2 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
35 #include "btrfs_inode.h"
36 #include "transaction.h"
38 static int g_verbose
= 0;
40 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
56 unsigned short buf_len
:15;
57 unsigned short reversed
:1;
61 * Average path length does not exceed 200 bytes, we'll have
62 * better packing in the slab and higher chance to satisfy
63 * a allocation later during send.
68 #define FS_PATH_INLINE_SIZE \
69 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
72 /* reused for each extent */
74 struct btrfs_root
*root
;
81 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
82 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
85 struct file
*send_filp
;
91 u64 cmd_send_size
[BTRFS_SEND_C_MAX
+ 1];
92 u64 flags
; /* 'flags' member of btrfs_ioctl_send_args is u64 */
94 struct btrfs_root
*send_root
;
95 struct btrfs_root
*parent_root
;
96 struct clone_root
*clone_roots
;
99 /* current state of the compare_tree call */
100 struct btrfs_path
*left_path
;
101 struct btrfs_path
*right_path
;
102 struct btrfs_key
*cmp_key
;
105 * infos of the currently processed inode. In case of deleted inodes,
106 * these are the values from the deleted inode.
111 int cur_inode_new_gen
;
112 int cur_inode_deleted
;
116 u64 cur_inode_last_extent
;
120 struct list_head new_refs
;
121 struct list_head deleted_refs
;
123 struct radix_tree_root name_cache
;
124 struct list_head name_cache_list
;
127 struct file_ra_state ra
;
132 * We process inodes by their increasing order, so if before an
133 * incremental send we reverse the parent/child relationship of
134 * directories such that a directory with a lower inode number was
135 * the parent of a directory with a higher inode number, and the one
136 * becoming the new parent got renamed too, we can't rename/move the
137 * directory with lower inode number when we finish processing it - we
138 * must process the directory with higher inode number first, then
139 * rename/move it and then rename/move the directory with lower inode
140 * number. Example follows.
142 * Tree state when the first send was performed:
154 * Tree state when the second (incremental) send is performed:
163 * The sequence of steps that lead to the second state was:
165 * mv /a/b/c/d /a/b/c2/d2
166 * mv /a/b/c /a/b/c2/d2/cc
168 * "c" has lower inode number, but we can't move it (2nd mv operation)
169 * before we move "d", which has higher inode number.
171 * So we just memorize which move/rename operations must be performed
172 * later when their respective parent is processed and moved/renamed.
175 /* Indexed by parent directory inode number. */
176 struct rb_root pending_dir_moves
;
179 * Reverse index, indexed by the inode number of a directory that
180 * is waiting for the move/rename of its immediate parent before its
181 * own move/rename can be performed.
183 struct rb_root waiting_dir_moves
;
186 * A directory that is going to be rm'ed might have a child directory
187 * which is in the pending directory moves index above. In this case,
188 * the directory can only be removed after the move/rename of its child
189 * is performed. Example:
209 * Sequence of steps that lead to the send snapshot:
210 * rm -f /a/b/c/foo.txt
212 * mv /a/b/c/x /a/b/YY
215 * When the child is processed, its move/rename is delayed until its
216 * parent is processed (as explained above), but all other operations
217 * like update utimes, chown, chgrp, etc, are performed and the paths
218 * that it uses for those operations must use the orphanized name of
219 * its parent (the directory we're going to rm later), so we need to
220 * memorize that name.
222 * Indexed by the inode number of the directory to be deleted.
224 struct rb_root orphan_dirs
;
227 struct pending_dir_move
{
229 struct list_head list
;
234 struct list_head update_refs
;
237 struct waiting_dir_move
{
241 * There might be some directory that could not be removed because it
242 * was waiting for this directory inode to be moved first. Therefore
243 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
249 struct orphan_dir_info
{
255 struct name_cache_entry
{
256 struct list_head list
;
258 * radix_tree has only 32bit entries but we need to handle 64bit inums.
259 * We use the lower 32bit of the 64bit inum to store it in the tree. If
260 * more then one inum would fall into the same entry, we use radix_list
261 * to store the additional entries. radix_list is also used to store
262 * entries where two entries have the same inum but different
265 struct list_head radix_list
;
271 int need_later_update
;
276 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
);
278 static struct waiting_dir_move
*
279 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
);
281 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
);
283 static int need_send_hole(struct send_ctx
*sctx
)
285 return (sctx
->parent_root
&& !sctx
->cur_inode_new
&&
286 !sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
&&
287 S_ISREG(sctx
->cur_inode_mode
));
290 static void fs_path_reset(struct fs_path
*p
)
293 p
->start
= p
->buf
+ p
->buf_len
- 1;
303 static struct fs_path
*fs_path_alloc(void)
307 p
= kmalloc(sizeof(*p
), GFP_NOFS
);
311 p
->buf
= p
->inline_buf
;
312 p
->buf_len
= FS_PATH_INLINE_SIZE
;
317 static struct fs_path
*fs_path_alloc_reversed(void)
329 static void fs_path_free(struct fs_path
*p
)
333 if (p
->buf
!= p
->inline_buf
)
338 static int fs_path_len(struct fs_path
*p
)
340 return p
->end
- p
->start
;
343 static int fs_path_ensure_buf(struct fs_path
*p
, int len
)
351 if (p
->buf_len
>= len
)
354 if (len
> PATH_MAX
) {
359 path_len
= p
->end
- p
->start
;
360 old_buf_len
= p
->buf_len
;
363 * First time the inline_buf does not suffice
365 if (p
->buf
== p
->inline_buf
) {
366 tmp_buf
= kmalloc(len
, GFP_NOFS
);
368 memcpy(tmp_buf
, p
->buf
, old_buf_len
);
370 tmp_buf
= krealloc(p
->buf
, len
, GFP_NOFS
);
376 * The real size of the buffer is bigger, this will let the fast path
377 * happen most of the time
379 p
->buf_len
= ksize(p
->buf
);
382 tmp_buf
= p
->buf
+ old_buf_len
- path_len
- 1;
383 p
->end
= p
->buf
+ p
->buf_len
- 1;
384 p
->start
= p
->end
- path_len
;
385 memmove(p
->start
, tmp_buf
, path_len
+ 1);
388 p
->end
= p
->start
+ path_len
;
393 static int fs_path_prepare_for_add(struct fs_path
*p
, int name_len
,
399 new_len
= p
->end
- p
->start
+ name_len
;
400 if (p
->start
!= p
->end
)
402 ret
= fs_path_ensure_buf(p
, new_len
);
407 if (p
->start
!= p
->end
)
409 p
->start
-= name_len
;
410 *prepared
= p
->start
;
412 if (p
->start
!= p
->end
)
423 static int fs_path_add(struct fs_path
*p
, const char *name
, int name_len
)
428 ret
= fs_path_prepare_for_add(p
, name_len
, &prepared
);
431 memcpy(prepared
, name
, name_len
);
437 static int fs_path_add_path(struct fs_path
*p
, struct fs_path
*p2
)
442 ret
= fs_path_prepare_for_add(p
, p2
->end
- p2
->start
, &prepared
);
445 memcpy(prepared
, p2
->start
, p2
->end
- p2
->start
);
451 static int fs_path_add_from_extent_buffer(struct fs_path
*p
,
452 struct extent_buffer
*eb
,
453 unsigned long off
, int len
)
458 ret
= fs_path_prepare_for_add(p
, len
, &prepared
);
462 read_extent_buffer(eb
, prepared
, off
, len
);
468 static int fs_path_copy(struct fs_path
*p
, struct fs_path
*from
)
472 p
->reversed
= from
->reversed
;
475 ret
= fs_path_add_path(p
, from
);
481 static void fs_path_unreverse(struct fs_path
*p
)
490 len
= p
->end
- p
->start
;
492 p
->end
= p
->start
+ len
;
493 memmove(p
->start
, tmp
, len
+ 1);
497 static struct btrfs_path
*alloc_path_for_send(void)
499 struct btrfs_path
*path
;
501 path
= btrfs_alloc_path();
504 path
->search_commit_root
= 1;
505 path
->skip_locking
= 1;
506 path
->need_commit_sem
= 1;
510 static int write_buf(struct file
*filp
, const void *buf
, u32 len
, loff_t
*off
)
520 ret
= vfs_write(filp
, (__force
const char __user
*)buf
+ pos
,
522 /* TODO handle that correctly */
523 /*if (ret == -ERESTARTSYS) {
542 static int tlv_put(struct send_ctx
*sctx
, u16 attr
, const void *data
, int len
)
544 struct btrfs_tlv_header
*hdr
;
545 int total_len
= sizeof(*hdr
) + len
;
546 int left
= sctx
->send_max_size
- sctx
->send_size
;
548 if (unlikely(left
< total_len
))
551 hdr
= (struct btrfs_tlv_header
*) (sctx
->send_buf
+ sctx
->send_size
);
552 hdr
->tlv_type
= cpu_to_le16(attr
);
553 hdr
->tlv_len
= cpu_to_le16(len
);
554 memcpy(hdr
+ 1, data
, len
);
555 sctx
->send_size
+= total_len
;
560 #define TLV_PUT_DEFINE_INT(bits) \
561 static int tlv_put_u##bits(struct send_ctx *sctx, \
562 u##bits attr, u##bits value) \
564 __le##bits __tmp = cpu_to_le##bits(value); \
565 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
568 TLV_PUT_DEFINE_INT(64)
570 static int tlv_put_string(struct send_ctx
*sctx
, u16 attr
,
571 const char *str
, int len
)
575 return tlv_put(sctx
, attr
, str
, len
);
578 static int tlv_put_uuid(struct send_ctx
*sctx
, u16 attr
,
581 return tlv_put(sctx
, attr
, uuid
, BTRFS_UUID_SIZE
);
584 static int tlv_put_btrfs_timespec(struct send_ctx
*sctx
, u16 attr
,
585 struct extent_buffer
*eb
,
586 struct btrfs_timespec
*ts
)
588 struct btrfs_timespec bts
;
589 read_extent_buffer(eb
, &bts
, (unsigned long)ts
, sizeof(bts
));
590 return tlv_put(sctx
, attr
, &bts
, sizeof(bts
));
594 #define TLV_PUT(sctx, attrtype, attrlen, data) \
596 ret = tlv_put(sctx, attrtype, attrlen, data); \
598 goto tlv_put_failure; \
601 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
603 ret = tlv_put_u##bits(sctx, attrtype, value); \
605 goto tlv_put_failure; \
608 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
609 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
610 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
611 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
612 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
614 ret = tlv_put_string(sctx, attrtype, str, len); \
616 goto tlv_put_failure; \
618 #define TLV_PUT_PATH(sctx, attrtype, p) \
620 ret = tlv_put_string(sctx, attrtype, p->start, \
621 p->end - p->start); \
623 goto tlv_put_failure; \
625 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
627 ret = tlv_put_uuid(sctx, attrtype, uuid); \
629 goto tlv_put_failure; \
631 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
633 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
635 goto tlv_put_failure; \
638 static int send_header(struct send_ctx
*sctx
)
640 struct btrfs_stream_header hdr
;
642 strcpy(hdr
.magic
, BTRFS_SEND_STREAM_MAGIC
);
643 hdr
.version
= cpu_to_le32(BTRFS_SEND_STREAM_VERSION
);
645 return write_buf(sctx
->send_filp
, &hdr
, sizeof(hdr
),
650 * For each command/item we want to send to userspace, we call this function.
652 static int begin_cmd(struct send_ctx
*sctx
, int cmd
)
654 struct btrfs_cmd_header
*hdr
;
656 if (WARN_ON(!sctx
->send_buf
))
659 BUG_ON(sctx
->send_size
);
661 sctx
->send_size
+= sizeof(*hdr
);
662 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
663 hdr
->cmd
= cpu_to_le16(cmd
);
668 static int send_cmd(struct send_ctx
*sctx
)
671 struct btrfs_cmd_header
*hdr
;
674 hdr
= (struct btrfs_cmd_header
*)sctx
->send_buf
;
675 hdr
->len
= cpu_to_le32(sctx
->send_size
- sizeof(*hdr
));
678 crc
= btrfs_crc32c(0, (unsigned char *)sctx
->send_buf
, sctx
->send_size
);
679 hdr
->crc
= cpu_to_le32(crc
);
681 ret
= write_buf(sctx
->send_filp
, sctx
->send_buf
, sctx
->send_size
,
684 sctx
->total_send_size
+= sctx
->send_size
;
685 sctx
->cmd_send_size
[le16_to_cpu(hdr
->cmd
)] += sctx
->send_size
;
692 * Sends a move instruction to user space
694 static int send_rename(struct send_ctx
*sctx
,
695 struct fs_path
*from
, struct fs_path
*to
)
699 verbose_printk("btrfs: send_rename %s -> %s\n", from
->start
, to
->start
);
701 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RENAME
);
705 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, from
);
706 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_TO
, to
);
708 ret
= send_cmd(sctx
);
716 * Sends a link instruction to user space
718 static int send_link(struct send_ctx
*sctx
,
719 struct fs_path
*path
, struct fs_path
*lnk
)
723 verbose_printk("btrfs: send_link %s -> %s\n", path
->start
, lnk
->start
);
725 ret
= begin_cmd(sctx
, BTRFS_SEND_C_LINK
);
729 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
730 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, lnk
);
732 ret
= send_cmd(sctx
);
740 * Sends an unlink instruction to user space
742 static int send_unlink(struct send_ctx
*sctx
, struct fs_path
*path
)
746 verbose_printk("btrfs: send_unlink %s\n", path
->start
);
748 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UNLINK
);
752 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
754 ret
= send_cmd(sctx
);
762 * Sends a rmdir instruction to user space
764 static int send_rmdir(struct send_ctx
*sctx
, struct fs_path
*path
)
768 verbose_printk("btrfs: send_rmdir %s\n", path
->start
);
770 ret
= begin_cmd(sctx
, BTRFS_SEND_C_RMDIR
);
774 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
776 ret
= send_cmd(sctx
);
784 * Helper function to retrieve some fields from an inode item.
786 static int __get_inode_info(struct btrfs_root
*root
, struct btrfs_path
*path
,
787 u64 ino
, u64
*size
, u64
*gen
, u64
*mode
, u64
*uid
,
791 struct btrfs_inode_item
*ii
;
792 struct btrfs_key key
;
795 key
.type
= BTRFS_INODE_ITEM_KEY
;
797 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
804 ii
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
805 struct btrfs_inode_item
);
807 *size
= btrfs_inode_size(path
->nodes
[0], ii
);
809 *gen
= btrfs_inode_generation(path
->nodes
[0], ii
);
811 *mode
= btrfs_inode_mode(path
->nodes
[0], ii
);
813 *uid
= btrfs_inode_uid(path
->nodes
[0], ii
);
815 *gid
= btrfs_inode_gid(path
->nodes
[0], ii
);
817 *rdev
= btrfs_inode_rdev(path
->nodes
[0], ii
);
822 static int get_inode_info(struct btrfs_root
*root
,
823 u64 ino
, u64
*size
, u64
*gen
,
824 u64
*mode
, u64
*uid
, u64
*gid
,
827 struct btrfs_path
*path
;
830 path
= alloc_path_for_send();
833 ret
= __get_inode_info(root
, path
, ino
, size
, gen
, mode
, uid
, gid
,
835 btrfs_free_path(path
);
839 typedef int (*iterate_inode_ref_t
)(int num
, u64 dir
, int index
,
844 * Helper function to iterate the entries in ONE btrfs_inode_ref or
845 * btrfs_inode_extref.
846 * The iterate callback may return a non zero value to stop iteration. This can
847 * be a negative value for error codes or 1 to simply stop it.
849 * path must point to the INODE_REF or INODE_EXTREF when called.
851 static int iterate_inode_ref(struct btrfs_root
*root
, struct btrfs_path
*path
,
852 struct btrfs_key
*found_key
, int resolve
,
853 iterate_inode_ref_t iterate
, void *ctx
)
855 struct extent_buffer
*eb
= path
->nodes
[0];
856 struct btrfs_item
*item
;
857 struct btrfs_inode_ref
*iref
;
858 struct btrfs_inode_extref
*extref
;
859 struct btrfs_path
*tmp_path
;
863 int slot
= path
->slots
[0];
870 unsigned long name_off
;
871 unsigned long elem_size
;
874 p
= fs_path_alloc_reversed();
878 tmp_path
= alloc_path_for_send();
885 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
886 ptr
= (unsigned long)btrfs_item_ptr(eb
, slot
,
887 struct btrfs_inode_ref
);
888 item
= btrfs_item_nr(slot
);
889 total
= btrfs_item_size(eb
, item
);
890 elem_size
= sizeof(*iref
);
892 ptr
= btrfs_item_ptr_offset(eb
, slot
);
893 total
= btrfs_item_size_nr(eb
, slot
);
894 elem_size
= sizeof(*extref
);
897 while (cur
< total
) {
900 if (found_key
->type
== BTRFS_INODE_REF_KEY
) {
901 iref
= (struct btrfs_inode_ref
*)(ptr
+ cur
);
902 name_len
= btrfs_inode_ref_name_len(eb
, iref
);
903 name_off
= (unsigned long)(iref
+ 1);
904 index
= btrfs_inode_ref_index(eb
, iref
);
905 dir
= found_key
->offset
;
907 extref
= (struct btrfs_inode_extref
*)(ptr
+ cur
);
908 name_len
= btrfs_inode_extref_name_len(eb
, extref
);
909 name_off
= (unsigned long)&extref
->name
;
910 index
= btrfs_inode_extref_index(eb
, extref
);
911 dir
= btrfs_inode_extref_parent(eb
, extref
);
915 start
= btrfs_ref_to_path(root
, tmp_path
, name_len
,
919 ret
= PTR_ERR(start
);
922 if (start
< p
->buf
) {
923 /* overflow , try again with larger buffer */
924 ret
= fs_path_ensure_buf(p
,
925 p
->buf_len
+ p
->buf
- start
);
928 start
= btrfs_ref_to_path(root
, tmp_path
,
933 ret
= PTR_ERR(start
);
936 BUG_ON(start
< p
->buf
);
940 ret
= fs_path_add_from_extent_buffer(p
, eb
, name_off
,
946 cur
+= elem_size
+ name_len
;
947 ret
= iterate(num
, dir
, index
, p
, ctx
);
954 btrfs_free_path(tmp_path
);
959 typedef int (*iterate_dir_item_t
)(int num
, struct btrfs_key
*di_key
,
960 const char *name
, int name_len
,
961 const char *data
, int data_len
,
965 * Helper function to iterate the entries in ONE btrfs_dir_item.
