if they were not already overwritten by other
4014 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
4015 ret
= did_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
4016 sctx
->cur_ino
, sctx
->cur_inode_gen
,
4017 cur
->name
, cur
->name_len
);
4022 * If we orphanized any ancestor before, we need
4023 * to recompute the full path for deleted names,
4024 * since any such path was computed before we
4025 * processed any references and orphanized any
4028 if (orphanized_ancestor
) {
4029 ret
= update_ref_path(sctx
, cur
);
4033 ret
= send_unlink(sctx
, cur
->full_path
);
4037 ret
= dup_ref(cur
, &check_dirs
);
4042 * If the inode is still orphan, unlink the orphan. This may
4043 * happen when a previous inode did overwrite the first ref
4044 * of this inode and no new refs were added for the current
4045 * inode. Unlinking does not mean that the inode is deleted in
4046 * all cases. There may still be links to this inode in other
4050 ret
= send_unlink(sctx
, valid_path
);
4057 * We did collect all parent dirs where cur_inode was once located. We
4058 * now go through all these dirs and check if they are pending for
4059 * deletion and if it's finally possible to perform the rmdir now.
4060 * We also update the inode stats of the parent dirs here.
4062 list_for_each_entry(cur
, &check_dirs
, list
) {
4064 * In case we had refs into dirs that were not processed yet,
4065 * we don't need to do the utime and rmdir logic for these dirs.
4066 * The dir will be processed later.
4068 if (cur
->dir
> sctx
->cur_ino
)
4071 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
4075 if (ret
== inode_state_did_create
||
4076 ret
== inode_state_no_change
) {
4077 /* TODO delayed utimes */
4078 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
4081 } else if (ret
== inode_state_did_delete
&&
4082 cur
->dir
!= last_dir_ino_rm
) {
4083 ret
= can_rmdir(sctx
, cur
->dir
, cur
->dir_gen
,
4088 ret
= get_cur_path(sctx
, cur
->dir
,
4089 cur
->dir_gen
, valid_path
);
4092 ret
= send_rmdir(sctx
, valid_path
);
4095 last_dir_ino_rm
= cur
->dir
;
4103 __free_recorded_refs(&check_dirs
);
4104 free_recorded_refs(sctx
);
4105 fs_path_free(valid_path
);
4109 static int record_ref(struct btrfs_root
*root
, int num
, u64 dir
, int index
,
4110 struct fs_path
*name
, void *ctx
, struct list_head
*refs
)
4113 struct send_ctx
*sctx
= ctx
;
4117 p
= fs_path_alloc();
4121 ret
= get_inode_info(root
, dir
, NULL
, &gen
, NULL
, NULL
,
4126 ret
= get_cur_path(sctx
, dir
, gen
, p
);
4129 ret
= fs_path_add_path(p
, name
);
4133 ret
= __record_ref(refs
, dir
, gen
, p
);
4141 static int __record_new_ref(int num
, u64 dir
, int index
,
4142 struct fs_path
*name
,
4145 struct send_ctx
*sctx
= ctx
;
4146 return record_ref(sctx
->send_root
, num
, dir
, index
, name
,
4147 ctx
, &sctx
->new_refs
);
4151 static int __record_deleted_ref(int num
, u64 dir
, int index
,
4152 struct fs_path
*name
,
4155 struct send_ctx
*sctx
= ctx
;
4156 return record_ref(sctx
->parent_root
, num
, dir
, index
, name
,
4157 ctx
, &sctx
->deleted_refs
);
4160 static int record_new_ref(struct send_ctx
*sctx
)
4164 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
4165 sctx
->cmp_key
, 0, __record_new_ref
, sctx
);
4174 static int record_deleted_ref(struct send_ctx
*sctx
)
4178 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
4179 sctx
->cmp_key
, 0, __record_deleted_ref
, sctx
);
4188 struct find_ref_ctx
{
4191 struct btrfs_root
*root
;
4192 struct fs_path
*name
;
4196 static int __find_iref(int num
, u64 dir
, int index
,
4197 struct fs_path
*name
,
4200 struct find_ref_ctx
*ctx
= ctx_
;
4204 if (dir
== ctx
->dir
&& fs_path_len(name
) == fs_path_len(ctx
->name
) &&
4205 strncmp(name
->start
, ctx
->name
->start
, fs_path_len(name
)) == 0) {
4207 * To avoid doing extra lookups we'll only do this if everything
4210 ret
= get_inode_info(ctx
->root
, dir
, NULL
, &dir_gen
, NULL
,
4214 if (dir_gen
!= ctx
->dir_gen
)
4216 ctx
->found_idx
= num
;
4222 static int find_iref(struct btrfs_root
*root
,
4223 struct btrfs_path
*path
,
4224 struct btrfs_key
*key
,
4225 u64 dir
, u64 dir_gen
, struct fs_path
*name
)
4228 struct find_ref_ctx ctx
;
4232 ctx
.dir_gen
= dir_gen
;
4236 ret
= iterate_inode_ref(root
, path
, key
, 0, __find_iref
, &ctx
);
4240 if (ctx
.found_idx
== -1)
4243 return ctx
.found_idx
;
4246 static int __record_changed_new_ref(int num
, u64 dir
, int index
,
4247 struct fs_path
*name
,
4252 struct send_ctx
*sctx
= ctx
;
4254 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &dir_gen
, NULL
,
4259 ret
= find_iref(sctx
->parent_root
, sctx
->right_path
,
4260 sctx
->cmp_key
, dir
, dir_gen
, name
);
4262 ret
= __record_new_ref(num
, dir
, index
, name
, sctx
);
4269 static int __record_changed_deleted_ref(int num
, u64 dir
, int index
,
4270 struct fs_path
*name
,
4275 struct send_ctx
*sctx
= ctx
;
4277 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &dir_gen
, NULL
,
4282 ret
= find_iref(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
4283 dir
, dir_gen
, name
);
4285 ret
= __record_deleted_ref(num
, dir
, index
, name
, sctx
);
4292 static int record_changed_ref(struct send_ctx
*sctx
)
4296 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
4297 sctx
->cmp_key
, 0, __record_changed_new_ref
, sctx
);
4300 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
4301 sctx
->cmp_key
, 0, __record_changed_deleted_ref
, sctx
);
4311 * Record and process all refs at once. Needed when an inode changes the
4312 * generation number, which means that it was deleted and recreated.
4314 static int process_all_refs(struct send_ctx
*sctx
,
4315 enum btrfs_compare_tree_result cmd
)
4318 struct btrfs_root
*root
;
4319 struct btrfs_path
*path
;
4320 struct btrfs_key key
;
4321 struct btrfs_key found_key
;
4322 struct extent_buffer
*eb
;
4324 iterate_inode_ref_t cb
;
4325 int pending_move
= 0;
4327 path
= alloc_path_for_send();
4331 if (cmd
== BTRFS_COMPARE_TREE_NEW
) {
4332 root
= sctx
->send_root
;
4333 cb
= __record_new_ref
;
4334 } else if (cmd
== BTRFS_COMPARE_TREE_DELETED
) {
4335 root
= sctx
->parent_root
;
4336 cb
= __record_deleted_ref
;
4338 btrfs_err(sctx
->send_root
->fs_info
,
4339 "Wrong command %d in process_all_refs", cmd
);
4344 key
.objectid
= sctx
->cmp_key
->objectid
;
4345 key
.type
= BTRFS_INODE_REF_KEY
;
4347 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4352 eb
= path
->nodes
[0];
4353 slot
= path
->slots
[0];
4354 if (slot
>= btrfs_header_nritems(eb
)) {
4355 ret
= btrfs_next_leaf(root
, path
);
4363 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4365 if (found_key
.objectid
!= key
.objectid
||
4366 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
4367 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
))
4370 ret
= iterate_inode_ref(root
, path
, &found_key
, 0, cb
, sctx
);
4376 btrfs_release_path(path
);
4379 * We don't actually care about pending_move as we are simply
4380 * re-creating this inode and will be rename'ing it into place once we
4381 * rename the parent directory.
