2 * Copyright (C) 2007 Oracle. 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.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "ref-cache.h"
35 #define PENDING_EXTENT_INSERT 0
36 #define PENDING_EXTENT_DELETE 1
37 #define PENDING_BACKREF_UPDATE 2
39 struct pending_extent_op
{
48 struct list_head list
;
52 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
53 struct btrfs_root
*root
, u64 parent
,
54 u64 root_objectid
, u64 ref_generation
,
55 u64 owner
, struct btrfs_key
*ins
,
57 static int update_reserved_extents(struct btrfs_root
*root
,
58 u64 bytenr
, u64 num
, int reserve
);
59 static int update_block_group(struct btrfs_trans_handle
*trans
,
60 struct btrfs_root
*root
,
61 u64 bytenr
, u64 num_bytes
, int alloc
,
63 static noinline
int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
64 struct btrfs_root
*root
,
65 u64 bytenr
, u64 num_bytes
, u64 parent
,
66 u64 root_objectid
, u64 ref_generation
,
67 u64 owner_objectid
, int pin
,
70 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
71 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
72 u64 flags
, int force
);
74 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
76 return (cache
->flags
& bits
) == bits
;
80 * this adds the block group to the fs_info rb tree for the block group
83 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
84 struct btrfs_block_group_cache
*block_group
)
87 struct rb_node
*parent
= NULL
;
88 struct btrfs_block_group_cache
*cache
;
90 spin_lock(&info
->block_group_cache_lock
);
91 p
= &info
->block_group_cache_tree
.rb_node
;
95 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
97 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
99 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
102 spin_unlock(&info
->block_group_cache_lock
);
107 rb_link_node(&block_group
->cache_node
, parent
, p
);
108 rb_insert_color(&block_group
->cache_node
,
109 &info
->block_group_cache_tree
);
110 spin_unlock(&info
->block_group_cache_lock
);
116 * This will return the block group at or after bytenr if contains is 0, else
117 * it will return the block group that contains the bytenr
119 static struct btrfs_block_group_cache
*
120 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
123 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
127 spin_lock(&info
->block_group_cache_lock
);
128 n
= info
->block_group_cache_tree
.rb_node
;
131 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
133 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
134 start
= cache
->key
.objectid
;
136 if (bytenr
< start
) {
137 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
140 } else if (bytenr
> start
) {
141 if (contains
&& bytenr
<= end
) {
152 atomic_inc(&ret
->count
);
153 spin_unlock(&info
->block_group_cache_lock
);
159 * this is only called by cache_block_group, since we could have freed extents
160 * we need to check the pinned_extents for any extents that can't be used yet
161 * since their free space will be released as soon as the transaction commits.
163 static int add_new_free_space(struct btrfs_block_group_cache
*block_group
,
164 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
166 u64 extent_start
, extent_end
, size
;
169 mutex_lock(&info
->pinned_mutex
);
170 while (start
< end
) {
171 ret
= find_first_extent_bit(&info
->pinned_extents
, start
,
172 &extent_start
, &extent_end
,
177 if (extent_start
== start
) {
178 start
= extent_end
+ 1;
179 } else if (extent_start
> start
&& extent_start
< end
) {
180 size
= extent_start
- start
;
181 ret
= btrfs_add_free_space(block_group
, start
,
184 start
= extent_end
+ 1;
192 ret
= btrfs_add_free_space(block_group
, start
, size
);
195 mutex_unlock(&info
->pinned_mutex
);
200 static int remove_sb_from_cache(struct btrfs_root
*root
,
201 struct btrfs_block_group_cache
*cache
)
208 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
209 bytenr
= btrfs_sb_offset(i
);
210 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
211 cache
->key
.objectid
, bytenr
, 0,
212 &logical
, &nr
, &stripe_len
);
215 btrfs_remove_free_space(cache
, logical
[nr
],
223 static int cache_block_group(struct btrfs_root
*root
,
224 struct btrfs_block_group_cache
*block_group
)
226 struct btrfs_path
*path
;
228 struct btrfs_key key
;
229 struct extent_buffer
*leaf
;
236 root
= root
->fs_info
->extent_root
;
238 if (block_group
->cached
)
241 path
= btrfs_alloc_path();
247 * we get into deadlocks with paths held by callers of this function.
248 * since the alloc_mutex is protecting things right now, just
249 * skip the locking here
251 path
->skip_locking
= 1;
252 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
255 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
256 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
261 leaf
= path
->nodes
[0];
262 slot
= path
->slots
[0];
263 if (slot
>= btrfs_header_nritems(leaf
)) {
264 ret
= btrfs_next_leaf(root
, path
);
272 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
273 if (key
.objectid
< block_group
->key
.objectid
)
276 if (key
.objectid
>= block_group
->key
.objectid
+
277 block_group
->key
.offset
)
280 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
281 add_new_free_space(block_group
, root
->fs_info
, last
,
284 last
= key
.objectid
+ key
.offset
;
290 add_new_free_space(block_group
, root
->fs_info
, last
,
291 block_group
->key
.objectid
+
292 block_group
->key
.offset
);
294 remove_sb_from_cache(root
, block_group
);
295 block_group
->cached
= 1;
298 btrfs_free_path(path
);
303 * return the block group that starts at or after bytenr
305 static struct btrfs_block_group_cache
*
306 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
308 struct btrfs_block_group_cache
*cache
;
310 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
316 * return the block group that contains teh given bytenr
318 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
319 struct btrfs_fs_info
*info
,
322 struct btrfs_block_group_cache
*cache
;
324 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
329 static inline void put_block_group(struct btrfs_block_group_cache
*cache
)
331 if (atomic_dec_and_test(&cache
->count
))
335 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
338 struct list_head
*head
= &info
->space_info
;
339 struct btrfs_space_info
*found
;
342 list_for_each_entry_rcu(found
, head
, list
) {
343 if (found
->flags
== flags
) {
353 * after adding space to the filesystem, we need to clear the full flags
354 * on all the space infos.
356 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
358 struct list_head
*head
= &info
->space_info
;
359 struct btrfs_space_info
*found
;
362 list_for_each_entry_rcu(found
, head
, list
)
367 static u64
div_factor(u64 num
, int factor
)
376 u64
btrfs_find_block_group(struct btrfs_root
*root
,
377 u64 search_start
, u64 search_hint
, int owner
)
379 struct btrfs_block_group_cache
*cache
;
381 u64 last
= max(search_hint
, search_start
);
388 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
392 spin_lock(&cache
->lock
);
393 last
= cache
->key
.objectid
+ cache
->key
.offset
;
394 used
= btrfs_block_group_used(&cache
->item
);
396 if ((full_search
|| !cache
->ro
) &&
397 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
398 if (used
+ cache
->pinned
+ cache
->reserved
<
399 div_factor(cache
->key
.offset
, factor
)) {
400 group_start
= cache
->key
.objectid
;
401 spin_unlock(&cache
->lock
);
402 put_block_group(cache
);
406 spin_unlock(&cache
->lock
);
407 put_block_group(cache
);
415 if (!full_search
&& factor
< 10) {
425 /* simple helper to search for an existing extent at a given offset */
426 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
429 struct btrfs_key key
;
430 struct btrfs_path
*path
;
432 path
= btrfs_alloc_path();
434 key
.objectid
= start
;
436 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
437 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
439 btrfs_free_path(path
);
444 * Back reference rules. Back refs have three main goals:
446 * 1) differentiate between all holders of references to an extent so that
447 * when a reference is dropped we can make sure it was a valid reference
448 * before freeing the extent.
450 * 2) Provide enough information to quickly find the holders of an extent
451 * if we notice a given block is corrupted or bad.
453 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
454 * maintenance. This is actually the same as #2, but with a slightly
455 * different use case.
457 * File extents can be referenced by:
459 * - multiple snapshots, subvolumes, or different generations in one subvol
460 * - different files inside a single subvolume
461 * - different offsets inside a file (bookend extents in file.c)
463 * The extent ref structure has fields for:
465 * - Objectid of the subvolume root
466 * - Generation number of the tree holding the reference
467 * - objectid of the file holding the reference
468 * - number of references holding by parent node (alway 1 for tree blocks)
470 * Btree leaf may hold multiple references to a file extent. In most cases,
471 * these references are from same file and the corresponding offsets inside
472 * the file are close together.
474 * When a file extent is allocated the fields are filled in:
475 * (root_key.objectid, trans->transid, inode objectid, 1)
477 * When a leaf is cow'd new references are added for every file extent found
478 * in the leaf. It looks similar to the create case, but trans->transid will
479 * be different when the block is cow'd.
481 * (root_key.objectid, trans->transid, inode objectid,
482 * number of references in the leaf)
484 * When a file extent is removed either during snapshot deletion or
485 * file truncation, we find the corresponding back reference and check
486 * the following fields:
488 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
491 * Btree extents can be referenced by:
493 * - Different subvolumes
494 * - Different generations of the same subvolume
496 * When a tree block is created, back references are inserted:
498 * (root->root_key.objectid, trans->transid, level, 1)
500 * When a tree block is cow'd, new back references are added for all the
501 * blocks it points to. If the tree block isn't in reference counted root,
502 * the old back references are removed. These new back references are of
503 * the form (trans->transid will have increased since creation):
505 * (root->root_key.objectid, trans->transid, level, 1)
507 * When a backref is in deleting, the following fields are checked:
509 * if backref was for a tree root:
510 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
512 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
514 * Back Reference Key composing:
516 * The key objectid corresponds to the first byte in the extent, the key
517 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
518 * byte of parent extent. If a extent is tree root, the key offset is set
519 * to the key objectid.
522 static noinline
int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
523 struct btrfs_root
*root
,
524 struct btrfs_path
*path
,
525 u64 bytenr
, u64 parent
,
526 u64 ref_root
, u64 ref_generation
,
527 u64 owner_objectid
, int del
)
529 struct btrfs_key key
;
530 struct btrfs_extent_ref
*ref
;
531 struct extent_buffer
*leaf
;
535 key
.objectid
= bytenr
;
536 key
.type
= BTRFS_EXTENT_REF_KEY
;
539 ret
= btrfs_search_slot(trans
, root
, &key
, path
, del
? -1 : 0, 1);
547 leaf
= path
->nodes
[0];
548 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
549 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
550 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
551 btrfs_ref_generation(leaf
, ref
) != ref_generation
||
552 (ref_objectid
!= owner_objectid
&&
553 ref_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
)) {
563 static noinline
int insert_extent_backref(struct btrfs_trans_handle
*trans
,
564 struct btrfs_root
*root
,
565 struct btrfs_path
*path
,
566 u64 bytenr
, u64 parent
,
567 u64 ref_root
, u64 ref_generation
,
571 struct btrfs_key key
;
572 struct extent_buffer
*leaf
;
573 struct btrfs_extent_ref
*ref
;
577 key
.objectid
= bytenr
;
578 key
.type
= BTRFS_EXTENT_REF_KEY
;
581 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(*ref
));
583 leaf
= path
->nodes
[0];
584 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
585 struct btrfs_extent_ref
);
586 btrfs_set_ref_root(leaf
, ref
, ref_root
);
587 btrfs_set_ref_generation(leaf
, ref
, ref_generation
);
588 btrfs_set_ref_objectid(leaf
, ref
, owner_objectid
);
589 btrfs_set_ref_num_refs(leaf
, ref
, refs_to_add
);
590 } else if (ret
== -EEXIST
) {
593 BUG_ON(owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
);
594 leaf
= path
->nodes
[0];
595 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
596 struct btrfs_extent_ref
);
597 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
598 btrfs_ref_generation(leaf
, ref
) != ref_generation
) {
604 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
605 BUG_ON(num_refs
== 0);
606 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
+ refs_to_add
);
608 existing_owner
= btrfs_ref_objectid(leaf
, ref
);
609 if (existing_owner
!= owner_objectid
&&
610 existing_owner
!= BTRFS_MULTIPLE_OBJECTIDS
) {
611 btrfs_set_ref_objectid(leaf
, ref
,
612 BTRFS_MULTIPLE_OBJECTIDS
);
618 btrfs_unlock_up_safe(path
, 1);
619 btrfs_mark_buffer_dirty(path
->nodes
[0]);
621 btrfs_release_path(root
, path
);
625 static noinline
int remove_extent_backref(struct btrfs_trans_handle
*trans
,
626 struct btrfs_root
*root
,
627 struct btrfs_path
*path
,
630 struct extent_buffer
*leaf
;
631 struct btrfs_extent_ref
*ref
;
635 leaf
= path
->nodes
[0];
636 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
637 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
638 BUG_ON(num_refs
< refs_to_drop
);
639 num_refs
-= refs_to_drop
;
641 ret
= btrfs_del_item(trans
, root
, path
);
643 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
);
644 btrfs_mark_buffer_dirty(leaf
);
646 btrfs_release_path(root
, path
);
650 #ifdef BIO_RW_DISCARD
651 static void btrfs_issue_discard(struct block_device
*bdev
,
654 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
);
658 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
661 #ifdef BIO_RW_DISCARD
663 u64 map_length
= num_bytes
;
664 struct btrfs_multi_bio
*multi
= NULL
;
666 /* Tell the block device(s) that the sectors can be discarded */
667 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
668 bytenr
, &map_length
, &multi
, 0);
670 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
673 if (map_length
> num_bytes
)
674 map_length
= num_bytes
;
676 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
677 btrfs_issue_discard(stripe
->dev
->bdev
,
690 static int __btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
691 struct btrfs_root
*root
, u64 bytenr
,
693 u64 orig_parent
, u64 parent
,
694 u64 orig_root
, u64 ref_root
,
695 u64 orig_generation
, u64 ref_generation
,
699 int pin
= owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
;
701 ret
= btrfs_update_delayed_ref(trans
, bytenr
, num_bytes
,
702 orig_parent
, parent
, orig_root
,
703 ref_root
, orig_generation
,
704 ref_generation
, owner_objectid
, pin
);
709 int btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
710 struct btrfs_root
*root
, u64 bytenr
,
711 u64 num_bytes
, u64 orig_parent
, u64 parent
,
712 u64 ref_root
, u64 ref_generation
,
716 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
717 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
720 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
, num_bytes
,
721 orig_parent
, parent
, ref_root
,
722 ref_root
, ref_generation
,
723 ref_generation
, owner_objectid
);
726 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
727 struct btrfs_root
*root
, u64 bytenr
,
729 u64 orig_parent
, u64 parent
,
730 u64 orig_root
, u64 ref_root
,
731 u64 orig_generation
, u64 ref_generation
,
736 ret
= btrfs_add_delayed_ref(trans
, bytenr
, num_bytes
, parent
, ref_root
,
737 ref_generation
, owner_objectid
,
738 BTRFS_ADD_DELAYED_REF
, 0);
743 static noinline_for_stack
int add_extent_ref(struct btrfs_trans_handle
*trans
,
744 struct btrfs_root
*root
, u64 bytenr
,
745 u64 num_bytes
, u64 parent
, u64 ref_root
,
746 u64 ref_generation
, u64 owner_objectid
,
749 struct btrfs_path
*path
;
751 struct btrfs_key key
;
752 struct extent_buffer
*l
;
753 struct btrfs_extent_item
*item
;
756 path
= btrfs_alloc_path();
761 path
->leave_spinning
= 1;
762 key
.objectid
= bytenr
;
763 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
764 key
.offset
= num_bytes
;
766 /* first find the extent item and update its reference count */
767 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
770 btrfs_set_path_blocking(path
);
776 btrfs_free_path(path
);
781 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
782 if (key
.objectid
!= bytenr
) {
783 btrfs_print_leaf(root
->fs_info
->extent_root
, path
->nodes
[0]);
784 printk(KERN_ERR
"btrfs wanted %llu found %llu\n",
785 (unsigned long long)bytenr
,
786 (unsigned long long)key
.objectid
);
789 BUG_ON(key
.type
!= BTRFS_EXTENT_ITEM_KEY
);
791 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
793 refs
= btrfs_extent_refs(l
, item
);
794 btrfs_set_extent_refs(l
, item
, refs
+ refs_to_add
);
795 btrfs_unlock_up_safe(path
, 1);
797 btrfs_mark_buffer_dirty(path
->nodes
[0]);
799 btrfs_release_path(root
->fs_info
->extent_root
, path
);
802 path
->leave_spinning
= 1;
804 /* now insert the actual backref */
805 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
806 path
, bytenr
, parent
,
807 ref_root
, ref_generation
,
808 owner_objectid
, refs_to_add
);
810 btrfs_free_path(path
);
814 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
815 struct btrfs_root
*root
,
816 u64 bytenr
, u64 num_bytes
, u64 parent
,
817 u64 ref_root
, u64 ref_generation
,
821 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
822 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
825 ret
= __btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
, 0, parent
,
826 0, ref_root
, 0, ref_generation
,
831 static int drop_delayed_ref(struct btrfs_trans_handle
*trans
,
832 struct btrfs_root
*root
,
833 struct btrfs_delayed_ref_node
*node
)
836 struct btrfs_delayed_ref
*ref
= btrfs_delayed_node_to_ref(node
);
838 BUG_ON(node
->ref_mod
== 0);
839 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
, node
->num_bytes
,
840 node
->parent
, ref
->root
, ref
->generation
,
841 ref
->owner_objectid
, ref
->pin
, node
->ref_mod
);
846 /* helper function to actually process a single delayed ref entry */
847 static noinline
int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
848 struct btrfs_root
*root
,
849 struct btrfs_delayed_ref_node
*node
,
853 struct btrfs_delayed_ref
*ref
;
855 if (node
->parent
== (u64
)-1) {
856 struct btrfs_delayed_ref_head
*head
;
858 * we've hit the end of the chain and we were supposed
859 * to insert this extent into the tree. But, it got
860 * deleted before we ever needed to insert it, so all
861 * we have to do is clean up the accounting
863 if (insert_reserved
) {
864 update_reserved_extents(root
, node
->bytenr
,
867 head
= btrfs_delayed_node_to_head(node
);
868 mutex_unlock(&head
->mutex
);
872 ref
= btrfs_delayed_node_to_ref(node
);
873 if (ref
->action
== BTRFS_ADD_DELAYED_REF
) {
874 if (insert_reserved
) {
875 struct btrfs_key ins
;
877 ins
.objectid
= node
->bytenr
;
878 ins
.offset
= node
->num_bytes
;
879 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
881 /* record the full extent allocation */
882 ret
= __btrfs_alloc_reserved_extent(trans
, root
,
883 node
->parent
, ref
->root
,
884 ref
->generation
, ref
->owner_objectid
,
885 &ins
, node
->ref_mod
);
886 update_reserved_extents(root
, node
->bytenr
,
889 /* just add one backref */
890 ret
= add_extent_ref(trans
, root
, node
->bytenr
,
892 node
->parent
, ref
->root
, ref
->generation
,
893 ref
->owner_objectid
, node
->ref_mod
);
896 } else if (ref
->action
== BTRFS_DROP_DELAYED_REF
) {
897 WARN_ON(insert_reserved
);
898 ret
= drop_delayed_ref(trans
, root
, node
);
903 static noinline
struct btrfs_delayed_ref_node
*
904 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
906 struct rb_node
*node
;
907 struct btrfs_delayed_ref_node
*ref
;
908 int action
= BTRFS_ADD_DELAYED_REF
;
911 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
912 * this prevents ref count from going down to zero when
913 * there still are pending delayed ref.