966 * The iterate callback may return a non zero value to stop iteration. This can
967 * be a negative value for error codes or 1 to simply stop it.
969 * path must point to the dir item when called.
971 static int iterate_dir_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
972 struct btrfs_key
*found_key
,
973 iterate_dir_item_t iterate
, void *ctx
)
976 struct extent_buffer
*eb
;
977 struct btrfs_item
*item
;
978 struct btrfs_dir_item
*di
;
979 struct btrfs_key di_key
;
992 * Start with a small buffer (1 page). If later we end up needing more
993 * space, which can happen for xattrs on a fs with a leaf size greater
994 * then the page size, attempt to increase the buffer. Typically xattr
998 buf
= kmalloc(buf_len
, GFP_NOFS
);
1004 eb
= path
->nodes
[0];
1005 slot
= path
->slots
[0];
1006 item
= btrfs_item_nr(slot
);
1007 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
1010 total
= btrfs_item_size(eb
, item
);
1013 while (cur
< total
) {
1014 name_len
= btrfs_dir_name_len(eb
, di
);
1015 data_len
= btrfs_dir_data_len(eb
, di
);
1016 type
= btrfs_dir_type(eb
, di
);
1017 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
1019 if (type
== BTRFS_FT_XATTR
) {
1020 if (name_len
> XATTR_NAME_MAX
) {
1021 ret
= -ENAMETOOLONG
;
1024 if (name_len
+ data_len
> BTRFS_MAX_XATTR_SIZE(root
)) {
1032 if (name_len
+ data_len
> PATH_MAX
) {
1033 ret
= -ENAMETOOLONG
;
1038 if (name_len
+ data_len
> buf_len
) {
1039 buf_len
= name_len
+ data_len
;
1040 if (is_vmalloc_addr(buf
)) {
1044 char *tmp
= krealloc(buf
, buf_len
,
1045 GFP_NOFS
| __GFP_NOWARN
);
1052 buf
= vmalloc(buf_len
);
1060 read_extent_buffer(eb
, buf
, (unsigned long)(di
+ 1),
1061 name_len
+ data_len
);
1063 len
= sizeof(*di
) + name_len
+ data_len
;
1064 di
= (struct btrfs_dir_item
*)((char *)di
+ len
);
1067 ret
= iterate(num
, &di_key
, buf
, name_len
, buf
+ name_len
,
1068 data_len
, type
, ctx
);
1084 static int __copy_first_ref(int num
, u64 dir
, int index
,
1085 struct fs_path
*p
, void *ctx
)
1088 struct fs_path
*pt
= ctx
;
1090 ret
= fs_path_copy(pt
, p
);
1094 /* we want the first only */
1099 * Retrieve the first path of an inode. If an inode has more then one
1100 * ref/hardlink, this is ignored.
1102 static int get_inode_path(struct btrfs_root
*root
,
1103 u64 ino
, struct fs_path
*path
)
1106 struct btrfs_key key
, found_key
;
1107 struct btrfs_path
*p
;
1109 p
= alloc_path_for_send();
1113 fs_path_reset(path
);
1116 key
.type
= BTRFS_INODE_REF_KEY
;
1119 ret
= btrfs_search_slot_for_read(root
, &key
, p
, 1, 0);
1126 btrfs_item_key_to_cpu(p
->nodes
[0], &found_key
, p
->slots
[0]);
1127 if (found_key
.objectid
!= ino
||
1128 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1129 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1134 ret
= iterate_inode_ref(root
, p
, &found_key
, 1,
1135 __copy_first_ref
, path
);
1145 struct backref_ctx
{
1146 struct send_ctx
*sctx
;
1148 struct btrfs_path
*path
;
1149 /* number of total found references */
1153 * used for clones found in send_root. clones found behind cur_objectid
1154 * and cur_offset are not considered as allowed clones.
1159 /* may be truncated in case it's the last extent in a file */
1162 /* data offset in the file extent item */
1165 /* Just to check for bugs in backref resolving */
1169 static int __clone_root_cmp_bsearch(const void *key
, const void *elt
)
1171 u64 root
= (u64
)(uintptr_t)key
;
1172 struct clone_root
*cr
= (struct clone_root
*)elt
;
1174 if (root
< cr
->root
->objectid
)
1176 if (root
> cr
->root
->objectid
)
1181 static int __clone_root_cmp_sort(const void *e1
, const void *e2
)
1183 struct clone_root
*cr1
= (struct clone_root
*)e1
;
1184 struct clone_root
*cr2
= (struct clone_root
*)e2
;
1186 if (cr1
->root
->objectid
< cr2
->root
->objectid
)
1188 if (cr1
->root
->objectid
> cr2
->root
->objectid
)
1194 * Called for every backref that is found for the current extent.
1195 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1197 static int __iterate_backrefs(u64 ino
, u64 offset
, u64 root
, void *ctx_
)
1199 struct backref_ctx
*bctx
= ctx_
;
1200 struct clone_root
*found
;
1204 /* First check if the root is in the list of accepted clone sources */
1205 found
= bsearch((void *)(uintptr_t)root
, bctx
->sctx
->clone_roots
,
1206 bctx
->sctx
->clone_roots_cnt
,
1207 sizeof(struct clone_root
),
1208 __clone_root_cmp_bsearch
);
1212 if (found
->root
== bctx
->sctx
->send_root
&&
1213 ino
== bctx
->cur_objectid
&&
1214 offset
== bctx
->cur_offset
) {
1215 bctx
->found_itself
= 1;
1219 * There are inodes that have extents that lie behind its i_size. Don't
1220 * accept clones from these extents.
1222 ret
= __get_inode_info(found
->root
, bctx
->path
, ino
, &i_size
, NULL
, NULL
,
1224 btrfs_release_path(bctx
->path
);
1228 if (offset
+ bctx
->data_offset
+ bctx
->extent_len
> i_size
)
1232 * Make sure we don't consider clones from send_root that are
1233 * behind the current inode/offset.
1235 if (found
->root
== bctx
->sctx
->send_root
) {
1237 * TODO for the moment we don't accept clones from the inode
1238 * that is currently send. We may change this when
1239 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1242 if (ino
>= bctx
->cur_objectid
)
1245 if (ino
> bctx
->cur_objectid
)
1247 if (offset
+ bctx
->extent_len
> bctx
->cur_offset
)
1253 found
->found_refs
++;
1254 if (ino
< found
->ino
) {
1256 found
->offset
= offset
;
1257 } else if (found
->ino
== ino
) {
1259 * same extent found more then once in the same file.
1261 if (found
->offset
> offset
+ bctx
->extent_len
)
1262 found
->offset
= offset
;
1269 * Given an inode, offset and extent item, it finds a good clone for a clone
1270 * instruction. Returns -ENOENT when none could be found. The function makes
1271 * sure that the returned clone is usable at the point where sending is at the
1272 * moment. This means, that no clones are accepted which lie behind the current
1275 * path must point to the extent item when called.
1277 static int find_extent_clone(struct send_ctx
*sctx
,
1278 struct btrfs_path
*path
,
1279 u64 ino
, u64 data_offset
,
1281 struct clone_root
**found
)
1288 u64 extent_item_pos
;
1290 struct btrfs_file_extent_item
*fi
;
1291 struct extent_buffer
*eb
= path
->nodes
[0];
1292 struct backref_ctx
*backref_ctx
= NULL
;
1293 struct clone_root
*cur_clone_root
;
1294 struct btrfs_key found_key
;
1295 struct btrfs_path
*tmp_path
;
1299 tmp_path
= alloc_path_for_send();
1303 /* We only use this path under the commit sem */
1304 tmp_path
->need_commit_sem
= 0;
1306 backref_ctx
= kmalloc(sizeof(*backref_ctx
), GFP_NOFS
);
1312 backref_ctx
->path
= tmp_path
;
1314 if (data_offset
>= ino_size
) {
1316 * There may be extents that lie behind the file's size.
1317 * I at least had this in combination with snapshotting while
1318 * writing large files.
1324 fi
= btrfs_item_ptr(eb
, path
->slots
[0],
1325 struct btrfs_file_extent_item
);
1326 extent_type
= btrfs_file_extent_type(eb
, fi
);
1327 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1331 compressed
= btrfs_file_extent_compression(eb
, fi
);
1333 num_bytes
= btrfs_file_extent_num_bytes(eb
, fi
);
1334 disk_byte
= btrfs_file_extent_disk_bytenr(eb
, fi
);
1335 if (disk_byte
== 0) {
1339 logical
= disk_byte
+ btrfs_file_extent_offset(eb
, fi
);
1341 down_read(&sctx
->send_root
->fs_info
->commit_root_sem
);
1342 ret
= extent_from_logical(sctx
->send_root
->fs_info
, disk_byte
, tmp_path
,
1343 &found_key
, &flags
);
1344 up_read(&sctx
->send_root
->fs_info
->commit_root_sem
);
1345 btrfs_release_path(tmp_path
);
1349 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1355 * Setup the clone roots.
1357 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1358 cur_clone_root
= sctx
->clone_roots
+ i
;
1359 cur_clone_root
->ino
= (u64
)-1;
1360 cur_clone_root
->offset
= 0;
1361 cur_clone_root
->found_refs
= 0;
1364 backref_ctx
->sctx
= sctx
;
1365 backref_ctx
->found
= 0;
1366 backref_ctx
->cur_objectid
= ino
;
1367 backref_ctx
->cur_offset
= data_offset
;
1368 backref_ctx
->found_itself
= 0;
1369 backref_ctx
->extent_len
= num_bytes
;
1371 * For non-compressed extents iterate_extent_inodes() gives us extent
1372 * offsets that already take into account the data offset, but not for
1373 * compressed extents, since the offset is logical and not relative to
1374 * the physical extent locations. We must take this into account to
1375 * avoid sending clone offsets that go beyond the source file's size,
1376 * which would result in the clone ioctl failing with -EINVAL on the
1379 if (compressed
== BTRFS_COMPRESS_NONE
)
1380 backref_ctx
->data_offset
= 0;
1382 backref_ctx
->data_offset
= btrfs_file_extent_offset(eb
, fi
);
1385 * The last extent of a file may be too large due to page alignment.
1386 * We need to adjust extent_len in this case so that the checks in
1387 * __iterate_backrefs work.
1389 if (data_offset
+ num_bytes
>= ino_size
)
1390 backref_ctx
->extent_len
= ino_size
- data_offset
;
1393 * Now collect all backrefs.
1395 if (compressed
== BTRFS_COMPRESS_NONE
)
1396 extent_item_pos
= logical
- found_key
.objectid
;
1398 extent_item_pos
= 0;
1399 ret
= iterate_extent_inodes(sctx
->send_root
->fs_info
,
1400 found_key
.objectid
, extent_item_pos
, 1,
1401 __iterate_backrefs
, backref_ctx
);
1406 if (!backref_ctx
->found_itself
) {
1407 /* found a bug in backref code? */
1409 btrfs_err(sctx
->send_root
->fs_info
, "did not find backref in "
1410 "send_root. inode=%llu, offset=%llu, "
1411 "disk_byte=%llu found extent=%llu",
1412 ino
, data_offset
, disk_byte
, found_key
.objectid
);
1416 verbose_printk(KERN_DEBUG
"btrfs: find_extent_clone: data_offset=%llu, "
1418 "num_bytes=%llu, logical=%llu\n",
1419 data_offset
, ino
, num_bytes
, logical
);
1421 if (!backref_ctx
->found
)
1422 verbose_printk("btrfs: no clones found\n");
1424 cur_clone_root
= NULL
;
1425 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++) {
1426 if (sctx
->clone_roots
[i
].found_refs
) {
1427 if (!cur_clone_root
)
1428 cur_clone_root
= sctx
->clone_roots
+ i
;
1429 else if (sctx
->clone_roots
[i
].root
== sctx
->send_root
)
1430 /* prefer clones from send_root over others */
1431 cur_clone_root
= sctx
->clone_roots
+ i
;
1436 if (cur_clone_root
) {
1437 *found
= cur_clone_root
;
1444 btrfs_free_path(tmp_path
);
1449 static int read_symlink(struct btrfs_root
*root
,
1451 struct fs_path
*dest
)
1454 struct btrfs_path
*path
;
1455 struct btrfs_key key
;
1456 struct btrfs_file_extent_item
*ei
;
1462 path
= alloc_path_for_send();
1467 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1469 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1474 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1475 struct btrfs_file_extent_item
);
1476 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
1477 compression
= btrfs_file_extent_compression(path
->nodes
[0], ei
);
1478 BUG_ON(type
!= BTRFS_FILE_EXTENT_INLINE
);
1479 BUG_ON(compression
);
1481 off
= btrfs_file_extent_inline_start(ei
);
1482 len
= btrfs_file_extent_inline_len(path
->nodes
[0], path
->slots
[0], ei
);
1484 ret
= fs_path_add_from_extent_buffer(dest
, path
->nodes
[0], off
, len
);
1487 btrfs_free_path(path
);
1492 * Helper function to generate a file name that is unique in the root of
1493 * send_root and parent_root. This is used to generate names for orphan inodes.
1495 static int gen_unique_name(struct send_ctx
*sctx
,
1497 struct fs_path
*dest
)
1500 struct btrfs_path
*path
;
1501 struct btrfs_dir_item
*di
;
1506 path
= alloc_path_for_send();
1511 len
= snprintf(tmp
, sizeof(tmp
), "o%llu-%llu-%llu",
1513 ASSERT(len
< sizeof(tmp
));
1515 di
= btrfs_lookup_dir_item(NULL
, sctx
->send_root
,
1516 path
, BTRFS_FIRST_FREE_OBJECTID
,
1517 tmp
, strlen(tmp
), 0);
1518 btrfs_release_path(path
);
1524 /* not unique, try again */
1529 if (!sctx
->parent_root
) {
1535 di
= btrfs_lookup_dir_item(NULL
, sctx
->parent_root
,
1536 path
, BTRFS_FIRST_FREE_OBJECTID
,
1537 tmp
, strlen(tmp
), 0);
1538 btrfs_release_path(path
);
1544 /* not unique, try again */
1552 ret
= fs_path_add(dest
, tmp
, strlen(tmp
));
1555 btrfs_free_path(path
);
1560 inode_state_no_change
,
1561 inode_state_will_create
,
1562 inode_state_did_create
,
1563 inode_state_will_delete
,
1564 inode_state_did_delete
,
1567 static int get_cur_inode_state(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1575 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &left_gen
, NULL
, NULL
,
1577 if (ret
< 0 && ret
!= -ENOENT
)
1581 if (!sctx
->parent_root
) {
1582 right_ret
= -ENOENT
;
1584 ret
= get_inode_info(sctx
->parent_root
, ino
, NULL
, &right_gen
,
1585 NULL
, NULL
, NULL
, NULL
);
1586 if (ret
< 0 && ret
!= -ENOENT
)
1591 if (!left_ret
&& !right_ret
) {
1592 if (left_gen
== gen
&& right_gen
== gen
) {
1593 ret
= inode_state_no_change
;
1594 } else if (left_gen
== gen
) {
1595 if (ino
< sctx
->send_progress
)
1596 ret
= inode_state_did_create
;
1598 ret
= inode_state_will_create
;
1599 } else if (right_gen
== gen
) {
1600 if (ino
< sctx
->send_progress
)
1601 ret
= inode_state_did_delete
;
1603 ret
= inode_state_will_delete
;
1607 } else if (!left_ret
) {
1608 if (left_gen
== gen
) {
1609 if (ino
< sctx
->send_progress
)
1610 ret
= inode_state_did_create
;
1612 ret
= inode_state_will_create
;
1616 } else if (!right_ret
) {
1617 if (right_gen
== gen
) {
1618 if (ino
< sctx
->send_progress
)
1619 ret
= inode_state_did_delete
;
1621 ret
= inode_state_will_delete
;
1633 static int is_inode_existent(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1637 ret
= get_cur_inode_state(sctx
, ino
, gen
);
1641 if (ret
== inode_state_no_change
||
1642 ret
== inode_state_did_create
||
1643 ret
== inode_state_will_delete
)
1653 * Helper function to lookup a dir item in a dir.
1655 static int lookup_dir_item_inode(struct btrfs_root
*root
,
1656 u64 dir
, const char *name
, int name_len
,
1661 struct btrfs_dir_item
*di
;
1662 struct btrfs_key key
;
1663 struct btrfs_path
*path
;
1665 path
= alloc_path_for_send();
1669 di
= btrfs_lookup_dir_item(NULL
, root
, path
,
1670 dir
, name
, name_len
, 0);
1679 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
1680 if (key
.type
== BTRFS_ROOT_ITEM_KEY
) {
1684 *found_inode
= key
.objectid
;
1685 *found_type
= btrfs_dir_type(path
->nodes
[0], di
);
1688 btrfs_free_path(path
);
1693 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1694 * generation of the parent dir and the name of the dir entry.
1696 static int get_first_ref(struct btrfs_root
*root
, u64 ino
,
1697 u64
*dir
, u64
*dir_gen
, struct fs_path
*name
)
1700 struct btrfs_key key
;
1701 struct btrfs_key found_key
;
1702 struct btrfs_path
*path
;
1706 path
= alloc_path_for_send();
1711 key
.type
= BTRFS_INODE_REF_KEY
;
1714 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 1, 0);
1718 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1720 if (ret
|| found_key
.objectid
!= ino
||
1721 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
1722 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
)) {
1727 if (found_key
.type
== BTRFS_INODE_REF_KEY
) {
1728 struct btrfs_inode_ref
*iref
;
1729 iref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1730 struct btrfs_inode_ref
);
1731 len
= btrfs_inode_ref_name_len(path
->nodes
[0], iref
);
1732 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1733 (unsigned long)(iref
+ 1),
1735 parent_dir
= found_key
.offset
;
1737 struct btrfs_inode_extref
*extref
;
1738 extref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1739 struct btrfs_inode_extref
);
1740 len
= btrfs_inode_extref_name_len(path
->nodes
[0], extref
);
1741 ret
= fs_path_add_from_extent_buffer(name
, path
->nodes
[0],
1742 (unsigned long)&extref
->name
, len
);
1743 parent_dir
= btrfs_inode_extref_parent(path
->nodes
[0], extref
);
1747 btrfs_release_path(path
);
1750 ret
= get_inode_info(root
, parent_dir
, NULL
, dir_gen
, NULL
,
1759 btrfs_free_path(path
);
1763 static int is_first_ref(struct btrfs_root
*root
,
1765 const char *name
, int name_len
)
1768 struct fs_path
*tmp_name
;
1771 tmp_name
= fs_path_alloc();
1775 ret
= get_first_ref(root
, ino
, &tmp_dir
, NULL
, tmp_name
);
1779 if (dir
!= tmp_dir
|| name_len
!= fs_path_len(tmp_name
)) {
1784 ret
= !memcmp(tmp_name
->start
, name
, name_len
);
1787 fs_path_free(tmp_name
);
1792 * Used by process_recorded_refs to determine if a new ref would overwrite an
1793 * already existing ref. In case it detects an overwrite, it returns the
1794 * inode/gen in who_ino/who_gen.