4383 ret
= process_recorded_refs(sctx
, &pending_move
);
4385 btrfs_free_path(path
);
4389 static int send_set_xattr(struct send_ctx
*sctx
,
4390 struct fs_path
*path
,
4391 const char *name
, int name_len
,
4392 const char *data
, int data_len
)
4396 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SET_XATTR
);
4400 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
4401 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
4402 TLV_PUT(sctx
, BTRFS_SEND_A_XATTR_DATA
, data
, data_len
);
4404 ret
= send_cmd(sctx
);
4411 static int send_remove_xattr(struct send_ctx
*sctx
,
4412 struct fs_path
*path
,
4413 const char *name
, int name_len
)
4417 ret
= begin_cmd(sctx
, BTRFS_SEND_C_REMOVE_XATTR
);
4421 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
4422 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
4424 ret
= send_cmd(sctx
);
4431 static int __process_new_xattr(int num
, struct btrfs_key
*di_key
,
4432 const char *name
, int name_len
,
4433 const char *data
, int data_len
,
4437 struct send_ctx
*sctx
= ctx
;
4439 struct posix_acl_xattr_header dummy_acl
;
4441 p
= fs_path_alloc();
4446 * This hack is needed because empty acls are stored as zero byte
4447 * data in xattrs. Problem with that is, that receiving these zero byte
4448 * acls will fail later. To fix this, we send a dummy acl list that
4449 * only contains the version number and no entries.
4451 if (!strncmp(name
, XATTR_NAME_POSIX_ACL_ACCESS
, name_len
) ||
4452 !strncmp(name
, XATTR_NAME_POSIX_ACL_DEFAULT
, name_len
)) {
4453 if (data_len
== 0) {
4454 dummy_acl
.a_version
=
4455 cpu_to_le32(POSIX_ACL_XATTR_VERSION
);
4456 data
= (char *)&dummy_acl
;
4457 data_len
= sizeof(dummy_acl
);
4461 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4465 ret
= send_set_xattr(sctx
, p
, name
, name_len
, data
, data_len
);
4472 static int __process_deleted_xattr(int num
, struct btrfs_key
*di_key
,
4473 const char *name
, int name_len
,
4474 const char *data
, int data_len
,
4478 struct send_ctx
*sctx
= ctx
;
4481 p
= fs_path_alloc();
4485 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4489 ret
= send_remove_xattr(sctx
, p
, name
, name_len
);
4496 static int process_new_xattr(struct send_ctx
*sctx
)
4500 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4501 sctx
->cmp_key
, __process_new_xattr
, sctx
);
4506 static int process_deleted_xattr(struct send_ctx
*sctx
)
4508 return iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4509 sctx
->cmp_key
, __process_deleted_xattr
, sctx
);
4512 struct find_xattr_ctx
{
4520 static int __find_xattr(int num
, struct btrfs_key
*di_key
,
4521 const char *name
, int name_len
,
4522 const char *data
, int data_len
,
4523 u8 type
, void *vctx
)
4525 struct find_xattr_ctx
*ctx
= vctx
;
4527 if (name_len
== ctx
->name_len
&&
4528 strncmp(name
, ctx
->name
, name_len
) == 0) {
4529 ctx
->found_idx
= num
;
4530 ctx
->found_data_len
= data_len
;
4531 ctx
->found_data
= kmemdup(data
, data_len
, GFP_KERNEL
);
4532 if (!ctx
->found_data
)
4539 static int find_xattr(struct btrfs_root
*root
,
4540 struct btrfs_path
*path
,
4541 struct btrfs_key
*key
,
4542 const char *name
, int name_len
,
4543 char **data
, int *data_len
)
4546 struct find_xattr_ctx ctx
;
4549 ctx
.name_len
= name_len
;
4551 ctx
.found_data
= NULL
;
4552 ctx
.found_data_len
= 0;
4554 ret
= iterate_dir_item(root
, path
, key
, __find_xattr
, &ctx
);
4558 if (ctx
.found_idx
== -1)
4561 *data
= ctx
.found_data
;
4562 *data_len
= ctx
.found_data_len
;
4564 kfree(ctx
.found_data
);
4566 return ctx
.found_idx
;
4570 static int __process_changed_new_xattr(int num
, struct btrfs_key
*di_key
,
4571 const char *name
, int name_len
,
4572 const char *data
, int data_len
,
4576 struct send_ctx
*sctx
= ctx
;
4577 char *found_data
= NULL
;
4578 int found_data_len
= 0;
4580 ret
= find_xattr(sctx
->parent_root
, sctx
->right_path
,
4581 sctx
->cmp_key
, name
, name_len
, &found_data
,
4583 if (ret
== -ENOENT
) {
4584 ret
= __process_new_xattr(num
, di_key
, name
, name_len
, data
,
4585 data_len
, type
, ctx
);
4586 } else if (ret
>= 0) {
4587 if (data_len
!= found_data_len
||
4588 memcmp(data
, found_data
, data_len
)) {
4589 ret
= __process_new_xattr(num
, di_key
, name
, name_len
,
4590 data
, data_len
, type
, ctx
);
4600 static int __process_changed_deleted_xattr(int num
, struct btrfs_key
*di_key
,
4601 const char *name
, int name_len
,
4602 const char *data
, int data_len
,
4606 struct send_ctx
*sctx
= ctx
;
4608 ret
= find_xattr(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
4609 name
, name_len
, NULL
, NULL
);
4611 ret
= __process_deleted_xattr(num
, di_key
, name
, name_len
, data
,
4612 data_len
, type
, ctx
);
4619 static int process_changed_xattr(struct send_ctx
*sctx
)
4623 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4624 sctx
->cmp_key
, __process_changed_new_xattr
, sctx
);
4627 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4628 sctx
->cmp_key
, __process_changed_deleted_xattr
, sctx
);
4634 static int process_all_new_xattrs(struct send_ctx
*sctx
)
4637 struct btrfs_root
*root
;
4638 struct btrfs_path
*path
;
4639 struct btrfs_key key
;
4640 struct btrfs_key found_key
;
4641 struct extent_buffer
*eb
;
4644 path
= alloc_path_for_send();
4648 root
= sctx
->send_root
;
4650 key
.objectid
= sctx
->cmp_key
->objectid
;
4651 key
.type
= BTRFS_XATTR_ITEM_KEY
;
4653 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4658 eb
= path
->nodes
[0];
4659 slot
= path
->slots
[0];
4660 if (slot
>= btrfs_header_nritems(eb
)) {
4661 ret
= btrfs_next_leaf(root
, path
);
4664 } else if (ret
> 0) {
4671 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4672 if (found_key
.objectid
!= key
.objectid
||
4673 found_key
.type
!= key
.type
) {
4678 ret
= iterate_dir_item(root
, path
, &found_key
,
4679 __process_new_xattr
, sctx
);
4687 btrfs_free_path(path
);
4691 static ssize_t
fill_read_buf(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4693 struct btrfs_root
*root
= sctx
->send_root
;
4694 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4695 struct inode
*inode
;
4698 struct btrfs_key key
;
4699 pgoff_t index
= offset
>> PAGE_SHIFT
;
4701 unsigned pg_offset
= offset
& ~PAGE_MASK
;
4704 key
.objectid
= sctx
->cur_ino
;
4705 key
.type
= BTRFS_INODE_ITEM_KEY
;
4708 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
4710 return PTR_ERR(inode
);
4712 if (offset
+ len
> i_size_read(inode
)) {
4713 if (offset
> i_size_read(inode
))
4716 len
= offset
- i_size_read(inode
);
4721 last_index
= (offset
+ len
- 1) >> PAGE_SHIFT
;
4723 /* initial readahead */
4724 memset(&sctx
->ra
, 0, sizeof(struct file_ra_state
));
4725 file_ra_state_init(&sctx
->ra
, inode
->i_mapping
);
4726 page_cache_sync_readahead(inode
->i_mapping
, &sctx
->ra
, NULL
, index
,
4727 last_index
- index
+ 1);
4729 while (index
<= last_index
) {
4730 unsigned cur_len
= min_t(unsigned, len
,
4731 PAGE_SIZE
- pg_offset
);
4732 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_KERNEL
);
4738 if (!PageUptodate(page
)) {
4739 btrfs_readpage(NULL
, page
);
4741 if (!PageUptodate(page
)) {
4750 memcpy(sctx
->read_buf
+ ret
, addr
+ pg_offset
, cur_len
);
4765 * Read some bytes from the current inode/file and send a write command to
4768 static int send_write(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4770 struct btrfs_fs_info
*fs_info
= sctx
->send_root
->fs_info
;
4773 ssize_t num_read
= 0;
4775 p
= fs_path_alloc();
4779 btrfs_debug(fs_info
, "send_write offset=%llu, len=%d", offset
, len
);
4781 num_read
= fill_read_buf(sctx
, offset
, len
);
4782 if (num_read
<= 0) {
4788 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4792 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4796 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4797 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4798 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, num_read
);
4800 ret
= send_cmd(sctx
);
4811 * Send a clone command to user space.
