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.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/falloc.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/slab.h>
33 #include <linux/btrfs.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 static struct kmem_cache
*btrfs_inode_defrag_cachep
;
46 * when auto defrag is enabled we
47 * queue up these defrag structs to remember which
48 * inodes need defragging passes
51 struct rb_node rb_node
;
55 * transid where the defrag was added, we search for
56 * extents newer than this
63 /* last offset we were able to defrag */
66 /* if we've wrapped around back to zero once already */
70 static int __compare_inode_defrag(struct inode_defrag
*defrag1
,
71 struct inode_defrag
*defrag2
)
73 if (defrag1
->root
> defrag2
->root
)
75 else if (defrag1
->root
< defrag2
->root
)
77 else if (defrag1
->ino
> defrag2
->ino
)
79 else if (defrag1
->ino
< defrag2
->ino
)
85 /* pop a record for an inode into the defrag tree. The lock
86 * must be held already
88 * If you're inserting a record for an older transid than an
89 * existing record, the transid already in the tree is lowered
91 * If an existing record is found the defrag item you
94 static int __btrfs_add_inode_defrag(struct inode
*inode
,
95 struct inode_defrag
*defrag
)
97 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
98 struct inode_defrag
*entry
;
100 struct rb_node
*parent
= NULL
;
103 p
= &root
->fs_info
->defrag_inodes
.rb_node
;
106 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
108 ret
= __compare_inode_defrag(defrag
, entry
);
110 p
= &parent
->rb_left
;
112 p
= &parent
->rb_right
;
114 /* if we're reinserting an entry for
115 * an old defrag run, make sure to
116 * lower the transid of our existing record
118 if (defrag
->transid
< entry
->transid
)
119 entry
->transid
= defrag
->transid
;
120 if (defrag
->last_offset
> entry
->last_offset
)
121 entry
->last_offset
= defrag
->last_offset
;
125 set_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
);
126 rb_link_node(&defrag
->rb_node
, parent
, p
);
127 rb_insert_color(&defrag
->rb_node
, &root
->fs_info
->defrag_inodes
);
131 static inline int __need_auto_defrag(struct btrfs_root
*root
)
133 if (!btrfs_test_opt(root
, AUTO_DEFRAG
))
136 if (btrfs_fs_closing(root
->fs_info
))
143 * insert a defrag record for this inode if auto defrag is
146 int btrfs_add_inode_defrag(struct btrfs_trans_handle
*trans
,
149 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
150 struct inode_defrag
*defrag
;
154 if (!__need_auto_defrag(root
))
157 if (test_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
))
161 transid
= trans
->transid
;
163 transid
= BTRFS_I(inode
)->root
->last_trans
;
165 defrag
= kmem_cache_zalloc(btrfs_inode_defrag_cachep
, GFP_NOFS
);
169 defrag
->ino
= btrfs_ino(inode
);
170 defrag
->transid
= transid
;
171 defrag
->root
= root
->root_key
.objectid
;
173 spin_lock(&root
->fs_info
->defrag_inodes_lock
);
174 if (!test_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
)) {
176 * If we set IN_DEFRAG flag and evict the inode from memory,
177 * and then re-read this inode, this new inode doesn't have
178 * IN_DEFRAG flag. At the case, we may find the existed defrag.
180 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
182 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
184 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
186 spin_unlock(&root
->fs_info
->defrag_inodes_lock
);
191 * Requeue the defrag object. If there is a defrag object that points to
192 * the same inode in the tree, we will merge them together (by
193 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
195 void btrfs_requeue_inode_defrag(struct inode
*inode
,
196 struct inode_defrag
*defrag
)
198 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
201 if (!__need_auto_defrag(root
))
205 * Here we don't check the IN_DEFRAG flag, because we need merge
208 spin_lock(&root
->fs_info
->defrag_inodes_lock
);
209 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
210 spin_unlock(&root
->fs_info
->defrag_inodes_lock
);
215 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
219 * pick the defragable inode that we want, if it doesn't exist, we will get
222 static struct inode_defrag
*
223 btrfs_pick_defrag_inode(struct btrfs_fs_info
*fs_info
, u64 root
, u64 ino
)
225 struct inode_defrag
*entry
= NULL
;
226 struct inode_defrag tmp
;
228 struct rb_node
*parent
= NULL
;
234 spin_lock(&fs_info
->defrag_inodes_lock
);
235 p
= fs_info
->defrag_inodes
.rb_node
;
238 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
240 ret
= __compare_inode_defrag(&tmp
, entry
);
244 p
= parent
->rb_right
;
249 if (parent
&& __compare_inode_defrag(&tmp
, entry
) > 0) {
250 parent
= rb_next(parent
);
252 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
258 rb_erase(parent
, &fs_info
->defrag_inodes
);
259 spin_unlock(&fs_info
->defrag_inodes_lock
);
263 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info
*fs_info
)
265 struct inode_defrag
*defrag
;
266 struct rb_node
*node
;
268 spin_lock(&fs_info
->defrag_inodes_lock
);
269 node
= rb_first(&fs_info
->defrag_inodes
);
271 rb_erase(node
, &fs_info
->defrag_inodes
);
272 defrag
= rb_entry(node
, struct inode_defrag
, rb_node
);
273 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
275 if (need_resched()) {
276 spin_unlock(&fs_info
->defrag_inodes_lock
);
278 spin_lock(&fs_info
->defrag_inodes_lock
);
281 node
= rb_first(&fs_info
->defrag_inodes
);
283 spin_unlock(&fs_info
->defrag_inodes_lock
);
286 #define BTRFS_DEFRAG_BATCH 1024
288 static int __btrfs_run_defrag_inode(struct btrfs_fs_info
*fs_info
,
289 struct inode_defrag
*defrag
)
291 struct btrfs_root
*inode_root
;
293 struct btrfs_key key
;
294 struct btrfs_ioctl_defrag_range_args range
;
300 key
.objectid
= defrag
->root
;
301 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
302 key
.offset
= (u64
)-1;
304 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
306 inode_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
307 if (IS_ERR(inode_root
)) {
308 ret
= PTR_ERR(inode_root
);
311 if (btrfs_root_refs(&inode_root
->root_item
) == 0) {
316 key
.objectid
= defrag
->ino
;
317 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
319 inode
= btrfs_iget(fs_info
->sb
, &key
, inode_root
, NULL
);
321 ret
= PTR_ERR(inode
);
324 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
326 /* do a chunk of defrag */
327 clear_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
);
328 memset(&range
, 0, sizeof(range
));
330 range
.start
= defrag
->last_offset
;
332 sb_start_write(fs_info
->sb
);
333 num_defrag
= btrfs_defrag_file(inode
, NULL
, &range
, defrag
->transid
,
335 sb_end_write(fs_info
->sb
);
337 * if we filled the whole defrag batch, there
338 * must be more work to do. Queue this defrag
341 if (num_defrag
== BTRFS_DEFRAG_BATCH
) {
342 defrag
->last_offset
= range
.start
;
343 btrfs_requeue_inode_defrag(inode
, defrag
);
344 } else if (defrag
->last_offset
&& !defrag
->cycled
) {
346 * we didn't fill our defrag batch, but
347 * we didn't start at zero. Make sure we loop
348 * around to the start of the file.