1795 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1796 * to make sure later references to the overwritten inode are possible.
1797 * Orphanizing is however only required for the first ref of an inode.
1798 * process_recorded_refs does an additional is_first_ref check to see if
1799 * orphanizing is really required.
1801 static int will_overwrite_ref(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
1802 const char *name
, int name_len
,
1803 u64
*who_ino
, u64
*who_gen
)
1807 u64 other_inode
= 0;
1810 if (!sctx
->parent_root
)
1813 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1818 * If we have a parent root we need to verify that the parent dir was
1819 * not delted and then re-created, if it was then we have no overwrite
1820 * and we can just unlink this entry.
1822 if (sctx
->parent_root
) {
1823 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &gen
, NULL
,
1825 if (ret
< 0 && ret
!= -ENOENT
)
1835 ret
= lookup_dir_item_inode(sctx
->parent_root
, dir
, name
, name_len
,
1836 &other_inode
, &other_type
);
1837 if (ret
< 0 && ret
!= -ENOENT
)
1845 * Check if the overwritten ref was already processed. If yes, the ref
1846 * was already unlinked/moved, so we can safely assume that we will not
1847 * overwrite anything at this point in time.
1849 if (other_inode
> sctx
->send_progress
) {
1850 ret
= get_inode_info(sctx
->parent_root
, other_inode
, NULL
,
1851 who_gen
, NULL
, NULL
, NULL
, NULL
);
1856 *who_ino
= other_inode
;
1866 * Checks if the ref was overwritten by an already processed inode. This is
1867 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1868 * thus the orphan name needs be used.
1869 * process_recorded_refs also uses it to avoid unlinking of refs that were
1872 static int did_overwrite_ref(struct send_ctx
*sctx
,
1873 u64 dir
, u64 dir_gen
,
1874 u64 ino
, u64 ino_gen
,
1875 const char *name
, int name_len
)
1882 if (!sctx
->parent_root
)
1885 ret
= is_inode_existent(sctx
, dir
, dir_gen
);
1889 /* check if the ref was overwritten by another ref */
1890 ret
= lookup_dir_item_inode(sctx
->send_root
, dir
, name
, name_len
,
1891 &ow_inode
, &other_type
);
1892 if (ret
< 0 && ret
!= -ENOENT
)
1895 /* was never and will never be overwritten */
1900 ret
= get_inode_info(sctx
->send_root
, ow_inode
, NULL
, &gen
, NULL
, NULL
,
1905 if (ow_inode
== ino
&& gen
== ino_gen
) {
1911 * We know that it is or will be overwritten. Check this now.
1912 * The current inode being processed might have been the one that caused
1913 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1914 * the current inode being processed.
1916 if ((ow_inode
< sctx
->send_progress
) ||
1917 (ino
!= sctx
->cur_ino
&& ow_inode
== sctx
->cur_ino
&&
1918 gen
== sctx
->cur_inode_gen
))
1928 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1929 * that got overwritten. This is used by process_recorded_refs to determine
1930 * if it has to use the path as returned by get_cur_path or the orphan name.
1932 static int did_overwrite_first_ref(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
1935 struct fs_path
*name
= NULL
;
1939 if (!sctx
->parent_root
)
1942 name
= fs_path_alloc();
1946 ret
= get_first_ref(sctx
->parent_root
, ino
, &dir
, &dir_gen
, name
);
1950 ret
= did_overwrite_ref(sctx
, dir
, dir_gen
, ino
, gen
,
1951 name
->start
, fs_path_len(name
));
1959 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1960 * so we need to do some special handling in case we have clashes. This function
1961 * takes care of this with the help of name_cache_entry::radix_list.
1962 * In case of error, nce is kfreed.
1964 static int name_cache_insert(struct send_ctx
*sctx
,
1965 struct name_cache_entry
*nce
)
1968 struct list_head
*nce_head
;
1970 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
1971 (unsigned long)nce
->ino
);
1973 nce_head
= kmalloc(sizeof(*nce_head
), GFP_NOFS
);
1978 INIT_LIST_HEAD(nce_head
);
1980 ret
= radix_tree_insert(&sctx
->name_cache
, nce
->ino
, nce_head
);
1987 list_add_tail(&nce
->radix_list
, nce_head
);
1988 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
1989 sctx
->name_cache_size
++;
1994 static void name_cache_delete(struct send_ctx
*sctx
,
1995 struct name_cache_entry
*nce
)
1997 struct list_head
*nce_head
;
1999 nce_head
= radix_tree_lookup(&sctx
->name_cache
,
2000 (unsigned long)nce
->ino
);
2002 btrfs_err(sctx
->send_root
->fs_info
,
2003 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2004 nce
->ino
, sctx
->name_cache_size
);
2007 list_del(&nce
->radix_list
);
2008 list_del(&nce
->list
);
2009 sctx
->name_cache_size
--;
2012 * We may not get to the final release of nce_head if the lookup fails
2014 if (nce_head
&& list_empty(nce_head
)) {
2015 radix_tree_delete(&sctx
->name_cache
, (unsigned long)nce
->ino
);
2020 static struct name_cache_entry
*name_cache_search(struct send_ctx
*sctx
,
2023 struct list_head
*nce_head
;
2024 struct name_cache_entry
*cur
;
2026 nce_head
= radix_tree_lookup(&sctx
->name_cache
, (unsigned long)ino
);
2030 list_for_each_entry(cur
, nce_head
, radix_list
) {
2031 if (cur
->ino
== ino
&& cur
->gen
== gen
)
2038 * Removes the entry from the list and adds it back to the end. This marks the
2039 * entry as recently used so that name_cache_clean_unused does not remove it.
2041 static void name_cache_used(struct send_ctx
*sctx
, struct name_cache_entry
*nce
)
2043 list_del(&nce
->list
);
2044 list_add_tail(&nce
->list
, &sctx
->name_cache_list
);
2048 * Remove some entries from the beginning of name_cache_list.
2050 static void name_cache_clean_unused(struct send_ctx
*sctx
)
2052 struct name_cache_entry
*nce
;
2054 if (sctx
->name_cache_size
< SEND_CTX_NAME_CACHE_CLEAN_SIZE
)
2057 while (sctx
->name_cache_size
> SEND_CTX_MAX_NAME_CACHE_SIZE
) {
2058 nce
= list_entry(sctx
->name_cache_list
.next
,
2059 struct name_cache_entry
, list
);
2060 name_cache_delete(sctx
, nce
);
2065 static void name_cache_free(struct send_ctx
*sctx
)
2067 struct name_cache_entry
*nce
;
2069 while (!list_empty(&sctx
->name_cache_list
)) {
2070 nce
= list_entry(sctx
->name_cache_list
.next
,
2071 struct name_cache_entry
, list
);
2072 name_cache_delete(sctx
, nce
);
2078 * Used by get_cur_path for each ref up to the root.
2079 * Returns 0 if it succeeded.
2080 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2081 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2082 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2083 * Returns <0 in case of error.
2085 static int __get_cur_name_and_parent(struct send_ctx
*sctx
,
2089 struct fs_path
*dest
)
2093 struct name_cache_entry
*nce
= NULL
;
2096 * First check if we already did a call to this function with the same
2097 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2098 * return the cached result.
2100 nce
= name_cache_search(sctx
, ino
, gen
);
2102 if (ino
< sctx
->send_progress
&& nce
->need_later_update
) {
2103 name_cache_delete(sctx
, nce
);
2107 name_cache_used(sctx
, nce
);
2108 *parent_ino
= nce
->parent_ino
;
2109 *parent_gen
= nce
->parent_gen
;
2110 ret
= fs_path_add(dest
, nce
->name
, nce
->name_len
);
2119 * If the inode is not existent yet, add the orphan name and return 1.
2120 * This should only happen for the parent dir that we determine in
2123 ret
= is_inode_existent(sctx
, ino
, gen
);
2128 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2136 * Depending on whether the inode was already processed or not, use
2137 * send_root or parent_root for ref lookup.
2139 if (ino
< sctx
->send_progress
)
2140 ret
= get_first_ref(sctx
->send_root
, ino
,
2141 parent_ino
, parent_gen
, dest
);
2143 ret
= get_first_ref(sctx
->parent_root
, ino
,
2144 parent_ino
, parent_gen
, dest
);
2149 * Check if the ref was overwritten by an inode's ref that was processed
2150 * earlier. If yes, treat as orphan and return 1.
2152 ret
= did_overwrite_ref(sctx
, *parent_ino
, *parent_gen
, ino
, gen
,
2153 dest
->start
, dest
->end
- dest
->start
);
2157 fs_path_reset(dest
);
2158 ret
= gen_unique_name(sctx
, ino
, gen
, dest
);
2166 * Store the result of the lookup in the name cache.
2168 nce
= kmalloc(sizeof(*nce
) + fs_path_len(dest
) + 1, GFP_NOFS
);
2176 nce
->parent_ino
= *parent_ino
;
2177 nce
->parent_gen
= *parent_gen
;
2178 nce
->name_len
= fs_path_len(dest
);
2180 strcpy(nce
->name
, dest
->start
);
2182 if (ino
< sctx
->send_progress
)
2183 nce
->need_later_update
= 0;
2185 nce
->need_later_update
= 1;
2187 nce_ret
= name_cache_insert(sctx
, nce
);
2190 name_cache_clean_unused(sctx
);
2197 * Magic happens here. This function returns the first ref to an inode as it
2198 * would look like while receiving the stream at this point in time.
2199 * We walk the path up to the root. For every inode in between, we check if it
2200 * was already processed/sent. If yes, we continue with the parent as found
2201 * in send_root. If not, we continue with the parent as found in parent_root.
2202 * If we encounter an inode that was deleted at this point in time, we use the
2203 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2204 * that were not created yet and overwritten inodes/refs.
2206 * When do we have have orphan inodes:
2207 * 1. When an inode is freshly created and thus no valid refs are available yet
2208 * 2. When a directory lost all it's refs (deleted) but still has dir items
2209 * inside which were not processed yet (pending for move/delete). If anyone
2210 * tried to get the path to the dir items, it would get a path inside that
2212 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2213 * of an unprocessed inode. If in that case the first ref would be
2214 * overwritten, the overwritten inode gets "orphanized". Later when we
2215 * process this overwritten inode, it is restored at a new place by moving
2218 * sctx->send_progress tells this function at which point in time receiving
2221 static int get_cur_path(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2222 struct fs_path
*dest
)
2225 struct fs_path
*name
= NULL
;
2226 u64 parent_inode
= 0;
2230 name
= fs_path_alloc();
2237 fs_path_reset(dest
);
2239 while (!stop
&& ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
2240 struct waiting_dir_move
*wdm
;
2242 fs_path_reset(name
);
2244 if (is_waiting_for_rm(sctx
, ino
)) {
2245 ret
= gen_unique_name(sctx
, ino
, gen
, name
);
2248 ret
= fs_path_add_path(dest
, name
);
2252 wdm
= get_waiting_dir_move(sctx
, ino
);
2253 if (wdm
&& wdm
->orphanized
) {
2254 ret
= gen_unique_name(sctx
, ino
, gen
, name
);
2257 ret
= get_first_ref(sctx
->parent_root
, ino
,
2258 &parent_inode
, &parent_gen
, name
);
2260 ret
= __get_cur_name_and_parent(sctx
, ino
, gen
,
2270 ret
= fs_path_add_path(dest
, name
);
2281 fs_path_unreverse(dest
);
2286 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2288 static int send_subvol_begin(struct send_ctx
*sctx
)
2291 struct btrfs_root
*send_root
= sctx
->send_root
;
2292 struct btrfs_root
*parent_root
= sctx
->parent_root
;
2293 struct btrfs_path
*path
;
2294 struct btrfs_key key
;
2295 struct btrfs_root_ref
*ref
;
2296 struct extent_buffer
*leaf
;
2300 path
= btrfs_alloc_path();
2304 name
= kmalloc(BTRFS_PATH_NAME_MAX
, GFP_NOFS
);
2306 btrfs_free_path(path
);
2310 key
.objectid
= send_root
->objectid
;
2311 key
.type
= BTRFS_ROOT_BACKREF_KEY
;
2314 ret
= btrfs_search_slot_for_read(send_root
->fs_info
->tree_root
,
2323 leaf
= path
->nodes
[0];
2324 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2325 if (key
.type
!= BTRFS_ROOT_BACKREF_KEY
||
2326 key
.objectid
!= send_root
->objectid
) {
2330 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
2331 namelen
= btrfs_root_ref_name_len(leaf
, ref
);
2332 read_extent_buffer(leaf
, name
, (unsigned long)(ref
+ 1), namelen
);
2333 btrfs_release_path(path
);
2336 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SNAPSHOT
);
2340 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SUBVOL
);
2345 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_PATH
, name
, namelen
);
2346 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_UUID
,
2347 sctx
->send_root
->root_item
.uuid
);
2348 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CTRANSID
,
2349 le64_to_cpu(sctx
->send_root
->root_item
.ctransid
));
2351 if (!btrfs_is_empty_uuid(parent_root
->root_item
.received_uuid
))
2352 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2353 parent_root
->root_item
.received_uuid
);
2355 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
2356 parent_root
->root_item
.uuid
);
2357 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
2358 le64_to_cpu(sctx
->parent_root
->root_item
.ctransid
));
2361 ret
= send_cmd(sctx
);
2365 btrfs_free_path(path
);
2370 static int send_truncate(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 size
)
2375 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino
, size
);
2377 p
= fs_path_alloc();
2381 ret
= begin_cmd(sctx
, BTRFS_SEND_C_TRUNCATE
);
2385 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2388 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2389 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, size
);
2391 ret
= send_cmd(sctx
);
2399 static int send_chmod(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 mode
)
2404 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino
, mode
);
2406 p
= fs_path_alloc();
2410 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHMOD
);
2414 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2417 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2418 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
& 07777);
2420 ret
= send_cmd(sctx
);
2428 static int send_chown(struct send_ctx
*sctx
, u64 ino
, u64 gen
, u64 uid
, u64 gid
)
2433 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino
, uid
, gid
);
2435 p
= fs_path_alloc();
2439 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CHOWN
);
2443 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2446 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2447 TLV_PUT_U64(sctx
, BTRFS_SEND_A_UID
, uid
);
2448 TLV_PUT_U64(sctx
, BTRFS_SEND_A_GID
, gid
);
2450 ret
= send_cmd(sctx
);
2458 static int send_utimes(struct send_ctx
*sctx
, u64 ino
, u64 gen
)
2461 struct fs_path
*p
= NULL
;
2462 struct btrfs_inode_item
*ii
;
2463 struct btrfs_path
*path
= NULL
;
2464 struct extent_buffer
*eb
;
2465 struct btrfs_key key
;
2468 verbose_printk("btrfs: send_utimes %llu\n", ino
);
2470 p
= fs_path_alloc();
2474 path
= alloc_path_for_send();
2481 key
.type
= BTRFS_INODE_ITEM_KEY
;
2483 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2487 eb
= path
->nodes
[0];
2488 slot
= path
->slots
[0];
2489 ii
= btrfs_item_ptr(eb
, slot
, struct btrfs_inode_item
);
2491 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UTIMES
);
2495 ret
= get_cur_path(sctx
, ino
, gen
, p
);
2498 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2499 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_ATIME
, eb
, &ii
->atime
);
2500 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_MTIME
, eb
, &ii
->mtime
);
2501 TLV_PUT_BTRFS_TIMESPEC(sctx
, BTRFS_SEND_A_CTIME
, eb
, &ii
->ctime
);
2502 /* TODO Add otime support when the otime patches get into upstream */
2504 ret
= send_cmd(sctx
);
2509 btrfs_free_path(path
);
2514 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2515 * a valid path yet because we did not process the refs yet. So, the inode
2516 * is created as orphan.
2518 static int send_create_inode(struct send_ctx
*sctx
, u64 ino
)
2527 verbose_printk("btrfs: send_create_inode %llu\n", ino
);
2529 p
= fs_path_alloc();
2533 if (ino
!= sctx
->cur_ino
) {
2534 ret
= get_inode_info(sctx
->send_root
, ino
, NULL
, &gen
, &mode
,
2539 gen
= sctx
->cur_inode_gen
;
2540 mode
= sctx
->cur_inode_mode
;
2541 rdev
= sctx
->cur_inode_rdev
;
2544 if (S_ISREG(mode
)) {
2545 cmd
= BTRFS_SEND_C_MKFILE
;
2546 } else if (S_ISDIR(mode
)) {
2547 cmd
= BTRFS_SEND_C_MKDIR
;
2548 } else if (S_ISLNK(mode
)) {
2549 cmd
= BTRFS_SEND_C_SYMLINK
;
2550 } else if (S_ISCHR(mode
) || S_ISBLK(mode
)) {
2551 cmd
= BTRFS_SEND_C_MKNOD
;
2552 } else if (S_ISFIFO(mode
)) {
2553 cmd
= BTRFS_SEND_C_MKFIFO
;
2554 } else if (S_ISSOCK(mode
)) {
2555 cmd
= BTRFS_SEND_C_MKSOCK
;
2557 btrfs_warn(sctx
->send_root
->fs_info
, "unexpected inode type %o",
2558 (int)(mode
& S_IFMT
));
2563 ret
= begin_cmd(sctx
, cmd
);
2567 ret
= gen_unique_name(sctx
, ino
, gen
, p
);
2571 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
2572 TLV_PUT_U64(sctx
, BTRFS_SEND_A_INO
, ino
);
2574 if (S_ISLNK(mode
)) {
2576 ret
= read_symlink(sctx
->send_root
, ino
, p
);
2579 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH_LINK
, p
);
2580 } else if (S_ISCHR(mode
) || S_ISBLK(mode
) ||
2581 S_ISFIFO(mode
) || S_ISSOCK(mode
)) {
2582 TLV_PUT_U64(sctx
, BTRFS_SEND_A_RDEV
, new_encode_dev(rdev
));
2583 TLV_PUT_U64(sctx
, BTRFS_SEND_A_MODE
, mode
);
2586 ret
= send_cmd(sctx
);
2598 * We need some special handling for inodes that get processed before the parent
2599 * directory got created. See process_recorded_refs for details.
