2 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
35 #include "btrfs_inode.h"
36 #include "transaction.h"
38 static int g_verbose
= 0;
40 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
58 unsigned int reversed
:1;
59 unsigned int virtual_mem
:1;
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
;
112 u64 cur_inode_last_extent
;
116 struct list_head new_refs
;
117 struct list_head deleted_refs
;
119 struct radix_tree_root name_cache
;
120 struct list_head name_cache_list
;
126 * We process inodes by their increasing order, so if before an
127 * incremental send we reverse the parent/child relationship of
128 * directories such that a directory with a lower inode number was
129 * the parent of a directory with a higher inode number, and the one
130 * becoming the new parent got renamed too, we can't rename/move the
131 * directory with lower inode number when we finish processing it - we
132 * must process the directory with higher inode number first, then
133 * rename/move it and then rename/move the directory with lower inode
134 * number. Example follows.
136 * Tree state when the first send was performed:
148 * Tree state when the second (incremental) send is performed:
157 * The sequence of steps that lead to the second state was:
159 * mv /a/b/c/d /a/b/c2/d2
160 * mv /a/b/c /a/b/c2/d2/cc
162 * "c" has lower inode number, but we can't move it (2nd mv operation)
163 * before we move "d", which has higher inode number.
165 * So we just memorize which move/rename operations must be performed
166 * later when their respective parent is processed and moved/renamed.
169 /* Indexed by parent directory inode number. */
170 struct rb_root pending_dir_moves
;
173 * Reverse index, indexed by the inode number of a directory that
174 * is waiting for the move/rename of its immediate parent before its
175 * own move/rename can be performed.
177 struct rb_root waiting_dir_moves
;
180 struct pending_dir_move
{
182 struct list_head list
;
186 struct list_head update_refs
;
189 struct waiting_dir_move
{
194 struct name_cache_entry
{
195 struct list_head list
;
197 * radix_tree has only 32bit entries but we need to handle 64bit inums.
198 * We use the lower 32bit of the 64bit inum to store it in the tree. If
199 * more then one inum would fall into the same entry, we use radix_list
200 * to store the additional entries. radix_list is also used to store
201 * entries where two entries have the same inum but different
204 struct list_head radix_list
;
210 int need_later_update
;
215 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
);
217 static int need_send_hole(struct send_ctx
*sctx
)
219 return (sctx
->parent_root
&& !sctx
->cur_inode_new
&&
220 !sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
&&
221 S_ISREG(sctx
->cur_inode_mode
));
224 static void fs_path_reset(struct fs_path
*p
)
227 p
->start
= p
->buf
+ p
->buf_len
- 1;
237 static struct fs_path
*fs_path_alloc(void)
241 p
= kmalloc(sizeof(*p
), GFP_NOFS
);
246 p
->buf
= p
->inline_buf
;
247 p
->buf_len
= FS_PATH_INLINE_SIZE
;
252 static struct fs_path
*fs_path_alloc_reversed(void)
264 static void fs_path_free(struct fs_path
*p
)
268 if (p
->buf
!= p
->inline_buf
) {
277 static int fs_path_len(struct fs_path
*p
)
279 return p
->end
- p
->start
;
282 static int fs_path_ensure_buf(struct fs_path
*p
, int len
)
290 if (p
->buf_len
>= len
)
293 path_len
= p
->end
- p
->start
;
294 old_buf_len
= p
->buf_len
;
295 len
= PAGE_ALIGN(len
);
297 if (p
->buf
== p
->inline_buf
) {
298 tmp_buf
= kmalloc(len
, GFP_NOFS
| __GFP_NOWARN
);
300 tmp_buf
= vmalloc(len
);
305 memcpy(tmp_buf
, p
->buf
, p
->buf_len
);
309 if (p
->virtual_mem
) {
310 tmp_buf
= vmalloc(len
);
313 memcpy(tmp_buf
, p
->buf
, p
->buf_len
);
316 tmp_buf
= krealloc(p
->buf
, len
, GFP_NOFS
);
318 tmp_buf
= vmalloc(len
);
321 memcpy(tmp_buf
, p
->buf
, p
->buf_len
);
330 tmp_buf
= p
->buf
+ old_buf_len
- path_len
- 1;
331 p
->end
= p
->buf
+ p
->buf_len
- 1;
332 p
->start
= p
->end
- path_len
;
333 memmove(p
->start
, tmp_buf
, path_len
+ 1);
336 p
->end
= p
->start
+ path_len
;
341 static int fs_path_prepare_for_add(struct fs_path
*p
, int name_len
)
346 new_len
= p
->end
- p
->start
+ name_len
;
347 if (p
->start
!= p
->end
)
349 ret
= fs_path_ensure_buf(p
, new_len
);
354 if (p
->start
!= p
->end
)
356 p
->start
-= name_len
;
357 p
->prepared
= p
->start
;
359 if (p
->start
!= p
->end
)
361 p
->prepared
= p
->end
;
370 static int fs_path_add(struct fs_path
*p
, const char *name
, int name_len
)
374 ret
= fs_path_prepare_for_add(p
, name_len
);
377 memcpy(p
->prepared
, name
, name_len
);
384 static int fs_path_add_path(struct fs_path
*p
, struct fs_path
*p2
)
388 ret
= fs_path_prepare_for_add(p
, p2
->end
- p2
->start
);
391 memcpy(p
->prepared
, p2
->start
, p2
->end
- p2
->start
);
398 static int fs_path_add_from_extent_buffer(struct fs_path
*p
,
399 struct extent_buffer
*eb
,
400 unsigned long off
, int len
)
404 ret
= fs_path_prepare_for_add(p
, len
);
408 read_extent_buffer(eb
, p
->prepared
, off
, len
);
415 static int fs_path_copy(struct fs_path
*p
, struct fs_path
*from
)
419 p
->reversed
= from
->reversed
;
422 ret
= fs_path_add_path(p
, from
);
428 static void fs_path_unreverse(struct fs_path
*p
)
437 len
= p
->end
- p
->start
;
439 p
->end
= p
->start
+ len
;
440 memmove(p
->start
, tmp
, len
+ 1);
444 static struct btrfs_path
*alloc_path_for_send(void)
446 struct btrfs_path
*path
;
448 path
= btrfs_alloc_path();
451 path
->search_commit_root
= 1;
452 path
->skip_locking
= 1;
456 static int write_buf(struct file
*filp
, const void *buf
, u32 len
, loff_t
*off
)
466 ret
= vfs_write(filp
, (char *)buf
+ pos
, len
- pos
, off
);
467 /* TODO handle that correctly */
468 /*if (ret == -ERESTARTSYS) {
487 static int tlv_put(struct send_ctx
*sctx
, u16 attr
, const void *data
, int len
)
489 struct btrfs_tlv_header
*hdr
;
490 int total_len
= sizeof(*hdr
) + len
;
491 int left
= sctx
->send_max_size
- sctx
->send_size
;
493 if (unlikely(left
< total_len
))
496 hdr
= (struct btrfs_tlv_header
*) (sctx
->send_buf
+ sctx
->send_size
);
497 hdr
->tlv_type
= cpu_to_le16(attr
);
498 hdr
->tlv_len
= cpu_to_le16(len
);
499 memcpy(hdr
+ 1, data
, len
);
500 sctx
->send_size
+= total_len
;
505 #define TLV_PUT_DEFINE_INT(bits) \
506 static int tlv_put_u##bits(struct send_ctx *sctx, \
507 u##bits attr, u##bits value) \
509 __le##bits __tmp = cpu_to_le##bits(value); \
510 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
513 TLV_PUT_DEFINE_INT(64)
515 static int tlv_put_string(struct send_ctx
*sctx
, u16 attr
,
516 const char *str
, int len
)
520 return tlv_put(sctx
, attr
, str
, len
);
523 static int tlv_put_uuid(struct send_ctx
*sctx
, u16 attr
,
526 return tlv_put(sctx
, attr
, uuid
, BTRFS_UUID_SIZE
);
529 static int tlv_put_btrfs_timespec(struct send_ctx
*sctx
, u16 attr
,
530 struct extent_buffer
*eb
,
531 struct btrfs_timespec
*ts
)
533 struct btrfs_timespec bts
;
534 read_extent_buffer(eb
, &bts
, (unsigned long)ts
, sizeof(bts
));
535 return tlv_put(sctx
, attr
, &bts
, sizeof(bts
));
539 #define TLV_PUT(sctx, attrtype, attrlen, data) \
541 ret = tlv_put(sctx, attrtype, attrlen, data); \
543 goto tlv_put_failure; \
546 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
548 ret = tlv_put_u##bits(sctx, attrtype, value); \
550 goto tlv_put_failure; \
553 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
554 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
555 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
556 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
557 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
559 ret = tlv_put_string(sctx, attrtype, str, len); \
561 goto tlv_put_failure; \
563 #define TLV_PUT_PATH(sctx, attrtype, p) \
565 ret = tlv_put_string(sctx, attrtype, p->start, \
566 p->end - p->start); \
568 goto tlv_put_failure; \
570 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
572 ret = tlv_put_uuid(sctx, attrtype, uuid); \
574 goto tlv_put_failure; \
576 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
578 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
580 goto tlv_put_failure; \
583 static int send_header(struct send_ctx
*sctx
)
585 struct btrfs_stream_header hdr
;
587 strcpy(hdr
.magic
, BTRFS_SEND_STREAM_MAGIC
);
588 hdr
.version
= cpu_to_le32(BTRFS_SEND_STREAM_VERSION
);
590 return write_buf(sctx
->send_filp
, &hdr
, sizeof(hdr
),
595 * For each command/item we want to send to userspace, we call this function.
597 static int begin_cmd(struct send_ctx
*sctx
, int cmd
)
599 struct btrfs_cmd_header
*hdr
;
601 if (WARN_ON(!sctx
->send_buf
))
604 BUG_ON(sctx
->send_size
);
606 sctx
->send_size
+= sizeof(*hdr
);
607 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
608 hdr
->cmd
= cpu_to_le16(cmd
);
613 static int send_cmd(struct send_ctx
*sctx
)
616 struct btrfs_cmd_header
*hdr
;
619 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
620 hdr
->len
= cpu_to_le32(sctx
->send_size
- sizeof(*hdr
));
623 crc
= btrfs_crc32c(0, (unsigned char *)sctx
->send_buf
, sctx
->send_size
);
624 hdr
->crc
= cpu_to_le32(crc
);
626 ret
= write_buf(sctx
->send_filp
, sctx
->send_buf
, sctx
->send_size
,
629 sctx
->total_send_size
+= sctx
->send_size
;
630 sctx
->cmd_send_size
[le16_to_cpu(hdr
->cmd
)] += sctx
->send_size
;
637 * Sends a move instruction to user space
639 static int send_rename(struct send_ctx
*sctx
,
640 struct fs_path
*from
, struct fs_path
*to
)
644 verbose_printk("btrfs: send_rename %s -> %s\n", from
->start
, to
->start
);
646 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RENAME
);
650 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, from
);
651 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_TO
, to
);
653 ret
= send_cmd(sctx
);
661 * Sends a link instruction to user space
663 static int send_link(struct send_ctx
*sctx
,
664 struct fs_path
*path
, struct fs_path
*lnk
)
668 verbose_printk("btrfs: send_link %s -> %s\n", path
->start
, lnk
->start
);
670 ret
= begin_cmd(sctx
, BTRFS_SEND_C_LINK
);
674 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
675 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, lnk
);
677 ret
= send_cmd(sctx
);
685 * Sends an unlink instruction to user space
687 static int send_unlink(struct send_ctx
*sctx
, struct fs_path
*path
)
691 verbose_printk("btrfs: send_unlink %s\n", path
->start
);
693 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UNLINK
);
697 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
699 ret
= send_cmd(sctx
);
707 * Sends a rmdir instruction to user space
709 static int send_rmdir(struct send_ctx
*sctx
, struct fs_path
*path
)
713 verbose_printk("btrfs: send_rmdir %s\n", path
->start
);
715 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RMDIR
);
719 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
721 ret
= send_cmd(sctx
);
729 * Helper function to retrieve some fields from an inode item.
731 static int get_inode_info(struct btrfs_root
*root
,
732 u64 ino
, u64
*size
, u64
*gen
,
733 u64
*mode
, u64
*uid
, u64
*gid
,
737 struct btrfs_inode_item
*ii
;
738 struct btrfs_key key
;
739 struct btrfs_path
*path
;
741 path
= alloc_path_for_send();
746 key
.type
= BTRFS_INODE_ITEM_KEY
;
748 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
756 ii
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
757 struct btrfs_inode_item
);
759 *size
= btrfs_inode_size(path
->nodes
[0], ii
);
761 *gen
= btrfs_inode_generation(path
->nodes
[0], ii
);
763 *mode
= btrfs_inode_mode(path
->nodes
[0], ii
);
765 *uid
= btrfs_inode_uid(path
->nodes
[0], ii
);
767 *gid
= btrfs_inode_gid(path
->nodes
[0], ii
);
769 *rdev
= btrfs_inode_rdev(path
->nodes
[0], ii
);
772 btrfs_free_path(path
);
776 typedef int (*iterate_inode_ref_t
)(int num
, u64 dir
, int index
,
781 * Helper function to iterate the entries in ONE btrfs_inode_ref or
782 * btrfs_inode_extref.
783 * The iterate callback may return a non zero value to stop iteration. This can
784 * be a negative value for error codes or 1 to simply stop it.
786 * path must point to the INODE_REF or INODE_EXTREF when called.
788 static int iterate_inode_ref(struct btrfs_root
*root
, struct btrfs_path
*path
,
789 struct btrfs_key
*found_key
, int resolve
,
790 iterate_inode_ref_t iterate
, void *ctx
)
792 struct extent_buffer
*eb
= path
->nodes
[0];
793 struct btrfs_item
*item
;
794 struct btrfs_inode_ref
*iref
;
795 struct btrfs_inode_extref
*extref
;
796 struct btrfs_path
*tmp_path
;
800 int slot
= path
->slots
[0];
807 unsigned long name_off
;
808 unsigned long elem_size
;
811 p
= fs_path_alloc_reversed();
815 tmp_path
= alloc_path_for_send();
822 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
823 ptr
= (unsigned long)btrfs_item_ptr(eb
, slot
,
824 struct btrfs_inode_ref
);
825 item
= btrfs_item_nr(slot
);
826 total
= btrfs_item_size(eb
, item
);
827 elem_size
= sizeof(*iref
);
829 ptr
= btrfs_item_ptr_offset(eb
, slot
);
830 total
= btrfs_item_size_nr(eb
, slot
);
831 elem_size
= sizeof(*extref
);
834 while (cur
< total
) {
837 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
838 iref
= (struct btrfs_inode_ref
*)(ptr
+ cur
);
839 name_len
= btrfs_inode_ref_name_len(eb
, iref
);
840 name_off
= (unsigned long)(iref
+ 1);
841 index
= btrfs_inode_ref_index(eb
, iref
);
842 dir
= found_key
->offset
;
844 extref
= (struct btrfs_inode_extref
*)(ptr
+ cur
);
845 name_len
= btrfs_inode_extref_name_len(eb
, extref
);
846 name_off
= (unsigned long)&extref
->name
;
847 index
= btrfs_inode_extref_index(eb
, extref
);
848 dir
= btrfs_inode_extref_parent(eb
, extref
);
852 start
= btrfs_ref_to_path(root
, tmp_path
, name_len
,
856 ret
= PTR_ERR(start
);
859 if (start
< p
->buf
) {
860 /* overflow , try again with larger buffer */
861 ret
= fs_path_ensure_buf(p
,
862 p
->buf_len
+ p
->buf
- start
);
865 start
= btrfs_ref_to_path(root
, tmp_path
,
870 ret
= PTR_ERR(start
);
873 BUG_ON(start
< p
->buf
);
877 ret
= fs_path_add_from_extent_buffer(p
, eb
, name_off
,
883 cur
+= elem_size
+ name_len
;
884 ret
= iterate(num
, dir
, index
, p
, ctx
);
891 btrfs_free_path(tmp_path
);
896 typedef int (*iterate_dir_item_t
)(int num
, struct btrfs_key
*di_key
,
897 const char *name
, int name_len
,
898 const char *data
, int data_len
,
902 * Helper function to iterate the entries in ONE btrfs_dir_item.