350 defrag
->last_offset
= 0;
352 btrfs_requeue_inode_defrag(inode
, defrag
);
354 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
360 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
361 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
366 * run through the list of inodes in the FS that need
369 int btrfs_run_defrag_inodes(struct btrfs_fs_info
*fs_info
)
371 struct inode_defrag
*defrag
;
373 u64 root_objectid
= 0;
375 atomic_inc(&fs_info
->defrag_running
);
377 /* Pause the auto defragger. */
378 if (test_bit(BTRFS_FS_STATE_REMOUNTING
,
382 if (!__need_auto_defrag(fs_info
->tree_root
))
385 /* find an inode to defrag */
386 defrag
= btrfs_pick_defrag_inode(fs_info
, root_objectid
,
389 if (root_objectid
|| first_ino
) {
398 first_ino
= defrag
->ino
+ 1;
399 root_objectid
= defrag
->root
;
401 __btrfs_run_defrag_inode(fs_info
, defrag
);
403 atomic_dec(&fs_info
->defrag_running
);
406 * during unmount, we use the transaction_wait queue to
407 * wait for the defragger to stop
409 wake_up(&fs_info
->transaction_wait
);
413 /* simple helper to fault in pages and copy. This should go away
414 * and be replaced with calls into generic code.
416 static noinline
int btrfs_copy_from_user(loff_t pos
, int num_pages
,
418 struct page
**prepared_pages
,
422 size_t total_copied
= 0;
424 int offset
= pos
& (PAGE_CACHE_SIZE
- 1);
426 while (write_bytes
> 0) {
427 size_t count
= min_t(size_t,
428 PAGE_CACHE_SIZE
- offset
, write_bytes
);
429 struct page
*page
= prepared_pages
[pg
];
431 * Copy data from userspace to the current page
433 * Disable pagefault to avoid recursive lock since
434 * the pages are already locked
437 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, count
);
440 /* Flush processor's dcache for this page */
441 flush_dcache_page(page
);
444 * if we get a partial write, we can end up with
445 * partially up to date pages. These add
446 * a lot of complexity, so make sure they don't
447 * happen by forcing this copy to be retried.
449 * The rest of the btrfs_file_write code will fall
450 * back to page at a time copies after we return 0.
452 if (!PageUptodate(page
) && copied
< count
)
455 iov_iter_advance(i
, copied
);
456 write_bytes
-= copied
;
457 total_copied
+= copied
;
459 /* Return to btrfs_file_aio_write to fault page */
460 if (unlikely(copied
== 0))
463 if (unlikely(copied
< PAGE_CACHE_SIZE
- offset
)) {
474 * unlocks pages after btrfs_file_write is done with them
476 void btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
479 for (i
= 0; i
< num_pages
; i
++) {
480 /* page checked is some magic around finding pages that
481 * have been modified without going through btrfs_set_page_dirty
484 ClearPageChecked(pages
[i
]);
485 unlock_page(pages
[i
]);
486 mark_page_accessed(pages
[i
]);
487 page_cache_release(pages
[i
]);
492 * after copy_from_user, pages need to be dirtied and we need to make
493 * sure holes are created between the current EOF and the start of
494 * any next extents (if required).
496 * this also makes the decision about creating an inline extent vs
497 * doing real data extents, marking pages dirty and delalloc as required.
499 int btrfs_dirty_pages(struct btrfs_root
*root
, struct inode
*inode
,
500 struct page
**pages
, size_t num_pages
,
501 loff_t pos
, size_t write_bytes
,
502 struct extent_state
**cached
)
508 u64 end_of_last_block
;
509 u64 end_pos
= pos
+ write_bytes
;
510 loff_t isize
= i_size_read(inode
);
512 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
513 num_bytes
= ALIGN(write_bytes
+ pos
- start_pos
, root
->sectorsize
);
515 end_of_last_block
= start_pos
+ num_bytes
- 1;
516 err
= btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
,
521 for (i
= 0; i
< num_pages
; i
++) {
522 struct page
*p
= pages
[i
];
529 * we've only changed i_size in ram, and we haven't updated
530 * the disk i_size. There is no need to log the inode
534 i_size_write(inode
, end_pos
);
539 * this drops all the extents in the cache that intersect the range
540 * [start, end]. Existing extents are split as required.
542 void btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
,
545 struct extent_map
*em
;
546 struct extent_map
*split
= NULL
;
547 struct extent_map
*split2
= NULL
;
548 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
549 u64 len
= end
- start
+ 1;
556 WARN_ON(end
< start
);
557 if (end
== (u64
)-1) {
565 split
= alloc_extent_map();
567 split2
= alloc_extent_map();
568 if (!split
|| !split2
)
571 write_lock(&em_tree
->lock
);
572 em
= lookup_extent_mapping(em_tree
, start
, len
);
574 write_unlock(&em_tree
->lock
);
578 gen
= em
->generation
;
579 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
580 if (testend
&& em
->start
+ em
->len
>= start
+ len
) {
582 write_unlock(&em_tree
->lock
);
585 start
= em
->start
+ em
->len
;
587 len
= start
+ len
- (em
->start
+ em
->len
);
589 write_unlock(&em_tree
->lock
);
592 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
593 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
594 clear_bit(EXTENT_FLAG_LOGGING
, &flags
);
595 remove_extent_mapping(em_tree
, em
);
599 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
601 split
->start
= em
->start
;
602 split
->len
= start
- em
->start
;
603 split
->orig_start
= em
->orig_start
;
604 split
->block_start
= em
->block_start
;
607 split
->block_len
= em
->block_len
;
609 split
->block_len
= split
->len
;
610 split
->orig_block_len
= max(split
->block_len
,
612 split
->generation
= gen
;
613 split
->bdev
= em
->bdev
;
614 split
->flags
= flags
;
615 split
->compress_type
= em
->compress_type
;
616 ret
= add_extent_mapping(em_tree
, split
);
617 BUG_ON(ret
); /* Logic error */
618 list_move(&split
->list
, &em_tree
->modified_extents
);
619 free_extent_map(split
);
623 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
624 testend
&& em
->start
+ em
->len
> start
+ len
) {
625 u64 diff
= start
+ len
- em
->start
;
627 split
->start
= start
+ len
;
628 split
->len
= em
->start
+ em
->len
- (start
+ len
);
629 split
->bdev
= em
->bdev
;
630 split
->flags
= flags
;
631 split
->compress_type
= em
->compress_type
;
632 split
->generation
= gen
;
633 split
->orig_block_len
= max(em
->block_len
,
637 split
->block_len
= em
->block_len
;
638 split
->block_start
= em
->block_start
;
639 split
->orig_start
= em
->orig_start
;
641 split
->block_len
= split
->len
;
642 split
->block_start
= em
->block_start
+ diff
;
643 split
->orig_start
= em
->orig_start
;
646 ret
= add_extent_mapping(em_tree
, split
);
647 BUG_ON(ret
); /* Logic error */
648 list_move(&split
->list
, &em_tree
->modified_extents
);
649 free_extent_map(split
);
653 write_unlock(&em_tree
->lock
);
657 /* once for the tree*/
661 free_extent_map(split
);
663 free_extent_map(split2
);
667 * this is very complex, but the basic idea is to drop all extents
668 * in the range start - end. hint_block is filled in with a block number
669 * that would be a good hint to the block allocator for this file.
671 * If an extent intersects the range but is not entirely inside the range
672 * it is either truncated or split. Anything entirely inside the range
673 * is deleted from the tree.