4813 static int send_clone(struct send_ctx
*sctx
,
4814 u64 offset
, u32 len
,
4815 struct clone_root
*clone_root
)
4821 btrfs_debug(sctx
->send_root
->fs_info
,
4822 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4823 offset
, len
, clone_root
->root
->objectid
, clone_root
->ino
,
4824 clone_root
->offset
);
4826 p
= fs_path_alloc();
4830 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CLONE
);
4834 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4838 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4839 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_LEN
, len
);
4840 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4842 if (clone_root
->root
== sctx
->send_root
) {
4843 ret
= get_inode_info(sctx
->send_root
, clone_root
->ino
, NULL
,
4844 &gen
, NULL
, NULL
, NULL
, NULL
);
4847 ret
= get_cur_path(sctx
, clone_root
->ino
, gen
, p
);
4849 ret
= get_inode_path(clone_root
->root
, clone_root
->ino
, p
);
4855 * If the parent we're using has a received_uuid set then use that as
4856 * our clone source as that is what we will look for when doing a
4859 * This covers the case that we create a snapshot off of a received
4860 * subvolume and then use that as the parent and try to receive on a
4863 if (!btrfs_is_empty_uuid(clone_root
->root
->root_item
.received_uuid
))
4864 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4865 clone_root
->root
->root_item
.received_uuid
);
4867 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4868 clone_root
->root
->root_item
.uuid
);
4869 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
4870 le64_to_cpu(clone_root
->root
->root_item
.ctransid
));
4871 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_CLONE_PATH
, p
);
4872 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_OFFSET
,
4873 clone_root
->offset
);
4875 ret
= send_cmd(sctx
);
4884 * Send an update extent command to user space.
4886 static int send_update_extent(struct send_ctx
*sctx
,
4887 u64 offset
, u32 len
)
4892 p
= fs_path_alloc();
4896 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UPDATE_EXTENT
);
4900 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4904 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4905 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4906 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, len
);
4908 ret
= send_cmd(sctx
);
4916 static int send_hole(struct send_ctx
*sctx
, u64 end
)
4918 struct fs_path
*p
= NULL
;
4919 u64 offset
= sctx
->cur_inode_last_extent
;
4923 p
= fs_path_alloc();
4926 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4928 goto tlv_put_failure
;
4929 memset(sctx
->read_buf
, 0, BTRFS_SEND_READ_SIZE
);
4930 while (offset
< end
) {
4931 len
= min_t(u64
, end
- offset
, BTRFS_SEND_READ_SIZE
);
4933 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4936 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4937 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4938 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, len
);
4939 ret
= send_cmd(sctx
);
4949 static int send_extent_data(struct send_ctx
*sctx
,
4955 if (sctx
->flags
& BTRFS_SEND_FLAG_NO_FILE_DATA
)
4956 return send_update_extent(sctx
, offset
, len
);
4958 while (sent
< len
) {
4959 u64 size
= len
- sent
;
4962 if (size
> BTRFS_SEND_READ_SIZE
)
4963 size
= BTRFS_SEND_READ_SIZE
;
4964 ret
= send_write(sctx
, offset
+ sent
, size
);
4974 static int clone_range(struct send_ctx
*sctx
,
4975 struct clone_root
*clone_root
,
4976 const u64 disk_byte
,
4981 struct btrfs_path
*path
;
4982 struct btrfs_key key
;
4986 * Prevent cloning from a zero offset with a length matching the sector
4987 * size because in some scenarios this will make the receiver fail.
4989 * For example, if in the source filesystem the extent at offset 0
4990 * has a length of sectorsize and it was written using direct IO, then
4991 * it can never be an inline extent (even if compression is enabled).
4992 * Then this extent can be cloned in the original filesystem to a non
4993 * zero file offset, but it may not be possible to clone in the
4994 * destination filesystem because it can be inlined due to compression
4995 * on the destination filesystem (as the receiver's write operations are
4996 * always done using buffered IO). The same happens when the original
4997 * filesystem does not have compression enabled but the destination
5000 if (clone_root
->offset
== 0 &&
5001 len
== sctx
->send_root
->fs_info
->sectorsize
)
5002 return send_extent_data(sctx
, offset
, len
);
5004 path
= alloc_path_for_send();
5009 * We can't send a clone operation for the entire range if we find
5010 * extent items in the respective range in the source file that
5011 * refer to different extents or if we find holes.
5012 * So check for that and do a mix of clone and regular write/copy
5013 * operations if needed.
5017 * mkfs.btrfs -f /dev/sda
5018 * mount /dev/sda /mnt
5019 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5020 * cp --reflink=always /mnt/foo /mnt/bar
5021 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5022 * btrfs subvolume snapshot -r /mnt /mnt/snap
5024 * If when we send the snapshot and we are processing file bar (which
5025 * has a higher inode number than foo) we blindly send a clone operation
5026 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5027 * a file bar that matches the content of file foo - iow, doesn't match
5028 * the content from bar in the original filesystem.
5030 key
.objectid
= clone_root
->ino
;
5031 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5032 key
.offset
= clone_root
->offset
;
5033 ret
= btrfs_search_slot(NULL
, clone_root
->root
, &key
, path
, 0, 0);
5036 if (ret
> 0 && path
->slots
[0] > 0) {
5037 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0] - 1);
5038 if (key
.objectid
== clone_root
->ino
&&
5039 key
.type
== BTRFS_EXTENT_DATA_KEY
)
5044 struct extent_buffer
*leaf
= path
->nodes
[0];
5045 int slot
= path
->slots
[0];
5046 struct btrfs_file_extent_item
*ei
;
5051 if (slot
>= btrfs_header_nritems(leaf
)) {
5052 ret
= btrfs_next_leaf(clone_root
->root
, path
);
5060 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5063 * We might have an implicit trailing hole (NO_HOLES feature
5064 * enabled). We deal with it after leaving this loop.
5066 if (key
.objectid
!= clone_root
->ino
||
5067 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5070 ei
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5071 type
= btrfs_file_extent_type(leaf
, ei
);
5072 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5073 ext_len
= btrfs_file_extent_inline_len(leaf
, slot
, ei
);
5074 ext_len
= PAGE_ALIGN(ext_len
);
5076 ext_len
= btrfs_file_extent_num_bytes(leaf
, ei
);
5079 if (key
.offset
+ ext_len
<= clone_root
->offset
)
5082 if (key
.offset
> clone_root
->offset
) {
5083 /* Implicit hole, NO_HOLES feature enabled. */
5084 u64 hole_len
= key
.offset
- clone_root
->offset
;
5088 ret
= send_extent_data(sctx
, offset
, hole_len
);
5096 clone_root
->offset
+= hole_len
;
5097 data_offset
+= hole_len
;
5100 if (key
.offset
>= clone_root
->offset
+ len
)
5103 clone_len
= min_t(u64
, ext_len
, len
);
5105 if (btrfs_file_extent_disk_bytenr(leaf
, ei
) == disk_byte
&&
5106 btrfs_file_extent_offset(leaf
, ei
) == data_offset
)
5107 ret
= send_clone(sctx
, offset
, clone_len
, clone_root
);
5109 ret
= send_extent_data(sctx
, offset
, clone_len
);
5117 offset
+= clone_len
;
5118 clone_root
->offset
+= clone_len
;
5119 data_offset
+= clone_len
;
5125 ret
= send_extent_data(sctx
, offset
, len
);
5129 btrfs_free_path(path
);
5133 static int send_write_or_clone(struct send_ctx
*sctx
,
5134 struct btrfs_path
*path
,
5135 struct btrfs_key
*key
,
5136 struct clone_root
*clone_root
)
5139 struct btrfs_file_extent_item
*ei
;
5140 u64 offset
= key
->offset
;
5143 u64 bs
= sctx
->send_root
->fs_info
->sb
->s_blocksize
;
5145 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5146 struct btrfs_file_extent_item
);
5147 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
5148 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5149 len
= btrfs_file_extent_inline_len(path
->nodes
[0],
5150 path
->slots
[0], ei
);
5152 * it is possible the inline item won't cover the whole page,
5153 * but there may be items after this page. Make
5154 * sure to send the whole thing
5156 len
= PAGE_ALIGN(len
);
5158 len
= btrfs_file_extent_num_bytes(path
->nodes
[0], ei
);
5161 if (offset
+ len
> sctx
->cur_inode_size
)
5162 len
= sctx
->cur_inode_size
- offset
;
5168 if (clone_root
&& IS_ALIGNED(offset
+ len
, bs
)) {
5172 disk_byte
= btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
);
5173 data_offset
= btrfs_file_extent_offset(path
->nodes
[0], ei
);
5174 ret
= clone_range(sctx
, clone_root
, disk_byte
, data_offset
,
5177 ret
= send_extent_data(sctx
, offset
, len
);
5183 static int is_extent_unchanged(struct send_ctx
*sctx
,
5184 struct btrfs_path
*left_path
,
5185 struct btrfs_key
*ekey
)
5188 struct btrfs_key key
;
5189 struct btrfs_path
*path
= NULL
;
5190 struct extent_buffer
*eb
;
5192 struct btrfs_key found_key
;
5193 struct btrfs_file_extent_item
*ei
;
5198 u64 left_offset_fixed
;
5206 path
= alloc_path_for_send();
5210 eb
= left_path
->nodes
[0];
5211 slot
= left_path
->slots
[0];
5212 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
5213 left_type
= btrfs_file_extent_type(eb
, ei
);
5215 if (left_type
!= BTRFS_FILE_EXTENT_REG
) {
5219 left_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
5220 left_len
= btrfs_file_extent_num_bytes(eb
, ei
);
5221 left_offset
= btrfs_file_extent_offset(eb
, ei
);
5222 left_gen
= btrfs_file_extent_generation(eb
, ei
);
5225 * Following comments will refer to these graphics. L is the left
5226 * extents which we are checking at the moment. 1-8 are the right
5227 * extents that we iterate.