903 * The iterate callback may return a non zero value to stop iteration. This can
904 * be a negative value for error codes or 1 to simply stop it.
906 * path must point to the dir item when called.
908 static int iterate_dir_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
909 struct btrfs_key
*found_key
,
910 iterate_dir_item_t iterate
, void *ctx
)
913 struct extent_buffer
*eb
;
914 struct btrfs_item
*item
;
915 struct btrfs_dir_item
*di
;
916 struct btrfs_key di_key
;
931 buf
= kmalloc(buf_len
, GFP_NOFS
);
938 slot
= path
->slots
[0];
939 item
= btrfs_item_nr(slot
);
940 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
943 total
= btrfs_item_size(eb
, item
);
946 while (cur
< total
) {
947 name_len
= btrfs_dir_name_len(eb
, di
);
948 data_len
= btrfs_dir_data_len(eb
, di
);
949 type
= btrfs_dir_type(eb
, di
);
950 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
952 if (name_len
+ data_len
> buf_len
) {
953 buf_len
= PAGE_ALIGN(name_len
+ data_len
);
955 buf2
= vmalloc(buf_len
);
962 buf2
= krealloc(buf
, buf_len
, GFP_NOFS
);
964 buf2
= vmalloc(buf_len
);
978 read_extent_buffer(eb
, buf
, (unsigned long)(di
+ 1),
979 name_len
+ data_len
);
981 len
= sizeof(*di
) + name_len
+ data_len
;
982 di
= (struct btrfs_dir_item
*)((char *)di
+ len
);
985 ret
= iterate(num
, &di_key
, buf
, name_len
, buf
+ name_len
,
986 data_len
, type
, ctx
);
1005 static int __copy_first_ref(int num
, u64 dir
, int index
,
1006 struct fs_path
*p
, void *ctx
)
1009 struct fs_path
*pt
= ctx
;
1011 ret
= fs_path_copy(pt
, p
);
1015 /* we want the first only */
1020 * Retrieve the first path of an inode. If an inode has more then one
1021 * ref/hardlink, this is ignored.
1023 static int get_inode_path(struct btrfs_root
*root
,
1024 u64 ino
, struct fs_path
*path
)
1027 struct btrfs_key key
, found_key
;
1028 struct btrfs_path
*p
;
1030 p
= alloc_path_for_send();
1034 fs_path_reset(path
);
1037 key
.type
= BTRFS_INODE_REF_KEY
;
1040 ret
= btrfs_search_slot_for_read(root
, &key
, p
, 1, 0);
1047 btrfs_item_key_to_cpu(p
->nodes
[0], &found_key
, p
->slots
[0]);
1048 if (found_key
.objectid
!= ino
||
1049 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1050 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1055 ret
= iterate_inode_ref(root
, p
, &found_key
, 1,
1056 __copy_first_ref
, path
);
1066 struct backref_ctx
{
1067 struct send_ctx
*sctx
;
1069 /* number of total found references */
1073 * used for clones found in send_root. clones found behind cur_objectid
1074 * and cur_offset are not considered as allowed clones.
1079 /* may be truncated in case it's the last extent in a file */
1082 /* Just to check for bugs in backref resolving */
1086 static int __clone_root_cmp_bsearch(const void *key
, const void *elt
)
1088 u64 root
= (u64
)(uintptr_t)key
;
1089 struct clone_root
*cr
= (struct clone_root
*)elt
;
1091 if (root
< cr
->root
->objectid
)
1093 if (root
> cr
->root
->objectid
)
1098 static int __clone_root_cmp_sort(const void *e1
, const void *e2
)
1100 struct clone_root
*cr1
= (struct clone_root
*)e1
;
1101 struct clone_root
*cr2
= (struct clone_root
*)e2
;
1103 if (cr1
->root
->objectid
< cr2
->root
->objectid
)
1105 if (cr1
->root
->objectid
> cr2
->root
->objectid
)
1111 * Called for every backref that is found for the current extent.
1112 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1114 static int __iterate_backrefs(u64 ino
, u64 offset
, u64 root
, void *ctx_
)
1116 struct backref_ctx
*bctx
= ctx_
;
1117 struct clone_root
*found
;
1121 /* First check if the root is in the list of accepted clone sources */
1122 found
= bsearch((void *)(uintptr_t)root
, bctx
->sctx
->clone_roots
,
1123 bctx
->sctx
->clone_roots_cnt
,
1124 sizeof(struct clone_root
),
1125 __clone_root_cmp_bsearch
);
1129 if (found
->root
== bctx
->sctx
->send_root
&&
1130 ino
== bctx
->cur_objectid
&&
1131 offset
== bctx
->cur_offset
) {
1132 bctx
->found_itself
= 1;
1136 * There are inodes that have extents that lie behind its i_size. Don't
1137 * accept clones from these extents.
1139 ret
= get_inode_info(found
->root
, ino
, &i_size
, NULL
, NULL
, NULL
, NULL
,
1144 if (offset
+ bctx
->extent_len
> i_size
)
1148 * Make sure we don't consider clones from send_root that are
1149 * behind the current inode/offset.
1151 if (found
->root
== bctx
->sctx
->send_root
) {
1153 * TODO for the moment we don't accept clones from the inode
1154 * that is currently send. We may change this when
1155 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1158 if (ino
>= bctx
->cur_objectid
)
1161 if (ino
> bctx
->cur_objectid
)
1163 if (offset
+ bctx
->extent_len
> bctx
->cur_offset
)
1169 found
->found_refs
++;
1170 if (ino
< found
->ino
) {
1172 found
->offset
= offset
;
1173 } else if (found
->ino
== ino
) {
1175 * same extent found more then once in the same file.
1177 if (found
->offset
> offset
+ bctx
->extent_len
)
1178 found
->offset
= offset
;
1185 * Given an inode, offset and extent item, it finds a good clone for a clone
1186 * instruction. Returns -ENOENT when none could be found. The function makes
1187 * sure that the returned clone is usable at the point where sending is at the
1188 * moment. This means, that no clones are accepted which lie behind the current
1191 * path must point to the extent item when called.
1193 static int find_extent_clone(struct send_ctx
*sctx
,
1194 struct btrfs_path
*path
,
1195 u64 ino
, u64 data_offset
,
1197 struct clone_root
**found
)
1204 u64 extent_item_pos
;
1206 struct btrfs_file_extent_item
*fi
;
1207 struct extent_buffer
*eb
= path
->nodes
[0];
1208 struct backref_ctx
*backref_ctx
= NULL
;
1209 struct clone_root
*cur_clone_root
;
1210 struct btrfs_key found_key
;
1211 struct btrfs_path
*tmp_path
;
1215 tmp_path
= alloc_path_for_send();
1219 backref_ctx
= kmalloc(sizeof(*backref_ctx
), GFP_NOFS
);
1225 if (data_offset
>= ino_size
) {
1227 * There may be extents that lie behind the file's size.
1228 * I at least had this in combination with snapshotting while
1229 * writing large files.
1235 fi
= btrfs_item_ptr(eb
, path
->slots
[0],
1236 struct btrfs_file_extent_item
);
1237 extent_type
= btrfs_file_extent_type(eb
, fi
);
1238 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1242 compressed
= btrfs_file_extent_compression(eb
, fi
);
1244 num_bytes
= btrfs_file_extent_num_bytes(eb
, fi
);
1245 disk_byte
= btrfs_file_extent_disk_bytenr(eb
, fi
);
1246 if (disk_byte
== 0) {
1250 logical
= disk_byte
+ btrfs_file_extent_offset(eb
, fi
);
1252 ret
= extent_from_logical(sctx
->send_root
->fs_info
, disk_byte
, tmp_path
,
1253 &found_key
, &flags
);
1254 btrfs_release_path(tmp_path
);
1258 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1264 * Setup the clone roots.
1266 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1267 cur_clone_root
= sctx
->clone_roots
+ i
;
1268 cur_clone_root
->ino
= (u64
)-1;
1269 cur_clone_root
->offset
= 0;
1270 cur_clone_root
->found_refs
= 0;
1273 backref_ctx
->sctx
= sctx
;
1274 backref_ctx
->found
= 0;
1275 backref_ctx
->cur_objectid
= ino
;
1276 backref_ctx
->cur_offset
= data_offset
;
1277 backref_ctx
->found_itself
= 0;
1278 backref_ctx
->extent_len
= num_bytes
;
1281 * The last extent of a file may be too large due to page alignment.
1282 * We need to adjust extent_len in this case so that the checks in
1283 * __iterate_backrefs work.
1285 if (data_offset
+ num_bytes
>= ino_size
)
1286 backref_ctx
->extent_len
= ino_size
- data_offset
;
1289 * Now collect all backrefs.
1291 if (compressed
== BTRFS_COMPRESS_NONE
)
1292 extent_item_pos
= logical
- found_key
.objectid
;
1294 extent_item_pos
= 0;
1296 extent_item_pos
= logical
- found_key
.objectid
;
1297 ret
= iterate_extent_inodes(sctx
->send_root
->fs_info
,
1298 found_key
.objectid
, extent_item_pos
, 1,
1299 __iterate_backrefs
, backref_ctx
);
1304 if (!backref_ctx
->found_itself
) {
1305 /* found a bug in backref code? */
1307 btrfs_err(sctx
->send_root
->fs_info
, "did not find backref in "
1308 "send_root. inode=%llu, offset=%llu, "
1309 "disk_byte=%llu found extent=%llu\n",
1310 ino
, data_offset
, disk_byte
, found_key
.objectid
);
1314 verbose_printk(KERN_DEBUG
"btrfs: find_extent_clone: data_offset=%llu, "
1316 "num_bytes=%llu, logical=%llu\n",
1317 data_offset
, ino
, num_bytes
, logical
);
1319 if (!backref_ctx
->found
)
1320 verbose_printk("btrfs: no clones found\n");
1322 cur_clone_root
= NULL
;
1323 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1324 if (sctx
->clone_roots
[i
].found_refs
) {
1325 if (!cur_clone_root
)
1326 cur_clone_root
= sctx
->clone_roots
+ i
;
1327 else if (sctx
->clone_roots
[i
].root
== sctx
->send_root
)
1328 /* prefer clones from send_root over others */
1329 cur_clone_root
= sctx
->clone_roots
+ i
;
1334 if (cur_clone_root
) {
1335 *found
= cur_clone_root
;
1342 btrfs_free_path(tmp_path
);
1347 static int read_symlink(struct btrfs_root
*root
,
1349 struct fs_path
*dest
)
1352 struct btrfs_path
*path
;
1353 struct btrfs_key key
;
1354 struct btrfs_file_extent_item
*ei
;
1360 path
= alloc_path_for_send();
1365 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1367 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1372 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1373 struct btrfs_file_extent_item
);
1374 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
1375 compression
= btrfs_file_extent_compression(path
->nodes
[0], ei
);
1376 BUG_ON(type
!= BTRFS_FILE_EXTENT_INLINE
);
1377 BUG_ON(compression
);
1379 off
= btrfs_file_extent_inline_start(ei
);
1380 len
= btrfs_file_extent_inline_len(path
->nodes
[0], path
->slots
[0], ei
);
1382 ret
= fs_path_add_from_extent_buffer(dest
, path
->nodes
[0], off
, len
);
1385 btrfs_free_path(path
);
1390 * Helper function to generate a file name that is unique in the root of
1391 * send_root and parent_root. This is used to generate names for orphan inodes.
1393 static int gen_unique_name(struct send_ctx
*sctx
,
1395 struct fs_path
*dest
)
1398 struct btrfs_path
*path
;
1399 struct btrfs_dir_item
*di
;
1404 path
= alloc_path_for_send();
1409 len
= snprintf(tmp
, sizeof(tmp
), "o%llu-%llu-%llu",
1411 if (len
>= sizeof(tmp
)) {
1412 /* should really not happen */
1417 di
= btrfs_lookup_dir_item(NULL
, sctx
->send_root
,
1418 path
, BTRFS_FIRST_FREE_OBJECTID
,
1419 tmp
, strlen(tmp
), 0);
1420 btrfs_release_path(path
);
1426 /* not unique, try again */
1431 if (!sctx
->parent_root
) {
1437 di
= btrfs_lookup_dir_item(NULL
, sctx
->parent_root
,
1438 path
, BTRFS_FIRST_FREE_OBJECTID
,
1439 tmp
, strlen(tmp
), 0);
1440 btrfs_release_path(path
);
1446 /* not unique, try again */
1454 ret
= fs_path_add(dest
, tmp
, strlen(tmp
));
1457 btrfs_free_path(path
);
1462 inode_state_no_change
,
1463 inode_state_will_create
,
1464 inode_state_did_create
,
1465 inode_state_will_delete
,
1466 inode_state_did_delete
,
1469 static int get_cur_inode_state(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1477 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &left_gen
, NULL
, NULL
,
1479 if (ret
< 0 && ret
!= -ENOENT
)
1483 if (!sctx
->parent_root
) {
1484 right_ret
= -ENOENT
;
1486 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
, &right_gen
,
1487 NULL
, NULL
, NULL
, NULL
);
1488 if (ret
< 0 && ret
!= -ENOENT
)
1493 if (!left_ret
&& !right_ret
) {
1494 if (left_gen
== gen
&& right_gen
== gen
) {
1495 ret
= inode_state_no_change
;
1496 } else if (left_gen
== gen
) {
1497 if (ino
< sctx
->send_progress
)
1498 ret
= inode_state_did_create
;
1500 ret
= inode_state_will_create
;
1501 } else if (right_gen
== gen
) {
1502 if (ino
< sctx
->send_progress
)
1503 ret
= inode_state_did_delete
;
1505 ret
= inode_state_will_delete
;
1509 } else if (!left_ret
) {
1510 if (left_gen
== gen
) {
1511 if (ino
< sctx
->send_progress
)
1512 ret
= inode_state_did_create
;
1514 ret
= inode_state_will_create
;
1518 } else if (!right_ret
) {
1519 if (right_gen
== gen
) {
1520 if (ino
< sctx
->send_progress
)
1521 ret
= inode_state_did_delete
;
1523 ret
= inode_state_will_delete
;
1535 static int is_inode_existent(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1539 ret
= get_cur_inode_state(sctx
, ino
, gen
);
1543 if (ret
== inode_state_no_change
||
1544 ret
== inode_state_did_create
||
1545 ret
== inode_state_will_delete
)
1555 * Helper function to lookup a dir item in a dir.