2600 * This function does the check if we already created the dir out of order.
2602 static int did_create_dir(struct send_ctx
*sctx
, u64 dir
)
2605 struct btrfs_path
*path
= NULL
;
2606 struct btrfs_key key
;
2607 struct btrfs_key found_key
;
2608 struct btrfs_key di_key
;
2609 struct extent_buffer
*eb
;
2610 struct btrfs_dir_item
*di
;
2613 path
= alloc_path_for_send();
2620 key
.type
= BTRFS_DIR_INDEX_KEY
;
2622 ret
= btrfs_search_slot(NULL
, sctx
->send_root
, &key
, path
, 0, 0);
2627 eb
= path
->nodes
[0];
2628 slot
= path
->slots
[0];
2629 if (slot
>= btrfs_header_nritems(eb
)) {
2630 ret
= btrfs_next_leaf(sctx
->send_root
, path
);
2633 } else if (ret
> 0) {
2640 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
2641 if (found_key
.objectid
!= key
.objectid
||
2642 found_key
.type
!= key
.type
) {
2647 di
= btrfs_item_ptr(eb
, slot
, struct btrfs_dir_item
);
2648 btrfs_dir_item_key_to_cpu(eb
, di
, &di_key
);
2650 if (di_key
.type
!= BTRFS_ROOT_ITEM_KEY
&&
2651 di_key
.objectid
< sctx
->send_progress
) {
2660 btrfs_free_path(path
);
2665 * Only creates the inode if it is:
2666 * 1. Not a directory
2667 * 2. Or a directory which was not created already due to out of order
2668 * directories. See did_create_dir and process_recorded_refs for details.
2670 static int send_create_inode_if_needed(struct send_ctx
*sctx
)
2674 if (S_ISDIR(sctx
->cur_inode_mode
)) {
2675 ret
= did_create_dir(sctx
, sctx
->cur_ino
);
2684 ret
= send_create_inode(sctx
, sctx
->cur_ino
);
2692 struct recorded_ref
{
2693 struct list_head list
;
2696 struct fs_path
*full_path
;
2704 * We need to process new refs before deleted refs, but compare_tree gives us
2705 * everything mixed. So we first record all refs and later process them.
2706 * This function is a helper to record one ref.
2708 static int __record_ref(struct list_head
*head
, u64 dir
,
2709 u64 dir_gen
, struct fs_path
*path
)
2711 struct recorded_ref
*ref
;
2713 ref
= kmalloc(sizeof(*ref
), GFP_NOFS
);
2718 ref
->dir_gen
= dir_gen
;
2719 ref
->full_path
= path
;
2721 ref
->name
= (char *)kbasename(ref
->full_path
->start
);
2722 ref
->name_len
= ref
->full_path
->end
- ref
->name
;
2723 ref
->dir_path
= ref
->full_path
->start
;
2724 if (ref
->name
== ref
->full_path
->start
)
2725 ref
->dir_path_len
= 0;
2727 ref
->dir_path_len
= ref
->full_path
->end
-
2728 ref
->full_path
->start
- 1 - ref
->name_len
;
2730 list_add_tail(&ref
->list
, head
);
2734 static int dup_ref(struct recorded_ref
*ref
, struct list_head
*list
)
2736 struct recorded_ref
*new;
2738 new = kmalloc(sizeof(*ref
), GFP_NOFS
);
2742 new->dir
= ref
->dir
;
2743 new->dir_gen
= ref
->dir_gen
;
2744 new->full_path
= NULL
;
2745 INIT_LIST_HEAD(&new->list
);
2746 list_add_tail(&new->list
, list
);
2750 static void __free_recorded_refs(struct list_head
*head
)
2752 struct recorded_ref
*cur
;
2754 while (!list_empty(head
)) {
2755 cur
= list_entry(head
->next
, struct recorded_ref
, list
);
2756 fs_path_free(cur
->full_path
);
2757 list_del(&cur
->list
);
2762 static void free_recorded_refs(struct send_ctx
*sctx
)
2764 __free_recorded_refs(&sctx
->new_refs
);
2765 __free_recorded_refs(&sctx
->deleted_refs
);
2769 * Renames/moves a file/dir to its orphan name. Used when the first
2770 * ref of an unprocessed inode gets overwritten and for all non empty
2773 static int orphanize_inode(struct send_ctx
*sctx
, u64 ino
, u64 gen
,
2774 struct fs_path
*path
)
2777 struct fs_path
*orphan
;
2779 orphan
= fs_path_alloc();
2783 ret
= gen_unique_name(sctx
, ino
, gen
, orphan
);
2787 ret
= send_rename(sctx
, path
, orphan
);
2790 fs_path_free(orphan
);
2794 static struct orphan_dir_info
*
2795 add_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2797 struct rb_node
**p
= &sctx
->orphan_dirs
.rb_node
;
2798 struct rb_node
*parent
= NULL
;
2799 struct orphan_dir_info
*entry
, *odi
;
2801 odi
= kmalloc(sizeof(*odi
), GFP_NOFS
);
2803 return ERR_PTR(-ENOMEM
);
2809 entry
= rb_entry(parent
, struct orphan_dir_info
, node
);
2810 if (dir_ino
< entry
->ino
) {
2812 } else if (dir_ino
> entry
->ino
) {
2813 p
= &(*p
)->rb_right
;
2820 rb_link_node(&odi
->node
, parent
, p
);
2821 rb_insert_color(&odi
->node
, &sctx
->orphan_dirs
);
2825 static struct orphan_dir_info
*
2826 get_orphan_dir_info(struct send_ctx
*sctx
, u64 dir_ino
)
2828 struct rb_node
*n
= sctx
->orphan_dirs
.rb_node
;
2829 struct orphan_dir_info
*entry
;
2832 entry
= rb_entry(n
, struct orphan_dir_info
, node
);
2833 if (dir_ino
< entry
->ino
)
2835 else if (dir_ino
> entry
->ino
)
2843 static int is_waiting_for_rm(struct send_ctx
*sctx
, u64 dir_ino
)
2845 struct orphan_dir_info
*odi
= get_orphan_dir_info(sctx
, dir_ino
);
2850 static void free_orphan_dir_info(struct send_ctx
*sctx
,
2851 struct orphan_dir_info
*odi
)
2855 rb_erase(&odi
->node
, &sctx
->orphan_dirs
);
2860 * Returns 1 if a directory can be removed at this point in time.
2861 * We check this by iterating all dir items and checking if the inode behind
2862 * the dir item was already processed.
2864 static int can_rmdir(struct send_ctx
*sctx
, u64 dir
, u64 dir_gen
,
2868 struct btrfs_root
*root
= sctx
->parent_root
;
2869 struct btrfs_path
*path
;
2870 struct btrfs_key key
;
2871 struct btrfs_key found_key
;
2872 struct btrfs_key loc
;
2873 struct btrfs_dir_item
*di
;
2876 * Don't try to rmdir the top/root subvolume dir.
2878 if (dir
== BTRFS_FIRST_FREE_OBJECTID
)
2881 path
= alloc_path_for_send();
2886 key
.type
= BTRFS_DIR_INDEX_KEY
;
2888 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2893 struct waiting_dir_move
*dm
;
2895 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
2896 ret
= btrfs_next_leaf(root
, path
);
2903 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2905 if (found_key
.objectid
!= key
.objectid
||
2906 found_key
.type
!= key
.type
)
2909 di
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2910 struct btrfs_dir_item
);
2911 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &loc
);
2913 dm
= get_waiting_dir_move(sctx
, loc
.objectid
);
2915 struct orphan_dir_info
*odi
;
2917 odi
= add_orphan_dir_info(sctx
, dir
);
2923 dm
->rmdir_ino
= dir
;
2928 if (loc
.objectid
> send_progress
) {
2939 btrfs_free_path(path
);
2943 static int is_waiting_for_move(struct send_ctx
*sctx
, u64 ino
)
2945 struct waiting_dir_move
*entry
= get_waiting_dir_move(sctx
, ino
);
2947 return entry
!= NULL
;
2950 static int add_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
, bool orphanized
)
2952 struct rb_node
**p
= &sctx
->waiting_dir_moves
.rb_node
;
2953 struct rb_node
*parent
= NULL
;
2954 struct waiting_dir_move
*entry
, *dm
;
2956 dm
= kmalloc(sizeof(*dm
), GFP_NOFS
);
2961 dm
->orphanized
= orphanized
;
2965 entry
= rb_entry(parent
, struct waiting_dir_move
, node
);
2966 if (ino
< entry
->ino
) {
2968 } else if (ino
> entry
->ino
) {
2969 p
= &(*p
)->rb_right
;
2976 rb_link_node(&dm
->node
, parent
, p
);
2977 rb_insert_color(&dm
->node
, &sctx
->waiting_dir_moves
);
2981 static struct waiting_dir_move
*
2982 get_waiting_dir_move(struct send_ctx
*sctx
, u64 ino
)
2984 struct rb_node
*n
= sctx
->waiting_dir_moves
.rb_node
;
2985 struct waiting_dir_move
*entry
;
2988 entry
= rb_entry(n
, struct waiting_dir_move
, node
);
2989 if (ino
< entry
->ino
)
2991 else if (ino
> entry
->ino
)
2999 static void free_waiting_dir_move(struct send_ctx
*sctx
,
3000 struct waiting_dir_move
*dm
)
3004 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
3008 static int add_pending_dir_move(struct send_ctx
*sctx
,
3012 struct list_head
*new_refs
,
3013 struct list_head
*deleted_refs
,
3014 const bool is_orphan
)
3016 struct rb_node
**p
= &sctx
->pending_dir_moves
.rb_node
;
3017 struct rb_node
*parent
= NULL
;
3018 struct pending_dir_move
*entry
= NULL
, *pm
;
3019 struct recorded_ref
*cur
;
3023 pm
= kmalloc(sizeof(*pm
), GFP_NOFS
);
3026 pm
->parent_ino
= parent_ino
;
3029 pm
->is_orphan
= is_orphan
;
3030 INIT_LIST_HEAD(&pm
->list
);
3031 INIT_LIST_HEAD(&pm
->update_refs
);
3032 RB_CLEAR_NODE(&pm
->node
);
3036 entry
= rb_entry(parent
, struct pending_dir_move
, node
);
3037 if (parent_ino
< entry
->parent_ino
) {
3039 } else if (parent_ino
> entry
->parent_ino
) {
3040 p
= &(*p
)->rb_right
;
3047 list_for_each_entry(cur
, deleted_refs
, list
) {
3048 ret
= dup_ref(cur
, &pm
->update_refs
);
3052 list_for_each_entry(cur
, new_refs
, list
) {
3053 ret
= dup_ref(cur
, &pm
->update_refs
);
3058 ret
= add_waiting_dir_move(sctx
, pm
->ino
, is_orphan
);
3063 list_add_tail(&pm
->list
, &entry
->list
);
3065 rb_link_node(&pm
->node
, parent
, p
);
3066 rb_insert_color(&pm
->node
, &sctx
->pending_dir_moves
);
3071 __free_recorded_refs(&pm
->update_refs
);
3077 static struct pending_dir_move
*get_pending_dir_moves(struct send_ctx
*sctx
,
3080 struct rb_node
*n
= sctx
->pending_dir_moves
.rb_node
;
3081 struct pending_dir_move
*entry
;
3084 entry
= rb_entry(n
, struct pending_dir_move
, node
);
3085 if (parent_ino
< entry
->parent_ino
)
3087 else if (parent_ino
> entry
->parent_ino
)
3095 static int apply_dir_move(struct send_ctx
*sctx
, struct pending_dir_move
*pm
)
3097 struct fs_path
*from_path
= NULL
;
3098 struct fs_path
*to_path
= NULL
;
3099 struct fs_path
*name
= NULL
;
3100 u64 orig_progress
= sctx
->send_progress
;
3101 struct recorded_ref
*cur
;
3102 u64 parent_ino
, parent_gen
;
3103 struct waiting_dir_move
*dm
= NULL
;
3107 name
= fs_path_alloc();
3108 from_path
= fs_path_alloc();
3109 if (!name
|| !from_path
) {
3114 dm
= get_waiting_dir_move(sctx
, pm
->ino
);
3116 rmdir_ino
= dm
->rmdir_ino
;
3117 free_waiting_dir_move(sctx
, dm
);
3119 if (pm
->is_orphan
) {
3120 ret
= gen_unique_name(sctx
, pm
->ino
,
3121 pm
->gen
, from_path
);
3123 ret
= get_first_ref(sctx
->parent_root
, pm
->ino
,
3124 &parent_ino
, &parent_gen
, name
);
3127 ret
= get_cur_path(sctx
, parent_ino
, parent_gen
,
3131 ret
= fs_path_add_path(from_path
, name
);
3136 sctx
->send_progress
= sctx
->cur_ino
+ 1;
3137 fs_path_reset(name
);
3140 ret
= get_cur_path(sctx
, pm
->ino
, pm
->gen
, to_path
);
3144 ret
= send_rename(sctx
, from_path
, to_path
);
3149 struct orphan_dir_info
*odi
;
3151 odi
= get_orphan_dir_info(sctx
, rmdir_ino
);
3153 /* already deleted */
3156 ret
= can_rmdir(sctx
, rmdir_ino
, odi
->gen
, sctx
->cur_ino
+ 1);
3162 name
= fs_path_alloc();
3167 ret
= get_cur_path(sctx
, rmdir_ino
, odi
->gen
, name
);
3170 ret
= send_rmdir(sctx
, name
);
3173 free_orphan_dir_info(sctx
, odi
);
3177 ret
= send_utimes(sctx
, pm
->ino
, pm
->gen
);
3182 * After rename/move, need to update the utimes of both new parent(s)
3183 * and old parent(s).
3185 list_for_each_entry(cur
, &pm
->update_refs
, list
) {
3186 if (cur
->dir
== rmdir_ino
)
3188 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
3195 fs_path_free(from_path
);
3196 fs_path_free(to_path
);
3197 sctx
->send_progress
= orig_progress
;
3202 static void free_pending_move(struct send_ctx
*sctx
, struct pending_dir_move
*m
)
3204 if (!list_empty(&m
->list
))
3206 if (!RB_EMPTY_NODE(&m
->node
))
3207 rb_erase(&m
->node
, &sctx
->pending_dir_moves
);
3208 __free_recorded_refs(&m
->update_refs
);
3212 static void tail_append_pending_moves(struct pending_dir_move
*moves
,
3213 struct list_head
*stack
)
3215 if (list_empty(&moves
->list
)) {
3216 list_add_tail(&moves
->list
, stack
);
3219 list_splice_init(&moves
->list
, &list
);
3220 list_add_tail(&moves
->list
, stack
);
3221 list_splice_tail(&list
, stack
);
3225 static int apply_children_dir_moves(struct send_ctx
*sctx
)
3227 struct pending_dir_move
*pm
;
3228 struct list_head stack
;
3229 u64 parent_ino
= sctx
->cur_ino
;
3232 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3236 INIT_LIST_HEAD(&stack
);
3237 tail_append_pending_moves(pm
, &stack
);
3239 while (!list_empty(&stack
)) {
3240 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3241 parent_ino
= pm
->ino
;
3242 ret
= apply_dir_move(sctx
, pm
);
3243 free_pending_move(sctx
, pm
);
3246 pm
= get_pending_dir_moves(sctx
, parent_ino
);
3248 tail_append_pending_moves(pm
, &stack
);
3253 while (!list_empty(&stack
)) {
3254 pm
= list_first_entry(&stack
, struct pending_dir_move
, list
);
3255 free_pending_move(sctx
, pm
);
3261 * We might need to delay a directory rename even when no ancestor directory
3262 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3263 * renamed. This happens when we rename a directory to the old name (the name
3264 * in the parent root) of some other unrelated directory that got its rename
3265 * delayed due to some ancestor with higher number that got renamed.
3271 * |---- a/ (ino 257)
3272 * | |---- file (ino 260)
3274 * |---- b/ (ino 258)
3275 * |---- c/ (ino 259)
3279 * |---- a/ (ino 258)
3280 * |---- x/ (ino 259)
3281 * |---- y/ (ino 257)
3282 * |----- file (ino 260)
3284 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3285 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3286 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3289 * 1 - rename 259 from 'c' to 'x'
3290 * 2 - rename 257 from 'a' to 'x/y'
3291 * 3 - rename 258 from 'b' to 'a'
3293 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3294 * be done right away and < 0 on error.
3296 static int wait_for_dest_dir_move(struct send_ctx
*sctx
,
3297 struct recorded_ref
*parent_ref
,
3298 const bool is_orphan
)
3300 struct btrfs_path
*path
;
3301 struct btrfs_key key
;
3302 struct btrfs_key di_key
;
3303 struct btrfs_dir_item
*di
;
3308 if (RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
))
3311 path
= alloc_path_for_send();
3315 key
.objectid
= parent_ref
->dir
;
3316 key
.type
= BTRFS_DIR_ITEM_KEY
;
3317 key
.offset
= btrfs_name_hash(parent_ref
->name
, parent_ref
->name_len
);
3319 ret
= btrfs_search_slot(NULL
, sctx
->parent_root
, &key
, path
, 0, 0);
3322 } else if (ret
> 0) {
3327 di
= btrfs_match_dir_item_name(sctx
->parent_root
, path
,
3328 parent_ref
->name
, parent_ref
->name_len
);
3334 * di_key.objectid has the number of the inode that has a dentry in the
3335 * parent directory with the same name that sctx->cur_ino is being
3336 * renamed to. We need to check if that inode is in the send root as
3337 * well and if it is currently marked as an inode with a pending rename,
3338 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3339 * that it happens after that other inode is renamed.
