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/compat.h>
31 #include <linux/slab.h>
32 #include <linux/btrfs.h>
33 #include <linux/uio.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "print-tree.h"
43 #include "compression.h"
45 static struct kmem_cache
*btrfs_inode_defrag_cachep
;
47 * when auto defrag is enabled we
48 * queue up these defrag structs to remember which
49 * inodes need defragging passes
52 struct rb_node rb_node
;
56 * transid where the defrag was added, we search for
57 * extents newer than this
64 /* last offset we were able to defrag */
67 /* if we've wrapped around back to zero once already */
71 static int __compare_inode_defrag(struct inode_defrag
*defrag1
,
72 struct inode_defrag
*defrag2
)
74 if (defrag1
->root
> defrag2
->root
)
76 else if (defrag1
->root
< defrag2
->root
)
78 else if (defrag1
->ino
> defrag2
->ino
)
80 else if (defrag1
->ino
< defrag2
->ino
)
86 /* pop a record for an inode into the defrag tree. The lock
87 * must be held already
89 * If you're inserting a record for an older transid than an
90 * existing record, the transid already in the tree is lowered
92 * If an existing record is found the defrag item you
95 static int __btrfs_add_inode_defrag(struct btrfs_inode
*inode
,
96 struct inode_defrag
*defrag
)
98 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
99 struct inode_defrag
*entry
;
101 struct rb_node
*parent
= NULL
;
104 p
= &fs_info
->defrag_inodes
.rb_node
;
107 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
109 ret
= __compare_inode_defrag(defrag
, entry
);
111 p
= &parent
->rb_left
;
113 p
= &parent
->rb_right
;
115 /* if we're reinserting an entry for
116 * an old defrag run, make sure to
117 * lower the transid of our existing record
119 if (defrag
->transid
< entry
->transid
)
120 entry
->transid
= defrag
->transid
;
121 if (defrag
->last_offset
> entry
->last_offset
)
122 entry
->last_offset
= defrag
->last_offset
;
126 set_bit(BTRFS_INODE_IN_DEFRAG
, &inode
->runtime_flags
);
127 rb_link_node(&defrag
->rb_node
, parent
, p
);
128 rb_insert_color(&defrag
->rb_node
, &fs_info
->defrag_inodes
);
132 static inline int __need_auto_defrag(struct btrfs_fs_info
*fs_info
)
134 if (!btrfs_test_opt(fs_info
, AUTO_DEFRAG
))
137 if (btrfs_fs_closing(fs_info
))
144 * insert a defrag record for this inode if auto defrag is
147 int btrfs_add_inode_defrag(struct btrfs_trans_handle
*trans
,
148 struct btrfs_inode
*inode
)
150 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
151 struct btrfs_root
*root
= inode
->root
;
152 struct inode_defrag
*defrag
;
156 if (!__need_auto_defrag(fs_info
))
159 if (test_bit(BTRFS_INODE_IN_DEFRAG
, &inode
->runtime_flags
))
163 transid
= trans
->transid
;
165 transid
= inode
->root
->last_trans
;
167 defrag
= kmem_cache_zalloc(btrfs_inode_defrag_cachep
, GFP_NOFS
);
171 defrag
->ino
= btrfs_ino(inode
);
172 defrag
->transid
= transid
;
173 defrag
->root
= root
->root_key
.objectid
;
175 spin_lock(&fs_info
->defrag_inodes_lock
);
176 if (!test_bit(BTRFS_INODE_IN_DEFRAG
, &inode
->runtime_flags
)) {
178 * If we set IN_DEFRAG flag and evict the inode from memory,
179 * and then re-read this inode, this new inode doesn't have
180 * IN_DEFRAG flag. At the case, we may find the existed defrag.
182 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
184 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
186 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
188 spin_unlock(&fs_info
->defrag_inodes_lock
);
193 * Requeue the defrag object. If there is a defrag object that points to
194 * the same inode in the tree, we will merge them together (by
195 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
197 static void btrfs_requeue_inode_defrag(struct btrfs_inode
*inode
,
198 struct inode_defrag
*defrag
)
200 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
203 if (!__need_auto_defrag(fs_info
))
207 * Here we don't check the IN_DEFRAG flag, because we need merge
210 spin_lock(&fs_info
->defrag_inodes_lock
);
211 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
212 spin_unlock(&fs_info
->defrag_inodes_lock
);
217 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
221 * pick the defragable inode that we want, if it doesn't exist, we will get
224 static struct inode_defrag
*
225 btrfs_pick_defrag_inode(struct btrfs_fs_info
*fs_info
, u64 root
, u64 ino
)
227 struct inode_defrag
*entry
= NULL
;
228 struct inode_defrag tmp
;
230 struct rb_node
*parent
= NULL
;
236 spin_lock(&fs_info
->defrag_inodes_lock
);
237 p
= fs_info
->defrag_inodes
.rb_node
;
240 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
242 ret
= __compare_inode_defrag(&tmp
, entry
);
246 p
= parent
->rb_right
;
251 if (parent
&& __compare_inode_defrag(&tmp
, entry
) > 0) {
252 parent
= rb_next(parent
);
254 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
260 rb_erase(parent
, &fs_info
->defrag_inodes
);
261 spin_unlock(&fs_info
->defrag_inodes_lock
);
265 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info
*fs_info
)
267 struct inode_defrag
*defrag
;
268 struct rb_node
*node
;
270 spin_lock(&fs_info
->defrag_inodes_lock
);
271 node
= rb_first(&fs_info
->defrag_inodes
);
273 rb_erase(node
, &fs_info
->defrag_inodes
);
274 defrag
= rb_entry(node
, struct inode_defrag
, rb_node
);
275 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
277 cond_resched_lock(&fs_info
->defrag_inodes_lock
);
279 node
= rb_first(&fs_info
->defrag_inodes
);
281 spin_unlock(&fs_info
->defrag_inodes_lock
);
284 #define BTRFS_DEFRAG_BATCH 1024
286 static int __btrfs_run_defrag_inode(struct btrfs_fs_info
*fs_info
,
287 struct inode_defrag
*defrag
)
289 struct btrfs_root
*inode_root
;
291 struct btrfs_key key
;
292 struct btrfs_ioctl_defrag_range_args range
;
298 key
.objectid
= defrag
->root
;
299 key
.type
= BTRFS_ROOT_ITEM_KEY
;
300 key
.offset
= (u64
)-1;
302 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
304 inode_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
305 if (IS_ERR(inode_root
)) {
306 ret
= PTR_ERR(inode_root
);
310 key
.objectid
= defrag
->ino
;
311 key
.type
= BTRFS_INODE_ITEM_KEY
;
313 inode
= btrfs_iget(fs_info
->sb
, &key
, inode_root
, NULL
);
315 ret
= PTR_ERR(inode
);
318 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
320 /* do a chunk of defrag */
321 clear_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
);
322 memset(&range
, 0, sizeof(range
));
324 range
.start
= defrag
->last_offset
;
326 sb_start_write(fs_info
->sb
);
327 num_defrag
= btrfs_defrag_file(inode
, NULL
, &range
, defrag
->transid
,
329 sb_end_write(fs_info
->sb
);
331 * if we filled the whole defrag batch, there
332 * must be more work to do. Queue this defrag
335 if (num_defrag
== BTRFS_DEFRAG_BATCH
) {
336 defrag
->last_offset
= range
.start
;
337 btrfs_requeue_inode_defrag(BTRFS_I(inode
), defrag
);
338 } else if (defrag
->last_offset
&& !defrag
->cycled
) {
340 * we didn't fill our defrag batch, but
341 * we didn't start at zero. Make sure we loop
342 * around to the start of the file.
344 defrag
->last_offset
= 0;
346 btrfs_requeue_inode_defrag(BTRFS_I(inode
), defrag
);
348 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
354 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
355 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
360 * run through the list of inodes in the FS that need
363 int btrfs_run_defrag_inodes(struct btrfs_fs_info
*fs_info
)
365 struct inode_defrag
*defrag
;
367 u64 root_objectid
= 0;
369 atomic_inc(&fs_info
->defrag_running
);
371 /* Pause the auto defragger. */
372 if (test_bit(BTRFS_FS_STATE_REMOUNTING
,
376 if (!__need_auto_defrag(fs_info
))
379 /* find an inode to defrag */
380 defrag
= btrfs_pick_defrag_inode(fs_info
, root_objectid
,
383 if (root_objectid
|| first_ino
) {
392 first_ino
= defrag
->ino
+ 1;
393 root_objectid
= defrag
->root
;
395 __btrfs_run_defrag_inode(fs_info
, defrag
);
397 atomic_dec(&fs_info
->defrag_running
);
400 * during unmount, we use the transaction_wait queue to
401 * wait for the defragger to stop
403 wake_up(&fs_info
->transaction_wait
);
407 /* simple helper to fault in pages and copy. This should go away
408 * and be replaced with calls into generic code.