1557 static int lookup_dir_item_inode(struct btrfs_root
*root
,
1558 u64 dir
, const char *name
, int name_len
,
1563 struct btrfs_dir_item
*di
;
1564 struct btrfs_key key
;
1565 struct btrfs_path
*path
;
1567 path
= alloc_path_for_send();
1571 di
= btrfs_lookup_dir_item(NULL
, root
, path
,
1572 dir
, name
, name_len
, 0);
1581 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
1582 *found_inode
= key
.objectid
;
1583 *found_type
= btrfs_dir_type(path
->nodes
[0], di
);
1586 btrfs_free_path(path
);
1591 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1592 * generation of the parent dir and the name of the dir entry.
1594 static int get_first_ref(struct btrfs_root
*root
, u64 ino
,
1595 u64
*dir
, u64
*dir_gen
, struct fs_path
*name
)
1598 struct btrfs_key key
;
1599 struct btrfs_key found_key
;
1600 struct btrfs_path
*path
;
1604 path
= alloc_path_for_send();
1609 key
.type
= BTRFS_INODE_REF_KEY
;
1612 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
1616 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1618 if (ret
|| found_key
.objectid
!= ino
||
1619 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1620 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1625 if (key
.type
== BTRFS_INODE_REF_KEY
) {
1626 struct btrfs_inode_ref
*iref
;
1627 iref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1628 struct btrfs_inode_ref
);
1629 len
= btrfs_inode_ref_name_len(path
->nodes
[0], iref
);
1630 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1631 (unsigned long)(iref
+ 1),
1633 parent_dir
= found_key
.offset
;
1635 struct btrfs_inode_extref
*extref
;
1636 extref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1637 struct btrfs_inode_extref
);
1638 len
= btrfs_inode_extref_name_len(path
->nodes
[0], extref
);
1639 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1640 (unsigned long)&extref
->name
, len
);
1641 parent_dir
= btrfs_inode_extref_parent(path
->nodes
[0], extref
);
1645 btrfs_release_path(path
);
1647 ret
= get_inode_info(root
, parent_dir
, NULL
, dir_gen
, NULL
, NULL
,
1655 btrfs_free_path(path
);
1659 static int is_first_ref(struct btrfs_root
*root
,
1661 const char *name
, int name_len
)
1664 struct fs_path
*tmp_name
;
1668 tmp_name
= fs_path_alloc();
1672 ret
= get_first_ref(root
, ino
, &tmp_dir
, &tmp_dir_gen
, tmp_name
);
1676 if (dir
!= tmp_dir
|| name_len
!= fs_path_len(tmp_name
)) {
1681 ret
= !memcmp(tmp_name
->start
, name
, name_len
);
1684 fs_path_free(tmp_name
);
1689 * Used by process_recorded_refs to determine if a new ref would overwrite an
1690 * already existing ref. In case it detects an overwrite, it returns the
1691 * inode/gen in who_ino/who_gen.
1692 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1693 * to make sure later references to the overwritten inode are possible.
1694 * Orphanizing is however only required for the first ref of an inode.
1695 * process_recorded_refs does an additional is_first_ref check to see if
1696 * orphanizing is really required.
1698 static int will_overwrite_ref(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
1699 const char *name
, int name_len
,
1700 u64
*who_ino
, u64
*who_gen
)
1704 u64 other_inode
= 0;
1707 if (!sctx
->parent_root
)
1710 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1715 * If we have a parent root we need to verify that the parent dir was
1716 * not delted and then re-created, if it was then we have no overwrite
1717 * and we can just unlink this entry.
1719 if (sctx
->parent_root
) {
1720 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &gen
, NULL
,
1722 if (ret
< 0 && ret
!= -ENOENT
)
1732 ret
= lookup_dir_item_inode(sctx
->parent_root
, dir
, name
, name_len
,
1733 &other_inode
, &other_type
);
1734 if (ret
< 0 && ret
!= -ENOENT
)
1742 * Check if the overwritten ref was already processed. If yes, the ref
1743 * was already unlinked/moved, so we can safely assume that we will not
1744 * overwrite anything at this point in time.
1746 if (other_inode
> sctx
->send_progress
) {
1747 ret
= get_inode_info(sctx
->parent_root
, other_inode
, NULL
,
1748 who_gen
, NULL
, NULL
, NULL
, NULL
);
1753 *who_ino
= other_inode
;
1763 * Checks if the ref was overwritten by an already processed inode. This is
1764 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1765 * thus the orphan name needs be used.
1766 * process_recorded_refs also uses it to avoid unlinking of refs that were
1769 static int did_overwrite_ref(struct send_ctx
*sctx
,
1770 u64 dir
, u64 dir_gen
,
1771 u64 ino
, u64 ino_gen
,
1772 const char *name
, int name_len
)
1779 if (!sctx
->parent_root
)
1782 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1786 /* check if the ref was overwritten by another ref */
1787 ret
= lookup_dir_item_inode(sctx
->send_root
, dir
, name
, name_len
,
1788 &ow_inode
, &other_type
);
1789 if (ret
< 0 && ret
!= -ENOENT
)
1792 /* was never and will never be overwritten */
1797 ret
= get_inode_info(sctx
->send_root
, ow_inode
, NULL
, &gen
, NULL
, NULL
,
1802 if (ow_inode
== ino
&& gen
== ino_gen
) {
1807 /* we know that it is or will be overwritten. check this now */
1808 if (ow_inode
< sctx
->send_progress
)
1818 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1819 * that got overwritten. This is used by process_recorded_refs to determine
1820 * if it has to use the path as returned by get_cur_path or the orphan name.
1822 static int did_overwrite_first_ref(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1825 struct fs_path
*name
= NULL
;
1829 if (!sctx
->parent_root
)
1832 name
= fs_path_alloc();
1836 ret
= get_first_ref(sctx
->parent_root
, ino
, &dir
, &dir_gen
, name
);
1840 ret
= did_overwrite_ref(sctx
, dir
, dir_gen
, ino
, gen
,
1841 name
->start
, fs_path_len(name
));
1849 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1850 * so we need to do some special handling in case we have clashes. This function
1851 * takes care of this with the help of name_cache_entry::radix_list.
1852 * In case of error, nce is kfreed.
1854 static int name_cache_insert(struct send_ctx
*sctx
,
1855 struct name_cache_entry
*nce
)
1858 struct list_head
*nce_head
;
1860 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
1861 (unsigned long)nce
->ino
);
1863 nce_head
= kmalloc(sizeof(*nce_head
), GFP_NOFS
);
1868 INIT_LIST_HEAD(nce_head
);
1870 ret
= radix_tree_insert(&sctx
->name_cache
, nce
->ino
, nce_head
);
1877 list_add_tail(&nce
->radix_list
, nce_head
);
1878 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
1879 sctx
->name_cache_size
++;
1884 static void name_cache_delete(struct send_ctx
*sctx
,
1885 struct name_cache_entry
*nce
)
1887 struct list_head
*nce_head
;
1889 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
1890 (unsigned long)nce
->ino
);
1893 list_del(&nce
->radix_list
);
1894 list_del(&nce
->list
);
1895 sctx
->name_cache_size
--;
1897 if (list_empty(nce_head
)) {
1898 radix_tree_delete(&sctx
->name_cache
, (unsigned long)nce
->ino
);
1903 static struct name_cache_entry
*name_cache_search(struct send_ctx
*sctx
,
1906 struct list_head
*nce_head
;
1907 struct name_cache_entry
*cur
;
1909 nce_head
= radix_tree_lookup(&sctx
->name_cache
, (unsigned long)ino
);
1913 list_for_each_entry(cur
, nce_head
, radix_list
) {
1914 if (cur
->ino
== ino
&& cur
->gen
== gen
)
1921 * Removes the entry from the list and adds it back to the end. This marks the
1922 * entry as recently used so that name_cache_clean_unused does not remove it.
1924 static void name_cache_used(struct send_ctx
*sctx
, struct name_cache_entry
*nce
)
1926 list_del(&nce
->list
);
1927 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
1931 * Remove some entries from the beginning of name_cache_list.
1933 static void name_cache_clean_unused(struct send_ctx
*sctx
)
1935 struct name_cache_entry
*nce
;
1937 if (sctx
->name_cache_size
< SEND_CTX_NAME_CACHE_CLEAN_SIZE
)
1940 while (sctx
->name_cache_size
> SEND_CTX_MAX_NAME_CACHE_SIZE
) {
1941 nce
= list_entry(sctx
->name_cache_list
.next
,
1942 struct name_cache_entry
, list
);
1943 name_cache_delete(sctx
, nce
);
1948 static void name_cache_free(struct send_ctx
*sctx
)
1950 struct name_cache_entry
*nce
;
1952 while (!list_empty(&sctx
->name_cache_list
)) {
1953 nce
= list_entry(sctx
->name_cache_list
.next
,
1954 struct name_cache_entry
, list
);
1955 name_cache_delete(sctx
, nce
);
1961 * Used by get_cur_path for each ref up to the root.
1962 * Returns 0 if it succeeded.
1963 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1964 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1965 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1966 * Returns <0 in case of error.
1968 static int __get_cur_name_and_parent(struct send_ctx
*sctx
,
1970 int skip_name_cache
,
1973 struct fs_path
*dest
)
1977 struct btrfs_path
*path
= NULL
;
1978 struct name_cache_entry
*nce
= NULL
;
1980 if (skip_name_cache
)
1983 * First check if we already did a call to this function with the same
1984 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1985 * return the cached result.
1987 nce
= name_cache_search(sctx
, ino
, gen
);
1989 if (ino
< sctx
->send_progress
&& nce
->need_later_update
) {
1990 name_cache_delete(sctx
, nce
);
1994 name_cache_used(sctx
, nce
);
1995 *parent_ino
= nce
->parent_ino
;
1996 *parent_gen
= nce
->parent_gen
;
1997 ret
= fs_path_add(dest
, nce
->name
, nce
->name_len
);
2005 path
= alloc_path_for_send();
2010 * If the inode is not existent yet, add the orphan name and return 1.
2011 * This should only happen for the parent dir that we determine in
2014 ret
= is_inode_existent(sctx
, ino
, gen
);
2019 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2028 * Depending on whether the inode was already processed or not, use
2029 * send_root or parent_root for ref lookup.
2031 if (ino
< sctx
->send_progress
&& !skip_name_cache
)
2032 ret
= get_first_ref(sctx
->send_root
, ino
,
2033 parent_ino
, parent_gen
, dest
);
2035 ret
= get_first_ref(sctx
->parent_root
, ino
,
2036 parent_ino
, parent_gen
, dest
);
2041 * Check if the ref was overwritten by an inode's ref that was processed
2042 * earlier. If yes, treat as orphan and return 1.
2044 ret
= did_overwrite_ref(sctx
, *parent_ino
, *parent_gen
, ino
, gen
,
2045 dest
->start
, dest
->end
- dest
->start
);
2049 fs_path_reset(dest
);
2050 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2055 if (skip_name_cache
)
2060 * Store the result of the lookup in the name cache.
2062 nce
= kmalloc(sizeof(*nce
) + fs_path_len(dest
) + 1, GFP_NOFS
);
2070 nce
->parent_ino
= *parent_ino
;
2071 nce
->parent_gen
= *parent_gen
;
2072 nce
->name_len
= fs_path_len(dest
);
2074 strcpy(nce
->name
, dest
->start
);
2076 if (ino
< sctx
->send_progress
)
2077 nce
->need_later_update
= 0;
2079 nce
->need_later_update
= 1;
2081 nce_ret
= name_cache_insert(sctx
, nce
);
2084 name_cache_clean_unused(sctx
);
2087 btrfs_free_path(path
);
2092 * Magic happens here. This function returns the first ref to an inode as it
2093 * would look like while receiving the stream at this point in time.
2094 * We walk the path up to the root. For every inode in between, we check if it
2095 * was already processed/sent. If yes, we continue with the parent as found
2096 * in send_root. If not, we continue with the parent as found in parent_root.
2097 * If we encounter an inode that was deleted at this point in time, we use the
2098 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2099 * that were not created yet and overwritten inodes/refs.