675 int __btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
676 struct btrfs_root
*root
, struct inode
*inode
,
677 struct btrfs_path
*path
, u64 start
, u64 end
,
678 u64
*drop_end
, int drop_cache
)
680 struct extent_buffer
*leaf
;
681 struct btrfs_file_extent_item
*fi
;
682 struct btrfs_key key
;
683 struct btrfs_key new_key
;
684 u64 ino
= btrfs_ino(inode
);
685 u64 search_start
= start
;
688 u64 extent_offset
= 0;
695 int modify_tree
= -1;
696 int update_refs
= (root
->ref_cows
|| root
== root
->fs_info
->tree_root
);
700 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
702 if (start
>= BTRFS_I(inode
)->disk_i_size
)
707 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
708 search_start
, modify_tree
);
711 if (ret
> 0 && path
->slots
[0] > 0 && search_start
== start
) {
712 leaf
= path
->nodes
[0];
713 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0] - 1);
714 if (key
.objectid
== ino
&&
715 key
.type
== BTRFS_EXTENT_DATA_KEY
)
720 leaf
= path
->nodes
[0];
721 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
723 ret
= btrfs_next_leaf(root
, path
);
730 leaf
= path
->nodes
[0];
734 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
735 if (key
.objectid
> ino
||
736 key
.type
> BTRFS_EXTENT_DATA_KEY
|| key
.offset
>= end
)
739 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
740 struct btrfs_file_extent_item
);
741 extent_type
= btrfs_file_extent_type(leaf
, fi
);
743 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
744 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
745 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
746 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
747 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
748 extent_end
= key
.offset
+
749 btrfs_file_extent_num_bytes(leaf
, fi
);
750 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
751 extent_end
= key
.offset
+
752 btrfs_file_extent_inline_len(leaf
, fi
);
755 extent_end
= search_start
;
758 if (extent_end
<= search_start
) {
764 search_start
= max(key
.offset
, start
);
765 if (recow
|| !modify_tree
) {
767 btrfs_release_path(path
);
772 * | - range to drop - |
773 * | -------- extent -------- |
775 if (start
> key
.offset
&& end
< extent_end
) {
777 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
779 memcpy(&new_key
, &key
, sizeof(new_key
));
780 new_key
.offset
= start
;
781 ret
= btrfs_duplicate_item(trans
, root
, path
,
783 if (ret
== -EAGAIN
) {
784 btrfs_release_path(path
);
790 leaf
= path
->nodes
[0];
791 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
792 struct btrfs_file_extent_item
);
793 btrfs_set_file_extent_num_bytes(leaf
, fi
,
796 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
797 struct btrfs_file_extent_item
);
799 extent_offset
+= start
- key
.offset
;
800 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
801 btrfs_set_file_extent_num_bytes(leaf
, fi
,
803 btrfs_mark_buffer_dirty(leaf
);
805 if (update_refs
&& disk_bytenr
> 0) {
806 ret
= btrfs_inc_extent_ref(trans
, root
,
807 disk_bytenr
, num_bytes
, 0,
808 root
->root_key
.objectid
,
810 start
- extent_offset
, 0);
811 BUG_ON(ret
); /* -ENOMEM */
816 * | ---- range to drop ----- |
817 * | -------- extent -------- |
819 if (start
<= key
.offset
&& end
< extent_end
) {
820 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
822 memcpy(&new_key
, &key
, sizeof(new_key
));
823 new_key
.offset
= end
;
824 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
826 extent_offset
+= end
- key
.offset
;
827 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
828 btrfs_set_file_extent_num_bytes(leaf
, fi
,
830 btrfs_mark_buffer_dirty(leaf
);
831 if (update_refs
&& disk_bytenr
> 0)
832 inode_sub_bytes(inode
, end
- key
.offset
);
836 search_start
= extent_end
;
838 * | ---- range to drop ----- |
839 * | -------- extent -------- |
841 if (start
> key
.offset
&& end
>= extent_end
) {
843 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
845 btrfs_set_file_extent_num_bytes(leaf
, fi
,
847 btrfs_mark_buffer_dirty(leaf
);
848 if (update_refs
&& disk_bytenr
> 0)
849 inode_sub_bytes(inode
, extent_end
- start
);
850 if (end
== extent_end
)
858 * | ---- range to drop ----- |
859 * | ------ extent ------ |
861 if (start
<= key
.offset
&& end
>= extent_end
) {
863 del_slot
= path
->slots
[0];
866 BUG_ON(del_slot
+ del_nr
!= path
->slots
[0]);
871 extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
872 inode_sub_bytes(inode
,
873 extent_end
- key
.offset
);
874 extent_end
= ALIGN(extent_end
,
876 } else if (update_refs
&& disk_bytenr
> 0) {
877 ret
= btrfs_free_extent(trans
, root
,
878 disk_bytenr
, num_bytes
, 0,
879 root
->root_key
.objectid
,
880 key
.objectid
, key
.offset
-
882 BUG_ON(ret
); /* -ENOMEM */
883 inode_sub_bytes(inode
,
884 extent_end
- key
.offset
);
887 if (end
== extent_end
)
890 if (path
->slots
[0] + 1 < btrfs_header_nritems(leaf
)) {
895 ret
= btrfs_del_items(trans
, root
, path
, del_slot
,
898 btrfs_abort_transaction(trans
, root
, ret
);
905 btrfs_release_path(path
);
912 if (!ret
&& del_nr
> 0) {
913 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
915 btrfs_abort_transaction(trans
, root
, ret
);
919 *drop_end
= found
? min(end
, extent_end
) : end
;
920 btrfs_release_path(path
);
924 int btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
925 struct btrfs_root
*root
, struct inode
*inode
, u64 start
,
926 u64 end
, int drop_cache
)
928 struct btrfs_path
*path
;
931 path
= btrfs_alloc_path();
934 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, start
, end
, NULL
,
936 btrfs_free_path(path
);
940 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
941 u64 objectid
, u64 bytenr
, u64 orig_offset
,
942 u64
*start
, u64
*end
)
944 struct btrfs_file_extent_item
*fi
;
945 struct btrfs_key key
;
948 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
951 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
952 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
955 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
956 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
957 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
958 btrfs_file_extent_offset(leaf
, fi
) != key
.offset
- orig_offset
||
959 btrfs_file_extent_compression(leaf
, fi
) ||
960 btrfs_file_extent_encryption(leaf
, fi
) ||
961 btrfs_file_extent_other_encoding(leaf
, fi
))
964 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
965 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
974 * Mark extent in the range start - end as written.