3341 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &di_key
);
3342 if (di_key
.type
!= BTRFS_INODE_ITEM_KEY
) {
3347 ret
= get_inode_info(sctx
->parent_root
, di_key
.objectid
, NULL
,
3348 &left_gen
, NULL
, NULL
, NULL
, NULL
);
3351 ret
= get_inode_info(sctx
->send_root
, di_key
.objectid
, NULL
,
3352 &right_gen
, NULL
, NULL
, NULL
, NULL
);
3359 /* Different inode, no need to delay the rename of sctx->cur_ino */
3360 if (right_gen
!= left_gen
) {
3365 if (is_waiting_for_move(sctx
, di_key
.objectid
)) {
3366 ret
= add_pending_dir_move(sctx
,
3368 sctx
->cur_inode_gen
,
3371 &sctx
->deleted_refs
,
3377 btrfs_free_path(path
);
3382 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3383 * Return 1 if true, 0 if false and < 0 on error.
3385 static int is_ancestor(struct btrfs_root
*root
,
3389 struct fs_path
*fs_path
)
3393 while (ino
> BTRFS_FIRST_FREE_OBJECTID
) {
3398 fs_path_reset(fs_path
);
3399 ret
= get_first_ref(root
, ino
, &parent
, &parent_gen
, fs_path
);
3401 if (ret
== -ENOENT
&& ino
== ino2
)
3406 return parent_gen
== ino1_gen
? 1 : 0;
3412 static int wait_for_parent_move(struct send_ctx
*sctx
,
3413 struct recorded_ref
*parent_ref
,
3414 const bool is_orphan
)
3417 u64 ino
= parent_ref
->dir
;
3418 u64 parent_ino_before
, parent_ino_after
;
3419 struct fs_path
*path_before
= NULL
;
3420 struct fs_path
*path_after
= NULL
;
3423 path_after
= fs_path_alloc();
3424 path_before
= fs_path_alloc();
3425 if (!path_after
|| !path_before
) {
3431 * Our current directory inode may not yet be renamed/moved because some
3432 * ancestor (immediate or not) has to be renamed/moved first. So find if
3433 * such ancestor exists and make sure our own rename/move happens after
3434 * that ancestor is processed to avoid path build infinite loops (done
3435 * at get_cur_path()).
3437 while (ino
> BTRFS_FIRST_FREE_OBJECTID
) {
3438 if (is_waiting_for_move(sctx
, ino
)) {
3440 * If the current inode is an ancestor of ino in the
3441 * parent root, we need to delay the rename of the
3442 * current inode, otherwise don't delayed the rename
3443 * because we can end up with a circular dependency
3444 * of renames, resulting in some directories never
3445 * getting the respective rename operations issued in
3446 * the send stream or getting into infinite path build
3449 ret
= is_ancestor(sctx
->parent_root
,
3450 sctx
->cur_ino
, sctx
->cur_inode_gen
,
3455 fs_path_reset(path_before
);
3456 fs_path_reset(path_after
);
3458 ret
= get_first_ref(sctx
->send_root
, ino
, &parent_ino_after
,
3462 ret
= get_first_ref(sctx
->parent_root
, ino
, &parent_ino_before
,
3464 if (ret
< 0 && ret
!= -ENOENT
) {
3466 } else if (ret
== -ENOENT
) {
3471 len1
= fs_path_len(path_before
);
3472 len2
= fs_path_len(path_after
);
3473 if (ino
> sctx
->cur_ino
&&
3474 (parent_ino_before
!= parent_ino_after
|| len1
!= len2
||
3475 memcmp(path_before
->start
, path_after
->start
, len1
))) {
3479 ino
= parent_ino_after
;
3483 fs_path_free(path_before
);
3484 fs_path_free(path_after
);
3487 ret
= add_pending_dir_move(sctx
,
3489 sctx
->cur_inode_gen
,
3492 &sctx
->deleted_refs
,
3502 * This does all the move/link/unlink/rmdir magic.
3504 static int process_recorded_refs(struct send_ctx
*sctx
, int *pending_move
)
3507 struct recorded_ref
*cur
;
3508 struct recorded_ref
*cur2
;
3509 struct list_head check_dirs
;
3510 struct fs_path
*valid_path
= NULL
;
3513 int did_overwrite
= 0;
3515 u64 last_dir_ino_rm
= 0;
3516 bool can_rename
= true;
3518 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx
->cur_ino
);
3521 * This should never happen as the root dir always has the same ref
3522 * which is always '..'
3524 BUG_ON(sctx
->cur_ino
<= BTRFS_FIRST_FREE_OBJECTID
);
3525 INIT_LIST_HEAD(&check_dirs
);
3527 valid_path
= fs_path_alloc();
3534 * First, check if the first ref of the current inode was overwritten
3535 * before. If yes, we know that the current inode was already orphanized
3536 * and thus use the orphan name. If not, we can use get_cur_path to
3537 * get the path of the first ref as it would like while receiving at
3538 * this point in time.
3539 * New inodes are always orphan at the beginning, so force to use the
3540 * orphan name in this case.
3541 * The first ref is stored in valid_path and will be updated if it
3542 * gets moved around.
3544 if (!sctx
->cur_inode_new
) {
3545 ret
= did_overwrite_first_ref(sctx
, sctx
->cur_ino
,
3546 sctx
->cur_inode_gen
);
3552 if (sctx
->cur_inode_new
|| did_overwrite
) {
3553 ret
= gen_unique_name(sctx
, sctx
->cur_ino
,
3554 sctx
->cur_inode_gen
, valid_path
);
3559 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3565 list_for_each_entry(cur
, &sctx
->new_refs
, list
) {
3567 * We may have refs where the parent directory does not exist
3568 * yet. This happens if the parent directories inum is higher
3569 * the the current inum. To handle this case, we create the
3570 * parent directory out of order. But we need to check if this
3571 * did already happen before due to other refs in the same dir.
3573 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3576 if (ret
== inode_state_will_create
) {
3579 * First check if any of the current inodes refs did
3580 * already create the dir.
3582 list_for_each_entry(cur2
, &sctx
->new_refs
, list
) {
3585 if (cur2
->dir
== cur
->dir
) {
3592 * If that did not happen, check if a previous inode
3593 * did already create the dir.
3596 ret
= did_create_dir(sctx
, cur
->dir
);
3600 ret
= send_create_inode(sctx
, cur
->dir
);
3607 * Check if this new ref would overwrite the first ref of
3608 * another unprocessed inode. If yes, orphanize the
3609 * overwritten inode. If we find an overwritten ref that is
3610 * not the first ref, simply unlink it.
3612 ret
= will_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3613 cur
->name
, cur
->name_len
,
3614 &ow_inode
, &ow_gen
);
3618 ret
= is_first_ref(sctx
->parent_root
,
3619 ow_inode
, cur
->dir
, cur
->name
,
3624 struct name_cache_entry
*nce
;
3626 ret
= orphanize_inode(sctx
, ow_inode
, ow_gen
,
3631 * Make sure we clear our orphanized inode's
3632 * name from the name cache. This is because the
3633 * inode ow_inode might be an ancestor of some
3634 * other inode that will be orphanized as well
3635 * later and has an inode number greater than
3636 * sctx->send_progress. We need to prevent
3637 * future name lookups from using the old name
3638 * and get instead the orphan name.
3640 nce
= name_cache_search(sctx
, ow_inode
, ow_gen
);
3642 name_cache_delete(sctx
, nce
);
3646 ret
= send_unlink(sctx
, cur
->full_path
);
3652 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->parent_root
) {
3653 ret
= wait_for_dest_dir_move(sctx
, cur
, is_orphan
);
3662 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->parent_root
&&
3664 ret
= wait_for_parent_move(sctx
, cur
, is_orphan
);
3674 * link/move the ref to the new place. If we have an orphan
3675 * inode, move it and update valid_path. If not, link or move
3676 * it depending on the inode mode.
3678 if (is_orphan
&& can_rename
) {
3679 ret
= send_rename(sctx
, valid_path
, cur
->full_path
);
3683 ret
= fs_path_copy(valid_path
, cur
->full_path
);
3686 } else if (can_rename
) {
3687 if (S_ISDIR(sctx
->cur_inode_mode
)) {
3689 * Dirs can't be linked, so move it. For moved
3690 * dirs, we always have one new and one deleted
3691 * ref. The deleted ref is ignored later.
3693 ret
= send_rename(sctx
, valid_path
,
3696 ret
= fs_path_copy(valid_path
,
3701 ret
= send_link(sctx
, cur
->full_path
,
3707 ret
= dup_ref(cur
, &check_dirs
);
3712 if (S_ISDIR(sctx
->cur_inode_mode
) && sctx
->cur_inode_deleted
) {
3714 * Check if we can already rmdir the directory. If not,
3715 * orphanize it. For every dir item inside that gets deleted
3716 * later, we do this check again and rmdir it then if possible.
3717 * See the use of check_dirs for more details.
3719 ret
= can_rmdir(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
3724 ret
= send_rmdir(sctx
, valid_path
);
3727 } else if (!is_orphan
) {
3728 ret
= orphanize_inode(sctx
, sctx
->cur_ino
,
3729 sctx
->cur_inode_gen
, valid_path
);
3735 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
3736 ret
= dup_ref(cur
, &check_dirs
);
3740 } else if (S_ISDIR(sctx
->cur_inode_mode
) &&
3741 !list_empty(&sctx
->deleted_refs
)) {
3743 * We have a moved dir. Add the old parent to check_dirs
3745 cur
= list_entry(sctx
->deleted_refs
.next
, struct recorded_ref
,
3747 ret
= dup_ref(cur
, &check_dirs
);
3750 } else if (!S_ISDIR(sctx
->cur_inode_mode
)) {
3752 * We have a non dir inode. Go through all deleted refs and
3753 * unlink them if they were not already overwritten by other
3756 list_for_each_entry(cur
, &sctx
->deleted_refs
, list
) {
3757 ret
= did_overwrite_ref(sctx
, cur
->dir
, cur
->dir_gen
,
3758 sctx
->cur_ino
, sctx
->cur_inode_gen
,
3759 cur
->name
, cur
->name_len
);
3763 ret
= send_unlink(sctx
, cur
->full_path
);
3767 ret
= dup_ref(cur
, &check_dirs
);
3772 * If the inode is still orphan, unlink the orphan. This may
3773 * happen when a previous inode did overwrite the first ref
3774 * of this inode and no new refs were added for the current
3775 * inode. Unlinking does not mean that the inode is deleted in
3776 * all cases. There may still be links to this inode in other
3780 ret
= send_unlink(sctx
, valid_path
);
3787 * We did collect all parent dirs where cur_inode was once located. We
3788 * now go through all these dirs and check if they are pending for
3789 * deletion and if it's finally possible to perform the rmdir now.
3790 * We also update the inode stats of the parent dirs here.
3792 list_for_each_entry(cur
, &check_dirs
, list
) {
3794 * In case we had refs into dirs that were not processed yet,
3795 * we don't need to do the utime and rmdir logic for these dirs.
3796 * The dir will be processed later.
3798 if (cur
->dir
> sctx
->cur_ino
)
3801 ret
= get_cur_inode_state(sctx
, cur
->dir
, cur
->dir_gen
);
3805 if (ret
== inode_state_did_create
||
3806 ret
== inode_state_no_change
) {
3807 /* TODO delayed utimes */
3808 ret
= send_utimes(sctx
, cur
->dir
, cur
->dir_gen
);
3811 } else if (ret
== inode_state_did_delete
&&
3812 cur
->dir
!= last_dir_ino_rm
) {
3813 ret
= can_rmdir(sctx
, cur
->dir
, cur
->dir_gen
,
3818 ret
= get_cur_path(sctx
, cur
->dir
,
3819 cur
->dir_gen
, valid_path
);
3822 ret
= send_rmdir(sctx
, valid_path
);
3825 last_dir_ino_rm
= cur
->dir
;
3833 __free_recorded_refs(&check_dirs
);
3834 free_recorded_refs(sctx
);
3835 fs_path_free(valid_path
);
3839 static int record_ref(struct btrfs_root
*root
, int num
, u64 dir
, int index
,
3840 struct fs_path
*name
, void *ctx
, struct list_head
*refs
)
3843 struct send_ctx
*sctx
= ctx
;
3847 p
= fs_path_alloc();
3851 ret
= get_inode_info(root
, dir
, NULL
, &gen
, NULL
, NULL
,
3856 ret
= get_cur_path(sctx
, dir
, gen
, p
);
3859 ret
= fs_path_add_path(p
, name
);
3863 ret
= __record_ref(refs
, dir
, gen
, p
);
3871 static int __record_new_ref(int num
, u64 dir
, int index
,
3872 struct fs_path
*name
,
3875 struct send_ctx
*sctx
= ctx
;
3876 return record_ref(sctx
->send_root
, num
, dir
, index
, name
,
3877 ctx
, &sctx
->new_refs
);
3881 static int __record_deleted_ref(int num
, u64 dir
, int index
,
3882 struct fs_path
*name
,
3885 struct send_ctx
*sctx
= ctx
;
3886 return record_ref(sctx
->parent_root
, num
, dir
, index
, name
,
3887 ctx
, &sctx
->deleted_refs
);
3890 static int record_new_ref(struct send_ctx
*sctx
)
3894 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
3895 sctx
->cmp_key
, 0, __record_new_ref
, sctx
);
3904 static int record_deleted_ref(struct send_ctx
*sctx
)
3908 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
3909 sctx
->cmp_key
, 0, __record_deleted_ref
, sctx
);
3918 struct find_ref_ctx
{
3921 struct btrfs_root
*root
;
3922 struct fs_path
*name
;
3926 static int __find_iref(int num
, u64 dir
, int index
,
3927 struct fs_path
*name
,
3930 struct find_ref_ctx
*ctx
= ctx_
;
3934 if (dir
== ctx
->dir
&& fs_path_len(name
) == fs_path_len(ctx
->name
) &&
3935 strncmp(name
->start
, ctx
->name
->start
, fs_path_len(name
)) == 0) {
3937 * To avoid doing extra lookups we'll only do this if everything
3940 ret
= get_inode_info(ctx
->root
, dir
, NULL
, &dir_gen
, NULL
,
3944 if (dir_gen
!= ctx
->dir_gen
)
3946 ctx
->found_idx
= num
;
3952 static int find_iref(struct btrfs_root
*root
,
3953 struct btrfs_path
*path
,
3954 struct btrfs_key
*key
,
3955 u64 dir
, u64 dir_gen
, struct fs_path
*name
)
3958 struct find_ref_ctx ctx
;
3962 ctx
.dir_gen
= dir_gen
;
3966 ret
= iterate_inode_ref(root
, path
, key
, 0, __find_iref
, &ctx
);
3970 if (ctx
.found_idx
== -1)
3973 return ctx
.found_idx
;
3976 static int __record_changed_new_ref(int num
, u64 dir
, int index
,
3977 struct fs_path
*name
,
3982 struct send_ctx
*sctx
= ctx
;
3984 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &dir_gen
, NULL
,
3989 ret
= find_iref(sctx
->parent_root
, sctx
->right_path
,
3990 sctx
->cmp_key
, dir
, dir_gen
, name
);
3992 ret
= __record_new_ref(num
, dir
, index
, name
, sctx
);
3999 static int __record_changed_deleted_ref(int num
, u64 dir
, int index
,
4000 struct fs_path
*name
,
4005 struct send_ctx
*sctx
= ctx
;
4007 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &dir_gen
, NULL
,
4012 ret
= find_iref(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
4013 dir
, dir_gen
, name
);
4015 ret
= __record_deleted_ref(num
, dir
, index
, name
, sctx
);
4022 static int record_changed_ref(struct send_ctx
*sctx
)
4026 ret
= iterate_inode_ref(sctx
->send_root
, sctx
->left_path
,
4027 sctx
->cmp_key
, 0, __record_changed_new_ref
, sctx
);
4030 ret
= iterate_inode_ref(sctx
->parent_root
, sctx
->right_path
,
4031 sctx
->cmp_key
, 0, __record_changed_deleted_ref
, sctx
);
4041 * Record and process all refs at once. Needed when an inode changes the
4042 * generation number, which means that it was deleted and recreated.
4044 static int process_all_refs(struct send_ctx
*sctx
,
4045 enum btrfs_compare_tree_result cmd
)
4048 struct btrfs_root
*root
;
4049 struct btrfs_path
*path
;
4050 struct btrfs_key key
;
4051 struct btrfs_key found_key
;
4052 struct extent_buffer
*eb
;
4054 iterate_inode_ref_t cb
;
4055 int pending_move
= 0;
4057 path
= alloc_path_for_send();
4061 if (cmd
== BTRFS_COMPARE_TREE_NEW
) {
4062 root
= sctx
->send_root
;
4063 cb
= __record_new_ref
;
4064 } else if (cmd
== BTRFS_COMPARE_TREE_DELETED
) {
4065 root
= sctx
->parent_root
;
4066 cb
= __record_deleted_ref
;
4068 btrfs_err(sctx
->send_root
->fs_info
,
4069 "Wrong command %d in process_all_refs", cmd
);
4074 key
.objectid
= sctx
->cmp_key
->objectid
;
4075 key
.type
= BTRFS_INODE_REF_KEY
;
4077 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4082 eb
= path
->nodes
[0];
4083 slot
= path
->slots
[0];
4084 if (slot
>= btrfs_header_nritems(eb
)) {
4085 ret
= btrfs_next_leaf(root
, path
);
4093 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4095 if (found_key
.objectid
!= key
.objectid
||
4096 (found_key
.type
!= BTRFS_INODE_REF_KEY
&&
4097 found_key
.type
!= BTRFS_INODE_EXTREF_KEY
))
4100 ret
= iterate_inode_ref(root
, path
, &found_key
, 0, cb
, sctx
);
4106 btrfs_release_path(path
);
4108 ret
= process_recorded_refs(sctx
, &pending_move
);
4109 /* Only applicable to an incremental send. */
4110 ASSERT(pending_move
== 0);
4113 btrfs_free_path(path
);
4117 static int send_set_xattr(struct send_ctx
*sctx
,
4118 struct fs_path
*path
,
4119 const char *name
, int name_len
,
4120 const char *data
, int data_len
)
4124 ret
= begin_cmd(sctx
, BTRFS_SEND_C_SET_XATTR
);
4128 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
4129 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
4130 TLV_PUT(sctx
, BTRFS_SEND_A_XATTR_DATA
, data
, data_len
);
4132 ret
= send_cmd(sctx
);
4139 static int send_remove_xattr(struct send_ctx
*sctx
,
4140 struct fs_path
*path
,
4141 const char *name
, int name_len
)
4145 ret
= begin_cmd(sctx
, BTRFS_SEND_C_REMOVE_XATTR
);
4149 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, path
);
4150 TLV_PUT_STRING(sctx
, BTRFS_SEND_A_XATTR_NAME
, name
, name_len
);
4152 ret
= send_cmd(sctx
);
4159 static int __process_new_xattr(int num
, struct btrfs_key
*di_key
,
4160 const char *name
, int name_len
,
4161 const char *data
, int data_len
,
4165 struct send_ctx
*sctx
= ctx
;
4167 posix_acl_xattr_header dummy_acl
;
4169 p
= fs_path_alloc();
4174 * This hack is needed because empty acl's are stored as zero byte
4175 * data in xattrs. Problem with that is, that receiving these zero byte
4176 * acl's will fail later. To fix this, we send a dummy acl list that
4177 * only contains the version number and no entries.