410 static noinline
int btrfs_copy_from_user(loff_t pos
, size_t write_bytes
,
411 struct page
**prepared_pages
,
415 size_t total_copied
= 0;
417 int offset
= pos
& (PAGE_SIZE
- 1);
419 while (write_bytes
> 0) {
420 size_t count
= min_t(size_t,
421 PAGE_SIZE
- offset
, write_bytes
);
422 struct page
*page
= prepared_pages
[pg
];
424 * Copy data from userspace to the current page
426 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, count
);
428 /* Flush processor's dcache for this page */
429 flush_dcache_page(page
);
432 * if we get a partial write, we can end up with
433 * partially up to date pages. These add
434 * a lot of complexity, so make sure they don't
435 * happen by forcing this copy to be retried.
437 * The rest of the btrfs_file_write code will fall
438 * back to page at a time copies after we return 0.
440 if (!PageUptodate(page
) && copied
< count
)
443 iov_iter_advance(i
, copied
);
444 write_bytes
-= copied
;
445 total_copied
+= copied
;
447 /* Return to btrfs_file_write_iter to fault page */
448 if (unlikely(copied
== 0))
451 if (copied
< PAGE_SIZE
- offset
) {
462 * unlocks pages after btrfs_file_write is done with them
464 static void btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
467 for (i
= 0; i
< num_pages
; i
++) {
468 /* page checked is some magic around finding pages that
469 * have been modified without going through btrfs_set_page_dirty
470 * clear it here. There should be no need to mark the pages
471 * accessed as prepare_pages should have marked them accessed
472 * in prepare_pages via find_or_create_page()
474 ClearPageChecked(pages
[i
]);
475 unlock_page(pages
[i
]);
480 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode
*inode
,
483 struct extent_state
**cached_state
)
485 u64 search_start
= start
;
486 const u64 end
= start
+ len
- 1;
488 while (search_start
< end
) {
489 const u64 search_len
= end
- search_start
+ 1;
490 struct extent_map
*em
;
494 em
= btrfs_get_extent(inode
, NULL
, 0, search_start
,
499 if (em
->block_start
!= EXTENT_MAP_HOLE
)
503 if (em
->start
< search_start
)
504 em_len
-= search_start
- em
->start
;
505 if (em_len
> search_len
)
508 ret
= set_extent_bit(&inode
->io_tree
, search_start
,
509 search_start
+ em_len
- 1,
511 NULL
, cached_state
, GFP_NOFS
);
513 search_start
= extent_map_end(em
);
522 * after copy_from_user, pages need to be dirtied and we need to make
523 * sure holes are created between the current EOF and the start of
524 * any next extents (if required).
526 * this also makes the decision about creating an inline extent vs
527 * doing real data extents, marking pages dirty and delalloc as required.
529 int btrfs_dirty_pages(struct inode
*inode
, struct page
**pages
,
530 size_t num_pages
, loff_t pos
, size_t write_bytes
,
531 struct extent_state
**cached
)
533 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
538 u64 end_of_last_block
;
539 u64 end_pos
= pos
+ write_bytes
;
540 loff_t isize
= i_size_read(inode
);
541 unsigned int extra_bits
= 0;
543 start_pos
= pos
& ~((u64
) fs_info
->sectorsize
- 1);
544 num_bytes
= round_up(write_bytes
+ pos
- start_pos
,
545 fs_info
->sectorsize
);
547 end_of_last_block
= start_pos
+ num_bytes
- 1;
550 * The pages may have already been dirty, clear out old accounting so
551 * we can set things up properly
553 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start_pos
, end_of_last_block
,
554 EXTENT_DIRTY
| EXTENT_DELALLOC
|
555 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 0, 0, cached
,
558 if (!btrfs_is_free_space_inode(BTRFS_I(inode
))) {
559 if (start_pos
>= isize
&&
560 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)) {
562 * There can't be any extents following eof in this case
563 * so just set the delalloc new bit for the range
566 extra_bits
|= EXTENT_DELALLOC_NEW
;
568 err
= btrfs_find_new_delalloc_bytes(BTRFS_I(inode
),
576 err
= btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
,
577 extra_bits
, cached
, 0);
581 for (i
= 0; i
< num_pages
; i
++) {
582 struct page
*p
= pages
[i
];
589 * we've only changed i_size in ram, and we haven't updated
590 * the disk i_size. There is no need to log the inode
594 i_size_write(inode
, end_pos
);
599 * this drops all the extents in the cache that intersect the range
600 * [start, end]. Existing extents are split as required.
602 void btrfs_drop_extent_cache(struct btrfs_inode
*inode
, u64 start
, u64 end
,
605 struct extent_map
*em
;
606 struct extent_map
*split
= NULL
;
607 struct extent_map
*split2
= NULL
;
608 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
609 u64 len
= end
- start
+ 1;
617 WARN_ON(end
< start
);
618 if (end
== (u64
)-1) {
627 split
= alloc_extent_map();
629 split2
= alloc_extent_map();
630 if (!split
|| !split2
)
633 write_lock(&em_tree
->lock
);
634 em
= lookup_extent_mapping(em_tree
, start
, len
);
636 write_unlock(&em_tree
->lock
);
640 gen
= em
->generation
;
641 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
642 if (testend
&& em
->start
+ em
->len
>= start
+ len
) {
644 write_unlock(&em_tree
->lock
);
647 start
= em
->start
+ em
->len
;
649 len
= start
+ len
- (em
->start
+ em
->len
);
651 write_unlock(&em_tree
->lock
);
654 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
655 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
656 clear_bit(EXTENT_FLAG_LOGGING
, &flags
);
657 modified
= !list_empty(&em
->list
);
661 if (em
->start
< start
) {
662 split
->start
= em
->start
;
663 split
->len
= start
- em
->start
;
665 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
) {
666 split
->orig_start
= em
->orig_start
;
667 split
->block_start
= em
->block_start
;
670 split
->block_len
= em
->block_len
;
672 split
->block_len
= split
->len
;
673 split
->orig_block_len
= max(split
->block_len
,
675 split
->ram_bytes
= em
->ram_bytes
;
677 split
->orig_start
= split
->start
;
678 split
->block_len
= 0;
679 split
->block_start
= em
->block_start
;
680 split
->orig_block_len
= 0;
681 split
->ram_bytes
= split
->len
;
684 split
->generation
= gen
;
685 split
->bdev
= em
->bdev
;
686 split
->flags
= flags
;
687 split
->compress_type
= em
->compress_type
;
688 replace_extent_mapping(em_tree
, em
, split
, modified
);
689 free_extent_map(split
);
693 if (testend
&& em
->start
+ em
->len
> start
+ len
) {
694 u64 diff
= start
+ len
- em
->start
;
696 split
->start
= start
+ len
;
697 split
->len
= em
->start
+ em
->len
- (start
+ len
);
698 split
->bdev
= em
->bdev
;
699 split
->flags
= flags
;
700 split
->compress_type
= em
->compress_type
;
701 split
->generation
= gen
;
703 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
) {
704 split
->orig_block_len
= max(em
->block_len
,
707 split
->ram_bytes
= em
->ram_bytes
;
709 split
->block_len
= em
->block_len
;
710 split
->block_start
= em
->block_start
;
711 split
->orig_start
= em
->orig_start
;
713 split
->block_len
= split
->len
;
714 split
->block_start
= em
->block_start
716 split
->orig_start
= em
->orig_start
;
719 split
->ram_bytes
= split
->len
;
720 split
->orig_start
= split
->start
;
721 split
->block_len
= 0;
722 split
->block_start
= em
->block_start
;
723 split
->orig_block_len
= 0;
726 if (extent_map_in_tree(em
)) {
727 replace_extent_mapping(em_tree
, em
, split
,
730 ret
= add_extent_mapping(em_tree
, split
,
732 ASSERT(ret
== 0); /* Logic error */
734 free_extent_map(split
);
738 if (extent_map_in_tree(em
))
739 remove_extent_mapping(em_tree
, em
);
740 write_unlock(&em_tree
->lock
);
744 /* once for the tree*/
748 free_extent_map(split
);
750 free_extent_map(split2
);
754 * this is very complex, but the basic idea is to drop all extents
755 * in the range start - end. hint_block is filled in with a block number
756 * that would be a good hint to the block allocator for this file.