915 node
= rb_prev(&head
->node
.rb_node
);
919 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
921 if (ref
->bytenr
!= head
->node
.bytenr
)
923 if (btrfs_delayed_node_to_ref(ref
)->action
== action
)
925 node
= rb_prev(node
);
927 if (action
== BTRFS_ADD_DELAYED_REF
) {
928 action
= BTRFS_DROP_DELAYED_REF
;
934 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
935 struct btrfs_root
*root
,
936 struct list_head
*cluster
)
938 struct btrfs_delayed_ref_root
*delayed_refs
;
939 struct btrfs_delayed_ref_node
*ref
;
940 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
943 int must_insert_reserved
= 0;
945 delayed_refs
= &trans
->transaction
->delayed_refs
;
948 /* pick a new head ref from the cluster list */
949 if (list_empty(cluster
))
952 locked_ref
= list_entry(cluster
->next
,
953 struct btrfs_delayed_ref_head
, cluster
);
955 /* grab the lock that says we are going to process
956 * all the refs for this head */
957 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
960 * we may have dropped the spin lock to get the head
961 * mutex lock, and that might have given someone else
962 * time to free the head. If that's true, it has been
963 * removed from our list and we can move on.
965 if (ret
== -EAGAIN
) {
973 * record the must insert reserved flag before we
974 * drop the spin lock.
976 must_insert_reserved
= locked_ref
->must_insert_reserved
;
977 locked_ref
->must_insert_reserved
= 0;
980 * locked_ref is the head node, so we have to go one
981 * node back for any delayed ref updates
983 ref
= select_delayed_ref(locked_ref
);
985 /* All delayed refs have been processed, Go ahead
986 * and send the head node to run_one_delayed_ref,
987 * so that any accounting fixes can happen
989 ref
= &locked_ref
->node
;
990 list_del_init(&locked_ref
->cluster
);
995 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
996 delayed_refs
->num_entries
--;
997 spin_unlock(&delayed_refs
->lock
);
999 ret
= run_one_delayed_ref(trans
, root
, ref
,
1000 must_insert_reserved
);
1002 btrfs_put_delayed_ref(ref
);
1006 spin_lock(&delayed_refs
->lock
);
1012 * this starts processing the delayed reference count updates and
1013 * extent insertions we have queued up so far. count can be
1014 * 0, which means to process everything in the tree at the start
1015 * of the run (but not newly added entries), or it can be some target
1016 * number you'd like to process.
1018 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
1019 struct btrfs_root
*root
, unsigned long count
)
1021 struct rb_node
*node
;
1022 struct btrfs_delayed_ref_root
*delayed_refs
;
1023 struct btrfs_delayed_ref_node
*ref
;
1024 struct list_head cluster
;
1026 int run_all
= count
== (unsigned long)-1;
1029 if (root
== root
->fs_info
->extent_root
)
1030 root
= root
->fs_info
->tree_root
;
1032 delayed_refs
= &trans
->transaction
->delayed_refs
;
1033 INIT_LIST_HEAD(&cluster
);
1035 spin_lock(&delayed_refs
->lock
);
1037 count
= delayed_refs
->num_entries
* 2;
1041 if (!(run_all
|| run_most
) &&
1042 delayed_refs
->num_heads_ready
< 64)
1046 * go find something we can process in the rbtree. We start at
1047 * the beginning of the tree, and then build a cluster
1048 * of refs to process starting at the first one we are able to
1051 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
1052 delayed_refs
->run_delayed_start
);
1056 ret
= run_clustered_refs(trans
, root
, &cluster
);
1059 count
-= min_t(unsigned long, ret
, count
);
1066 node
= rb_first(&delayed_refs
->root
);
1069 count
= (unsigned long)-1;
1072 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
1074 if (btrfs_delayed_ref_is_head(ref
)) {
1075 struct btrfs_delayed_ref_head
*head
;
1077 head
= btrfs_delayed_node_to_head(ref
);
1078 atomic_inc(&ref
->refs
);
1080 spin_unlock(&delayed_refs
->lock
);
1081 mutex_lock(&head
->mutex
);
1082 mutex_unlock(&head
->mutex
);
1084 btrfs_put_delayed_ref(ref
);
1088 node
= rb_next(node
);
1090 spin_unlock(&delayed_refs
->lock
);
1091 schedule_timeout(1);
1095 spin_unlock(&delayed_refs
->lock
);
1099 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
1100 struct btrfs_root
*root
, u64 objectid
, u64 bytenr
)
1102 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1103 struct btrfs_path
*path
;
1104 struct extent_buffer
*leaf
;
1105 struct btrfs_extent_ref
*ref_item
;
1106 struct btrfs_key key
;
1107 struct btrfs_key found_key
;
1113 key
.objectid
= bytenr
;
1114 key
.offset
= (u64
)-1;
1115 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1117 path
= btrfs_alloc_path();
1118 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
1124 if (path
->slots
[0] == 0)
1128 leaf
= path
->nodes
[0];
1129 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1131 if (found_key
.objectid
!= bytenr
||
1132 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
1135 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
1137 leaf
= path
->nodes
[0];
1138 nritems
= btrfs_header_nritems(leaf
);
1139 if (path
->slots
[0] >= nritems
) {
1140 ret
= btrfs_next_leaf(extent_root
, path
);
1147 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1148 if (found_key
.objectid
!= bytenr
)
1151 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
1156 ref_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1157 struct btrfs_extent_ref
);
1158 ref_root
= btrfs_ref_root(leaf
, ref_item
);
1159 if ((ref_root
!= root
->root_key
.objectid
&&
1160 ref_root
!= BTRFS_TREE_LOG_OBJECTID
) ||
1161 objectid
!= btrfs_ref_objectid(leaf
, ref_item
)) {
1165 if (btrfs_ref_generation(leaf
, ref_item
) <= last_snapshot
) {
1174 btrfs_free_path(path
);
1178 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1179 struct extent_buffer
*buf
, u32 nr_extents
)
1181 struct btrfs_key key
;
1182 struct btrfs_file_extent_item
*fi
;
1190 if (!root
->ref_cows
)
1193 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
1195 root_gen
= root
->root_key
.offset
;
1198 root_gen
= trans
->transid
- 1;
1201 level
= btrfs_header_level(buf
);
1202 nritems
= btrfs_header_nritems(buf
);
1205 struct btrfs_leaf_ref
*ref
;
1206 struct btrfs_extent_info
*info
;
1208 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
1214 ref
->root_gen
= root_gen
;
1215 ref
->bytenr
= buf
->start
;
1216 ref
->owner
= btrfs_header_owner(buf
);
1217 ref
->generation
= btrfs_header_generation(buf
);
1218 ref
->nritems
= nr_extents
;
1219 info
= ref
->extents
;
1221 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
1223 btrfs_item_key_to_cpu(buf
, &key
, i
);
1224 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1226 fi
= btrfs_item_ptr(buf
, i
,
1227 struct btrfs_file_extent_item
);
1228 if (btrfs_file_extent_type(buf
, fi
) ==
1229 BTRFS_FILE_EXTENT_INLINE
)
1231 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1232 if (disk_bytenr
== 0)
1235 info
->bytenr
= disk_bytenr
;
1237 btrfs_file_extent_disk_num_bytes(buf
, fi
);
1238 info
->objectid
= key
.objectid
;
1239 info
->offset
= key
.offset
;
1243 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
1244 if (ret
== -EEXIST
&& shared
) {
1245 struct btrfs_leaf_ref
*old
;
1246 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
1248 btrfs_remove_leaf_ref(root
, old
);
1249 btrfs_free_leaf_ref(root
, old
);
1250 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
1253 btrfs_free_leaf_ref(root
, ref
);
1259 /* when a block goes through cow, we update the reference counts of
1260 * everything that block points to. The internal pointers of the block
1261 * can be in just about any order, and it is likely to have clusters of
1262 * things that are close together and clusters of things that are not.
1264 * To help reduce the seeks that come with updating all of these reference
1265 * counts, sort them by byte number before actual updates are done.
1267 * struct refsort is used to match byte number to slot in the btree block.
1268 * we sort based on the byte number and then use the slot to actually
1271 * struct refsort is smaller than strcut btrfs_item and smaller than
1272 * struct btrfs_key_ptr. Since we're currently limited to the page size
1273 * for a btree block, there's no way for a kmalloc of refsorts for a
1274 * single node to be bigger than a page.
1282 * for passing into sort()
1284 static int refsort_cmp(const void *a_void
, const void *b_void
)
1286 const struct refsort
*a
= a_void
;
1287 const struct refsort
*b
= b_void
;
1289 if (a
->bytenr
< b
->bytenr
)
1291 if (a
->bytenr
> b
->bytenr
)
1297 noinline
int btrfs_inc_ref(struct btrfs_trans_handle
*trans
,
1298 struct btrfs_root
*root
,
1299 struct extent_buffer
*orig_buf
,
1300 struct extent_buffer
*buf
, u32
*nr_extents
)
1306 u64 orig_generation
;
1307 struct refsort
*sorted
;
1309 u32 nr_file_extents
= 0;
1310 struct btrfs_key key
;
1311 struct btrfs_file_extent_item
*fi
;
1318 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
1319 u64
, u64
, u64
, u64
, u64
, u64
, u64
, u64
, u64
);
1321 ref_root
= btrfs_header_owner(buf
);
1322 ref_generation
= btrfs_header_generation(buf
);
1323 orig_root
= btrfs_header_owner(orig_buf
);
1324 orig_generation
= btrfs_header_generation(orig_buf
);
1326 nritems
= btrfs_header_nritems(buf
);
1327 level
= btrfs_header_level(buf
);
1329 sorted
= kmalloc(sizeof(struct refsort
) * nritems
, GFP_NOFS
);
1332 if (root
->ref_cows
) {
1333 process_func
= __btrfs_inc_extent_ref
;
1336 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1339 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1341 process_func
= __btrfs_update_extent_ref
;
1345 * we make two passes through the items. In the first pass we
1346 * only record the byte number and slot. Then we sort based on
1347 * byte number and do the actual work based on the sorted results
1349 for (i
= 0; i
< nritems
; i
++) {
1352 btrfs_item_key_to_cpu(buf
, &key
, i
);
1353 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1355 fi
= btrfs_item_ptr(buf
, i
,
1356 struct btrfs_file_extent_item
);
1357 if (btrfs_file_extent_type(buf
, fi
) ==
1358 BTRFS_FILE_EXTENT_INLINE
)
1360 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1365 sorted
[refi
].bytenr
= bytenr
;
1366 sorted
[refi
].slot
= i
;
1369 bytenr
= btrfs_node_blockptr(buf
, i
);
1370 sorted
[refi
].bytenr
= bytenr
;
1371 sorted
[refi
].slot
= i
;
1376 * if refi == 0, we didn't actually put anything into the sorted
1377 * array and we're done
1382 sort(sorted
, refi
, sizeof(struct refsort
), refsort_cmp
, NULL
);
1384 for (i
= 0; i
< refi
; i
++) {
1386 slot
= sorted
[i
].slot
;
1387 bytenr
= sorted
[i
].bytenr
;
1390 btrfs_item_key_to_cpu(buf
, &key
, slot
);
1391 fi
= btrfs_item_ptr(buf
, slot
,
1392 struct btrfs_file_extent_item
);
1394 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1398 ret
= process_func(trans
, root
, bytenr
,
1399 btrfs_file_extent_disk_num_bytes(buf
, fi
),
1400 orig_buf
->start
, buf
->start
,
1401 orig_root
, ref_root
,
1402 orig_generation
, ref_generation
,
1411 ret
= process_func(trans
, root
, bytenr
, buf
->len
,
1412 orig_buf
->start
, buf
->start
,
1413 orig_root
, ref_root
,
1414 orig_generation
, ref_generation
,
1427 *nr_extents
= nr_file_extents
;
1429 *nr_extents
= nritems
;
1438 int btrfs_update_ref(struct btrfs_trans_handle
*trans
,
1439 struct btrfs_root
*root
, struct extent_buffer
*orig_buf
,
1440 struct extent_buffer
*buf
, int start_slot
, int nr
)
1447 u64 orig_generation
;
1448 struct btrfs_key key
;
1449 struct btrfs_file_extent_item
*fi
;
1455 BUG_ON(start_slot
< 0);
1456 BUG_ON(start_slot
+ nr
> btrfs_header_nritems(buf
));
1458 ref_root
= btrfs_header_owner(buf
);
1459 ref_generation
= btrfs_header_generation(buf
);
1460 orig_root
= btrfs_header_owner(orig_buf
);
1461 orig_generation
= btrfs_header_generation(orig_buf
);
1462 level
= btrfs_header_level(buf
);
1464 if (!root
->ref_cows
) {
1466 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1469 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1473 for (i
= 0, slot
= start_slot
; i
< nr
; i
++, slot
++) {
1476 btrfs_item_key_to_cpu(buf
, &key
, slot
);
1477 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1479 fi
= btrfs_item_ptr(buf
, slot
,
1480 struct btrfs_file_extent_item
);
1481 if (btrfs_file_extent_type(buf
, fi
) ==
1482 BTRFS_FILE_EXTENT_INLINE
)
1484 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1487 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1488 btrfs_file_extent_disk_num_bytes(buf
, fi
),
1489 orig_buf
->start
, buf
->start
,
1490 orig_root
, ref_root
, orig_generation
,
1491 ref_generation
, key
.objectid
);
1495 bytenr
= btrfs_node_blockptr(buf
, slot
);
1496 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1497 buf
->len
, orig_buf
->start
,
1498 buf
->start
, orig_root
, ref_root
,
1499 orig_generation
, ref_generation
,
1511 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
1512 struct btrfs_root
*root
,
1513 struct btrfs_path
*path
,
1514 struct btrfs_block_group_cache
*cache
)
1517 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1519 struct extent_buffer
*leaf
;
1521 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
1526 leaf
= path
->nodes
[0];
1527 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1528 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
1529 btrfs_mark_buffer_dirty(leaf
);
1530 btrfs_release_path(extent_root
, path
);
1538 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
1539 struct btrfs_root
*root
)
1541 struct btrfs_block_group_cache
*cache
, *entry
;
1545 struct btrfs_path
*path
;
1548 path
= btrfs_alloc_path();
1554 spin_lock(&root
->fs_info
->block_group_cache_lock
);
1555 for (n
= rb_first(&root
->fs_info
->block_group_cache_tree
);
1556 n
; n
= rb_next(n
)) {
1557 entry
= rb_entry(n
, struct btrfs_block_group_cache
,
1564 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
1570 last
+= cache
->key
.offset
;
1572 err
= write_one_cache_group(trans
, root
,
1575 * if we fail to write the cache group, we want
1576 * to keep it marked dirty in hopes that a later
1584 btrfs_free_path(path
);
1588 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
1590 struct btrfs_block_group_cache
*block_group
;
1593 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
1594 if (!block_group
|| block_group
->ro
)
1597 put_block_group(block_group
);
1601 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
1602 u64 total_bytes
, u64 bytes_used
,
1603 struct btrfs_space_info
**space_info
)
1605 struct btrfs_space_info
*found
;
1607 found
= __find_space_info(info
, flags
);
1609 spin_lock(&found
->lock
);
1610 found
->total_bytes
+= total_bytes
;
1611 found
->bytes_used
+= bytes_used
;
1613 spin_unlock(&found
->lock
);
1614 *space_info
= found
;
1617 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
1621 INIT_LIST_HEAD(&found
->block_groups
);
1622 init_rwsem(&found
->groups_sem
);
1623 spin_lock_init(&found
->lock
);
1624 found
->flags
= flags
;
1625 found
->total_bytes
= total_bytes
;
1626 found
->bytes_used
= bytes_used
;
1627 found
->bytes_pinned
= 0;
1628 found
->bytes_reserved
= 0;
1629 found
->bytes_readonly
= 0;
1630 found
->bytes_delalloc
= 0;
1632 found
->force_alloc
= 0;
1633 *space_info
= found
;
1634 list_add_rcu(&found
->list
, &info
->space_info
);
1638 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1640 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
1641 BTRFS_BLOCK_GROUP_RAID1
|
1642 BTRFS_BLOCK_GROUP_RAID10
|
1643 BTRFS_BLOCK_GROUP_DUP
);
1645 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1646 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1647 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1648 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1649 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1650 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1654 static void set_block_group_readonly(struct btrfs_block_group_cache
*cache
)
1656 spin_lock(&cache
->space_info
->lock
);
1657 spin_lock(&cache
->lock
);
1659 cache
->space_info
->bytes_readonly
+= cache
->key
.offset
-
1660 btrfs_block_group_used(&cache
->item
);
1663 spin_unlock(&cache
->lock
);
1664 spin_unlock(&cache
->space_info
->lock
);
1667 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
1669 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
1671 if (num_devices
== 1)
1672 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
1673 if (num_devices
< 4)
1674 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
1676 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
1677 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
1678 BTRFS_BLOCK_GROUP_RAID10
))) {
1679 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
1682 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
1683 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
1684 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
1687 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
1688 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
1689 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
1690 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
1691 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
1695 static u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, u64 data
)
1697 struct btrfs_fs_info
*info
= root
->fs_info
;
1701 alloc_profile
= info
->avail_data_alloc_bits
&
1702 info
->data_alloc_profile
;
1703 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
1704 } else if (root
== root
->fs_info
->chunk_root
) {
1705 alloc_profile
= info
->avail_system_alloc_bits
&
1706 info
->system_alloc_profile
;
1707 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
1709 alloc_profile
= info
->avail_metadata_alloc_bits
&
1710 info
->metadata_alloc_profile
;
1711 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
1714 return btrfs_reduce_alloc_profile(root
, data
);
1717 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
1721 alloc_target
= btrfs_get_alloc_profile(root
, 1);
1722 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
1727 * for now this just makes sure we have at least 5% of our metadata space free
1730 int btrfs_check_metadata_free_space(struct btrfs_root
*root
)
1732 struct btrfs_fs_info
*info
= root
->fs_info
;
1733 struct btrfs_space_info
*meta_sinfo
;
1734 u64 alloc_target
, thresh
;
1735 int committed
= 0, ret
;
1737 /* get the space info for where the metadata will live */
1738 alloc_target
= btrfs_get_alloc_profile(root
, 0);
1739 meta_sinfo
= __find_space_info(info
, alloc_target
);
1742 spin_lock(&meta_sinfo
->lock
);
1743 if (!meta_sinfo
->full
)
1744 thresh
= meta_sinfo
->total_bytes
* 80;
1746 thresh
= meta_sinfo
->total_bytes
* 95;
1748 do_div(thresh
, 100);
1750 if (meta_sinfo
->bytes_used
+ meta_sinfo
->bytes_reserved
+
1751 meta_sinfo
->bytes_pinned
+ meta_sinfo
->bytes_readonly
> thresh
) {
1752 struct btrfs_trans_handle
*trans
;
1753 if (!meta_sinfo
->full
) {
1754 meta_sinfo
->force_alloc
= 1;
1755 spin_unlock(&meta_sinfo
->lock
);
1757 trans
= btrfs_start_transaction(root
, 1);
1761 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
1762 2 * 1024 * 1024, alloc_target
, 0);
1763 btrfs_end_transaction(trans
, root
);
1766 spin_unlock(&meta_sinfo
->lock
);
1770 trans
= btrfs_join_transaction(root
, 1);
1773 ret
= btrfs_commit_transaction(trans
, root
);
1780 spin_unlock(&meta_sinfo
->lock
);
1786 * This will check the space that the inode allocates from to make sure we have
1787 * enough space for bytes.