2101 * When do we have have orphan inodes:
2102 * 1. When an inode is freshly created and thus no valid refs are available yet
2103 * 2. When a directory lost all it's refs (deleted) but still has dir items
2104 * inside which were not processed yet (pending for move/delete). If anyone
2105 * tried to get the path to the dir items, it would get a path inside that
2107 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2108 * of an unprocessed inode. If in that case the first ref would be
2109 * overwritten, the overwritten inode gets "orphanized". Later when we
2110 * process this overwritten inode, it is restored at a new place by moving
2113 * sctx->send_progress tells this function at which point in time receiving
2116 static int get_cur_path(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2117 struct fs_path
*dest
)
2120 struct fs_path
*name
= NULL
;
2121 u64 parent_inode
= 0;
2124 u64 start_ino
= ino
;
2125 u64 start_gen
= gen
;
2126 int skip_name_cache
= 0;
2128 name
= fs_path_alloc();
2134 if (is_waiting_for_move(sctx
, ino
))
2135 skip_name_cache
= 1;
2139 fs_path_reset(dest
);
2141 while (!stop
&& ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
2142 fs_path_reset(name
);
2144 ret
= __get_cur_name_and_parent(sctx
, ino
, gen
, skip_name_cache
,
2145 &parent_inode
, &parent_gen
, name
);
2151 if (!skip_name_cache
&&
2152 is_waiting_for_move(sctx
, parent_inode
)) {
2156 skip_name_cache
= 1;
2160 ret
= fs_path_add_path(dest
, name
);
2171 fs_path_unreverse(dest
);
2176 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2178 static int send_subvol_begin(struct send_ctx
*sctx
)
2181 struct btrfs_root
*send_root
= sctx
->send_root
;
2182 struct btrfs_root
*parent_root
= sctx
->parent_root
;
2183 struct btrfs_path
*path
;
2184 struct btrfs_key key
;
2185 struct btrfs_root_ref
*ref
;
2186 struct extent_buffer
*leaf
;
2190 path
= btrfs_alloc_path();
2194 name
= kmalloc(BTRFS_PATH_NAME_MAX
, GFP_NOFS
);
2196 btrfs_free_path(path
);
2200 key
.objectid
= send_root
->objectid
;
2201 key
.type
= BTRFS_ROOT_BACKREF_KEY
;
2204 ret
= btrfs_search_slot_for_read(send_root
->fs_info
->tree_root
,
2213 leaf
= path
->nodes
[0];
2214 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2215 if (key
.type
!= BTRFS_ROOT_BACKREF_KEY
||
2216 key
.objectid
!= send_root
->objectid
) {
2220 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
2221 namelen
= btrfs_root_ref_name_len(leaf
, ref
);
2222 read_extent_buffer(leaf
, name
, (unsigned long)(ref
+ 1), namelen
);
2223 btrfs_release_path(path
);
2226 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SNAPSHOT
);
2230 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SUBVOL
);
2235 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_PATH
, name
, namelen
);
2236 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_UUID
,
2237 sctx
->send_root
->root_item
.uuid
);
2238 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CTRANSID
,
2239 le64_to_cpu(sctx
->send_root
->root_item
.ctransid
));
2241 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2242 sctx
->parent_root
->root_item
.uuid
);
2243 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
2244 le64_to_cpu(sctx
->parent_root
->root_item
.ctransid
));
2247 ret
= send_cmd(sctx
);
2251 btrfs_free_path(path
);
2256 static int send_truncate(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 size
)
2261 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino
, size
);
2263 p
= fs_path_alloc();
2267 ret
= begin_cmd(sctx
, BTRFS_SEND_C_TRUNCATE
);
2271 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2274 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2275 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, size
);
2277 ret
= send_cmd(sctx
);
2285 static int send_chmod(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 mode
)
2290 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino
, mode
);
2292 p
= fs_path_alloc();
2296 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHMOD
);
2300 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2303 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2304 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
& 07777);
2306 ret
= send_cmd(sctx
);
2314 static int send_chown(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 uid
, u64 gid
)
2319 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino
, uid
, gid
);
2321 p
= fs_path_alloc();
2325 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHOWN
);
2329 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2332 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2333 TLV_PUT_U64(sctx
, BTRFS_SEND_A_UID
, uid
);
2334 TLV_PUT_U64(sctx
, BTRFS_SEND_A_GID
, gid
);
2336 ret
= send_cmd(sctx
);
2344 static int send_utimes(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
2347 struct fs_path
*p
= NULL
;
2348 struct btrfs_inode_item
*ii
;
2349 struct btrfs_path
*path
= NULL
;
2350 struct extent_buffer
*eb
;
2351 struct btrfs_key key
;
2354 verbose_printk("btrfs: send_utimes %llu\n", ino
);
2356 p
= fs_path_alloc();
2360 path
= alloc_path_for_send();
2367 key
.type
= BTRFS_INODE_ITEM_KEY
;
2369 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2373 eb
= path
->nodes
[0];
2374 slot
= path
->slots
[0];
2375 ii
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
2377 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UTIMES
);
2381 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2384 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2385 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_ATIME
, eb
,
2386 btrfs_inode_atime(ii
));
2387 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_MTIME
, eb
,
2388 btrfs_inode_mtime(ii
));
2389 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_CTIME
, eb
,
2390 btrfs_inode_ctime(ii
));
2391 /* TODO Add otime support when the otime patches get into upstream */
2393 ret
= send_cmd(sctx
);
2398 btrfs_free_path(path
);
2403 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2404 * a valid path yet because we did not process the refs yet. So, the inode
2405 * is created as orphan.
2407 static int send_create_inode(struct send_ctx
*sctx
, u64 ino
)
2416 verbose_printk("btrfs: send_create_inode %llu\n", ino
);
2418 p
= fs_path_alloc();
2422 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &gen
, &mode
, NULL
,
2427 if (S_ISREG(mode
)) {
2428 cmd
= BTRFS_SEND_C_MKFILE
;
2429 } else if (S_ISDIR(mode
)) {
2430 cmd
= BTRFS_SEND_C_MKDIR
;
2431 } else if (S_ISLNK(mode
)) {
2432 cmd
= BTRFS_SEND_C_SYMLINK
;
2433 } else if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
2434 cmd
= BTRFS_SEND_C_MKNOD
;
2435 } else if (S_ISFIFO(mode
)) {
2436 cmd
= BTRFS_SEND_C_MKFIFO
;
2437 } else if (S_ISSOCK(mode
)) {
2438 cmd
= BTRFS_SEND_C_MKSOCK
;
2440 printk(KERN_WARNING
"btrfs: unexpected inode type %o",
2441 (int)(mode
& S_IFMT
));
2446 ret
= begin_cmd(sctx
, cmd
);
2450 ret
= gen_unique_name(sctx
, ino
, gen
, p
);
2454 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2455 TLV_PUT_U64(sctx
, BTRFS_SEND_A_INO
, ino
);
2457 if (S_ISLNK(mode
)) {
2459 ret
= read_symlink(sctx
->send_root
, ino
, p
);
2462 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, p
);
2463 } else if (S_ISCHR(mode
) || S_ISBLK(mode
) ||
2464 S_ISFIFO(mode
) || S_ISSOCK(mode
)) {
2465 TLV_PUT_U64(sctx
, BTRFS_SEND_A_RDEV
, new_encode_dev(rdev
));
2466 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
);
2469 ret
= send_cmd(sctx
);
2481 * We need some special handling for inodes that get processed before the parent
2482 * directory got created. See process_recorded_refs for details.
2483 * This function does the check if we already created the dir out of order.
2485 static int did_create_dir(struct send_ctx
*sctx
, u64 dir
)
2488 struct btrfs_path
*path
= NULL
;
2489 struct btrfs_key key
;
2490 struct btrfs_key found_key
;
2491 struct btrfs_key di_key
;
2492 struct extent_buffer
*eb
;
2493 struct btrfs_dir_item
*di
;
2496 path
= alloc_path_for_send();
2503 key
.type
= BTRFS_DIR_INDEX_KEY
;
2506 ret
= btrfs_search_slot_for_read(sctx
->send_root
, &key
, path
,
2511 eb
= path
->nodes
[0];
2512 slot
= path
->slots
[0];
2513 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
2515 if (ret
|| found_key
.objectid
!= key
.objectid
||
2516 found_key
.type
!= key
.type
) {
2521 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
2522 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
2524 if (di_key
.type
!= BTRFS_ROOT_ITEM_KEY
&&
2525 di_key
.objectid
< sctx
->send_progress
) {
2530 key
.offset
= found_key
.offset
+ 1;
2531 btrfs_release_path(path
);
2535 btrfs_free_path(path
);
2540 * Only creates the inode if it is:
2541 * 1. Not a directory
2542 * 2. Or a directory which was not created already due to out of order
2543 * directories. See did_create_dir and process_recorded_refs for details.
2545 static int send_create_inode_if_needed(struct send_ctx
*sctx
)
2549 if (S_ISDIR(sctx
->cur_inode_mode
)) {
2550 ret
= did_create_dir(sctx
, sctx
->cur_ino
);
2559 ret
= send_create_inode(sctx
, sctx
->cur_ino
);
2567 struct recorded_ref
{
2568 struct list_head list
;
2571 struct fs_path
*full_path
;
2579 * We need to process new refs before deleted refs, but compare_tree gives us
2580 * everything mixed. So we first record all refs and later process them.
2581 * This function is a helper to record one ref.
2583 static int record_ref(struct list_head
*head
, u64 dir
,
2584 u64 dir_gen
, struct fs_path
*path
)
2586 struct recorded_ref
*ref
;
2588 ref
= kmalloc(sizeof(*ref
), GFP_NOFS
);
2593 ref
->dir_gen
= dir_gen
;
2594 ref
->full_path
= path
;
2596 ref
->name
= (char *)kbasename(ref
->full_path
->start
);
2597 ref
->name_len
= ref
->full_path
->end
- ref
->name
;
2598 ref
->dir_path
= ref
->full_path
->start
;
2599 if (ref
->name
== ref
->full_path
->start
)
2600 ref
->dir_path_len
= 0;
2602 ref
->dir_path_len
= ref
->full_path
->end
-
2603 ref
->full_path
->start
- 1 - ref
->name_len
;
2605 list_add_tail(&ref
->list
, head
);
2609 static int dup_ref(struct recorded_ref
*ref
, struct list_head
*list
)
2611 struct recorded_ref
*new;
2613 new = kmalloc(sizeof(*ref
), GFP_NOFS
);
2617 new->dir
= ref
->dir
;
2618 new->dir_gen
= ref
->dir_gen
;
2619 new->full_path
= NULL
;
2620 INIT_LIST_HEAD(&new->list
);
2621 list_add_tail(&new->list
, list
);
2625 static void __free_recorded_refs(struct list_head
*head
)
2627 struct recorded_ref
*cur
;
2629 while (!list_empty(head
)) {
2630 cur
= list_entry(head
->next
, struct recorded_ref
, list
);
2631 fs_path_free(cur
->full_path
);
2632 list_del(&cur
->list
);
2637 static void free_recorded_refs(struct send_ctx
*sctx
)
2639 __free_recorded_refs(&sctx
->new_refs
);
2640 __free_recorded_refs(&sctx
->deleted_refs
);
2644 * Renames/moves a file/dir to its orphan name. Used when the first
2645 * ref of an unprocessed inode gets overwritten and for all non empty
2648 static int orphanize_inode(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2649 struct fs_path
*path
)
2652 struct fs_path
*orphan
;
2654 orphan
= fs_path_alloc();
2658 ret
= gen_unique_name(sctx
, ino
, gen
, orphan
);
2662 ret
= send_rename(sctx
, path
, orphan
);
2665 fs_path_free(orphan
);
2670 * Returns 1 if a directory can be removed at this point in time.
2671 * We check this by iterating all dir items and checking if the inode behind
2672 * the dir item was already processed.
2674 static int can_rmdir(struct send_ctx
*sctx
, u64 dir
, u64 send_progress
)
2677 struct btrfs_root
*root
= sctx
->parent_root
;
2678 struct btrfs_path
*path
;
2679 struct btrfs_key key
;
2680 struct btrfs_key found_key
;
2681 struct btrfs_key loc
;
2682 struct btrfs_dir_item
*di
;
2685 * Don't try to rmdir the top/root subvolume dir.
2687 if (dir
== BTRFS_FIRST_FREE_OBJECTID
)
2690 path
= alloc_path_for_send();
2695 key
.type
= BTRFS_DIR_INDEX_KEY
;
2699 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
2703 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2706 if (ret
|| found_key
.objectid
!= key
.objectid
||
2707 found_key
.type
!= key
.type
) {
2711 di
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2712 struct btrfs_dir_item
);
2713 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &loc
);
2715 if (loc
.objectid
> send_progress
) {
2720 btrfs_release_path(path
);
2721 key
.offset
= found_key
.offset
+ 1;
2727 btrfs_free_path(path
);
2731 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
)
2733 struct rb_node
*n
= sctx
->waiting_dir_moves
.rb_node
;
2734 struct waiting_dir_move
*entry
;
2737 entry
= rb_entry(n
, struct waiting_dir_move
, node
);
2738 if (ino
< entry
->ino
)
2740 else if (ino
> entry
->ino
)
2748 static int add_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
)
2750 struct rb_node
**p
= &sctx
->waiting_dir_moves
.rb_node
;
2751 struct rb_node
*parent
= NULL
;
2752 struct waiting_dir_move
*entry
, *dm
;
2754 dm
= kmalloc(sizeof(*dm
), GFP_NOFS
);
2761 entry
= rb_entry(parent
, struct waiting_dir_move
, node
);
2762 if (ino
< entry
->ino
) {
2764 } else if (ino
> entry
->ino
) {
2765 p
= &(*p
)->rb_right
;
2772 rb_link_node(&dm
->node
, parent
, p
);
2773 rb_insert_color(&dm
->node
, &sctx
->waiting_dir_moves
);
2777 static int del_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
)
2779 struct rb_node
*n
= sctx
->waiting_dir_moves
.rb_node
;
2780 struct waiting_dir_move
*entry
;
2783 entry
= rb_entry(n
, struct waiting_dir_move
, node
);
2784 if (ino
< entry
->ino
) {
2786 } else if (ino
> entry
->ino
) {
2789 rb_erase(&entry
->node
, &sctx
->waiting_dir_moves
);
2797 static int add_pending_dir_move(struct send_ctx
*sctx
, u64 parent_ino
)
2799 struct rb_node
**p
= &sctx
->pending_dir_moves
.rb_node
;
2800 struct rb_node
*parent
= NULL
;
2801 struct pending_dir_move
*entry
, *pm
;
2802 struct recorded_ref
*cur
;
2806 pm
= kmalloc(sizeof(*pm
), GFP_NOFS
);
2809 pm
->parent_ino
= parent_ino
;
2810 pm
->ino
= sctx
->cur_ino
;
2811 pm
->gen
= sctx
->cur_inode_gen
;
2812 INIT_LIST_HEAD(&pm
->list
);
2813 INIT_LIST_HEAD(&pm
->update_refs
);
2814 RB_CLEAR_NODE(&pm
->node
);
2818 entry
= rb_entry(parent
, struct pending_dir_move
, node
);
2819 if (parent_ino
< entry
->parent_ino
) {
2821 } else if (parent_ino
> entry
->parent_ino
) {
2822 p
= &(*p
)->rb_right
;
2829 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
2830 ret
= dup_ref(cur
, &pm
->update_refs
);
2834 list_for_each_entry(cur
, &sctx
->new_refs
, list
) {
2835 ret
= dup_ref(cur
, &pm
->update_refs
);
2840 ret
= add_waiting_dir_move(sctx
, pm
->ino
);
2845 list_add_tail(&pm
->list
, &entry
->list
);
2847 rb_link_node(&pm
->node
, parent
, p
);
2848 rb_insert_color(&pm
->node
, &sctx
->pending_dir_moves
);
2853 __free_recorded_refs(&pm
->update_refs
);
2859 static struct pending_dir_move
*get_pending_dir_moves(struct send_ctx
*sctx
,
2862 struct rb_node
*n
= sctx
->pending_dir_moves
.rb_node
;
2863 struct pending_dir_move
*entry
;
2866 entry
= rb_entry(n
, struct pending_dir_move
, node
);
2867 if (parent_ino
< entry
->parent_ino
)
2869 else if (parent_ino
> entry
->parent_ino
)
2877 static int apply_dir_move(struct send_ctx
*sctx
, struct pending_dir_move
*pm
)
2879 struct fs_path
*from_path
= NULL
;
2880 struct fs_path
*to_path
= NULL
;
2881 u64 orig_progress
= sctx
->send_progress
;
2882 struct recorded_ref
*cur
;
2885 from_path
= fs_path_alloc();
2889 sctx
->send_progress
= pm
->ino
;
2890 ret
= get_cur_path(sctx
, pm
->ino
, pm
->gen
, from_path
);
2894 to_path
= fs_path_alloc();
2900 sctx
->send_progress
= sctx
->cur_ino
+ 1;
2901 ret
= del_waiting_dir_move(sctx
, pm
->ino
);
2904 ret
= get_cur_path(sctx
, pm
->ino
, pm
->gen
, to_path
);
2908 ret
= send_rename(sctx
, from_path
, to_path
);
2912 ret
= send_utimes(sctx
, pm
->ino
, pm
->gen
);
2917 * After rename/move, need to update the utimes of both new parent(s)
2918 * and old parent(s).