758 * If an extent intersects the range but is not entirely inside the range
759 * it is either truncated or split. Anything entirely inside the range
760 * is deleted from the tree.
762 int __btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
763 struct btrfs_root
*root
, struct inode
*inode
,
764 struct btrfs_path
*path
, u64 start
, u64 end
,
765 u64
*drop_end
, int drop_cache
,
767 u32 extent_item_size
,
770 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
771 struct extent_buffer
*leaf
;
772 struct btrfs_file_extent_item
*fi
;
773 struct btrfs_key key
;
774 struct btrfs_key new_key
;
775 u64 ino
= btrfs_ino(BTRFS_I(inode
));
776 u64 search_start
= start
;
779 u64 extent_offset
= 0;
781 u64 last_end
= start
;
787 int modify_tree
= -1;
790 int leafs_visited
= 0;
793 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, end
- 1, 0);
795 if (start
>= BTRFS_I(inode
)->disk_i_size
&& !replace_extent
)
798 update_refs
= (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
799 root
== fs_info
->tree_root
);
802 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
803 search_start
, modify_tree
);
806 if (ret
> 0 && path
->slots
[0] > 0 && search_start
== start
) {
807 leaf
= path
->nodes
[0];
808 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0] - 1);
809 if (key
.objectid
== ino
&&
810 key
.type
== BTRFS_EXTENT_DATA_KEY
)
816 leaf
= path
->nodes
[0];
817 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
819 ret
= btrfs_next_leaf(root
, path
);
827 leaf
= path
->nodes
[0];
831 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
833 if (key
.objectid
> ino
)
835 if (WARN_ON_ONCE(key
.objectid
< ino
) ||
836 key
.type
< BTRFS_EXTENT_DATA_KEY
) {
841 if (key
.type
> BTRFS_EXTENT_DATA_KEY
|| key
.offset
>= end
)
844 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
845 struct btrfs_file_extent_item
);
846 extent_type
= btrfs_file_extent_type(leaf
, fi
);
848 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
849 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
850 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
851 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
852 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
853 extent_end
= key
.offset
+
854 btrfs_file_extent_num_bytes(leaf
, fi
);
855 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
856 extent_end
= key
.offset
+
857 btrfs_file_extent_inline_len(leaf
,
865 * Don't skip extent items representing 0 byte lengths. They
866 * used to be created (bug) if while punching holes we hit
867 * -ENOSPC condition. So if we find one here, just ensure we
868 * delete it, otherwise we would insert a new file extent item
869 * with the same key (offset) as that 0 bytes length file
870 * extent item in the call to setup_items_for_insert() later
873 if (extent_end
== key
.offset
&& extent_end
>= search_start
) {
874 last_end
= extent_end
;
875 goto delete_extent_item
;
878 if (extent_end
<= search_start
) {
884 search_start
= max(key
.offset
, start
);
885 if (recow
|| !modify_tree
) {
887 btrfs_release_path(path
);
892 * | - range to drop - |
893 * | -------- extent -------- |
895 if (start
> key
.offset
&& end
< extent_end
) {
897 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
902 memcpy(&new_key
, &key
, sizeof(new_key
));
903 new_key
.offset
= start
;
904 ret
= btrfs_duplicate_item(trans
, root
, path
,
906 if (ret
== -EAGAIN
) {
907 btrfs_release_path(path
);
913 leaf
= path
->nodes
[0];
914 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
915 struct btrfs_file_extent_item
);
916 btrfs_set_file_extent_num_bytes(leaf
, fi
,
919 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
920 struct btrfs_file_extent_item
);
922 extent_offset
+= start
- key
.offset
;
923 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
924 btrfs_set_file_extent_num_bytes(leaf
, fi
,
926 btrfs_mark_buffer_dirty(leaf
);
928 if (update_refs
&& disk_bytenr
> 0) {
929 ret
= btrfs_inc_extent_ref(trans
, root
,
930 disk_bytenr
, num_bytes
, 0,
931 root
->root_key
.objectid
,
933 start
- extent_offset
);
934 BUG_ON(ret
); /* -ENOMEM */
939 * From here on out we will have actually dropped something, so
940 * last_end can be updated.
942 last_end
= extent_end
;
945 * | ---- range to drop ----- |
946 * | -------- extent -------- |
948 if (start
<= key
.offset
&& end
< extent_end
) {
949 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
954 memcpy(&new_key
, &key
, sizeof(new_key
));
955 new_key
.offset
= end
;
956 btrfs_set_item_key_safe(fs_info
, path
, &new_key
);
958 extent_offset
+= end
- key
.offset
;
959 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
960 btrfs_set_file_extent_num_bytes(leaf
, fi
,
962 btrfs_mark_buffer_dirty(leaf
);
963 if (update_refs
&& disk_bytenr
> 0)
964 inode_sub_bytes(inode
, end
- key
.offset
);
968 search_start
= extent_end
;
970 * | ---- range to drop ----- |
971 * | -------- extent -------- |
973 if (start
> key
.offset
&& end
>= extent_end
) {
975 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
980 btrfs_set_file_extent_num_bytes(leaf
, fi
,
982 btrfs_mark_buffer_dirty(leaf
);
983 if (update_refs
&& disk_bytenr
> 0)
984 inode_sub_bytes(inode
, extent_end
- start
);
985 if (end
== extent_end
)
993 * | ---- range to drop ----- |
994 * | ------ extent ------ |
996 if (start
<= key
.offset
&& end
>= extent_end
) {
999 del_slot
= path
->slots
[0];
1002 BUG_ON(del_slot
+ del_nr
!= path
->slots
[0]);
1007 extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1008 inode_sub_bytes(inode
,
1009 extent_end
- key
.offset
);
1010 extent_end
= ALIGN(extent_end
,
1011 fs_info
->sectorsize
);
1012 } else if (update_refs
&& disk_bytenr
> 0) {
1013 ret
= btrfs_free_extent(trans
, root
,
1014 disk_bytenr
, num_bytes
, 0,
1015 root
->root_key
.objectid
,
1016 key
.objectid
, key
.offset
-
1018 BUG_ON(ret
); /* -ENOMEM */
1019 inode_sub_bytes(inode
,
1020 extent_end
- key
.offset
);
1023 if (end
== extent_end
)
1026 if (path
->slots
[0] + 1 < btrfs_header_nritems(leaf
)) {
1031 ret
= btrfs_del_items(trans
, root
, path
, del_slot
,
1034 btrfs_abort_transaction(trans
, ret
);
1041 btrfs_release_path(path
);
1048 if (!ret
&& del_nr
> 0) {
1050 * Set path->slots[0] to first slot, so that after the delete
1051 * if items are move off from our leaf to its immediate left or
1052 * right neighbor leafs, we end up with a correct and adjusted
1053 * path->slots[0] for our insertion (if replace_extent != 0).
1055 path
->slots
[0] = del_slot
;
1056 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
1058 btrfs_abort_transaction(trans
, ret
);
1061 leaf
= path
->nodes
[0];
1063 * If btrfs_del_items() was called, it might have deleted a leaf, in
1064 * which case it unlocked our path, so check path->locks[0] matches a
1067 if (!ret
&& replace_extent
&& leafs_visited
== 1 &&
1068 (path
->locks
[0] == BTRFS_WRITE_LOCK_BLOCKING
||
1069 path
->locks
[0] == BTRFS_WRITE_LOCK
) &&
1070 btrfs_leaf_free_space(fs_info
, leaf
) >=
1071 sizeof(struct btrfs_item
) + extent_item_size
) {
1074 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1076 if (!del_nr
&& path
->slots
[0] < btrfs_header_nritems(leaf
)) {
1077 struct btrfs_key slot_key
;
1079 btrfs_item_key_to_cpu(leaf
, &slot_key
, path
->slots
[0]);
1080 if (btrfs_comp_cpu_keys(&key
, &slot_key
) > 0)
1083 setup_items_for_insert(root
, path
, &key
,
1086 sizeof(struct btrfs_item
) +
1087 extent_item_size
, 1);
1091 if (!replace_extent
|| !(*key_inserted
))
1092 btrfs_release_path(path
);
1094 *drop_end
= found
? min(end
, last_end
) : end
;
1098 int btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
1099 struct btrfs_root
*root
, struct inode
*inode
, u64 start
,
1100 u64 end
, int drop_cache
)
1102 struct btrfs_path
*path
;
1105 path
= btrfs_alloc_path();
1108 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, start
, end
, NULL
,
1109 drop_cache
, 0, 0, NULL
);
1110 btrfs_free_path(path
);
1114 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
1115 u64 objectid
, u64 bytenr
, u64 orig_offset
,
1116 u64
*start
, u64
*end
)
1118 struct btrfs_file_extent_item
*fi
;
1119 struct btrfs_key key
;
1122 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
1125 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
1126 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
1129 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
1130 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
1131 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
1132 btrfs_file_extent_offset(leaf
, fi
) != key
.offset
- orig_offset
||
1133 btrfs_file_extent_compression(leaf
, fi
) ||
1134 btrfs_file_extent_encryption(leaf
, fi
) ||
1135 btrfs_file_extent_other_encoding(leaf
, fi
))
1138 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
1139 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
1142 *start
= key
.offset
;
1148 * Mark extent in the range start - end as written.