1789 int btrfs_check_data_free_space(struct btrfs_root
*root
, struct inode
*inode
,
1792 struct btrfs_space_info
*data_sinfo
;
1793 int ret
= 0, committed
= 0;
1795 /* make sure bytes are sectorsize aligned */
1796 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
1798 data_sinfo
= BTRFS_I(inode
)->space_info
;
1800 /* make sure we have enough space to handle the data first */
1801 spin_lock(&data_sinfo
->lock
);
1802 if (data_sinfo
->total_bytes
- data_sinfo
->bytes_used
-
1803 data_sinfo
->bytes_delalloc
- data_sinfo
->bytes_reserved
-
1804 data_sinfo
->bytes_pinned
- data_sinfo
->bytes_readonly
-
1805 data_sinfo
->bytes_may_use
< bytes
) {
1806 struct btrfs_trans_handle
*trans
;
1809 * if we don't have enough free bytes in this space then we need
1810 * to alloc a new chunk.
1812 if (!data_sinfo
->full
) {
1815 data_sinfo
->force_alloc
= 1;
1816 spin_unlock(&data_sinfo
->lock
);
1818 alloc_target
= btrfs_get_alloc_profile(root
, 1);
1819 trans
= btrfs_start_transaction(root
, 1);
1823 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
1824 bytes
+ 2 * 1024 * 1024,
1826 btrfs_end_transaction(trans
, root
);
1831 spin_unlock(&data_sinfo
->lock
);
1833 /* commit the current transaction and try again */
1836 trans
= btrfs_join_transaction(root
, 1);
1839 ret
= btrfs_commit_transaction(trans
, root
);
1845 printk(KERN_ERR
"no space left, need %llu, %llu delalloc bytes"
1846 ", %llu bytes_used, %llu bytes_reserved, "
1847 "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
1848 "%llu total\n", bytes
, data_sinfo
->bytes_delalloc
,
1849 data_sinfo
->bytes_used
, data_sinfo
->bytes_reserved
,
1850 data_sinfo
->bytes_pinned
, data_sinfo
->bytes_readonly
,
1851 data_sinfo
->bytes_may_use
, data_sinfo
->total_bytes
);
1854 data_sinfo
->bytes_may_use
+= bytes
;
1855 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
1856 spin_unlock(&data_sinfo
->lock
);
1858 return btrfs_check_metadata_free_space(root
);
1862 * if there was an error for whatever reason after calling
1863 * btrfs_check_data_free_space, call this so we can cleanup the counters.
1865 void btrfs_free_reserved_data_space(struct btrfs_root
*root
,
1866 struct inode
*inode
, u64 bytes
)
1868 struct btrfs_space_info
*data_sinfo
;
1870 /* make sure bytes are sectorsize aligned */
1871 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
1873 data_sinfo
= BTRFS_I(inode
)->space_info
;
1874 spin_lock(&data_sinfo
->lock
);
1875 data_sinfo
->bytes_may_use
-= bytes
;
1876 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
1877 spin_unlock(&data_sinfo
->lock
);
1880 /* called when we are adding a delalloc extent to the inode's io_tree */
1881 void btrfs_delalloc_reserve_space(struct btrfs_root
*root
, struct inode
*inode
,
1884 struct btrfs_space_info
*data_sinfo
;
1886 /* get the space info for where this inode will be storing its data */
1887 data_sinfo
= BTRFS_I(inode
)->space_info
;
1889 /* make sure we have enough space to handle the data first */
1890 spin_lock(&data_sinfo
->lock
);
1891 data_sinfo
->bytes_delalloc
+= bytes
;
1894 * we are adding a delalloc extent without calling
1895 * btrfs_check_data_free_space first. This happens on a weird
1896 * writepage condition, but shouldn't hurt our accounting
1898 if (unlikely(bytes
> BTRFS_I(inode
)->reserved_bytes
)) {
1899 data_sinfo
->bytes_may_use
-= BTRFS_I(inode
)->reserved_bytes
;
1900 BTRFS_I(inode
)->reserved_bytes
= 0;
1902 data_sinfo
->bytes_may_use
-= bytes
;
1903 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
1906 spin_unlock(&data_sinfo
->lock
);
1909 /* called when we are clearing an delalloc extent from the inode's io_tree */
1910 void btrfs_delalloc_free_space(struct btrfs_root
*root
, struct inode
*inode
,
1913 struct btrfs_space_info
*info
;
1915 info
= BTRFS_I(inode
)->space_info
;
1917 spin_lock(&info
->lock
);
1918 info
->bytes_delalloc
-= bytes
;
1919 spin_unlock(&info
->lock
);
1922 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
1923 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
1924 u64 flags
, int force
)
1926 struct btrfs_space_info
*space_info
;
1930 mutex_lock(&extent_root
->fs_info
->chunk_mutex
);
1932 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
1934 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
1936 ret
= update_space_info(extent_root
->fs_info
, flags
,
1940 BUG_ON(!space_info
);
1942 spin_lock(&space_info
->lock
);
1943 if (space_info
->force_alloc
) {
1945 space_info
->force_alloc
= 0;
1947 if (space_info
->full
) {
1948 spin_unlock(&space_info
->lock
);
1952 thresh
= space_info
->total_bytes
- space_info
->bytes_readonly
;
1953 thresh
= div_factor(thresh
, 6);
1955 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
1956 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
) {
1957 spin_unlock(&space_info
->lock
);
1960 spin_unlock(&space_info
->lock
);
1962 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
1964 space_info
->full
= 1;
1966 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
1970 static int update_block_group(struct btrfs_trans_handle
*trans
,
1971 struct btrfs_root
*root
,
1972 u64 bytenr
, u64 num_bytes
, int alloc
,
1975 struct btrfs_block_group_cache
*cache
;
1976 struct btrfs_fs_info
*info
= root
->fs_info
;
1977 u64 total
= num_bytes
;
1982 cache
= btrfs_lookup_block_group(info
, bytenr
);
1985 byte_in_group
= bytenr
- cache
->key
.objectid
;
1986 WARN_ON(byte_in_group
> cache
->key
.offset
);
1988 spin_lock(&cache
->space_info
->lock
);
1989 spin_lock(&cache
->lock
);
1991 old_val
= btrfs_block_group_used(&cache
->item
);
1992 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1994 old_val
+= num_bytes
;
1995 cache
->space_info
->bytes_used
+= num_bytes
;
1997 cache
->space_info
->bytes_readonly
-= num_bytes
;
1998 btrfs_set_block_group_used(&cache
->item
, old_val
);
1999 spin_unlock(&cache
->lock
);
2000 spin_unlock(&cache
->space_info
->lock
);
2002 old_val
-= num_bytes
;
2003 cache
->space_info
->bytes_used
-= num_bytes
;
2005 cache
->space_info
->bytes_readonly
+= num_bytes
;
2006 btrfs_set_block_group_used(&cache
->item
, old_val
);
2007 spin_unlock(&cache
->lock
);
2008 spin_unlock(&cache
->space_info
->lock
);
2012 ret
= btrfs_discard_extent(root
, bytenr
,
2016 ret
= btrfs_add_free_space(cache
, bytenr
,
2021 put_block_group(cache
);
2023 bytenr
+= num_bytes
;
2028 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
2030 struct btrfs_block_group_cache
*cache
;
2033 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
2037 bytenr
= cache
->key
.objectid
;
2038 put_block_group(cache
);
2043 int btrfs_update_pinned_extents(struct btrfs_root
*root
,
2044 u64 bytenr
, u64 num
, int pin
)
2047 struct btrfs_block_group_cache
*cache
;
2048 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2050 WARN_ON(!mutex_is_locked(&root
->fs_info
->pinned_mutex
));
2052 set_extent_dirty(&fs_info
->pinned_extents
,
2053 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
2055 clear_extent_dirty(&fs_info
->pinned_extents
,
2056 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
2058 mutex_unlock(&root
->fs_info
->pinned_mutex
);
2061 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
2063 len
= min(num
, cache
->key
.offset
-
2064 (bytenr
- cache
->key
.objectid
));
2066 spin_lock(&cache
->space_info
->lock
);
2067 spin_lock(&cache
->lock
);
2068 cache
->pinned
+= len
;
2069 cache
->space_info
->bytes_pinned
+= len
;
2070 spin_unlock(&cache
->lock
);
2071 spin_unlock(&cache
->space_info
->lock
);
2072 fs_info
->total_pinned
+= len
;
2074 spin_lock(&cache
->space_info
->lock
);
2075 spin_lock(&cache
->lock
);
2076 cache
->pinned
-= len
;
2077 cache
->space_info
->bytes_pinned
-= len
;
2078 spin_unlock(&cache
->lock
);
2079 spin_unlock(&cache
->space_info
->lock
);
2080 fs_info
->total_pinned
-= len
;
2082 btrfs_add_free_space(cache
, bytenr
, len
);
2084 put_block_group(cache
);
2091 static int update_reserved_extents(struct btrfs_root
*root
,
2092 u64 bytenr
, u64 num
, int reserve
)
2095 struct btrfs_block_group_cache
*cache
;
2096 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2099 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
2101 len
= min(num
, cache
->key
.offset
-
2102 (bytenr
- cache
->key
.objectid
));
2104 spin_lock(&cache
->space_info
->lock
);
2105 spin_lock(&cache
->lock
);
2107 cache
->reserved
+= len
;
2108 cache
->space_info
->bytes_reserved
+= len
;
2110 cache
->reserved
-= len
;
2111 cache
->space_info
->bytes_reserved
-= len
;
2113 spin_unlock(&cache
->lock
);
2114 spin_unlock(&cache
->space_info
->lock
);
2115 put_block_group(cache
);
2122 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
2127 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
2130 mutex_lock(&root
->fs_info
->pinned_mutex
);
2132 ret
= find_first_extent_bit(pinned_extents
, last
,
2133 &start
, &end
, EXTENT_DIRTY
);
2136 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
2139 mutex_unlock(&root
->fs_info
->pinned_mutex
);
2143 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
2144 struct btrfs_root
*root
,
2145 struct extent_io_tree
*unpin
)
2152 mutex_lock(&root
->fs_info
->pinned_mutex
);
2153 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
2158 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
2160 /* unlocks the pinned mutex */
2161 btrfs_update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
2162 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
2166 mutex_unlock(&root
->fs_info
->pinned_mutex
);
2170 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
2171 struct btrfs_root
*root
,
2172 struct btrfs_path
*path
,
2173 u64 bytenr
, u64 num_bytes
, int is_data
,
2174 struct extent_buffer
**must_clean
)
2177 struct extent_buffer
*buf
;
2182 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
2186 /* we can reuse a block if it hasn't been written
2187 * and it is from this transaction. We can't
2188 * reuse anything from the tree log root because
2189 * it has tiny sub-transactions.
2191 if (btrfs_buffer_uptodate(buf
, 0) &&
2192 btrfs_try_tree_lock(buf
)) {
2193 u64 header_owner
= btrfs_header_owner(buf
);
2194 u64 header_transid
= btrfs_header_generation(buf
);
2195 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
2196 header_owner
!= BTRFS_TREE_RELOC_OBJECTID
&&
2197 header_owner
!= BTRFS_DATA_RELOC_TREE_OBJECTID
&&
2198 header_transid
== trans
->transid
&&
2199 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
2203 btrfs_tree_unlock(buf
);
2205 free_extent_buffer(buf
);
2207 btrfs_set_path_blocking(path
);
2208 mutex_lock(&root
->fs_info
->pinned_mutex
);
2209 /* unlocks the pinned mutex */
2210 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
2217 * remove an extent from the root, returns 0 on success
2219 static int __free_extent(struct btrfs_trans_handle
*trans
,
2220 struct btrfs_root
*root
,
2221 u64 bytenr
, u64 num_bytes
, u64 parent
,
2222 u64 root_objectid
, u64 ref_generation
,
2223 u64 owner_objectid
, int pin
, int mark_free
,
2226 struct btrfs_path
*path
;
2227 struct btrfs_key key
;
2228 struct btrfs_fs_info
*info
= root
->fs_info
;
2229 struct btrfs_root
*extent_root
= info
->extent_root
;
2230 struct extent_buffer
*leaf
;
2232 int extent_slot
= 0;
2233 int found_extent
= 0;
2235 struct btrfs_extent_item
*ei
;
2238 key
.objectid
= bytenr
;
2239 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
2240 key
.offset
= num_bytes
;
2241 path
= btrfs_alloc_path();
2246 path
->leave_spinning
= 1;
2247 ret
= lookup_extent_backref(trans
, extent_root
, path
,
2248 bytenr
, parent
, root_objectid
,
2249 ref_generation
, owner_objectid
, 1);
2251 struct btrfs_key found_key
;
2252 extent_slot
= path
->slots
[0];
2253 while (extent_slot
> 0) {
2255 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2257 if (found_key
.objectid
!= bytenr
)
2259 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2260 found_key
.offset
== num_bytes
) {
2264 if (path
->slots
[0] - extent_slot
> 5)
2267 if (!found_extent
) {
2268 ret
= remove_extent_backref(trans
, extent_root
, path
,
2271 btrfs_release_path(extent_root
, path
);
2272 path
->leave_spinning
= 1;
2273 ret
= btrfs_search_slot(trans
, extent_root
,
2276 printk(KERN_ERR
"umm, got %d back from search"
2277 ", was looking for %llu\n", ret
,
2278 (unsigned long long)bytenr
);
2279 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
2282 extent_slot
= path
->slots
[0];
2285 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
2287 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
2288 "parent %llu root %llu gen %llu owner %llu\n",
2289 (unsigned long long)bytenr
,
2290 (unsigned long long)parent
,
2291 (unsigned long long)root_objectid
,
2292 (unsigned long long)ref_generation
,
2293 (unsigned long long)owner_objectid
);
2296 leaf
= path
->nodes
[0];
2297 ei
= btrfs_item_ptr(leaf
, extent_slot
,
2298 struct btrfs_extent_item
);
2299 refs
= btrfs_extent_refs(leaf
, ei
);
2302 * we're not allowed to delete the extent item if there
2303 * are other delayed ref updates pending
2306 BUG_ON(refs
< refs_to_drop
);
2307 refs
-= refs_to_drop
;
2308 btrfs_set_extent_refs(leaf
, ei
, refs
);
2309 btrfs_mark_buffer_dirty(leaf
);
2311 if (refs
== 0 && found_extent
&&
2312 path
->slots
[0] == extent_slot
+ 1) {
2313 struct btrfs_extent_ref
*ref
;
2314 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
2315 struct btrfs_extent_ref
);
2316 BUG_ON(btrfs_ref_num_refs(leaf
, ref
) != refs_to_drop
);
2317 /* if the back ref and the extent are next to each other
2318 * they get deleted below in one shot
2320 path
->slots
[0] = extent_slot
;
2322 } else if (found_extent
) {
2323 /* otherwise delete the extent back ref */
2324 ret
= remove_extent_backref(trans
, extent_root
, path
,
2327 /* if refs are 0, we need to setup the path for deletion */
2329 btrfs_release_path(extent_root
, path
);
2330 path
->leave_spinning
= 1;
2331 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
2340 struct extent_buffer
*must_clean
= NULL
;
2343 ret
= pin_down_bytes(trans
, root
, path
,
2345 owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
,
2352 /* block accounting for super block */
2353 spin_lock(&info
->delalloc_lock
);
2354 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
2355 btrfs_set_super_bytes_used(&info
->super_copy
,
2356 super_used
- num_bytes
);
2358 /* block accounting for root item */
2359 root_used
= btrfs_root_used(&root
->root_item
);
2360 btrfs_set_root_used(&root
->root_item
,
2361 root_used
- num_bytes
);
2362 spin_unlock(&info
->delalloc_lock
);
2365 * it is going to be very rare for someone to be waiting
2366 * on the block we're freeing. del_items might need to
2367 * schedule, so rather than get fancy, just force it
2371 btrfs_set_lock_blocking(must_clean
);
2373 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
2376 btrfs_release_path(extent_root
, path
);
2379 clean_tree_block(NULL
, root
, must_clean
);
2380 btrfs_tree_unlock(must_clean
);
2381 free_extent_buffer(must_clean
);
2384 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
2385 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
2388 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
2389 bytenr
>> PAGE_CACHE_SHIFT
,
2390 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
2393 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
2397 btrfs_free_path(path
);
2402 * remove an extent from the root, returns 0 on success
2404 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2405 struct btrfs_root
*root
,
2406 u64 bytenr
, u64 num_bytes
, u64 parent
,
2407 u64 root_objectid
, u64 ref_generation
,
2408 u64 owner_objectid
, int pin
,
2411 WARN_ON(num_bytes
< root
->sectorsize
);
2414 * if metadata always pin
2415 * if data pin when any transaction has committed this
2417 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
||
2418 ref_generation
!= trans
->transid
)
2421 if (ref_generation
!= trans
->transid
)
2424 return __free_extent(trans
, root
, bytenr
, num_bytes
, parent
,
2425 root_objectid
, ref_generation
,
2426 owner_objectid
, pin
, pin
== 0, refs_to_drop
);
2430 * when we free an extent, it is possible (and likely) that we free the last
2431 * delayed ref for that extent as well. This searches the delayed ref tree for
2432 * a given extent, and if there are no other delayed refs to be processed, it
2433 * removes it from the tree.
