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>
27 #include "print-tree.h"
28 #include "transaction.h"
31 #include "ref-cache.h"
34 #define PENDING_EXTENT_INSERT 0
35 #define PENDING_EXTENT_DELETE 1
36 #define PENDING_BACKREF_UPDATE 2
38 struct pending_extent_op
{
47 struct list_head list
;
51 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
52 struct btrfs_root
*extent_root
, int all
);
53 static int del_pending_extents(struct btrfs_trans_handle
*trans
,
54 struct btrfs_root
*extent_root
, int all
);
55 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 bytenr
, u64 num_bytes
, int is_data
);
58 static int update_block_group(struct btrfs_trans_handle
*trans
,
59 struct btrfs_root
*root
,
60 u64 bytenr
, u64 num_bytes
, int alloc
,
63 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
65 return (cache
->flags
& bits
) == bits
;
69 * this adds the block group to the fs_info rb tree for the block group
72 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
73 struct btrfs_block_group_cache
*block_group
)
76 struct rb_node
*parent
= NULL
;
77 struct btrfs_block_group_cache
*cache
;
79 spin_lock(&info
->block_group_cache_lock
);
80 p
= &info
->block_group_cache_tree
.rb_node
;
84 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
86 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
88 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
91 spin_unlock(&info
->block_group_cache_lock
);
96 rb_link_node(&block_group
->cache_node
, parent
, p
);
97 rb_insert_color(&block_group
->cache_node
,
98 &info
->block_group_cache_tree
);
99 spin_unlock(&info
->block_group_cache_lock
);
105 * This will return the block group at or after bytenr if contains is 0, else
106 * it will return the block group that contains the bytenr
108 static struct btrfs_block_group_cache
*
109 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
112 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
116 spin_lock(&info
->block_group_cache_lock
);
117 n
= info
->block_group_cache_tree
.rb_node
;
120 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
122 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
123 start
= cache
->key
.objectid
;
125 if (bytenr
< start
) {
126 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
129 } else if (bytenr
> start
) {
130 if (contains
&& bytenr
<= end
) {
141 atomic_inc(&ret
->count
);
142 spin_unlock(&info
->block_group_cache_lock
);
148 * this is only called by cache_block_group, since we could have freed extents
149 * we need to check the pinned_extents for any extents that can't be used yet
150 * since their free space will be released as soon as the transaction commits.
152 static int add_new_free_space(struct btrfs_block_group_cache
*block_group
,
153 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
155 u64 extent_start
, extent_end
, size
;
158 mutex_lock(&info
->pinned_mutex
);
159 while (start
< end
) {
160 ret
= find_first_extent_bit(&info
->pinned_extents
, start
,
161 &extent_start
, &extent_end
,
166 if (extent_start
== start
) {
167 start
= extent_end
+ 1;
168 } else if (extent_start
> start
&& extent_start
< end
) {
169 size
= extent_start
- start
;
170 ret
= btrfs_add_free_space(block_group
, start
,
173 start
= extent_end
+ 1;
181 ret
= btrfs_add_free_space(block_group
, start
, size
);
184 mutex_unlock(&info
->pinned_mutex
);
189 static int remove_sb_from_cache(struct btrfs_root
*root
,
190 struct btrfs_block_group_cache
*cache
)
197 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
198 bytenr
= btrfs_sb_offset(i
);
199 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
200 cache
->key
.objectid
, bytenr
, 0,
201 &logical
, &nr
, &stripe_len
);
204 btrfs_remove_free_space(cache
, logical
[nr
],
212 static int cache_block_group(struct btrfs_root
*root
,
213 struct btrfs_block_group_cache
*block_group
)
215 struct btrfs_path
*path
;
217 struct btrfs_key key
;
218 struct extent_buffer
*leaf
;
225 root
= root
->fs_info
->extent_root
;
227 if (block_group
->cached
)
230 path
= btrfs_alloc_path();
236 * we get into deadlocks with paths held by callers of this function.
237 * since the alloc_mutex is protecting things right now, just
238 * skip the locking here
240 path
->skip_locking
= 1;
241 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
244 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
245 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
250 leaf
= path
->nodes
[0];
251 slot
= path
->slots
[0];
252 if (slot
>= btrfs_header_nritems(leaf
)) {
253 ret
= btrfs_next_leaf(root
, path
);
261 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
262 if (key
.objectid
< block_group
->key
.objectid
)
265 if (key
.objectid
>= block_group
->key
.objectid
+
266 block_group
->key
.offset
)
269 if (btrfs_key_type(&key
) == BTRFS_EXTENT_ITEM_KEY
) {
270 add_new_free_space(block_group
, root
->fs_info
, last
,
273 last
= key
.objectid
+ key
.offset
;
279 add_new_free_space(block_group
, root
->fs_info
, last
,
280 block_group
->key
.objectid
+
281 block_group
->key
.offset
);
283 remove_sb_from_cache(root
, block_group
);
284 block_group
->cached
= 1;
287 btrfs_free_path(path
);
292 * return the block group that starts at or after bytenr
294 static struct btrfs_block_group_cache
*
295 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
297 struct btrfs_block_group_cache
*cache
;
299 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
305 * return the block group that contains teh given bytenr
307 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
308 struct btrfs_fs_info
*info
,
311 struct btrfs_block_group_cache
*cache
;
313 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
318 static inline void put_block_group(struct btrfs_block_group_cache
*cache
)
320 if (atomic_dec_and_test(&cache
->count
))
324 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
327 struct list_head
*head
= &info
->space_info
;
328 struct list_head
*cur
;
329 struct btrfs_space_info
*found
;
330 list_for_each(cur
, head
) {
331 found
= list_entry(cur
, struct btrfs_space_info
, list
);
332 if (found
->flags
== flags
)
338 static u64
div_factor(u64 num
, int factor
)
347 u64
btrfs_find_block_group(struct btrfs_root
*root
,
348 u64 search_start
, u64 search_hint
, int owner
)
350 struct btrfs_block_group_cache
*cache
;
352 u64 last
= max(search_hint
, search_start
);
359 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
363 spin_lock(&cache
->lock
);
364 last
= cache
->key
.objectid
+ cache
->key
.offset
;
365 used
= btrfs_block_group_used(&cache
->item
);
367 if ((full_search
|| !cache
->ro
) &&
368 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
369 if (used
+ cache
->pinned
+ cache
->reserved
<
370 div_factor(cache
->key
.offset
, factor
)) {
371 group_start
= cache
->key
.objectid
;
372 spin_unlock(&cache
->lock
);
373 put_block_group(cache
);
377 spin_unlock(&cache
->lock
);
378 put_block_group(cache
);
386 if (!full_search
&& factor
< 10) {
396 /* simple helper to search for an existing extent at a given offset */
397 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
400 struct btrfs_key key
;
401 struct btrfs_path
*path
;
403 path
= btrfs_alloc_path();
405 key
.objectid
= start
;
407 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
408 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
410 btrfs_free_path(path
);
415 * Back reference rules. Back refs have three main goals:
417 * 1) differentiate between all holders of references to an extent so that
418 * when a reference is dropped we can make sure it was a valid reference
419 * before freeing the extent.
421 * 2) Provide enough information to quickly find the holders of an extent
422 * if we notice a given block is corrupted or bad.
424 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
425 * maintenance. This is actually the same as #2, but with a slightly
426 * different use case.
428 * File extents can be referenced by:
430 * - multiple snapshots, subvolumes, or different generations in one subvol
431 * - different files inside a single subvolume
432 * - different offsets inside a file (bookend extents in file.c)
434 * The extent ref structure has fields for:
436 * - Objectid of the subvolume root
437 * - Generation number of the tree holding the reference
438 * - objectid of the file holding the reference
439 * - number of references holding by parent node (alway 1 for tree blocks)
441 * Btree leaf may hold multiple references to a file extent. In most cases,
442 * these references are from same file and the corresponding offsets inside
443 * the file are close together.
445 * When a file extent is allocated the fields are filled in:
446 * (root_key.objectid, trans->transid, inode objectid, 1)
448 * When a leaf is cow'd new references are added for every file extent found
449 * in the leaf. It looks similar to the create case, but trans->transid will
450 * be different when the block is cow'd.
452 * (root_key.objectid, trans->transid, inode objectid,
453 * number of references in the leaf)
455 * When a file extent is removed either during snapshot deletion or
456 * file truncation, we find the corresponding back reference and check
457 * the following fields:
459 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
462 * Btree extents can be referenced by:
464 * - Different subvolumes
465 * - Different generations of the same subvolume
467 * When a tree block is created, back references are inserted:
469 * (root->root_key.objectid, trans->transid, level, 1)
471 * When a tree block is cow'd, new back references are added for all the
472 * blocks it points to. If the tree block isn't in reference counted root,
473 * the old back references are removed. These new back references are of
474 * the form (trans->transid will have increased since creation):
476 * (root->root_key.objectid, trans->transid, level, 1)
478 * When a backref is in deleting, the following fields are checked:
480 * if backref was for a tree root:
481 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
483 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
485 * Back Reference Key composing:
487 * The key objectid corresponds to the first byte in the extent, the key
488 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
489 * byte of parent extent. If a extent is tree root, the key offset is set
490 * to the key objectid.
493 static noinline
int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
494 struct btrfs_root
*root
,
495 struct btrfs_path
*path
,
496 u64 bytenr
, u64 parent
,
497 u64 ref_root
, u64 ref_generation
,
498 u64 owner_objectid
, int del
)
500 struct btrfs_key key
;
501 struct btrfs_extent_ref
*ref
;
502 struct extent_buffer
*leaf
;
506 key
.objectid
= bytenr
;
507 key
.type
= BTRFS_EXTENT_REF_KEY
;
510 ret
= btrfs_search_slot(trans
, root
, &key
, path
, del
? -1 : 0, 1);
518 leaf
= path
->nodes
[0];
519 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
520 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
521 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
522 btrfs_ref_generation(leaf
, ref
) != ref_generation
||
523 (ref_objectid
!= owner_objectid
&&
524 ref_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
)) {
535 * updates all the backrefs that are pending on update_list for the
538 static noinline
int update_backrefs(struct btrfs_trans_handle
*trans
,
539 struct btrfs_root
*extent_root
,
540 struct btrfs_path
*path
,
541 struct list_head
*update_list
)
543 struct btrfs_key key
;
544 struct btrfs_extent_ref
*ref
;
545 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
546 struct pending_extent_op
*op
;
547 struct extent_buffer
*leaf
;
549 struct list_head
*cur
= update_list
->next
;
551 u64 ref_root
= extent_root
->root_key
.objectid
;
553 op
= list_entry(cur
, struct pending_extent_op
, list
);
556 key
.objectid
= op
->bytenr
;
557 key
.type
= BTRFS_EXTENT_REF_KEY
;
558 key
.offset
= op
->orig_parent
;
560 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 1);
563 leaf
= path
->nodes
[0];
566 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
568 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
570 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
571 btrfs_ref_generation(leaf
, ref
) != op
->orig_generation
||
572 (ref_objectid
!= op
->level
&&
573 ref_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
)) {
574 printk(KERN_ERR
"btrfs couldn't find %llu, parent %llu, "
575 "root %llu, owner %u\n",
576 (unsigned long long)op
->bytenr
,
577 (unsigned long long)op
->orig_parent
,
578 (unsigned long long)ref_root
, op
->level
);
579 btrfs_print_leaf(extent_root
, leaf
);
583 key
.objectid
= op
->bytenr
;
584 key
.offset
= op
->parent
;
585 key
.type
= BTRFS_EXTENT_REF_KEY
;
586 ret
= btrfs_set_item_key_safe(trans
, extent_root
, path
, &key
);
588 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
589 btrfs_set_ref_generation(leaf
, ref
, op
->generation
);
593 list_del_init(&op
->list
);
594 unlock_extent(&info
->extent_ins
, op
->bytenr
,
595 op
->bytenr
+ op
->num_bytes
- 1, GFP_NOFS
);
598 if (cur
== update_list
) {
599 btrfs_mark_buffer_dirty(path
->nodes
[0]);
600 btrfs_release_path(extent_root
, path
);
604 op
= list_entry(cur
, struct pending_extent_op
, list
);
607 while (path
->slots
[0] < btrfs_header_nritems(leaf
)) {
608 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
609 if (key
.objectid
== op
->bytenr
&&
610 key
.type
== BTRFS_EXTENT_REF_KEY
)
615 btrfs_mark_buffer_dirty(path
->nodes
[0]);
616 btrfs_release_path(extent_root
, path
);
623 static noinline
int insert_extents(struct btrfs_trans_handle
*trans
,
624 struct btrfs_root
*extent_root
,
625 struct btrfs_path
*path
,
626 struct list_head
*insert_list
, int nr
)
628 struct btrfs_key
*keys
;
630 struct pending_extent_op
*op
;
631 struct extent_buffer
*leaf
;
632 struct list_head
*cur
= insert_list
->next
;
633 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
634 u64 ref_root
= extent_root
->root_key
.objectid
;
635 int i
= 0, last
= 0, ret
;
641 keys
= kzalloc(total
* sizeof(struct btrfs_key
), GFP_NOFS
);
645 data_size
= kzalloc(total
* sizeof(u32
), GFP_NOFS
);
651 list_for_each_entry(op
, insert_list
, list
) {
652 keys
[i
].objectid
= op
->bytenr
;
653 keys
[i
].offset
= op
->num_bytes
;
654 keys
[i
].type
= BTRFS_EXTENT_ITEM_KEY
;
655 data_size
[i
] = sizeof(struct btrfs_extent_item
);
658 keys
[i
].objectid
= op
->bytenr
;
659 keys
[i
].offset
= op
->parent
;
660 keys
[i
].type
= BTRFS_EXTENT_REF_KEY
;
661 data_size
[i
] = sizeof(struct btrfs_extent_ref
);
665 op
= list_entry(cur
, struct pending_extent_op
, list
);
669 ret
= btrfs_insert_some_items(trans
, extent_root
, path
,
670 keys
+i
, data_size
+i
, total
-i
);
676 leaf
= path
->nodes
[0];
677 for (c
= 0; c
< ret
; c
++) {
678 int ref_first
= keys
[i
].type
== BTRFS_EXTENT_REF_KEY
;
681 * if the first item we inserted was a backref, then
682 * the EXTENT_ITEM will be the odd c's, else it will
685 if ((ref_first
&& (c
% 2)) ||
686 (!ref_first
&& !(c
% 2))) {
687 struct btrfs_extent_item
*itm
;
689 itm
= btrfs_item_ptr(leaf
, path
->slots
[0] + c
,
690 struct btrfs_extent_item
);
691 btrfs_set_extent_refs(path
->nodes
[0], itm
, 1);
694 struct btrfs_extent_ref
*ref
;
696 ref
= btrfs_item_ptr(leaf
, path
->slots
[0] + c
,
697 struct btrfs_extent_ref
);
698 btrfs_set_ref_root(leaf
, ref
, ref_root
);
699 btrfs_set_ref_generation(leaf
, ref
,
701 btrfs_set_ref_objectid(leaf
, ref
, op
->level
);
702 btrfs_set_ref_num_refs(leaf
, ref
, 1);
707 * using del to see when its ok to free up the
708 * pending_extent_op. In the case where we insert the
709 * last item on the list in order to help do batching
710 * we need to not free the extent op until we actually
711 * insert the extent_item
714 unlock_extent(&info
->extent_ins
, op
->bytenr
,
715 op
->bytenr
+ op
->num_bytes
- 1,
718 list_del_init(&op
->list
);
720 if (cur
!= insert_list
)
722 struct pending_extent_op
,
726 btrfs_mark_buffer_dirty(leaf
);
727 btrfs_release_path(extent_root
, path
);
730 * Ok backref's and items usually go right next to eachother,
731 * but if we could only insert 1 item that means that we
732 * inserted on the end of a leaf, and we have no idea what may
733 * be on the next leaf so we just play it safe. In order to
734 * try and help this case we insert the last thing on our
735 * insert list so hopefully it will end up being the last
736 * thing on the leaf and everything else will be before it,
737 * which will let us insert a whole bunch of items at the same
740 if (ret
== 1 && !last
&& (i
+ ret
< total
)) {
742 * last: where we will pick up the next time around
743 * i: our current key to insert, will be total - 1
744 * cur: the current op we are screwing with
749 cur
= insert_list
->prev
;
750 op
= list_entry(cur
, struct pending_extent_op
, list
);
753 * ok we successfully inserted the last item on the
754 * list, lets reset everything
756 * i: our current key to insert, so where we left off
758 * last: done with this
759 * cur: the op we are messing with
761 * total: since we inserted the last key, we need to
762 * decrement total so we dont overflow
768 cur
= insert_list
->next
;
769 op
= list_entry(cur
, struct pending_extent_op
,
784 static noinline
int insert_extent_backref(struct btrfs_trans_handle
*trans
,
785 struct btrfs_root
*root
,
786 struct btrfs_path
*path
,
787 u64 bytenr
, u64 parent
,
788 u64 ref_root
, u64 ref_generation
,
791 struct btrfs_key key
;
792 struct extent_buffer
*leaf
;
793 struct btrfs_extent_ref
*ref
;
797 key
.objectid
= bytenr
;
798 key
.type
= BTRFS_EXTENT_REF_KEY
;
801 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(*ref
));
803 leaf
= path
->nodes
[0];
804 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
805 struct btrfs_extent_ref
);
806 btrfs_set_ref_root(leaf
, ref
, ref_root
);
807 btrfs_set_ref_generation(leaf
, ref
, ref_generation
);
808 btrfs_set_ref_objectid(leaf
, ref
, owner_objectid
);
809 btrfs_set_ref_num_refs(leaf
, ref
, 1);
810 } else if (ret
== -EEXIST
) {
812 BUG_ON(owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