976 * This changes extent type from 'pre-allocated' to 'regular'. If only
977 * part of extent is marked as written, the extent will be split into
980 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
981 struct inode
*inode
, u64 start
, u64 end
)
983 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
984 struct extent_buffer
*leaf
;
985 struct btrfs_path
*path
;
986 struct btrfs_file_extent_item
*fi
;
987 struct btrfs_key key
;
988 struct btrfs_key new_key
;
1000 u64 ino
= btrfs_ino(inode
);
1002 path
= btrfs_alloc_path();
1009 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1012 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1015 if (ret
> 0 && path
->slots
[0] > 0)
1018 leaf
= path
->nodes
[0];
1019 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1020 BUG_ON(key
.objectid
!= ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
);
1021 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1022 struct btrfs_file_extent_item
);
1023 BUG_ON(btrfs_file_extent_type(leaf
, fi
) !=
1024 BTRFS_FILE_EXTENT_PREALLOC
);
1025 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
1026 BUG_ON(key
.offset
> start
|| extent_end
< end
);
1028 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1029 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1030 orig_offset
= key
.offset
- btrfs_file_extent_offset(leaf
, fi
);
1031 memcpy(&new_key
, &key
, sizeof(new_key
));
1033 if (start
== key
.offset
&& end
< extent_end
) {
1036 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1037 ino
, bytenr
, orig_offset
,
1038 &other_start
, &other_end
)) {
1039 new_key
.offset
= end
;
1040 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
1041 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1042 struct btrfs_file_extent_item
);
1043 btrfs_set_file_extent_generation(leaf
, fi
,
1045 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1047 btrfs_set_file_extent_offset(leaf
, fi
,
1049 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1050 struct btrfs_file_extent_item
);
1051 btrfs_set_file_extent_generation(leaf
, fi
,
1053 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1055 btrfs_mark_buffer_dirty(leaf
);
1060 if (start
> key
.offset
&& end
== extent_end
) {
1063 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1064 ino
, bytenr
, orig_offset
,
1065 &other_start
, &other_end
)) {
1066 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1067 struct btrfs_file_extent_item
);
1068 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1069 start
- key
.offset
);
1070 btrfs_set_file_extent_generation(leaf
, fi
,
1073 new_key
.offset
= start
;
1074 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
1076 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1077 struct btrfs_file_extent_item
);
1078 btrfs_set_file_extent_generation(leaf
, fi
,
1080 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1082 btrfs_set_file_extent_offset(leaf
, fi
,
1083 start
- orig_offset
);
1084 btrfs_mark_buffer_dirty(leaf
);
1089 while (start
> key
.offset
|| end
< extent_end
) {
1090 if (key
.offset
== start
)
1093 new_key
.offset
= split
;
1094 ret
= btrfs_duplicate_item(trans
, root
, path
, &new_key
);
1095 if (ret
== -EAGAIN
) {
1096 btrfs_release_path(path
);
1100 btrfs_abort_transaction(trans
, root
, ret
);
1104 leaf
= path
->nodes
[0];
1105 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1106 struct btrfs_file_extent_item
);
1107 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1108 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1109 split
- key
.offset
);
1111 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1112 struct btrfs_file_extent_item
);
1114 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1115 btrfs_set_file_extent_offset(leaf
, fi
, split
- orig_offset
);
1116 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1117 extent_end
- split
);
1118 btrfs_mark_buffer_dirty(leaf
);
1120 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
, 0,
1121 root
->root_key
.objectid
,
1122 ino
, orig_offset
, 0);
1123 BUG_ON(ret
); /* -ENOMEM */
1125 if (split
== start
) {
1128 BUG_ON(start
!= key
.offset
);
1137 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1138 ino
, bytenr
, orig_offset
,
1139 &other_start
, &other_end
)) {
1141 btrfs_release_path(path
);
1144 extent_end
= other_end
;
1145 del_slot
= path
->slots
[0] + 1;
1147 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1148 0, root
->root_key
.objectid
,
1149 ino
, orig_offset
, 0);
1150 BUG_ON(ret
); /* -ENOMEM */
1154 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1155 ino
, bytenr
, orig_offset
,
1156 &other_start
, &other_end
)) {
1158 btrfs_release_path(path
);
1161 key
.offset
= other_start
;
1162 del_slot
= path
->slots
[0];
1164 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1165 0, root
->root_key
.objectid
,
1166 ino
, orig_offset
, 0);
1167 BUG_ON(ret
); /* -ENOMEM */
1170 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1171 struct btrfs_file_extent_item
);
1172 btrfs_set_file_extent_type(leaf
, fi
,
1173 BTRFS_FILE_EXTENT_REG
);
1174 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1175 btrfs_mark_buffer_dirty(leaf
);
1177 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
1178 struct btrfs_file_extent_item
);
1179 btrfs_set_file_extent_type(leaf
, fi
,
1180 BTRFS_FILE_EXTENT_REG
);
1181 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1182 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1183 extent_end
- key
.offset
);
1184 btrfs_mark_buffer_dirty(leaf
);
1186 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
1188 btrfs_abort_transaction(trans
, root
, ret
);
1193 btrfs_free_path(path
);
1198 * on error we return an unlocked page and the error value
1199 * on success we return a locked page and 0
1201 static int prepare_uptodate_page(struct page
*page
, u64 pos
,
1202 bool force_uptodate
)
1206 if (((pos
& (PAGE_CACHE_SIZE
- 1)) || force_uptodate
) &&
1207 !PageUptodate(page
)) {
1208 ret
= btrfs_readpage(NULL
, page
);
1212 if (!PageUptodate(page
)) {
1221 * this gets pages into the page cache and locks them down, it also properly
1222 * waits for data=ordered extents to finish before allowing the pages to be
1225 static noinline
int prepare_pages(struct btrfs_root
*root
, struct file
*file
,
1226 struct page
**pages
, size_t num_pages
,
1227 loff_t pos
, unsigned long first_index
,
1228 size_t write_bytes
, bool force_uptodate
)
1230 struct extent_state
*cached_state
= NULL
;
1232 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
1233 struct inode
*inode
= file_inode(file
);
1234 gfp_t mask
= btrfs_alloc_write_mask(inode
->i_mapping
);
1240 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
1241 last_pos
= ((u64
)index
+ num_pages
) << PAGE_CACHE_SHIFT
;
1244 for (i
= 0; i
< num_pages
; i
++) {
1245 pages
[i
] = find_or_create_page(inode
->i_mapping
, index
+ i
,
1246 mask
| __GFP_WRITE
);
1254 err
= prepare_uptodate_page(pages
[i
], pos
,
1256 if (i
== num_pages
- 1)
1257 err
= prepare_uptodate_page(pages
[i
],
1258 pos
+ write_bytes
, false);
1260 page_cache_release(pages
[i
]);
1264 wait_on_page_writeback(pages
[i
]);
1267 if (start_pos
< inode
->i_size
) {
1268 struct btrfs_ordered_extent
*ordered
;
1269 lock_extent_bits(&BTRFS_I(inode
)->io_tree
,
1270 start_pos
, last_pos
- 1, 0, &cached_state
);
1271 ordered
= btrfs_lookup_first_ordered_extent(inode
,
1274 ordered
->file_offset
+ ordered
->len
> start_pos
&&
1275 ordered
->file_offset
< last_pos
) {
1276 btrfs_put_ordered_extent(ordered
);
1277 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1278 start_pos
, last_pos
- 1,
1279 &cached_state
, GFP_NOFS
);
1280 for (i
= 0; i
< num_pages
; i
++) {
1281 unlock_page(pages
[i
]);
1282 page_cache_release(pages
[i
]);
1284 btrfs_wait_ordered_range(inode
, start_pos