2435 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
2436 struct btrfs_root
*root
, u64 bytenr
)
2438 struct btrfs_delayed_ref_head
*head
;
2439 struct btrfs_delayed_ref_root
*delayed_refs
;
2440 struct btrfs_delayed_ref_node
*ref
;
2441 struct rb_node
*node
;
2444 delayed_refs
= &trans
->transaction
->delayed_refs
;
2445 spin_lock(&delayed_refs
->lock
);
2446 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2450 node
= rb_prev(&head
->node
.rb_node
);
2454 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2456 /* there are still entries for this ref, we can't drop it */
2457 if (ref
->bytenr
== bytenr
)
2461 * waiting for the lock here would deadlock. If someone else has it
2462 * locked they are already in the process of dropping it anyway
2464 if (!mutex_trylock(&head
->mutex
))
2468 * at this point we have a head with no other entries. Go
2469 * ahead and process it.
2471 head
->node
.in_tree
= 0;
2472 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
2474 delayed_refs
->num_entries
--;
2477 * we don't take a ref on the node because we're removing it from the
2478 * tree, so we just steal the ref the tree was holding.
2480 delayed_refs
->num_heads
--;
2481 if (list_empty(&head
->cluster
))
2482 delayed_refs
->num_heads_ready
--;
2484 list_del_init(&head
->cluster
);
2485 spin_unlock(&delayed_refs
->lock
);
2487 ret
= run_one_delayed_ref(trans
, root
->fs_info
->tree_root
,
2488 &head
->node
, head
->must_insert_reserved
);
2490 btrfs_put_delayed_ref(&head
->node
);
2493 spin_unlock(&delayed_refs
->lock
);
2497 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2498 struct btrfs_root
*root
,
2499 u64 bytenr
, u64 num_bytes
, u64 parent
,
2500 u64 root_objectid
, u64 ref_generation
,
2501 u64 owner_objectid
, int pin
)
2506 * tree log blocks never actually go into the extent allocation
2507 * tree, just update pinning info and exit early.
2509 * data extents referenced by the tree log do need to have
2510 * their reference counts bumped.
2512 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
&&
2513 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
2514 mutex_lock(&root
->fs_info
->pinned_mutex
);
2516 /* unlocks the pinned mutex */
2517 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
2518 update_reserved_extents(root
, bytenr
, num_bytes
, 0);
2521 ret
= btrfs_add_delayed_ref(trans
, bytenr
, num_bytes
, parent
,
2522 root_objectid
, ref_generation
,
2524 BTRFS_DROP_DELAYED_REF
, 1);
2526 ret
= check_ref_cleanup(trans
, root
, bytenr
);
2532 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
2534 u64 mask
= ((u64
)root
->stripesize
- 1);
2535 u64 ret
= (val
+ mask
) & ~mask
;
2540 * walks the btree of allocated extents and find a hole of a given size.
2541 * The key ins is changed to record the hole:
2542 * ins->objectid == block start
2543 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2544 * ins->offset == number of blocks
2545 * Any available blocks before search_start are skipped.
2547 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
2548 struct btrfs_root
*orig_root
,
2549 u64 num_bytes
, u64 empty_size
,
2550 u64 search_start
, u64 search_end
,
2551 u64 hint_byte
, struct btrfs_key
*ins
,
2552 u64 exclude_start
, u64 exclude_nr
,
2556 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
2557 u64 total_needed
= num_bytes
;
2558 u64
*last_ptr
= NULL
;
2559 u64 last_wanted
= 0;
2560 struct btrfs_block_group_cache
*block_group
= NULL
;
2561 int chunk_alloc_done
= 0;
2562 int empty_cluster
= 2 * 1024 * 1024;
2563 int allowed_chunk_alloc
= 0;
2564 struct list_head
*head
= NULL
, *cur
= NULL
;
2567 struct btrfs_space_info
*space_info
;
2569 WARN_ON(num_bytes
< root
->sectorsize
);
2570 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
2574 if (orig_root
->ref_cows
|| empty_size
)
2575 allowed_chunk_alloc
= 1;
2577 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
2578 last_ptr
= &root
->fs_info
->last_alloc
;
2579 if (!btrfs_test_opt(root
, SSD
))
2580 empty_cluster
= 64 * 1024;
2583 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
))
2584 last_ptr
= &root
->fs_info
->last_data_alloc
;
2588 hint_byte
= *last_ptr
;
2589 last_wanted
= *last_ptr
;
2591 empty_size
+= empty_cluster
;
2595 search_start
= max(search_start
, first_logical_byte(root
, 0));
2596 search_start
= max(search_start
, hint_byte
);
2598 if (last_wanted
&& search_start
!= last_wanted
) {
2600 empty_size
+= empty_cluster
;
2603 total_needed
+= empty_size
;
2604 block_group
= btrfs_lookup_block_group(root
->fs_info
, search_start
);
2606 block_group
= btrfs_lookup_first_block_group(root
->fs_info
,
2608 space_info
= __find_space_info(root
->fs_info
, data
);
2610 down_read(&space_info
->groups_sem
);
2612 struct btrfs_free_space
*free_space
;
2614 * the only way this happens if our hint points to a block
2615 * group thats not of the proper type, while looping this
2616 * should never happen
2622 goto new_group_no_lock
;
2624 if (unlikely(!block_group
->cached
)) {
2625 mutex_lock(&block_group
->cache_mutex
);
2626 ret
= cache_block_group(root
, block_group
);
2627 mutex_unlock(&block_group
->cache_mutex
);
2632 mutex_lock(&block_group
->alloc_mutex
);
2633 if (unlikely(!block_group_bits(block_group
, data
)))
2636 if (unlikely(block_group
->ro
))
2639 free_space
= btrfs_find_free_space(block_group
, search_start
,
2642 u64 start
= block_group
->key
.objectid
;
2643 u64 end
= block_group
->key
.objectid
+
2644 block_group
->key
.offset
;
2646 search_start
= stripe_align(root
, free_space
->offset
);
2648 /* move on to the next group */
2649 if (search_start
+ num_bytes
>= search_end
)
2652 /* move on to the next group */
2653 if (search_start
+ num_bytes
> end
)
2656 if (last_wanted
&& search_start
!= last_wanted
) {
2657 total_needed
+= empty_cluster
;
2658 empty_size
+= empty_cluster
;
2661 * if search_start is still in this block group
2662 * then we just re-search this block group
2664 if (search_start
>= start
&&
2665 search_start
< end
) {
2666 mutex_unlock(&block_group
->alloc_mutex
);
2670 /* else we go to the next block group */
2674 if (exclude_nr
> 0 &&
2675 (search_start
+ num_bytes
> exclude_start
&&
2676 search_start
< exclude_start
+ exclude_nr
)) {
2677 search_start
= exclude_start
+ exclude_nr
;
2679 * if search_start is still in this block group
2680 * then we just re-search this block group
2682 if (search_start
>= start
&&
2683 search_start
< end
) {
2684 mutex_unlock(&block_group
->alloc_mutex
);
2689 /* else we go to the next block group */
2693 ins
->objectid
= search_start
;
2694 ins
->offset
= num_bytes
;
2696 btrfs_remove_free_space_lock(block_group
, search_start
,
2698 /* we are all good, lets return */
2699 mutex_unlock(&block_group
->alloc_mutex
);
2703 mutex_unlock(&block_group
->alloc_mutex
);
2704 put_block_group(block_group
);
2707 /* don't try to compare new allocations against the
2708 * last allocation any more
2713 * Here's how this works.
2714 * loop == 0: we were searching a block group via a hint
2715 * and didn't find anything, so we start at
2716 * the head of the block groups and keep searching
2717 * loop == 1: we're searching through all of the block groups
2718 * if we hit the head again we have searched
2719 * all of the block groups for this space and we
2720 * need to try and allocate, if we cant error out.
2721 * loop == 2: we allocated more space and are looping through
2722 * all of the block groups again.
2725 head
= &space_info
->block_groups
;
2728 } else if (loop
== 1 && cur
== head
) {
2731 /* at this point we give up on the empty_size
2732 * allocations and just try to allocate the min
2735 * The extra_loop field was set if an empty_size
2736 * allocation was attempted above, and if this
2737 * is try we need to try the loop again without
2738 * the additional empty_size.
2740 total_needed
-= empty_size
;
2742 keep_going
= extra_loop
;
2745 if (allowed_chunk_alloc
&& !chunk_alloc_done
) {
2746 up_read(&space_info
->groups_sem
);
2747 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
2748 2 * 1024 * 1024, data
, 1);
2749 down_read(&space_info
->groups_sem
);
2752 head
= &space_info
->block_groups
;
2754 * we've allocated a new chunk, keep
2758 chunk_alloc_done
= 1;
2759 } else if (!allowed_chunk_alloc
) {
2760 space_info
->force_alloc
= 1;
2769 } else if (cur
== head
) {
2773 block_group
= list_entry(cur
, struct btrfs_block_group_cache
,
2775 atomic_inc(&block_group
->count
);
2777 search_start
= block_group
->key
.objectid
;
2781 /* we found what we needed */
2782 if (ins
->objectid
) {
2783 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
2784 trans
->block_group
= block_group
->key
.objectid
;
2787 *last_ptr
= ins
->objectid
+ ins
->offset
;
2790 printk(KERN_ERR
"btrfs searching for %llu bytes, "
2791 "num_bytes %llu, loop %d, allowed_alloc %d\n",
2792 (unsigned long long)total_needed
,
2793 (unsigned long long)num_bytes
,
2794 loop
, allowed_chunk_alloc
);
2798 put_block_group(block_group
);
2800 up_read(&space_info
->groups_sem
);
2804 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
)
2806 struct btrfs_block_group_cache
*cache
;
2808 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
2809 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
2810 info
->bytes_pinned
- info
->bytes_reserved
),
2811 (info
->full
) ? "" : "not ");
2812 printk(KERN_INFO
"space_info total=%llu, pinned=%llu, delalloc=%llu,"
2813 " may_use=%llu, used=%llu\n", info
->total_bytes
,
2814 info
->bytes_pinned
, info
->bytes_delalloc
, info
->bytes_may_use
,
2817 down_read(&info
->groups_sem
);
2818 list_for_each_entry(cache
, &info
->block_groups
, list
) {
2819 spin_lock(&cache
->lock
);
2820 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
2821 "%llu pinned %llu reserved\n",
2822 (unsigned long long)cache
->key
.objectid
,
2823 (unsigned long long)cache
->key
.offset
,
2824 (unsigned long long)btrfs_block_group_used(&cache
->item
),
2825 (unsigned long long)cache
->pinned
,
2826 (unsigned long long)cache
->reserved
);
2827 btrfs_dump_free_space(cache
, bytes
);
2828 spin_unlock(&cache
->lock
);
2830 up_read(&info
->groups_sem
);
2833 static int __btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2834 struct btrfs_root
*root
,
2835 u64 num_bytes
, u64 min_alloc_size
,
2836 u64 empty_size
, u64 hint_byte
,
2837 u64 search_end
, struct btrfs_key
*ins
,
2841 u64 search_start
= 0;
2842 struct btrfs_fs_info
*info
= root
->fs_info
;
2844 data
= btrfs_get_alloc_profile(root
, data
);
2847 * the only place that sets empty_size is btrfs_realloc_node, which
2848 * is not called recursively on allocations
2850 if (empty_size
|| root
->ref_cows
) {
2851 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
2852 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2854 BTRFS_BLOCK_GROUP_METADATA
|
2855 (info
->metadata_alloc_profile
&
2856 info
->avail_metadata_alloc_bits
), 0);
2858 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2859 num_bytes
+ 2 * 1024 * 1024, data
, 0);
2862 WARN_ON(num_bytes
< root
->sectorsize
);
2863 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
2864 search_start
, search_end
, hint_byte
, ins
,
2865 trans
->alloc_exclude_start
,
2866 trans
->alloc_exclude_nr
, data
);
2868 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
2869 num_bytes
= num_bytes
>> 1;
2870 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
2871 num_bytes
= max(num_bytes
, min_alloc_size
);
2872 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2873 num_bytes
, data
, 1);
2877 struct btrfs_space_info
*sinfo
;
2879 sinfo
= __find_space_info(root
->fs_info
, data
);
2880 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
2881 "wanted %llu\n", (unsigned long long)data
,
2882 (unsigned long long)num_bytes
);
2883 dump_space_info(sinfo
, num_bytes
);
2890 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
2892 struct btrfs_block_group_cache
*cache
;
2895 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2897 printk(KERN_ERR
"Unable to find block group for %llu\n",
2898 (unsigned long long)start
);
2902 ret
= btrfs_discard_extent(root
, start
, len
);
2904 btrfs_add_free_space(cache
, start
, len
);
2905 put_block_group(cache
);
2906 update_reserved_extents(root
, start
, len
, 0);
2911 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
2912 struct btrfs_root
*root
,
2913 u64 num_bytes
, u64 min_alloc_size
,
2914 u64 empty_size
, u64 hint_byte
,
2915 u64 search_end
, struct btrfs_key
*ins
,
2919 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, min_alloc_size
,
2920 empty_size
, hint_byte
, search_end
, ins
,
2922 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
2926 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
2927 struct btrfs_root
*root
, u64 parent
,
2928 u64 root_objectid
, u64 ref_generation
,
2929 u64 owner
, struct btrfs_key
*ins
,
2935 u64 num_bytes
= ins
->offset
;
2937 struct btrfs_fs_info
*info
= root
->fs_info
;
2938 struct btrfs_root
*extent_root
= info
->extent_root
;
2939 struct btrfs_extent_item
*extent_item
;
2940 struct btrfs_extent_ref
*ref
;
2941 struct btrfs_path
*path
;
2942 struct btrfs_key keys
[2];
2945 parent
= ins
->objectid
;
2947 /* block accounting for super block */
2948 spin_lock(&info
->delalloc_lock
);
2949 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
2950 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
2952 /* block accounting for root item */
2953 root_used
= btrfs_root_used(&root
->root_item
);
2954 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
2955 spin_unlock(&info
->delalloc_lock
);
2957 memcpy(&keys
[0], ins
, sizeof(*ins
));
2958 keys
[1].objectid
= ins
->objectid
;
2959 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
2960 keys
[1].offset
= parent
;
2961 sizes
[0] = sizeof(*extent_item
);
2962 sizes
[1] = sizeof(*ref
);
2964 path
= btrfs_alloc_path();
2967 path
->leave_spinning
= 1;
2968 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
2972 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2973 struct btrfs_extent_item
);
2974 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, ref_mod
);
2975 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
2976 struct btrfs_extent_ref
);
2978 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
2979 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
2980 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
2981 btrfs_set_ref_num_refs(path
->nodes
[0], ref
, ref_mod
);
2983 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2985 trans
->alloc_exclude_start
= 0;
2986 trans
->alloc_exclude_nr
= 0;
2987 btrfs_free_path(path
);
2992 ret
= update_block_group(trans
, root
, ins
->objectid
,
2995 printk(KERN_ERR
"btrfs update block group failed for %llu "
2996 "%llu\n", (unsigned long long)ins
->objectid
,
2997 (unsigned long long)ins
->offset
);
3004 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
3005 struct btrfs_root
*root
, u64 parent
,
3006 u64 root_objectid
, u64 ref_generation
,
3007 u64 owner
, struct btrfs_key
*ins
)
3011 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
)
3014 ret
= btrfs_add_delayed_ref(trans
, ins
->objectid
,
3015 ins
->offset
, parent
, root_objectid
,
3016 ref_generation
, owner
,
3017 BTRFS_ADD_DELAYED_EXTENT
, 0);
3023 * this is used by the tree logging recovery code. It records that
3024 * an extent has been allocated and makes sure to clear the free
3025 * space cache bits as well
3027 int btrfs_alloc_logged_extent(struct btrfs_trans_handle
*trans
,
3028 struct btrfs_root
*root
, u64 parent
,
3029 u64 root_objectid
, u64 ref_generation
,
3030 u64 owner
, struct btrfs_key
*ins
)
3033 struct btrfs_block_group_cache
*block_group
;
3035 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
3036 mutex_lock(&block_group
->cache_mutex
);
3037 cache_block_group(root
, block_group
);
3038 mutex_unlock(&block_group
->cache_mutex
);
3040 ret
= btrfs_remove_free_space(block_group
, ins
->objectid
,
3043 put_block_group(block_group
);
3044 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
, root_objectid
,
3045 ref_generation
, owner
, ins
, 1);
3050 * finds a free extent and does all the dirty work required for allocation
3051 * returns the key for the extent through ins, and a tree buffer for
3052 * the first block of the extent through buf.