);
813 leaf
= path
->nodes
[0];
814 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
815 struct btrfs_extent_ref
);
816 if (btrfs_ref_root(leaf
, ref
) != ref_root
||
817 btrfs_ref_generation(leaf
, ref
) != ref_generation
) {
823 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
824 BUG_ON(num_refs
== 0);
825 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
+ 1);
827 existing_owner
= btrfs_ref_objectid(leaf
, ref
);
828 if (existing_owner
!= owner_objectid
&&
829 existing_owner
!= BTRFS_MULTIPLE_OBJECTIDS
) {
830 btrfs_set_ref_objectid(leaf
, ref
,
831 BTRFS_MULTIPLE_OBJECTIDS
);
837 btrfs_mark_buffer_dirty(path
->nodes
[0]);
839 btrfs_release_path(root
, path
);
843 static noinline
int remove_extent_backref(struct btrfs_trans_handle
*trans
,
844 struct btrfs_root
*root
,
845 struct btrfs_path
*path
)
847 struct extent_buffer
*leaf
;
848 struct btrfs_extent_ref
*ref
;
852 leaf
= path
->nodes
[0];
853 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_ref
);
854 num_refs
= btrfs_ref_num_refs(leaf
, ref
);
855 BUG_ON(num_refs
== 0);
858 ret
= btrfs_del_item(trans
, root
, path
);
860 btrfs_set_ref_num_refs(leaf
, ref
, num_refs
);
861 btrfs_mark_buffer_dirty(leaf
);
863 btrfs_release_path(root
, path
);
867 #ifdef BIO_RW_DISCARD
868 static void btrfs_issue_discard(struct block_device
*bdev
,
871 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
);
875 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
878 #ifdef BIO_RW_DISCARD
880 u64 map_length
= num_bytes
;
881 struct btrfs_multi_bio
*multi
= NULL
;
883 /* Tell the block device(s) that the sectors can be discarded */
884 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
885 bytenr
, &map_length
, &multi
, 0);
887 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
890 if (map_length
> num_bytes
)
891 map_length
= num_bytes
;
893 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
894 btrfs_issue_discard(stripe
->dev
->bdev
,
907 static noinline
int free_extents(struct btrfs_trans_handle
*trans
,
908 struct btrfs_root
*extent_root
,
909 struct list_head
*del_list
)
911 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
912 struct btrfs_path
*path
;
913 struct btrfs_key key
, found_key
;
914 struct extent_buffer
*leaf
;
915 struct list_head
*cur
;
916 struct pending_extent_op
*op
;
917 struct btrfs_extent_item
*ei
;
918 int ret
, num_to_del
, extent_slot
= 0, found_extent
= 0;
922 path
= btrfs_alloc_path();
928 /* search for the backref for the current ref we want to delete */
929 cur
= del_list
->next
;
930 op
= list_entry(cur
, struct pending_extent_op
, list
);
931 ret
= lookup_extent_backref(trans
, extent_root
, path
, op
->bytenr
,
933 extent_root
->root_key
.objectid
,
934 op
->orig_generation
, op
->level
, 1);
936 printk(KERN_ERR
"btrfs unable to find backref byte nr %llu "
937 "root %llu gen %llu owner %u\n",
938 (unsigned long long)op
->bytenr
,
939 (unsigned long long)extent_root
->root_key
.objectid
,
940 (unsigned long long)op
->orig_generation
, op
->level
);
941 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
946 extent_slot
= path
->slots
[0];
951 * if we aren't the first item on the leaf we can move back one and see
952 * if our ref is right next to our extent item
954 if (likely(extent_slot
)) {
956 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
958 if (found_key
.objectid
== op
->bytenr
&&
959 found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
960 found_key
.offset
== op
->num_bytes
) {
967 * if we didn't find the extent we need to delete the backref and then
968 * search for the extent item key so we can update its ref count
971 key
.objectid
= op
->bytenr
;
972 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
973 key
.offset
= op
->num_bytes
;
975 ret
= remove_extent_backref(trans
, extent_root
, path
);
977 btrfs_release_path(extent_root
, path
);
978 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, -1, 1);
980 extent_slot
= path
->slots
[0];
983 /* this is where we update the ref count for the extent */
984 leaf
= path
->nodes
[0];
985 ei
= btrfs_item_ptr(leaf
, extent_slot
, struct btrfs_extent_item
);
986 refs
= btrfs_extent_refs(leaf
, ei
);
989 btrfs_set_extent_refs(leaf
, ei
, refs
);
991 btrfs_mark_buffer_dirty(leaf
);
994 * This extent needs deleting. The reason cur_slot is extent_slot +
995 * num_to_del is because extent_slot points to the slot where the extent
996 * is, and if the backref was not right next to the extent we will be
997 * deleting at least 1 item, and will want to start searching at the
998 * slot directly next to extent_slot. However if we did find the
999 * backref next to the extent item them we will be deleting at least 2
1000 * items and will want to start searching directly after the ref slot
1003 struct list_head
*pos
, *n
, *end
;
1004 int cur_slot
= extent_slot
+num_to_del
;
1008 path
->slots
[0] = extent_slot
;
1009 bytes_freed
= op
->num_bytes
;
1011 mutex_lock(&info
->pinned_mutex
);
1012 ret
= pin_down_bytes(trans
, extent_root
, op
->bytenr
,
1013 op
->num_bytes
, op
->level
>=
1014 BTRFS_FIRST_FREE_OBJECTID
);
1015 mutex_unlock(&info
->pinned_mutex
);
1020 * we need to see if we can delete multiple things at once, so
1021 * start looping through the list of extents we are wanting to
1022 * delete and see if their extent/backref's are right next to
1023 * eachother and the extents only have 1 ref
1025 for (pos
= cur
->next
; pos
!= del_list
; pos
= pos
->next
) {
1026 struct pending_extent_op
*tmp
;
1028 tmp
= list_entry(pos
, struct pending_extent_op
, list
);
1030 /* we only want to delete extent+ref at this stage */
1031 if (cur_slot
>= btrfs_header_nritems(leaf
) - 1)
1034 btrfs_item_key_to_cpu(leaf
, &found_key
, cur_slot
);
1035 if (found_key
.objectid
!= tmp
->bytenr
||
1036 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
||
1037 found_key
.offset
!= tmp
->num_bytes
)
1040 /* check to make sure this extent only has one ref */
1041 ei
= btrfs_item_ptr(leaf
, cur_slot
,
1042 struct btrfs_extent_item
);
1043 if (btrfs_extent_refs(leaf
, ei
) != 1)
1046 btrfs_item_key_to_cpu(leaf
, &found_key
, cur_slot
+1);
1047 if (found_key
.objectid
!= tmp
->bytenr
||
1048 found_key
.type
!= BTRFS_EXTENT_REF_KEY
||
1049 found_key
.offset
!= tmp
->orig_parent
)
1053 * the ref is right next to the extent, we can set the
1054 * ref count to 0 since we will delete them both now
1056 btrfs_set_extent_refs(leaf
, ei
, 0);
1058 /* pin down the bytes for this extent */
1059 mutex_lock(&info
->pinned_mutex
);
1060 ret
= pin_down_bytes(trans
, extent_root
, tmp
->bytenr
,
1061 tmp
->num_bytes
, tmp
->level
>=
1062 BTRFS_FIRST_FREE_OBJECTID
);
1063 mutex_unlock(&info
->pinned_mutex
);
1067 * use the del field to tell if we need to go ahead and
1068 * free up the extent when we delete the item or not.
1071 bytes_freed
+= tmp
->num_bytes
;
1078 /* update the free space counters */
1079 spin_lock(&info
->delalloc_lock
);
1080 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
1081 btrfs_set_super_bytes_used(&info
->super_copy
,
1082 super_used
- bytes_freed
);
1084 root_used
= btrfs_root_used(&extent_root
->root_item
);
1085 btrfs_set_root_used(&extent_root
->root_item
,
1086 root_used
- bytes_freed
);
1087 spin_unlock(&info
->delalloc_lock
);
1089 /* delete the items */
1090 ret
= btrfs_del_items(trans
, extent_root
, path
,
1091 path
->slots
[0], num_to_del
);
1095 * loop through the extents we deleted and do the cleanup work
1098 for (pos
= cur
, n
= pos
->next
; pos
!= end
;
1099 pos
= n
, n
= pos
->next
) {
1100 struct pending_extent_op
*tmp
;
1101 tmp
= list_entry(pos
, struct pending_extent_op
, list
);
1104 * remember tmp->del tells us wether or not we pinned
1107 ret
= update_block_group(trans
, extent_root
,
1108 tmp
->bytenr
, tmp
->num_bytes
, 0,
1112 list_del_init(&tmp
->list
);
1113 unlock_extent(&info
->extent_ins
, tmp
->bytenr
,
1114 tmp
->bytenr
+ tmp
->num_bytes
- 1,
1118 } else if (refs
&& found_extent
) {
1120 * the ref and extent were right next to eachother, but the
1121 * extent still has a ref, so just free the backref and keep
1124 ret
= remove_extent_backref(trans
, extent_root
, path
);
1127 list_del_init(&op
->list
);
1128 unlock_extent(&info
->extent_ins
, op
->bytenr
,
1129 op
->bytenr
+ op
->num_bytes
- 1, GFP_NOFS
);
1133 * the extent has multiple refs and the backref we were looking
1134 * for was not right next to it, so just unlock and go next,
1137 list_del_init(&op
->list
);
1138 unlock_extent(&info
->extent_ins
, op
->bytenr
,
1139 op
->bytenr
+ op
->num_bytes
- 1, GFP_NOFS
);
1143 btrfs_release_path(extent_root
, path
);
1144 if (!list_empty(del_list
))
1148 btrfs_free_path(path
);
1152 static int __btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
1153 struct btrfs_root
*root
, u64 bytenr
,
1154 u64 orig_parent
, u64 parent
,
1155 u64 orig_root
, u64 ref_root
,
1156 u64 orig_generation
, u64 ref_generation
,
1160 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1161 struct btrfs_path
*path
;
1163 if (root
== root
->fs_info
->extent_root
) {
1164 struct pending_extent_op
*extent_op
;
1167 BUG_ON(owner_objectid
>= BTRFS_MAX_LEVEL
);
1168 num_bytes
= btrfs_level_size(root
, (int)owner_objectid
);
1169 mutex_lock(&root
->fs_info
->extent_ins_mutex
);
1170 if (test_range_bit(&root
->fs_info
->extent_ins
, bytenr
,
1171 bytenr
+ num_bytes
- 1, EXTENT_WRITEBACK
, 0)) {
1173 ret
= get_state_private(&root
->fs_info
->extent_ins
,
1176 extent_op
= (struct pending_extent_op
*)
1177 (unsigned long)priv
;
1178 BUG_ON(extent_op
->parent
!= orig_parent
);
1179 BUG_ON(extent_op
->generation
!= orig_generation
);
1181 extent_op
->parent
= parent
;
1182 extent_op
->generation
= ref_generation
;
1184 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
1187 extent_op
->type
= PENDING_BACKREF_UPDATE
;
1188 extent_op
->bytenr
= bytenr
;
1189 extent_op
->num_bytes
= num_bytes
;
1190 extent_op
->parent
= parent
;
1191 extent_op
->orig_parent
= orig_parent
;
1192 extent_op
->generation
= ref_generation
;
1193 extent_op
->orig_generation
= orig_generation
;
1194 extent_op
->level
= (int)owner_objectid
;
1195 INIT_LIST_HEAD(&extent_op
->list
);
1198 set_extent_bits(&root
->fs_info
->extent_ins
,
1199 bytenr
, bytenr
+ num_bytes
- 1,
1200 EXTENT_WRITEBACK
, GFP_NOFS
);
1201 set_state_private(&root
->fs_info
->extent_ins
,
1202 bytenr
, (unsigned long)extent_op
);
1204 mutex_unlock(&root
->fs_info
->extent_ins_mutex
);
1208 path
= btrfs_alloc_path();
1211 ret
= lookup_extent_backref(trans
, extent_root
, path
,
1212 bytenr
, orig_parent
, orig_root
,
1213 orig_generation
, owner_objectid
, 1);
1216 ret
= remove_extent_backref(trans
, extent_root
, path
);
1219 ret
= insert_extent_backref(trans
, extent_root
, path
, bytenr
,
1220 parent
, ref_root
, ref_generation
,
1223 finish_current_insert(trans
, extent_root
, 0);
1224 del_pending_extents(trans
, extent_root
, 0);
1226 btrfs_free_path(path
);
1230 int btrfs_update_extent_ref(struct btrfs_trans_handle
*trans
,
1231 struct btrfs_root
*root
, u64 bytenr
,
1232 u64 orig_parent
, u64 parent
,
1233 u64 ref_root
, u64 ref_generation
,
1237 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
1238 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
1240 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
, orig_parent
,
1241 parent
, ref_root
, ref_root
,
1242 ref_generation
, ref_generation
,
1247 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1248 struct btrfs_root
*root
, u64 bytenr
,
1249 u64 orig_parent
, u64 parent
,
1250 u64 orig_root
, u64 ref_root
,
1251 u64 orig_generation
, u64 ref_generation
,
1254 struct btrfs_path
*path
;
1256 struct btrfs_key key
;
1257 struct extent_buffer
*l
;
1258 struct btrfs_extent_item
*item
;
1261 path
= btrfs_alloc_path();
1266 key
.objectid
= bytenr
;
1267 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1268 key
.offset
= (u64
)-1;
1270 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
1274 BUG_ON(ret
== 0 || path
->slots
[0] == 0);
1279 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
1280 if (key
.objectid
!= bytenr
) {
1281 btrfs_print_leaf(root
->fs_info
->extent_root
, path
->nodes
[0]);
1282 printk(KERN_ERR
"btrfs wanted %llu found %llu\n",
1283 (unsigned long long)bytenr
,
1284 (unsigned long long)key
.objectid
);
1287 BUG_ON(key
.type
!= BTRFS_EXTENT_ITEM_KEY
);
1289 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
1290 refs
= btrfs_extent_refs(l
, item
);
1291 btrfs_set_extent_refs(l
, item
, refs
+ 1);
1292 btrfs_mark_buffer_dirty(path
->nodes
[0]);
1294 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1297 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1298 path
, bytenr
, parent
,
1299 ref_root
, ref_generation
,
1302 finish_current_insert(trans
, root
->fs_info
->extent_root
, 0);
1303 del_pending_extents(trans
, root
->fs_info
->extent_root
, 0);
1305 btrfs_free_path(path
);
1309 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1310 struct btrfs_root
*root
,
1311 u64 bytenr
, u64 num_bytes
, u64 parent
,
1312 u64 ref_root
, u64 ref_generation
,
1316 if (ref_root
== BTRFS_TREE_LOG_OBJECTID
&&
1317 owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
1319 ret
= __btrfs_inc_extent_ref(trans
, root
, bytenr
, 0, parent
,
1320 0, ref_root
, 0, ref_generation
,
1325 int btrfs_extent_post_op(struct btrfs_trans_handle
*trans
,
1326 struct btrfs_root
*root
)
1328 finish_current_insert(trans
, root
->fs_info
->extent_root
, 1);
1329 del_pending_extents(trans
, root
->fs_info
->extent_root
, 1);
1333 int btrfs_lookup_extent_ref(struct btrfs_trans_handle
*trans
,
1334 struct btrfs_root
*root
, u64 bytenr
,
1335 u64 num_bytes
, u32
*refs
)
1337 struct btrfs_path
*path
;
1339 struct btrfs_key key
;
1340 struct extent_buffer
*l
;
1341 struct btrfs_extent_item
*item
;
1343 WARN_ON(num_bytes
< root
->sectorsize
);
1344 path
= btrfs_alloc_path();
1346 key
.objectid
= bytenr
;
1347 key
.offset
= num_bytes
;
1348 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
1349 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
, path
,
1354 btrfs_print_leaf(root
, path
->nodes
[0]);
1355 printk(KERN_INFO
"btrfs failed to find block number %llu\n",
1356 (unsigned long long)bytenr
);
1360 item
= btrfs_item_ptr(l
, path
->slots
[0], struct btrfs_extent_item
);
1361 *refs
= btrfs_extent_refs(l
, item
);
1363 btrfs_free_path(path
);
1367 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
1368 struct btrfs_root
*root
, u64 objectid
, u64 bytenr
)
1370 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1371 struct btrfs_path
*path
;
1372 struct extent_buffer
*leaf
;
1373 struct btrfs_extent_ref
*ref_item
;
1374 struct btrfs_key key
;
1375 struct btrfs_key found_key
;
1381 key
.objectid
= bytenr
;
1382 key
.offset
= (u64
)-1;
1383 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1385 path
= btrfs_alloc_path();
1386 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
1392 if (path
->slots
[0] == 0)
1396 leaf
= path
->nodes
[0];
1397 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1399 if (found_key
.objectid
!= bytenr
||
1400 found_key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
1403 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
1405 leaf
= path
->nodes
[0];
1406 nritems
= btrfs_header_nritems(leaf
);
1407 if (path
->slots
[0] >= nritems
) {
1408 ret
= btrfs_next_leaf(extent_root
, path
);
1415 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1416 if (found_key
.objectid
!= bytenr
)
1419 if (found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
1424 ref_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1425 struct btrfs_extent_ref
);
1426 ref_root
= btrfs_ref_root(leaf
, ref_item
);
1427 if ((ref_root
!= root
->root_key
.objectid
&&
1428 ref_root
!= BTRFS_TREE_LOG_OBJECTID
) ||
1429 objectid
!= btrfs_ref_objectid(leaf
, ref_item
)) {
1433 if (btrfs_ref_generation(leaf
, ref_item
) <= last_snapshot
) {
1442 btrfs_free_path(path
);
1446 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1447 struct extent_buffer
*buf
, u32 nr_extents
)
1449 struct btrfs_key key
;
1450 struct btrfs_file_extent_item
*fi
;
1458 if (!root
->ref_cows
)
1461 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
1463 root_gen
= root
->root_key
.offset
;
1466 root_gen
= trans
->transid
- 1;
1469 level
= btrfs_header_level(buf
);
1470 nritems
= btrfs_header_nritems(buf
);
1473 struct btrfs_leaf_ref
*ref
;
1474 struct btrfs_extent_info
*info
;
1476 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
1482 ref
->root_gen
= root_gen
;
1483 ref
->bytenr
= buf
->start
;
1484 ref
->owner
= btrfs_header_owner(buf
);
1485 ref
->generation
= btrfs_header_generation(buf
);
1486 ref
->nritems
= nr_extents
;
1487 info
= ref
->extents
;
1489 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
1491 btrfs_item_key_to_cpu(buf
, &key
, i
);
1492 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1494 fi
= btrfs_item_ptr(buf
, i
,
1495 struct btrfs_file_extent_item
);
1496 if (btrfs_file_extent_type(buf
, fi
) ==
1497 BTRFS_FILE_EXTENT_INLINE
)
1499 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1500 if (disk_bytenr
== 0)
1503 info
->bytenr
= disk_bytenr
;
1505 btrfs_file_extent_disk_num_bytes(buf
, fi
);
1506 info
->objectid
= key
.objectid
;
1507 info
->offset
= key
.offset
;
1511 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
1512 if (ret
== -EEXIST
&& shared
) {
1513 struct btrfs_leaf_ref
*old
;
1514 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
1516 btrfs_remove_leaf_ref(root
, old
);
1517 btrfs_free_leaf_ref(root
, old
);
1518 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
1521 btrfs_free_leaf_ref(root
, ref
);
1527 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1528 struct extent_buffer
*orig_buf
, struct extent_buffer
*buf
,
1535 u64 orig_generation
;
1537 u32 nr_file_extents
= 0;
1538 struct btrfs_key key
;
1539 struct btrfs_file_extent_item
*fi
;
1544 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
1545 u64
, u64
, u64
, u64
, u64
, u64
, u64
, u64
);
1547 ref_root
= btrfs_header_owner(buf
);
1548 ref_generation