1150 * This changes extent type from 'pre-allocated' to 'regular'. If only
1151 * part of extent is marked as written, the extent will be split into
1154 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
1155 struct btrfs_inode
*inode
, u64 start
, u64 end
)
1157 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1158 struct btrfs_root
*root
= inode
->root
;
1159 struct extent_buffer
*leaf
;
1160 struct btrfs_path
*path
;
1161 struct btrfs_file_extent_item
*fi
;
1162 struct btrfs_key key
;
1163 struct btrfs_key new_key
;
1175 u64 ino
= btrfs_ino(inode
);
1177 path
= btrfs_alloc_path();
1184 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1187 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1190 if (ret
> 0 && path
->slots
[0] > 0)
1193 leaf
= path
->nodes
[0];
1194 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1195 if (key
.objectid
!= ino
||
1196 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
1198 btrfs_abort_transaction(trans
, ret
);
1201 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1202 struct btrfs_file_extent_item
);
1203 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_PREALLOC
) {
1205 btrfs_abort_transaction(trans
, ret
);
1208 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
1209 if (key
.offset
> start
|| extent_end
< end
) {
1211 btrfs_abort_transaction(trans
, ret
);
1215 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1216 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1217 orig_offset
= key
.offset
- btrfs_file_extent_offset(leaf
, fi
);
1218 memcpy(&new_key
, &key
, sizeof(new_key
));
1220 if (start
== key
.offset
&& end
< extent_end
) {
1223 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1224 ino
, bytenr
, orig_offset
,
1225 &other_start
, &other_end
)) {
1226 new_key
.offset
= end
;
1227 btrfs_set_item_key_safe(fs_info
, path
, &new_key
);
1228 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1229 struct btrfs_file_extent_item
);
1230 btrfs_set_file_extent_generation(leaf
, fi
,
1232 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1234 btrfs_set_file_extent_offset(leaf
, fi
,
1236 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1237 struct btrfs_file_extent_item
);
1238 btrfs_set_file_extent_generation(leaf
, fi
,
1240 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1242 btrfs_mark_buffer_dirty(leaf
);
1247 if (start
> key
.offset
&& end
== extent_end
) {
1250 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1251 ino
, bytenr
, orig_offset
,
1252 &other_start
, &other_end
)) {
1253 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1254 struct btrfs_file_extent_item
);
1255 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1256 start
- key
.offset
);
1257 btrfs_set_file_extent_generation(leaf
, fi
,
1260 new_key
.offset
= start
;
1261 btrfs_set_item_key_safe(fs_info
, path
, &new_key
);
1263 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1264 struct btrfs_file_extent_item
);
1265 btrfs_set_file_extent_generation(leaf
, fi
,
1267 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1269 btrfs_set_file_extent_offset(leaf
, fi
,
1270 start
- orig_offset
);
1271 btrfs_mark_buffer_dirty(leaf
);
1276 while (start
> key
.offset
|| end
< extent_end
) {
1277 if (key
.offset
== start
)
1280 new_key
.offset
= split
;
1281 ret
= btrfs_duplicate_item(trans
, root
, path
, &new_key
);
1282 if (ret
== -EAGAIN
) {
1283 btrfs_release_path(path
);
1287 btrfs_abort_transaction(trans
, ret
);
1291 leaf
= path
->nodes
[0];
1292 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1293 struct btrfs_file_extent_item
);
1294 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1295 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1296 split
- key
.offset
);
1298 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1299 struct btrfs_file_extent_item
);
1301 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1302 btrfs_set_file_extent_offset(leaf
, fi
, split
- orig_offset
);
1303 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1304 extent_end
- split
);
1305 btrfs_mark_buffer_dirty(leaf
);
1307 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
,
1308 0, root
->root_key
.objectid
,
1311 btrfs_abort_transaction(trans
, ret
);
1315 if (split
== start
) {
1318 if (start
!= key
.offset
) {
1320 btrfs_abort_transaction(trans
, ret
);
1331 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1332 ino
, bytenr
, orig_offset
,
1333 &other_start
, &other_end
)) {
1335 btrfs_release_path(path
);
1338 extent_end
= other_end
;
1339 del_slot
= path
->slots
[0] + 1;
1341 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1342 0, root
->root_key
.objectid
,
1345 btrfs_abort_transaction(trans
, ret
);
1351 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1352 ino
, bytenr
, orig_offset
,
1353 &other_start
, &other_end
)) {
1355 btrfs_release_path(path
);
1358 key
.offset
= other_start
;
1359 del_slot
= path
->slots
[0];
1361 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1362 0, root
->root_key
.objectid
,
1365 btrfs_abort_transaction(trans
, ret
);
1370 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1371 struct btrfs_file_extent_item
);
1372 btrfs_set_file_extent_type(leaf
, fi
,
1373 BTRFS_FILE_EXTENT_REG
);
1374 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1375 btrfs_mark_buffer_dirty(leaf
);
1377 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
1378 struct btrfs_file_extent_item
);
1379 btrfs_set_file_extent_type(leaf
, fi
,
1380 BTRFS_FILE_EXTENT_REG
);
1381 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1382 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1383 extent_end
- key
.offset
);
1384 btrfs_mark_buffer_dirty(leaf
);
1386 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
1388 btrfs_abort_transaction(trans
, ret
);
1393 btrfs_free_path(path
);
1398 * on error we return an unlocked page and the error value
1399 * on success we return a locked page and 0
1401 static int prepare_uptodate_page(struct inode
*inode
,
1402 struct page
*page
, u64 pos
,
1403 bool force_uptodate
)
1407 if (((pos
& (PAGE_SIZE
- 1)) || force_uptodate
) &&
1408 !PageUptodate(page
)) {
1409 ret
= btrfs_readpage(NULL
, page
);
1413 if (!PageUptodate(page
)) {
1417 if (page
->mapping
!= inode
->i_mapping
) {
1426 * this just gets pages into the page cache and locks them down.
1428 static noinline
int prepare_pages(struct inode
*inode
, struct page
**pages
,
1429 size_t num_pages
, loff_t pos
,
1430 size_t write_bytes
, bool force_uptodate
)
1433 unsigned long index
= pos
>> PAGE_SHIFT
;
1434 gfp_t mask
= btrfs_alloc_write_mask(inode
->i_mapping
);
1438 for (i
= 0; i
< num_pages
; i
++) {
1440 pages
[i
] = find_or_create_page(inode
->i_mapping
, index
+ i
,
1441 mask
| __GFP_WRITE
);
1449 err
= prepare_uptodate_page(inode
, pages
[i
], pos
,
1451 if (!err
&& i
== num_pages
- 1)
1452 err
= prepare_uptodate_page(inode
, pages
[i
],
1453 pos
+ write_bytes
, false);
1456 if (err
== -EAGAIN
) {
1463 wait_on_page_writeback(pages
[i
]);
1468 while (faili
>= 0) {
1469 unlock_page(pages
[faili
]);
1470 put_page(pages
[faili
]);
1478 * This function locks the extent and properly waits for data=ordered extents
1479 * to finish before allowing the pages to be modified if need.