3054 * returns 0 if everything worked, non-zero otherwise.
3056 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
3057 struct btrfs_root
*root
,
3058 u64 num_bytes
, u64 parent
, u64 min_alloc_size
,
3059 u64 root_objectid
, u64 ref_generation
,
3060 u64 owner_objectid
, u64 empty_size
, u64 hint_byte
,
3061 u64 search_end
, struct btrfs_key
*ins
, u64 data
)
3064 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
,
3065 min_alloc_size
, empty_size
, hint_byte
,
3066 search_end
, ins
, data
);
3068 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
3069 ret
= btrfs_add_delayed_ref(trans
, ins
->objectid
,
3070 ins
->offset
, parent
, root_objectid
,
3071 ref_generation
, owner_objectid
,
3072 BTRFS_ADD_DELAYED_EXTENT
, 0);
3075 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
3079 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
3080 struct btrfs_root
*root
,
3081 u64 bytenr
, u32 blocksize
,
3084 struct extent_buffer
*buf
;
3086 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
3088 return ERR_PTR(-ENOMEM
);
3089 btrfs_set_header_generation(buf
, trans
->transid
);
3090 btrfs_set_buffer_lockdep_class(buf
, level
);
3091 btrfs_tree_lock(buf
);
3092 clean_tree_block(trans
, root
, buf
);
3094 btrfs_set_lock_blocking(buf
);
3095 btrfs_set_buffer_uptodate(buf
);
3097 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
3098 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
3099 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
3101 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
3102 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
3104 trans
->blocks_used
++;
3105 /* this returns a buffer locked for blocking */
3110 * helper function to allocate a block for a given tree
3111 * returns the tree buffer or NULL.
3113 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
3114 struct btrfs_root
*root
,
3115 u32 blocksize
, u64 parent
,
3122 struct btrfs_key ins
;
3124 struct extent_buffer
*buf
;
3126 ret
= btrfs_alloc_extent(trans
, root
, blocksize
, parent
, blocksize
,
3127 root_objectid
, ref_generation
, level
,
3128 empty_size
, hint
, (u64
)-1, &ins
, 0);
3131 return ERR_PTR(ret
);
3134 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
3139 int btrfs_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
3140 struct btrfs_root
*root
, struct extent_buffer
*leaf
)
3143 u64 leaf_generation
;
3144 struct refsort
*sorted
;
3145 struct btrfs_key key
;
3146 struct btrfs_file_extent_item
*fi
;
3153 BUG_ON(!btrfs_is_leaf(leaf
));
3154 nritems
= btrfs_header_nritems(leaf
);
3155 leaf_owner
= btrfs_header_owner(leaf
);
3156 leaf_generation
= btrfs_header_generation(leaf
);
3158 sorted
= kmalloc(sizeof(*sorted
) * nritems
, GFP_NOFS
);
3159 /* we do this loop twice. The first time we build a list
3160 * of the extents we have a reference on, then we sort the list
3161 * by bytenr. The second time around we actually do the
3164 for (i
= 0; i
< nritems
; i
++) {
3168 btrfs_item_key_to_cpu(leaf
, &key
, i
);
3170 /* only extents have references, skip everything else */
3171 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
3174 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
3176 /* inline extents live in the btree, they don't have refs */
3177 if (btrfs_file_extent_type(leaf
, fi
) ==
3178 BTRFS_FILE_EXTENT_INLINE
)
3181 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3183 /* holes don't have refs */
3184 if (disk_bytenr
== 0)
3187 sorted
[refi
].bytenr
= disk_bytenr
;
3188 sorted
[refi
].slot
= i
;
3195 sort(sorted
, refi
, sizeof(struct refsort
), refsort_cmp
, NULL
);
3197 for (i
= 0; i
< refi
; i
++) {
3200 disk_bytenr
= sorted
[i
].bytenr
;
3201 slot
= sorted
[i
].slot
;
3205 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
3206 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
3209 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
3211 ret
= btrfs_free_extent(trans
, root
, disk_bytenr
,
3212 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
3213 leaf
->start
, leaf_owner
, leaf_generation
,
3217 atomic_inc(&root
->fs_info
->throttle_gen
);
3218 wake_up(&root
->fs_info
->transaction_throttle
);
3226 static noinline
int cache_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
3227 struct btrfs_root
*root
,
3228 struct btrfs_leaf_ref
*ref
)
3232 struct btrfs_extent_info
*info
;
3233 struct refsort
*sorted
;
3235 if (ref
->nritems
== 0)
3238 sorted
= kmalloc(sizeof(*sorted
) * ref
->nritems
, GFP_NOFS
);
3239 for (i
= 0; i
< ref
->nritems
; i
++) {
3240 sorted
[i
].bytenr
= ref
->extents
[i
].bytenr
;
3243 sort(sorted
, ref
->nritems
, sizeof(struct refsort
), refsort_cmp
, NULL
);
3246 * the items in the ref were sorted when the ref was inserted
3247 * into the ref cache, so this is already in order
3249 for (i
= 0; i
< ref
->nritems
; i
++) {
3250 info
= ref
->extents
+ sorted
[i
].slot
;
3251 ret
= btrfs_free_extent(trans
, root
, info
->bytenr
,
3252 info
->num_bytes
, ref
->bytenr
,
3253 ref
->owner
, ref
->generation
,
3256 atomic_inc(&root
->fs_info
->throttle_gen
);
3257 wake_up(&root
->fs_info
->transaction_throttle
);
3268 static int drop_snap_lookup_refcount(struct btrfs_trans_handle
*trans
,
3269 struct btrfs_root
*root
, u64 start
,
3274 ret
= btrfs_lookup_extent_ref(trans
, root
, start
, len
, refs
);
3277 #if 0 /* some debugging code in case we see problems here */
3278 /* if the refs count is one, it won't get increased again. But
3279 * if the ref count is > 1, someone may be decreasing it at
3280 * the same time we are.
3283 struct extent_buffer
*eb
= NULL
;
3284 eb
= btrfs_find_create_tree_block(root
, start
, len
);
3286 btrfs_tree_lock(eb
);
3288 mutex_lock(&root
->fs_info
->alloc_mutex
);
3289 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
3291 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3294 btrfs_tree_unlock(eb
);
3295 free_extent_buffer(eb
);
3298 printk(KERN_ERR
"btrfs block %llu went down to one "
3299 "during drop_snap\n", (unsigned long long)start
);
3310 * this is used while deleting old snapshots, and it drops the refs
3311 * on a whole subtree starting from a level 1 node.
3313 * The idea is to sort all the leaf pointers, and then drop the
3314 * ref on all the leaves in order. Most of the time the leaves
3315 * will have ref cache entries, so no leaf IOs will be required to
3316 * find the extents they have references on.
3318 * For each leaf, any references it has are also dropped in order
3320 * This ends up dropping the references in something close to optimal
3321 * order for reading and modifying the extent allocation tree.
3323 static noinline
int drop_level_one_refs(struct btrfs_trans_handle
*trans
,
3324 struct btrfs_root
*root
,
3325 struct btrfs_path
*path
)
3330 struct extent_buffer
*eb
= path
->nodes
[1];
3331 struct extent_buffer
*leaf
;
3332 struct btrfs_leaf_ref
*ref
;
3333 struct refsort
*sorted
= NULL
;
3334 int nritems
= btrfs_header_nritems(eb
);
3338 int slot
= path
->slots
[1];
3339 u32 blocksize
= btrfs_level_size(root
, 0);
3345 root_owner
= btrfs_header_owner(eb
);
3346 root_gen
= btrfs_header_generation(eb
);
3347 sorted
= kmalloc(sizeof(*sorted
) * nritems
, GFP_NOFS
);
3350 * step one, sort all the leaf pointers so we don't scribble
3351 * randomly into the extent allocation tree
3353 for (i
= slot
; i
< nritems
; i
++) {
3354 sorted
[refi
].bytenr
= btrfs_node_blockptr(eb
, i
);
3355 sorted
[refi
].slot
= i
;
3360 * nritems won't be zero, but if we're picking up drop_snapshot
3361 * after a crash, slot might be > 0, so double check things
3367 sort(sorted
, refi
, sizeof(struct refsort
), refsort_cmp
, NULL
);
3370 * the first loop frees everything the leaves point to
3372 for (i
= 0; i
< refi
; i
++) {
3375 bytenr
= sorted
[i
].bytenr
;
3378 * check the reference count on this leaf. If it is > 1
3379 * we just decrement it below and don't update any
3380 * of the refs the leaf points to.
3382 ret
= drop_snap_lookup_refcount(trans
, root
, bytenr
,
3388 ptr_gen
= btrfs_node_ptr_generation(eb
, sorted
[i
].slot
);
3391 * the leaf only had one reference, which means the
3392 * only thing pointing to this leaf is the snapshot
3393 * we're deleting. It isn't possible for the reference
3394 * count to increase again later
3396 * The reference cache is checked for the leaf,
3397 * and if found we'll be able to drop any refs held by
3398 * the leaf without needing to read it in.
3400 ref
= btrfs_lookup_leaf_ref(root
, bytenr
);
3401 if (ref
&& ref
->generation
!= ptr_gen
) {
3402 btrfs_free_leaf_ref(root
, ref
);
3406 ret
= cache_drop_leaf_ref(trans
, root
, ref
);
3408 btrfs_remove_leaf_ref(root
, ref
);
3409 btrfs_free_leaf_ref(root
, ref
);
3412 * the leaf wasn't in the reference cache, so
3413 * we have to read it.
3415 leaf
= read_tree_block(root
, bytenr
, blocksize
,
3417 ret
= btrfs_drop_leaf_ref(trans
, root
, leaf
);
3419 free_extent_buffer(leaf
);
3421 atomic_inc(&root
->fs_info
->throttle_gen
);
3422 wake_up(&root
->fs_info
->transaction_throttle
);
3427 * run through the loop again to free the refs on the leaves.
3428 * This is faster than doing it in the loop above because
3429 * the leaves are likely to be clustered together. We end up
3430 * working in nice chunks on the extent allocation tree.
3432 for (i
= 0; i
< refi
; i
++) {
3433 bytenr
= sorted
[i
].bytenr
;
3434 ret
= btrfs_free_extent(trans
, root
, bytenr
,
3435 blocksize
, eb
->start
,
3436 root_owner
, root_gen
, 0, 1);
3439 atomic_inc(&root
->fs_info
->throttle_gen
);
3440 wake_up(&root
->fs_info
->transaction_throttle
);
3447 * update the path to show we've processed the entire level 1
3448 * node. This will get saved into the root's drop_snapshot_progress
3449 * field so these drops are not repeated again if this transaction
3452 path
->slots
[1] = nritems
;
3457 * helper function for drop_snapshot, this walks down the tree dropping ref
3458 * counts as it goes.
3460 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
3461 struct btrfs_root
*root
,
3462 struct btrfs_path
*path
, int *level
)
3468 struct extent_buffer
*next
;
3469 struct extent_buffer
*cur
;
3470 struct extent_buffer
*parent
;
3475 WARN_ON(*level
< 0);
3476 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
3477 ret
= drop_snap_lookup_refcount(trans
, root
, path
->nodes
[*level
]->start
,
3478 path
->nodes
[*level
]->len
, &refs
);
3484 * walk down to the last node level and free all the leaves
3486 while (*level
>= 0) {
3487 WARN_ON(*level
< 0);
3488 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
3489 cur
= path
->nodes
[*level
];
3491 if (btrfs_header_level(cur
) != *level
)
3494 if (path
->slots
[*level
] >=
3495 btrfs_header_nritems(cur
))
3498 /* the new code goes down to level 1 and does all the
3499 * leaves pointed to that node in bulk. So, this check
3500 * for level 0 will always be false.
3502 * But, the disk format allows the drop_snapshot_progress
3503 * field in the root to leave things in a state where
3504 * a leaf will need cleaning up here. If someone crashes
3505 * with the old code and then boots with the new code,
3506 * we might find a leaf here.
3509 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
3515 * once we get to level one, process the whole node
3516 * at once, including everything below it.
3519 ret
= drop_level_one_refs(trans
, root
, path
);
3524 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
3525 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
3526 blocksize
= btrfs_level_size(root
, *level
- 1);
3528 ret
= drop_snap_lookup_refcount(trans
, root
, bytenr
,
3533 * if there is more than one reference, we don't need
3534 * to read that node to drop any references it has. We
3535 * just drop the ref we hold on that node and move on to the
3536 * next slot in this level.
3539 parent
= path
->nodes
[*level
];
3540 root_owner
= btrfs_header_owner(parent
);
3541 root_gen
= btrfs_header_generation(parent
);
3542 path
->slots
[*level
]++;
3544 ret
= btrfs_free_extent(trans
, root
, bytenr
,
3545 blocksize
, parent
->start
,
3546 root_owner
, root_gen
,
3550 atomic_inc(&root
->fs_info
->throttle_gen
);
3551 wake_up(&root
->fs_info
->transaction_throttle
);
3558 * we need to keep freeing things in the next level down.
3559 * read the block and loop around to process it
3561 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
3562 WARN_ON(*level
<= 0);
3563 if (path
->nodes
[*level
-1])
3564 free_extent_buffer(path
->nodes
[*level
-1]);
3565 path
->nodes
[*level
-1] = next
;
3566 *level
= btrfs_header_level(next
);
3567 path
->slots
[*level
] = 0;
3571 WARN_ON(*level
< 0);
3572 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
3574 if (path
->nodes
[*level
] == root
->node
) {
3575 parent
= path
->nodes
[*level
];
3576 bytenr
= path
->nodes
[*level
]->start
;
3578 parent
= path
->nodes
[*level
+ 1];
3579 bytenr
= btrfs_node_blockptr(parent
, path
->slots
[*level
+ 1]);
3582 blocksize
= btrfs_level_size(root
, *level
);
3583 root_owner
= btrfs_header_owner(parent
);
3584 root_gen
= btrfs_header_generation(parent
);
3587 * cleanup and free the reference on the last node
3590 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
3591 parent
->start
, root_owner
, root_gen
,
3593 free_extent_buffer(path
->nodes
[*level
]);
3594 path
->nodes
[*level
] = NULL
;
3604 * helper function for drop_subtree, this function is similar to
3605 * walk_down_tree. The main difference is that it checks reference
3606 * counts while tree blocks are locked.