= btrfs_header_generation(buf
);
1549 orig_root
= btrfs_header_owner(orig_buf
);
1550 orig_generation
= btrfs_header_generation(orig_buf
);
1552 nritems
= btrfs_header_nritems(buf
);
1553 level
= btrfs_header_level(buf
);
1555 if (root
->ref_cows
) {
1556 process_func
= __btrfs_inc_extent_ref
;
1559 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1562 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1564 process_func
= __btrfs_update_extent_ref
;
1567 for (i
= 0; i
< nritems
; i
++) {
1570 btrfs_item_key_to_cpu(buf
, &key
, i
);
1571 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1573 fi
= btrfs_item_ptr(buf
, i
,
1574 struct btrfs_file_extent_item
);
1575 if (btrfs_file_extent_type(buf
, fi
) ==
1576 BTRFS_FILE_EXTENT_INLINE
)
1578 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1584 ret
= process_func(trans
, root
, bytenr
,
1585 orig_buf
->start
, buf
->start
,
1586 orig_root
, ref_root
,
1587 orig_generation
, ref_generation
,
1596 bytenr
= btrfs_node_blockptr(buf
, i
);
1597 ret
= process_func(trans
, root
, bytenr
,
1598 orig_buf
->start
, buf
->start
,
1599 orig_root
, ref_root
,
1600 orig_generation
, ref_generation
,
1612 *nr_extents
= nr_file_extents
;
1614 *nr_extents
= nritems
;
1622 int btrfs_update_ref(struct btrfs_trans_handle
*trans
,
1623 struct btrfs_root
*root
, struct extent_buffer
*orig_buf
,
1624 struct extent_buffer
*buf
, int start_slot
, int nr
)
1631 u64 orig_generation
;
1632 struct btrfs_key key
;
1633 struct btrfs_file_extent_item
*fi
;
1639 BUG_ON(start_slot
< 0);
1640 BUG_ON(start_slot
+ nr
> btrfs_header_nritems(buf
));
1642 ref_root
= btrfs_header_owner(buf
);
1643 ref_generation
= btrfs_header_generation(buf
);
1644 orig_root
= btrfs_header_owner(orig_buf
);
1645 orig_generation
= btrfs_header_generation(orig_buf
);
1646 level
= btrfs_header_level(buf
);
1648 if (!root
->ref_cows
) {
1650 root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
1653 root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
)
1657 for (i
= 0, slot
= start_slot
; i
< nr
; i
++, slot
++) {
1660 btrfs_item_key_to_cpu(buf
, &key
, slot
);
1661 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
1663 fi
= btrfs_item_ptr(buf
, slot
,
1664 struct btrfs_file_extent_item
);
1665 if (btrfs_file_extent_type(buf
, fi
) ==
1666 BTRFS_FILE_EXTENT_INLINE
)
1668 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
1671 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1672 orig_buf
->start
, buf
->start
,
1673 orig_root
, ref_root
,
1674 orig_generation
, ref_generation
,
1679 bytenr
= btrfs_node_blockptr(buf
, slot
);
1680 ret
= __btrfs_update_extent_ref(trans
, root
, bytenr
,
1681 orig_buf
->start
, buf
->start
,
1682 orig_root
, ref_root
,
1683 orig_generation
, ref_generation
,
1695 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
1696 struct btrfs_root
*root
,
1697 struct btrfs_path
*path
,
1698 struct btrfs_block_group_cache
*cache
)
1702 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
1704 struct extent_buffer
*leaf
;
1706 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
1711 leaf
= path
->nodes
[0];
1712 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
1713 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
1714 btrfs_mark_buffer_dirty(leaf
);
1715 btrfs_release_path(extent_root
, path
);
1717 finish_current_insert(trans
, extent_root
, 0);
1718 pending_ret
= del_pending_extents(trans
, extent_root
, 0);
1727 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
1728 struct btrfs_root
*root
)
1730 struct btrfs_block_group_cache
*cache
, *entry
;
1734 struct btrfs_path
*path
;
1737 path
= btrfs_alloc_path();
1743 spin_lock(&root
->fs_info
->block_group_cache_lock
);
1744 for (n
= rb_first(&root
->fs_info
->block_group_cache_tree
);
1745 n
; n
= rb_next(n
)) {
1746 entry
= rb_entry(n
, struct btrfs_block_group_cache
,
1753 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
1759 last
+= cache
->key
.offset
;
1761 err
= write_one_cache_group(trans
, root
,
1764 * if we fail to write the cache group, we want
1765 * to keep it marked dirty in hopes that a later
1773 btrfs_free_path(path
);
1777 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
1779 struct btrfs_block_group_cache
*block_group
;
1782 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
1783 if (!block_group
|| block_group
->ro
)
1786 put_block_group(block_group
);
1790 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
1791 u64 total_bytes
, u64 bytes_used
,
1792 struct btrfs_space_info
**space_info
)
1794 struct btrfs_space_info
*found
;
1796 found
= __find_space_info(info
, flags
);
1798 spin_lock(&found
->lock
);
1799 found
->total_bytes
+= total_bytes
;
1800 found
->bytes_used
+= bytes_used
;
1802 spin_unlock(&found
->lock
);
1803 *space_info
= found
;
1806 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
1810 list_add(&found
->list
, &info
->space_info
);
1811 INIT_LIST_HEAD(&found
->block_groups
);
1812 init_rwsem(&found
->groups_sem
);
1813 spin_lock_init(&found
->lock
);
1814 found
->flags
= flags
;
1815 found
->total_bytes
= total_bytes
;
1816 found
->bytes_used
= bytes_used
;
1817 found
->bytes_pinned
= 0;
1818 found
->bytes_reserved
= 0;
1819 found
->bytes_readonly
= 0;
1821 found
->force_alloc
= 0;
1822 *space_info
= found
;
1826 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
1828 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
1829 BTRFS_BLOCK_GROUP_RAID1
|
1830 BTRFS_BLOCK_GROUP_RAID10
|
1831 BTRFS_BLOCK_GROUP_DUP
);
1833 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
1834 fs_info
->avail_data_alloc_bits
|= extra_flags
;
1835 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
1836 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
1837 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
1838 fs_info
->avail_system_alloc_bits
|= extra_flags
;
1842 static void set_block_group_readonly(struct btrfs_block_group_cache
*cache
)
1844 spin_lock(&cache
->space_info
->lock
);
1845 spin_lock(&cache
->lock
);
1847 cache
->space_info
->bytes_readonly
+= cache
->key
.offset
-
1848 btrfs_block_group_used(&cache
->item
);
1851 spin_unlock(&cache
->lock
);
1852 spin_unlock(&cache
->space_info
->lock
);
1855 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
1857 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
1859 if (num_devices
== 1)
1860 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
1861 if (num_devices
< 4)
1862 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
1864 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
1865 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
1866 BTRFS_BLOCK_GROUP_RAID10
))) {
1867 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
1870 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
1871 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
1872 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
1875 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
1876 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
1877 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
1878 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
1879 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
1883 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
1884 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
1885 u64 flags
, int force
)
1887 struct btrfs_space_info
*space_info
;
1891 mutex_lock(&extent_root
->fs_info
->chunk_mutex
);
1893 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
1895 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
1897 ret
= update_space_info(extent_root
->fs_info
, flags
,
1901 BUG_ON(!space_info
);
1903 spin_lock(&space_info
->lock
);
1904 if (space_info
->force_alloc
) {
1906 space_info
->force_alloc
= 0;
1908 if (space_info
->full
) {
1909 spin_unlock(&space_info
->lock
);
1913 thresh
= space_info
->total_bytes
- space_info
->bytes_readonly
;
1914 thresh
= div_factor(thresh
, 6);
1916 (space_info
->bytes_used
+ space_info
->bytes_pinned
+
1917 space_info
->bytes_reserved
+ alloc_bytes
) < thresh
) {
1918 spin_unlock(&space_info
->lock
);
1921 spin_unlock(&space_info
->lock
);
1923 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
1925 space_info
->full
= 1;
1927 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
1931 static int update_block_group(struct btrfs_trans_handle
*trans
,
1932 struct btrfs_root
*root
,
1933 u64 bytenr
, u64 num_bytes
, int alloc
,
1936 struct btrfs_block_group_cache
*cache
;
1937 struct btrfs_fs_info
*info
= root
->fs_info
;
1938 u64 total
= num_bytes
;
1943 cache
= btrfs_lookup_block_group(info
, bytenr
);
1946 byte_in_group
= bytenr
- cache
->key
.objectid
;
1947 WARN_ON(byte_in_group
> cache
->key
.offset
);
1949 spin_lock(&cache
->space_info
->lock
);
1950 spin_lock(&cache
->lock
);
1952 old_val
= btrfs_block_group_used(&cache
->item
);
1953 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
1955 old_val
+= num_bytes
;
1956 cache
->space_info
->bytes_used
+= num_bytes
;
1958 cache
->space_info
->bytes_readonly
-= num_bytes
;
1959 btrfs_set_block_group_used(&cache
->item
, old_val
);
1960 spin_unlock(&cache
->lock
);
1961 spin_unlock(&cache
->space_info
->lock
);
1963 old_val
-= num_bytes
;
1964 cache
->space_info
->bytes_used
-= num_bytes
;
1966 cache
->space_info
->bytes_readonly
+= num_bytes
;
1967 btrfs_set_block_group_used(&cache
->item
, old_val
);
1968 spin_unlock(&cache
->lock
);
1969 spin_unlock(&cache
->space_info
->lock
);
1973 ret
= btrfs_discard_extent(root
, bytenr
,
1977 ret
= btrfs_add_free_space(cache
, bytenr
,
1982 put_block_group(cache
);
1984 bytenr
+= num_bytes
;
1989 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
1991 struct btrfs_block_group_cache
*cache
;
1994 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
1998 bytenr
= cache
->key
.objectid
;
1999 put_block_group(cache
);
2004 int btrfs_update_pinned_extents(struct btrfs_root
*root
,
2005 u64 bytenr
, u64 num
, int pin
)
2008 struct btrfs_block_group_cache
*cache
;
2009 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2011 WARN_ON(!mutex_is_locked(&root
->fs_info
->pinned_mutex
));
2013 set_extent_dirty(&fs_info
->pinned_extents
,
2014 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
2016 clear_extent_dirty(&fs_info
->pinned_extents
,
2017 bytenr
, bytenr
+ num
- 1, GFP_NOFS
);
2020 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
2022 len
= min(num
, cache
->key
.offset
-
2023 (bytenr
- cache
->key
.objectid
));
2025 spin_lock(&cache
->space_info
->lock
);
2026 spin_lock(&cache
->lock
);
2027 cache
->pinned
+= len
;
2028 cache
->space_info
->bytes_pinned
+= len
;
2029 spin_unlock(&cache
->lock
);
2030 spin_unlock(&cache
->space_info
->lock
);
2031 fs_info
->total_pinned
+= len
;
2033 spin_lock(&cache
->space_info
->lock
);
2034 spin_lock(&cache
->lock
);
2035 cache
->pinned
-= len
;
2036 cache
->space_info
->bytes_pinned
-= len
;
2037 spin_unlock(&cache
->lock
);
2038 spin_unlock(&cache
->space_info
->lock
);
2039 fs_info
->total_pinned
-= len
;
2041 btrfs_add_free_space(cache
, bytenr
, len
);
2043 put_block_group(cache
);
2050 static int update_reserved_extents(struct btrfs_root
*root
,
2051 u64 bytenr
, u64 num
, int reserve
)
2054 struct btrfs_block_group_cache
*cache
;
2055 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2058 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
2060 len
= min(num
, cache
->key
.offset
-
2061 (bytenr
- cache
->key
.objectid
));
2063 spin_lock(&cache
->space_info
->lock
);
2064 spin_lock(&cache
->lock
);
2066 cache
->reserved
+= len
;
2067 cache
->space_info
->bytes_reserved
+= len
;
2069 cache
->reserved
-= len
;
2070 cache
->space_info
->bytes_reserved
-= len
;
2072 spin_unlock(&cache
->lock
);
2073 spin_unlock(&cache
->space_info
->lock
);
2074 put_block_group(cache
);
2081 int btrfs_copy_pinned(struct btrfs_root
*root
, struct extent_io_tree
*copy
)
2086 struct extent_io_tree
*pinned_extents
= &root
->fs_info
->pinned_extents
;
2089 mutex_lock(&root
->fs_info
->pinned_mutex
);
2091 ret
= find_first_extent_bit(pinned_extents
, last
,
2092 &start
, &end
, EXTENT_DIRTY
);
2095 set_extent_dirty(copy
, start
, end
, GFP_NOFS
);
2098 mutex_unlock(&root
->fs_info
->pinned_mutex
);
2102 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
2103 struct btrfs_root
*root
,
2104 struct extent_io_tree
*unpin
)
2110 mutex_lock(&root
->fs_info
->pinned_mutex
);
2112 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
2117 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
2119 btrfs_update_pinned_extents(root
, start
, end
+ 1 - start
, 0);
2120 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
2122 if (need_resched()) {
2123 mutex_unlock(&root
->fs_info
->pinned_mutex
);
2125 mutex_lock(&root
->fs_info
->pinned_mutex
);
2128 mutex_unlock(&root
->fs_info
->pinned_mutex
);
2132 static int finish_current_insert(struct btrfs_trans_handle
*trans
,
2133 struct btrfs_root
*extent_root
, int all
)
2140 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
2141 struct btrfs_path
*path
;
2142 struct pending_extent_op
*extent_op
, *tmp
;
2143 struct list_head insert_list
, update_list
;
2145 int num_inserts
= 0, max_inserts
;
2147 path
= btrfs_alloc_path();
2148 INIT_LIST_HEAD(&insert_list
);
2149 INIT_LIST_HEAD(&update_list
);
2151 max_inserts
= extent_root
->leafsize
/
2152 (2 * sizeof(struct btrfs_key
) + 2 * sizeof(struct btrfs_item
) +
2153 sizeof(struct btrfs_extent_ref
) +
2154 sizeof(struct btrfs_extent_item
));
2156 mutex_lock(&info
->extent_ins_mutex
);
2158 ret
= find_first_extent_bit(&info
->extent_ins
, search
, &start
,
2159 &end
, EXTENT_WRITEBACK
);
2161 if (skipped
&& all
&& !num_inserts
) {
2166 mutex_unlock(&info
->extent_ins_mutex
);
2170 ret
= try_lock_extent(&info
->extent_ins
, start
, end
, GFP_NOFS
);
2174 if (need_resched()) {
2175 mutex_unlock(&info
->extent_ins_mutex
);
2177 mutex_lock(&info
->extent_ins_mutex
);
2182 ret
= get_state_private(&info
->extent_ins
, start
, &priv
);
2184 extent_op
= (struct pending_extent_op
*)(unsigned long) priv
;
2186 if (extent_op
->type
== PENDING_EXTENT_INSERT
) {
2188 list_add_tail(&extent_op
->list
, &insert_list
);
2190 if (num_inserts
== max_inserts
) {
2191 mutex_unlock(&info
->extent_ins_mutex
);
2194 } else if (extent_op
->type
== PENDING_BACKREF_UPDATE
) {
2195 list_add_tail(&extent_op
->list
, &update_list
);
2203 * process the update list, clear the writeback bit for it, and if
2204 * somebody marked this thing for deletion then just unlock it and be
2205 * done, the free_extents will handle it
2207 mutex_lock(&info
->extent_ins_mutex
);
2208 list_for_each_entry_safe(extent_op
, tmp
, &update_list
, list
) {
2209 clear_extent_bits(&info
->extent_ins
, extent_op
->bytenr
,
2210 extent_op
->bytenr
+ extent_op
->num_bytes
- 1,
2211 EXTENT_WRITEBACK
, GFP_NOFS
);
2212 if (extent_op
->del
) {
2213 list_del_init(&extent_op
->list
);
2214 unlock_extent(&info
->extent_ins
, extent_op
->bytenr
,
2215 extent_op
->bytenr
+ extent_op
->num_bytes
2220 mutex_unlock(&info
->extent_ins_mutex
);
2223 * still have things left on the update list, go ahead an update
2226 if (!list_empty(&update_list
)) {
2227 ret
= update_backrefs(trans
, extent_root
, path
, &update_list
);
2232 * if no inserts need to be done, but we skipped some extents and we
2233 * need to make sure everything is cleaned then reset everything and
2234 * go back to the beginning
2236 if (!num_inserts
&& all
&& skipped
) {
2239 INIT_LIST_HEAD(&update_list
);
2240 INIT_LIST_HEAD(&insert_list
);
2242 } else if (!num_inserts
) {
2247 * process the insert extents list. Again if we are deleting this
2248 * extent, then just unlock it, pin down the bytes if need be, and be
2249 * done with it. Saves us from having to actually insert the extent
2250 * into the tree and then subsequently come along and delete it
2252 mutex_lock(&info
->extent_ins_mutex
);
2253 list_for_each_entry_safe(extent_op
, tmp
, &insert_list
, list
) {
2254 clear_extent_bits(&info
->extent_ins
, extent_op
->bytenr
,
2255 extent_op
->bytenr
+ extent_op
->num_bytes
- 1,
2256 EXTENT_WRITEBACK
, GFP_NOFS
);
2257 if (extent_op
->del
) {
2259 list_del_init(&extent_op
->list
);
2260 unlock_extent(&info
->extent_ins
, extent_op
->bytenr
,
2261 extent_op
->bytenr
+ extent_op
->num_bytes
2264 mutex_lock(&extent_root
->fs_info
->pinned_mutex
);
2265 ret
= pin_down_bytes(trans
, extent_root
,
2267 extent_op
->num_bytes
, 0);
2268 mutex_unlock(&extent_root
->fs_info
->pinned_mutex
);
2270 spin_lock(&info
->delalloc_lock
);
2271 used
= btrfs_super_bytes_used(&info
->super_copy
);
2272 btrfs_set_super_bytes_used(&info
->super_copy
,
2273 used
- extent_op
->num_bytes
);
2274 used
= btrfs_root_used(&extent_root
->root_item
);
2275 btrfs_set_root_used(&extent_root
->root_item
,
2276 used
- extent_op
->num_bytes
);
2277 spin_unlock(&info
->delalloc_lock
);
2279 ret
= update_block_group(trans
, extent_root
,
2281 extent_op
->num_bytes
,
2288 mutex_unlock(&info
->extent_ins_mutex
);
2290 ret
= insert_extents(trans
, extent_root
, path
, &insert_list
,
2295 * if we broke out of the loop in order to insert stuff because we hit
2296 * the maximum number of inserts at a time we can handle, then loop
2297 * back and pick up where we left off
2299 if (num_inserts
== max_inserts
) {
2300 INIT_LIST_HEAD(&insert_list
);
2301 INIT_LIST_HEAD(&update_list
);
2307 * again, if we need to make absolutely sure there are no more pending
2308 * extent operations left and we know that we skipped some, go back to
2309 * the beginning and do it all again
2311 if (all
&& skipped
) {
2312 INIT_LIST_HEAD(&insert_list
);
2313 INIT_LIST_HEAD(&update_list
);
2320 btrfs_free_path(path
);
2324 static int pin_down_bytes(struct btrfs_trans_handle
*trans
,
2325 struct btrfs_root
*root
,
2326 u64 bytenr
, u64 num_bytes
, int is_data
)
2329 struct extent_buffer
*buf
;
2334 buf
= btrfs_find_tree_block(root
, bytenr
, num_bytes
);
2338 /* we can reuse a block if it hasn't been written
2339 * and it is from this transaction. We can't
2340 * reuse anything from the tree log root because
2341 * it has tiny sub-transactions.