1482 * 1 - the extent is locked
1483 * 0 - the extent is not locked, and everything is OK
1484 * -EAGAIN - need re-prepare the pages
1485 * the other < 0 number - Something wrong happens
1488 lock_and_cleanup_extent_if_need(struct btrfs_inode
*inode
, struct page
**pages
,
1489 size_t num_pages
, loff_t pos
,
1491 u64
*lockstart
, u64
*lockend
,
1492 struct extent_state
**cached_state
)
1494 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1500 start_pos
= round_down(pos
, fs_info
->sectorsize
);
1501 last_pos
= start_pos
1502 + round_up(pos
+ write_bytes
- start_pos
,
1503 fs_info
->sectorsize
) - 1;
1505 if (start_pos
< inode
->vfs_inode
.i_size
) {
1506 struct btrfs_ordered_extent
*ordered
;
1508 lock_extent_bits(&inode
->io_tree
, start_pos
, last_pos
,
1510 ordered
= btrfs_lookup_ordered_range(inode
, start_pos
,
1511 last_pos
- start_pos
+ 1);
1513 ordered
->file_offset
+ ordered
->len
> start_pos
&&
1514 ordered
->file_offset
<= last_pos
) {
1515 unlock_extent_cached(&inode
->io_tree
, start_pos
,
1516 last_pos
, cached_state
, GFP_NOFS
);
1517 for (i
= 0; i
< num_pages
; i
++) {
1518 unlock_page(pages
[i
]);
1521 btrfs_start_ordered_extent(&inode
->vfs_inode
,
1523 btrfs_put_ordered_extent(ordered
);
1527 btrfs_put_ordered_extent(ordered
);
1529 *lockstart
= start_pos
;
1530 *lockend
= last_pos
;
1535 * It's possible the pages are dirty right now, but we don't want
1536 * to clean them yet because copy_from_user may catch a page fault
1537 * and we might have to fall back to one page at a time. If that
1538 * happens, we'll unlock these pages and we'd have a window where
1539 * reclaim could sneak in and drop the once-dirty page on the floor
1540 * without writing it.
1542 * We have the pages locked and the extent range locked, so there's
1543 * no way someone can start IO on any dirty pages in this range.
1545 * We'll call btrfs_dirty_pages() later on, and that will flip around
1546 * delalloc bits and dirty the pages as required.
1548 for (i
= 0; i
< num_pages
; i
++) {
1549 set_page_extent_mapped(pages
[i
]);
1550 WARN_ON(!PageLocked(pages
[i
]));
1556 static noinline
int check_can_nocow(struct btrfs_inode
*inode
, loff_t pos
,
1557 size_t *write_bytes
)
1559 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1560 struct btrfs_root
*root
= inode
->root
;
1561 struct btrfs_ordered_extent
*ordered
;
1562 u64 lockstart
, lockend
;
1566 ret
= btrfs_start_write_no_snapshotting(root
);
1570 lockstart
= round_down(pos
, fs_info
->sectorsize
);
1571 lockend
= round_up(pos
+ *write_bytes
,
1572 fs_info
->sectorsize
) - 1;
1575 lock_extent(&inode
->io_tree
, lockstart
, lockend
);
1576 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
1577 lockend
- lockstart
+ 1);
1581 unlock_extent(&inode
->io_tree
, lockstart
, lockend
);
1582 btrfs_start_ordered_extent(&inode
->vfs_inode
, ordered
, 1);
1583 btrfs_put_ordered_extent(ordered
);
1586 num_bytes
= lockend
- lockstart
+ 1;
1587 ret
= can_nocow_extent(&inode
->vfs_inode
, lockstart
, &num_bytes
,
1591 btrfs_end_write_no_snapshotting(root
);
1593 *write_bytes
= min_t(size_t, *write_bytes
,
1594 num_bytes
- pos
+ lockstart
);
1597 unlock_extent(&inode
->io_tree
, lockstart
, lockend
);
1602 static noinline ssize_t
__btrfs_buffered_write(struct file
*file
,
1606 struct inode
*inode
= file_inode(file
);
1607 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1608 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1609 struct page
**pages
= NULL
;
1610 struct extent_state
*cached_state
= NULL
;
1611 struct extent_changeset
*data_reserved
= NULL
;
1612 u64 release_bytes
= 0;
1615 size_t num_written
= 0;
1618 bool only_release_metadata
= false;
1619 bool force_page_uptodate
= false;
1621 nrptrs
= min(DIV_ROUND_UP(iov_iter_count(i
), PAGE_SIZE
),
1622 PAGE_SIZE
/ (sizeof(struct page
*)));
1623 nrptrs
= min(nrptrs
, current
->nr_dirtied_pause
- current
->nr_dirtied
);
1624 nrptrs
= max(nrptrs
, 8);
1625 pages
= kmalloc_array(nrptrs
, sizeof(struct page
*), GFP_KERNEL
);
1629 while (iov_iter_count(i
) > 0) {
1630 size_t offset
= pos
& (PAGE_SIZE
- 1);
1631 size_t sector_offset
;
1632 size_t write_bytes
= min(iov_iter_count(i
),
1633 nrptrs
* (size_t)PAGE_SIZE
-
1635 size_t num_pages
= DIV_ROUND_UP(write_bytes
+ offset
,
1637 size_t reserve_bytes
;
1640 size_t dirty_sectors
;
1644 WARN_ON(num_pages
> nrptrs
);
1647 * Fault pages before locking them in prepare_pages
1648 * to avoid recursive lock
1650 if (unlikely(iov_iter_fault_in_readable(i
, write_bytes
))) {
1655 sector_offset
= pos
& (fs_info
->sectorsize
- 1);
1656 reserve_bytes
= round_up(write_bytes
+ sector_offset
,
1657 fs_info
->sectorsize
);
1659 extent_changeset_release(data_reserved
);
1660 ret
= btrfs_check_data_free_space(inode
, &data_reserved
, pos
,
1663 if ((BTRFS_I(inode
)->flags
& (BTRFS_INODE_NODATACOW
|
1664 BTRFS_INODE_PREALLOC
)) &&
1665 check_can_nocow(BTRFS_I(inode
), pos
,
1666 &write_bytes
) > 0) {
1668 * For nodata cow case, no need to reserve
1671 only_release_metadata
= true;
1673 * our prealloc extent may be smaller than
1674 * write_bytes, so scale down.
1676 num_pages
= DIV_ROUND_UP(write_bytes
+ offset
,
1678 reserve_bytes
= round_up(write_bytes
+
1680 fs_info
->sectorsize
);
1686 WARN_ON(reserve_bytes
== 0);
1687 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
),
1690 if (!only_release_metadata
)
1691 btrfs_free_reserved_data_space(inode
,
1695 btrfs_end_write_no_snapshotting(root
);
1699 release_bytes
= reserve_bytes
;
1702 * This is going to setup the pages array with the number of
1703 * pages we want, so we don't really need to worry about the
1704 * contents of pages from loop to loop
1706 ret
= prepare_pages(inode
, pages
, num_pages
,
1708 force_page_uptodate
);
1710 btrfs_delalloc_release_extents(BTRFS_I(inode
),
1715 extents_locked
= lock_and_cleanup_extent_if_need(
1716 BTRFS_I(inode
), pages
,
1717 num_pages
, pos
, write_bytes
, &lockstart
,
1718 &lockend
, &cached_state
);
1719 if (extents_locked
< 0) {
1720 if (extents_locked
== -EAGAIN
)
1722 btrfs_delalloc_release_extents(BTRFS_I(inode
),
1724 ret
= extents_locked
;
1728 copied
= btrfs_copy_from_user(pos
, write_bytes
, pages
, i
);
1730 num_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, reserve_bytes
);
1731 dirty_sectors
= round_up(copied
+ sector_offset
,
1732 fs_info
->sectorsize
);
1733 dirty_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, dirty_sectors
);
1736 * if we have trouble faulting in the pages, fall
1737 * back to one page at a time
1739 if (copied
< write_bytes
)
1743 force_page_uptodate
= true;
1747 force_page_uptodate
= false;
1748 dirty_pages
= DIV_ROUND_UP(copied
+ offset
,
1752 if (num_sectors
> dirty_sectors
) {
1753 /* release everything except the sectors we dirtied */
1754 release_bytes
-= dirty_sectors
<<
1755 fs_info
->sb
->s_blocksize_bits
;
1756 if (only_release_metadata
) {
1757 btrfs_delalloc_release_metadata(BTRFS_I(inode
),
1762 __pos
= round_down(pos
,
1763 fs_info
->sectorsize
) +
1764 (dirty_pages
<< PAGE_SHIFT
);
1765 btrfs_delalloc_release_space(inode
,
1766 data_reserved
, __pos
,
1771 release_bytes
= round_up(copied
+ sector_offset
,
1772 fs_info
->sectorsize
);
1775 ret
= btrfs_dirty_pages(inode
, pages
, dirty_pages
,
1778 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1779 lockstart
, lockend
, &cached_state
,
1781 btrfs_delalloc_release_extents(BTRFS_I(inode
), reserve_bytes
);
1783 btrfs_drop_pages(pages
, num_pages
);
1788 if (only_release_metadata
)
1789 btrfs_end_write_no_snapshotting(root
);
1791 if (only_release_metadata
&& copied
> 0) {
1792 lockstart
= round_down(pos
,
1793 fs_info
->sectorsize
);
1794 lockend
= round_up(pos
+ copied
,
1795 fs_info
->sectorsize
) - 1;
1797 set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
1798 lockend
, EXTENT_NORESERVE
, NULL
,
1800 only_release_metadata
= false;
1803 btrfs_drop_pages(pages
, num_pages
);
1807 balance_dirty_pages_ratelimited(inode
->i_mapping
);
1808 if (dirty_pages
< (fs_info
->nodesize
>> PAGE_SHIFT
) + 1)
1809 btrfs_btree_balance_dirty(fs_info
);
1812 num_written
+= copied
;
1817 if (release_bytes
) {
1818 if (only_release_metadata
) {
1819 btrfs_end_write_no_snapshotting(root
);
1820 btrfs_delalloc_release_metadata(BTRFS_I(inode
),
1823 btrfs_delalloc_release_space(inode
, data_reserved
,
1824 round_down(pos
, fs_info
->sectorsize
),
1829 extent_changeset_free(data_reserved
);
1830 return num_written
? num_written
: ret
;
1833 static ssize_t
__btrfs_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
1835 struct file
*file
= iocb
->ki_filp
;
1836 struct inode
*inode
= file_inode(file
);
1837 loff_t pos
= iocb
->ki_pos
;
1839 ssize_t written_buffered
;
1843 written
= generic_file_direct_write(iocb
, from
);
1845 if (written
< 0 || !iov_iter_count(from
))
1849 written_buffered
= __btrfs_buffered_write(file
, from
, pos
);
1850 if (written_buffered
< 0) {
1851 err
= written_buffered
;
1855 * Ensure all data is persisted. We want the next direct IO read to be
1856 * able to read what was just written.