3608 static noinline
int walk_down_subtree(struct btrfs_trans_handle
*trans
,
3609 struct btrfs_root
*root
,
3610 struct btrfs_path
*path
, int *level
)
3612 struct extent_buffer
*next
;
3613 struct extent_buffer
*cur
;
3614 struct extent_buffer
*parent
;
3621 cur
= path
->nodes
[*level
];
3622 ret
= btrfs_lookup_extent_ref(trans
, root
, cur
->start
, cur
->len
,
3628 while (*level
>= 0) {
3629 cur
= path
->nodes
[*level
];
3631 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
3633 clean_tree_block(trans
, root
, cur
);
3636 if (path
->slots
[*level
] >= btrfs_header_nritems(cur
)) {
3637 clean_tree_block(trans
, root
, cur
);
3641 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
3642 blocksize
= btrfs_level_size(root
, *level
- 1);
3643 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
3645 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
3646 btrfs_tree_lock(next
);
3647 btrfs_set_lock_blocking(next
);
3649 ret
= btrfs_lookup_extent_ref(trans
, root
, bytenr
, blocksize
,
3653 parent
= path
->nodes
[*level
];
3654 ret
= btrfs_free_extent(trans
, root
, bytenr
,
3655 blocksize
, parent
->start
,
3656 btrfs_header_owner(parent
),
3657 btrfs_header_generation(parent
),
3660 path
->slots
[*level
]++;
3661 btrfs_tree_unlock(next
);
3662 free_extent_buffer(next
);
3666 *level
= btrfs_header_level(next
);
3667 path
->nodes
[*level
] = next
;
3668 path
->slots
[*level
] = 0;
3669 path
->locks
[*level
] = 1;
3673 parent
= path
->nodes
[*level
+ 1];
3674 bytenr
= path
->nodes
[*level
]->start
;
3675 blocksize
= path
->nodes
[*level
]->len
;
3677 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
3678 parent
->start
, btrfs_header_owner(parent
),
3679 btrfs_header_generation(parent
), *level
, 1);
3682 if (path
->locks
[*level
]) {
3683 btrfs_tree_unlock(path
->nodes
[*level
]);
3684 path
->locks
[*level
] = 0;
3686 free_extent_buffer(path
->nodes
[*level
]);
3687 path
->nodes
[*level
] = NULL
;
3694 * helper for dropping snapshots. This walks back up the tree in the path
3695 * to find the first node higher up where we haven't yet gone through
3698 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
3699 struct btrfs_root
*root
,
3700 struct btrfs_path
*path
,
3701 int *level
, int max_level
)
3705 struct btrfs_root_item
*root_item
= &root
->root_item
;
3710 for (i
= *level
; i
< max_level
&& path
->nodes
[i
]; i
++) {
3711 slot
= path
->slots
[i
];
3712 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
3713 struct extent_buffer
*node
;
3714 struct btrfs_disk_key disk_key
;
3717 * there is more work to do in this level.
3718 * Update the drop_progress marker to reflect
3719 * the work we've done so far, and then bump
3722 node
= path
->nodes
[i
];
3725 WARN_ON(*level
== 0);
3726 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
3727 memcpy(&root_item
->drop_progress
,
3728 &disk_key
, sizeof(disk_key
));
3729 root_item
->drop_level
= i
;
3732 struct extent_buffer
*parent
;
3735 * this whole node is done, free our reference
3736 * on it and go up one level
3738 if (path
->nodes
[*level
] == root
->node
)
3739 parent
= path
->nodes
[*level
];
3741 parent
= path
->nodes
[*level
+ 1];
3743 root_owner
= btrfs_header_owner(parent
);
3744 root_gen
= btrfs_header_generation(parent
);
3746 clean_tree_block(trans
, root
, path
->nodes
[*level
]);
3747 ret
= btrfs_free_extent(trans
, root
,
3748 path
->nodes
[*level
]->start
,
3749 path
->nodes
[*level
]->len
,
3750 parent
->start
, root_owner
,
3751 root_gen
, *level
, 1);
3753 if (path
->locks
[*level
]) {
3754 btrfs_tree_unlock(path
->nodes
[*level
]);
3755 path
->locks
[*level
] = 0;
3757 free_extent_buffer(path
->nodes
[*level
]);
3758 path
->nodes
[*level
] = NULL
;
3766 * drop the reference count on the tree rooted at 'snap'. This traverses
3767 * the tree freeing any blocks that have a ref count of zero after being
3770 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
3776 struct btrfs_path
*path
;
3780 struct btrfs_root_item
*root_item
= &root
->root_item
;
3782 WARN_ON(!mutex_is_locked(&root
->fs_info
->drop_mutex
));
3783 path
= btrfs_alloc_path();
3786 level
= btrfs_header_level(root
->node
);
3788 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
3789 path
->nodes
[level
] = root
->node
;
3790 extent_buffer_get(root
->node
);
3791 path
->slots
[level
] = 0;
3793 struct btrfs_key key
;
3794 struct btrfs_disk_key found_key
;
3795 struct extent_buffer
*node
;
3797 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
3798 level
= root_item
->drop_level
;
3799 path
->lowest_level
= level
;
3800 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3805 node
= path
->nodes
[level
];
3806 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
3807 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
3808 sizeof(found_key
)));
3810 * unlock our path, this is safe because only this
3811 * function is allowed to delete this snapshot
3813 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
3814 if (path
->nodes
[i
] && path
->locks
[i
]) {
3816 btrfs_tree_unlock(path
->nodes
[i
]);
3821 unsigned long update
;
3822 wret
= walk_down_tree(trans
, root
, path
, &level
);
3828 wret
= walk_up_tree(trans
, root
, path
, &level
,
3834 if (trans
->transaction
->in_commit
||
3835 trans
->transaction
->delayed_refs
.flushing
) {
3839 atomic_inc(&root
->fs_info
->throttle_gen
);
3840 wake_up(&root
->fs_info
->transaction_throttle
);
3841 for (update_count
= 0; update_count
< 16; update_count
++) {
3842 update
= trans
->delayed_ref_updates
;
3843 trans
->delayed_ref_updates
= 0;
3845 btrfs_run_delayed_refs(trans
, root
, update
);
3850 for (i
= 0; i
<= orig_level
; i
++) {
3851 if (path
->nodes
[i
]) {
3852 free_extent_buffer(path
->nodes
[i
]);
3853 path
->nodes
[i
] = NULL
;
3857 btrfs_free_path(path
);
3861 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
3862 struct btrfs_root
*root
,
3863 struct extent_buffer
*node
,
3864 struct extent_buffer
*parent
)
3866 struct btrfs_path
*path
;
3872 path
= btrfs_alloc_path();
3875 btrfs_assert_tree_locked(parent
);
3876 parent_level
= btrfs_header_level(parent
);
3877 extent_buffer_get(parent
);
3878 path
->nodes
[parent_level
] = parent
;
3879 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
3881 btrfs_assert_tree_locked(node
);
3882 level
= btrfs_header_level(node
);
3883 extent_buffer_get(node
);
3884 path
->nodes
[level
] = node
;
3885 path
->slots
[level
] = 0;
3888 wret
= walk_down_subtree(trans
, root
, path
, &level
);
3894 wret
= walk_up_tree(trans
, root
, path
, &level
, parent_level
);
3901 btrfs_free_path(path
);
3905 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
3908 return min(last
, start
+ nr
- 1);
3911 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
3916 unsigned long first_index
;
3917 unsigned long last_index
;
3920 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3921 struct file_ra_state
*ra
;
3922 struct btrfs_ordered_extent
*ordered
;
3923 unsigned int total_read
= 0;
3924 unsigned int total_dirty
= 0;
3927 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
3929 mutex_lock(&inode
->i_mutex
);
3930 first_index
= start
>> PAGE_CACHE_SHIFT
;
3931 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
3933 /* make sure the dirty trick played by the caller work */
3934 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
3935 first_index
, last_index
);
3939 file_ra_state_init(ra
, inode
->i_mapping
);
3941 for (i
= first_index
; i
<= last_index
; i
++) {
3942 if (total_read
% ra
->ra_pages
== 0) {
3943 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
3944 calc_ra(i
, last_index
, ra
->ra_pages
));
3948 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
3950 page
= grab_cache_page(inode
->i_mapping
, i
);
3955 if (!PageUptodate(page
)) {
3956 btrfs_readpage(NULL
, page
);
3958 if (!PageUptodate(page
)) {
3960 page_cache_release(page
);
3965 wait_on_page_writeback(page
);
3967 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3968 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3969 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3971 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3973 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3975 page_cache_release(page
);
3976 btrfs_start_ordered_extent(inode
, ordered
, 1);
3977 btrfs_put_ordered_extent(ordered
);
3980 set_page_extent_mapped(page
);
3982 if (i
== first_index
)
3983 set_extent_bits(io_tree
, page_start
, page_end
,
3984 EXTENT_BOUNDARY
, GFP_NOFS
);
3985 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
3987 set_page_dirty(page
);
3990 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3992 page_cache_release(page
);
3997 mutex_unlock(&inode
->i_mutex
);
3998 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
4002 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
4003 struct btrfs_key
*extent_key
,
4006 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
4007 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
4008 struct extent_map
*em
;
4009 u64 start
= extent_key
->objectid
- offset
;
4010 u64 end
= start
+ extent_key
->offset
- 1;
4012 em
= alloc_extent_map(GFP_NOFS
);
4013 BUG_ON(!em
|| IS_ERR(em
));
4016 em
->len
= extent_key
->offset
;
4017 em
->block_len
= extent_key
->offset
;
4018 em
->block_start
= extent_key
->objectid
;
4019 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4020 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
4022 /* setup extent map to cheat btrfs_readpage */
4023 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
4026 spin_lock(&em_tree
->lock
);
4027 ret
= add_extent_mapping(em_tree
, em
);
4028 spin_unlock(&em_tree
->lock
);
4029 if (ret
!= -EEXIST
) {
4030 free_extent_map(em
);
4033 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
4035 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
4037 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
4040 struct btrfs_ref_path
{
4042 u64 nodes
[BTRFS_MAX_LEVEL
];
4044 u64 root_generation
;
4051 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
4052 u64 new_nodes
[BTRFS_MAX_LEVEL
];
4055 struct disk_extent
{
4066 static int is_cowonly_root(u64 root_objectid
)
4068 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
4069 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
4070 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
4071 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
4072 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
4073 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
4078 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
4079 struct btrfs_root
*extent_root
,
4080 struct btrfs_ref_path
*ref_path
,
4083 struct extent_buffer
*leaf
;
4084 struct btrfs_path
*path
;
4085 struct btrfs_extent_ref
*ref
;
4086 struct btrfs_key key
;
4087 struct btrfs_key found_key
;
4093 path
= btrfs_alloc_path();
4098 ref_path
->lowest_level
= -1;
4099 ref_path
->current_level
= -1;
4100 ref_path
->shared_level
= -1;
4104 level
= ref_path
->current_level
- 1;
4105 while (level
>= -1) {
4107 if (level
< ref_path
->lowest_level
)
4111 bytenr
= ref_path
->nodes
[level
];
4113 bytenr
= ref_path
->extent_start
;
4114 BUG_ON(bytenr
== 0);
4116 parent
= ref_path
->nodes
[level
+ 1];
4117 ref_path
->nodes
[level
+ 1] = 0;
4118 ref_path
->current_level
= level
;
4119 BUG_ON(parent
== 0);
4121 key
.objectid
= bytenr
;
4122 key
.offset
= parent
+ 1;
4123 key
.type
= BTRFS_EXTENT_REF_KEY
;
4125 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
4130 leaf
= path
->nodes
[0];
4131 nritems
= btrfs_header_nritems(leaf
);
4132 if (path
->slots
[0] >= nritems
) {
4133 ret
= btrfs_next_leaf(extent_root
, path
);
4138 leaf
= path
->nodes
[0];
4141 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4142 if (found_key
.objectid
== bytenr
&&
4143 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
4144 if (level
< ref_path
->shared_level
)
4145 ref_path
->shared_level
= level
;
4150 btrfs_release_path(extent_root
, path
);
4153 /* reached lowest level */
4157 level
= ref_path
->current_level
;
4158 while (level
< BTRFS_MAX_LEVEL
- 1) {
4162 bytenr
= ref_path
->nodes
[level
];
4164 bytenr
= ref_path
->extent_start
;
4166 BUG_ON(bytenr
== 0);
4168 key
.objectid
= bytenr
;
4170 key
.type
= BTRFS_EXTENT_REF_KEY
;
4172 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
4176 leaf
= path
->nodes
[0];
4177 nritems
= btrfs_header_nritems(leaf
);
4178 if (path
->slots
[0] >= nritems
) {
4179 ret
= btrfs_next_leaf(extent_root
, path
);
4183 /* the extent was freed by someone */
4184 if (ref_path
->lowest_level
== level
)
4186 btrfs_release_path(extent_root
, path
);
4189 leaf
= path
->nodes
[0];
4192 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4193 if (found_key
.objectid
!= bytenr
||
4194 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
4195 /* the extent was freed by someone */
4196 if (ref_path
->lowest_level
== level
) {
4200 btrfs_release_path(extent_root
, path
);
4204 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
4205 struct btrfs_extent_ref
);
4206 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
4207 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4209 level
= (int)ref_objectid
;
4210 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
4211 ref_path
->lowest_level
= level
;
4212 ref_path
->current_level
= level
;
4213 ref_path
->nodes
[level
] = bytenr
;
4215 WARN_ON(ref_objectid
!= level
);
4218 WARN_ON(level
!= -1);
4222 if (ref_path
->lowest_level
== level
) {
4223 ref_path
->owner_objectid
= ref_objectid
;
4224 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
4228 * the block is tree root or the block isn't in reference
4231 if (found_key
.objectid
== found_key
.offset
||
4232 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
4233 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
4234 ref_path
->root_generation
=
4235 btrfs_ref_generation(leaf
, ref
);
4237 /* special reference from the tree log */
4238 ref_path
->nodes
[0] = found_key
.offset
;
4239 ref_path
->current_level
= 0;
4246 BUG_ON(ref_path
->nodes
[level
] != 0);
4247 ref_path
->nodes
[level
] = found_key
.offset
;
4248 ref_path
->current_level
= level
;
4251 * the reference was created in the running transaction,
4252 * no need to continue walking up.
4254 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
4255 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
4256 ref_path
->root_generation
=
4257 btrfs_ref_generation(leaf
, ref
);
4262 btrfs_release_path(extent_root
, path
);
4265 /* reached max tree level, but no tree root found. */
4268 btrfs_free_path(path
);
4272 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
4273 struct btrfs_root
*extent_root
,
4274 struct btrfs_ref_path
*ref_path
,
4277 memset(ref_path
, 0, sizeof(*ref_path
));
4278 ref_path
->extent_start
= extent_start
;
4280 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
4283 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
4284 struct btrfs_root
*extent_root
,
4285 struct btrfs_ref_path
*ref_path
)
4287 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
4290 static noinline
int get_new_locations(struct inode
*reloc_inode
,
4291 struct btrfs_key
*extent_key
,
4292 u64 offset
, int no_fragment
,
4293 struct disk_extent
**extents
,
4296 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
4297 struct btrfs_path
*path
;
4298 struct btrfs_file_extent_item
*fi
;
4299 struct extent_buffer
*leaf
;
4300 struct disk_extent
*exts
= *extents
;
4301 struct btrfs_key found_key
;
4306 int max
= *nr_extents
;
4309 WARN_ON(!no_fragment
&& *extents
);
4312 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
4317 path
= btrfs_alloc_path();
4320 cur_pos
= extent_key
->objectid
- offset
;
4321 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
4322 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
4332 leaf
= path
->nodes
[0];
4333 nritems
= btrfs_header_nritems(leaf
);
4334 if (path
->slots
[0] >= nritems
) {
4335 ret
= btrfs_next_leaf(root
, path
);
4340 leaf
= path
->nodes
[0];
4343 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4344 if (found_key
.offset
!= cur_pos
||
4345 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
4346 found_key
.objectid
!= reloc_inode
->i_ino
)
4349 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4350 struct btrfs_file_extent_item
);
4351 if (btrfs_file_extent_type(leaf
, fi
) !=
4352 BTRFS_FILE_EXTENT_REG
||
4353 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
4357 struct disk_extent
*old
= exts
;
4359 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
4360 memcpy(exts
, old
, sizeof(*exts
) * nr
);
4361 if (old
!= *extents
)
4365 exts
[nr
].disk_bytenr
=
4366 btrfs_file_extent_disk_bytenr(leaf
, fi
);
4367 exts
[nr
].disk_num_bytes
=
4368 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4369 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
4370 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4371 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
4372 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
4373 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
4374 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
4376 BUG_ON(exts
[nr
].offset
> 0);
4377 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
4378 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
4380 cur_pos
+= exts
[nr
].num_bytes
;
4383 if (cur_pos
+ offset
>= last_byte
)
4393 BUG_ON(cur_pos
+ offset
> last_byte
);
4394 if (cur_pos
+ offset
< last_byte
) {
4400 btrfs_free_path(path
);
4402 if (exts
!= *extents
)
4411 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
4412 struct btrfs_root
*root
,
4413 struct btrfs_path
*path
,
4414 struct btrfs_key
*extent_key
,
4415 struct btrfs_key
*leaf_key
,
4416 struct btrfs_ref_path
*ref_path
,
4417 struct disk_extent
*new_extents
,
4420 struct extent_buffer
*leaf
;
4421 struct btrfs_file_extent_item
*fi
;
4422 struct inode
*inode
= NULL
;
4423 struct btrfs_key key
;
4428 u64 search_end
= (u64
)-1;
4431 int extent_locked
= 0;
4435 memcpy(&key
, leaf_key
, sizeof(key
));
4436 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
4437 if (key
.objectid
< ref_path
->owner_objectid
||
4438 (key
.objectid
== ref_path
->owner_objectid
&&
4439 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
4440 key
.objectid
= ref_path
->owner_objectid
;
4441 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4447 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4451 leaf
= path
->nodes
[0];
4452 nritems
= btrfs_header_nritems(leaf
);
4454 if (extent_locked
&& ret
> 0) {
4456 * the file extent item was modified by someone
4457 * before the extent got locked.