2343 if (btrfs_buffer_uptodate(buf
, 0) &&
2344 btrfs_try_tree_lock(buf
)) {
2345 u64 header_owner
= btrfs_header_owner(buf
);
2346 u64 header_transid
= btrfs_header_generation(buf
);
2347 if (header_owner
!= BTRFS_TREE_LOG_OBJECTID
&&
2348 header_owner
!= BTRFS_TREE_RELOC_OBJECTID
&&
2349 header_transid
== trans
->transid
&&
2350 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
2351 clean_tree_block(NULL
, root
, buf
);
2352 btrfs_tree_unlock(buf
);
2353 free_extent_buffer(buf
);
2356 btrfs_tree_unlock(buf
);
2358 free_extent_buffer(buf
);
2360 btrfs_update_pinned_extents(root
, bytenr
, num_bytes
, 1);
2367 * remove an extent from the root, returns 0 on success
2369 static int __free_extent(struct btrfs_trans_handle
*trans
,
2370 struct btrfs_root
*root
,
2371 u64 bytenr
, u64 num_bytes
, u64 parent
,
2372 u64 root_objectid
, u64 ref_generation
,
2373 u64 owner_objectid
, int pin
, int mark_free
)
2375 struct btrfs_path
*path
;
2376 struct btrfs_key key
;
2377 struct btrfs_fs_info
*info
= root
->fs_info
;
2378 struct btrfs_root
*extent_root
= info
->extent_root
;
2379 struct extent_buffer
*leaf
;
2381 int extent_slot
= 0;
2382 int found_extent
= 0;
2384 struct btrfs_extent_item
*ei
;
2387 key
.objectid
= bytenr
;
2388 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
2389 key
.offset
= num_bytes
;
2390 path
= btrfs_alloc_path();
2395 ret
= lookup_extent_backref(trans
, extent_root
, path
,
2396 bytenr
, parent
, root_objectid
,
2397 ref_generation
, owner_objectid
, 1);
2399 struct btrfs_key found_key
;
2400 extent_slot
= path
->slots
[0];
2401 while (extent_slot
> 0) {
2403 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2405 if (found_key
.objectid
!= bytenr
)
2407 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2408 found_key
.offset
== num_bytes
) {
2412 if (path
->slots
[0] - extent_slot
> 5)
2415 if (!found_extent
) {
2416 ret
= remove_extent_backref(trans
, extent_root
, path
);
2418 btrfs_release_path(extent_root
, path
);
2419 ret
= btrfs_search_slot(trans
, extent_root
,
2422 printk(KERN_ERR
"umm, got %d back from search"
2423 ", was looking for %llu\n", ret
,
2424 (unsigned long long)bytenr
);
2425 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
2428 extent_slot
= path
->slots
[0];
2431 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
2433 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
2434 "root %llu gen %llu owner %llu\n",
2435 (unsigned long long)bytenr
,
2436 (unsigned long long)root_objectid
,
2437 (unsigned long long)ref_generation
,
2438 (unsigned long long)owner_objectid
);
2441 leaf
= path
->nodes
[0];
2442 ei
= btrfs_item_ptr(leaf
, extent_slot
,
2443 struct btrfs_extent_item
);
2444 refs
= btrfs_extent_refs(leaf
, ei
);
2447 btrfs_set_extent_refs(leaf
, ei
, refs
);
2449 btrfs_mark_buffer_dirty(leaf
);
2451 if (refs
== 0 && found_extent
&& path
->slots
[0] == extent_slot
+ 1) {
2452 struct btrfs_extent_ref
*ref
;
2453 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
2454 struct btrfs_extent_ref
);
2455 BUG_ON(btrfs_ref_num_refs(leaf
, ref
) != 1);
2456 /* if the back ref and the extent are next to each other
2457 * they get deleted below in one shot
2459 path
->slots
[0] = extent_slot
;
2461 } else if (found_extent
) {
2462 /* otherwise delete the extent back ref */
2463 ret
= remove_extent_backref(trans
, extent_root
, path
);
2465 /* if refs are 0, we need to setup the path for deletion */
2467 btrfs_release_path(extent_root
, path
);
2468 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
2479 mutex_lock(&root
->fs_info
->pinned_mutex
);
2480 ret
= pin_down_bytes(trans
, root
, bytenr
, num_bytes
,
2481 owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
);
2482 mutex_unlock(&root
->fs_info
->pinned_mutex
);
2487 /* block accounting for super block */
2488 spin_lock(&info
->delalloc_lock
);
2489 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
2490 btrfs_set_super_bytes_used(&info
->super_copy
,
2491 super_used
- num_bytes
);
2493 /* block accounting for root item */
2494 root_used
= btrfs_root_used(&root
->root_item
);
2495 btrfs_set_root_used(&root
->root_item
,
2496 root_used
- num_bytes
);
2497 spin_unlock(&info
->delalloc_lock
);
2498 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
2501 btrfs_release_path(extent_root
, path
);
2503 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
2504 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
2508 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0,
2512 btrfs_free_path(path
);
2513 finish_current_insert(trans
, extent_root
, 0);
2518 * find all the blocks marked as pending in the radix tree and remove
2519 * them from the extent map
2521 static int del_pending_extents(struct btrfs_trans_handle
*trans
,
2522 struct btrfs_root
*extent_root
, int all
)
2530 int nr
= 0, skipped
= 0;
2531 struct extent_io_tree
*pending_del
;
2532 struct extent_io_tree
*extent_ins
;
2533 struct pending_extent_op
*extent_op
;
2534 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
2535 struct list_head delete_list
;
2537 INIT_LIST_HEAD(&delete_list
);
2538 extent_ins
= &extent_root
->fs_info
->extent_ins
;
2539 pending_del
= &extent_root
->fs_info
->pending_del
;
2542 mutex_lock(&info
->extent_ins_mutex
);
2544 ret
= find_first_extent_bit(pending_del
, search
, &start
, &end
,
2547 if (all
&& skipped
&& !nr
) {
2551 mutex_unlock(&info
->extent_ins_mutex
);
2555 ret
= try_lock_extent(extent_ins
, start
, end
, GFP_NOFS
);
2560 if (need_resched()) {
2561 mutex_unlock(&info
->extent_ins_mutex
);
2563 mutex_lock(&info
->extent_ins_mutex
);
2570 ret
= get_state_private(pending_del
, start
, &priv
);
2572 extent_op
= (struct pending_extent_op
*)(unsigned long)priv
;
2574 clear_extent_bits(pending_del
, start
, end
, EXTENT_WRITEBACK
,
2576 if (!test_range_bit(extent_ins
, start
, end
,
2577 EXTENT_WRITEBACK
, 0)) {
2578 list_add_tail(&extent_op
->list
, &delete_list
);
2583 ret
= get_state_private(&info
->extent_ins
, start
,
2586 extent_op
= (struct pending_extent_op
*)
2587 (unsigned long)priv
;
2589 clear_extent_bits(&info
->extent_ins
, start
, end
,
2590 EXTENT_WRITEBACK
, GFP_NOFS
);
2592 if (extent_op
->type
== PENDING_BACKREF_UPDATE
) {
2593 list_add_tail(&extent_op
->list
, &delete_list
);
2599 mutex_lock(&extent_root
->fs_info
->pinned_mutex
);
2600 ret
= pin_down_bytes(trans
, extent_root
, start
,
2601 end
+ 1 - start
, 0);
2602 mutex_unlock(&extent_root
->fs_info
->pinned_mutex
);
2604 ret
= update_block_group(trans
, extent_root
, start
,
2605 end
+ 1 - start
, 0, ret
> 0);
2607 unlock_extent(extent_ins
, start
, end
, GFP_NOFS
);
2616 if (need_resched()) {
2617 mutex_unlock(&info
->extent_ins_mutex
);
2619 mutex_lock(&info
->extent_ins_mutex
);
2624 ret
= free_extents(trans
, extent_root
, &delete_list
);
2628 if (all
&& skipped
) {
2629 INIT_LIST_HEAD(&delete_list
);
2639 * remove an extent from the root, returns 0 on success
2641 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2642 struct btrfs_root
*root
,
2643 u64 bytenr
, u64 num_bytes
, u64 parent
,
2644 u64 root_objectid
, u64 ref_generation
,
2645 u64 owner_objectid
, int pin
)
2647 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2651 WARN_ON(num_bytes
< root
->sectorsize
);
2652 if (root
== extent_root
) {
2653 struct pending_extent_op
*extent_op
= NULL
;
2655 mutex_lock(&root
->fs_info
->extent_ins_mutex
);
2656 if (test_range_bit(&root
->fs_info
->extent_ins
, bytenr
,
2657 bytenr
+ num_bytes
- 1, EXTENT_WRITEBACK
, 0)) {
2659 ret
= get_state_private(&root
->fs_info
->extent_ins
,
2662 extent_op
= (struct pending_extent_op
*)
2663 (unsigned long)priv
;
2666 if (extent_op
->type
== PENDING_EXTENT_INSERT
) {
2667 mutex_unlock(&root
->fs_info
->extent_ins_mutex
);
2673 ref_generation
= extent_op
->orig_generation
;
2674 parent
= extent_op
->orig_parent
;
2677 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2680 extent_op
->type
= PENDING_EXTENT_DELETE
;
2681 extent_op
->bytenr
= bytenr
;
2682 extent_op
->num_bytes
= num_bytes
;
2683 extent_op
->parent
= parent
;
2684 extent_op
->orig_parent
= parent
;
2685 extent_op
->generation
= ref_generation
;
2686 extent_op
->orig_generation
= ref_generation
;
2687 extent_op
->level
= (int)owner_objectid
;
2688 INIT_LIST_HEAD(&extent_op
->list
);
2691 set_extent_bits(&root
->fs_info
->pending_del
,
2692 bytenr
, bytenr
+ num_bytes
- 1,
2693 EXTENT_WRITEBACK
, GFP_NOFS
);
2694 set_state_private(&root
->fs_info
->pending_del
,
2695 bytenr
, (unsigned long)extent_op
);
2696 mutex_unlock(&root
->fs_info
->extent_ins_mutex
);
2699 /* if metadata always pin */
2700 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
2701 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
2702 struct btrfs_block_group_cache
*cache
;
2704 /* btrfs_free_reserved_extent */
2705 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2707 btrfs_add_free_space(cache
, bytenr
, num_bytes
);
2708 put_block_group(cache
);
2709 update_reserved_extents(root
, bytenr
, num_bytes
, 0);
2715 /* if data pin when any transaction has committed this */
2716 if (ref_generation
!= trans
->transid
)
2719 ret
= __free_extent(trans
, root
, bytenr
, num_bytes
, parent
,
2720 root_objectid
, ref_generation
,
2721 owner_objectid
, pin
, pin
== 0);
2723 finish_current_insert(trans
, root
->fs_info
->extent_root
, 0);
2724 pending_ret
= del_pending_extents(trans
, root
->fs_info
->extent_root
, 0);
2725 return ret
? ret
: pending_ret
;
2728 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
2729 struct btrfs_root
*root
,
2730 u64 bytenr
, u64 num_bytes
, u64 parent
,
2731 u64 root_objectid
, u64 ref_generation
,
2732 u64 owner_objectid
, int pin
)
2736 ret
= __btrfs_free_extent(trans
, root
, bytenr
, num_bytes
, parent
,
2737 root_objectid
, ref_generation
,
2738 owner_objectid
, pin
);
2742 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
2744 u64 mask
= ((u64
)root
->stripesize
- 1);
2745 u64 ret
= (val
+ mask
) & ~mask
;
2750 * walks the btree of allocated extents and find a hole of a given size.
2751 * The key ins is changed to record the hole:
2752 * ins->objectid == block start
2753 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2754 * ins->offset == number of blocks
2755 * Any available blocks before search_start are skipped.
2757 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
2758 struct btrfs_root
*orig_root
,
2759 u64 num_bytes
, u64 empty_size
,
2760 u64 search_start
, u64 search_end
,
2761 u64 hint_byte
, struct btrfs_key
*ins
,
2762 u64 exclude_start
, u64 exclude_nr
,
2766 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
2767 u64 total_needed
= num_bytes
;
2768 u64
*last_ptr
= NULL
;
2769 u64 last_wanted
= 0;
2770 struct btrfs_block_group_cache
*block_group
= NULL
;
2771 int chunk_alloc_done
= 0;
2772 int empty_cluster
= 2 * 1024 * 1024;
2773 int allowed_chunk_alloc
= 0;
2774 struct list_head
*head
= NULL
, *cur
= NULL
;
2777 struct btrfs_space_info
*space_info
;
2779 WARN_ON(num_bytes
< root
->sectorsize
);
2780 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
2784 if (orig_root
->ref_cows
|| empty_size
)
2785 allowed_chunk_alloc
= 1;
2787 if (data
& BTRFS_BLOCK_GROUP_METADATA
) {
2788 last_ptr
= &root
->fs_info
->last_alloc
;
2789 empty_cluster
= 64 * 1024;
2792 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && btrfs_test_opt(root
, SSD
))
2793 last_ptr
= &root
->fs_info
->last_data_alloc
;
2797 hint_byte
= *last_ptr
;
2798 last_wanted
= *last_ptr
;
2800 empty_size
+= empty_cluster
;
2804 search_start
= max(search_start
, first_logical_byte(root
, 0));
2805 search_start
= max(search_start
, hint_byte
);
2807 if (last_wanted
&& search_start
!= last_wanted
) {
2809 empty_size
+= empty_cluster
;
2812 total_needed
+= empty_size
;
2813 block_group
= btrfs_lookup_block_group(root
->fs_info
, search_start
);
2815 block_group
= btrfs_lookup_first_block_group(root
->fs_info
,
2817 space_info
= __find_space_info(root
->fs_info
, data
);
2819 down_read(&space_info
->groups_sem
);
2821 struct btrfs_free_space
*free_space
;
2823 * the only way this happens if our hint points to a block
2824 * group thats not of the proper type, while looping this
2825 * should never happen
2831 goto new_group_no_lock
;
2833 if (unlikely(!block_group
->cached
)) {
2834 mutex_lock(&block_group
->cache_mutex
);
2835 ret
= cache_block_group(root
, block_group
);
2836 mutex_unlock(&block_group
->cache_mutex
);
2841 mutex_lock(&block_group
->alloc_mutex
);
2842 if (unlikely(!block_group_bits(block_group
, data
)))
2845 if (unlikely(block_group
->ro
))
2848 free_space
= btrfs_find_free_space(block_group
, search_start
,
2851 u64 start
= block_group
->key
.objectid
;
2852 u64 end
= block_group
->key
.objectid
+
2853 block_group
->key
.offset
;
2855 search_start
= stripe_align(root
, free_space
->offset
);
2857 /* move on to the next group */
2858 if (search_start
+ num_bytes
>= search_end
)
2861 /* move on to the next group */
2862 if (search_start
+ num_bytes
> end
)
2865 if (last_wanted
&& search_start
!= last_wanted
) {
2866 total_needed
+= empty_cluster
;
2867 empty_size
+= empty_cluster
;
2870 * if search_start is still in this block group
2871 * then we just re-search this block group
2873 if (search_start
>= start
&&
2874 search_start
< end
) {
2875 mutex_unlock(&block_group
->alloc_mutex
);
2879 /* else we go to the next block group */
2883 if (exclude_nr
> 0 &&
2884 (search_start
+ num_bytes
> exclude_start
&&
2885 search_start
< exclude_start
+ exclude_nr
)) {
2886 search_start
= exclude_start
+ exclude_nr
;
2888 * if search_start is still in this block group
2889 * then we just re-search this block group
2891 if (search_start
>= start
&&
2892 search_start
< end
) {
2893 mutex_unlock(&block_group
->alloc_mutex
);
2898 /* else we go to the next block group */
2902 ins
->objectid
= search_start
;
2903 ins
->offset
= num_bytes
;
2905 btrfs_remove_free_space_lock(block_group
, search_start
,
2907 /* we are all good, lets return */
2908 mutex_unlock(&block_group
->alloc_mutex
);
2912 mutex_unlock(&block_group
->alloc_mutex
);
2913 put_block_group(block_group
);
2916 /* don't try to compare new allocations against the
2917 * last allocation any more
2922 * Here's how this works.
2923 * loop == 0: we were searching a block group via a hint
2924 * and didn't find anything, so we start at
2925 * the head of the block groups and keep searching
2926 * loop == 1: we're searching through all of the block groups
2927 * if we hit the head again we have searched
2928 * all of the block groups for this space and we
2929 * need to try and allocate, if we cant error out.
2930 * loop == 2: we allocated more space and are looping through
2931 * all of the block groups again.
2934 head
= &space_info
->block_groups
;
2937 } else if (loop
== 1 && cur
== head
) {
2940 /* at this point we give up on the empty_size
2941 * allocations and just try to allocate the min
2944 * The extra_loop field was set if an empty_size
2945 * allocation was attempted above, and if this
2946 * is try we need to try the loop again without
2947 * the additional empty_size.