5230 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5233 * |--1--|-2b-|...(same as above)
5235 * Alternative situation. Happens on files where extents got split.
5237 * |-----------7-----------|-6-|
5239 * Alternative situation. Happens on files which got larger.
5242 * Nothing follows after 8.
5245 key
.objectid
= ekey
->objectid
;
5246 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5247 key
.offset
= ekey
->offset
;
5248 ret
= btrfs_search_slot_for_read(sctx
->parent_root
, &key
, path
, 0, 0);
5257 * Handle special case where the right side has no extents at all.
5259 eb
= path
->nodes
[0];
5260 slot
= path
->slots
[0];
5261 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5262 if (found_key
.objectid
!= key
.objectid
||
5263 found_key
.type
!= key
.type
) {
5264 /* If we're a hole then just pretend nothing changed */
5265 ret
= (left_disknr
) ? 0 : 1;
5270 * We're now on 2a, 2b or 7.
5273 while (key
.offset
< ekey
->offset
+ left_len
) {
5274 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
5275 right_type
= btrfs_file_extent_type(eb
, ei
);
5276 if (right_type
!= BTRFS_FILE_EXTENT_REG
&&
5277 right_type
!= BTRFS_FILE_EXTENT_INLINE
) {
5282 if (right_type
== BTRFS_FILE_EXTENT_INLINE
) {
5283 right_len
= btrfs_file_extent_inline_len(eb
, slot
, ei
);
5284 right_len
= PAGE_ALIGN(right_len
);
5286 right_len
= btrfs_file_extent_num_bytes(eb
, ei
);
5290 * Are we at extent 8? If yes, we know the extent is changed.
5291 * This may only happen on the first iteration.
5293 if (found_key
.offset
+ right_len
<= ekey
->offset
) {
5294 /* If we're a hole just pretend nothing changed */
5295 ret
= (left_disknr
) ? 0 : 1;
5300 * We just wanted to see if when we have an inline extent, what
5301 * follows it is a regular extent (wanted to check the above
5302 * condition for inline extents too). This should normally not
5303 * happen but it's possible for example when we have an inline
5304 * compressed extent representing data with a size matching
5305 * the page size (currently the same as sector size).
5307 if (right_type
== BTRFS_FILE_EXTENT_INLINE
) {
5312 right_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
5313 right_offset
= btrfs_file_extent_offset(eb
, ei
);
5314 right_gen
= btrfs_file_extent_generation(eb
, ei
);
5316 left_offset_fixed
= left_offset
;
5317 if (key
.offset
< ekey
->offset
) {
5318 /* Fix the right offset for 2a and 7. */
5319 right_offset
+= ekey
->offset
- key
.offset
;
5321 /* Fix the left offset for all behind 2a and 2b */
5322 left_offset_fixed
+= key
.offset
- ekey
->offset
;
5326 * Check if we have the same extent.
5328 if (left_disknr
!= right_disknr
||
5329 left_offset_fixed
!= right_offset
||
5330 left_gen
!= right_gen
) {
5336 * Go to the next extent.
5338 ret
= btrfs_next_item(sctx
->parent_root
, path
);
5342 eb
= path
->nodes
[0];
5343 slot
= path
->slots
[0];
5344 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5346 if (ret
|| found_key
.objectid
!= key
.objectid
||
5347 found_key
.type
!= key
.type
) {
5348 key
.offset
+= right_len
;
5351 if (found_key
.offset
!= key
.offset
+ right_len
) {
5359 * We're now behind the left extent (treat as unchanged) or at the end
5360 * of the right side (treat as changed).
5362 if (key
.offset
>= ekey
->offset
+ left_len
)
5369 btrfs_free_path(path
);
5373 static int get_last_extent(struct send_ctx
*sctx
, u64 offset
)
5375 struct btrfs_path
*path
;
5376 struct btrfs_root
*root
= sctx
->send_root
;
5377 struct btrfs_file_extent_item
*fi
;
5378 struct btrfs_key key
;
5383 path
= alloc_path_for_send();
5387 sctx
->cur_inode_last_extent
= 0;
5389 key
.objectid
= sctx
->cur_ino
;
5390 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5391 key
.offset
= offset
;
5392 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 0, 1);
5396 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
5397 if (key
.objectid
!= sctx
->cur_ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5400 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5401 struct btrfs_file_extent_item
);
5402 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
5403 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5404 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
5405 path
->slots
[0], fi
);
5406 extent_end
= ALIGN(key
.offset
+ size
,
5407 sctx
->send_root
->fs_info
->sectorsize
);
5409 extent_end
= key
.offset
+
5410 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
5412 sctx
->cur_inode_last_extent
= extent_end
;
5414 btrfs_free_path(path
);
5418 static int range_is_hole_in_parent(struct send_ctx
*sctx
,
5422 struct btrfs_path
*path
;
5423 struct btrfs_key key
;
5424 struct btrfs_root
*root
= sctx
->parent_root
;
5425 u64 search_start
= start
;
5428 path
= alloc_path_for_send();
5432 key
.objectid
= sctx
->cur_ino
;
5433 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5434 key
.offset
= search_start
;
5435 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5438 if (ret
> 0 && path
->slots
[0] > 0)
5441 while (search_start
< end
) {
5442 struct extent_buffer
*leaf
= path
->nodes
[0];
5443 int slot
= path
->slots
[0];
5444 struct btrfs_file_extent_item
*fi
;
5447 if (slot
>= btrfs_header_nritems(leaf
)) {
5448 ret
= btrfs_next_leaf(root
, path
);
5456 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5457 if (key
.objectid
< sctx
->cur_ino
||
5458 key
.type
< BTRFS_EXTENT_DATA_KEY
)
5460 if (key
.objectid
> sctx
->cur_ino
||
5461 key
.type
> BTRFS_EXTENT_DATA_KEY
||
5465 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5466 if (btrfs_file_extent_type(leaf
, fi
) ==
5467 BTRFS_FILE_EXTENT_INLINE
) {
5468 u64 size
= btrfs_file_extent_inline_len(leaf
, slot
, fi
);
5470 extent_end
= ALIGN(key
.offset
+ size
,
5471 root
->fs_info
->sectorsize
);
5473 extent_end
= key
.offset
+
5474 btrfs_file_extent_num_bytes(leaf
, fi
);
5476 if (extent_end
<= start
)
5478 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0) {
5479 search_start
= extent_end
;
5489 btrfs_free_path(path
);
5493 static int maybe_send_hole(struct send_ctx
*sctx
, struct btrfs_path
*path
,
5494 struct btrfs_key
*key
)
5496 struct btrfs_file_extent_item
*fi
;
5501 if (sctx
->cur_ino
!= key
->objectid
|| !need_send_hole(sctx
))
5504 if (sctx
->cur_inode_last_extent
== (u64
)-1) {
5505 ret
= get_last_extent(sctx
, key
->offset
- 1);
5510 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5511 struct btrfs_file_extent_item
);
5512 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
5513 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5514 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
5515 path
->slots
[0], fi
);
5516 extent_end
= ALIGN(key
->offset
+ size
,
5517 sctx
->send_root
->fs_info
->sectorsize
);
5519 extent_end
= key
->offset
+
5520 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
5523 if (path
->slots
[0] == 0 &&
5524 sctx
->cur_inode_last_extent
< key
->offset
) {
5526 * We might have skipped entire leafs that contained only
5527 * file extent items for our current inode. These leafs have
5528 * a generation number smaller (older) than the one in the
5529 * current leaf and the leaf our last extent came from, and
5530 * are located between these 2 leafs.