4459 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4460 lock_end
, GFP_NOFS
);
4464 if (path
->slots
[0] >= nritems
) {
4465 if (++nr_scaned
> 2)
4468 BUG_ON(extent_locked
);
4469 ret
= btrfs_next_leaf(root
, path
);
4474 leaf
= path
->nodes
[0];
4475 nritems
= btrfs_header_nritems(leaf
);
4478 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4480 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
4481 if ((key
.objectid
> ref_path
->owner_objectid
) ||
4482 (key
.objectid
== ref_path
->owner_objectid
&&
4483 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
4484 key
.offset
>= search_end
)
4488 if (inode
&& key
.objectid
!= inode
->i_ino
) {
4489 BUG_ON(extent_locked
);
4490 btrfs_release_path(root
, path
);
4491 mutex_unlock(&inode
->i_mutex
);
4497 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
4502 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4503 struct btrfs_file_extent_item
);
4504 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4505 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
4506 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
4507 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
4508 extent_key
->objectid
)) {
4514 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4515 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
4517 if (search_end
== (u64
)-1) {
4518 search_end
= key
.offset
- ext_offset
+
4519 btrfs_file_extent_ram_bytes(leaf
, fi
);
4522 if (!extent_locked
) {
4523 lock_start
= key
.offset
;
4524 lock_end
= lock_start
+ num_bytes
- 1;
4526 if (lock_start
> key
.offset
||
4527 lock_end
+ 1 < key
.offset
+ num_bytes
) {
4528 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4529 lock_start
, lock_end
, GFP_NOFS
);
4535 btrfs_release_path(root
, path
);
4537 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
4538 key
.objectid
, root
);
4539 if (inode
->i_state
& I_NEW
) {
4540 BTRFS_I(inode
)->root
= root
;
4541 BTRFS_I(inode
)->location
.objectid
=
4543 BTRFS_I(inode
)->location
.type
=
4544 BTRFS_INODE_ITEM_KEY
;
4545 BTRFS_I(inode
)->location
.offset
= 0;
4546 btrfs_read_locked_inode(inode
);
4547 unlock_new_inode(inode
);
4550 * some code call btrfs_commit_transaction while
4551 * holding the i_mutex, so we can't use mutex_lock
4554 if (is_bad_inode(inode
) ||
4555 !mutex_trylock(&inode
->i_mutex
)) {
4558 key
.offset
= (u64
)-1;
4563 if (!extent_locked
) {
4564 struct btrfs_ordered_extent
*ordered
;
4566 btrfs_release_path(root
, path
);
4568 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4569 lock_end
, GFP_NOFS
);
4570 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4573 ordered
->file_offset
<= lock_end
&&
4574 ordered
->file_offset
+ ordered
->len
> lock_start
) {
4575 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4576 lock_start
, lock_end
, GFP_NOFS
);
4577 btrfs_start_ordered_extent(inode
, ordered
, 1);
4578 btrfs_put_ordered_extent(ordered
);
4579 key
.offset
+= num_bytes
;
4583 btrfs_put_ordered_extent(ordered
);
4589 if (nr_extents
== 1) {
4590 /* update extent pointer in place */
4591 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4592 new_extents
[0].disk_bytenr
);
4593 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4594 new_extents
[0].disk_num_bytes
);
4595 btrfs_mark_buffer_dirty(leaf
);
4597 btrfs_drop_extent_cache(inode
, key
.offset
,
4598 key
.offset
+ num_bytes
- 1, 0);
4600 ret
= btrfs_inc_extent_ref(trans
, root
,
4601 new_extents
[0].disk_bytenr
,
4602 new_extents
[0].disk_num_bytes
,
4604 root
->root_key
.objectid
,
4609 ret
= btrfs_free_extent(trans
, root
,
4610 extent_key
->objectid
,
4613 btrfs_header_owner(leaf
),
4614 btrfs_header_generation(leaf
),
4618 btrfs_release_path(root
, path
);
4619 key
.offset
+= num_bytes
;
4627 * drop old extent pointer at first, then insert the
4628 * new pointers one bye one
4630 btrfs_release_path(root
, path
);
4631 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
4632 key
.offset
+ num_bytes
,
4633 key
.offset
, &alloc_hint
);
4636 for (i
= 0; i
< nr_extents
; i
++) {
4637 if (ext_offset
>= new_extents
[i
].num_bytes
) {
4638 ext_offset
-= new_extents
[i
].num_bytes
;
4641 extent_len
= min(new_extents
[i
].num_bytes
-
4642 ext_offset
, num_bytes
);
4644 ret
= btrfs_insert_empty_item(trans
, root
,
4649 leaf
= path
->nodes
[0];
4650 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4651 struct btrfs_file_extent_item
);
4652 btrfs_set_file_extent_generation(leaf
, fi
,
4654 btrfs_set_file_extent_type(leaf
, fi
,
4655 BTRFS_FILE_EXTENT_REG
);
4656 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4657 new_extents
[i
].disk_bytenr
);
4658 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4659 new_extents
[i
].disk_num_bytes
);
4660 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
4661 new_extents
[i
].ram_bytes
);
4663 btrfs_set_file_extent_compression(leaf
, fi
,
4664 new_extents
[i
].compression
);
4665 btrfs_set_file_extent_encryption(leaf
, fi
,
4666 new_extents
[i
].encryption
);
4667 btrfs_set_file_extent_other_encoding(leaf
, fi
,
4668 new_extents
[i
].other_encoding
);
4670 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4672 ext_offset
+= new_extents
[i
].offset
;
4673 btrfs_set_file_extent_offset(leaf
, fi
,
4675 btrfs_mark_buffer_dirty(leaf
);
4677 btrfs_drop_extent_cache(inode
, key
.offset
,
4678 key
.offset
+ extent_len
- 1, 0);
4680 ret
= btrfs_inc_extent_ref(trans
, root
,
4681 new_extents
[i
].disk_bytenr
,
4682 new_extents
[i
].disk_num_bytes
,
4684 root
->root_key
.objectid
,
4685 trans
->transid
, key
.objectid
);
4687 btrfs_release_path(root
, path
);
4689 inode_add_bytes(inode
, extent_len
);
4692 num_bytes
-= extent_len
;
4693 key
.offset
+= extent_len
;
4698 BUG_ON(i
>= nr_extents
);
4702 if (extent_locked
) {
4703 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4704 lock_end
, GFP_NOFS
);
4708 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
4709 key
.offset
>= search_end
)
4716 btrfs_release_path(root
, path
);
4718 mutex_unlock(&inode
->i_mutex
);
4719 if (extent_locked
) {
4720 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4721 lock_end
, GFP_NOFS
);
4728 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
4729 struct btrfs_root
*root
,
4730 struct extent_buffer
*buf
, u64 orig_start
)
4735 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
4736 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
4738 level
= btrfs_header_level(buf
);
4740 struct btrfs_leaf_ref
*ref
;
4741 struct btrfs_leaf_ref
*orig_ref
;
4743 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
4747 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
4749 btrfs_free_leaf_ref(root
, orig_ref
);
4753 ref
->nritems
= orig_ref
->nritems
;
4754 memcpy(ref
->extents
, orig_ref
->extents
,
4755 sizeof(ref
->extents
[0]) * ref
->nritems
);
4757 btrfs_free_leaf_ref(root
, orig_ref
);
4759 ref
->root_gen
= trans
->transid
;
4760 ref
->bytenr
= buf
->start
;
4761 ref
->owner
= btrfs_header_owner(buf
);
4762 ref
->generation
= btrfs_header_generation(buf
);
4764 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
4766 btrfs_free_leaf_ref(root
, ref
);
4771 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
4772 struct extent_buffer
*leaf
,
4773 struct btrfs_block_group_cache
*group
,
4774 struct btrfs_root
*target_root
)
4776 struct btrfs_key key
;
4777 struct inode
*inode
= NULL
;
4778 struct btrfs_file_extent_item
*fi
;
4780 u64 skip_objectid
= 0;
4784 nritems
= btrfs_header_nritems(leaf
);
4785 for (i
= 0; i
< nritems
; i
++) {
4786 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4787 if (key
.objectid
== skip_objectid
||
4788 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4790 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4791 if (btrfs_file_extent_type(leaf
, fi
) ==
4792 BTRFS_FILE_EXTENT_INLINE
)
4794 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
4796 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
4798 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
4799 key
.objectid
, target_root
, 1);
4802 skip_objectid
= key
.objectid
;
4805 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4807 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4808 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4809 btrfs_drop_extent_cache(inode
, key
.offset
,
4810 key
.offset
+ num_bytes
- 1, 1);
4811 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4812 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4819 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
4820 struct btrfs_root
*root
,
4821 struct extent_buffer
*leaf
,
4822 struct btrfs_block_group_cache
*group
,
4823 struct inode
*reloc_inode
)
4825 struct btrfs_key key
;
4826 struct btrfs_key extent_key
;
4827 struct btrfs_file_extent_item
*fi
;
4828 struct btrfs_leaf_ref
*ref
;
4829 struct disk_extent
*new_extent
;
4838 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
4839 BUG_ON(!new_extent
);
4841 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
4845 nritems
= btrfs_header_nritems(leaf
);
4846 for (i
= 0; i
< nritems
; i
++) {
4847 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4848 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
4850 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4851 if (btrfs_file_extent_type(leaf
, fi
) ==
4852 BTRFS_FILE_EXTENT_INLINE
)
4854 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4855 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4860 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
4861 bytenr
+ num_bytes
<= group
->key
.objectid
)
4864 extent_key
.objectid
= bytenr
;
4865 extent_key
.offset
= num_bytes
;
4866 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4868 ret
= get_new_locations(reloc_inode
, &extent_key
,
4869 group
->key
.objectid
, 1,
4870 &new_extent
, &nr_extent
);
4875 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
4876 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
4877 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
4878 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
4880 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4881 new_extent
->disk_bytenr
);
4882 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4883 new_extent
->disk_num_bytes
);
4884 btrfs_mark_buffer_dirty(leaf
);
4886 ret
= btrfs_inc_extent_ref(trans
, root
,
4887 new_extent
->disk_bytenr
,
4888 new_extent
->disk_num_bytes
,
4890 root
->root_key
.objectid
,
4891 trans
->transid
, key
.objectid
);
4894 ret
= btrfs_free_extent(trans
, root
,
4895 bytenr
, num_bytes
, leaf
->start
,
4896 btrfs_header_owner(leaf
),
4897 btrfs_header_generation(leaf
),
4903 BUG_ON(ext_index
+ 1 != ref
->nritems
);
4904 btrfs_free_leaf_ref(root
, ref
);
4908 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
4909 struct btrfs_root
*root
)
4911 struct btrfs_root
*reloc_root
;
4914 if (root
->reloc_root
) {
4915 reloc_root
= root
->reloc_root
;
4916 root
->reloc_root
= NULL
;
4917 list_add(&reloc_root
->dead_list
,
4918 &root
->fs_info
->dead_reloc_roots
);
4920 btrfs_set_root_bytenr(&reloc_root
->root_item
,
4921 reloc_root
->node
->start
);
4922 btrfs_set_root_level(&root
->root_item
,
4923 btrfs_header_level(reloc_root
->node
));
4924 memset(&reloc_root
->root_item
.drop_progress
, 0,
4925 sizeof(struct btrfs_disk_key
));
4926 reloc_root
->root_item
.drop_level
= 0;
4928 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
4929 &reloc_root
->root_key
,
4930 &reloc_root
->root_item
);
4936 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
4938 struct btrfs_trans_handle
*trans
;
4939 struct btrfs_root
*reloc_root
;
4940 struct btrfs_root
*prev_root
= NULL
;
4941 struct list_head dead_roots
;
4945 INIT_LIST_HEAD(&dead_roots
);
4946 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
4948 while (!list_empty(&dead_roots
)) {
4949 reloc_root
= list_entry(dead_roots
.prev
,
4950 struct btrfs_root
, dead_list
);
4951 list_del_init(&reloc_root
->dead_list
);
4953 BUG_ON(reloc_root
->commit_root
!= NULL
);
4955 trans
= btrfs_join_transaction(root
, 1);
4958 mutex_lock(&root
->fs_info
->drop_mutex
);
4959 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
4962 mutex_unlock(&root
->fs_info
->drop_mutex
);
4964 nr
= trans
->blocks_used
;
4965 ret
= btrfs_end_transaction(trans
, root
);
4967 btrfs_btree_balance_dirty(root
, nr
);
4970 free_extent_buffer(reloc_root
->node
);
4972 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
4973 &reloc_root
->root_key
);
4975 mutex_unlock(&root
->fs_info
->drop_mutex
);
4977 nr
= trans
->blocks_used
;
4978 ret
= btrfs_end_transaction(trans
, root
);
4980 btrfs_btree_balance_dirty(root
, nr
);
4983 prev_root
= reloc_root
;
4986 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
4992 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
4994 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
4998 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
5000 struct btrfs_root
*reloc_root
;
5001 struct btrfs_trans_handle
*trans
;
5002 struct btrfs_key location
;
5006 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
5007 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
5009 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
5010 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
5013 trans
= btrfs_start_transaction(root
, 1);
5015 ret
= btrfs_commit_transaction(trans
, root
);
5019 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
5020 location
.offset
= (u64
)-1;
5021 location
.type
= BTRFS_ROOT_ITEM_KEY
;
5023 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
5024 BUG_ON(!reloc_root
);
5025 btrfs_orphan_cleanup(reloc_root
);
5029 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
5030 struct btrfs_root
*root
)
5032 struct btrfs_root
*reloc_root
;
5033 struct extent_buffer
*eb
;
5034 struct btrfs_root_item
*root_item
;
5035 struct btrfs_key root_key
;
5038 BUG_ON(!root
->ref_cows
);
5039 if (root
->reloc_root
)
5042 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
5045 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
5046 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
5049 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
5050 root_key
.offset
= root
->root_key
.objectid
;
5051 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
5053 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
5054 btrfs_set_root_refs(root_item
, 0);
5055 btrfs_set_root_bytenr(root_item
, eb
->start
);
5056 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
5057 btrfs_set_root_generation(root_item
, trans
->transid
);
5059 btrfs_tree_unlock(eb
);
5060 free_extent_buffer(eb
);
5062 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
5063 &root_key
, root_item
);
5067 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
5069 BUG_ON(!reloc_root
);
5070 reloc_root
->last_trans
= trans
->transid
;
5071 reloc_root
->commit_root
= NULL
;
5072 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
5074 root
->reloc_root
= reloc_root
;
5079 * Core function of space balance.
5081 * The idea is using reloc trees to relocate tree blocks in reference
5082 * counted roots. There is one reloc tree for each subvol, and all
5083 * reloc trees share same root key objectid. Reloc trees are snapshots
5084 * of the latest committed roots of subvols (root->commit_root).
5086 * To relocate a tree block referenced by a subvol, there are two steps.
5087 * COW the block through subvol's reloc tree, then update block pointer
5088 * in the subvol to point to the new block. Since all reloc trees share
5089 * same root key objectid, doing special handing for tree blocks owned
5090 * by them is easy. Once a tree block has been COWed in one reloc tree,
5091 * we can use the resulting new block directly when the same block is
5092 * required to COW again through other reloc trees. By this way, relocated
5093 * tree blocks are shared between reloc trees, so they are also shared
5096 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
5097 struct btrfs_root
*root
,
5098 struct btrfs_path
*path
,
5099 struct btrfs_key
*first_key
,
5100 struct btrfs_ref_path
*ref_path
,
5101 struct btrfs_block_group_cache
*group
,
5102 struct inode
*reloc_inode
)
5104 struct btrfs_root
*reloc_root
;
5105 struct extent_buffer
*eb
= NULL
;
5106 struct btrfs_key
*keys
;
5110 int lowest_level
= 0;
5113 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
5114 lowest_level
= ref_path
->owner_objectid
;
5116 if (!root
->ref_cows
) {
5117 path
->lowest_level
= lowest_level
;
5118 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
5120 path
->lowest_level
= 0;
5121 btrfs_release_path(root
, path
);
5125 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
5126 ret
= init_reloc_tree(trans
, root
);
5128 reloc_root
= root
->reloc_root
;
5130 shared_level
= ref_path
->shared_level
;
5131 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
5133 keys
= ref_path
->node_keys
;
5134 nodes
= ref_path
->new_nodes
;
5135 memset(&keys
[shared_level
+ 1], 0,
5136 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
5137 memset(&nodes
[shared_level
+ 1], 0,
5138 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
5140 if (nodes
[lowest_level
] == 0) {
5141 path
->lowest_level
= lowest_level
;
5142 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
5145 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
5146 eb
= path
->nodes
[level
];
5147 if (!eb
|| eb
== reloc_root
->node
)
5149 nodes
[level
] = eb
->start
;
5151 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
5153 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
5156 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5157 eb
= path
->nodes
[0];
5158 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
5159 group
, reloc_inode
);
5162 btrfs_release_path(reloc_root
, path
);
5164 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
5170 * replace tree blocks in the fs tree with tree blocks in
5173 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
5176 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5177 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
5180 extent_buffer_get(path
->nodes
[0]);
5181 eb
= path
->nodes
[0];
5182 btrfs_release_path(reloc_root
, path
);
5183 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
5185 free_extent_buffer(eb
);
5188 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
5189 path
->lowest_level
= 0;
5193 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
5194 struct btrfs_root
*root
,
5195 struct btrfs_path
*path
,
5196 struct btrfs_key
*first_key
,
5197 struct btrfs_ref_path
*ref_path
)
5201 ret
= relocate_one_path(trans
, root
, path
, first_key
,
5202 ref_path
, NULL
, NULL
);
5208 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
5209 struct btrfs_root
*extent_root
,
5210 struct btrfs_path
*path
,
5211 struct btrfs_key
*extent_key
)
5215 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
5218 ret
= btrfs_del_item(trans
, extent_root
, path
);
5220 btrfs_release_path(extent_root
, path
);
5224 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
5225 struct btrfs_ref_path
*ref_path
)
5227 struct btrfs_key root_key
;
5229 root_key
.objectid
= ref_path
->root_objectid
;
5230 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
5231 if (is_cowonly_root(ref_path
->root_objectid
))
5232 root_key
.offset
= 0;
5234 root_key
.offset
= (u64
)-1;
5236 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
5239 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
5240 struct btrfs_path
*path
,
5241 struct btrfs_key
*extent_key
,
5242 struct btrfs_block_group_cache
*group
,
5243 struct inode
*reloc_inode
, int pass
)
5245 struct btrfs_trans_handle
*trans
;
5246 struct btrfs_root
*found_root
;
5247 struct btrfs_ref_path
*ref_path
= NULL
;
5248 struct disk_extent
*new_extents
= NULL
;
5253 struct btrfs_key first_key
;
5257 trans
= btrfs_start_transaction(extent_root
, 1);
5260 if (extent_key
->objectid
== 0) {
5261 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
5265 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
5271 for (loops
= 0; ; loops
++) {
5273 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
5274 extent_key
->objectid
);
5276 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
5283 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
5284 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
5287 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
5288 BUG_ON(!found_root
);
5290 * for reference counted tree, only process reference paths
5291 * rooted at the latest committed root.