,
1285 last_pos
- start_pos
);
1289 btrfs_put_ordered_extent(ordered
);
1291 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start_pos
,
1292 last_pos
- 1, EXTENT_DIRTY
| EXTENT_DELALLOC
|
1293 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
1294 0, 0, &cached_state
, GFP_NOFS
);
1295 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1296 start_pos
, last_pos
- 1, &cached_state
,
1299 for (i
= 0; i
< num_pages
; i
++) {
1300 if (clear_page_dirty_for_io(pages
[i
]))
1301 account_page_redirty(pages
[i
]);
1302 set_page_extent_mapped(pages
[i
]);
1303 WARN_ON(!PageLocked(pages
[i
]));
1307 while (faili
>= 0) {
1308 unlock_page(pages
[faili
]);
1309 page_cache_release(pages
[faili
]);
1316 static noinline ssize_t
__btrfs_buffered_write(struct file
*file
,
1320 struct inode
*inode
= file_inode(file
);
1321 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1322 struct page
**pages
= NULL
;
1323 unsigned long first_index
;
1324 size_t num_written
= 0;
1327 bool force_page_uptodate
= false;
1329 nrptrs
= min((iov_iter_count(i
) + PAGE_CACHE_SIZE
- 1) /
1330 PAGE_CACHE_SIZE
, PAGE_CACHE_SIZE
/
1331 (sizeof(struct page
*)));
1332 nrptrs
= min(nrptrs
, current
->nr_dirtied_pause
- current
->nr_dirtied
);
1333 nrptrs
= max(nrptrs
, 8);
1334 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
1338 first_index
= pos
>> PAGE_CACHE_SHIFT
;
1340 while (iov_iter_count(i
) > 0) {
1341 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1342 size_t write_bytes
= min(iov_iter_count(i
),
1343 nrptrs
* (size_t)PAGE_CACHE_SIZE
-
1345 size_t num_pages
= (write_bytes
+ offset
+
1346 PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1350 WARN_ON(num_pages
> nrptrs
);
1353 * Fault pages before locking them in prepare_pages
1354 * to avoid recursive lock
1356 if (unlikely(iov_iter_fault_in_readable(i
, write_bytes
))) {
1361 ret
= btrfs_delalloc_reserve_space(inode
,
1362 num_pages
<< PAGE_CACHE_SHIFT
);
1367 * This is going to setup the pages array with the number of
1368 * pages we want, so we don't really need to worry about the
1369 * contents of pages from loop to loop
1371 ret
= prepare_pages(root
, file
, pages
, num_pages
,
1372 pos
, first_index
, write_bytes
,
1373 force_page_uptodate
);
1375 btrfs_delalloc_release_space(inode
,
1376 num_pages
<< PAGE_CACHE_SHIFT
);
1380 copied
= btrfs_copy_from_user(pos
, num_pages
,
1381 write_bytes
, pages
, i
);
1384 * if we have trouble faulting in the pages, fall
1385 * back to one page at a time
1387 if (copied
< write_bytes
)
1391 force_page_uptodate
= true;
1394 force_page_uptodate
= false;
1395 dirty_pages
= (copied
+ offset
+
1396 PAGE_CACHE_SIZE
- 1) >>
1401 * If we had a short copy we need to release the excess delaloc
1402 * bytes we reserved. We need to increment outstanding_extents
1403 * because btrfs_delalloc_release_space will decrement it, but
1404 * we still have an outstanding extent for the chunk we actually
1407 if (num_pages
> dirty_pages
) {
1409 spin_lock(&BTRFS_I(inode
)->lock
);
1410 BTRFS_I(inode
)->outstanding_extents
++;
1411 spin_unlock(&BTRFS_I(inode
)->lock
);
1413 btrfs_delalloc_release_space(inode
,
1414 (num_pages
- dirty_pages
) <<
1419 ret
= btrfs_dirty_pages(root
, inode
, pages
,
1420 dirty_pages
, pos
, copied
,
1423 btrfs_delalloc_release_space(inode
,
1424 dirty_pages
<< PAGE_CACHE_SHIFT
);
1425 btrfs_drop_pages(pages
, num_pages
);
1430 btrfs_drop_pages(pages
, num_pages
);
1434 balance_dirty_pages_ratelimited(inode
->i_mapping
);
1435 if (dirty_pages
< (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
1436 btrfs_btree_balance_dirty(root
);
1439 num_written
+= copied
;
1444 return num_written
? num_written
: ret
;
1447 static ssize_t
__btrfs_direct_write(struct kiocb
*iocb
,
1448 const struct iovec
*iov
,
1449 unsigned long nr_segs
, loff_t pos
,
1450 loff_t
*ppos
, size_t count
, size_t ocount
)
1452 struct file
*file
= iocb
->ki_filp
;
1455 ssize_t written_buffered
;
1459 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
, ppos
,
1462 if (written
< 0 || written
== count
)
1467 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
1468 written_buffered
= __btrfs_buffered_write(file
, &i
, pos
);
1469 if (written_buffered
< 0) {
1470 err
= written_buffered
;
1473 endbyte
= pos
+ written_buffered
- 1;
1474 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
1477 written
+= written_buffered
;
1478 *ppos
= pos
+ written_buffered
;
1479 invalidate_mapping_pages(file
->f_mapping
, pos
>> PAGE_CACHE_SHIFT
,
1480 endbyte
>> PAGE_CACHE_SHIFT
);
1482 return written
? written
: err
;
1485 static void update_time_for_write(struct inode
*inode
)
1487 struct timespec now
;
1489 if (IS_NOCMTIME(inode
))
1492 now
= current_fs_time(inode
->i_sb
);
1493 if (!timespec_equal(&inode
->i_mtime
, &now
))
1494 inode
->i_mtime
= now
;
1496 if (!timespec_equal(&inode
->i_ctime
, &now
))
1497 inode
->i_ctime
= now
;
1499 if (IS_I_VERSION(inode
))
1500 inode_inc_iversion(inode
);
1503 static ssize_t
btrfs_file_aio_write(struct kiocb
*iocb
,
1504 const struct iovec
*iov
,
1505 unsigned long nr_segs
, loff_t pos
)
1507 struct file
*file
= iocb
->ki_filp
;
1508 struct inode
*inode
= file_inode(file
);
1509 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1510 loff_t
*ppos
= &iocb
->ki_pos
;
1512 ssize_t num_written
= 0;
1514 size_t count
, ocount
;
1515 bool sync
= (file
->f_flags
& O_DSYNC
) || IS_SYNC(file
->f_mapping
->host
);
1517 sb_start_write(inode
->i_sb
);
1519 mutex_lock(&inode
->i_mutex
);
1521 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
1523 mutex_unlock(&inode
->i_mutex
);
1528 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
1529 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1531 mutex_unlock(&inode
->i_mutex
);
1536 mutex_unlock(&inode
->i_mutex
);
1540 err
= file_remove_suid(file
);
1542 mutex_unlock(&inode
->i_mutex
);
1547 * If BTRFS flips readonly due to some impossible error
1548 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1549 * although we have opened a file as writable, we have
1550 * to stop this write operation to ensure FS consistency.
1552 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
1553 mutex_unlock(&inode
->i_mutex
);
1559 * We reserve space for updating the inode when we reserve space for the
1560 * extent we are going to write, so we will enospc out there. We don't
1561 * need to start yet another transaction to update the inode as we will
1562 * update the inode when we finish writing whatever data we write.
1564 update_time_for_write(inode
);
1566 start_pos
= round_down(pos
, root
->sectorsize
);
1567 if (start_pos
> i_size_read(inode
)) {
1568 err
= btrfs_cont_expand(inode
, i_size_read(inode
), start_pos
);
1570 mutex_unlock(&inode
->i_mutex
);
1576 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
1578 if (unlikely(file
->f_flags
& O_DIRECT
)) {
1579 num_written
= __btrfs_direct_write(iocb
, iov
, nr_segs
,
1580 pos
, ppos
, count
, ocount
);
1584 iov_iter_init(&i
, iov
, nr_segs
, count
, num_written
);
1586 num_written
= __btrfs_buffered_write(file
, &i
, pos
);
1587 if (num_written
> 0)
1588 *ppos
= pos
+ num_written
;
1591 mutex_unlock(&inode
->i_mutex
);
1594 * we want to make sure fsync finds this change
1595 * but we haven't joined a transaction running right now.
1597 * Later on, someone is sure to update the inode and get the
1598 * real transid recorded.
1600 * We set last_trans now to the fs_info generation + 1,
1601 * this will either be one more than the running transaction
1602 * or the generation used for the next transaction if there isn't
1603 * one running right now.
1605 * We also have to set last_sub_trans to the current log transid,
1606 * otherwise subsequent syncs to a file that's been synced in this
1607 * transaction will appear to have already occured.