2920 list_for_each_entry(cur
, &pm
->update_refs
, list
) {
2921 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
2927 fs_path_free(from_path
);
2928 fs_path_free(to_path
);
2929 sctx
->send_progress
= orig_progress
;
2934 static void free_pending_move(struct send_ctx
*sctx
, struct pending_dir_move
*m
)
2936 if (!list_empty(&m
->list
))
2938 if (!RB_EMPTY_NODE(&m
->node
))
2939 rb_erase(&m
->node
, &sctx
->pending_dir_moves
);
2940 __free_recorded_refs(&m
->update_refs
);
2944 static void tail_append_pending_moves(struct pending_dir_move
*moves
,
2945 struct list_head
*stack
)
2947 if (list_empty(&moves
->list
)) {
2948 list_add_tail(&moves
->list
, stack
);
2951 list_splice_init(&moves
->list
, &list
);
2952 list_add_tail(&moves
->list
, stack
);
2953 list_splice_tail(&list
, stack
);
2957 static int apply_children_dir_moves(struct send_ctx
*sctx
)
2959 struct pending_dir_move
*pm
;
2960 struct list_head stack
;
2961 u64 parent_ino
= sctx
->cur_ino
;
2964 pm
= get_pending_dir_moves(sctx
, parent_ino
);
2968 INIT_LIST_HEAD(&stack
);
2969 tail_append_pending_moves(pm
, &stack
);
2971 while (!list_empty(&stack
)) {
2972 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
2973 parent_ino
= pm
->ino
;
2974 ret
= apply_dir_move(sctx
, pm
);
2975 free_pending_move(sctx
, pm
);
2978 pm
= get_pending_dir_moves(sctx
, parent_ino
);
2980 tail_append_pending_moves(pm
, &stack
);
2985 while (!list_empty(&stack
)) {
2986 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
2987 free_pending_move(sctx
, pm
);
2992 static int wait_for_parent_move(struct send_ctx
*sctx
,
2993 struct recorded_ref
*parent_ref
)
2996 u64 ino
= parent_ref
->dir
;
2997 u64 parent_ino_before
, parent_ino_after
;
2998 u64 new_gen
, old_gen
;
2999 struct fs_path
*path_before
= NULL
;
3000 struct fs_path
*path_after
= NULL
;
3003 if (parent_ref
->dir
<= sctx
->cur_ino
)
3006 if (is_waiting_for_move(sctx
, ino
))
3009 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
, &old_gen
,
3010 NULL
, NULL
, NULL
, NULL
);
3016 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &new_gen
,
3017 NULL
, NULL
, NULL
, NULL
);
3021 if (new_gen
!= old_gen
)
3024 path_before
= fs_path_alloc();
3028 ret
= get_first_ref(sctx
->parent_root
, ino
, &parent_ino_before
,
3030 if (ret
== -ENOENT
) {
3033 } else if (ret
< 0) {
3037 path_after
= fs_path_alloc();
3043 ret
= get_first_ref(sctx
->send_root
, ino
, &parent_ino_after
,
3045 if (ret
== -ENOENT
) {
3048 } else if (ret
< 0) {
3052 len1
= fs_path_len(path_before
);
3053 len2
= fs_path_len(path_after
);
3054 if ((parent_ino_before
!= parent_ino_after
) && (len1
!= len2
||
3055 memcmp(path_before
->start
, path_after
->start
, len1
))) {
3062 fs_path_free(path_before
);
3063 fs_path_free(path_after
);
3069 * This does all the move/link/unlink/rmdir magic.
3071 static int process_recorded_refs(struct send_ctx
*sctx
, int *pending_move
)
3074 struct recorded_ref
*cur
;
3075 struct recorded_ref
*cur2
;
3076 struct list_head check_dirs
;
3077 struct fs_path
*valid_path
= NULL
;
3080 int did_overwrite
= 0;
3083 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx
->cur_ino
);
3086 * This should never happen as the root dir always has the same ref
3087 * which is always '..'
3089 BUG_ON(sctx
->cur_ino
<= BTRFS_FIRST_FREE_OBJECTID
);
3090 INIT_LIST_HEAD(&check_dirs
);
3092 valid_path
= fs_path_alloc();
3099 * First, check if the first ref of the current inode was overwritten
3100 * before. If yes, we know that the current inode was already orphanized
3101 * and thus use the orphan name. If not, we can use get_cur_path to
3102 * get the path of the first ref as it would like while receiving at
3103 * this point in time.
3104 * New inodes are always orphan at the beginning, so force to use the
3105 * orphan name in this case.
3106 * The first ref is stored in valid_path and will be updated if it
3107 * gets moved around.
3109 if (!sctx
->cur_inode_new
) {
3110 ret
= did_overwrite_first_ref(sctx
, sctx
->cur_ino
,
3111 sctx
->cur_inode_gen
);
3117 if (sctx
->cur_inode_new
|| did_overwrite
) {
3118 ret
= gen_unique_name(sctx
, sctx
->cur_ino
,
3119 sctx
->cur_inode_gen
, valid_path
);
3124 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3130 list_for_each_entry(cur
, &sctx
->new_refs
, list
) {
3132 * We may have refs where the parent directory does not exist
3133 * yet. This happens if the parent directories inum is higher
3134 * the the current inum. To handle this case, we create the
3135 * parent directory out of order. But we need to check if this
3136 * did already happen before due to other refs in the same dir.
3138 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3141 if (ret
== inode_state_will_create
) {
3144 * First check if any of the current inodes refs did
3145 * already create the dir.
3147 list_for_each_entry(cur2
, &sctx
->new_refs
, list
) {
3150 if (cur2
->dir
== cur
->dir
) {
3157 * If that did not happen, check if a previous inode
3158 * did already create the dir.
3161 ret
= did_create_dir(sctx
, cur
->dir
);
3165 ret
= send_create_inode(sctx
, cur
->dir
);
3172 * Check if this new ref would overwrite the first ref of
3173 * another unprocessed inode. If yes, orphanize the
3174 * overwritten inode. If we find an overwritten ref that is
3175 * not the first ref, simply unlink it.
3177 ret
= will_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3178 cur
->name
, cur
->name_len
,
3179 &ow_inode
, &ow_gen
);
3183 ret
= is_first_ref(sctx
->parent_root
,
3184 ow_inode
, cur
->dir
, cur
->name
,
3189 ret
= orphanize_inode(sctx
, ow_inode
, ow_gen
,
3194 ret
= send_unlink(sctx
, cur
->full_path
);
3201 * link/move the ref to the new place. If we have an orphan
3202 * inode, move it and update valid_path. If not, link or move
3203 * it depending on the inode mode.
3206 ret
= send_rename(sctx
, valid_path
, cur
->full_path
);
3210 ret
= fs_path_copy(valid_path
, cur
->full_path
);
3214 if (S_ISDIR(sctx
->cur_inode_mode
)) {
3216 * Dirs can't be linked, so move it. For moved
3217 * dirs, we always have one new and one deleted
3218 * ref. The deleted ref is ignored later.
3220 if (wait_for_parent_move(sctx
, cur
)) {
3221 ret
= add_pending_dir_move(sctx
,
3225 ret
= send_rename(sctx
, valid_path
,
3228 ret
= fs_path_copy(valid_path
,
3234 ret
= send_link(sctx
, cur
->full_path
,
3240 ret
= dup_ref(cur
, &check_dirs
);
3245 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->cur_inode_deleted
) {
3247 * Check if we can already rmdir the directory. If not,
3248 * orphanize it. For every dir item inside that gets deleted
3249 * later, we do this check again and rmdir it then if possible.
3250 * See the use of check_dirs for more details.
3252 ret
= can_rmdir(sctx
, sctx
->cur_ino
, sctx
->cur_ino
);
3256 ret
= send_rmdir(sctx
, valid_path
);
3259 } else if (!is_orphan
) {
3260 ret
= orphanize_inode(sctx
, sctx
->cur_ino
,
3261 sctx
->cur_inode_gen
, valid_path
);
3267 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
3268 ret
= dup_ref(cur
, &check_dirs
);
3272 } else if (S_ISDIR(sctx
->cur_inode_mode
) &&
3273 !list_empty(&sctx
->deleted_refs
)) {
3275 * We have a moved dir. Add the old parent to check_dirs
3277 cur
= list_entry(sctx
->deleted_refs
.next
, struct recorded_ref
,
3279 ret
= dup_ref(cur
, &check_dirs
);
3282 } else if (!S_ISDIR(sctx
->cur_inode_mode
)) {
3284 * We have a non dir inode. Go through all deleted refs and
3285 * unlink them if they were not already overwritten by other
3288 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
3289 ret
= did_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3290 sctx
->cur_ino
, sctx
->cur_inode_gen
,
3291 cur
->name
, cur
->name_len
);
3295 ret
= send_unlink(sctx
, cur
->full_path
);
3299 ret
= dup_ref(cur
, &check_dirs
);
3304 * If the inode is still orphan, unlink the orphan. This may
3305 * happen when a previous inode did overwrite the first ref
3306 * of this inode and no new refs were added for the current
3307 * inode. Unlinking does not mean that the inode is deleted in
3308 * all cases. There may still be links to this inode in other
3312 ret
= send_unlink(sctx
, valid_path
);
3319 * We did collect all parent dirs where cur_inode was once located. We
3320 * now go through all these dirs and check if they are pending for
3321 * deletion and if it's finally possible to perform the rmdir now.
3322 * We also update the inode stats of the parent dirs here.
3324 list_for_each_entry(cur
, &check_dirs
, list
) {
3326 * In case we had refs into dirs that were not processed yet,
3327 * we don't need to do the utime and rmdir logic for these dirs.
3328 * The dir will be processed later.
3330 if (cur
->dir
> sctx
->cur_ino
)
3333 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3337 if (ret
== inode_state_did_create
||
3338 ret
== inode_state_no_change
) {
3339 /* TODO delayed utimes */
3340 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
3343 } else if (ret
== inode_state_did_delete
) {
3344 ret
= can_rmdir(sctx
, cur
->dir
, sctx
->cur_ino
);
3348 ret
= get_cur_path(sctx
, cur
->dir
,
3349 cur
->dir_gen
, valid_path
);
3352 ret
= send_rmdir(sctx
, valid_path
);
3362 __free_recorded_refs(&check_dirs
);
3363 free_recorded_refs(sctx
);
3364 fs_path_free(valid_path
);
3368 static int __record_new_ref(int num
, u64 dir
, int index
,
3369 struct fs_path
*name
,
3373 struct send_ctx
*sctx
= ctx
;
3377 p
= fs_path_alloc();
3381 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &gen
, NULL
, NULL
,
3386 ret
= get_cur_path(sctx
, dir
, gen
, p
);
3389 ret
= fs_path_add_path(p
, name
);
3393 ret
= record_ref(&sctx
->new_refs
, dir
, gen
, p
);
3401 static int __record_deleted_ref(int num
, u64 dir
, int index
,
3402 struct fs_path
*name
,
3406 struct send_ctx
*sctx
= ctx
;
3410 p
= fs_path_alloc();
3414 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &gen
, NULL
, NULL
,
3419 ret
= get_cur_path(sctx
, dir
, gen
, p
);
3422 ret
= fs_path_add_path(p
, name
);
3426 ret
= record_ref(&sctx
->deleted_refs
, dir
, gen
, p
);
3434 static int record_new_ref(struct send_ctx
*sctx
)
3438 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
3439 sctx
->cmp_key
, 0, __record_new_ref
, sctx
);
3448 static int record_deleted_ref(struct send_ctx
*sctx
)
3452 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
3453 sctx
->cmp_key
, 0, __record_deleted_ref
, sctx
);
3462 struct find_ref_ctx
{
3465 struct btrfs_root
*root
;
3466 struct fs_path
*name
;
3470 static int __find_iref(int num
, u64 dir
, int index
,
3471 struct fs_path
*name
,
3474 struct find_ref_ctx
*ctx
= ctx_
;
3478 if (dir
== ctx
->dir
&& fs_path_len(name
) == fs_path_len(ctx
->name
) &&
3479 strncmp(name
->start
, ctx
->name
->start
, fs_path_len(name
)) == 0) {
3481 * To avoid doing extra lookups we'll only do this if everything
3484 ret
= get_inode_info(ctx
->root
, dir
, NULL
, &dir_gen
, NULL
,
3488 if (dir_gen
!= ctx
->dir_gen
)
3490 ctx
->found_idx
= num
;
3496 static int find_iref(struct btrfs_root
*root
,
3497 struct btrfs_path
*path
,
3498 struct btrfs_key
*key
,
3499 u64 dir
, u64 dir_gen
, struct fs_path
*name
)
3502 struct find_ref_ctx ctx
;
3506 ctx
.dir_gen
= dir_gen
;
3510 ret
= iterate_inode_ref(root
, path
, key
, 0, __find_iref
, &ctx
);
3514 if (ctx
.found_idx
== -1)
3517 return ctx
.found_idx
;
3520 static int __record_changed_new_ref(int num
, u64 dir
, int index
,
3521 struct fs_path
*name
,
3526 struct send_ctx
*sctx
= ctx
;
3528 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &dir_gen
, NULL
,
3533 ret
= find_iref(sctx
->parent_root
, sctx
->right_path
,
3534 sctx
->cmp_key
, dir
, dir_gen
, name
);
3536 ret
= __record_new_ref(num
, dir
, index
, name
, sctx
);
3543 static int __record_changed_deleted_ref(int num
, u64 dir
, int index
,
3544 struct fs_path
*name
,
3549 struct send_ctx
*sctx
= ctx
;
3551 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &dir_gen
, NULL
,
3556 ret
= find_iref(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
3557 dir
, dir_gen
, name
);
3559 ret
= __record_deleted_ref(num
, dir
, index
, name
, sctx
);
3566 static int record_changed_ref(struct send_ctx
*sctx
)
3570 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
3571 sctx
->cmp_key
, 0, __record_changed_new_ref
, sctx
);
3574 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
3575 sctx
->cmp_key
, 0, __record_changed_deleted_ref
, sctx
);
3585 * Record and process all refs at once. Needed when an inode changes the
3586 * generation number, which means that it was deleted and recreated.