5532 ret
= get_last_extent(sctx
, key
->offset
- 1);
5537 if (sctx
->cur_inode_last_extent
< key
->offset
) {
5538 ret
= range_is_hole_in_parent(sctx
,
5539 sctx
->cur_inode_last_extent
,
5544 ret
= send_hole(sctx
, key
->offset
);
5548 sctx
->cur_inode_last_extent
= extent_end
;
5552 static int process_extent(struct send_ctx
*sctx
,
5553 struct btrfs_path
*path
,
5554 struct btrfs_key
*key
)
5556 struct clone_root
*found_clone
= NULL
;
5559 if (S_ISLNK(sctx
->cur_inode_mode
))
5562 if (sctx
->parent_root
&& !sctx
->cur_inode_new
) {
5563 ret
= is_extent_unchanged(sctx
, path
, key
);
5571 struct btrfs_file_extent_item
*ei
;
5574 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5575 struct btrfs_file_extent_item
);
5576 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
5577 if (type
== BTRFS_FILE_EXTENT_PREALLOC
||
5578 type
== BTRFS_FILE_EXTENT_REG
) {
5580 * The send spec does not have a prealloc command yet,
5581 * so just leave a hole for prealloc'ed extents until
5582 * we have enough commands queued up to justify rev'ing
5585 if (type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5590 /* Have a hole, just skip it. */
5591 if (btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
) == 0) {
5598 ret
= find_extent_clone(sctx
, path
, key
->objectid
, key
->offset
,
5599 sctx
->cur_inode_size
, &found_clone
);
5600 if (ret
!= -ENOENT
&& ret
< 0)
5603 ret
= send_write_or_clone(sctx
, path
, key
, found_clone
);
5607 ret
= maybe_send_hole(sctx
, path
, key
);
5612 static int process_all_extents(struct send_ctx
*sctx
)
5615 struct btrfs_root
*root
;
5616 struct btrfs_path
*path
;
5617 struct btrfs_key key
;
5618 struct btrfs_key found_key
;
5619 struct extent_buffer
*eb
;
5622 root
= sctx
->send_root
;
5623 path
= alloc_path_for_send();
5627 key
.objectid
= sctx
->cmp_key
->objectid
;
5628 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5630 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5635 eb
= path
->nodes
[0];
5636 slot
= path
->slots
[0];
5638 if (slot
>= btrfs_header_nritems(eb
)) {
5639 ret
= btrfs_next_leaf(root
, path
);
5642 } else if (ret
> 0) {
5649 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5651 if (found_key
.objectid
!= key
.objectid
||
5652 found_key
.type
!= key
.type
) {
5657 ret
= process_extent(sctx
, path
, &found_key
);
5665 btrfs_free_path(path
);
5669 static int process_recorded_refs_if_needed(struct send_ctx
*sctx
, int at_end
,
5671 int *refs_processed
)
5675 if (sctx
->cur_ino
== 0)
5677 if (!at_end
&& sctx
->cur_ino
== sctx
->cmp_key
->objectid
&&
5678 sctx
->cmp_key
->type
<= BTRFS_INODE_EXTREF_KEY
)
5680 if (list_empty(&sctx
->new_refs
) && list_empty(&sctx
->deleted_refs
))
5683 ret
= process_recorded_refs(sctx
, pending_move
);
5687 *refs_processed
= 1;
5692 static int finish_inode_if_needed(struct send_ctx
*sctx
, int at_end
)
5703 int pending_move
= 0;
5704 int refs_processed
= 0;
5706 ret
= process_recorded_refs_if_needed(sctx
, at_end
, &pending_move
,
5712 * We have processed the refs and thus need to advance send_progress.
5713 * Now, calls to get_cur_xxx will take the updated refs of the current
5714 * inode into account.
5716 * On the other hand, if our current inode is a directory and couldn't
5717 * be moved/renamed because its parent was renamed/moved too and it has
5718 * a higher inode number, we can only move/rename our current inode
5719 * after we moved/renamed its parent. Therefore in this case operate on
5720 * the old path (pre move/rename) of our current inode, and the
5721 * move/rename will be performed later.
5723 if (refs_processed
&& !pending_move
)
5724 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5726 if (sctx
->cur_ino
== 0 || sctx
->cur_inode_deleted
)
5728 if (!at_end
&& sctx
->cmp_key
->objectid
== sctx
->cur_ino
)
5731 ret
= get_inode_info(sctx
->send_root
, sctx
->cur_ino
, NULL
, NULL
,
5732 &left_mode
, &left_uid
, &left_gid
, NULL
);
5736 if (!sctx
->parent_root
|| sctx
->cur_inode_new
) {
5738 if (!S_ISLNK(sctx
->cur_inode_mode
))
5741 ret
= get_inode_info(sctx
->parent_root
, sctx
->cur_ino
,
5742 NULL
, NULL
, &right_mode
, &right_uid
,
5747 if (left_uid
!= right_uid
|| left_gid
!= right_gid
)
5749 if (!S_ISLNK(sctx
->cur_inode_mode
) && left_mode
!= right_mode
)
5753 if (S_ISREG(sctx
->cur_inode_mode
)) {
5754 if (need_send_hole(sctx
)) {
5755 if (sctx
->cur_inode_last_extent
== (u64
)-1 ||
5756 sctx
->cur_inode_last_extent
<
5757 sctx
->cur_inode_size
) {
5758 ret
= get_last_extent(sctx
, (u64
)-1);
5762 if (sctx
->cur_inode_last_extent
<
5763 sctx
->cur_inode_size
) {
5764 ret
= send_hole(sctx
, sctx
->cur_inode_size
);
5769 ret
= send_truncate(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5770 sctx
->cur_inode_size
);
5776 ret
= send_chown(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5777 left_uid
, left_gid
);
5782 ret
= send_chmod(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5789 * If other directory inodes depended on our current directory
5790 * inode's move/rename, now do their move/rename operations.
5792 if (!is_waiting_for_move(sctx
, sctx
->cur_ino
)) {
5793 ret
= apply_children_dir_moves(sctx
);
5797 * Need to send that every time, no matter if it actually
5798 * changed between the two trees as we have done changes to
5799 * the inode before. If our inode is a directory and it's
5800 * waiting to be moved/renamed, we will send its utimes when
5801 * it's moved/renamed, therefore we don't need to do it here.
5803 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5804 ret
= send_utimes(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
);
5813 static int changed_inode(struct send_ctx
*sctx
,
5814 enum btrfs_compare_tree_result result
)
5817 struct btrfs_key
*key
= sctx
->cmp_key
;
5818 struct btrfs_inode_item
*left_ii
= NULL
;
5819 struct btrfs_inode_item
*right_ii
= NULL
;
5823 sctx
->cur_ino
= key
->objectid
;
5824 sctx
->cur_inode_new_gen
= 0;
5825 sctx
->cur_inode_last_extent
= (u64
)-1;
5828 * Set send_progress to current inode. This will tell all get_cur_xxx
5829 * functions that the current inode's refs are not updated yet. Later,
5830 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5832 sctx
->send_progress
= sctx
->cur_ino
;
5834 if (result
== BTRFS_COMPARE_TREE_NEW
||
5835 result
== BTRFS_COMPARE_TREE_CHANGED
) {
5836 left_ii
= btrfs_item_ptr(sctx
->left_path
->nodes
[0],
5837 sctx
->left_path
->slots
[0],
5838 struct btrfs_inode_item
);
5839 left_gen
= btrfs_inode_generation(sctx
->left_path
->nodes
[0],
5842 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5843 sctx
->right_path
->slots
[0],
5844 struct btrfs_inode_item
);
5845 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5848 if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5849 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5850 sctx
->right_path
->slots
[0],
5851 struct btrfs_inode_item
);
5853 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5857 * The cur_ino = root dir case is special here. We can't treat
5858 * the inode as deleted+reused because it would generate a
5859 * stream that tries to delete/mkdir the root dir.