1609 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
+ 1;
1610 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
1611 if (num_written
> 0 || num_written
== -EIOCBQUEUED
) {
1612 err
= generic_write_sync(file
, pos
, num_written
);
1613 if (err
< 0 && num_written
> 0)
1618 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
1620 sb_end_write(inode
->i_sb
);
1621 current
->backing_dev_info
= NULL
;
1622 return num_written
? num_written
: err
;
1625 int btrfs_release_file(struct inode
*inode
, struct file
*filp
)
1628 * ordered_data_close is set by settattr when we are about to truncate
1629 * a file from a non-zero size to a zero size. This tries to
1630 * flush down new bytes that may have been written if the
1631 * application were using truncate to replace a file in place.
1633 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
1634 &BTRFS_I(inode
)->runtime_flags
)) {
1635 struct btrfs_trans_handle
*trans
;
1636 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1639 * We need to block on a committing transaction to keep us from
1640 * throwing a ordered operation on to the list and causing
1641 * something like sync to deadlock trying to flush out this
1644 trans
= btrfs_start_transaction(root
, 0);
1646 return PTR_ERR(trans
);
1647 btrfs_add_ordered_operation(trans
, BTRFS_I(inode
)->root
, inode
);
1648 btrfs_end_transaction(trans
, root
);
1649 if (inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
1650 filemap_flush(inode
->i_mapping
);
1652 if (filp
->private_data
)
1653 btrfs_ioctl_trans_end(filp
);
1658 * fsync call for both files and directories. This logs the inode into
1659 * the tree log instead of forcing full commits whenever possible.
1661 * It needs to call filemap_fdatawait so that all ordered extent updates are
1662 * in the metadata btree are up to date for copying to the log.
1664 * It drops the inode mutex before doing the tree log commit. This is an
1665 * important optimization for directories because holding the mutex prevents
1666 * new operations on the dir while we write to disk.
1668 int btrfs_sync_file(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1670 struct dentry
*dentry
= file
->f_path
.dentry
;
1671 struct inode
*inode
= dentry
->d_inode
;
1672 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1674 struct btrfs_trans_handle
*trans
;
1677 trace_btrfs_sync_file(file
, datasync
);
1680 * We write the dirty pages in the range and wait until they complete
1681 * out of the ->i_mutex. If so, we can flush the dirty pages by
1682 * multi-task, and make the performance up. See
1683 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1685 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
1686 ret
= filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
1687 if (!ret
&& test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1688 &BTRFS_I(inode
)->runtime_flags
))
1689 ret
= filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
1690 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
1694 mutex_lock(&inode
->i_mutex
);
1697 * We flush the dirty pages again to avoid some dirty pages in the
1700 atomic_inc(&root
->log_batch
);
1701 full_sync
= test_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1702 &BTRFS_I(inode
)->runtime_flags
);
1704 btrfs_wait_ordered_range(inode
, start
, end
- start
+ 1);
1705 atomic_inc(&root
->log_batch
);
1708 * check the transaction that last modified this inode
1709 * and see if its already been committed
1711 if (!BTRFS_I(inode
)->last_trans
) {
1712 mutex_unlock(&inode
->i_mutex
);
1717 * if the last transaction that changed this file was before
1718 * the current transaction, we can bail out now without any
1722 if (btrfs_inode_in_log(inode
, root
->fs_info
->generation
) ||
1723 BTRFS_I(inode
)->last_trans
<=
1724 root
->fs_info
->last_trans_committed
) {
1725 BTRFS_I(inode
)->last_trans
= 0;
1728 * We'v had everything committed since the last time we were
1729 * modified so clear this flag in case it was set for whatever
1730 * reason, it's no longer relevant.
1732 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1733 &BTRFS_I(inode
)->runtime_flags
);
1734 mutex_unlock(&inode
->i_mutex
);
1739 * ok we haven't committed the transaction yet, lets do a commit
1741 if (file
->private_data
)
1742 btrfs_ioctl_trans_end(file
);
1744 trans
= btrfs_start_transaction(root
, 0);
1745 if (IS_ERR(trans
)) {
1746 ret
= PTR_ERR(trans
);
1747 mutex_unlock(&inode
->i_mutex
);
1751 ret
= btrfs_log_dentry_safe(trans
, root
, dentry
);
1753 mutex_unlock(&inode
->i_mutex
);
1757 /* we've logged all the items and now have a consistent
1758 * version of the file in the log. It is possible that
1759 * someone will come in and modify the file, but that's
1760 * fine because the log is consistent on disk, and we
1761 * have references to all of the file's extents
1763 * It is possible that someone will come in and log the
1764 * file again, but that will end up using the synchronization
1765 * inside btrfs_sync_log to keep things safe.
1767 mutex_unlock(&inode
->i_mutex
);
1769 if (ret
!= BTRFS_NO_LOG_SYNC
) {
1772 * If we didn't already wait for ordered extents we need
1776 btrfs_wait_ordered_range(inode
, start
,
1778 ret
= btrfs_commit_transaction(trans
, root
);
1780 ret
= btrfs_sync_log(trans
, root
);
1782 ret
= btrfs_end_transaction(trans
, root
);
1785 btrfs_wait_ordered_range(inode
, start
,
1788 ret
= btrfs_commit_transaction(trans
, root
);
1792 ret
= btrfs_end_transaction(trans
, root
);
1795 return ret
> 0 ? -EIO
: ret
;
1798 static const struct vm_operations_struct btrfs_file_vm_ops
= {
1799 .fault
= filemap_fault
,
1800 .page_mkwrite
= btrfs_page_mkwrite
,
1801 .remap_pages
= generic_file_remap_pages
,
1804 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
1806 struct address_space
*mapping
= filp
->f_mapping
;
1808 if (!mapping
->a_ops
->readpage
)
1811 file_accessed(filp
);
1812 vma
->vm_ops
= &btrfs_file_vm_ops
;
1817 static int hole_mergeable(struct inode
*inode
, struct extent_buffer
*leaf
,
1818 int slot
, u64 start
, u64 end
)
1820 struct btrfs_file_extent_item
*fi
;
1821 struct btrfs_key key
;
1823 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
1826 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
1827 if (key
.objectid
!= btrfs_ino(inode
) ||
1828 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
1831 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
1833 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
1836 if (btrfs_file_extent_disk_bytenr(leaf
, fi
))
1839 if (key
.offset
== end
)
1841 if (key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
) == start
)
1846 static int fill_holes(struct btrfs_trans_handle
*trans
, struct inode
*inode
,
1847 struct btrfs_path
*path
, u64 offset
, u64 end
)
1849 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1850 struct extent_buffer
*leaf
;
1851 struct btrfs_file_extent_item
*fi
;
1852 struct extent_map
*hole_em
;
1853 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1854 struct btrfs_key key
;
1857 key
.objectid
= btrfs_ino(inode
);
1858 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1859 key
.offset
= offset
;
1862 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1867 leaf
= path
->nodes
[0];
1868 if (hole_mergeable(inode
, leaf
, path
->slots
[0]-1, offset
, end
)) {