2949 total_needed
-= empty_size
;
2951 keep_going
= extra_loop
;
2954 if (allowed_chunk_alloc
&& !chunk_alloc_done
) {
2955 up_read(&space_info
->groups_sem
);
2956 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
2957 2 * 1024 * 1024, data
, 1);
2958 down_read(&space_info
->groups_sem
);
2961 head
= &space_info
->block_groups
;
2963 * we've allocated a new chunk, keep
2967 chunk_alloc_done
= 1;
2968 } else if (!allowed_chunk_alloc
) {
2969 space_info
->force_alloc
= 1;
2978 } else if (cur
== head
) {
2982 block_group
= list_entry(cur
, struct btrfs_block_group_cache
,
2984 atomic_inc(&block_group
->count
);
2986 search_start
= block_group
->key
.objectid
;
2990 /* we found what we needed */
2991 if (ins
->objectid
) {
2992 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
2993 trans
->block_group
= block_group
->key
.objectid
;
2996 *last_ptr
= ins
->objectid
+ ins
->offset
;
2999 printk(KERN_ERR
"btrfs searching for %llu bytes, "
3000 "num_bytes %llu, loop %d, allowed_alloc %d\n",
3001 (unsigned long long)total_needed
,
3002 (unsigned long long)num_bytes
,
3003 loop
, allowed_chunk_alloc
);
3007 put_block_group(block_group
);
3009 up_read(&space_info
->groups_sem
);
3013 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
)
3015 struct btrfs_block_group_cache
*cache
;
3016 struct list_head
*l
;
3018 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
3019 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
3020 info
->bytes_pinned
- info
->bytes_reserved
),
3021 (info
->full
) ? "" : "not ");
3023 down_read(&info
->groups_sem
);
3024 list_for_each(l
, &info
->block_groups
) {
3025 cache
= list_entry(l
, struct btrfs_block_group_cache
, list
);
3026 spin_lock(&cache
->lock
);
3027 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
3028 "%llu pinned %llu reserved\n",
3029 (unsigned long long)cache
->key
.objectid
,
3030 (unsigned long long)cache
->key
.offset
,
3031 (unsigned long long)btrfs_block_group_used(&cache
->item
),
3032 (unsigned long long)cache
->pinned
,
3033 (unsigned long long)cache
->reserved
);
3034 btrfs_dump_free_space(cache
, bytes
);
3035 spin_unlock(&cache
->lock
);
3037 up_read(&info
->groups_sem
);
3040 static int __btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
3041 struct btrfs_root
*root
,
3042 u64 num_bytes
, u64 min_alloc_size
,
3043 u64 empty_size
, u64 hint_byte
,
3044 u64 search_end
, struct btrfs_key
*ins
,
3048 u64 search_start
= 0;
3050 struct btrfs_fs_info
*info
= root
->fs_info
;
3053 alloc_profile
= info
->avail_data_alloc_bits
&
3054 info
->data_alloc_profile
;
3055 data
= BTRFS_BLOCK_GROUP_DATA
| alloc_profile
;
3056 } else if (root
== root
->fs_info
->chunk_root
) {
3057 alloc_profile
= info
->avail_system_alloc_bits
&
3058 info
->system_alloc_profile
;
3059 data
= BTRFS_BLOCK_GROUP_SYSTEM
| alloc_profile
;
3061 alloc_profile
= info
->avail_metadata_alloc_bits
&
3062 info
->metadata_alloc_profile
;
3063 data
= BTRFS_BLOCK_GROUP_METADATA
| alloc_profile
;
3066 data
= btrfs_reduce_alloc_profile(root
, data
);
3068 * the only place that sets empty_size is btrfs_realloc_node, which
3069 * is not called recursively on allocations
3071 if (empty_size
|| root
->ref_cows
) {
3072 if (!(data
& BTRFS_BLOCK_GROUP_METADATA
)) {
3073 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3075 BTRFS_BLOCK_GROUP_METADATA
|
3076 (info
->metadata_alloc_profile
&
3077 info
->avail_metadata_alloc_bits
), 0);
3079 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3080 num_bytes
+ 2 * 1024 * 1024, data
, 0);
3083 WARN_ON(num_bytes
< root
->sectorsize
);
3084 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
3085 search_start
, search_end
, hint_byte
, ins
,
3086 trans
->alloc_exclude_start
,
3087 trans
->alloc_exclude_nr
, data
);
3089 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
3090 num_bytes
= num_bytes
>> 1;
3091 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
3092 num_bytes
= max(num_bytes
, min_alloc_size
);
3093 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3094 num_bytes
, data
, 1);
3098 struct btrfs_space_info
*sinfo
;
3100 sinfo
= __find_space_info(root
->fs_info
, data
);
3101 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
3102 "wanted %llu\n", (unsigned long long)data
,
3103 (unsigned long long)num_bytes
);
3104 dump_space_info(sinfo
, num_bytes
);
3111 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
3113 struct btrfs_block_group_cache
*cache
;
3116 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
3118 printk(KERN_ERR
"Unable to find block group for %llu\n",
3119 (unsigned long long)start
);
3123 ret
= btrfs_discard_extent(root
, start
, len
);
3125 btrfs_add_free_space(cache
, start
, len
);
3126 put_block_group(cache
);
3127 update_reserved_extents(root
, start
, len
, 0);
3132 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
3133 struct btrfs_root
*root
,
3134 u64 num_bytes
, u64 min_alloc_size
,
3135 u64 empty_size
, u64 hint_byte
,
3136 u64 search_end
, struct btrfs_key
*ins
,
3140 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
, min_alloc_size
,
3141 empty_size
, hint_byte
, search_end
, ins
,
3143 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
3147 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
3148 struct btrfs_root
*root
, u64 parent
,
3149 u64 root_objectid
, u64 ref_generation
,
3150 u64 owner
, struct btrfs_key
*ins
)
3156 u64 num_bytes
= ins
->offset
;
3158 struct btrfs_fs_info
*info
= root
->fs_info
;
3159 struct btrfs_root
*extent_root
= info
->extent_root
;
3160 struct btrfs_extent_item
*extent_item
;
3161 struct btrfs_extent_ref
*ref
;
3162 struct btrfs_path
*path
;
3163 struct btrfs_key keys
[2];
3166 parent
= ins
->objectid
;
3168 /* block accounting for super block */
3169 spin_lock(&info
->delalloc_lock
);
3170 super_used
= btrfs_super_bytes_used(&info
->super_copy
);
3171 btrfs_set_super_bytes_used(&info
->super_copy
, super_used
+ num_bytes
);
3173 /* block accounting for root item */
3174 root_used
= btrfs_root_used(&root
->root_item
);
3175 btrfs_set_root_used(&root
->root_item
, root_used
+ num_bytes
);
3176 spin_unlock(&info
->delalloc_lock
);
3178 if (root
== extent_root
) {
3179 struct pending_extent_op
*extent_op
;
3181 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
3184 extent_op
->type
= PENDING_EXTENT_INSERT
;
3185 extent_op
->bytenr
= ins
->objectid
;
3186 extent_op
->num_bytes
= ins
->offset
;
3187 extent_op
->parent
= parent
;
3188 extent_op
->orig_parent
= 0;
3189 extent_op
->generation
= ref_generation
;
3190 extent_op
->orig_generation
= 0;
3191 extent_op
->level
= (int)owner
;
3192 INIT_LIST_HEAD(&extent_op
->list
);
3195 mutex_lock(&root
->fs_info
->extent_ins_mutex
);
3196 set_extent_bits(&root
->fs_info
->extent_ins
, ins
->objectid
,
3197 ins
->objectid
+ ins
->offset
- 1,
3198 EXTENT_WRITEBACK
, GFP_NOFS
);
3199 set_state_private(&root
->fs_info
->extent_ins
,
3200 ins
->objectid
, (unsigned long)extent_op
);
3201 mutex_unlock(&root
->fs_info
->extent_ins_mutex
);
3205 memcpy(&keys
[0], ins
, sizeof(*ins
));
3206 keys
[1].objectid
= ins
->objectid
;
3207 keys
[1].type
= BTRFS_EXTENT_REF_KEY
;
3208 keys
[1].offset
= parent
;
3209 sizes
[0] = sizeof(*extent_item
);
3210 sizes
[1] = sizeof(*ref
);
3212 path
= btrfs_alloc_path();
3215 ret
= btrfs_insert_empty_items(trans
, extent_root
, path
, keys
,
3219 extent_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3220 struct btrfs_extent_item
);
3221 btrfs_set_extent_refs(path
->nodes
[0], extent_item
, 1);
3222 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3223 struct btrfs_extent_ref
);
3225 btrfs_set_ref_root(path
->nodes
[0], ref
, root_objectid
);
3226 btrfs_set_ref_generation(path
->nodes
[0], ref
, ref_generation
);
3227 btrfs_set_ref_objectid(path
->nodes
[0], ref
, owner
);
3228 btrfs_set_ref_num_refs(path
->nodes
[0], ref
, 1);
3230 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3232 trans
->alloc_exclude_start
= 0;
3233 trans
->alloc_exclude_nr
= 0;
3234 btrfs_free_path(path
);
3235 finish_current_insert(trans
, extent_root
, 0);
3236 pending_ret
= del_pending_extents(trans
, extent_root
, 0);
3246 ret
= update_block_group(trans
, root
, ins
->objectid
,
3249 printk(KERN_ERR
"btrfs update block group failed for %llu "
3250 "%llu\n", (unsigned long long)ins
->objectid
,
3251 (unsigned long long)ins
->offset
);
3258 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle
*trans
,
3259 struct btrfs_root
*root
, u64 parent
,
3260 u64 root_objectid
, u64 ref_generation
,
3261 u64 owner
, struct btrfs_key
*ins
)
3265 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
)
3267 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
, root_objectid
,
3268 ref_generation
, owner
, ins
);
3269 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 0);
3274 * this is used by the tree logging recovery code. It records that
3275 * an extent has been allocated and makes sure to clear the free
3276 * space cache bits as well
3278 int btrfs_alloc_logged_extent(struct btrfs_trans_handle
*trans
,
3279 struct btrfs_root
*root
, u64 parent
,
3280 u64 root_objectid
, u64 ref_generation
,
3281 u64 owner
, struct btrfs_key
*ins
)
3284 struct btrfs_block_group_cache
*block_group
;
3286 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
3287 mutex_lock(&block_group
->cache_mutex
);
3288 cache_block_group(root
, block_group
);
3289 mutex_unlock(&block_group
->cache_mutex
);
3291 ret
= btrfs_remove_free_space(block_group
, ins
->objectid
,
3294 put_block_group(block_group
);
3295 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
, root_objectid
,
3296 ref_generation
, owner
, ins
);
3301 * finds a free extent and does all the dirty work required for allocation
3302 * returns the key for the extent through ins, and a tree buffer for
3303 * the first block of the extent through buf.
3305 * returns 0 if everything worked, non-zero otherwise.
3307 int btrfs_alloc_extent(struct btrfs_trans_handle
*trans
,
3308 struct btrfs_root
*root
,
3309 u64 num_bytes
, u64 parent
, u64 min_alloc_size
,
3310 u64 root_objectid
, u64 ref_generation
,
3311 u64 owner_objectid
, u64 empty_size
, u64 hint_byte
,
3312 u64 search_end
, struct btrfs_key
*ins
, u64 data
)
3316 ret
= __btrfs_reserve_extent(trans
, root
, num_bytes
,
3317 min_alloc_size
, empty_size
, hint_byte
,
3318 search_end
, ins
, data
);
3320 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
3321 ret
= __btrfs_alloc_reserved_extent(trans
, root
, parent
,
3322 root_objectid
, ref_generation
,
3323 owner_objectid
, ins
);
3327 update_reserved_extents(root
, ins
->objectid
, ins
->offset
, 1);
3332 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
3333 struct btrfs_root
*root
,
3334 u64 bytenr
, u32 blocksize
)
3336 struct extent_buffer
*buf
;
3338 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
3340 return ERR_PTR(-ENOMEM
);
3341 btrfs_set_header_generation(buf
, trans
->transid
);
3342 btrfs_tree_lock(buf
);
3343 clean_tree_block(trans
, root
, buf
);
3344 btrfs_set_buffer_uptodate(buf
);
3345 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
3346 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
3347 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
3349 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
3350 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
3352 trans
->blocks_used
++;
3357 * helper function to allocate a block for a given tree
3358 * returns the tree buffer or NULL.
3360 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
3361 struct btrfs_root
*root
,
3362 u32 blocksize
, u64 parent
,
3369 struct btrfs_key ins
;
3371 struct extent_buffer
*buf
;
3373 ret
= btrfs_alloc_extent(trans
, root
, blocksize
, parent
, blocksize
,
3374 root_objectid
, ref_generation
, level
,
3375 empty_size
, hint
, (u64
)-1, &ins
, 0);
3378 return ERR_PTR(ret
);
3381 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, blocksize
);
3385 int btrfs_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
3386 struct btrfs_root
*root
, struct extent_buffer
*leaf
)
3389 u64 leaf_generation
;
3390 struct btrfs_key key
;
3391 struct btrfs_file_extent_item
*fi
;
3396 BUG_ON(!btrfs_is_leaf(leaf
));
3397 nritems
= btrfs_header_nritems(leaf
);
3398 leaf_owner
= btrfs_header_owner(leaf
);
3399 leaf_generation
= btrfs_header_generation(leaf
);
3401 for (i
= 0; i
< nritems
; i
++) {
3405 btrfs_item_key_to_cpu(leaf
, &key
, i
);
3406 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
3408 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
3409 if (btrfs_file_extent_type(leaf
, fi
) ==
3410 BTRFS_FILE_EXTENT_INLINE
)
3413 * FIXME make sure to insert a trans record that
3414 * repeats the snapshot del on crash
3416 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3417 if (disk_bytenr
== 0)
3420 ret
= __btrfs_free_extent(trans
, root
, disk_bytenr
,
3421 btrfs_file_extent_disk_num_bytes(leaf
, fi
),
3422 leaf
->start
, leaf_owner
, leaf_generation
,
3426 atomic_inc(&root
->fs_info
->throttle_gen
);
3427 wake_up(&root
->fs_info
->transaction_throttle
);
3433 static noinline
int cache_drop_leaf_ref(struct btrfs_trans_handle
*trans
,
3434 struct btrfs_root
*root
,
3435 struct btrfs_leaf_ref
*ref
)
3439 struct btrfs_extent_info
*info
= ref
->extents
;
3441 for (i
= 0; i
< ref
->nritems
; i
++) {
3442 ret
= __btrfs_free_extent(trans
, root
, info
->bytenr
,
3443 info
->num_bytes
, ref
->bytenr
,
3444 ref
->owner
, ref
->generation
,
3447 atomic_inc(&root
->fs_info
->throttle_gen
);
3448 wake_up(&root
->fs_info
->transaction_throttle
);
3458 static int drop_snap_lookup_refcount(struct btrfs_root
*root
, u64 start
,
3463 ret
= btrfs_lookup_extent_ref(NULL
, root
, start
, len
, refs
);
3466 #if 0 /* some debugging code in case we see problems here */
3467 /* if the refs count is one, it won't get increased again. But
3468 * if the ref count is > 1, someone may be decreasing it at
3469 * the same time we are.
3472 struct extent_buffer
*eb
= NULL
;
3473 eb
= btrfs_find_create_tree_block(root
, start
, len
);
3475 btrfs_tree_lock(eb
);
3477 mutex_lock(&root
->fs_info
->alloc_mutex
);
3478 ret
= lookup_extent_ref(NULL
, root
, start
, len
, refs
);
3480 mutex_unlock(&root
->fs_info
->alloc_mutex
);
3483 btrfs_tree_unlock(eb
);
3484 free_extent_buffer(eb
);
3487 printk(KERN_ERR
"btrfs block %llu went down to one "
3488 "during drop_snap\n", (unsigned long long)start
);
3499 * helper function for drop_snapshot, this walks down the tree dropping ref
3500 * counts as it goes.
3502 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
3503 struct btrfs_root
*root
,
3504 struct btrfs_path
*path
, int *level
)
3510 struct extent_buffer
*next
;
3511 struct extent_buffer
*cur
;
3512 struct extent_buffer
*parent
;
3513 struct btrfs_leaf_ref
*ref
;
3518 WARN_ON(*level
< 0);
3519 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
3520 ret
= drop_snap_lookup_refcount(root
, path
->nodes
[*level
]->start
,
3521 path
->nodes
[*level
]->len
, &refs
);
3527 * walk down to the last node level and free all the leaves
3529 while (*level
>= 0) {
3530 WARN_ON(*level
< 0);
3531 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
3532 cur
= path
->nodes
[*level
];
3534 if (btrfs_header_level(cur
) != *level
)
3537 if (path
->slots
[*level
] >=
3538 btrfs_header_nritems(cur
))
3541 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
3545 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
3546 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
3547 blocksize
= btrfs_level_size(root
, *level
- 1);
3549 ret
= drop_snap_lookup_refcount(root
, bytenr
, blocksize
, &refs
);
3552 parent
= path
->nodes
[*level
];
3553 root_owner
= btrfs_header_owner(parent
);
3554 root_gen
= btrfs_header_generation(parent
);
3555 path
->slots
[*level
]++;
3557 ret
= __btrfs_free_extent(trans
, root
, bytenr
,
3558 blocksize
, parent
->start
,
3559 root_owner
, root_gen
,
3563 atomic_inc(&root
->fs_info
->throttle_gen
);
3564 wake_up(&root
->fs_info
->transaction_throttle
);
3570 * at this point, we have a single ref, and since the
3571 * only place referencing this extent is a dead root
3572 * the reference count should never go higher.
3573 * So, we don't need to check it again
3576 ref
= btrfs_lookup_leaf_ref(root
, bytenr
);
3577 if (ref
&& ref
->generation
!= ptr_gen
) {
3578 btrfs_free_leaf_ref(root
, ref
);
3582 ret
= cache_drop_leaf_ref(trans
, root
, ref
);
3584 btrfs_remove_leaf_ref(root
, ref
);
3585 btrfs_free_leaf_ref(root
, ref
);
3590 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
3591 if (!next
|| !btrfs_buffer_uptodate(next
, ptr_gen
)) {
3592 free_extent_buffer(next
);
3594 next
= read_tree_block(root
, bytenr
, blocksize
,
3599 * this is a debugging check and can go away
3600 * the ref should never go all the way down to 1
3603 ret
= lookup_extent_ref(NULL
, root
, bytenr
, blocksize
,
3609 WARN_ON(*level
<= 0);
3610 if (path
->nodes
[*level
-1])
3611 free_extent_buffer(path
->nodes
[*level
-1]);
3612 path
->nodes
[*level
-1] = next
;
3613 *level
= btrfs_header_level(next
);
3614 path
->slots
[*level
] = 0;
3618 WARN_ON(*level
< 0);
3619 WARN_ON(*level
>= BTRFS_MAX_LEVEL
);
3621 if (path
->nodes
[*level
] == root
->node
) {
3622 parent
= path
->nodes
[*level
];
3623 bytenr
= path
->nodes
[*level
]->start
;
3625 parent
= path
->nodes
[*level
+ 1];
3626 bytenr
= btrfs_node_blockptr(parent
, path
->slots
[*level
+ 1]);
3629 blocksize
= btrfs_level_size(root
, *level
);
3630 root_owner
= btrfs_header_owner(parent
);
3631 root_gen
= btrfs_header_generation(parent
);
3633 ret
= __btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
3634 parent
->start
, root_owner
, root_gen
,
3636 free_extent_buffer(path
->nodes
[*level
]);
3637 path
->nodes
[*level
] = NULL
;
3646 * helper function for drop_subtree, this function is similar to
3647 * walk_down_tree. The main difference is that it checks reference
3648 * counts while tree blocks are locked.