5861 if (left_gen
!= right_gen
&&
5862 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5863 sctx
->cur_inode_new_gen
= 1;
5866 if (result
== BTRFS_COMPARE_TREE_NEW
) {
5867 sctx
->cur_inode_gen
= left_gen
;
5868 sctx
->cur_inode_new
= 1;
5869 sctx
->cur_inode_deleted
= 0;
5870 sctx
->cur_inode_size
= btrfs_inode_size(
5871 sctx
->left_path
->nodes
[0], left_ii
);
5872 sctx
->cur_inode_mode
= btrfs_inode_mode(
5873 sctx
->left_path
->nodes
[0], left_ii
);
5874 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
5875 sctx
->left_path
->nodes
[0], left_ii
);
5876 if (sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5877 ret
= send_create_inode_if_needed(sctx
);
5878 } else if (result
== BTRFS_COMPARE_TREE_DELETED
) {
5879 sctx
->cur_inode_gen
= right_gen
;
5880 sctx
->cur_inode_new
= 0;
5881 sctx
->cur_inode_deleted
= 1;
5882 sctx
->cur_inode_size
= btrfs_inode_size(
5883 sctx
->right_path
->nodes
[0], right_ii
);
5884 sctx
->cur_inode_mode
= btrfs_inode_mode(
5885 sctx
->right_path
->nodes
[0], right_ii
);
5886 } else if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5888 * We need to do some special handling in case the inode was
5889 * reported as changed with a changed generation number. This
5890 * means that the original inode was deleted and new inode
5891 * reused the same inum. So we have to treat the old inode as
5892 * deleted and the new one as new.
5894 if (sctx
->cur_inode_new_gen
) {
5896 * First, process the inode as if it was deleted.
5898 sctx
->cur_inode_gen
= right_gen
;
5899 sctx
->cur_inode_new
= 0;
5900 sctx
->cur_inode_deleted
= 1;
5901 sctx
->cur_inode_size
= btrfs_inode_size(
5902 sctx
->right_path
->nodes
[0], right_ii
);
5903 sctx
->cur_inode_mode
= btrfs_inode_mode(
5904 sctx
->right_path
->nodes
[0], right_ii
);
5905 ret
= process_all_refs(sctx
,
5906 BTRFS_COMPARE_TREE_DELETED
);
5911 * Now process the inode as if it was new.
5913 sctx
->cur_inode_gen
= left_gen
;
5914 sctx
->cur_inode_new
= 1;
5915 sctx
->cur_inode_deleted
= 0;
5916 sctx
->cur_inode_size
= btrfs_inode_size(
5917 sctx
->left_path
->nodes
[0], left_ii
);
5918 sctx
->cur_inode_mode
= btrfs_inode_mode(
5919 sctx
->left_path
->nodes
[0], left_ii
);
5920 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
5921 sctx
->left_path
->nodes
[0], left_ii
);
5922 ret
= send_create_inode_if_needed(sctx
);
5926 ret
= process_all_refs(sctx
, BTRFS_COMPARE_TREE_NEW
);
5930 * Advance send_progress now as we did not get into
5931 * process_recorded_refs_if_needed in the new_gen case.
5933 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5936 * Now process all extents and xattrs of the inode as if
5937 * they were all new.
5939 ret
= process_all_extents(sctx
);
5942 ret
= process_all_new_xattrs(sctx
);
5946 sctx
->cur_inode_gen
= left_gen
;
5947 sctx
->cur_inode_new
= 0;
5948 sctx
->cur_inode_new_gen
= 0;
5949 sctx
->cur_inode_deleted
= 0;
5950 sctx
->cur_inode_size
= btrfs_inode_size(
5951 sctx
->left_path
->nodes
[0], left_ii
);
5952 sctx
->cur_inode_mode
= btrfs_inode_mode(
5953 sctx
->left_path
->nodes
[0], left_ii
);
5962 * We have to process new refs before deleted refs, but compare_trees gives us
5963 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5964 * first and later process them in process_recorded_refs.
5965 * For the cur_inode_new_gen case, we skip recording completely because
5966 * changed_inode did already initiate processing of refs. The reason for this is
5967 * that in this case, compare_tree actually compares the refs of 2 different
5968 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5969 * refs of the right tree as deleted and all refs of the left tree as new.
5971 static int changed_ref(struct send_ctx
*sctx
,
5972 enum btrfs_compare_tree_result result
)
5976 if (sctx
->cur_ino
!= sctx
->cmp_key
->objectid
) {
5977 inconsistent_snapshot_error(sctx
, result
, "reference");
5981 if (!sctx
->cur_inode_new_gen
&&
5982 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
5983 if (result
== BTRFS_COMPARE_TREE_NEW
)
5984 ret
= record_new_ref(sctx
);
5985 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
5986 ret
= record_deleted_ref(sctx
);
5987 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
5988 ret
= record_changed_ref(sctx
);
5995 * Process new/deleted/changed xattrs. We skip processing in the
5996 * cur_inode_new_gen case because changed_inode did already initiate processing
5997 * of xattrs. The reason is the same as in changed_ref
5999 static int changed_xattr(struct send_ctx
*sctx
,
6000 enum btrfs_compare_tree_result result
)
6004 if (sctx
->cur_ino
!= sctx
->cmp_key
->objectid
) {
6005 inconsistent_snapshot_error(sctx
, result
, "xattr");
6009 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
6010 if (result
== BTRFS_COMPARE_TREE_NEW
)
6011 ret
= process_new_xattr(sctx
);
6012 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
6013 ret
= process_deleted_xattr(sctx
);
6014 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
6015 ret
= process_changed_xattr(sctx
);
6022 * Process new/deleted/changed extents. We skip processing in the
6023 * cur_inode_new_gen case because changed_inode did already initiate processing
6024 * of extents. The reason is the same as in changed_ref
6026 static int changed_extent(struct send_ctx
*sctx
,
6027 enum btrfs_compare_tree_result result
)
6031 if (sctx
->cur_ino
!= sctx
->cmp_key
->objectid
) {
6033 if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
6034 struct extent_buffer
*leaf_l
;
6035 struct extent_buffer
*leaf_r
;
6036 struct btrfs_file_extent_item
*ei_l
;
6037 struct btrfs_file_extent_item
*ei_r
;
6039 leaf_l
= sctx
->left_path
->nodes
[0];
6040 leaf_r
= sctx
->right_path
->nodes
[0];
6041 ei_l
= btrfs_item_ptr(leaf_l
,
6042 sctx
->left_path
->slots
[0],
6043 struct btrfs_file_extent_item
);
6044 ei_r
= btrfs_item_ptr(leaf_r
,
6045 sctx
->right_path
->slots
[0],
6046 struct btrfs_file_extent_item
);
6049 * We may have found an extent item that has changed
6050 * only its disk_bytenr field and the corresponding
6051 * inode item was not updated. This case happens due to
6052 * very specific timings during relocation when a leaf
6053 * that contains file extent items is COWed while
6054 * relocation is ongoing and its in the stage where it
6055 * updates data pointers. So when this happens we can
6056 * safely ignore it since we know it's the same extent,
6057 * but just at different logical and physical locations
6058 * (when an extent is fully replaced with a new one, we
6059 * know the generation number must have changed too,
6060 * since snapshot creation implies committing the current
6061 * transaction, and the inode item must have been updated
6063 * This replacement of the disk_bytenr happens at
6064 * relocation.c:replace_file_extents() through
6065 * relocation.c:btrfs_reloc_cow_block().