1858 endbyte
= pos
+ written_buffered
- 1;
1859 err
= btrfs_fdatawrite_range(inode
, pos
, endbyte
);
1862 err
= filemap_fdatawait_range(inode
->i_mapping
, pos
, endbyte
);
1865 written
+= written_buffered
;
1866 iocb
->ki_pos
= pos
+ written_buffered
;
1867 invalidate_mapping_pages(file
->f_mapping
, pos
>> PAGE_SHIFT
,
1868 endbyte
>> PAGE_SHIFT
);
1870 return written
? written
: err
;
1873 static void update_time_for_write(struct inode
*inode
)
1875 struct timespec now
;
1877 if (IS_NOCMTIME(inode
))
1880 now
= current_time(inode
);
1881 if (!timespec_equal(&inode
->i_mtime
, &now
))
1882 inode
->i_mtime
= now
;
1884 if (!timespec_equal(&inode
->i_ctime
, &now
))
1885 inode
->i_ctime
= now
;
1887 if (IS_I_VERSION(inode
))
1888 inode_inc_iversion(inode
);
1891 static ssize_t
btrfs_file_write_iter(struct kiocb
*iocb
,
1892 struct iov_iter
*from
)
1894 struct file
*file
= iocb
->ki_filp
;
1895 struct inode
*inode
= file_inode(file
);
1896 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1897 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1900 ssize_t num_written
= 0;
1901 bool sync
= (file
->f_flags
& O_DSYNC
) || IS_SYNC(file
->f_mapping
->host
);
1904 size_t count
= iov_iter_count(from
);
1908 if (!(iocb
->ki_flags
& IOCB_DIRECT
) &&
1909 (iocb
->ki_flags
& IOCB_NOWAIT
))
1912 if (!inode_trylock(inode
)) {
1913 if (iocb
->ki_flags
& IOCB_NOWAIT
)
1918 err
= generic_write_checks(iocb
, from
);
1920 inode_unlock(inode
);
1925 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
1927 * We will allocate space in case nodatacow is not set,
1930 if (!(BTRFS_I(inode
)->flags
& (BTRFS_INODE_NODATACOW
|
1931 BTRFS_INODE_PREALLOC
)) ||
1932 check_can_nocow(BTRFS_I(inode
), pos
, &count
) <= 0) {
1933 inode_unlock(inode
);
1938 current
->backing_dev_info
= inode_to_bdi(inode
);
1939 err
= file_remove_privs(file
);
1941 inode_unlock(inode
);
1946 * If BTRFS flips readonly due to some impossible error
1947 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1948 * although we have opened a file as writable, we have
1949 * to stop this write operation to ensure FS consistency.
1951 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
)) {
1952 inode_unlock(inode
);
1958 * We reserve space for updating the inode when we reserve space for the
1959 * extent we are going to write, so we will enospc out there. We don't
1960 * need to start yet another transaction to update the inode as we will
1961 * update the inode when we finish writing whatever data we write.
1963 update_time_for_write(inode
);
1965 start_pos
= round_down(pos
, fs_info
->sectorsize
);
1966 oldsize
= i_size_read(inode
);
1967 if (start_pos
> oldsize
) {
1968 /* Expand hole size to cover write data, preventing empty gap */
1969 end_pos
= round_up(pos
+ count
,
1970 fs_info
->sectorsize
);
1971 err
= btrfs_cont_expand(inode
, oldsize
, end_pos
);
1973 inode_unlock(inode
);
1976 if (start_pos
> round_up(oldsize
, fs_info
->sectorsize
))
1981 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
1983 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1984 num_written
= __btrfs_direct_write(iocb
, from
);
1986 num_written
= __btrfs_buffered_write(file
, from
, pos
);
1987 if (num_written
> 0)
1988 iocb
->ki_pos
= pos
+ num_written
;
1990 pagecache_isize_extended(inode
, oldsize
,
1991 i_size_read(inode
));
1994 inode_unlock(inode
);
1997 * We also have to set last_sub_trans to the current log transid,
1998 * otherwise subsequent syncs to a file that's been synced in this
1999 * transaction will appear to have already occurred.
2001 spin_lock(&BTRFS_I(inode
)->lock
);
2002 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
2003 spin_unlock(&BTRFS_I(inode
)->lock
);
2004 if (num_written
> 0)
2005 num_written
= generic_write_sync(iocb
, num_written
);
2008 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
2010 current
->backing_dev_info
= NULL
;
2011 return num_written
? num_written
: err
;
2014 int btrfs_release_file(struct inode
*inode
, struct file
*filp
)
2016 struct btrfs_file_private
*private = filp
->private_data
;
2018 if (private && private->trans
)
2019 btrfs_ioctl_trans_end(filp
);
2020 if (private && private->filldir_buf
)
2021 kfree(private->filldir_buf
);
2023 filp
->private_data
= NULL
;
2026 * ordered_data_close is set by settattr when we are about to truncate
2027 * a file from a non-zero size to a zero size. This tries to
2028 * flush down new bytes that may have been written if the
2029 * application were using truncate to replace a file in place.
2031 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
2032 &BTRFS_I(inode
)->runtime_flags
))
2033 filemap_flush(inode
->i_mapping
);
2037 static int start_ordered_ops(struct inode
*inode
, loff_t start
, loff_t end
)
2040 struct blk_plug plug
;
2043 * This is only called in fsync, which would do synchronous writes, so
2044 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2045 * multiple disks using raid profile, a large IO can be split to
2046 * several segments of stripe length (currently 64K).
2048 blk_start_plug(&plug
);
2049 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
2050 ret
= btrfs_fdatawrite_range(inode
, start
, end
);
2051 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
2052 blk_finish_plug(&plug
);
2058 * fsync call for both files and directories. This logs the inode into
2059 * the tree log instead of forcing full commits whenever possible.
2061 * It needs to call filemap_fdatawait so that all ordered extent updates are
2062 * in the metadata btree are up to date for copying to the log.
2064 * It drops the inode mutex before doing the tree log commit. This is an
2065 * important optimization for directories because holding the mutex prevents
2066 * new operations on the dir while we write to disk.
2068 int btrfs_sync_file(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
2070 struct dentry
*dentry
= file_dentry(file
);
2071 struct inode
*inode
= d_inode(dentry
);
2072 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2073 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2074 struct btrfs_trans_handle
*trans
;
2075 struct btrfs_log_ctx ctx
;
2077 bool full_sync
= false;
2081 * The range length can be represented by u64, we have to do the typecasts
2082 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
2084 len
= (u64
)end
- (u64
)start
+ 1;
2085 trace_btrfs_sync_file(file
, datasync
);
2087 btrfs_init_log_ctx(&ctx
, inode
);
2090 * We write the dirty pages in the range and wait until they complete
2091 * out of the ->i_mutex. If so, we can flush the dirty pages by
2092 * multi-task, and make the performance up. See
2093 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2095 ret
= start_ordered_ops(inode
, start
, end
);
2102 * We take the dio_sem here because the tree log stuff can race with
2103 * lockless dio writes and get an extent map logged for an extent we
2104 * never waited on. We need it this high up for lockdep reasons.