3588 static int process_all_refs(struct send_ctx
*sctx
,
3589 enum btrfs_compare_tree_result cmd
)
3592 struct btrfs_root
*root
;
3593 struct btrfs_path
*path
;
3594 struct btrfs_key key
;
3595 struct btrfs_key found_key
;
3596 struct extent_buffer
*eb
;
3598 iterate_inode_ref_t cb
;
3599 int pending_move
= 0;
3601 path
= alloc_path_for_send();
3605 if (cmd
== BTRFS_COMPARE_TREE_NEW
) {
3606 root
= sctx
->send_root
;
3607 cb
= __record_new_ref
;
3608 } else if (cmd
== BTRFS_COMPARE_TREE_DELETED
) {
3609 root
= sctx
->parent_root
;
3610 cb
= __record_deleted_ref
;
3615 key
.objectid
= sctx
->cmp_key
->objectid
;
3616 key
.type
= BTRFS_INODE_REF_KEY
;
3619 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
3625 eb
= path
->nodes
[0];
3626 slot
= path
->slots
[0];
3627 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
3629 if (found_key
.objectid
!= key
.objectid
||
3630 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
3631 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
))
3634 ret
= iterate_inode_ref(root
, path
, &found_key
, 0, cb
, sctx
);
3635 btrfs_release_path(path
);
3639 key
.offset
= found_key
.offset
+ 1;
3641 btrfs_release_path(path
);
3643 ret
= process_recorded_refs(sctx
, &pending_move
);
3644 /* Only applicable to an incremental send. */
3645 ASSERT(pending_move
== 0);
3648 btrfs_free_path(path
);
3652 static int send_set_xattr(struct send_ctx
*sctx
,
3653 struct fs_path
*path
,
3654 const char *name
, int name_len
,
3655 const char *data
, int data_len
)
3659 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SET_XATTR
);
3663 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
3664 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
3665 TLV_PUT(sctx
, BTRFS_SEND_A_XATTR_DATA
, data
, data_len
);
3667 ret
= send_cmd(sctx
);
3674 static int send_remove_xattr(struct send_ctx
*sctx
,
3675 struct fs_path
*path
,
3676 const char *name
, int name_len
)
3680 ret
= begin_cmd(sctx
, BTRFS_SEND_C_REMOVE_XATTR
);
3684 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
3685 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
3687 ret
= send_cmd(sctx
);
3694 static int __process_new_xattr(int num
, struct btrfs_key
*di_key
,
3695 const char *name
, int name_len
,
3696 const char *data
, int data_len
,
3700 struct send_ctx
*sctx
= ctx
;
3702 posix_acl_xattr_header dummy_acl
;
3704 p
= fs_path_alloc();
3709 * This hack is needed because empty acl's are stored as zero byte
3710 * data in xattrs. Problem with that is, that receiving these zero byte
3711 * acl's will fail later. To fix this, we send a dummy acl list that
3712 * only contains the version number and no entries.
3714 if (!strncmp(name
, XATTR_NAME_POSIX_ACL_ACCESS
, name_len
) ||
3715 !strncmp(name
, XATTR_NAME_POSIX_ACL_DEFAULT
, name_len
)) {
3716 if (data_len
== 0) {
3717 dummy_acl
.a_version
=
3718 cpu_to_le32(POSIX_ACL_XATTR_VERSION
);
3719 data
= (char *)&dummy_acl
;
3720 data_len
= sizeof(dummy_acl
);
3724 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
3728 ret
= send_set_xattr(sctx
, p
, name
, name_len
, data
, data_len
);
3735 static int __process_deleted_xattr(int num
, struct btrfs_key
*di_key
,
3736 const char *name
, int name_len
,
3737 const char *data
, int data_len
,
3741 struct send_ctx
*sctx
= ctx
;
3744 p
= fs_path_alloc();
3748 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
3752 ret
= send_remove_xattr(sctx
, p
, name
, name_len
);
3759 static int process_new_xattr(struct send_ctx
*sctx
)
3763 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
3764 sctx
->cmp_key
, __process_new_xattr
, sctx
);
3769 static int process_deleted_xattr(struct send_ctx
*sctx
)
3773 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
3774 sctx
->cmp_key
, __process_deleted_xattr
, sctx
);
3779 struct find_xattr_ctx
{
3787 static int __find_xattr(int num
, struct btrfs_key
*di_key
,
3788 const char *name
, int name_len
,
3789 const char *data
, int data_len
,
3790 u8 type
, void *vctx
)
3792 struct find_xattr_ctx
*ctx
= vctx
;
3794 if (name_len
== ctx
->name_len
&&
3795 strncmp(name
, ctx
->name
, name_len
) == 0) {
3796 ctx
->found_idx
= num
;
3797 ctx
->found_data_len
= data_len
;
3798 ctx
->found_data
= kmemdup(data
, data_len
, GFP_NOFS
);
3799 if (!ctx
->found_data
)
3806 static int find_xattr(struct btrfs_root
*root
,
3807 struct btrfs_path
*path
,
3808 struct btrfs_key
*key
,
3809 const char *name
, int name_len
,
3810 char **data
, int *data_len
)
3813 struct find_xattr_ctx ctx
;
3816 ctx
.name_len
= name_len
;
3818 ctx
.found_data
= NULL
;
3819 ctx
.found_data_len
= 0;
3821 ret
= iterate_dir_item(root
, path
, key
, __find_xattr
, &ctx
);
3825 if (ctx
.found_idx
== -1)
3828 *data
= ctx
.found_data
;
3829 *data_len
= ctx
.found_data_len
;
3831 kfree(ctx
.found_data
);
3833 return ctx
.found_idx
;
3837 static int __process_changed_new_xattr(int num
, struct btrfs_key
*di_key
,
3838 const char *name
, int name_len
,
3839 const char *data
, int data_len
,
3843 struct send_ctx
*sctx
= ctx
;
3844 char *found_data
= NULL
;
3845 int found_data_len
= 0;
3847 ret
= find_xattr(sctx
->parent_root
, sctx
->right_path
,
3848 sctx
->cmp_key
, name
, name_len
, &found_data
,
3850 if (ret
== -ENOENT
) {
3851 ret
= __process_new_xattr(num
, di_key
, name
, name_len
, data
,
3852 data_len
, type
, ctx
);
3853 } else if (ret
>= 0) {
3854 if (data_len
!= found_data_len
||
3855 memcmp(data
, found_data
, data_len
)) {
3856 ret
= __process_new_xattr(num
, di_key
, name
, name_len
,
3857 data
, data_len
, type
, ctx
);
3867 static int __process_changed_deleted_xattr(int num
, struct btrfs_key
*di_key
,
3868 const char *name
, int name_len
,
3869 const char *data
, int data_len
,
3873 struct send_ctx
*sctx
= ctx
;
3875 ret
= find_xattr(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
3876 name
, name_len
, NULL
, NULL
);
3878 ret
= __process_deleted_xattr(num
, di_key
, name
, name_len
, data
,
3879 data_len
, type
, ctx
);
3886 static int process_changed_xattr(struct send_ctx
*sctx
)
3890 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
3891 sctx
->cmp_key
, __process_changed_new_xattr
, sctx
);
3894 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
3895 sctx
->cmp_key
, __process_changed_deleted_xattr
, sctx
);
3901 static int process_all_new_xattrs(struct send_ctx
*sctx
)
3904 struct btrfs_root
*root
;
3905 struct btrfs_path
*path
;
3906 struct btrfs_key key
;
3907 struct btrfs_key found_key
;
3908 struct extent_buffer
*eb
;
3911 path
= alloc_path_for_send();
3915 root
= sctx
->send_root
;
3917 key
.objectid
= sctx
->cmp_key
->objectid
;
3918 key
.type
= BTRFS_XATTR_ITEM_KEY
;
3921 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
3929 eb
= path
->nodes
[0];
3930 slot
= path
->slots
[0];
3931 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
3933 if (found_key
.objectid
!= key
.objectid
||
3934 found_key
.type
!= key
.type
) {
3939 ret
= iterate_dir_item(root
, path
, &found_key
,
3940 __process_new_xattr
, sctx
);
3944 btrfs_release_path(path
);
3945 key
.offset
= found_key
.offset
+ 1;
3949 btrfs_free_path(path
);
3953 static ssize_t
fill_read_buf(struct send_ctx
*sctx
, u64 offset
, u32 len
)
3955 struct btrfs_root
*root
= sctx
->send_root
;
3956 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3957 struct inode
*inode
;
3960 struct btrfs_key key
;
3961 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
3963 unsigned pg_offset
= offset
& ~PAGE_CACHE_MASK
;
3966 key
.objectid
= sctx
->cur_ino
;
3967 key
.type
= BTRFS_INODE_ITEM_KEY
;
3970 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
3972 return PTR_ERR(inode
);
3974 if (offset
+ len
> i_size_read(inode
)) {
3975 if (offset
> i_size_read(inode
))
3978 len
= offset
- i_size_read(inode
);
3983 last_index
= (offset
+ len
- 1) >> PAGE_CACHE_SHIFT
;
3984 while (index
<= last_index
) {
3985 unsigned cur_len
= min_t(unsigned, len
,
3986 PAGE_CACHE_SIZE
- pg_offset
);
3987 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
3993 if (!PageUptodate(page
)) {
3994 btrfs_readpage(NULL
, page
);
3996 if (!PageUptodate(page
)) {
3998 page_cache_release(page
);
4005 memcpy(sctx
->read_buf
+ ret
, addr
+ pg_offset
, cur_len
);
4008 page_cache_release(page
);
4020 * Read some bytes from the current inode/file and send a write command to
4023 static int send_write(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4027 ssize_t num_read
= 0;
4029 p
= fs_path_alloc();
4033 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset
, len
);
4035 num_read
= fill_read_buf(sctx
, offset
, len
);
4036 if (num_read
<= 0) {
4042 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4046 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4050 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4051 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4052 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, num_read
);
4054 ret
= send_cmd(sctx
);
4065 * Send a clone command to user space.
4067 static int send_clone(struct send_ctx
*sctx
,
4068 u64 offset
, u32 len
,
4069 struct clone_root
*clone_root
)
4075 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4076 "clone_inode=%llu, clone_offset=%llu\n", offset
, len
,
4077 clone_root
->root
->objectid
, clone_root
->ino
,
4078 clone_root
->offset
);
4080 p
= fs_path_alloc();
4084 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CLONE
);
4088 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4092 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4093 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_LEN
, len
);
4094 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4096 if (clone_root
->root
== sctx
->send_root
) {
4097 ret
= get_inode_info(sctx
->send_root
, clone_root
->ino
, NULL
,
4098 &gen
, NULL
, NULL
, NULL
, NULL
);
4101 ret
= get_cur_path(sctx
, clone_root
->ino
, gen
, p
);
4103 ret
= get_inode_path(clone_root
->root
, clone_root
->ino
, p
);
4108 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4109 clone_root
->root
->root_item
.uuid
);
4110 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
4111 le64_to_cpu(clone_root
->root
->root_item
.ctransid
));
4112 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_CLONE_PATH
, p
);
4113 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_OFFSET
,
4114 clone_root
->offset
);
4116 ret
= send_cmd(sctx
);
4125 * Send an update extent command to user space.
4127 static int send_update_extent(struct send_ctx
*sctx
,
4128 u64 offset
, u32 len
)
4133 p
= fs_path_alloc();
4137 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UPDATE_EXTENT
);
4141 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4145 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4146 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4147 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, len
);
4149 ret
= send_cmd(sctx
);
4157 static int send_hole(struct send_ctx
*sctx
, u64 end
)
4159 struct fs_path
*p
= NULL
;
4160 u64 offset
= sctx
->cur_inode_last_extent
;
4164 p
= fs_path_alloc();
4167 memset(sctx
->read_buf
, 0, BTRFS_SEND_READ_SIZE
);
4168 while (offset
< end
) {
4169 len
= min_t(u64
, end
- offset
, BTRFS_SEND_READ_SIZE
);
4171 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4174 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4177 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4178 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4179 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, len
);
4180 ret
= send_cmd(sctx
);
4190 static int send_write_or_clone(struct send_ctx
*sctx
,
4191 struct btrfs_path
*path
,
4192 struct btrfs_key
*key
,
4193 struct clone_root
*clone_root
)
4196 struct btrfs_file_extent_item
*ei
;
4197 u64 offset
= key
->offset
;
4202 u64 bs
= sctx
->send_root
->fs_info
->sb
->s_blocksize
;
4204 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4205 struct btrfs_file_extent_item
);
4206 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
4207 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4208 len
= btrfs_file_extent_inline_len(path
->nodes
[0],
4209 path
->slots
[0], ei
);
4211 * it is possible the inline item won't cover the whole page,
4212 * but there may be items after this page. Make
4213 * sure to send the whole thing
4215 len
= PAGE_CACHE_ALIGN(len
);
4217 len
= btrfs_file_extent_num_bytes(path
->nodes
[0], ei
);
4220 if (offset
+ len
> sctx
->cur_inode_size
)
4221 len
= sctx
->cur_inode_size
- offset
;
4227 if (clone_root
&& IS_ALIGNED(offset
+ len
, bs
)) {
4228 ret
= send_clone(sctx
, offset
, len
, clone_root
);
4229 } else if (sctx
->flags
& BTRFS_SEND_FLAG_NO_FILE_DATA
) {
4230 ret
= send_update_extent(sctx
, offset
, len
);
4234 if (l
> BTRFS_SEND_READ_SIZE
)
4235 l
= BTRFS_SEND_READ_SIZE
;
4236 ret
= send_write(sctx
, pos
+ offset
, l
);
4249 static int is_extent_unchanged(struct send_ctx
*sctx
,
4250 struct btrfs_path
*left_path
,
4251 struct btrfs_key
*ekey
)
4254 struct btrfs_key key
;
4255 struct btrfs_path
*path
= NULL
;
4256 struct extent_buffer
*eb
;
4258 struct btrfs_key found_key
;
4259 struct btrfs_file_extent_item
*ei
;
4264 u64 left_offset_fixed
;
4272 path
= alloc_path_for_send();
4276 eb
= left_path
->nodes
[0];
4277 slot
= left_path
->slots
[0];
4278 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
4279 left_type
= btrfs_file_extent_type(eb
, ei
);
4281 if (left_type
!= BTRFS_FILE_EXTENT_REG
) {
4285 left_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
4286 left_len
= btrfs_file_extent_num_bytes(eb
, ei
);
4287 left_offset
= btrfs_file_extent_offset(eb
, ei
);
4288 left_gen
= btrfs_file_extent_generation(eb
, ei
);
4291 * Following comments will refer to these graphics. L is the left
4292 * extents which we are checking at the moment. 1-8 are the right
4293 * extents that we iterate.
4296 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4299 * |--1--|-2b-|...(same as above)
4301 * Alternative situation. Happens on files where extents got split.
4303 * |-----------7-----------|-6-|
4305 * Alternative situation. Happens on files which got larger.
4308 * Nothing follows after 8.
4311 key
.objectid
= ekey
->objectid
;
4312 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4313 key
.offset
= ekey
->offset
;
4314 ret
= btrfs_search_slot_for_read(sctx
->parent_root
, &key
, path
, 0, 0);
4323 * Handle special case where the right side has no extents at all.
4325 eb
= path
->nodes
[0];
4326 slot
= path
->slots
[0];
4327 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4328 if (found_key
.objectid
!= key
.objectid
||
4329 found_key
.type
!= key
.type
) {
4330 /* If we're a hole then just pretend nothing changed */
4331 ret
= (left_disknr
) ? 0 : 1;
4336 * We're now on 2a, 2b or 7.
4339 while (key
.offset
< ekey
->offset
+ left_len
) {
4340 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
4341 right_type
= btrfs_file_extent_type(eb
, ei
);
4342 if (right_type
!= BTRFS_FILE_EXTENT_REG
) {
4347 right_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
4348 right_len
= btrfs_file_extent_num_bytes(eb
, ei
);
4349 right_offset
= btrfs_file_extent_offset(eb
, ei
);
4350 right_gen
= btrfs_file_extent_generation(eb
, ei
);
4353 * Are we at extent 8? If yes, we know the extent is changed.
4354 * This may only happen on the first iteration.