6067 if (btrfs_file_extent_generation(leaf_l
, ei_l
) ==
6068 btrfs_file_extent_generation(leaf_r
, ei_r
) &&
6069 btrfs_file_extent_ram_bytes(leaf_l
, ei_l
) ==
6070 btrfs_file_extent_ram_bytes(leaf_r
, ei_r
) &&
6071 btrfs_file_extent_compression(leaf_l
, ei_l
) ==
6072 btrfs_file_extent_compression(leaf_r
, ei_r
) &&
6073 btrfs_file_extent_encryption(leaf_l
, ei_l
) ==
6074 btrfs_file_extent_encryption(leaf_r
, ei_r
) &&
6075 btrfs_file_extent_other_encoding(leaf_l
, ei_l
) ==
6076 btrfs_file_extent_other_encoding(leaf_r
, ei_r
) &&
6077 btrfs_file_extent_type(leaf_l
, ei_l
) ==
6078 btrfs_file_extent_type(leaf_r
, ei_r
) &&
6079 btrfs_file_extent_disk_bytenr(leaf_l
, ei_l
) !=
6080 btrfs_file_extent_disk_bytenr(leaf_r
, ei_r
) &&
6081 btrfs_file_extent_disk_num_bytes(leaf_l
, ei_l
) ==
6082 btrfs_file_extent_disk_num_bytes(leaf_r
, ei_r
) &&
6083 btrfs_file_extent_offset(leaf_l
, ei_l
) ==
6084 btrfs_file_extent_offset(leaf_r
, ei_r
) &&
6085 btrfs_file_extent_num_bytes(leaf_l
, ei_l
) ==
6086 btrfs_file_extent_num_bytes(leaf_r
, ei_r
))
6090 inconsistent_snapshot_error(sctx
, result
, "extent");
6094 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
6095 if (result
!= BTRFS_COMPARE_TREE_DELETED
)
6096 ret
= process_extent(sctx
, sctx
->left_path
,
6103 static int dir_changed(struct send_ctx
*sctx
, u64 dir
)
6105 u64 orig_gen
, new_gen
;
6108 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &new_gen
, NULL
, NULL
,
6113 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &orig_gen
, NULL
,
6118 return (orig_gen
!= new_gen
) ? 1 : 0;
6121 static int compare_refs(struct send_ctx
*sctx
, struct btrfs_path
*path
,
6122 struct btrfs_key
*key
)
6124 struct btrfs_inode_extref
*extref
;
6125 struct extent_buffer
*leaf
;
6126 u64 dirid
= 0, last_dirid
= 0;
6133 /* Easy case, just check this one dirid */
6134 if (key
->type
== BTRFS_INODE_REF_KEY
) {
6135 dirid
= key
->offset
;
6137 ret
= dir_changed(sctx
, dirid
);
6141 leaf
= path
->nodes
[0];
6142 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
6143 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
6144 while (cur_offset
< item_size
) {
6145 extref
= (struct btrfs_inode_extref
*)(ptr
+
6147 dirid
= btrfs_inode_extref_parent(leaf
, extref
);
6148 ref_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
6149 cur_offset
+= ref_name_len
+ sizeof(*extref
);
6150 if (dirid
== last_dirid
)
6152 ret
= dir_changed(sctx
, dirid
);
6162 * Updates compare related fields in sctx and simply forwards to the actual
6163 * changed_xxx functions.
6165 static int changed_cb(struct btrfs_root
*left_root
,
6166 struct btrfs_root
*right_root
,
6167 struct btrfs_path
*left_path
,
6168 struct btrfs_path
*right_path
,
6169 struct btrfs_key
*key
,
6170 enum btrfs_compare_tree_result result
,
6174 struct send_ctx
*sctx
= ctx
;
6176 if (result
== BTRFS_COMPARE_TREE_SAME
) {
6177 if (key
->type
== BTRFS_INODE_REF_KEY
||
6178 key
->type
== BTRFS_INODE_EXTREF_KEY
) {
6179 ret
= compare_refs(sctx
, left_path
, key
);
6184 } else if (key
->type
== BTRFS_EXTENT_DATA_KEY
) {
6185 return maybe_send_hole(sctx
, left_path
, key
);
6189 result
= BTRFS_COMPARE_TREE_CHANGED
;
6193 sctx
->left_path
= left_path
;
6194 sctx
->right_path
= right_path
;
6195 sctx
->cmp_key
= key
;
6197 ret
= finish_inode_if_needed(sctx
, 0);
6201 /* Ignore non-FS objects */
6202 if (key
->objectid
== BTRFS_FREE_INO_OBJECTID
||
6203 key
->objectid
== BTRFS_FREE_SPACE_OBJECTID
)
6206 if (key
->type
== BTRFS_INODE_ITEM_KEY
)
6207 ret
= changed_inode(sctx
, result
);
6208 else if (key
->type
== BTRFS_INODE_REF_KEY
||
6209 key
->type
== BTRFS_INODE_EXTREF_KEY
)
6210 ret
= changed_ref(sctx
, result
);
6211 else if (key
->type
== BTRFS_XATTR_ITEM_KEY
)
6212 ret
= changed_xattr(sctx
, result
);
6213 else if (key
->type
== BTRFS_EXTENT_DATA_KEY
)
6214 ret
= changed_extent(sctx
, result
);
6220 static int full_send_tree(struct send_ctx
*sctx
)
6223 struct btrfs_root
*send_root
= sctx
->send_root
;
6224 struct btrfs_key key
;
6225 struct btrfs_key found_key
;
6226 struct btrfs_path
*path
;
6227 struct extent_buffer
*eb
;
6230 path
= alloc_path_for_send();
6234 key
.objectid
= BTRFS_FIRST_FREE_OBJECTID
;
6235 key
.type
= BTRFS_INODE_ITEM_KEY
;
6238 ret
= btrfs_search_slot_for_read(send_root
, &key
, path
, 1, 0);
6245 eb
= path
->nodes
[0];
6246 slot
= path
->slots
[0];
6247 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6249 ret
= changed_cb(send_root
, NULL
, path
, NULL
,
6250 &found_key
, BTRFS_COMPARE_TREE_NEW
, sctx
);
6254 key
.objectid
= found_key
.objectid
;
6255 key
.type
= found_key
.type
;
6256 key
.offset
= found_key
.offset
+ 1;
6258 ret
= btrfs_next_item(send_root
, path
);
6268 ret
= finish_inode_if_needed(sctx
, 1);
6271 btrfs_free_path(path
);
6275 static int send_subvol(struct send_ctx
*sctx
)
6279 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_STREAM_HEADER
)) {
6280 ret
= send_header(sctx
);
6285 ret
= send_subvol_begin(sctx
);
6289 if (sctx
->parent_root
) {
6290 ret
= btrfs_compare_trees(sctx
->send_root
, sctx
->parent_root
,
6294 ret
= finish_inode_if_needed(sctx
, 1);
6298 ret
= full_send_tree(sctx
);
6304 free_recorded_refs(sctx
);
6309 * If orphan cleanup did remove any orphans from a root, it means the tree
6310 * was modified and therefore the commit root is not the same as the current
6311 * root anymore. This is a problem, because send uses the commit root and
6312 * therefore can see inode items that don't exist in the current root anymore,
6313 * and for example make calls to btrfs_iget, which will do tree lookups based
6314 * on the current root and not on the commit root. Those lookups will fail,
6315 * returning a -ESTALE error, and making send fail with that error. So make
6316 * sure a send does not see any orphans we have just removed, and that it will
6317 * see the same inodes regardless of whether a transaction commit happened
6318 * before it started (meaning that the commit root will be the same as the
6319 * current root) or not.
6321 static int ensure_commit_roots_uptodate(struct send_ctx
*sctx
)
6324 struct btrfs_trans_handle
*trans
= NULL
;
6327 if (sctx
->parent_root
&&
6328 sctx
->parent_root
->node
!= sctx
->parent_root
->commit_root
)
6331 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
6332 if (sctx
->clone_roots
[i
].root
->node
!=
6333 sctx
->clone_roots
[i
].root
->commit_root
)
6337 return btrfs_end_transaction(trans
);
6342 /* Use any root, all fs roots will get their commit roots updated. */
6344 trans
= btrfs_join_transaction(sctx
->send_root
);
6346 return PTR_ERR(trans
);
6350 return btrfs_commit_transaction(trans
);
6353 static void btrfs_root_dec_send_in_progress(struct btrfs_root
* root
)
6355 spin_lock(&root
->root_item_lock
);
6356 root
->send_in_progress
--;
6358 * Not much left to do, we don't know why it's unbalanced and
6359 * can't blindly reset it to 0.
6361 if (root
->send_in_progress
< 0)
6362 btrfs_err(root
->fs_info
,
6363 "send_in_progres unbalanced %d root %llu",
6364 root
->send_in_progress
, root
->root_key
.objectid
);
6365 spin_unlock(&root
->root_item_lock
);
6368 long btrfs_ioctl_send(struct file
*mnt_file
, void __user
*arg_
)
6371 struct btrfs_root
*send_root
= BTRFS_I(file_inode(mnt_file
))->root
;
6372 struct btrfs_fs_info
*fs_info
= send_root
->fs_info
;
6373 struct btrfs_root
*clone_root
;
6374 struct btrfs_ioctl_send_args
*arg
= NULL
;
6375 struct btrfs_key key
;
6376 struct send_ctx
*sctx
= NULL
;
6378 u64
*clone_sources_tmp
= NULL
;
6379 int clone_sources_to_rollback
= 0;
6380 unsigned alloc_size
;
6381 int sort_clone_roots
= 0;
6384 if (!capable(CAP_SYS_ADMIN
))
6388 * The subvolume must remain read-only during send, protect against
6389 * making it RW. This also protects against deletion.