3650 static noinline
int walk_down_subtree(struct btrfs_trans_handle
*trans
,
3651 struct btrfs_root
*root
,
3652 struct btrfs_path
*path
, int *level
)
3654 struct extent_buffer
*next
;
3655 struct extent_buffer
*cur
;
3656 struct extent_buffer
*parent
;
3663 cur
= path
->nodes
[*level
];
3664 ret
= btrfs_lookup_extent_ref(trans
, root
, cur
->start
, cur
->len
,
3670 while (*level
>= 0) {
3671 cur
= path
->nodes
[*level
];
3673 ret
= btrfs_drop_leaf_ref(trans
, root
, cur
);
3675 clean_tree_block(trans
, root
, cur
);
3678 if (path
->slots
[*level
] >= btrfs_header_nritems(cur
)) {
3679 clean_tree_block(trans
, root
, cur
);
3683 bytenr
= btrfs_node_blockptr(cur
, path
->slots
[*level
]);
3684 blocksize
= btrfs_level_size(root
, *level
- 1);
3685 ptr_gen
= btrfs_node_ptr_generation(cur
, path
->slots
[*level
]);
3687 next
= read_tree_block(root
, bytenr
, blocksize
, ptr_gen
);
3688 btrfs_tree_lock(next
);
3690 ret
= btrfs_lookup_extent_ref(trans
, root
, bytenr
, blocksize
,
3694 parent
= path
->nodes
[*level
];
3695 ret
= btrfs_free_extent(trans
, root
, bytenr
,
3696 blocksize
, parent
->start
,
3697 btrfs_header_owner(parent
),
3698 btrfs_header_generation(parent
),
3701 path
->slots
[*level
]++;
3702 btrfs_tree_unlock(next
);
3703 free_extent_buffer(next
);
3707 *level
= btrfs_header_level(next
);
3708 path
->nodes
[*level
] = next
;
3709 path
->slots
[*level
] = 0;
3710 path
->locks
[*level
] = 1;
3714 parent
= path
->nodes
[*level
+ 1];
3715 bytenr
= path
->nodes
[*level
]->start
;
3716 blocksize
= path
->nodes
[*level
]->len
;
3718 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
3719 parent
->start
, btrfs_header_owner(parent
),
3720 btrfs_header_generation(parent
), *level
, 1);
3723 if (path
->locks
[*level
]) {
3724 btrfs_tree_unlock(path
->nodes
[*level
]);
3725 path
->locks
[*level
] = 0;
3727 free_extent_buffer(path
->nodes
[*level
]);
3728 path
->nodes
[*level
] = NULL
;
3735 * helper for dropping snapshots. This walks back up the tree in the path
3736 * to find the first node higher up where we haven't yet gone through
3739 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
3740 struct btrfs_root
*root
,
3741 struct btrfs_path
*path
,
3742 int *level
, int max_level
)
3746 struct btrfs_root_item
*root_item
= &root
->root_item
;
3751 for (i
= *level
; i
< max_level
&& path
->nodes
[i
]; i
++) {
3752 slot
= path
->slots
[i
];
3753 if (slot
< btrfs_header_nritems(path
->nodes
[i
]) - 1) {
3754 struct extent_buffer
*node
;
3755 struct btrfs_disk_key disk_key
;
3756 node
= path
->nodes
[i
];
3759 WARN_ON(*level
== 0);
3760 btrfs_node_key(node
, &disk_key
, path
->slots
[i
]);
3761 memcpy(&root_item
->drop_progress
,
3762 &disk_key
, sizeof(disk_key
));
3763 root_item
->drop_level
= i
;
3766 struct extent_buffer
*parent
;
3767 if (path
->nodes
[*level
] == root
->node
)
3768 parent
= path
->nodes
[*level
];
3770 parent
= path
->nodes
[*level
+ 1];
3772 root_owner
= btrfs_header_owner(parent
);
3773 root_gen
= btrfs_header_generation(parent
);
3775 clean_tree_block(trans
, root
, path
->nodes
[*level
]);
3776 ret
= btrfs_free_extent(trans
, root
,
3777 path
->nodes
[*level
]->start
,
3778 path
->nodes
[*level
]->len
,
3779 parent
->start
, root_owner
,
3780 root_gen
, *level
, 1);
3782 if (path
->locks
[*level
]) {
3783 btrfs_tree_unlock(path
->nodes
[*level
]);
3784 path
->locks
[*level
] = 0;
3786 free_extent_buffer(path
->nodes
[*level
]);
3787 path
->nodes
[*level
] = NULL
;
3795 * drop the reference count on the tree rooted at 'snap'. This traverses
3796 * the tree freeing any blocks that have a ref count of zero after being
3799 int btrfs_drop_snapshot(struct btrfs_trans_handle
*trans
, struct btrfs_root
3805 struct btrfs_path
*path
;
3808 struct btrfs_root_item
*root_item
= &root
->root_item
;
3810 WARN_ON(!mutex_is_locked(&root
->fs_info
->drop_mutex
));
3811 path
= btrfs_alloc_path();
3814 level
= btrfs_header_level(root
->node
);
3816 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
3817 path
->nodes
[level
] = root
->node
;
3818 extent_buffer_get(root
->node
);
3819 path
->slots
[level
] = 0;
3821 struct btrfs_key key
;
3822 struct btrfs_disk_key found_key
;
3823 struct extent_buffer
*node
;
3825 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
3826 level
= root_item
->drop_level
;
3827 path
->lowest_level
= level
;
3828 wret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3833 node
= path
->nodes
[level
];
3834 btrfs_node_key(node
, &found_key
, path
->slots
[level
]);
3835 WARN_ON(memcmp(&found_key
, &root_item
->drop_progress
,
3836 sizeof(found_key
)));
3838 * unlock our path, this is safe because only this
3839 * function is allowed to delete this snapshot
3841 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
3842 if (path
->nodes
[i
] && path
->locks
[i
]) {
3844 btrfs_tree_unlock(path
->nodes
[i
]);
3849 wret
= walk_down_tree(trans
, root
, path
, &level
);
3855 wret
= walk_up_tree(trans
, root
, path
, &level
,
3861 if (trans
->transaction
->in_commit
) {
3865 atomic_inc(&root
->fs_info
->throttle_gen
);
3866 wake_up(&root
->fs_info
->transaction_throttle
);
3868 for (i
= 0; i
<= orig_level
; i
++) {
3869 if (path
->nodes
[i
]) {
3870 free_extent_buffer(path
->nodes
[i
]);
3871 path
->nodes
[i
] = NULL
;
3875 btrfs_free_path(path
);
3879 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
3880 struct btrfs_root
*root
,
3881 struct extent_buffer
*node
,
3882 struct extent_buffer
*parent
)
3884 struct btrfs_path
*path
;
3890 path
= btrfs_alloc_path();
3893 BUG_ON(!btrfs_tree_locked(parent
));
3894 parent_level
= btrfs_header_level(parent
);
3895 extent_buffer_get(parent
);
3896 path
->nodes
[parent_level
] = parent
;
3897 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
3899 BUG_ON(!btrfs_tree_locked(node
));
3900 level
= btrfs_header_level(node
);
3901 extent_buffer_get(node
);
3902 path
->nodes
[level
] = node
;
3903 path
->slots
[level
] = 0;
3906 wret
= walk_down_subtree(trans
, root
, path
, &level
);
3912 wret
= walk_up_tree(trans
, root
, path
, &level
, parent_level
);
3919 btrfs_free_path(path
);
3923 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
3926 return min(last
, start
+ nr
- 1);
3929 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
3934 unsigned long first_index
;
3935 unsigned long last_index
;
3938 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3939 struct file_ra_state
*ra
;
3940 struct btrfs_ordered_extent
*ordered
;
3941 unsigned int total_read
= 0;
3942 unsigned int total_dirty
= 0;
3945 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
3947 mutex_lock(&inode
->i_mutex
);
3948 first_index
= start
>> PAGE_CACHE_SHIFT
;
3949 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
3951 /* make sure the dirty trick played by the caller work */
3952 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
3953 first_index
, last_index
);
3957 file_ra_state_init(ra
, inode
->i_mapping
);
3959 for (i
= first_index
; i
<= last_index
; i
++) {
3960 if (total_read
% ra
->ra_pages
== 0) {
3961 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
3962 calc_ra(i
, last_index
, ra
->ra_pages
));
3966 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
3968 page
= grab_cache_page(inode
->i_mapping
, i
);
3973 if (!PageUptodate(page
)) {
3974 btrfs_readpage(NULL
, page
);
3976 if (!PageUptodate(page
)) {
3978 page_cache_release(page
);
3983 wait_on_page_writeback(page
);
3985 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
3986 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3987 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3989 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3991 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3993 page_cache_release(page
);
3994 btrfs_start_ordered_extent(inode
, ordered
, 1);
3995 btrfs_put_ordered_extent(ordered
);
3998 set_page_extent_mapped(page
);
4000 if (i
== first_index
)
4001 set_extent_bits(io_tree
, page_start
, page_end
,
4002 EXTENT_BOUNDARY
, GFP_NOFS
);
4003 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4005 set_page_dirty(page
);
4008 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4010 page_cache_release(page
);
4015 mutex_unlock(&inode
->i_mutex
);
4016 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
4020 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
4021 struct btrfs_key
*extent_key
,
4024 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
4025 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
4026 struct extent_map
*em
;
4027 u64 start
= extent_key
->objectid
- offset
;
4028 u64 end
= start
+ extent_key
->offset
- 1;
4030 em
= alloc_extent_map(GFP_NOFS
);
4031 BUG_ON(!em
|| IS_ERR(em
));
4034 em
->len
= extent_key
->offset
;
4035 em
->block_len
= extent_key
->offset
;
4036 em
->block_start
= extent_key
->objectid
;
4037 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4038 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
4040 /* setup extent map to cheat btrfs_readpage */
4041 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
4044 spin_lock(&em_tree
->lock
);
4045 ret
= add_extent_mapping(em_tree
, em
);
4046 spin_unlock(&em_tree
->lock
);
4047 if (ret
!= -EEXIST
) {
4048 free_extent_map(em
);
4051 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
4053 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
4055 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
4058 struct btrfs_ref_path
{
4060 u64 nodes
[BTRFS_MAX_LEVEL
];
4062 u64 root_generation
;
4069 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
4070 u64 new_nodes
[BTRFS_MAX_LEVEL
];
4073 struct disk_extent
{
4084 static int is_cowonly_root(u64 root_objectid
)
4086 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
4087 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
4088 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
4089 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
4090 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
4091 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
4096 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
4097 struct btrfs_root
*extent_root
,
4098 struct btrfs_ref_path
*ref_path
,
4101 struct extent_buffer
*leaf
;
4102 struct btrfs_path
*path
;
4103 struct btrfs_extent_ref
*ref
;
4104 struct btrfs_key key
;
4105 struct btrfs_key found_key
;
4111 path
= btrfs_alloc_path();
4116 ref_path
->lowest_level
= -1;
4117 ref_path
->current_level
= -1;
4118 ref_path
->shared_level
= -1;
4122 level
= ref_path
->current_level
- 1;
4123 while (level
>= -1) {
4125 if (level
< ref_path
->lowest_level
)
4129 bytenr
= ref_path
->nodes
[level
];
4131 bytenr
= ref_path
->extent_start
;
4132 BUG_ON(bytenr
== 0);
4134 parent
= ref_path
->nodes
[level
+ 1];
4135 ref_path
->nodes
[level
+ 1] = 0;
4136 ref_path
->current_level
= level
;
4137 BUG_ON(parent
== 0);
4139 key
.objectid
= bytenr
;
4140 key
.offset
= parent
+ 1;
4141 key
.type
= BTRFS_EXTENT_REF_KEY
;
4143 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
4148 leaf
= path
->nodes
[0];
4149 nritems
= btrfs_header_nritems(leaf
);
4150 if (path
->slots
[0] >= nritems
) {
4151 ret
= btrfs_next_leaf(extent_root
, path
);
4156 leaf
= path
->nodes
[0];
4159 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4160 if (found_key
.objectid
== bytenr
&&
4161 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
4162 if (level
< ref_path
->shared_level
)
4163 ref_path
->shared_level
= level
;
4168 btrfs_release_path(extent_root
, path
);
4171 /* reached lowest level */
4175 level
= ref_path
->current_level
;
4176 while (level
< BTRFS_MAX_LEVEL
- 1) {
4180 bytenr
= ref_path
->nodes
[level
];
4182 bytenr
= ref_path
->extent_start
;
4184 BUG_ON(bytenr
== 0);
4186 key
.objectid
= bytenr
;
4188 key
.type
= BTRFS_EXTENT_REF_KEY
;
4190 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
4194 leaf
= path
->nodes
[0];
4195 nritems
= btrfs_header_nritems(leaf
);
4196 if (path
->slots
[0] >= nritems
) {
4197 ret
= btrfs_next_leaf(extent_root
, path
);
4201 /* the extent was freed by someone */
4202 if (ref_path
->lowest_level
== level
)
4204 btrfs_release_path(extent_root
, path
);
4207 leaf
= path
->nodes
[0];
4210 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4211 if (found_key
.objectid
!= bytenr
||
4212 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
4213 /* the extent was freed by someone */
4214 if (ref_path
->lowest_level
== level
) {
4218 btrfs_release_path(extent_root
, path
);
4222 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
4223 struct btrfs_extent_ref
);
4224 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
4225 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4227 level
= (int)ref_objectid
;
4228 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
4229 ref_path
->lowest_level
= level
;
4230 ref_path
->current_level
= level
;
4231 ref_path
->nodes
[level
] = bytenr
;
4233 WARN_ON(ref_objectid
!= level
);
4236 WARN_ON(level
!= -1);
4240 if (ref_path
->lowest_level
== level
) {
4241 ref_path
->owner_objectid
= ref_objectid
;
4242 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
4246 * the block is tree root or the block isn't in reference
4249 if (found_key
.objectid
== found_key
.offset
||
4250 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
4251 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
4252 ref_path
->root_generation
=
4253 btrfs_ref_generation(leaf
, ref
);
4255 /* special reference from the tree log */
4256 ref_path
->nodes
[0] = found_key
.offset
;
4257 ref_path
->current_level
= 0;
4264 BUG_ON(ref_path
->nodes
[level
] != 0);
4265 ref_path
->nodes
[level
] = found_key
.offset
;
4266 ref_path
->current_level
= level
;
4269 * the reference was created in the running transaction,
4270 * no need to continue walking up.
4272 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
4273 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
4274 ref_path
->root_generation
=
4275 btrfs_ref_generation(leaf
, ref
);
4280 btrfs_release_path(extent_root
, path
);
4283 /* reached max tree level, but no tree root found. */
4286 btrfs_free_path(path
);
4290 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
4291 struct btrfs_root
*extent_root
,
4292 struct btrfs_ref_path
*ref_path
,
4295 memset(ref_path
, 0, sizeof(*ref_path
));
4296 ref_path
->extent_start
= extent_start
;
4298 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
4301 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
4302 struct btrfs_root
*extent_root
,
4303 struct btrfs_ref_path
*ref_path
)
4305 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
4308 static noinline
int get_new_locations(struct inode
*reloc_inode
,
4309 struct btrfs_key
*extent_key
,
4310 u64 offset
, int no_fragment
,
4311 struct disk_extent
**extents
,
4314 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
4315 struct btrfs_path
*path
;
4316 struct btrfs_file_extent_item
*fi
;
4317 struct extent_buffer
*leaf
;
4318 struct disk_extent
*exts
= *extents
;
4319 struct btrfs_key found_key
;
4324 int max
= *nr_extents
;
4327 WARN_ON(!no_fragment
&& *extents
);
4330 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
4335 path
= btrfs_alloc_path();
4338 cur_pos
= extent_key
->objectid
- offset
;
4339 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
4340 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
4350 leaf
= path
->nodes
[0];
4351 nritems
= btrfs_header_nritems(leaf
);
4352 if (path
->slots
[0] >= nritems
) {
4353 ret
= btrfs_next_leaf(root
, path
);
4358 leaf
= path
->nodes
[0];
4361 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4362 if (found_key
.offset
!= cur_pos
||
4363 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
4364 found_key
.objectid
!= reloc_inode
->i_ino
)
4367 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4368 struct btrfs_file_extent_item
);
4369 if (btrfs_file_extent_type(leaf
, fi
) !=
4370 BTRFS_FILE_EXTENT_REG
||
4371 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
4375 struct disk_extent
*old
= exts
;
4377 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
4378 memcpy(exts
, old
, sizeof(*exts
) * nr
);
4379 if (old
!= *extents
)
4383 exts
[nr
].disk_bytenr
=
4384 btrfs_file_extent_disk_bytenr(leaf
, fi
);
4385 exts
[nr
].disk_num_bytes
=
4386 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4387 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
4388 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4389 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
4390 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
4391 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
4392 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
4394 BUG_ON(exts
[nr
].offset
> 0);
4395 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
4396 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
4398 cur_pos
+= exts
[nr
].num_bytes
;
4401 if (cur_pos
+ offset
>= last_byte
)
4411 BUG_ON(cur_pos
+ offset
> last_byte
);
4412 if (cur_pos
+ offset
< last_byte
) {
4418 btrfs_free_path(path
);
4420 if (exts
!= *extents
)
4429 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
4430 struct btrfs_root
*root
,
4431 struct btrfs_path
*path
,
4432 struct btrfs_key
*extent_key
,
4433 struct btrfs_key
*leaf_key
,
4434 struct btrfs_ref_path
*ref_path
,
4435 struct disk_extent
*new_extents
,
4438 struct extent_buffer
*leaf
;
4439 struct btrfs_file_extent_item
*fi
;
4440 struct inode
*inode
= NULL
;
4441 struct btrfs_key key
;
4449 int extent_locked
= 0;
4453 memcpy(&key
, leaf_key
, sizeof(key
));
4454 first_pos
= INT_LIMIT(loff_t
) - extent_key
->offset
;
4455 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
4456 if (key
.objectid
< ref_path
->owner_objectid
||
4457 (key
.objectid
== ref_path
->owner_objectid
&&
4458 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
4459 key
.objectid
= ref_path
->owner_objectid
;
4460 key
.type
= BTRFS_EXTENT_DATA_KEY
;
4466 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4470 leaf
= path
->nodes
[0];
4471 nritems
= btrfs_header_nritems(leaf
);
4473 if (extent_locked
&& ret
> 0) {
4475 * the file extent item was modified by someone
4476 * before the extent got locked.