4356 if (found_key
.offset
+ right_len
<= ekey
->offset
) {
4357 /* If we're a hole just pretend nothing changed */
4358 ret
= (left_disknr
) ? 0 : 1;
4362 left_offset_fixed
= left_offset
;
4363 if (key
.offset
< ekey
->offset
) {
4364 /* Fix the right offset for 2a and 7. */
4365 right_offset
+= ekey
->offset
- key
.offset
;
4367 /* Fix the left offset for all behind 2a and 2b */
4368 left_offset_fixed
+= key
.offset
- ekey
->offset
;
4372 * Check if we have the same extent.
4374 if (left_disknr
!= right_disknr
||
4375 left_offset_fixed
!= right_offset
||
4376 left_gen
!= right_gen
) {
4382 * Go to the next extent.
4384 ret
= btrfs_next_item(sctx
->parent_root
, path
);
4388 eb
= path
->nodes
[0];
4389 slot
= path
->slots
[0];
4390 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4392 if (ret
|| found_key
.objectid
!= key
.objectid
||
4393 found_key
.type
!= key
.type
) {
4394 key
.offset
+= right_len
;
4397 if (found_key
.offset
!= key
.offset
+ right_len
) {
4405 * We're now behind the left extent (treat as unchanged) or at the end
4406 * of the right side (treat as changed).
4408 if (key
.offset
>= ekey
->offset
+ left_len
)
4415 btrfs_free_path(path
);
4419 static int get_last_extent(struct send_ctx
*sctx
, u64 offset
)
4421 struct btrfs_path
*path
;
4422 struct btrfs_root
*root
= sctx
->send_root
;
4423 struct btrfs_file_extent_item
*fi
;
4424 struct btrfs_key key
;
4429 path
= alloc_path_for_send();
4433 sctx
->cur_inode_last_extent
= 0;
4435 key
.objectid
= sctx
->cur_ino
;
4436 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4437 key
.offset
= offset
;
4438 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 0, 1);
4442 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
4443 if (key
.objectid
!= sctx
->cur_ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4446 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4447 struct btrfs_file_extent_item
);
4448 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
4449 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4450 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
4451 path
->slots
[0], fi
);
4452 extent_end
= ALIGN(key
.offset
+ size
,
4453 sctx
->send_root
->sectorsize
);
4455 extent_end
= key
.offset
+
4456 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
4458 sctx
->cur_inode_last_extent
= extent_end
;
4460 btrfs_free_path(path
);
4464 static int maybe_send_hole(struct send_ctx
*sctx
, struct btrfs_path
*path
,
4465 struct btrfs_key
*key
)
4467 struct btrfs_file_extent_item
*fi
;
4472 if (sctx
->cur_ino
!= key
->objectid
|| !need_send_hole(sctx
))
4475 if (sctx
->cur_inode_last_extent
== (u64
)-1) {
4476 ret
= get_last_extent(sctx
, key
->offset
- 1);
4481 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4482 struct btrfs_file_extent_item
);
4483 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
4484 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4485 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
4486 path
->slots
[0], fi
);
4487 extent_end
= ALIGN(key
->offset
+ size
,
4488 sctx
->send_root
->sectorsize
);
4490 extent_end
= key
->offset
+
4491 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
4494 if (path
->slots
[0] == 0 &&
4495 sctx
->cur_inode_last_extent
< key
->offset
) {
4497 * We might have skipped entire leafs that contained only
4498 * file extent items for our current inode. These leafs have
4499 * a generation number smaller (older) than the one in the
4500 * current leaf and the leaf our last extent came from, and
4501 * are located between these 2 leafs.
4503 ret
= get_last_extent(sctx
, key
->offset
- 1);
4508 if (sctx
->cur_inode_last_extent
< key
->offset
)
4509 ret
= send_hole(sctx
, key
->offset
);
4510 sctx
->cur_inode_last_extent
= extent_end
;
4514 static int process_extent(struct send_ctx
*sctx
,
4515 struct btrfs_path
*path
,
4516 struct btrfs_key
*key
)
4518 struct clone_root
*found_clone
= NULL
;
4521 if (S_ISLNK(sctx
->cur_inode_mode
))
4524 if (sctx
->parent_root
&& !sctx
->cur_inode_new
) {
4525 ret
= is_extent_unchanged(sctx
, path
, key
);
4533 struct btrfs_file_extent_item
*ei
;
4536 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4537 struct btrfs_file_extent_item
);
4538 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
4539 if (type
== BTRFS_FILE_EXTENT_PREALLOC
||
4540 type
== BTRFS_FILE_EXTENT_REG
) {
4542 * The send spec does not have a prealloc command yet,
4543 * so just leave a hole for prealloc'ed extents until
4544 * we have enough commands queued up to justify rev'ing
4547 if (type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4552 /* Have a hole, just skip it. */
4553 if (btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
) == 0) {
4560 ret
= find_extent_clone(sctx
, path
, key
->objectid
, key
->offset
,
4561 sctx
->cur_inode_size
, &found_clone
);
4562 if (ret
!= -ENOENT
&& ret
< 0)
4565 ret
= send_write_or_clone(sctx
, path
, key
, found_clone
);
4569 ret
= maybe_send_hole(sctx
, path
, key
);
4574 static int process_all_extents(struct send_ctx
*sctx
)
4577 struct btrfs_root
*root
;
4578 struct btrfs_path
*path
;
4579 struct btrfs_key key
;
4580 struct btrfs_key found_key
;
4581 struct extent_buffer
*eb
;
4584 root
= sctx
->send_root
;
4585 path
= alloc_path_for_send();
4589 key
.objectid
= sctx
->cmp_key
->objectid
;
4590 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4592 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4597 eb
= path
->nodes
[0];
4598 slot
= path
->slots
[0];
4600 if (slot
>= btrfs_header_nritems(eb
)) {
4601 ret
= btrfs_next_leaf(root
, path
);
4604 } else if (ret
> 0) {
4611 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4613 if (found_key
.objectid
!= key
.objectid
||
4614 found_key
.type
!= key
.type
) {
4619 ret
= process_extent(sctx
, path
, &found_key
);
4627 btrfs_free_path(path
);
4631 static int process_recorded_refs_if_needed(struct send_ctx
*sctx
, int at_end
,
4633 int *refs_processed
)
4637 if (sctx
->cur_ino
== 0)
4639 if (!at_end
&& sctx
->cur_ino
== sctx
->cmp_key
->objectid
&&
4640 sctx
->cmp_key
->type
<= BTRFS_INODE_EXTREF_KEY
)
4642 if (list_empty(&sctx
->new_refs
) && list_empty(&sctx
->deleted_refs
))
4645 ret
= process_recorded_refs(sctx
, pending_move
);
4649 *refs_processed
= 1;
4654 static int finish_inode_if_needed(struct send_ctx
*sctx
, int at_end
)
4665 int pending_move
= 0;
4666 int refs_processed
= 0;
4668 ret
= process_recorded_refs_if_needed(sctx
, at_end
, &pending_move
,
4674 * We have processed the refs and thus need to advance send_progress.
4675 * Now, calls to get_cur_xxx will take the updated refs of the current
4676 * inode into account.
4678 * On the other hand, if our current inode is a directory and couldn't
4679 * be moved/renamed because its parent was renamed/moved too and it has
4680 * a higher inode number, we can only move/rename our current inode
4681 * after we moved/renamed its parent. Therefore in this case operate on
4682 * the old path (pre move/rename) of our current inode, and the
4683 * move/rename will be performed later.
4685 if (refs_processed
&& !pending_move
)
4686 sctx
->send_progress
= sctx
->cur_ino
+ 1;
4688 if (sctx
->cur_ino
== 0 || sctx
->cur_inode_deleted
)
4690 if (!at_end
&& sctx
->cmp_key
->objectid
== sctx
->cur_ino
)
4693 ret
= get_inode_info(sctx
->send_root
, sctx
->cur_ino
, NULL
, NULL
,
4694 &left_mode
, &left_uid
, &left_gid
, NULL
);
4698 if (!sctx
->parent_root
|| sctx
->cur_inode_new
) {
4700 if (!S_ISLNK(sctx
->cur_inode_mode
))
4703 ret
= get_inode_info(sctx
->parent_root
, sctx
->cur_ino
,
4704 NULL
, NULL
, &right_mode
, &right_uid
,
4709 if (left_uid
!= right_uid
|| left_gid
!= right_gid
)
4711 if (!S_ISLNK(sctx
->cur_inode_mode
) && left_mode
!= right_mode
)
4715 if (S_ISREG(sctx
->cur_inode_mode
)) {
4716 if (need_send_hole(sctx
)) {
4717 if (sctx
->cur_inode_last_extent
== (u64
)-1) {
4718 ret
= get_last_extent(sctx
, (u64
)-1);
4722 if (sctx
->cur_inode_last_extent
<
4723 sctx
->cur_inode_size
) {
4724 ret
= send_hole(sctx
, sctx
->cur_inode_size
);
4729 ret
= send_truncate(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
4730 sctx
->cur_inode_size
);
4736 ret
= send_chown(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
4737 left_uid
, left_gid
);
4742 ret
= send_chmod(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
4749 * If other directory inodes depended on our current directory
4750 * inode's move/rename, now do their move/rename operations.
4752 if (!is_waiting_for_move(sctx
, sctx
->cur_ino
)) {
4753 ret
= apply_children_dir_moves(sctx
);
4759 * Need to send that every time, no matter if it actually
4760 * changed between the two trees as we have done changes to
4763 sctx
->send_progress
= sctx
->cur_ino
+ 1;
4764 ret
= send_utimes(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
);
4772 static int changed_inode(struct send_ctx
*sctx
,
4773 enum btrfs_compare_tree_result result
)
4776 struct btrfs_key
*key
= sctx
->cmp_key
;
4777 struct btrfs_inode_item
*left_ii
= NULL
;
4778 struct btrfs_inode_item
*right_ii
= NULL
;
4782 sctx
->cur_ino
= key
->objectid
;
4783 sctx
->cur_inode_new_gen
= 0;
4784 sctx
->cur_inode_last_extent
= (u64
)-1;
4787 * Set send_progress to current inode. This will tell all get_cur_xxx
4788 * functions that the current inode's refs are not updated yet. Later,
4789 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4791 sctx
->send_progress
= sctx
->cur_ino
;
4793 if (result
== BTRFS_COMPARE_TREE_NEW
||
4794 result
== BTRFS_COMPARE_TREE_CHANGED
) {
4795 left_ii
= btrfs_item_ptr(sctx
->left_path
->nodes
[0],
4796 sctx
->left_path
->slots
[0],
4797 struct btrfs_inode_item
);
4798 left_gen
= btrfs_inode_generation(sctx
->left_path
->nodes
[0],
4801 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
4802 sctx
->right_path
->slots
[0],
4803 struct btrfs_inode_item
);
4804 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
4807 if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
4808 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
4809 sctx
->right_path
->slots
[0],
4810 struct btrfs_inode_item
);
4812 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
4816 * The cur_ino = root dir case is special here. We can't treat
4817 * the inode as deleted+reused because it would generate a
4818 * stream that tries to delete/mkdir the root dir.
4820 if (left_gen
!= right_gen
&&
4821 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
4822 sctx
->cur_inode_new_gen
= 1;
4825 if (result
== BTRFS_COMPARE_TREE_NEW
) {
4826 sctx
->cur_inode_gen
= left_gen
;
4827 sctx
->cur_inode_new
= 1;
4828 sctx
->cur_inode_deleted
= 0;
4829 sctx
->cur_inode_size
= btrfs_inode_size(
4830 sctx
->left_path
->nodes
[0], left_ii
);
4831 sctx
->cur_inode_mode
= btrfs_inode_mode(
4832 sctx
->left_path
->nodes
[0], left_ii
);
4833 if (sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
4834 ret
= send_create_inode_if_needed(sctx
);
4835 } else if (result
== BTRFS_COMPARE_TREE_DELETED
) {
4836 sctx
->cur_inode_gen
= right_gen
;
4837 sctx
->cur_inode_new
= 0;
4838 sctx
->cur_inode_deleted
= 1;
4839 sctx
->cur_inode_size
= btrfs_inode_size(
4840 sctx
->right_path
->nodes
[0], right_ii
);
4841 sctx
->cur_inode_mode
= btrfs_inode_mode(
4842 sctx
->right_path
->nodes
[0], right_ii
);
4843 } else if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
4845 * We need to do some special handling in case the inode was
4846 * reported as changed with a changed generation number. This
4847 * means that the original inode was deleted and new inode
4848 * reused the same inum. So we have to treat the old inode as
4849 * deleted and the new one as new.
4851 if (sctx
->cur_inode_new_gen
) {
4853 * First, process the inode as if it was deleted.
4855 sctx
->cur_inode_gen
= right_gen
;
4856 sctx
->cur_inode_new
= 0;
4857 sctx
->cur_inode_deleted
= 1;
4858 sctx
->cur_inode_size
= btrfs_inode_size(
4859 sctx
->right_path
->nodes
[0], right_ii
);
4860 sctx
->cur_inode_mode
= btrfs_inode_mode(
4861 sctx
->right_path
->nodes
[0], right_ii
);
4862 ret
= process_all_refs(sctx
,
4863 BTRFS_COMPARE_TREE_DELETED
);
4868 * Now process the inode as if it was new.
4870 sctx
->cur_inode_gen
= left_gen
;
4871 sctx
->cur_inode_new
= 1;
4872 sctx
->cur_inode_deleted
= 0;
4873 sctx
->cur_inode_size
= btrfs_inode_size(
4874 sctx
->left_path
->nodes
[0], left_ii
);
4875 sctx
->cur_inode_mode
= btrfs_inode_mode(
4876 sctx
->left_path
->nodes
[0], left_ii
);
4877 ret
= send_create_inode_if_needed(sctx
);
4881 ret
= process_all_refs(sctx
, BTRFS_COMPARE_TREE_NEW
);
4885 * Advance send_progress now as we did not get into
4886 * process_recorded_refs_if_needed in the new_gen case.
4888 sctx
->send_progress
= sctx
->cur_ino
+ 1;
4891 * Now process all extents and xattrs of the inode as if
4892 * they were all new.
4894 ret
= process_all_extents(sctx
);
4897 ret
= process_all_new_xattrs(sctx
);
4901 sctx
->cur_inode_gen
= left_gen
;
4902 sctx
->cur_inode_new
= 0;
4903 sctx
->cur_inode_new_gen
= 0;
4904 sctx
->cur_inode_deleted
= 0;
4905 sctx
->cur_inode_size
= btrfs_inode_size(
4906 sctx
->left_path
->nodes
[0], left_ii
);
4907 sctx
->cur_inode_mode
= btrfs_inode_mode(
4908 sctx
->left_path
->nodes
[0], left_ii
);
4917 * We have to process new refs before deleted refs, but compare_trees gives us
4918 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4919 * first and later process them in process_recorded_refs.
4920 * For the cur_inode_new_gen case, we skip recording completely because
4921 * changed_inode did already initiate processing of refs. The reason for this is
4922 * that in this case, compare_tree actually compares the refs of 2 different
4923 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4924 * refs of the right tree as deleted and all refs of the left tree as new.