1872 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1873 struct btrfs_file_extent_item
);
1874 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) +
1876 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1877 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
1878 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1879 btrfs_mark_buffer_dirty(leaf
);
1883 if (hole_mergeable(inode
, leaf
, path
->slots
[0]+1, offset
, end
)) {
1887 key
.offset
= offset
;
1888 btrfs_set_item_key_safe(trans
, root
, path
, &key
);
1889 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1890 struct btrfs_file_extent_item
);
1891 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) + end
-
1893 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1894 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
1895 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1896 btrfs_mark_buffer_dirty(leaf
);
1899 btrfs_release_path(path
);
1901 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
1902 0, 0, end
- offset
, 0, end
- offset
,
1908 btrfs_release_path(path
);
1910 hole_em
= alloc_extent_map();
1912 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
1913 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1914 &BTRFS_I(inode
)->runtime_flags
);
1916 hole_em
->start
= offset
;
1917 hole_em
->len
= end
- offset
;
1918 hole_em
->orig_start
= offset
;
1920 hole_em
->block_start
= EXTENT_MAP_HOLE
;
1921 hole_em
->block_len
= 0;
1922 hole_em
->orig_block_len
= 0;
1923 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1924 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
1925 hole_em
->generation
= trans
->transid
;
1928 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
1929 write_lock(&em_tree
->lock
);
1930 ret
= add_extent_mapping(em_tree
, hole_em
);
1932 list_move(&hole_em
->list
,
1933 &em_tree
->modified_extents
);
1934 write_unlock(&em_tree
->lock
);
1935 } while (ret
== -EEXIST
);
1936 free_extent_map(hole_em
);
1938 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1939 &BTRFS_I(inode
)->runtime_flags
);
1945 static int btrfs_punch_hole(struct inode
*inode
, loff_t offset
, loff_t len
)
1947 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1948 struct extent_state
*cached_state
= NULL
;
1949 struct btrfs_path
*path
;
1950 struct btrfs_block_rsv
*rsv
;
1951 struct btrfs_trans_handle
*trans
;
1952 u64 lockstart
= round_up(offset
, BTRFS_I(inode
)->root
->sectorsize
);
1953 u64 lockend
= round_down(offset
+ len
,
1954 BTRFS_I(inode
)->root
->sectorsize
) - 1;
1955 u64 cur_offset
= lockstart
;
1956 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
1960 bool same_page
= ((offset
>> PAGE_CACHE_SHIFT
) ==
1961 ((offset
+ len
- 1) >> PAGE_CACHE_SHIFT
));
1963 btrfs_wait_ordered_range(inode
, offset
, len
);
1965 mutex_lock(&inode
->i_mutex
);
1967 * We needn't truncate any page which is beyond the end of the file
1968 * because we are sure there is no data there.
1971 * Only do this if we are in the same page and we aren't doing the
1974 if (same_page
&& len
< PAGE_CACHE_SIZE
) {
1975 if (offset
< round_up(inode
->i_size
, PAGE_CACHE_SIZE
))
1976 ret
= btrfs_truncate_page(inode
, offset
, len
, 0);
1977 mutex_unlock(&inode
->i_mutex
);
1981 /* zero back part of the first page */
1982 if (offset
< round_up(inode
->i_size
, PAGE_CACHE_SIZE
)) {
1983 ret
= btrfs_truncate_page(inode
, offset
, 0, 0);
1985 mutex_unlock(&inode
->i_mutex
);
1990 /* zero the front end of the last page */
1991 if (offset
+ len
< round_up(inode
->i_size
, PAGE_CACHE_SIZE
)) {
1992 ret
= btrfs_truncate_page(inode
, offset
+ len
, 0, 1);
1994 mutex_unlock(&inode
->i_mutex
);
1999 if (lockend
< lockstart
) {
2000 mutex_unlock(&inode
->i_mutex
);
2005 struct btrfs_ordered_extent
*ordered
;
2007 truncate_pagecache_range(inode
, lockstart
, lockend
);
2009 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2011 ordered
= btrfs_lookup_first_ordered_extent(inode
, lockend
);
2014 * We need to make sure we have no ordered extents in this range
2015 * and nobody raced in and read a page in this range, if we did
2016 * we need to try again.
2019 (ordered
->file_offset
+ ordered
->len
< lockstart
||
2020 ordered
->file_offset
> lockend
)) &&
2021 !test_range_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
2022 lockend
, EXTENT_UPTODATE
, 0,
2025 btrfs_put_ordered_extent(ordered
);
2029 btrfs_put_ordered_extent(ordered
);
2030 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
,
2031 lockend
, &cached_state
, GFP_NOFS
);
2032 btrfs_wait_ordered_range(inode
, lockstart
,
2033 lockend
- lockstart
+ 1);
2036 path
= btrfs_alloc_path();
2042 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2047 rsv
->size
= btrfs_calc_trunc_metadata_size(root
, 1);
2051 * 1 - update the inode
2052 * 1 - removing the extents in the range
2053 * 1 - adding the hole extent
2055 trans
= btrfs_start_transaction(root
, 3);
2056 if (IS_ERR(trans
)) {
2057 err
= PTR_ERR(trans
);
2061 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
2064 trans
->block_rsv
= rsv
;
2066 while (cur_offset
< lockend
) {
2067 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
2068 cur_offset
, lockend
+ 1,
2073 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2075 ret
= fill_holes(trans
, inode
, path
, cur_offset
, drop_end
);
2081 cur_offset
= drop_end
;
2083 ret
= btrfs_update_inode(trans
, root
, inode
);
2089 btrfs_end_transaction(trans
, root
);
2090 btrfs_btree_balance_dirty(root
);
2092 trans
= btrfs_start_transaction(root
, 3);
2093 if (IS_ERR(trans
)) {
2094 ret
= PTR_ERR(trans
);
2099 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
2101 BUG_ON(ret
); /* shouldn't happen */
2102 trans
->block_rsv
= rsv
;
2110 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2111 ret
= fill_holes(trans
, inode
, path
, cur_offset
, drop_end
);
2121 inode_inc_iversion(inode
);
2122 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2124 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2125 ret
= btrfs_update_inode(trans
, root
, inode
);
2126 btrfs_end_transaction(trans
, root
);
2127 btrfs_btree_balance_dirty(root
);
2129 btrfs_free_path(path
);
2130 btrfs_free_block_rsv(root
, rsv
);
2132 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2133 &cached_state
, GFP_NOFS
);
2134 mutex_unlock(&inode
->i_mutex
);
2140 static long btrfs_fallocate(struct file
*file
, int mode
,
2141 loff_t offset
, loff_t len
)
2143 struct inode
*inode
= file_inode(file
);
2144 struct extent_state
*cached_state
= NULL
;
2145 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2152 struct extent_map
*em
;
2153 int blocksize
= BTRFS_I(inode
)->root
->sectorsize
;
2156 alloc_start
= round_down(offset
, blocksize
);
2157 alloc_end
= round_up(offset
+ len
, blocksize
);
2159 /* Make sure we aren't being give some crap mode */
2160 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2163 if (mode
& FALLOC_FL_PUNCH_HOLE
)
2164 return btrfs_punch_hole(inode
, offset
, len
);
2167 * Make sure we have enough space before we do the
2170 ret
= btrfs_check_data_free_space(inode
, alloc_end
- alloc_start
);
2173 if (root
->fs_info
->quota_enabled
) {
2174 ret
= btrfs_qgroup_reserve(root
, alloc_end
- alloc_start
);
2176 goto out_reserve_fail
;
2180 * wait for ordered IO before we have any locks. We'll loop again
2181 * below with the locks held.