5293 if (found_root
->ref_cows
&&
5294 ref_path
->root_generation
!= found_root
->root_key
.offset
)
5297 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5300 * copy data extents to new locations
5302 u64 group_start
= group
->key
.objectid
;
5303 ret
= relocate_data_extent(reloc_inode
,
5312 level
= ref_path
->owner_objectid
;
5315 if (prev_block
!= ref_path
->nodes
[level
]) {
5316 struct extent_buffer
*eb
;
5317 u64 block_start
= ref_path
->nodes
[level
];
5318 u64 block_size
= btrfs_level_size(found_root
, level
);
5320 eb
= read_tree_block(found_root
, block_start
,
5322 btrfs_tree_lock(eb
);
5323 BUG_ON(level
!= btrfs_header_level(eb
));
5326 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
5328 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
5330 btrfs_tree_unlock(eb
);
5331 free_extent_buffer(eb
);
5332 prev_block
= block_start
;
5335 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
5336 btrfs_record_root_in_trans(found_root
);
5337 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
5338 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5340 * try to update data extent references while
5341 * keeping metadata shared between snapshots.
5344 ret
= relocate_one_path(trans
, found_root
,
5345 path
, &first_key
, ref_path
,
5346 group
, reloc_inode
);
5352 * use fallback method to process the remaining
5356 u64 group_start
= group
->key
.objectid
;
5357 new_extents
= kmalloc(sizeof(*new_extents
),
5360 ret
= get_new_locations(reloc_inode
,
5368 ret
= replace_one_extent(trans
, found_root
,
5370 &first_key
, ref_path
,
5371 new_extents
, nr_extents
);
5373 ret
= relocate_tree_block(trans
, found_root
, path
,
5374 &first_key
, ref_path
);
5381 btrfs_end_transaction(trans
, extent_root
);
5387 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
5390 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
5391 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
5393 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
5394 if (num_devices
== 1) {
5395 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
5396 stripped
= flags
& ~stripped
;
5398 /* turn raid0 into single device chunks */
5399 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
5402 /* turn mirroring into duplication */
5403 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
5404 BTRFS_BLOCK_GROUP_RAID10
))
5405 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
5408 /* they already had raid on here, just return */
5409 if (flags
& stripped
)
5412 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
5413 stripped
= flags
& ~stripped
;
5415 /* switch duplicated blocks with raid1 */
5416 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
5417 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
5419 /* turn single device chunks into raid0 */
5420 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
5425 static int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
5426 struct btrfs_block_group_cache
*shrink_block_group
,
5429 struct btrfs_trans_handle
*trans
;
5430 u64 new_alloc_flags
;
5433 spin_lock(&shrink_block_group
->lock
);
5434 if (btrfs_block_group_used(&shrink_block_group
->item
) > 0) {
5435 spin_unlock(&shrink_block_group
->lock
);
5437 trans
= btrfs_start_transaction(root
, 1);
5438 spin_lock(&shrink_block_group
->lock
);
5440 new_alloc_flags
= update_block_group_flags(root
,
5441 shrink_block_group
->flags
);
5442 if (new_alloc_flags
!= shrink_block_group
->flags
) {
5444 btrfs_block_group_used(&shrink_block_group
->item
);
5446 calc
= shrink_block_group
->key
.offset
;
5448 spin_unlock(&shrink_block_group
->lock
);
5450 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5451 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
5453 btrfs_end_transaction(trans
, root
);
5455 spin_unlock(&shrink_block_group
->lock
);
5459 static int __insert_orphan_inode(struct btrfs_trans_handle
*trans
,
5460 struct btrfs_root
*root
,
5461 u64 objectid
, u64 size
)
5463 struct btrfs_path
*path
;
5464 struct btrfs_inode_item
*item
;
5465 struct extent_buffer
*leaf
;
5468 path
= btrfs_alloc_path();
5472 path
->leave_spinning
= 1;
5473 ret
= btrfs_insert_empty_inode(trans
, root
, path
, objectid
);
5477 leaf
= path
->nodes
[0];
5478 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_inode_item
);
5479 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
5480 btrfs_set_inode_generation(leaf
, item
, 1);
5481 btrfs_set_inode_size(leaf
, item
, size
);
5482 btrfs_set_inode_mode(leaf
, item
, S_IFREG
| 0600);
5483 btrfs_set_inode_flags(leaf
, item
, BTRFS_INODE_NOCOMPRESS
);
5484 btrfs_mark_buffer_dirty(leaf
);
5485 btrfs_release_path(root
, path
);
5487 btrfs_free_path(path
);
5491 static noinline
struct inode
*create_reloc_inode(struct btrfs_fs_info
*fs_info
,
5492 struct btrfs_block_group_cache
*group
)
5494 struct inode
*inode
= NULL
;
5495 struct btrfs_trans_handle
*trans
;
5496 struct btrfs_root
*root
;
5497 struct btrfs_key root_key
;
5498 u64 objectid
= BTRFS_FIRST_FREE_OBJECTID
;
5501 root_key
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
5502 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
5503 root_key
.offset
= (u64
)-1;
5504 root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
5506 return ERR_CAST(root
);
5508 trans
= btrfs_start_transaction(root
, 1);
5511 err
= btrfs_find_free_objectid(trans
, root
, objectid
, &objectid
);
5515 err
= __insert_orphan_inode(trans
, root
, objectid
, group
->key
.offset
);
5518 err
= btrfs_insert_file_extent(trans
, root
, objectid
, 0, 0, 0,
5519 group
->key
.offset
, 0, group
->key
.offset
,
5523 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
5524 if (inode
->i_state
& I_NEW
) {
5525 BTRFS_I(inode
)->root
= root
;
5526 BTRFS_I(inode
)->location
.objectid
= objectid
;
5527 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
5528 BTRFS_I(inode
)->location
.offset
= 0;
5529 btrfs_read_locked_inode(inode
);
5530 unlock_new_inode(inode
);
5531 BUG_ON(is_bad_inode(inode
));
5535 BTRFS_I(inode
)->index_cnt
= group
->key
.objectid
;
5537 err
= btrfs_orphan_add(trans
, inode
);
5539 btrfs_end_transaction(trans
, root
);
5543 inode
= ERR_PTR(err
);
5548 int btrfs_reloc_clone_csums(struct inode
*inode
, u64 file_pos
, u64 len
)
5551 struct btrfs_ordered_sum
*sums
;
5552 struct btrfs_sector_sum
*sector_sum
;
5553 struct btrfs_ordered_extent
*ordered
;
5554 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5555 struct list_head list
;
5560 INIT_LIST_HEAD(&list
);
5562 ordered
= btrfs_lookup_ordered_extent(inode
, file_pos
);
5563 BUG_ON(ordered
->file_offset
!= file_pos
|| ordered
->len
!= len
);
5565 disk_bytenr
= file_pos
+ BTRFS_I(inode
)->index_cnt
;
5566 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, disk_bytenr
,
5567 disk_bytenr
+ len
- 1, &list
);
5569 while (!list_empty(&list
)) {
5570 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
5571 list_del_init(&sums
->list
);
5573 sector_sum
= sums
->sums
;
5574 sums
->bytenr
= ordered
->start
;
5577 while (offset
< sums
->len
) {
5578 sector_sum
->bytenr
+= ordered
->start
- disk_bytenr
;
5580 offset
+= root
->sectorsize
;
5583 btrfs_add_ordered_sum(inode
, ordered
, sums
);
5585 btrfs_put_ordered_extent(ordered
);
5589 int btrfs_relocate_block_group(struct btrfs_root
*root
, u64 group_start
)
5591 struct btrfs_trans_handle
*trans
;
5592 struct btrfs_path
*path
;
5593 struct btrfs_fs_info
*info
= root
->fs_info
;
5594 struct extent_buffer
*leaf
;
5595 struct inode
*reloc_inode
;
5596 struct btrfs_block_group_cache
*block_group
;
5597 struct btrfs_key key
;
5606 root
= root
->fs_info
->extent_root
;
5608 block_group
= btrfs_lookup_block_group(info
, group_start
);
5609 BUG_ON(!block_group
);
5611 printk(KERN_INFO
"btrfs relocating block group %llu flags %llu\n",
5612 (unsigned long long)block_group
->key
.objectid
,
5613 (unsigned long long)block_group
->flags
);
5615 path
= btrfs_alloc_path();
5618 reloc_inode
= create_reloc_inode(info
, block_group
);
5619 BUG_ON(IS_ERR(reloc_inode
));
5621 __alloc_chunk_for_shrink(root
, block_group
, 1);
5622 set_block_group_readonly(block_group
);
5624 btrfs_start_delalloc_inodes(info
->tree_root
);
5625 btrfs_wait_ordered_extents(info
->tree_root
, 0);
5630 key
.objectid
= block_group
->key
.objectid
;
5633 cur_byte
= key
.objectid
;
5635 trans
= btrfs_start_transaction(info
->tree_root
, 1);
5636 btrfs_commit_transaction(trans
, info
->tree_root
);
5638 mutex_lock(&root
->fs_info
->cleaner_mutex
);
5639 btrfs_clean_old_snapshots(info
->tree_root
);
5640 btrfs_remove_leaf_refs(info
->tree_root
, (u64
)-1, 1);
5641 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
5643 trans
= btrfs_start_transaction(info
->tree_root
, 1);
5644 btrfs_commit_transaction(trans
, info
->tree_root
);
5647 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5651 leaf
= path
->nodes
[0];
5652 nritems
= btrfs_header_nritems(leaf
);
5653 if (path
->slots
[0] >= nritems
) {
5654 ret
= btrfs_next_leaf(root
, path
);
5661 leaf
= path
->nodes
[0];
5662 nritems
= btrfs_header_nritems(leaf
);
5665 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
5667 if (key
.objectid
>= block_group
->key
.objectid
+
5668 block_group
->key
.offset
)
5671 if (progress
&& need_resched()) {
5672 btrfs_release_path(root
, path
);
5679 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
||
5680 key
.objectid
+ key
.offset
<= cur_byte
) {
5686 cur_byte
= key
.objectid
+ key
.offset
;
5687 btrfs_release_path(root
, path
);
5689 __alloc_chunk_for_shrink(root
, block_group
, 0);
5690 ret
= relocate_one_extent(root
, path
, &key
, block_group
,
5696 key
.objectid
= cur_byte
;
5701 btrfs_release_path(root
, path
);
5704 btrfs_wait_ordered_range(reloc_inode
, 0, (u64
)-1);
5705 invalidate_mapping_pages(reloc_inode
->i_mapping
, 0, -1);
5708 if (total_found
> 0) {
5709 printk(KERN_INFO
"btrfs found %llu extents in pass %d\n",
5710 (unsigned long long)total_found
, pass
);
5712 if (total_found
== skipped
&& pass
> 2) {
5714 reloc_inode
= create_reloc_inode(info
, block_group
);
5720 /* delete reloc_inode */
5723 /* unpin extents in this range */
5724 trans
= btrfs_start_transaction(info
->tree_root
, 1);
5725 btrfs_commit_transaction(trans
, info
->tree_root
);
5727 spin_lock(&block_group
->lock
);
5728 WARN_ON(block_group
->pinned
> 0);
5729 WARN_ON(block_group
->reserved
> 0);
5730 WARN_ON(btrfs_block_group_used(&block_group
->item
) > 0);
5731 spin_unlock(&block_group
->lock
);
5732 put_block_group(block_group
);
5735 btrfs_free_path(path
);
5739 static int find_first_block_group(struct btrfs_root
*root
,
5740 struct btrfs_path
*path
, struct btrfs_key
*key
)
5743 struct btrfs_key found_key
;
5744 struct extent_buffer
*leaf
;
5747 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
5752 slot
= path
->slots
[0];
5753 leaf
= path
->nodes
[0];
5754 if (slot
>= btrfs_header_nritems(leaf
)) {
5755 ret
= btrfs_next_leaf(root
, path
);
5762 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5764 if (found_key
.objectid
>= key
->objectid
&&
5765 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
5776 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
5778 struct btrfs_block_group_cache
*block_group
;
5779 struct btrfs_space_info
*space_info
;
5782 spin_lock(&info
->block_group_cache_lock
);
5783 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
5784 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
5786 rb_erase(&block_group
->cache_node
,
5787 &info
->block_group_cache_tree
);
5788 spin_unlock(&info
->block_group_cache_lock
);
5790 btrfs_remove_free_space_cache(block_group
);
5791 down_write(&block_group
->space_info
->groups_sem
);
5792 list_del(&block_group
->list
);
5793 up_write(&block_group
->space_info
->groups_sem
);
5795 WARN_ON(atomic_read(&block_group
->count
) != 1);
5798 spin_lock(&info
->block_group_cache_lock
);
5800 spin_unlock(&info
->block_group_cache_lock
);
5802 /* now that all the block groups are freed, go through and
5803 * free all the space_info structs. This is only called during
5804 * the final stages of unmount, and so we know nobody is
5805 * using them. We call synchronize_rcu() once before we start,
5806 * just to be on the safe side.
5810 while(!list_empty(&info
->space_info
)) {
5811 space_info
= list_entry(info
->space_info
.next
,
5812 struct btrfs_space_info
,
5815 list_del(&space_info
->list
);
5821 int btrfs_read_block_groups(struct btrfs_root
*root
)
5823 struct btrfs_path
*path
;
5825 struct btrfs_block_group_cache
*cache
;
5826 struct btrfs_fs_info
*info
= root
->fs_info
;
5827 struct btrfs_space_info
*space_info
;
5828 struct btrfs_key key
;
5829 struct btrfs_key found_key
;
5830 struct extent_buffer
*leaf
;
5832 root
= info
->extent_root
;
5835 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
5836 path
= btrfs_alloc_path();
5841 ret
= find_first_block_group(root
, path
, &key
);
5849 leaf
= path
->nodes
[0];
5850 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5851 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5857 atomic_set(&cache
->count
, 1);
5858 spin_lock_init(&cache
->lock
);
5859 mutex_init(&cache
->alloc_mutex
);
5860 mutex_init(&cache
->cache_mutex
);
5861 INIT_LIST_HEAD(&cache
->list
);
5862 read_extent_buffer(leaf
, &cache
->item
,
5863 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
5864 sizeof(cache
->item
));
5865 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
5867 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
5868 btrfs_release_path(root
, path
);
5869 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
5871 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
5872 btrfs_block_group_used(&cache
->item
),
5875 cache
->space_info
= space_info
;
5876 down_write(&space_info
->groups_sem
);
5877 list_add_tail(&cache
->list
, &space_info
->block_groups
);
5878 up_write(&space_info
->groups_sem
);
5880 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5883 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
5884 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
5885 set_block_group_readonly(cache
);
5889 btrfs_free_path(path
);
5893 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
5894 struct btrfs_root
*root
, u64 bytes_used
,
5895 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
5899 struct btrfs_root
*extent_root
;
5900 struct btrfs_block_group_cache
*cache
;
5902 extent_root
= root
->fs_info
->extent_root
;
5904 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
5906 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5910 cache
->key
.objectid
= chunk_offset
;
5911 cache
->key
.offset
= size
;
5912 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
5913 atomic_set(&cache
->count
, 1);
5914 spin_lock_init(&cache
->lock
);
5915 mutex_init(&cache
->alloc_mutex
);
5916 mutex_init(&cache
->cache_mutex
);
5917 INIT_LIST_HEAD(&cache
->list
);
5919 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
5920 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
5921 cache
->flags
= type
;
5922 btrfs_set_block_group_flags(&cache
->item
, type
);
5924 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
5925 &cache
->space_info
);
5927 down_write(&cache
->space_info
->groups_sem
);
5928 list_add_tail(&cache
->list
, &cache
->space_info
->block_groups
);
5929 up_write(&cache
->space_info
->groups_sem
);
5931 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5934 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
5935 sizeof(cache
->item
));
5938 set_avail_alloc_bits(extent_root
->fs_info
, type
);
5943 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
5944 struct btrfs_root
*root
, u64 group_start
)
5946 struct btrfs_path
*path
;
5947 struct btrfs_block_group_cache
*block_group
;
5948 struct btrfs_key key
;
5951 root
= root
->fs_info
->extent_root
;
5953 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
5954 BUG_ON(!block_group
);
5955 BUG_ON(!block_group
->ro
);
5957 memcpy(&key
, &block_group
->key
, sizeof(key
));
5959 path
= btrfs_alloc_path();
5962 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5963 rb_erase(&block_group
->cache_node
,
5964 &root
->fs_info
->block_group_cache_tree
);
5965 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5966 btrfs_remove_free_space_cache(block_group
);
5967 down_write(&block_group
->space_info
->groups_sem
);
5968 list_del(&block_group
->list
);
5969 up_write(&block_group
->space_info
->groups_sem
);
5971 spin_lock(&block_group
->space_info
->lock
);
5972 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
5973 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
5974 spin_unlock(&block_group
->space_info
->lock
);
5975 block_group
->space_info
->full
= 0;
5977 put_block_group(block_group
);
5978 put_block_group(block_group
);
5980 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
5986 ret
= btrfs_del_item(trans
, root
, path
);
5988 btrfs_free_path(path
);