2106 down_write(&BTRFS_I(inode
)->dio_sem
);
2108 atomic_inc(&root
->log_batch
);
2109 full_sync
= test_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
2110 &BTRFS_I(inode
)->runtime_flags
);
2112 * We might have have had more pages made dirty after calling
2113 * start_ordered_ops and before acquiring the inode's i_mutex.
2117 * For a full sync, we need to make sure any ordered operations
2118 * start and finish before we start logging the inode, so that
2119 * all extents are persisted and the respective file extent
2120 * items are in the fs/subvol btree.
2122 ret
= btrfs_wait_ordered_range(inode
, start
, len
);
2125 * Start any new ordered operations before starting to log the
2126 * inode. We will wait for them to finish in btrfs_sync_log().
2128 * Right before acquiring the inode's mutex, we might have new
2129 * writes dirtying pages, which won't immediately start the
2130 * respective ordered operations - that is done through the
2131 * fill_delalloc callbacks invoked from the writepage and
2132 * writepages address space operations. So make sure we start
2133 * all ordered operations before starting to log our inode. Not
2134 * doing this means that while logging the inode, writeback
2135 * could start and invoke writepage/writepages, which would call
2136 * the fill_delalloc callbacks (cow_file_range,
2137 * submit_compressed_extents). These callbacks add first an
2138 * extent map to the modified list of extents and then create
2139 * the respective ordered operation, which means in
2140 * tree-log.c:btrfs_log_inode() we might capture all existing
2141 * ordered operations (with btrfs_get_logged_extents()) before
2142 * the fill_delalloc callback adds its ordered operation, and by
2143 * the time we visit the modified list of extent maps (with
2144 * btrfs_log_changed_extents()), we see and process the extent
2145 * map they created. We then use the extent map to construct a
2146 * file extent item for logging without waiting for the
2147 * respective ordered operation to finish - this file extent
2148 * item points to a disk location that might not have yet been
2149 * written to, containing random data - so after a crash a log
2150 * replay will make our inode have file extent items that point
2151 * to disk locations containing invalid data, as we returned
2152 * success to userspace without waiting for the respective
2153 * ordered operation to finish, because it wasn't captured by
2154 * btrfs_get_logged_extents().
2156 ret
= start_ordered_ops(inode
, start
, end
);
2159 up_write(&BTRFS_I(inode
)->dio_sem
);
2160 inode_unlock(inode
);
2163 atomic_inc(&root
->log_batch
);
2166 * If the last transaction that changed this file was before the current
2167 * transaction and we have the full sync flag set in our inode, we can
2168 * bail out now without any syncing.
2170 * Note that we can't bail out if the full sync flag isn't set. This is
2171 * because when the full sync flag is set we start all ordered extents
2172 * and wait for them to fully complete - when they complete they update
2173 * the inode's last_trans field through:
2175 * btrfs_finish_ordered_io() ->
2176 * btrfs_update_inode_fallback() ->
2177 * btrfs_update_inode() ->
2178 * btrfs_set_inode_last_trans()
2180 * So we are sure that last_trans is up to date and can do this check to
2181 * bail out safely. For the fast path, when the full sync flag is not
2182 * set in our inode, we can not do it because we start only our ordered
2183 * extents and don't wait for them to complete (that is when
2184 * btrfs_finish_ordered_io runs), so here at this point their last_trans
2185 * value might be less than or equals to fs_info->last_trans_committed,
2186 * and setting a speculative last_trans for an inode when a buffered
2187 * write is made (such as fs_info->generation + 1 for example) would not
2188 * be reliable since after setting the value and before fsync is called
2189 * any number of transactions can start and commit (transaction kthread
2190 * commits the current transaction periodically), and a transaction
2191 * commit does not start nor waits for ordered extents to complete.
2194 if (btrfs_inode_in_log(BTRFS_I(inode
), fs_info
->generation
) ||
2195 (full_sync
&& BTRFS_I(inode
)->last_trans
<=
2196 fs_info
->last_trans_committed
) ||
2197 (!btrfs_have_ordered_extents_in_range(inode
, start
, len
) &&
2198 BTRFS_I(inode
)->last_trans
2199 <= fs_info
->last_trans_committed
)) {
2201 * We've had everything committed since the last time we were
2202 * modified so clear this flag in case it was set for whatever
2203 * reason, it's no longer relevant.
2205 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
2206 &BTRFS_I(inode
)->runtime_flags
);
2208 * An ordered extent might have started before and completed
2209 * already with io errors, in which case the inode was not
2210 * updated and we end up here. So check the inode's mapping
2211 * for any errors that might have happened since we last
2212 * checked called fsync.
2214 ret
= filemap_check_wb_err(inode
->i_mapping
, file
->f_wb_err
);
2215 up_write(&BTRFS_I(inode
)->dio_sem
);
2216 inode_unlock(inode
);
2221 * ok we haven't committed the transaction yet, lets do a commit
2223 if (file
->private_data
)
2224 btrfs_ioctl_trans_end(file
);
2227 * We use start here because we will need to wait on the IO to complete
2228 * in btrfs_sync_log, which could require joining a transaction (for
2229 * example checking cross references in the nocow path). If we use join
2230 * here we could get into a situation where we're waiting on IO to
2231 * happen that is blocked on a transaction trying to commit. With start
2232 * we inc the extwriter counter, so we wait for all extwriters to exit
2233 * before we start blocking join'ers. This comment is to keep somebody
2234 * from thinking they are super smart and changing this to
2235 * btrfs_join_transaction *cough*Josef*cough*.
2237 trans
= btrfs_start_transaction(root
, 0);
2238 if (IS_ERR(trans
)) {
2239 ret
= PTR_ERR(trans
);
2240 up_write(&BTRFS_I(inode
)->dio_sem
);
2241 inode_unlock(inode
);
2246 ret
= btrfs_log_dentry_safe(trans
, root
, dentry
, start
, end
, &ctx
);
2248 /* Fallthrough and commit/free transaction. */
2252 /* we've logged all the items and now have a consistent
2253 * version of the file in the log. It is possible that
2254 * someone will come in and modify the file, but that's
2255 * fine because the log is consistent on disk, and we
2256 * have references to all of the file's extents
2258 * It is possible that someone will come in and log the
2259 * file again, but that will end up using the synchronization
2260 * inside btrfs_sync_log to keep things safe.
2262 up_write(&BTRFS_I(inode
)->dio_sem
);
2263 inode_unlock(inode
);
2266 * If any of the ordered extents had an error, just return it to user
2267 * space, so that the application knows some writes didn't succeed and
2268 * can take proper action (retry for e.g.). Blindly committing the
2269 * transaction in this case, would fool userspace that everything was
2270 * successful. And we also want to make sure our log doesn't contain
2271 * file extent items pointing to extents that weren't fully written to -
2272 * just like in the non fast fsync path, where we check for the ordered
2273 * operation's error flag before writing to the log tree and return -EIO
2274 * if any of them had this flag set (btrfs_wait_ordered_range) -
2275 * therefore we need to check for errors in the ordered operations,
2276 * which are indicated by ctx.io_err.
2279 btrfs_end_transaction(trans
);
2284 if (ret
!= BTRFS_NO_LOG_SYNC
) {
2286 ret
= btrfs_sync_log(trans
, root
, &ctx
);
2288 ret
= btrfs_end_transaction(trans
);
2293 ret
= btrfs_wait_ordered_range(inode
, start
, len
);
2295 btrfs_end_transaction(trans
);
2299 ret
= btrfs_commit_transaction(trans
);
2301 ret
= btrfs_end_transaction(trans
);
2304 ASSERT(list_empty(&ctx
.list
));
2305 err
= file_check_and_advance_wb_err(file
);
2308 return ret
> 0 ? -EIO
: ret
;
2311 static const struct vm_operations_struct btrfs_file_vm_ops
= {
2312 .fault
= filemap_fault
,
2313 .map_pages
= filemap_map_pages
,
2314 .page_mkwrite
= btrfs_page_mkwrite
,
2317 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
2319 struct address_space
*mapping
= filp
->f_mapping
;
2321 if (!mapping
->a_ops
->readpage
)
2324 file_accessed(filp
);
2325 vma
->vm_ops
= &btrfs_file_vm_ops
;
2330 static int hole_mergeable(struct btrfs_inode
*inode
, struct extent_buffer
*leaf
,
2331 int slot
, u64 start
, u64 end
)
2333 struct btrfs_file_extent_item
*fi
;
2334 struct btrfs_key key
;
2336 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
2339 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2340 if (key
.objectid
!= btrfs_ino(inode
) ||
2341 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2344 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
2346 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2349 if (btrfs_file_extent_disk_bytenr(leaf
, fi
))
2352 if (key
.offset
== end
)
2354 if (key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
) == start
)
2359 static int fill_holes(struct btrfs_trans_handle
*trans
,
2360 struct btrfs_inode
*inode
,
2361 struct btrfs_path
*path
, u64 offset
, u64 end
)
2363 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
2364 struct btrfs_root
*root
= inode
->root
;
2365 struct extent_buffer
*leaf
;
2366 struct btrfs_file_extent_item
*fi
;
2367 struct extent_map
*hole_em
;
2368 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
2369 struct btrfs_key key
;
2372 if (btrfs_fs_incompat(fs_info
, NO_HOLES
))
2375 key
.objectid
= btrfs_ino(inode
);
2376 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2377 key
.offset
= offset
;
2379 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2382 * We should have dropped this offset, so if we find it then
2383 * something has gone horribly wrong.