2183 btrfs_wait_ordered_range(inode
, alloc_start
, alloc_end
- alloc_start
);
2185 mutex_lock(&inode
->i_mutex
);
2186 ret
= inode_newsize_ok(inode
, alloc_end
);
2190 if (alloc_start
> inode
->i_size
) {
2191 ret
= btrfs_cont_expand(inode
, i_size_read(inode
),
2197 locked_end
= alloc_end
- 1;
2199 struct btrfs_ordered_extent
*ordered
;
2201 /* the extent lock is ordered inside the running
2204 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, alloc_start
,
2205 locked_end
, 0, &cached_state
);
2206 ordered
= btrfs_lookup_first_ordered_extent(inode
,
2209 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
2210 ordered
->file_offset
< alloc_end
) {
2211 btrfs_put_ordered_extent(ordered
);
2212 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
2213 alloc_start
, locked_end
,
2214 &cached_state
, GFP_NOFS
);
2216 * we can't wait on the range with the transaction
2217 * running or with the extent lock held
2219 btrfs_wait_ordered_range(inode
, alloc_start
,
2220 alloc_end
- alloc_start
);
2223 btrfs_put_ordered_extent(ordered
);
2228 cur_offset
= alloc_start
;
2232 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2233 alloc_end
- cur_offset
, 0);
2234 if (IS_ERR_OR_NULL(em
)) {
2241 last_byte
= min(extent_map_end(em
), alloc_end
);
2242 actual_end
= min_t(u64
, extent_map_end(em
), offset
+ len
);
2243 last_byte
= ALIGN(last_byte
, blocksize
);
2245 if (em
->block_start
== EXTENT_MAP_HOLE
||
2246 (cur_offset
>= inode
->i_size
&&
2247 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
2248 ret
= btrfs_prealloc_file_range(inode
, mode
, cur_offset
,
2249 last_byte
- cur_offset
,
2250 1 << inode
->i_blkbits
,
2255 free_extent_map(em
);
2258 } else if (actual_end
> inode
->i_size
&&
2259 !(mode
& FALLOC_FL_KEEP_SIZE
)) {
2261 * We didn't need to allocate any more space, but we
2262 * still extended the size of the file so we need to
2265 inode
->i_ctime
= CURRENT_TIME
;
2266 i_size_write(inode
, actual_end
);
2267 btrfs_ordered_update_i_size(inode
, actual_end
, NULL
);
2269 free_extent_map(em
);
2271 cur_offset
= last_byte
;
2272 if (cur_offset
>= alloc_end
) {
2277 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
2278 &cached_state
, GFP_NOFS
);
2280 mutex_unlock(&inode
->i_mutex
);
2281 if (root
->fs_info
->quota_enabled
)
2282 btrfs_qgroup_free(root
, alloc_end
- alloc_start
);
2284 /* Let go of our reservation. */
2285 btrfs_free_reserved_data_space(inode
, alloc_end
- alloc_start
);
2289 static int find_desired_extent(struct inode
*inode
, loff_t
*offset
, int whence
)
2291 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2292 struct extent_map
*em
;
2293 struct extent_state
*cached_state
= NULL
;
2294 u64 lockstart
= *offset
;
2295 u64 lockend
= i_size_read(inode
);
2296 u64 start
= *offset
;
2297 u64 orig_start
= *offset
;
2298 u64 len
= i_size_read(inode
);
2302 lockend
= max_t(u64
, root
->sectorsize
, lockend
);
2303 if (lockend
<= lockstart
)
2304 lockend
= lockstart
+ root
->sectorsize
;
2307 len
= lockend
- lockstart
+ 1;
2309 len
= max_t(u64
, len
, root
->sectorsize
);
2310 if (inode
->i_size
== 0)
2313 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
, 0,
2317 * Delalloc is such a pain. If we have a hole and we have pending
2318 * delalloc for a portion of the hole we will get back a hole that
2319 * exists for the entire range since it hasn't been actually written
2320 * yet. So to take care of this case we need to look for an extent just
2321 * before the position we want in case there is outstanding delalloc
2324 if (whence
== SEEK_HOLE
&& start
!= 0) {
2325 if (start
<= root
->sectorsize
)
2326 em
= btrfs_get_extent_fiemap(inode
, NULL
, 0, 0,
2327 root
->sectorsize
, 0);
2329 em
= btrfs_get_extent_fiemap(inode
, NULL
, 0,
2330 start
- root
->sectorsize
,
2331 root
->sectorsize
, 0);
2336 last_end
= em
->start
+ em
->len
;
2337 if (em
->block_start
== EXTENT_MAP_DELALLOC
)
2338 last_end
= min_t(u64
, last_end
, inode
->i_size
);
2339 free_extent_map(em
);
2343 em
= btrfs_get_extent_fiemap(inode
, NULL
, 0, start
, len
, 0);
2349 if (em
->block_start
== EXTENT_MAP_HOLE
) {
2350 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2351 if (last_end
<= orig_start
) {
2352 free_extent_map(em
);
2358 if (whence
== SEEK_HOLE
) {
2360 free_extent_map(em
);
2364 if (whence
== SEEK_DATA
) {
2365 if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
2366 if (start
>= inode
->i_size
) {
2367 free_extent_map(em
);
2373 if (!test_bit(EXTENT_FLAG_PREALLOC
,
2376 free_extent_map(em
);
2382 start
= em
->start
+ em
->len
;
2383 last_end
= em
->start
+ em
->len
;
2385 if (em
->block_start
== EXTENT_MAP_DELALLOC
)
2386 last_end
= min_t(u64
, last_end
, inode
->i_size
);
2388 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2389 free_extent_map(em
);
2393 free_extent_map(em
);
2397 *offset
= min(*offset
, inode
->i_size
);
2399 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2400 &cached_state
, GFP_NOFS
);
2404 static loff_t
btrfs_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2406 struct inode
*inode
= file
->f_mapping
->host
;
2409 mutex_lock(&inode
->i_mutex
);
2413 offset
= generic_file_llseek(file
, offset
, whence
);
2417 if (offset
>= i_size_read(inode
)) {
2418 mutex_unlock(&inode
->i_mutex
);
2422 ret
= find_desired_extent(inode
, &offset
, whence
);
2424 mutex_unlock(&inode
->i_mutex
);
2429 if (offset
< 0 && !(file
->f_mode
& FMODE_UNSIGNED_OFFSET
)) {
2433 if (offset
> inode
->i_sb
->s_maxbytes
) {
2438 /* Special lock needed here? */
2439 if (offset
!= file
->f_pos
) {
2440 file
->f_pos
= offset
;
2441 file
->f_version
= 0;
2444 mutex_unlock(&inode
->i_mutex
);
2448 const struct file_operations btrfs_file_operations
= {
2449 .llseek
= btrfs_file_llseek
,
2450 .read
= do_sync_read
,
2451 .write
= do_sync_write
,
2452 .aio_read
= generic_file_aio_read
,
2453 .splice_read
= generic_file_splice_read
,
2454 .aio_write
= btrfs_file_aio_write
,
2455 .mmap
= btrfs_file_mmap
,
2456 .open
= generic_file_open
,
2457 .release
= btrfs_release_file
,
2458 .fsync
= btrfs_sync_file
,
2459 .fallocate
= btrfs_fallocate
,
2460 .unlocked_ioctl
= btrfs_ioctl
,
2461 #ifdef CONFIG_COMPAT
2462 .compat_ioctl
= btrfs_ioctl
,
2466 void btrfs_auto_defrag_exit(void)
2468 if (btrfs_inode_defrag_cachep
)
2469 kmem_cache_destroy(btrfs_inode_defrag_cachep
);
2472 int btrfs_auto_defrag_init(void)
2474 btrfs_inode_defrag_cachep
= kmem_cache_create("btrfs_inode_defrag",
2475 sizeof(struct inode_defrag
), 0,
2476 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
2478 if (!btrfs_inode_defrag_cachep
)