4179 if (!strncmp(name
, XATTR_NAME_POSIX_ACL_ACCESS
, name_len
) ||
4180 !strncmp(name
, XATTR_NAME_POSIX_ACL_DEFAULT
, name_len
)) {
4181 if (data_len
== 0) {
4182 dummy_acl
.a_version
=
4183 cpu_to_le32(POSIX_ACL_XATTR_VERSION
);
4184 data
= (char *)&dummy_acl
;
4185 data_len
= sizeof(dummy_acl
);
4189 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4193 ret
= send_set_xattr(sctx
, p
, name
, name_len
, data
, data_len
);
4200 static int __process_deleted_xattr(int num
, struct btrfs_key
*di_key
,
4201 const char *name
, int name_len
,
4202 const char *data
, int data_len
,
4206 struct send_ctx
*sctx
= ctx
;
4209 p
= fs_path_alloc();
4213 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4217 ret
= send_remove_xattr(sctx
, p
, name
, name_len
);
4224 static int process_new_xattr(struct send_ctx
*sctx
)
4228 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4229 sctx
->cmp_key
, __process_new_xattr
, sctx
);
4234 static int process_deleted_xattr(struct send_ctx
*sctx
)
4238 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4239 sctx
->cmp_key
, __process_deleted_xattr
, sctx
);
4244 struct find_xattr_ctx
{
4252 static int __find_xattr(int num
, struct btrfs_key
*di_key
,
4253 const char *name
, int name_len
,
4254 const char *data
, int data_len
,
4255 u8 type
, void *vctx
)
4257 struct find_xattr_ctx
*ctx
= vctx
;
4259 if (name_len
== ctx
->name_len
&&
4260 strncmp(name
, ctx
->name
, name_len
) == 0) {
4261 ctx
->found_idx
= num
;
4262 ctx
->found_data_len
= data_len
;
4263 ctx
->found_data
= kmemdup(data
, data_len
, GFP_NOFS
);
4264 if (!ctx
->found_data
)
4271 static int find_xattr(struct btrfs_root
*root
,
4272 struct btrfs_path
*path
,
4273 struct btrfs_key
*key
,
4274 const char *name
, int name_len
,
4275 char **data
, int *data_len
)
4278 struct find_xattr_ctx ctx
;
4281 ctx
.name_len
= name_len
;
4283 ctx
.found_data
= NULL
;
4284 ctx
.found_data_len
= 0;
4286 ret
= iterate_dir_item(root
, path
, key
, __find_xattr
, &ctx
);
4290 if (ctx
.found_idx
== -1)
4293 *data
= ctx
.found_data
;
4294 *data_len
= ctx
.found_data_len
;
4296 kfree(ctx
.found_data
);
4298 return ctx
.found_idx
;
4302 static int __process_changed_new_xattr(int num
, struct btrfs_key
*di_key
,
4303 const char *name
, int name_len
,
4304 const char *data
, int data_len
,
4308 struct send_ctx
*sctx
= ctx
;
4309 char *found_data
= NULL
;
4310 int found_data_len
= 0;
4312 ret
= find_xattr(sctx
->parent_root
, sctx
->right_path
,
4313 sctx
->cmp_key
, name
, name_len
, &found_data
,
4315 if (ret
== -ENOENT
) {
4316 ret
= __process_new_xattr(num
, di_key
, name
, name_len
, data
,
4317 data_len
, type
, ctx
);
4318 } else if (ret
>= 0) {
4319 if (data_len
!= found_data_len
||
4320 memcmp(data
, found_data
, data_len
)) {
4321 ret
= __process_new_xattr(num
, di_key
, name
, name_len
,
4322 data
, data_len
, type
, ctx
);
4332 static int __process_changed_deleted_xattr(int num
, struct btrfs_key
*di_key
,
4333 const char *name
, int name_len
,
4334 const char *data
, int data_len
,
4338 struct send_ctx
*sctx
= ctx
;
4340 ret
= find_xattr(sctx
->send_root
, sctx
->left_path
, sctx
->cmp_key
,
4341 name
, name_len
, NULL
, NULL
);
4343 ret
= __process_deleted_xattr(num
, di_key
, name
, name_len
, data
,
4344 data_len
, type
, ctx
);
4351 static int process_changed_xattr(struct send_ctx
*sctx
)
4355 ret
= iterate_dir_item(sctx
->send_root
, sctx
->left_path
,
4356 sctx
->cmp_key
, __process_changed_new_xattr
, sctx
);
4359 ret
= iterate_dir_item(sctx
->parent_root
, sctx
->right_path
,
4360 sctx
->cmp_key
, __process_changed_deleted_xattr
, sctx
);
4366 static int process_all_new_xattrs(struct send_ctx
*sctx
)
4369 struct btrfs_root
*root
;
4370 struct btrfs_path
*path
;
4371 struct btrfs_key key
;
4372 struct btrfs_key found_key
;
4373 struct extent_buffer
*eb
;
4376 path
= alloc_path_for_send();
4380 root
= sctx
->send_root
;
4382 key
.objectid
= sctx
->cmp_key
->objectid
;
4383 key
.type
= BTRFS_XATTR_ITEM_KEY
;
4385 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4390 eb
= path
->nodes
[0];
4391 slot
= path
->slots
[0];
4392 if (slot
>= btrfs_header_nritems(eb
)) {
4393 ret
= btrfs_next_leaf(root
, path
);
4396 } else if (ret
> 0) {
4403 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4404 if (found_key
.objectid
!= key
.objectid
||
4405 found_key
.type
!= key
.type
) {
4410 ret
= iterate_dir_item(root
, path
, &found_key
,
4411 __process_new_xattr
, sctx
);
4419 btrfs_free_path(path
);
4423 static ssize_t
fill_read_buf(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4425 struct btrfs_root
*root
= sctx
->send_root
;
4426 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4427 struct inode
*inode
;
4430 struct btrfs_key key
;
4431 pgoff_t index
= offset
>> PAGE_CACHE_SHIFT
;
4433 unsigned pg_offset
= offset
& ~PAGE_CACHE_MASK
;
4436 key
.objectid
= sctx
->cur_ino
;
4437 key
.type
= BTRFS_INODE_ITEM_KEY
;
4440 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
4442 return PTR_ERR(inode
);
4444 if (offset
+ len
> i_size_read(inode
)) {
4445 if (offset
> i_size_read(inode
))
4448 len
= offset
- i_size_read(inode
);
4453 last_index
= (offset
+ len
- 1) >> PAGE_CACHE_SHIFT
;
4455 /* initial readahead */
4456 memset(&sctx
->ra
, 0, sizeof(struct file_ra_state
));
4457 file_ra_state_init(&sctx
->ra
, inode
->i_mapping
);
4458 btrfs_force_ra(inode
->i_mapping
, &sctx
->ra
, NULL
, index
,
4459 last_index
- index
+ 1);
4461 while (index
<= last_index
) {
4462 unsigned cur_len
= min_t(unsigned, len
,
4463 PAGE_CACHE_SIZE
- pg_offset
);
4464 page
= find_or_create_page(inode
->i_mapping
, index
, GFP_NOFS
);
4470 if (!PageUptodate(page
)) {
4471 btrfs_readpage(NULL
, page
);
4473 if (!PageUptodate(page
)) {
4475 page_cache_release(page
);
4482 memcpy(sctx
->read_buf
+ ret
, addr
+ pg_offset
, cur_len
);
4485 page_cache_release(page
);
4497 * Read some bytes from the current inode/file and send a write command to
4500 static int send_write(struct send_ctx
*sctx
, u64 offset
, u32 len
)
4504 ssize_t num_read
= 0;
4506 p
= fs_path_alloc();
4510 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset
, len
);
4512 num_read
= fill_read_buf(sctx
, offset
, len
);
4513 if (num_read
<= 0) {
4519 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4523 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4527 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4528 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4529 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, num_read
);
4531 ret
= send_cmd(sctx
);
4542 * Send a clone command to user space.
4544 static int send_clone(struct send_ctx
*sctx
,
4545 u64 offset
, u32 len
,
4546 struct clone_root
*clone_root
)
4552 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4553 "clone_inode=%llu, clone_offset=%llu\n", offset
, len
,
4554 clone_root
->root
->objectid
, clone_root
->ino
,
4555 clone_root
->offset
);
4557 p
= fs_path_alloc();
4561 ret
= begin_cmd(sctx
, BTRFS_SEND_C_CLONE
);
4565 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4569 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4570 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_LEN
, len
);
4571 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4573 if (clone_root
->root
== sctx
->send_root
) {
4574 ret
= get_inode_info(sctx
->send_root
, clone_root
->ino
, NULL
,
4575 &gen
, NULL
, NULL
, NULL
, NULL
);
4578 ret
= get_cur_path(sctx
, clone_root
->ino
, gen
, p
);
4580 ret
= get_inode_path(clone_root
->root
, clone_root
->ino
, p
);
4586 * If the parent we're using has a received_uuid set then use that as
4587 * our clone source as that is what we will look for when doing a
4590 * This covers the case that we create a snapshot off of a received
4591 * subvolume and then use that as the parent and try to receive on a
4594 if (!btrfs_is_empty_uuid(clone_root
->root
->root_item
.received_uuid
))
4595 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4596 clone_root
->root
->root_item
.received_uuid
);
4598 TLV_PUT_UUID(sctx
, BTRFS_SEND_A_CLONE_UUID
,
4599 clone_root
->root
->root_item
.uuid
);
4600 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_CTRANSID
,
4601 le64_to_cpu(clone_root
->root
->root_item
.ctransid
));
4602 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_CLONE_PATH
, p
);
4603 TLV_PUT_U64(sctx
, BTRFS_SEND_A_CLONE_OFFSET
,
4604 clone_root
->offset
);
4606 ret
= send_cmd(sctx
);
4615 * Send an update extent command to user space.
4617 static int send_update_extent(struct send_ctx
*sctx
,
4618 u64 offset
, u32 len
)
4623 p
= fs_path_alloc();
4627 ret
= begin_cmd(sctx
, BTRFS_SEND_C_UPDATE_EXTENT
);
4631 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4635 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4636 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4637 TLV_PUT_U64(sctx
, BTRFS_SEND_A_SIZE
, len
);
4639 ret
= send_cmd(sctx
);
4647 static int send_hole(struct send_ctx
*sctx
, u64 end
)
4649 struct fs_path
*p
= NULL
;
4650 u64 offset
= sctx
->cur_inode_last_extent
;
4654 p
= fs_path_alloc();
4657 ret
= get_cur_path(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
, p
);
4659 goto tlv_put_failure
;
4660 memset(sctx
->read_buf
, 0, BTRFS_SEND_READ_SIZE
);
4661 while (offset
< end
) {
4662 len
= min_t(u64
, end
- offset
, BTRFS_SEND_READ_SIZE
);
4664 ret
= begin_cmd(sctx
, BTRFS_SEND_C_WRITE
);
4667 TLV_PUT_PATH(sctx
, BTRFS_SEND_A_PATH
, p
);
4668 TLV_PUT_U64(sctx
, BTRFS_SEND_A_FILE_OFFSET
, offset
);
4669 TLV_PUT(sctx
, BTRFS_SEND_A_DATA
, sctx
->read_buf
, len
);
4670 ret
= send_cmd(sctx
);
4680 static int send_extent_data(struct send_ctx
*sctx
,
4686 if (sctx
->flags
& BTRFS_SEND_FLAG_NO_FILE_DATA
)
4687 return send_update_extent(sctx
, offset
, len
);
4689 while (sent
< len
) {
4690 u64 size
= len
- sent
;
4693 if (size
> BTRFS_SEND_READ_SIZE
)
4694 size
= BTRFS_SEND_READ_SIZE
;
4695 ret
= send_write(sctx
, offset
+ sent
, size
);
4705 static int clone_range(struct send_ctx
*sctx
,
4706 struct clone_root
*clone_root
,
4707 const u64 disk_byte
,
4712 struct btrfs_path
*path
;
4713 struct btrfs_key key
;
4716 path
= alloc_path_for_send();
4721 * We can't send a clone operation for the entire range if we find
4722 * extent items in the respective range in the source file that
4723 * refer to different extents or if we find holes.
4724 * So check for that and do a mix of clone and regular write/copy
4725 * operations if needed.
4729 * mkfs.btrfs -f /dev/sda
4730 * mount /dev/sda /mnt
4731 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
4732 * cp --reflink=always /mnt/foo /mnt/bar
4733 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
4734 * btrfs subvolume snapshot -r /mnt /mnt/snap
4736 * If when we send the snapshot and we are processing file bar (which
4737 * has a higher inode number than foo) we blindly send a clone operation
4738 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
4739 * a file bar that matches the content of file foo - iow, doesn't match
4740 * the content from bar in the original filesystem.
4742 key
.objectid
= clone_root
->ino
;
4743 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4744 key
.offset
= clone_root
->offset
;
4745 ret
= btrfs_search_slot(NULL
, clone_root
->root
, &key
, path
, 0, 0);
4748 if (ret
> 0 && path
->slots
[0] > 0) {
4749 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0] - 1);
4750 if (key
.objectid
== clone_root
->ino
&&
4751 key
.type
== BTRFS_EXTENT_DATA_KEY
)
4756 struct extent_buffer
*leaf
= path
->nodes
[0];
4757 int slot
= path
->slots
[0];
4758 struct btrfs_file_extent_item
*ei
;
4763 if (slot
>= btrfs_header_nritems(leaf
)) {
4764 ret
= btrfs_next_leaf(clone_root
->root
, path
);
4772 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
4775 * We might have an implicit trailing hole (NO_HOLES feature
4776 * enabled). We deal with it after leaving this loop.
4778 if (key
.objectid
!= clone_root
->ino
||
4779 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4782 ei
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4783 type
= btrfs_file_extent_type(leaf
, ei
);
4784 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4785 ext_len
= btrfs_file_extent_inline_len(leaf
, slot
, ei
);
4786 ext_len
= PAGE_CACHE_ALIGN(ext_len
);
4788 ext_len
= btrfs_file_extent_num_bytes(leaf
, ei
);
4791 if (key
.offset
+ ext_len
<= clone_root
->offset
)
4794 if (key
.offset
> clone_root
->offset
) {
4795 /* Implicit hole, NO_HOLES feature enabled. */
4796 u64 hole_len
= key
.offset
- clone_root
->offset
;
4800 ret
= send_extent_data(sctx
, offset
, hole_len
);
4808 clone_root
->offset
+= hole_len
;
4809 data_offset
+= hole_len
;
4812 if (key
.offset
>= clone_root
->offset
+ len
)
4815 clone_len
= min_t(u64
, ext_len
, len
);
4817 if (btrfs_file_extent_disk_bytenr(leaf
, ei
) == disk_byte
&&
4818 btrfs_file_extent_offset(leaf
, ei
) == data_offset
)
4819 ret
= send_clone(sctx
, offset
, clone_len
, clone_root
);
4821 ret
= send_extent_data(sctx
, offset
, clone_len
);
4829 offset
+= clone_len
;
4830 clone_root
->offset
+= clone_len
;
4831 data_offset
+= clone_len
;
4837 ret
= send_extent_data(sctx
, offset
, len
);
4841 btrfs_free_path(path
);
4845 static int send_write_or_clone(struct send_ctx
*sctx
,
4846 struct btrfs_path
*path
,
4847 struct btrfs_key
*key
,
4848 struct clone_root
*clone_root
)
4851 struct btrfs_file_extent_item
*ei
;
4852 u64 offset
= key
->offset
;
4855 u64 bs
= sctx
->send_root
->fs_info
->sb
->s_blocksize
;
4857 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4858 struct btrfs_file_extent_item
);
4859 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
4860 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
4861 len
= btrfs_file_extent_inline_len(path
->nodes
[0],
4862 path
->slots
[0], ei
);
4864 * it is possible the inline item won't cover the whole page,
4865 * but there may be items after this page. Make
4866 * sure to send the whole thing
4868 len
= PAGE_CACHE_ALIGN(len
);
4870 len
= btrfs_file_extent_num_bytes(path
->nodes
[0], ei
);
4873 if (offset
+ len
> sctx
->cur_inode_size
)
4874 len
= sctx
->cur_inode_size
- offset
;
4880 if (clone_root
&& IS_ALIGNED(offset
+ len
, bs
)) {
4884 disk_byte
= btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
);
4885 data_offset
= btrfs_file_extent_offset(path
->nodes
[0], ei
);
4886 ret
= clone_range(sctx
, clone_root
, disk_byte
, data_offset
,
4889 ret
= send_extent_data(sctx
, offset
, len
);
4895 static int is_extent_unchanged(struct send_ctx
*sctx
,
4896 struct btrfs_path
*left_path
,
4897 struct btrfs_key
*ekey
)
4900 struct btrfs_key key
;
4901 struct btrfs_path
*path
= NULL
;
4902 struct extent_buffer
*eb
;
4904 struct btrfs_key found_key
;
4905 struct btrfs_file_extent_item
*ei
;
4910 u64 left_offset_fixed
;
4918 path
= alloc_path_for_send();
4922 eb
= left_path
->nodes
[0];
4923 slot
= left_path
->slots
[0];
4924 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
4925 left_type
= btrfs_file_extent_type(eb
, ei
);
4927 if (left_type
!= BTRFS_FILE_EXTENT_REG
) {
4931 left_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
4932 left_len
= btrfs_file_extent_num_bytes(eb
, ei
);
4933 left_offset
= btrfs_file_extent_offset(eb
, ei
);
4934 left_gen
= btrfs_file_extent_generation(eb
, ei
);
4937 * Following comments will refer to these graphics. L is the left
4938 * extents which we are checking at the moment. 1-8 are the right
4939 * extents that we iterate.