4926 static int changed_ref(struct send_ctx
*sctx
,
4927 enum btrfs_compare_tree_result result
)
4931 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
4933 if (!sctx
->cur_inode_new_gen
&&
4934 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
4935 if (result
== BTRFS_COMPARE_TREE_NEW
)
4936 ret
= record_new_ref(sctx
);
4937 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
4938 ret
= record_deleted_ref(sctx
);
4939 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
4940 ret
= record_changed_ref(sctx
);
4947 * Process new/deleted/changed xattrs. We skip processing in the
4948 * cur_inode_new_gen case because changed_inode did already initiate processing
4949 * of xattrs. The reason is the same as in changed_ref
4951 static int changed_xattr(struct send_ctx
*sctx
,
4952 enum btrfs_compare_tree_result result
)
4956 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
4958 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
4959 if (result
== BTRFS_COMPARE_TREE_NEW
)
4960 ret
= process_new_xattr(sctx
);
4961 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
4962 ret
= process_deleted_xattr(sctx
);
4963 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
4964 ret
= process_changed_xattr(sctx
);
4971 * Process new/deleted/changed extents. We skip processing in the
4972 * cur_inode_new_gen case because changed_inode did already initiate processing
4973 * of extents. The reason is the same as in changed_ref
4975 static int changed_extent(struct send_ctx
*sctx
,
4976 enum btrfs_compare_tree_result result
)
4980 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
4982 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
4983 if (result
!= BTRFS_COMPARE_TREE_DELETED
)
4984 ret
= process_extent(sctx
, sctx
->left_path
,
4991 static int dir_changed(struct send_ctx
*sctx
, u64 dir
)
4993 u64 orig_gen
, new_gen
;
4996 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &new_gen
, NULL
, NULL
,
5001 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &orig_gen
, NULL
,
5006 return (orig_gen
!= new_gen
) ? 1 : 0;
5009 static int compare_refs(struct send_ctx
*sctx
, struct btrfs_path
*path
,
5010 struct btrfs_key
*key
)
5012 struct btrfs_inode_extref
*extref
;
5013 struct extent_buffer
*leaf
;
5014 u64 dirid
= 0, last_dirid
= 0;
5021 /* Easy case, just check this one dirid */
5022 if (key
->type
== BTRFS_INODE_REF_KEY
) {
5023 dirid
= key
->offset
;
5025 ret
= dir_changed(sctx
, dirid
);
5029 leaf
= path
->nodes
[0];
5030 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
5031 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
5032 while (cur_offset
< item_size
) {
5033 extref
= (struct btrfs_inode_extref
*)(ptr
+
5035 dirid
= btrfs_inode_extref_parent(leaf
, extref
);
5036 ref_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
5037 cur_offset
+= ref_name_len
+ sizeof(*extref
);
5038 if (dirid
== last_dirid
)
5040 ret
= dir_changed(sctx
, dirid
);
5050 * Updates compare related fields in sctx and simply forwards to the actual
5051 * changed_xxx functions.
5053 static int changed_cb(struct btrfs_root
*left_root
,
5054 struct btrfs_root
*right_root
,
5055 struct btrfs_path
*left_path
,
5056 struct btrfs_path
*right_path
,
5057 struct btrfs_key
*key
,
5058 enum btrfs_compare_tree_result result
,
5062 struct send_ctx
*sctx
= ctx
;
5064 if (result
== BTRFS_COMPARE_TREE_SAME
) {
5065 if (key
->type
== BTRFS_INODE_REF_KEY
||
5066 key
->type
== BTRFS_INODE_EXTREF_KEY
) {
5067 ret
= compare_refs(sctx
, left_path
, key
);
5072 } else if (key
->type
== BTRFS_EXTENT_DATA_KEY
) {
5073 return maybe_send_hole(sctx
, left_path
, key
);
5077 result
= BTRFS_COMPARE_TREE_CHANGED
;
5081 sctx
->left_path
= left_path
;
5082 sctx
->right_path
= right_path
;
5083 sctx
->cmp_key
= key
;
5085 ret
= finish_inode_if_needed(sctx
, 0);
5089 /* Ignore non-FS objects */
5090 if (key
->objectid
== BTRFS_FREE_INO_OBJECTID
||
5091 key
->objectid
== BTRFS_FREE_SPACE_OBJECTID
)
5094 if (key
->type
== BTRFS_INODE_ITEM_KEY
)
5095 ret
= changed_inode(sctx
, result
);
5096 else if (key
->type
== BTRFS_INODE_REF_KEY
||
5097 key
->type
== BTRFS_INODE_EXTREF_KEY
)
5098 ret
= changed_ref(sctx
, result
);
5099 else if (key
->type
== BTRFS_XATTR_ITEM_KEY
)
5100 ret
= changed_xattr(sctx
, result
);
5101 else if (key
->type
== BTRFS_EXTENT_DATA_KEY
)
5102 ret
= changed_extent(sctx
, result
);
5108 static int full_send_tree(struct send_ctx
*sctx
)
5111 struct btrfs_root
*send_root
= sctx
->send_root
;
5112 struct btrfs_key key
;
5113 struct btrfs_key found_key
;
5114 struct btrfs_path
*path
;
5115 struct extent_buffer
*eb
;
5120 path
= alloc_path_for_send();
5124 spin_lock(&send_root
->root_item_lock
);
5125 start_ctransid
= btrfs_root_ctransid(&send_root
->root_item
);
5126 spin_unlock(&send_root
->root_item_lock
);
5128 key
.objectid
= BTRFS_FIRST_FREE_OBJECTID
;
5129 key
.type
= BTRFS_INODE_ITEM_KEY
;
5133 * Make sure the tree has not changed after re-joining. We detect this
5134 * by comparing start_ctransid and ctransid. They should always match.
5136 spin_lock(&send_root
->root_item_lock
);
5137 ctransid
= btrfs_root_ctransid(&send_root
->root_item
);
5138 spin_unlock(&send_root
->root_item_lock
);
5140 if (ctransid
!= start_ctransid
) {
5141 WARN(1, KERN_WARNING
"BTRFS: the root that you're trying to "
5142 "send was modified in between. This is "
5143 "probably a bug.\n");
5148 ret
= btrfs_search_slot_for_read(send_root
, &key
, path
, 1, 0);
5155 eb
= path
->nodes
[0];
5156 slot
= path
->slots
[0];
5157 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5159 ret
= changed_cb(send_root
, NULL
, path
, NULL
,
5160 &found_key
, BTRFS_COMPARE_TREE_NEW
, sctx
);
5164 key
.objectid
= found_key
.objectid
;
5165 key
.type
= found_key
.type
;
5166 key
.offset
= found_key
.offset
+ 1;
5168 ret
= btrfs_next_item(send_root
, path
);
5178 ret
= finish_inode_if_needed(sctx
, 1);
5181 btrfs_free_path(path
);
5185 static int send_subvol(struct send_ctx
*sctx
)
5189 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_STREAM_HEADER
)) {
5190 ret
= send_header(sctx
);
5195 ret
= send_subvol_begin(sctx
);
5199 if (sctx
->parent_root
) {
5200 ret
= btrfs_compare_trees(sctx
->send_root
, sctx
->parent_root
,
5204 ret
= finish_inode_if_needed(sctx
, 1);
5208 ret
= full_send_tree(sctx
);
5214 free_recorded_refs(sctx
);
5218 static void btrfs_root_dec_send_in_progress(struct btrfs_root
* root
)
5220 spin_lock(&root
->root_item_lock
);
5221 root
->send_in_progress
--;
5223 * Not much left to do, we don't know why it's unbalanced and
5224 * can't blindly reset it to 0.
5226 if (root
->send_in_progress
< 0)
5227 btrfs_err(root
->fs_info
,
5228 "send_in_progres unbalanced %d root %llu\n",
5229 root
->send_in_progress
, root
->root_key
.objectid
);
5230 spin_unlock(&root
->root_item_lock
);
5233 long btrfs_ioctl_send(struct file
*mnt_file
, void __user
*arg_
)
5236 struct btrfs_root
*send_root
;
5237 struct btrfs_root
*clone_root
;
5238 struct btrfs_fs_info
*fs_info
;
5239 struct btrfs_ioctl_send_args
*arg
= NULL
;
5240 struct btrfs_key key
;
5241 struct send_ctx
*sctx
= NULL
;
5243 u64
*clone_sources_tmp
= NULL
;
5244 int clone_sources_to_rollback
= 0;
5245 int sort_clone_roots
= 0;
5248 if (!capable(CAP_SYS_ADMIN
))
5251 send_root
= BTRFS_I(file_inode(mnt_file
))->root
;
5252 fs_info
= send_root
->fs_info
;
5255 * The subvolume must remain read-only during send, protect against
5258 spin_lock(&send_root
->root_item_lock
);
5259 send_root
->send_in_progress
++;
5260 spin_unlock(&send_root
->root_item_lock
);
5263 * This is done when we lookup the root, it should already be complete
5264 * by the time we get here.
5266 WARN_ON(send_root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
);
5269 * Userspace tools do the checks and warn the user if it's
5272 if (!btrfs_root_readonly(send_root
)) {
5277 arg
= memdup_user(arg_
, sizeof(*arg
));
5284 if (!access_ok(VERIFY_READ
, arg
->clone_sources
,
5285 sizeof(*arg
->clone_sources
) *
5286 arg
->clone_sources_count
)) {
5291 if (arg
->flags
& ~BTRFS_SEND_FLAG_MASK
) {
5296 sctx
= kzalloc(sizeof(struct send_ctx
), GFP_NOFS
);
5302 INIT_LIST_HEAD(&sctx
->new_refs
);
5303 INIT_LIST_HEAD(&sctx
->deleted_refs
);
5304 INIT_RADIX_TREE(&sctx
->name_cache
, GFP_NOFS
);
5305 INIT_LIST_HEAD(&sctx
->name_cache_list
);
5307 sctx
->flags
= arg
->flags
;
5309 sctx
->send_filp
= fget(arg
->send_fd
);
5310 if (!sctx
->send_filp
) {
5315 sctx
->send_root
= send_root
;
5316 sctx
->clone_roots_cnt
= arg
->clone_sources_count
;
5318 sctx
->send_max_size
= BTRFS_SEND_BUF_SIZE
;
5319 sctx
->send_buf
= vmalloc(sctx
->send_max_size
);
5320 if (!sctx
->send_buf
) {
5325 sctx
->read_buf
= vmalloc(BTRFS_SEND_READ_SIZE
);
5326 if (!sctx
->read_buf
) {
5331 sctx
->pending_dir_moves
= RB_ROOT
;
5332 sctx
->waiting_dir_moves
= RB_ROOT
;
5334 sctx
->clone_roots
= vzalloc(sizeof(struct clone_root
) *
5335 (arg
->clone_sources_count
+ 1));
5336 if (!sctx
->clone_roots
) {
5341 if (arg
->clone_sources_count
) {
5342 clone_sources_tmp
= vmalloc(arg
->clone_sources_count
*
5343 sizeof(*arg
->clone_sources
));
5344 if (!clone_sources_tmp
) {
5349 ret
= copy_from_user(clone_sources_tmp
, arg
->clone_sources
,
5350 arg
->clone_sources_count
*
5351 sizeof(*arg
->clone_sources
));
5357 for (i
= 0; i
< arg
->clone_sources_count
; i
++) {
5358 key
.objectid
= clone_sources_tmp
[i
];
5359 key
.type
= BTRFS_ROOT_ITEM_KEY
;
5360 key
.offset
= (u64
)-1;
5362 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5364 clone_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
5365 if (IS_ERR(clone_root
)) {
5366 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5367 ret
= PTR_ERR(clone_root
);
5370 clone_sources_to_rollback
= i
+ 1;
5371 spin_lock(&clone_root
->root_item_lock
);
5372 clone_root
->send_in_progress
++;
5373 if (!btrfs_root_readonly(clone_root
)) {
5374 spin_unlock(&clone_root
->root_item_lock
);
5375 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5379 spin_unlock(&clone_root
->root_item_lock
);
5380 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5382 sctx
->clone_roots
[i
].root
= clone_root
;
5384 vfree(clone_sources_tmp
);
5385 clone_sources_tmp
= NULL
;
5388 if (arg
->parent_root
) {
5389 key
.objectid
= arg
->parent_root
;
5390 key
.type
= BTRFS_ROOT_ITEM_KEY
;
5391 key
.offset
= (u64
)-1;
5393 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5395 sctx
->parent_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
5396 if (IS_ERR(sctx
->parent_root
)) {
5397 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5398 ret
= PTR_ERR(sctx
->parent_root
);
5402 spin_lock(&sctx
->parent_root
->root_item_lock
);
5403 sctx
->parent_root
->send_in_progress
++;
5404 if (!btrfs_root_readonly(sctx
->parent_root
)) {
5405 spin_unlock(&sctx
->parent_root
->root_item_lock
);
5406 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5410 spin_unlock(&sctx
->parent_root
->root_item_lock
);
5412 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5416 * Clones from send_root are allowed, but only if the clone source
5417 * is behind the current send position. This is checked while searching
5418 * for possible clone sources.
5420 sctx
->clone_roots
[sctx
->clone_roots_cnt
++].root
= sctx
->send_root
;
5422 /* We do a bsearch later */
5423 sort(sctx
->clone_roots
, sctx
->clone_roots_cnt
,
5424 sizeof(*sctx
->clone_roots
), __clone_root_cmp_sort
,
5426 sort_clone_roots
= 1;
5428 ret
= send_subvol(sctx
);
5432 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_END_CMD
)) {
5433 ret
= begin_cmd(sctx
, BTRFS_SEND_C_END
);
5436 ret
= send_cmd(sctx
);
5442 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
));
5443 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
)) {
5445 struct pending_dir_move
*pm
;
5447 n
= rb_first(&sctx
->pending_dir_moves
);
5448 pm
= rb_entry(n
, struct pending_dir_move
, node
);
5449 while (!list_empty(&pm
->list
)) {
5450 struct pending_dir_move
*pm2
;
5452 pm2
= list_first_entry(&pm
->list
,
5453 struct pending_dir_move
, list
);
5454 free_pending_move(sctx
, pm2
);
5456 free_pending_move(sctx
, pm
);
5459 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
));
5460 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
)) {
5462 struct waiting_dir_move
*dm
;
5464 n
= rb_first(&sctx
->waiting_dir_moves
);
5465 dm
= rb_entry(n
, struct waiting_dir_move
, node
);
5466 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
5470 if (sort_clone_roots
) {
5471 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
5472 btrfs_root_dec_send_in_progress(
5473 sctx
->clone_roots
[i
].root
);
5475 for (i
= 0; sctx
&& i
< clone_sources_to_rollback
; i
++)
5476 btrfs_root_dec_send_in_progress(
5477 sctx
->clone_roots
[i
].root
);
5479 btrfs_root_dec_send_in_progress(send_root
);
5481 if (sctx
&& !IS_ERR_OR_NULL(sctx
->parent_root
))
5482 btrfs_root_dec_send_in_progress(sctx
->parent_root
);
5485 vfree(clone_sources_tmp
);
5488 if (sctx
->send_filp
)
5489 fput(sctx
->send_filp
);
5491 vfree(sctx
->clone_roots
);
5492 vfree(sctx
->send_buf
);
5493 vfree(sctx
->read_buf
);
5495 name_cache_free(sctx
);