4478 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4479 lock_end
, GFP_NOFS
);
4483 if (path
->slots
[0] >= nritems
) {
4484 if (++nr_scaned
> 2)
4487 BUG_ON(extent_locked
);
4488 ret
= btrfs_next_leaf(root
, path
);
4493 leaf
= path
->nodes
[0];
4494 nritems
= btrfs_header_nritems(leaf
);
4497 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4499 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
4500 if ((key
.objectid
> ref_path
->owner_objectid
) ||
4501 (key
.objectid
== ref_path
->owner_objectid
&&
4502 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
4503 (key
.offset
>= first_pos
+ extent_key
->offset
))
4507 if (inode
&& key
.objectid
!= inode
->i_ino
) {
4508 BUG_ON(extent_locked
);
4509 btrfs_release_path(root
, path
);
4510 mutex_unlock(&inode
->i_mutex
);
4516 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
4521 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4522 struct btrfs_file_extent_item
);
4523 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4524 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
4525 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
4526 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
4527 extent_key
->objectid
)) {
4533 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4534 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
4536 if (first_pos
> key
.offset
- ext_offset
)
4537 first_pos
= key
.offset
- ext_offset
;
4539 if (!extent_locked
) {
4540 lock_start
= key
.offset
;
4541 lock_end
= lock_start
+ num_bytes
- 1;
4543 if (lock_start
> key
.offset
||
4544 lock_end
+ 1 < key
.offset
+ num_bytes
) {
4545 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4546 lock_start
, lock_end
, GFP_NOFS
);
4552 btrfs_release_path(root
, path
);
4554 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
4555 key
.objectid
, root
);
4556 if (inode
->i_state
& I_NEW
) {
4557 BTRFS_I(inode
)->root
= root
;
4558 BTRFS_I(inode
)->location
.objectid
=
4560 BTRFS_I(inode
)->location
.type
=
4561 BTRFS_INODE_ITEM_KEY
;
4562 BTRFS_I(inode
)->location
.offset
= 0;
4563 btrfs_read_locked_inode(inode
);
4564 unlock_new_inode(inode
);
4567 * some code call btrfs_commit_transaction while
4568 * holding the i_mutex, so we can't use mutex_lock
4571 if (is_bad_inode(inode
) ||
4572 !mutex_trylock(&inode
->i_mutex
)) {
4575 key
.offset
= (u64
)-1;
4580 if (!extent_locked
) {
4581 struct btrfs_ordered_extent
*ordered
;
4583 btrfs_release_path(root
, path
);
4585 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4586 lock_end
, GFP_NOFS
);
4587 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4590 ordered
->file_offset
<= lock_end
&&
4591 ordered
->file_offset
+ ordered
->len
> lock_start
) {
4592 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4593 lock_start
, lock_end
, GFP_NOFS
);
4594 btrfs_start_ordered_extent(inode
, ordered
, 1);
4595 btrfs_put_ordered_extent(ordered
);
4596 key
.offset
+= num_bytes
;
4600 btrfs_put_ordered_extent(ordered
);
4606 if (nr_extents
== 1) {
4607 /* update extent pointer in place */
4608 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4609 new_extents
[0].disk_bytenr
);
4610 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4611 new_extents
[0].disk_num_bytes
);
4612 btrfs_mark_buffer_dirty(leaf
);
4614 btrfs_drop_extent_cache(inode
, key
.offset
,
4615 key
.offset
+ num_bytes
- 1, 0);
4617 ret
= btrfs_inc_extent_ref(trans
, root
,
4618 new_extents
[0].disk_bytenr
,
4619 new_extents
[0].disk_num_bytes
,
4621 root
->root_key
.objectid
,
4626 ret
= btrfs_free_extent(trans
, root
,
4627 extent_key
->objectid
,
4630 btrfs_header_owner(leaf
),
4631 btrfs_header_generation(leaf
),
4635 btrfs_release_path(root
, path
);
4636 key
.offset
+= num_bytes
;
4644 * drop old extent pointer at first, then insert the
4645 * new pointers one bye one
4647 btrfs_release_path(root
, path
);
4648 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
4649 key
.offset
+ num_bytes
,
4650 key
.offset
, &alloc_hint
);
4653 for (i
= 0; i
< nr_extents
; i
++) {
4654 if (ext_offset
>= new_extents
[i
].num_bytes
) {
4655 ext_offset
-= new_extents
[i
].num_bytes
;
4658 extent_len
= min(new_extents
[i
].num_bytes
-
4659 ext_offset
, num_bytes
);
4661 ret
= btrfs_insert_empty_item(trans
, root
,
4666 leaf
= path
->nodes
[0];
4667 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4668 struct btrfs_file_extent_item
);
4669 btrfs_set_file_extent_generation(leaf
, fi
,
4671 btrfs_set_file_extent_type(leaf
, fi
,
4672 BTRFS_FILE_EXTENT_REG
);
4673 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4674 new_extents
[i
].disk_bytenr
);
4675 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4676 new_extents
[i
].disk_num_bytes
);
4677 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
4678 new_extents
[i
].ram_bytes
);
4680 btrfs_set_file_extent_compression(leaf
, fi
,
4681 new_extents
[i
].compression
);
4682 btrfs_set_file_extent_encryption(leaf
, fi
,
4683 new_extents
[i
].encryption
);
4684 btrfs_set_file_extent_other_encoding(leaf
, fi
,
4685 new_extents
[i
].other_encoding
);
4687 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4689 ext_offset
+= new_extents
[i
].offset
;
4690 btrfs_set_file_extent_offset(leaf
, fi
,
4692 btrfs_mark_buffer_dirty(leaf
);
4694 btrfs_drop_extent_cache(inode
, key
.offset
,
4695 key
.offset
+ extent_len
- 1, 0);
4697 ret
= btrfs_inc_extent_ref(trans
, root
,
4698 new_extents
[i
].disk_bytenr
,
4699 new_extents
[i
].disk_num_bytes
,
4701 root
->root_key
.objectid
,
4702 trans
->transid
, key
.objectid
);
4704 btrfs_release_path(root
, path
);
4706 inode_add_bytes(inode
, extent_len
);
4709 num_bytes
-= extent_len
;
4710 key
.offset
+= extent_len
;
4715 BUG_ON(i
>= nr_extents
);
4719 if (extent_locked
) {
4720 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4721 lock_end
, GFP_NOFS
);
4725 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
4726 key
.offset
>= first_pos
+ extent_key
->offset
)
4733 btrfs_release_path(root
, path
);
4735 mutex_unlock(&inode
->i_mutex
);
4736 if (extent_locked
) {
4737 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
4738 lock_end
, GFP_NOFS
);
4745 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
4746 struct btrfs_root
*root
,
4747 struct extent_buffer
*buf
, u64 orig_start
)
4752 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
4753 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
4755 level
= btrfs_header_level(buf
);
4757 struct btrfs_leaf_ref
*ref
;
4758 struct btrfs_leaf_ref
*orig_ref
;
4760 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
4764 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
4766 btrfs_free_leaf_ref(root
, orig_ref
);
4770 ref
->nritems
= orig_ref
->nritems
;
4771 memcpy(ref
->extents
, orig_ref
->extents
,
4772 sizeof(ref
->extents
[0]) * ref
->nritems
);
4774 btrfs_free_leaf_ref(root
, orig_ref
);
4776 ref
->root_gen
= trans
->transid
;
4777 ref
->bytenr
= buf
->start
;
4778 ref
->owner
= btrfs_header_owner(buf
);
4779 ref
->generation
= btrfs_header_generation(buf
);
4780 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
4782 btrfs_free_leaf_ref(root
, ref
);
4787 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
4788 struct extent_buffer
*leaf
,
4789 struct btrfs_block_group_cache
*group
,
4790 struct btrfs_root
*target_root
)
4792 struct btrfs_key key
;
4793 struct inode
*inode
= NULL
;
4794 struct btrfs_file_extent_item
*fi
;
4796 u64 skip_objectid
= 0;
4800 nritems
= btrfs_header_nritems(leaf
);
4801 for (i
= 0; i
< nritems
; i
++) {
4802 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4803 if (key
.objectid
== skip_objectid
||
4804 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4806 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4807 if (btrfs_file_extent_type(leaf
, fi
) ==
4808 BTRFS_FILE_EXTENT_INLINE
)
4810 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
4812 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
4814 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
4815 key
.objectid
, target_root
, 1);
4818 skip_objectid
= key
.objectid
;
4821 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
4823 lock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4824 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4825 btrfs_drop_extent_cache(inode
, key
.offset
,
4826 key
.offset
+ num_bytes
- 1, 1);
4827 unlock_extent(&BTRFS_I(inode
)->io_tree
, key
.offset
,
4828 key
.offset
+ num_bytes
- 1, GFP_NOFS
);
4835 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
4836 struct btrfs_root
*root
,
4837 struct extent_buffer
*leaf
,
4838 struct btrfs_block_group_cache
*group
,
4839 struct inode
*reloc_inode
)
4841 struct btrfs_key key
;
4842 struct btrfs_key extent_key
;
4843 struct btrfs_file_extent_item
*fi
;
4844 struct btrfs_leaf_ref
*ref
;
4845 struct disk_extent
*new_extent
;
4854 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
4855 BUG_ON(!new_extent
);
4857 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
4861 nritems
= btrfs_header_nritems(leaf
);
4862 for (i
= 0; i
< nritems
; i
++) {
4863 btrfs_item_key_to_cpu(leaf
, &key
, i
);
4864 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
4866 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
4867 if (btrfs_file_extent_type(leaf
, fi
) ==
4868 BTRFS_FILE_EXTENT_INLINE
)
4870 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4871 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
4876 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
4877 bytenr
+ num_bytes
<= group
->key
.objectid
)
4880 extent_key
.objectid
= bytenr
;
4881 extent_key
.offset
= num_bytes
;
4882 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4884 ret
= get_new_locations(reloc_inode
, &extent_key
,
4885 group
->key
.objectid
, 1,
4886 &new_extent
, &nr_extent
);
4891 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
4892 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
4893 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
4894 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
4896 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
4897 new_extent
->disk_bytenr
);
4898 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
4899 new_extent
->disk_num_bytes
);
4900 btrfs_mark_buffer_dirty(leaf
);
4902 ret
= btrfs_inc_extent_ref(trans
, root
,
4903 new_extent
->disk_bytenr
,
4904 new_extent
->disk_num_bytes
,
4906 root
->root_key
.objectid
,
4907 trans
->transid
, key
.objectid
);
4909 ret
= btrfs_free_extent(trans
, root
,
4910 bytenr
, num_bytes
, leaf
->start
,
4911 btrfs_header_owner(leaf
),
4912 btrfs_header_generation(leaf
),
4918 BUG_ON(ext_index
+ 1 != ref
->nritems
);
4919 btrfs_free_leaf_ref(root
, ref
);
4923 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
4924 struct btrfs_root
*root
)
4926 struct btrfs_root
*reloc_root
;
4929 if (root
->reloc_root
) {
4930 reloc_root
= root
->reloc_root
;
4931 root
->reloc_root
= NULL
;
4932 list_add(&reloc_root
->dead_list
,
4933 &root
->fs_info
->dead_reloc_roots
);
4935 btrfs_set_root_bytenr(&reloc_root
->root_item
,
4936 reloc_root
->node
->start
);
4937 btrfs_set_root_level(&root
->root_item
,
4938 btrfs_header_level(reloc_root
->node
));
4939 memset(&reloc_root
->root_item
.drop_progress
, 0,
4940 sizeof(struct btrfs_disk_key
));
4941 reloc_root
->root_item
.drop_level
= 0;
4943 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
4944 &reloc_root
->root_key
,
4945 &reloc_root
->root_item
);
4951 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
4953 struct btrfs_trans_handle
*trans
;
4954 struct btrfs_root
*reloc_root
;
4955 struct btrfs_root
*prev_root
= NULL
;
4956 struct list_head dead_roots
;
4960 INIT_LIST_HEAD(&dead_roots
);
4961 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
4963 while (!list_empty(&dead_roots
)) {
4964 reloc_root
= list_entry(dead_roots
.prev
,
4965 struct btrfs_root
, dead_list
);
4966 list_del_init(&reloc_root
->dead_list
);
4968 BUG_ON(reloc_root
->commit_root
!= NULL
);
4970 trans
= btrfs_join_transaction(root
, 1);
4973 mutex_lock(&root
->fs_info
->drop_mutex
);
4974 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
4977 mutex_unlock(&root
->fs_info
->drop_mutex
);
4979 nr
= trans
->blocks_used
;
4980 ret
= btrfs_end_transaction(trans
, root
);
4982 btrfs_btree_balance_dirty(root
, nr
);
4985 free_extent_buffer(reloc_root
->node
);
4987 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
4988 &reloc_root
->root_key
);
4990 mutex_unlock(&root
->fs_info
->drop_mutex
);
4992 nr
= trans
->blocks_used
;
4993 ret
= btrfs_end_transaction(trans
, root
);
4995 btrfs_btree_balance_dirty(root
, nr
);
4998 prev_root
= reloc_root
;
5001 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
5007 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
5009 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
5013 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
5015 struct btrfs_root
*reloc_root
;
5016 struct btrfs_trans_handle
*trans
;
5017 struct btrfs_key location
;
5021 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
5022 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
5024 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
5025 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
5028 trans
= btrfs_start_transaction(root
, 1);
5030 ret
= btrfs_commit_transaction(trans
, root
);
5034 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
5035 location
.offset
= (u64
)-1;
5036 location
.type
= BTRFS_ROOT_ITEM_KEY
;
5038 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
5039 BUG_ON(!reloc_root
);
5040 btrfs_orphan_cleanup(reloc_root
);
5044 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
5045 struct btrfs_root
*root
)
5047 struct btrfs_root
*reloc_root
;
5048 struct extent_buffer
*eb
;
5049 struct btrfs_root_item
*root_item
;
5050 struct btrfs_key root_key
;
5053 BUG_ON(!root
->ref_cows
);
5054 if (root
->reloc_root
)
5057 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
5060 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
5061 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
5064 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
5065 root_key
.offset
= root
->root_key
.objectid
;
5066 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
5068 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
5069 btrfs_set_root_refs(root_item
, 0);
5070 btrfs_set_root_bytenr(root_item
, eb
->start
);
5071 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
5072 btrfs_set_root_generation(root_item
, trans
->transid
);
5074 btrfs_tree_unlock(eb
);
5075 free_extent_buffer(eb
);
5077 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
5078 &root_key
, root_item
);
5082 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
5084 BUG_ON(!reloc_root
);
5085 reloc_root
->last_trans
= trans
->transid
;
5086 reloc_root
->commit_root
= NULL
;
5087 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
5089 root
->reloc_root
= reloc_root
;
5094 * Core function of space balance.
5096 * The idea is using reloc trees to relocate tree blocks in reference
5097 * counted roots. There is one reloc tree for each subvol, and all
5098 * reloc trees share same root key objectid. Reloc trees are snapshots
5099 * of the latest committed roots of subvols (root->commit_root).
5101 * To relocate a tree block referenced by a subvol, there are two steps.
5102 * COW the block through subvol's reloc tree, then update block pointer
5103 * in the subvol to point to the new block. Since all reloc trees share
5104 * same root key objectid, doing special handing for tree blocks owned
5105 * by them is easy. Once a tree block has been COWed in one reloc tree,
5106 * we can use the resulting new block directly when the same block is
5107 * required to COW again through other reloc trees. By this way, relocated
5108 * tree blocks are shared between reloc trees, so they are also shared
5111 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
5112 struct btrfs_root
*root
,
5113 struct btrfs_path
*path
,
5114 struct btrfs_key
*first_key
,
5115 struct btrfs_ref_path
*ref_path
,
5116 struct btrfs_block_group_cache
*group
,
5117 struct inode
*reloc_inode
)
5119 struct btrfs_root
*reloc_root
;
5120 struct extent_buffer
*eb
= NULL
;
5121 struct btrfs_key
*keys
;
5125 int lowest_level
= 0;
5128 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
5129 lowest_level
= ref_path
->owner_objectid
;
5131 if (!root
->ref_cows
) {
5132 path
->lowest_level
= lowest_level
;
5133 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
5135 path
->lowest_level
= 0;
5136 btrfs_release_path(root
, path
);
5140 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
5141 ret
= init_reloc_tree(trans
, root
);
5143 reloc_root
= root
->reloc_root
;
5145 shared_level
= ref_path
->shared_level
;
5146 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
5148 keys
= ref_path
->node_keys
;
5149 nodes
= ref_path
->new_nodes
;
5150 memset(&keys
[shared_level
+ 1], 0,
5151 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
5152 memset(&nodes
[shared_level
+ 1], 0,
5153 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
5155 if (nodes
[lowest_level
] == 0) {
5156 path
->lowest_level
= lowest_level
;
5157 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
5160 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
5161 eb
= path
->nodes
[level
];
5162 if (!eb
|| eb
== reloc_root
->node
)
5164 nodes
[level
] = eb
->start
;
5166 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
5168 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
5171 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5172 eb
= path
->nodes
[0];
5173 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
5174 group
, reloc_inode
);
5177 btrfs_release_path(reloc_root
, path
);
5179 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
5185 * replace tree blocks in the fs tree with tree blocks in
5188 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
5191 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5192 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
5195 extent_buffer_get(path
->nodes
[0]);
5196 eb
= path
->nodes
[0];
5197 btrfs_release_path(reloc_root
, path
);
5198 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
5200 free_extent_buffer(eb
);
5203 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
5204 path
->lowest_level
= 0;
5208 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
5209 struct btrfs_root
*root
,
5210 struct btrfs_path
*path
,
5211 struct btrfs_key
*first_key
,
5212 struct btrfs_ref_path
*ref_path
)
5216 ret
= relocate_one_path(trans
, root
, path
, first_key
,
5217 ref_path
, NULL
, NULL
);
5220 if (root
== root
->fs_info
->extent_root
)
5221 btrfs_extent_post_op(trans
, root
);
5226 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
5227 struct btrfs_root
*extent_root
,
5228 struct btrfs_path
*path
,
5229 struct btrfs_key
*extent_key
)
5233 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
5236 ret
= btrfs_del_item(trans
, extent_root
, path
);
5238 btrfs_release_path(extent_root
, path
);
5242 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
5243 struct btrfs_ref_path
*ref_path
)
5245 struct btrfs_key root_key
;
5247 root_key
.objectid
= ref_path
->root_objectid
;
5248 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
5249 if (is_cowonly_root(ref_path
->root_objectid
))
5250 root_key
.offset
= 0;
5252 root_key
.offset
= (u64
)-1;
5254 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
5257 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
5258 struct btrfs_path
*path
,
5259 struct btrfs_key
*extent_key
,
5260 struct btrfs_block_group_cache
*group
,
5261 struct inode
*reloc_inode
, int pass
)
5263 struct btrfs_trans_handle
*trans
;
5264 struct btrfs_root
*found_root
;
5265 struct btrfs_ref_path
*ref_path
= NULL
;
5266 struct disk_extent
*new_extents
= NULL
;
5271 struct btrfs_key first_key
;
5275 trans
= btrfs_start_transaction(extent_root
, 1);
5278 if (extent_key
->objectid
== 0) {
5279 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
5283 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
5289 for (loops
= 0; ; loops
++) {
5291 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
5292 extent_key
->objectid
);
5294 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
5301 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
5302 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
5305 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
5306 BUG_ON(!found_root
);
5308 * for reference counted tree, only process reference paths
5309 * rooted at the latest committed root.