4942 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4945 * |--1--|-2b-|...(same as above)
4947 * Alternative situation. Happens on files where extents got split.
4949 * |-----------7-----------|-6-|
4951 * Alternative situation. Happens on files which got larger.
4954 * Nothing follows after 8.
4957 key
.objectid
= ekey
->objectid
;
4958 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4959 key
.offset
= ekey
->offset
;
4960 ret
= btrfs_search_slot_for_read(sctx
->parent_root
, &key
, path
, 0, 0);
4969 * Handle special case where the right side has no extents at all.
4971 eb
= path
->nodes
[0];
4972 slot
= path
->slots
[0];
4973 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4974 if (found_key
.objectid
!= key
.objectid
||
4975 found_key
.type
!= key
.type
) {
4976 /* If we're a hole then just pretend nothing changed */
4977 ret
= (left_disknr
) ? 0 : 1;
4982 * We're now on 2a, 2b or 7.
4985 while (key
.offset
< ekey
->offset
+ left_len
) {
4986 ei
= btrfs_item_ptr(eb
, slot
, struct btrfs_file_extent_item
);
4987 right_type
= btrfs_file_extent_type(eb
, ei
);
4988 if (right_type
!= BTRFS_FILE_EXTENT_REG
) {
4993 right_disknr
= btrfs_file_extent_disk_bytenr(eb
, ei
);
4994 right_len
= btrfs_file_extent_num_bytes(eb
, ei
);
4995 right_offset
= btrfs_file_extent_offset(eb
, ei
);
4996 right_gen
= btrfs_file_extent_generation(eb
, ei
);
4999 * Are we at extent 8? If yes, we know the extent is changed.
5000 * This may only happen on the first iteration.
5002 if (found_key
.offset
+ right_len
<= ekey
->offset
) {
5003 /* If we're a hole just pretend nothing changed */
5004 ret
= (left_disknr
) ? 0 : 1;
5008 left_offset_fixed
= left_offset
;
5009 if (key
.offset
< ekey
->offset
) {
5010 /* Fix the right offset for 2a and 7. */
5011 right_offset
+= ekey
->offset
- key
.offset
;
5013 /* Fix the left offset for all behind 2a and 2b */
5014 left_offset_fixed
+= key
.offset
- ekey
->offset
;
5018 * Check if we have the same extent.
5020 if (left_disknr
!= right_disknr
||
5021 left_offset_fixed
!= right_offset
||
5022 left_gen
!= right_gen
) {
5028 * Go to the next extent.
5030 ret
= btrfs_next_item(sctx
->parent_root
, path
);
5034 eb
= path
->nodes
[0];
5035 slot
= path
->slots
[0];
5036 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5038 if (ret
|| found_key
.objectid
!= key
.objectid
||
5039 found_key
.type
!= key
.type
) {
5040 key
.offset
+= right_len
;
5043 if (found_key
.offset
!= key
.offset
+ right_len
) {
5051 * We're now behind the left extent (treat as unchanged) or at the end
5052 * of the right side (treat as changed).
5054 if (key
.offset
>= ekey
->offset
+ left_len
)
5061 btrfs_free_path(path
);
5065 static int get_last_extent(struct send_ctx
*sctx
, u64 offset
)
5067 struct btrfs_path
*path
;
5068 struct btrfs_root
*root
= sctx
->send_root
;
5069 struct btrfs_file_extent_item
*fi
;
5070 struct btrfs_key key
;
5075 path
= alloc_path_for_send();
5079 sctx
->cur_inode_last_extent
= 0;
5081 key
.objectid
= sctx
->cur_ino
;
5082 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5083 key
.offset
= offset
;
5084 ret
= btrfs_search_slot_for_read(root
, &key
, path
, 0, 1);
5088 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
5089 if (key
.objectid
!= sctx
->cur_ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5092 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5093 struct btrfs_file_extent_item
);
5094 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
5095 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5096 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
5097 path
->slots
[0], fi
);
5098 extent_end
= ALIGN(key
.offset
+ size
,
5099 sctx
->send_root
->sectorsize
);
5101 extent_end
= key
.offset
+
5102 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
5104 sctx
->cur_inode_last_extent
= extent_end
;
5106 btrfs_free_path(path
);
5110 static int maybe_send_hole(struct send_ctx
*sctx
, struct btrfs_path
*path
,
5111 struct btrfs_key
*key
)
5113 struct btrfs_file_extent_item
*fi
;
5118 if (sctx
->cur_ino
!= key
->objectid
|| !need_send_hole(sctx
))
5121 if (sctx
->cur_inode_last_extent
== (u64
)-1) {
5122 ret
= get_last_extent(sctx
, key
->offset
- 1);
5127 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5128 struct btrfs_file_extent_item
);
5129 type
= btrfs_file_extent_type(path
->nodes
[0], fi
);
5130 if (type
== BTRFS_FILE_EXTENT_INLINE
) {
5131 u64 size
= btrfs_file_extent_inline_len(path
->nodes
[0],
5132 path
->slots
[0], fi
);
5133 extent_end
= ALIGN(key
->offset
+ size
,
5134 sctx
->send_root
->sectorsize
);
5136 extent_end
= key
->offset
+
5137 btrfs_file_extent_num_bytes(path
->nodes
[0], fi
);
5140 if (path
->slots
[0] == 0 &&
5141 sctx
->cur_inode_last_extent
< key
->offset
) {
5143 * We might have skipped entire leafs that contained only
5144 * file extent items for our current inode. These leafs have
5145 * a generation number smaller (older) than the one in the
5146 * current leaf and the leaf our last extent came from, and
5147 * are located between these 2 leafs.
5149 ret
= get_last_extent(sctx
, key
->offset
- 1);
5154 if (sctx
->cur_inode_last_extent
< key
->offset
)
5155 ret
= send_hole(sctx
, key
->offset
);
5156 sctx
->cur_inode_last_extent
= extent_end
;
5160 static int process_extent(struct send_ctx
*sctx
,
5161 struct btrfs_path
*path
,
5162 struct btrfs_key
*key
)
5164 struct clone_root
*found_clone
= NULL
;
5167 if (S_ISLNK(sctx
->cur_inode_mode
))
5170 if (sctx
->parent_root
&& !sctx
->cur_inode_new
) {
5171 ret
= is_extent_unchanged(sctx
, path
, key
);
5179 struct btrfs_file_extent_item
*ei
;
5182 ei
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5183 struct btrfs_file_extent_item
);
5184 type
= btrfs_file_extent_type(path
->nodes
[0], ei
);
5185 if (type
== BTRFS_FILE_EXTENT_PREALLOC
||
5186 type
== BTRFS_FILE_EXTENT_REG
) {
5188 * The send spec does not have a prealloc command yet,
5189 * so just leave a hole for prealloc'ed extents until
5190 * we have enough commands queued up to justify rev'ing
5193 if (type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5198 /* Have a hole, just skip it. */
5199 if (btrfs_file_extent_disk_bytenr(path
->nodes
[0], ei
) == 0) {
5206 ret
= find_extent_clone(sctx
, path
, key
->objectid
, key
->offset
,
5207 sctx
->cur_inode_size
, &found_clone
);
5208 if (ret
!= -ENOENT
&& ret
< 0)
5211 ret
= send_write_or_clone(sctx
, path
, key
, found_clone
);
5215 ret
= maybe_send_hole(sctx
, path
, key
);
5220 static int process_all_extents(struct send_ctx
*sctx
)
5223 struct btrfs_root
*root
;
5224 struct btrfs_path
*path
;
5225 struct btrfs_key key
;
5226 struct btrfs_key found_key
;
5227 struct extent_buffer
*eb
;
5230 root
= sctx
->send_root
;
5231 path
= alloc_path_for_send();
5235 key
.objectid
= sctx
->cmp_key
->objectid
;
5236 key
.type
= BTRFS_EXTENT_DATA_KEY
;
5238 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5243 eb
= path
->nodes
[0];
5244 slot
= path
->slots
[0];
5246 if (slot
>= btrfs_header_nritems(eb
)) {
5247 ret
= btrfs_next_leaf(root
, path
);
5250 } else if (ret
> 0) {
5257 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5259 if (found_key
.objectid
!= key
.objectid
||
5260 found_key
.type
!= key
.type
) {
5265 ret
= process_extent(sctx
, path
, &found_key
);
5273 btrfs_free_path(path
);
5277 static int process_recorded_refs_if_needed(struct send_ctx
*sctx
, int at_end
,
5279 int *refs_processed
)
5283 if (sctx
->cur_ino
== 0)
5285 if (!at_end
&& sctx
->cur_ino
== sctx
->cmp_key
->objectid
&&
5286 sctx
->cmp_key
->type
<= BTRFS_INODE_EXTREF_KEY
)
5288 if (list_empty(&sctx
->new_refs
) && list_empty(&sctx
->deleted_refs
))
5291 ret
= process_recorded_refs(sctx
, pending_move
);
5295 *refs_processed
= 1;
5300 static int finish_inode_if_needed(struct send_ctx
*sctx
, int at_end
)
5311 int pending_move
= 0;
5312 int refs_processed
= 0;
5314 ret
= process_recorded_refs_if_needed(sctx
, at_end
, &pending_move
,
5320 * We have processed the refs and thus need to advance send_progress.
5321 * Now, calls to get_cur_xxx will take the updated refs of the current
5322 * inode into account.
5324 * On the other hand, if our current inode is a directory and couldn't
5325 * be moved/renamed because its parent was renamed/moved too and it has
5326 * a higher inode number, we can only move/rename our current inode
5327 * after we moved/renamed its parent. Therefore in this case operate on
5328 * the old path (pre move/rename) of our current inode, and the
5329 * move/rename will be performed later.
5331 if (refs_processed
&& !pending_move
)
5332 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5334 if (sctx
->cur_ino
== 0 || sctx
->cur_inode_deleted
)
5336 if (!at_end
&& sctx
->cmp_key
->objectid
== sctx
->cur_ino
)
5339 ret
= get_inode_info(sctx
->send_root
, sctx
->cur_ino
, NULL
, NULL
,
5340 &left_mode
, &left_uid
, &left_gid
, NULL
);
5344 if (!sctx
->parent_root
|| sctx
->cur_inode_new
) {
5346 if (!S_ISLNK(sctx
->cur_inode_mode
))
5349 ret
= get_inode_info(sctx
->parent_root
, sctx
->cur_ino
,
5350 NULL
, NULL
, &right_mode
, &right_uid
,
5355 if (left_uid
!= right_uid
|| left_gid
!= right_gid
)
5357 if (!S_ISLNK(sctx
->cur_inode_mode
) && left_mode
!= right_mode
)
5361 if (S_ISREG(sctx
->cur_inode_mode
)) {
5362 if (need_send_hole(sctx
)) {
5363 if (sctx
->cur_inode_last_extent
== (u64
)-1 ||
5364 sctx
->cur_inode_last_extent
<
5365 sctx
->cur_inode_size
) {
5366 ret
= get_last_extent(sctx
, (u64
)-1);
5370 if (sctx
->cur_inode_last_extent
<
5371 sctx
->cur_inode_size
) {
5372 ret
= send_hole(sctx
, sctx
->cur_inode_size
);
5377 ret
= send_truncate(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5378 sctx
->cur_inode_size
);
5384 ret
= send_chown(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5385 left_uid
, left_gid
);
5390 ret
= send_chmod(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
,
5397 * If other directory inodes depended on our current directory
5398 * inode's move/rename, now do their move/rename operations.
5400 if (!is_waiting_for_move(sctx
, sctx
->cur_ino
)) {
5401 ret
= apply_children_dir_moves(sctx
);
5405 * Need to send that every time, no matter if it actually
5406 * changed between the two trees as we have done changes to
5407 * the inode before. If our inode is a directory and it's
5408 * waiting to be moved/renamed, we will send its utimes when
5409 * it's moved/renamed, therefore we don't need to do it here.
5411 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5412 ret
= send_utimes(sctx
, sctx
->cur_ino
, sctx
->cur_inode_gen
);
5421 static int changed_inode(struct send_ctx
*sctx
,
5422 enum btrfs_compare_tree_result result
)
5425 struct btrfs_key
*key
= sctx
->cmp_key
;
5426 struct btrfs_inode_item
*left_ii
= NULL
;
5427 struct btrfs_inode_item
*right_ii
= NULL
;
5431 sctx
->cur_ino
= key
->objectid
;
5432 sctx
->cur_inode_new_gen
= 0;
5433 sctx
->cur_inode_last_extent
= (u64
)-1;
5436 * Set send_progress to current inode. This will tell all get_cur_xxx
5437 * functions that the current inode's refs are not updated yet. Later,
5438 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5440 sctx
->send_progress
= sctx
->cur_ino
;
5442 if (result
== BTRFS_COMPARE_TREE_NEW
||
5443 result
== BTRFS_COMPARE_TREE_CHANGED
) {
5444 left_ii
= btrfs_item_ptr(sctx
->left_path
->nodes
[0],
5445 sctx
->left_path
->slots
[0],
5446 struct btrfs_inode_item
);
5447 left_gen
= btrfs_inode_generation(sctx
->left_path
->nodes
[0],
5450 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5451 sctx
->right_path
->slots
[0],
5452 struct btrfs_inode_item
);
5453 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5456 if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5457 right_ii
= btrfs_item_ptr(sctx
->right_path
->nodes
[0],
5458 sctx
->right_path
->slots
[0],
5459 struct btrfs_inode_item
);
5461 right_gen
= btrfs_inode_generation(sctx
->right_path
->nodes
[0],
5465 * The cur_ino = root dir case is special here. We can't treat
5466 * the inode as deleted+reused because it would generate a
5467 * stream that tries to delete/mkdir the root dir.
5469 if (left_gen
!= right_gen
&&
5470 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5471 sctx
->cur_inode_new_gen
= 1;
5474 if (result
== BTRFS_COMPARE_TREE_NEW
) {
5475 sctx
->cur_inode_gen
= left_gen
;
5476 sctx
->cur_inode_new
= 1;
5477 sctx
->cur_inode_deleted
= 0;
5478 sctx
->cur_inode_size
= btrfs_inode_size(
5479 sctx
->left_path
->nodes
[0], left_ii
);
5480 sctx
->cur_inode_mode
= btrfs_inode_mode(
5481 sctx
->left_path
->nodes
[0], left_ii
);
5482 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
5483 sctx
->left_path
->nodes
[0], left_ii
);
5484 if (sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
5485 ret
= send_create_inode_if_needed(sctx
);
5486 } else if (result
== BTRFS_COMPARE_TREE_DELETED
) {
5487 sctx
->cur_inode_gen
= right_gen
;
5488 sctx
->cur_inode_new
= 0;
5489 sctx
->cur_inode_deleted
= 1;
5490 sctx
->cur_inode_size
= btrfs_inode_size(
5491 sctx
->right_path
->nodes
[0], right_ii
);
5492 sctx
->cur_inode_mode
= btrfs_inode_mode(
5493 sctx
->right_path
->nodes
[0], right_ii
);
5494 } else if (result
== BTRFS_COMPARE_TREE_CHANGED
) {
5496 * We need to do some special handling in case the inode was
5497 * reported as changed with a changed generation number. This
5498 * means that the original inode was deleted and new inode
5499 * reused the same inum. So we have to treat the old inode as
5500 * deleted and the new one as new.
5502 if (sctx
->cur_inode_new_gen
) {
5504 * First, process the inode as if it was deleted.
5506 sctx
->cur_inode_gen
= right_gen
;
5507 sctx
->cur_inode_new
= 0;
5508 sctx
->cur_inode_deleted
= 1;
5509 sctx
->cur_inode_size
= btrfs_inode_size(
5510 sctx
->right_path
->nodes
[0], right_ii
);
5511 sctx
->cur_inode_mode
= btrfs_inode_mode(
5512 sctx
->right_path
->nodes
[0], right_ii
);
5513 ret
= process_all_refs(sctx
,
5514 BTRFS_COMPARE_TREE_DELETED
);
5519 * Now process the inode as if it was new.
5521 sctx
->cur_inode_gen
= left_gen
;
5522 sctx
->cur_inode_new
= 1;
5523 sctx
->cur_inode_deleted
= 0;
5524 sctx
->cur_inode_size
= btrfs_inode_size(
5525 sctx
->left_path
->nodes
[0], left_ii
);
5526 sctx
->cur_inode_mode
= btrfs_inode_mode(
5527 sctx
->left_path
->nodes
[0], left_ii
);
5528 sctx
->cur_inode_rdev
= btrfs_inode_rdev(
5529 sctx
->left_path
->nodes
[0], left_ii
);
5530 ret
= send_create_inode_if_needed(sctx
);
5534 ret
= process_all_refs(sctx
, BTRFS_COMPARE_TREE_NEW
);
5538 * Advance send_progress now as we did not get into
5539 * process_recorded_refs_if_needed in the new_gen case.
5541 sctx
->send_progress
= sctx
->cur_ino
+ 1;
5544 * Now process all extents and xattrs of the inode as if
5545 * they were all new.
5547 ret
= process_all_extents(sctx
);
5550 ret
= process_all_new_xattrs(sctx
);
5554 sctx
->cur_inode_gen
= left_gen
;
5555 sctx
->cur_inode_new
= 0;
5556 sctx
->cur_inode_new_gen
= 0;
5557 sctx
->cur_inode_deleted
= 0;
5558 sctx
->cur_inode_size
= btrfs_inode_size(
5559 sctx
->left_path
->nodes
[0], left_ii
);
5560 sctx
->cur_inode_mode
= btrfs_inode_mode(
5561 sctx
->left_path
->nodes
[0], left_ii
);
5570 * We have to process new refs before deleted refs, but compare_trees gives us
5571 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5572 * first and later process them in process_recorded_refs.