6391 spin_lock(&send_root
->root_item_lock
);
6392 send_root
->send_in_progress
++;
6393 spin_unlock(&send_root
->root_item_lock
);
6396 * This is done when we lookup the root, it should already be complete
6397 * by the time we get here.
6399 WARN_ON(send_root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
);
6402 * Userspace tools do the checks and warn the user if it's
6405 if (!btrfs_root_readonly(send_root
)) {
6410 arg
= memdup_user(arg_
, sizeof(*arg
));
6418 * Check that we don't overflow at later allocations, we request
6419 * clone_sources_count + 1 items, and compare to unsigned long inside
6422 if (arg
->clone_sources_count
>
6423 ULONG_MAX
/ sizeof(struct clone_root
) - 1) {
6428 if (!access_ok(VERIFY_READ
, arg
->clone_sources
,
6429 sizeof(*arg
->clone_sources
) *
6430 arg
->clone_sources_count
)) {
6435 if (arg
->flags
& ~BTRFS_SEND_FLAG_MASK
) {
6440 sctx
= kzalloc(sizeof(struct send_ctx
), GFP_KERNEL
);
6446 INIT_LIST_HEAD(&sctx
->new_refs
);
6447 INIT_LIST_HEAD(&sctx
->deleted_refs
);
6448 INIT_RADIX_TREE(&sctx
->name_cache
, GFP_KERNEL
);
6449 INIT_LIST_HEAD(&sctx
->name_cache_list
);
6451 sctx
->flags
= arg
->flags
;
6453 sctx
->send_filp
= fget(arg
->send_fd
);
6454 if (!sctx
->send_filp
) {
6459 sctx
->send_root
= send_root
;
6461 * Unlikely but possible, if the subvolume is marked for deletion but
6462 * is slow to remove the directory entry, send can still be started
6464 if (btrfs_root_dead(sctx
->send_root
)) {
6469 sctx
->clone_roots_cnt
= arg
->clone_sources_count
;
6471 sctx
->send_max_size
= BTRFS_SEND_BUF_SIZE
;
6472 sctx
->send_buf
= kvmalloc(sctx
->send_max_size
, GFP_KERNEL
);
6473 if (!sctx
->send_buf
) {
6478 sctx
->read_buf
= kvmalloc(BTRFS_SEND_READ_SIZE
, GFP_KERNEL
);
6479 if (!sctx
->read_buf
) {
6484 sctx
->pending_dir_moves
= RB_ROOT
;
6485 sctx
->waiting_dir_moves
= RB_ROOT
;
6486 sctx
->orphan_dirs
= RB_ROOT
;
6488 alloc_size
= sizeof(struct clone_root
) * (arg
->clone_sources_count
+ 1);
6490 sctx
->clone_roots
= kzalloc(alloc_size
, GFP_KERNEL
);
6491 if (!sctx
->clone_roots
) {
6496 alloc_size
= arg
->clone_sources_count
* sizeof(*arg
->clone_sources
);
6498 if (arg
->clone_sources_count
) {
6499 clone_sources_tmp
= kvmalloc(alloc_size
, GFP_KERNEL
);
6500 if (!clone_sources_tmp
) {
6505 ret
= copy_from_user(clone_sources_tmp
, arg
->clone_sources
,
6512 for (i
= 0; i
< arg
->clone_sources_count
; i
++) {
6513 key
.objectid
= clone_sources_tmp
[i
];
6514 key
.type
= BTRFS_ROOT_ITEM_KEY
;
6515 key
.offset
= (u64
)-1;
6517 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
6519 clone_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
6520 if (IS_ERR(clone_root
)) {
6521 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6522 ret
= PTR_ERR(clone_root
);
6525 spin_lock(&clone_root
->root_item_lock
);
6526 if (!btrfs_root_readonly(clone_root
) ||
6527 btrfs_root_dead(clone_root
)) {
6528 spin_unlock(&clone_root
->root_item_lock
);
6529 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6533 clone_root
->send_in_progress
++;
6534 spin_unlock(&clone_root
->root_item_lock
);
6535 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6537 sctx
->clone_roots
[i
].root
= clone_root
;
6538 clone_sources_to_rollback
= i
+ 1;
6540 kvfree(clone_sources_tmp
);
6541 clone_sources_tmp
= NULL
;
6544 if (arg
->parent_root
) {
6545 key
.objectid
= arg
->parent_root
;
6546 key
.type
= BTRFS_ROOT_ITEM_KEY
;
6547 key
.offset
= (u64
)-1;
6549 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
6551 sctx
->parent_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
6552 if (IS_ERR(sctx
->parent_root
)) {
6553 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6554 ret
= PTR_ERR(sctx
->parent_root
);
6558 spin_lock(&sctx
->parent_root
->root_item_lock
);
6559 sctx
->parent_root
->send_in_progress
++;
6560 if (!btrfs_root_readonly(sctx
->parent_root
) ||
6561 btrfs_root_dead(sctx
->parent_root
)) {
6562 spin_unlock(&sctx
->parent_root
->root_item_lock
);
6563 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6567 spin_unlock(&sctx
->parent_root
->root_item_lock
);
6569 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6573 * Clones from send_root are allowed, but only if the clone source
6574 * is behind the current send position. This is checked while searching
6575 * for possible clone sources.
6577 sctx
->clone_roots
[sctx
->clone_roots_cnt
++].root
= sctx
->send_root
;
6579 /* We do a bsearch later */
6580 sort(sctx
->clone_roots
, sctx
->clone_roots_cnt
,
6581 sizeof(*sctx
->clone_roots
), __clone_root_cmp_sort
,
6583 sort_clone_roots
= 1;
6585 ret
= ensure_commit_roots_uptodate(sctx
);
6589 current
->journal_info
= BTRFS_SEND_TRANS_STUB
;
6590 ret
= send_subvol(sctx
);
6591 current
->journal_info
= NULL
;
6595 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_END_CMD
)) {
6596 ret
= begin_cmd(sctx
, BTRFS_SEND_C_END
);
6599 ret
= send_cmd(sctx
);
6605 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
));
6606 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
)) {
6608 struct pending_dir_move
*pm
;
6610 n
= rb_first(&sctx
->pending_dir_moves
);
6611 pm
= rb_entry(n
, struct pending_dir_move
, node
);
6612 while (!list_empty(&pm
->list
)) {
6613 struct pending_dir_move
*pm2
;
6615 pm2
= list_first_entry(&pm
->list
,
6616 struct pending_dir_move
, list
);
6617 free_pending_move(sctx
, pm2
);
6619 free_pending_move(sctx
, pm
);
6622 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
));
6623 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
)) {
6625 struct waiting_dir_move
*dm
;
6627 n
= rb_first(&sctx
->waiting_dir_moves
);
6628 dm
= rb_entry(n
, struct waiting_dir_move
, node
);
6629 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
6633 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
));
6634 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
)) {
6636 struct orphan_dir_info
*odi
;
6638 n
= rb_first(&sctx
->orphan_dirs
);
6639 odi
= rb_entry(n
, struct orphan_dir_info
, node
);
6640 free_orphan_dir_info(sctx
, odi
);
6643 if (sort_clone_roots
) {
6644 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
6645 btrfs_root_dec_send_in_progress(
6646 sctx
->clone_roots
[i
].root
);
6648 for (i
= 0; sctx
&& i
< clone_sources_to_rollback
; i
++)
6649 btrfs_root_dec_send_in_progress(
6650 sctx
->clone_roots
[i
].root
);
6652 btrfs_root_dec_send_in_progress(send_root
);
6654 if (sctx
&& !IS_ERR_OR_NULL(sctx
->parent_root
))
6655 btrfs_root_dec_send_in_progress(sctx
->parent_root
);
6658 kvfree(clone_sources_tmp
);
6661 if (sctx
->send_filp
)
6662 fput(sctx
->send_filp
);
6664 kvfree(sctx
->clone_roots
);
6665 kvfree(sctx
->send_buf
);
6666 kvfree(sctx
->read_buf
);
6668 name_cache_free(sctx
);