5311 if (found_root
->ref_cows
&&
5312 ref_path
->root_generation
!= found_root
->root_key
.offset
)
5315 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5318 * copy data extents to new locations
5320 u64 group_start
= group
->key
.objectid
;
5321 ret
= relocate_data_extent(reloc_inode
,
5330 level
= ref_path
->owner_objectid
;
5333 if (prev_block
!= ref_path
->nodes
[level
]) {
5334 struct extent_buffer
*eb
;
5335 u64 block_start
= ref_path
->nodes
[level
];
5336 u64 block_size
= btrfs_level_size(found_root
, level
);
5338 eb
= read_tree_block(found_root
, block_start
,
5340 btrfs_tree_lock(eb
);
5341 BUG_ON(level
!= btrfs_header_level(eb
));
5344 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
5346 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
5348 btrfs_tree_unlock(eb
);
5349 free_extent_buffer(eb
);
5350 prev_block
= block_start
;
5353 btrfs_record_root_in_trans(found_root
);
5354 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
5356 * try to update data extent references while
5357 * keeping metadata shared between snapshots.
5360 ret
= relocate_one_path(trans
, found_root
,
5361 path
, &first_key
, ref_path
,
5362 group
, reloc_inode
);
5368 * use fallback method to process the remaining
5372 u64 group_start
= group
->key
.objectid
;
5373 new_extents
= kmalloc(sizeof(*new_extents
),
5376 ret
= get_new_locations(reloc_inode
,
5384 ret
= replace_one_extent(trans
, found_root
,
5386 &first_key
, ref_path
,
5387 new_extents
, nr_extents
);
5389 ret
= relocate_tree_block(trans
, found_root
, path
,
5390 &first_key
, ref_path
);
5397 btrfs_end_transaction(trans
, extent_root
);
5403 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
5406 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
5407 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
5409 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
5410 if (num_devices
== 1) {
5411 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
5412 stripped
= flags
& ~stripped
;
5414 /* turn raid0 into single device chunks */
5415 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
5418 /* turn mirroring into duplication */
5419 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
5420 BTRFS_BLOCK_GROUP_RAID10
))
5421 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
5424 /* they already had raid on here, just return */
5425 if (flags
& stripped
)
5428 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
5429 stripped
= flags
& ~stripped
;
5431 /* switch duplicated blocks with raid1 */
5432 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
5433 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
5435 /* turn single device chunks into raid0 */
5436 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
5441 static int __alloc_chunk_for_shrink(struct btrfs_root
*root
,
5442 struct btrfs_block_group_cache
*shrink_block_group
,
5445 struct btrfs_trans_handle
*trans
;
5446 u64 new_alloc_flags
;
5449 spin_lock(&shrink_block_group
->lock
);
5450 if (btrfs_block_group_used(&shrink_block_group
->item
) > 0) {
5451 spin_unlock(&shrink_block_group
->lock
);
5453 trans
= btrfs_start_transaction(root
, 1);
5454 spin_lock(&shrink_block_group
->lock
);
5456 new_alloc_flags
= update_block_group_flags(root
,
5457 shrink_block_group
->flags
);
5458 if (new_alloc_flags
!= shrink_block_group
->flags
) {
5460 btrfs_block_group_used(&shrink_block_group
->item
);
5462 calc
= shrink_block_group
->key
.offset
;
5464 spin_unlock(&shrink_block_group
->lock
);
5466 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5467 calc
+ 2 * 1024 * 1024, new_alloc_flags
, force
);
5469 btrfs_end_transaction(trans
, root
);
5471 spin_unlock(&shrink_block_group
->lock
);
5475 static int __insert_orphan_inode(struct btrfs_trans_handle
*trans
,
5476 struct btrfs_root
*root
,
5477 u64 objectid
, u64 size
)
5479 struct btrfs_path
*path
;
5480 struct btrfs_inode_item
*item
;
5481 struct extent_buffer
*leaf
;
5484 path
= btrfs_alloc_path();
5488 ret
= btrfs_insert_empty_inode(trans
, root
, path
, objectid
);
5492 leaf
= path
->nodes
[0];
5493 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_inode_item
);
5494 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
5495 btrfs_set_inode_generation(leaf
, item
, 1);
5496 btrfs_set_inode_size(leaf
, item
, size
);
5497 btrfs_set_inode_mode(leaf
, item
, S_IFREG
| 0600);
5498 btrfs_set_inode_flags(leaf
, item
, BTRFS_INODE_NOCOMPRESS
);
5499 btrfs_mark_buffer_dirty(leaf
);
5500 btrfs_release_path(root
, path
);
5502 btrfs_free_path(path
);
5506 static noinline
struct inode
*create_reloc_inode(struct btrfs_fs_info
*fs_info
,
5507 struct btrfs_block_group_cache
*group
)
5509 struct inode
*inode
= NULL
;
5510 struct btrfs_trans_handle
*trans
;
5511 struct btrfs_root
*root
;
5512 struct btrfs_key root_key
;
5513 u64 objectid
= BTRFS_FIRST_FREE_OBJECTID
;
5516 root_key
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
5517 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
5518 root_key
.offset
= (u64
)-1;
5519 root
= btrfs_read_fs_root_no_name(fs_info
, &root_key
);
5521 return ERR_CAST(root
);
5523 trans
= btrfs_start_transaction(root
, 1);
5526 err
= btrfs_find_free_objectid(trans
, root
, objectid
, &objectid
);
5530 err
= __insert_orphan_inode(trans
, root
, objectid
, group
->key
.offset
);
5533 err
= btrfs_insert_file_extent(trans
, root
, objectid
, 0, 0, 0,
5534 group
->key
.offset
, 0, group
->key
.offset
,
5538 inode
= btrfs_iget_locked(root
->fs_info
->sb
, objectid
, root
);
5539 if (inode
->i_state
& I_NEW
) {
5540 BTRFS_I(inode
)->root
= root
;
5541 BTRFS_I(inode
)->location
.objectid
= objectid
;
5542 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
5543 BTRFS_I(inode
)->location
.offset
= 0;
5544 btrfs_read_locked_inode(inode
);
5545 unlock_new_inode(inode
);
5546 BUG_ON(is_bad_inode(inode
));
5550 BTRFS_I(inode
)->index_cnt
= group
->key
.objectid
;
5552 err
= btrfs_orphan_add(trans
, inode
);
5554 btrfs_end_transaction(trans
, root
);
5558 inode
= ERR_PTR(err
);
5563 int btrfs_reloc_clone_csums(struct inode
*inode
, u64 file_pos
, u64 len
)
5566 struct btrfs_ordered_sum
*sums
;
5567 struct btrfs_sector_sum
*sector_sum
;
5568 struct btrfs_ordered_extent
*ordered
;
5569 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5570 struct list_head list
;
5575 INIT_LIST_HEAD(&list
);
5577 ordered
= btrfs_lookup_ordered_extent(inode
, file_pos
);
5578 BUG_ON(ordered
->file_offset
!= file_pos
|| ordered
->len
!= len
);
5580 disk_bytenr
= file_pos
+ BTRFS_I(inode
)->index_cnt
;
5581 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, disk_bytenr
,
5582 disk_bytenr
+ len
- 1, &list
);
5584 while (!list_empty(&list
)) {
5585 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
5586 list_del_init(&sums
->list
);
5588 sector_sum
= sums
->sums
;
5589 sums
->bytenr
= ordered
->start
;
5592 while (offset
< sums
->len
) {
5593 sector_sum
->bytenr
+= ordered
->start
- disk_bytenr
;
5595 offset
+= root
->sectorsize
;
5598 btrfs_add_ordered_sum(inode
, ordered
, sums
);
5600 btrfs_put_ordered_extent(ordered
);
5604 int btrfs_relocate_block_group(struct btrfs_root
*root
, u64 group_start
)
5606 struct btrfs_trans_handle
*trans
;
5607 struct btrfs_path
*path
;
5608 struct btrfs_fs_info
*info
= root
->fs_info
;
5609 struct extent_buffer
*leaf
;
5610 struct inode
*reloc_inode
;
5611 struct btrfs_block_group_cache
*block_group
;
5612 struct btrfs_key key
;
5621 root
= root
->fs_info
->extent_root
;
5623 block_group
= btrfs_lookup_block_group(info
, group_start
);
5624 BUG_ON(!block_group
);
5626 printk(KERN_INFO
"btrfs relocating block group %llu flags %llu\n",
5627 (unsigned long long)block_group
->key
.objectid
,
5628 (unsigned long long)block_group
->flags
);
5630 path
= btrfs_alloc_path();
5633 reloc_inode
= create_reloc_inode(info
, block_group
);
5634 BUG_ON(IS_ERR(reloc_inode
));
5636 __alloc_chunk_for_shrink(root
, block_group
, 1);
5637 set_block_group_readonly(block_group
);
5639 btrfs_start_delalloc_inodes(info
->tree_root
);
5640 btrfs_wait_ordered_extents(info
->tree_root
, 0);
5645 key
.objectid
= block_group
->key
.objectid
;
5648 cur_byte
= key
.objectid
;
5650 trans
= btrfs_start_transaction(info
->tree_root
, 1);
5651 btrfs_commit_transaction(trans
, info
->tree_root
);
5653 mutex_lock(&root
->fs_info
->cleaner_mutex
);
5654 btrfs_clean_old_snapshots(info
->tree_root
);
5655 btrfs_remove_leaf_refs(info
->tree_root
, (u64
)-1, 1);
5656 mutex_unlock(&root
->fs_info
->cleaner_mutex
);
5659 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5663 leaf
= path
->nodes
[0];
5664 nritems
= btrfs_header_nritems(leaf
);
5665 if (path
->slots
[0] >= nritems
) {
5666 ret
= btrfs_next_leaf(root
, path
);
5673 leaf
= path
->nodes
[0];
5674 nritems
= btrfs_header_nritems(leaf
);
5677 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
5679 if (key
.objectid
>= block_group
->key
.objectid
+
5680 block_group
->key
.offset
)
5683 if (progress
&& need_resched()) {
5684 btrfs_release_path(root
, path
);
5691 if (btrfs_key_type(&key
) != BTRFS_EXTENT_ITEM_KEY
||
5692 key
.objectid
+ key
.offset
<= cur_byte
) {
5698 cur_byte
= key
.objectid
+ key
.offset
;
5699 btrfs_release_path(root
, path
);
5701 __alloc_chunk_for_shrink(root
, block_group
, 0);
5702 ret
= relocate_one_extent(root
, path
, &key
, block_group
,
5708 key
.objectid
= cur_byte
;
5713 btrfs_release_path(root
, path
);
5716 btrfs_wait_ordered_range(reloc_inode
, 0, (u64
)-1);
5717 invalidate_mapping_pages(reloc_inode
->i_mapping
, 0, -1);
5720 if (total_found
> 0) {
5721 printk(KERN_INFO
"btrfs found %llu extents in pass %d\n",
5722 (unsigned long long)total_found
, pass
);
5724 if (total_found
== skipped
&& pass
> 2) {
5726 reloc_inode
= create_reloc_inode(info
, block_group
);
5732 /* delete reloc_inode */
5735 /* unpin extents in this range */
5736 trans
= btrfs_start_transaction(info
->tree_root
, 1);
5737 btrfs_commit_transaction(trans
, info
->tree_root
);
5739 spin_lock(&block_group
->lock
);
5740 WARN_ON(block_group
->pinned
> 0);
5741 WARN_ON(block_group
->reserved
> 0);
5742 WARN_ON(btrfs_block_group_used(&block_group
->item
) > 0);
5743 spin_unlock(&block_group
->lock
);
5744 put_block_group(block_group
);
5747 btrfs_free_path(path
);
5751 static int find_first_block_group(struct btrfs_root
*root
,
5752 struct btrfs_path
*path
, struct btrfs_key
*key
)
5755 struct btrfs_key found_key
;
5756 struct extent_buffer
*leaf
;
5759 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
5764 slot
= path
->slots
[0];
5765 leaf
= path
->nodes
[0];
5766 if (slot
>= btrfs_header_nritems(leaf
)) {
5767 ret
= btrfs_next_leaf(root
, path
);
5774 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5776 if (found_key
.objectid
>= key
->objectid
&&
5777 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
5788 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
5790 struct btrfs_block_group_cache
*block_group
;
5793 spin_lock(&info
->block_group_cache_lock
);
5794 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
5795 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
5797 rb_erase(&block_group
->cache_node
,
5798 &info
->block_group_cache_tree
);
5799 spin_unlock(&info
->block_group_cache_lock
);
5801 btrfs_remove_free_space_cache(block_group
);
5802 down_write(&block_group
->space_info
->groups_sem
);
5803 list_del(&block_group
->list
);
5804 up_write(&block_group
->space_info
->groups_sem
);
5806 WARN_ON(atomic_read(&block_group
->count
) != 1);
5809 spin_lock(&info
->block_group_cache_lock
);
5811 spin_unlock(&info
->block_group_cache_lock
);
5815 int btrfs_read_block_groups(struct btrfs_root
*root
)
5817 struct btrfs_path
*path
;
5819 struct btrfs_block_group_cache
*cache
;
5820 struct btrfs_fs_info
*info
= root
->fs_info
;
5821 struct btrfs_space_info
*space_info
;
5822 struct btrfs_key key
;
5823 struct btrfs_key found_key
;
5824 struct extent_buffer
*leaf
;
5826 root
= info
->extent_root
;
5829 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
5830 path
= btrfs_alloc_path();
5835 ret
= find_first_block_group(root
, path
, &key
);
5843 leaf
= path
->nodes
[0];
5844 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5845 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5851 atomic_set(&cache
->count
, 1);
5852 spin_lock_init(&cache
->lock
);
5853 mutex_init(&cache
->alloc_mutex
);
5854 mutex_init(&cache
->cache_mutex
);
5855 INIT_LIST_HEAD(&cache
->list
);
5856 read_extent_buffer(leaf
, &cache
->item
,
5857 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
5858 sizeof(cache
->item
));
5859 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
5861 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
5862 btrfs_release_path(root
, path
);
5863 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
5865 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
5866 btrfs_block_group_used(&cache
->item
),
5869 cache
->space_info
= space_info
;
5870 down_write(&space_info
->groups_sem
);
5871 list_add_tail(&cache
->list
, &space_info
->block_groups
);
5872 up_write(&space_info
->groups_sem
);
5874 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5877 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
5878 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
5879 set_block_group_readonly(cache
);
5883 btrfs_free_path(path
);
5887 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
5888 struct btrfs_root
*root
, u64 bytes_used
,
5889 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
5893 struct btrfs_root
*extent_root
;
5894 struct btrfs_block_group_cache
*cache
;
5896 extent_root
= root
->fs_info
->extent_root
;
5898 root
->fs_info
->last_trans_new_blockgroup
= trans
->transid
;
5900 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
5904 cache
->key
.objectid
= chunk_offset
;
5905 cache
->key
.offset
= size
;
5906 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
5907 atomic_set(&cache
->count
, 1);
5908 spin_lock_init(&cache
->lock
);
5909 mutex_init(&cache
->alloc_mutex
);
5910 mutex_init(&cache
->cache_mutex
);
5911 INIT_LIST_HEAD(&cache
->list
);
5913 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
5914 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
5915 cache
->flags
= type
;
5916 btrfs_set_block_group_flags(&cache
->item
, type
);
5918 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
5919 &cache
->space_info
);
5921 down_write(&cache
->space_info
->groups_sem
);
5922 list_add_tail(&cache
->list
, &cache
->space_info
->block_groups
);
5923 up_write(&cache
->space_info
->groups_sem
);
5925 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
5928 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
5929 sizeof(cache
->item
));
5932 finish_current_insert(trans
, extent_root
, 0);
5933 ret
= del_pending_extents(trans
, extent_root
, 0);
5935 set_avail_alloc_bits(extent_root
->fs_info
, type
);
5940 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
5941 struct btrfs_root
*root
, u64 group_start
)
5943 struct btrfs_path
*path
;
5944 struct btrfs_block_group_cache
*block_group
;
5945 struct btrfs_key key
;
5948 root
= root
->fs_info
->extent_root
;
5950 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
5951 BUG_ON(!block_group
);
5952 BUG_ON(!block_group
->ro
);
5954 memcpy(&key
, &block_group
->key
, sizeof(key
));
5956 path
= btrfs_alloc_path();
5959 btrfs_remove_free_space_cache(block_group
);
5960 rb_erase(&block_group
->cache_node
,
5961 &root
->fs_info
->block_group_cache_tree
);
5962 down_write(&block_group
->space_info
->groups_sem
);
5963 list_del(&block_group
->list
);
5964 up_write(&block_group
->space_info
->groups_sem
);
5966 spin_lock(&block_group
->space_info
->lock
);
5967 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
5968 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
5969 spin_unlock(&block_group
->space_info
->lock
);
5970 block_group
->space_info
->full
= 0;
5972 put_block_group(block_group
);
5973 put_block_group(block_group
);
5975 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
5981 ret
= btrfs_del_item(trans
, root
, path
);
5983 btrfs_free_path(path
);