5573 * For the cur_inode_new_gen case, we skip recording completely because
5574 * changed_inode did already initiate processing of refs. The reason for this is
5575 * that in this case, compare_tree actually compares the refs of 2 different
5576 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5577 * refs of the right tree as deleted and all refs of the left tree as new.
5579 static int changed_ref(struct send_ctx
*sctx
,
5580 enum btrfs_compare_tree_result result
)
5584 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
5586 if (!sctx
->cur_inode_new_gen
&&
5587 sctx
->cur_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
5588 if (result
== BTRFS_COMPARE_TREE_NEW
)
5589 ret
= record_new_ref(sctx
);
5590 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
5591 ret
= record_deleted_ref(sctx
);
5592 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
5593 ret
= record_changed_ref(sctx
);
5600 * Process new/deleted/changed xattrs. We skip processing in the
5601 * cur_inode_new_gen case because changed_inode did already initiate processing
5602 * of xattrs. The reason is the same as in changed_ref
5604 static int changed_xattr(struct send_ctx
*sctx
,
5605 enum btrfs_compare_tree_result result
)
5609 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
5611 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
5612 if (result
== BTRFS_COMPARE_TREE_NEW
)
5613 ret
= process_new_xattr(sctx
);
5614 else if (result
== BTRFS_COMPARE_TREE_DELETED
)
5615 ret
= process_deleted_xattr(sctx
);
5616 else if (result
== BTRFS_COMPARE_TREE_CHANGED
)
5617 ret
= process_changed_xattr(sctx
);
5624 * Process new/deleted/changed extents. We skip processing in the
5625 * cur_inode_new_gen case because changed_inode did already initiate processing
5626 * of extents. The reason is the same as in changed_ref
5628 static int changed_extent(struct send_ctx
*sctx
,
5629 enum btrfs_compare_tree_result result
)
5633 BUG_ON(sctx
->cur_ino
!= sctx
->cmp_key
->objectid
);
5635 if (!sctx
->cur_inode_new_gen
&& !sctx
->cur_inode_deleted
) {
5636 if (result
!= BTRFS_COMPARE_TREE_DELETED
)
5637 ret
= process_extent(sctx
, sctx
->left_path
,
5644 static int dir_changed(struct send_ctx
*sctx
, u64 dir
)
5646 u64 orig_gen
, new_gen
;
5649 ret
= get_inode_info(sctx
->send_root
, dir
, NULL
, &new_gen
, NULL
, NULL
,
5654 ret
= get_inode_info(sctx
->parent_root
, dir
, NULL
, &orig_gen
, NULL
,
5659 return (orig_gen
!= new_gen
) ? 1 : 0;
5662 static int compare_refs(struct send_ctx
*sctx
, struct btrfs_path
*path
,
5663 struct btrfs_key
*key
)
5665 struct btrfs_inode_extref
*extref
;
5666 struct extent_buffer
*leaf
;
5667 u64 dirid
= 0, last_dirid
= 0;
5674 /* Easy case, just check this one dirid */
5675 if (key
->type
== BTRFS_INODE_REF_KEY
) {
5676 dirid
= key
->offset
;
5678 ret
= dir_changed(sctx
, dirid
);
5682 leaf
= path
->nodes
[0];
5683 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
5684 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
5685 while (cur_offset
< item_size
) {
5686 extref
= (struct btrfs_inode_extref
*)(ptr
+
5688 dirid
= btrfs_inode_extref_parent(leaf
, extref
);
5689 ref_name_len
= btrfs_inode_extref_name_len(leaf
, extref
);
5690 cur_offset
+= ref_name_len
+ sizeof(*extref
);
5691 if (dirid
== last_dirid
)
5693 ret
= dir_changed(sctx
, dirid
);
5703 * Updates compare related fields in sctx and simply forwards to the actual
5704 * changed_xxx functions.
5706 static int changed_cb(struct btrfs_root
*left_root
,
5707 struct btrfs_root
*right_root
,
5708 struct btrfs_path
*left_path
,
5709 struct btrfs_path
*right_path
,
5710 struct btrfs_key
*key
,
5711 enum btrfs_compare_tree_result result
,
5715 struct send_ctx
*sctx
= ctx
;
5717 if (result
== BTRFS_COMPARE_TREE_SAME
) {
5718 if (key
->type
== BTRFS_INODE_REF_KEY
||
5719 key
->type
== BTRFS_INODE_EXTREF_KEY
) {
5720 ret
= compare_refs(sctx
, left_path
, key
);
5725 } else if (key
->type
== BTRFS_EXTENT_DATA_KEY
) {
5726 return maybe_send_hole(sctx
, left_path
, key
);
5730 result
= BTRFS_COMPARE_TREE_CHANGED
;
5734 sctx
->left_path
= left_path
;
5735 sctx
->right_path
= right_path
;
5736 sctx
->cmp_key
= key
;
5738 ret
= finish_inode_if_needed(sctx
, 0);
5742 /* Ignore non-FS objects */
5743 if (key
->objectid
== BTRFS_FREE_INO_OBJECTID
||
5744 key
->objectid
== BTRFS_FREE_SPACE_OBJECTID
)
5747 if (key
->type
== BTRFS_INODE_ITEM_KEY
)
5748 ret
= changed_inode(sctx
, result
);
5749 else if (key
->type
== BTRFS_INODE_REF_KEY
||
5750 key
->type
== BTRFS_INODE_EXTREF_KEY
)
5751 ret
= changed_ref(sctx
, result
);
5752 else if (key
->type
== BTRFS_XATTR_ITEM_KEY
)
5753 ret
= changed_xattr(sctx
, result
);
5754 else if (key
->type
== BTRFS_EXTENT_DATA_KEY
)
5755 ret
= changed_extent(sctx
, result
);
5761 static int full_send_tree(struct send_ctx
*sctx
)
5764 struct btrfs_root
*send_root
= sctx
->send_root
;
5765 struct btrfs_key key
;
5766 struct btrfs_key found_key
;
5767 struct btrfs_path
*path
;
5768 struct extent_buffer
*eb
;
5771 path
= alloc_path_for_send();
5775 key
.objectid
= BTRFS_FIRST_FREE_OBJECTID
;
5776 key
.type
= BTRFS_INODE_ITEM_KEY
;
5779 ret
= btrfs_search_slot_for_read(send_root
, &key
, path
, 1, 0);
5786 eb
= path
->nodes
[0];
5787 slot
= path
->slots
[0];
5788 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5790 ret
= changed_cb(send_root
, NULL
, path
, NULL
,
5791 &found_key
, BTRFS_COMPARE_TREE_NEW
, sctx
);
5795 key
.objectid
= found_key
.objectid
;
5796 key
.type
= found_key
.type
;
5797 key
.offset
= found_key
.offset
+ 1;
5799 ret
= btrfs_next_item(send_root
, path
);
5809 ret
= finish_inode_if_needed(sctx
, 1);
5812 btrfs_free_path(path
);
5816 static int send_subvol(struct send_ctx
*sctx
)
5820 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_STREAM_HEADER
)) {
5821 ret
= send_header(sctx
);
5826 ret
= send_subvol_begin(sctx
);
5830 if (sctx
->parent_root
) {
5831 ret
= btrfs_compare_trees(sctx
->send_root
, sctx
->parent_root
,
5835 ret
= finish_inode_if_needed(sctx
, 1);
5839 ret
= full_send_tree(sctx
);
5845 free_recorded_refs(sctx
);
5850 * If orphan cleanup did remove any orphans from a root, it means the tree
5851 * was modified and therefore the commit root is not the same as the current
5852 * root anymore. This is a problem, because send uses the commit root and
5853 * therefore can see inode items that don't exist in the current root anymore,
5854 * and for example make calls to btrfs_iget, which will do tree lookups based
5855 * on the current root and not on the commit root. Those lookups will fail,
5856 * returning a -ESTALE error, and making send fail with that error. So make
5857 * sure a send does not see any orphans we have just removed, and that it will
5858 * see the same inodes regardless of whether a transaction commit happened
5859 * before it started (meaning that the commit root will be the same as the
5860 * current root) or not.
5862 static int ensure_commit_roots_uptodate(struct send_ctx
*sctx
)
5865 struct btrfs_trans_handle
*trans
= NULL
;
5868 if (sctx
->parent_root
&&
5869 sctx
->parent_root
->node
!= sctx
->parent_root
->commit_root
)
5872 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
5873 if (sctx
->clone_roots
[i
].root
->node
!=
5874 sctx
->clone_roots
[i
].root
->commit_root
)
5878 return btrfs_end_transaction(trans
, sctx
->send_root
);
5883 /* Use any root, all fs roots will get their commit roots updated. */
5885 trans
= btrfs_join_transaction(sctx
->send_root
);
5887 return PTR_ERR(trans
);
5891 return btrfs_commit_transaction(trans
, sctx
->send_root
);
5894 static void btrfs_root_dec_send_in_progress(struct btrfs_root
* root
)
5896 spin_lock(&root
->root_item_lock
);
5897 root
->send_in_progress
--;
5899 * Not much left to do, we don't know why it's unbalanced and
5900 * can't blindly reset it to 0.
5902 if (root
->send_in_progress
< 0)
5903 btrfs_err(root
->fs_info
,
5904 "send_in_progres unbalanced %d root %llu",
5905 root
->send_in_progress
, root
->root_key
.objectid
);
5906 spin_unlock(&root
->root_item_lock
);
5909 long btrfs_ioctl_send(struct file
*mnt_file
, void __user
*arg_
)
5912 struct btrfs_root
*send_root
;
5913 struct btrfs_root
*clone_root
;
5914 struct btrfs_fs_info
*fs_info
;
5915 struct btrfs_ioctl_send_args
*arg
= NULL
;
5916 struct btrfs_key key
;
5917 struct send_ctx
*sctx
= NULL
;
5919 u64
*clone_sources_tmp
= NULL
;
5920 int clone_sources_to_rollback
= 0;
5921 int sort_clone_roots
= 0;
5924 if (!capable(CAP_SYS_ADMIN
))
5927 send_root
= BTRFS_I(file_inode(mnt_file
))->root
;
5928 fs_info
= send_root
->fs_info
;
5931 * The subvolume must remain read-only during send, protect against
5932 * making it RW. This also protects against deletion.
5934 spin_lock(&send_root
->root_item_lock
);
5935 send_root
->send_in_progress
++;
5936 spin_unlock(&send_root
->root_item_lock
);
5939 * This is done when we lookup the root, it should already be complete
5940 * by the time we get here.
5942 WARN_ON(send_root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
);
5945 * Userspace tools do the checks and warn the user if it's
5948 if (!btrfs_root_readonly(send_root
)) {
5953 arg
= memdup_user(arg_
, sizeof(*arg
));
5960 if (!access_ok(VERIFY_READ
, arg
->clone_sources
,
5961 sizeof(*arg
->clone_sources
) *
5962 arg
->clone_sources_count
)) {
5967 if (arg
->flags
& ~BTRFS_SEND_FLAG_MASK
) {
5972 sctx
= kzalloc(sizeof(struct send_ctx
), GFP_NOFS
);
5978 INIT_LIST_HEAD(&sctx
->new_refs
);
5979 INIT_LIST_HEAD(&sctx
->deleted_refs
);
5980 INIT_RADIX_TREE(&sctx
->name_cache
, GFP_NOFS
);
5981 INIT_LIST_HEAD(&sctx
->name_cache_list
);
5983 sctx
->flags
= arg
->flags
;
5985 sctx
->send_filp
= fget(arg
->send_fd
);
5986 if (!sctx
->send_filp
) {
5991 sctx
->send_root
= send_root
;
5993 * Unlikely but possible, if the subvolume is marked for deletion but
5994 * is slow to remove the directory entry, send can still be started
5996 if (btrfs_root_dead(sctx
->send_root
)) {
6001 sctx
->clone_roots_cnt
= arg
->clone_sources_count
;
6003 sctx
->send_max_size
= BTRFS_SEND_BUF_SIZE
;
6004 sctx
->send_buf
= vmalloc(sctx
->send_max_size
);
6005 if (!sctx
->send_buf
) {
6010 sctx
->read_buf
= vmalloc(BTRFS_SEND_READ_SIZE
);
6011 if (!sctx
->read_buf
) {
6016 sctx
->pending_dir_moves
= RB_ROOT
;
6017 sctx
->waiting_dir_moves
= RB_ROOT
;
6018 sctx
->orphan_dirs
= RB_ROOT
;
6020 sctx
->clone_roots
= vzalloc(sizeof(struct clone_root
) *
6021 (arg
->clone_sources_count
+ 1));
6022 if (!sctx
->clone_roots
) {
6027 if (arg
->clone_sources_count
) {
6028 clone_sources_tmp
= vmalloc(arg
->clone_sources_count
*
6029 sizeof(*arg
->clone_sources
));
6030 if (!clone_sources_tmp
) {
6035 ret
= copy_from_user(clone_sources_tmp
, arg
->clone_sources
,
6036 arg
->clone_sources_count
*
6037 sizeof(*arg
->clone_sources
));
6043 for (i
= 0; i
< arg
->clone_sources_count
; i
++) {
6044 key
.objectid
= clone_sources_tmp
[i
];
6045 key
.type
= BTRFS_ROOT_ITEM_KEY
;
6046 key
.offset
= (u64
)-1;
6048 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
6050 clone_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
6051 if (IS_ERR(clone_root
)) {
6052 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6053 ret
= PTR_ERR(clone_root
);
6056 spin_lock(&clone_root
->root_item_lock
);
6057 if (!btrfs_root_readonly(clone_root
) ||
6058 btrfs_root_dead(clone_root
)) {
6059 spin_unlock(&clone_root
->root_item_lock
);
6060 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6064 clone_root
->send_in_progress
++;
6065 spin_unlock(&clone_root
->root_item_lock
);
6066 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6068 sctx
->clone_roots
[i
].root
= clone_root
;
6069 clone_sources_to_rollback
= i
+ 1;
6071 vfree(clone_sources_tmp
);
6072 clone_sources_tmp
= NULL
;
6075 if (arg
->parent_root
) {
6076 key
.objectid
= arg
->parent_root
;
6077 key
.type
= BTRFS_ROOT_ITEM_KEY
;
6078 key
.offset
= (u64
)-1;
6080 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
6082 sctx
->parent_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
6083 if (IS_ERR(sctx
->parent_root
)) {
6084 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6085 ret
= PTR_ERR(sctx
->parent_root
);
6089 spin_lock(&sctx
->parent_root
->root_item_lock
);
6090 sctx
->parent_root
->send_in_progress
++;
6091 if (!btrfs_root_readonly(sctx
->parent_root
) ||
6092 btrfs_root_dead(sctx
->parent_root
)) {
6093 spin_unlock(&sctx
->parent_root
->root_item_lock
);
6094 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6098 spin_unlock(&sctx
->parent_root
->root_item_lock
);
6100 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
6104 * Clones from send_root are allowed, but only if the clone source
6105 * is behind the current send position. This is checked while searching
6106 * for possible clone sources.
6108 sctx
->clone_roots
[sctx
->clone_roots_cnt
++].root
= sctx
->send_root
;
6110 /* We do a bsearch later */
6111 sort(sctx
->clone_roots
, sctx
->clone_roots_cnt
,
6112 sizeof(*sctx
->clone_roots
), __clone_root_cmp_sort
,
6114 sort_clone_roots
= 1;
6116 ret
= ensure_commit_roots_uptodate(sctx
);
6120 current
->journal_info
= BTRFS_SEND_TRANS_STUB
;
6121 ret
= send_subvol(sctx
);
6122 current
->journal_info
= NULL
;
6126 if (!(sctx
->flags
& BTRFS_SEND_FLAG_OMIT_END_CMD
)) {
6127 ret
= begin_cmd(sctx
, BTRFS_SEND_C_END
);
6130 ret
= send_cmd(sctx
);
6136 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
));
6137 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->pending_dir_moves
)) {
6139 struct pending_dir_move
*pm
;
6141 n
= rb_first(&sctx
->pending_dir_moves
);
6142 pm
= rb_entry(n
, struct pending_dir_move
, node
);
6143 while (!list_empty(&pm
->list
)) {
6144 struct pending_dir_move
*pm2
;
6146 pm2
= list_first_entry(&pm
->list
,
6147 struct pending_dir_move
, list
);
6148 free_pending_move(sctx
, pm2
);
6150 free_pending_move(sctx
, pm
);
6153 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
));
6154 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->waiting_dir_moves
)) {
6156 struct waiting_dir_move
*dm
;
6158 n
= rb_first(&sctx
->waiting_dir_moves
);
6159 dm
= rb_entry(n
, struct waiting_dir_move
, node
);
6160 rb_erase(&dm
->node
, &sctx
->waiting_dir_moves
);
6164 WARN_ON(sctx
&& !ret
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
));
6165 while (sctx
&& !RB_EMPTY_ROOT(&sctx
->orphan_dirs
)) {
6167 struct orphan_dir_info
*odi
;
6169 n
= rb_first(&sctx
->orphan_dirs
);
6170 odi
= rb_entry(n
, struct orphan_dir_info
, node
);
6171 free_orphan_dir_info(sctx
, odi
);
6174 if (sort_clone_roots
) {
6175 for (i
= 0; i
< sctx
->clone_roots_cnt
; i
++)
6176 btrfs_root_dec_send_in_progress(
6177 sctx
->clone_roots
[i
].root
);
6179 for (i
= 0; sctx
&& i
< clone_sources_to_rollback
; i
++)
6180 btrfs_root_dec_send_in_progress(
6181 sctx
->clone_roots
[i
].root
);
6183 btrfs_root_dec_send_in_progress(send_root
);
6185 if (sctx
&& !IS_ERR_OR_NULL(sctx
->parent_root
))
6186 btrfs_root_dec_send_in_progress(sctx
->parent_root
);
6189 vfree(clone_sources_tmp
);
6192 if (sctx
->send_filp
)
6193 fput(sctx
->send_filp
);
6195 vfree(sctx
->clone_roots
);
6196 vfree(sctx
->send_buf
);
6197 vfree(sctx
->read_buf
);
6199 name_cache_free(sctx
);