2390 leaf
= path
->nodes
[0];
2391 if (hole_mergeable(inode
, leaf
, path
->slots
[0] - 1, offset
, end
)) {
2395 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2396 struct btrfs_file_extent_item
);
2397 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) +
2399 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2400 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
2401 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2402 btrfs_mark_buffer_dirty(leaf
);
2406 if (hole_mergeable(inode
, leaf
, path
->slots
[0], offset
, end
)) {
2409 key
.offset
= offset
;
2410 btrfs_set_item_key_safe(fs_info
, path
, &key
);
2411 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2412 struct btrfs_file_extent_item
);
2413 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) + end
-
2415 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2416 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
2417 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2418 btrfs_mark_buffer_dirty(leaf
);
2421 btrfs_release_path(path
);
2423 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
),
2424 offset
, 0, 0, end
- offset
, 0, end
- offset
, 0, 0, 0);
2429 btrfs_release_path(path
);
2431 hole_em
= alloc_extent_map();
2433 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
2434 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &inode
->runtime_flags
);
2436 hole_em
->start
= offset
;
2437 hole_em
->len
= end
- offset
;
2438 hole_em
->ram_bytes
= hole_em
->len
;
2439 hole_em
->orig_start
= offset
;
2441 hole_em
->block_start
= EXTENT_MAP_HOLE
;
2442 hole_em
->block_len
= 0;
2443 hole_em
->orig_block_len
= 0;
2444 hole_em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
2445 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
2446 hole_em
->generation
= trans
->transid
;
2449 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
2450 write_lock(&em_tree
->lock
);
2451 ret
= add_extent_mapping(em_tree
, hole_em
, 1);
2452 write_unlock(&em_tree
->lock
);
2453 } while (ret
== -EEXIST
);
2454 free_extent_map(hole_em
);
2456 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
2457 &inode
->runtime_flags
);
2464 * Find a hole extent on given inode and change start/len to the end of hole
2465 * extent.(hole/vacuum extent whose em->start <= start &&
2466 * em->start + em->len > start)
2467 * When a hole extent is found, return 1 and modify start/len.
2469 static int find_first_non_hole(struct inode
*inode
, u64
*start
, u64
*len
)
2471 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2472 struct extent_map
*em
;
2475 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0,
2476 round_down(*start
, fs_info
->sectorsize
),
2477 round_up(*len
, fs_info
->sectorsize
), 0);
2481 /* Hole or vacuum extent(only exists in no-hole mode) */
2482 if (em
->block_start
== EXTENT_MAP_HOLE
) {
2484 *len
= em
->start
+ em
->len
> *start
+ *len
?
2485 0 : *start
+ *len
- em
->start
- em
->len
;
2486 *start
= em
->start
+ em
->len
;
2488 free_extent_map(em
);
2492 static int btrfs_punch_hole(struct inode
*inode
, loff_t offset
, loff_t len
)
2494 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2495 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2496 struct extent_state
*cached_state
= NULL
;
2497 struct btrfs_path
*path
;
2498 struct btrfs_block_rsv
*rsv
;
2499 struct btrfs_trans_handle
*trans
;
2504 u64 orig_start
= offset
;
2506 u64 min_size
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2510 unsigned int rsv_count
;
2512 bool no_holes
= btrfs_fs_incompat(fs_info
, NO_HOLES
);
2514 bool truncated_block
= false;
2515 bool updated_inode
= false;
2517 ret
= btrfs_wait_ordered_range(inode
, offset
, len
);
2522 ino_size
= round_up(inode
->i_size
, fs_info
->sectorsize
);
2523 ret
= find_first_non_hole(inode
, &offset
, &len
);
2525 goto out_only_mutex
;
2527 /* Already in a large hole */
2529 goto out_only_mutex
;
2532 lockstart
= round_up(offset
, btrfs_inode_sectorsize(inode
));
2533 lockend
= round_down(offset
+ len
,
2534 btrfs_inode_sectorsize(inode
)) - 1;
2535 same_block
= (BTRFS_BYTES_TO_BLKS(fs_info
, offset
))
2536 == (BTRFS_BYTES_TO_BLKS(fs_info
, offset
+ len
- 1));
2538 * We needn't truncate any block which is beyond the end of the file
2539 * because we are sure there is no data there.
2542 * Only do this if we are in the same block and we aren't doing the
2545 if (same_block
&& len
< fs_info
->sectorsize
) {
2546 if (offset
< ino_size
) {
2547 truncated_block
= true;
2548 ret
= btrfs_truncate_block(inode
, offset
, len
, 0);
2552 goto out_only_mutex
;
2555 /* zero back part of the first block */
2556 if (offset
< ino_size
) {
2557 truncated_block
= true;
2558 ret
= btrfs_truncate_block(inode
, offset
, 0, 0);
2560 inode_unlock(inode
);
2565 /* Check the aligned pages after the first unaligned page,
2566 * if offset != orig_start, which means the first unaligned page
2567 * including several following pages are already in holes,
2568 * the extra check can be skipped */
2569 if (offset
== orig_start
) {
2570 /* after truncate page, check hole again */
2571 len
= offset
+ len
- lockstart
;
2573 ret
= find_first_non_hole(inode
, &offset
, &len
);
2575 goto out_only_mutex
;
2578 goto out_only_mutex
;
2583 /* Check the tail unaligned part is in a hole */
2584 tail_start
= lockend
+ 1;
2585 tail_len
= offset
+ len
- tail_start
;
2587 ret
= find_first_non_hole(inode
, &tail_start
, &tail_len
);
2588 if (unlikely(ret
< 0))
2589 goto out_only_mutex
;
2591 /* zero the front end of the last page */
2592 if (tail_start
+ tail_len
< ino_size
) {
2593 truncated_block
= true;
2594 ret
= btrfs_truncate_block(inode
,
2595 tail_start
+ tail_len
,
2598 goto out_only_mutex
;
2603 if (lockend
< lockstart
) {
2605 goto out_only_mutex
;
2609 struct btrfs_ordered_extent
*ordered
;
2611 truncate_pagecache_range(inode
, lockstart
, lockend
);
2613 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2615 ordered
= btrfs_lookup_first_ordered_extent(inode
, lockend
);
2618 * We need to make sure we have no ordered extents in this range
2619 * and nobody raced in and read a page in this range, if we did
2620 * we need to try again.
2623 (ordered
->file_offset
+ ordered
->len
<= lockstart
||
2624 ordered
->file_offset
> lockend
)) &&
2625 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)) {
2627 btrfs_put_ordered_extent(ordered
);
2631 btrfs_put_ordered_extent(ordered
);
2632 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
,
2633 lockend
, &cached_state
, GFP_NOFS
);
2634 ret
= btrfs_wait_ordered_range(inode
, lockstart
,
2635 lockend
- lockstart
+ 1);
2637 inode_unlock(inode
);
2642 path
= btrfs_alloc_path();
2648 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
2653 rsv
->size
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2657 * 1 - update the inode
2658 * 1 - removing the extents in the range
2659 * 1 - adding the hole extent if no_holes isn't set
2661 rsv_count
= no_holes
? 2 : 3;
2662 trans
= btrfs_start_transaction(root
, rsv_count
);
2663 if (IS_ERR(trans
)) {
2664 err
= PTR_ERR(trans
);
2668 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
, rsv
,
2671 trans
->block_rsv
= rsv
;
2673 cur_offset
= lockstart
;
2674 len
= lockend
- cur_offset
;
2675 while (cur_offset
< lockend
) {
2676 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
2677 cur_offset
, lockend
+ 1,
2678 &drop_end
, 1, 0, 0, NULL
);
2682 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
2684 if (cur_offset
< drop_end
&& cur_offset
< ino_size
) {
2685 ret
= fill_holes(trans
, BTRFS_I(inode
), path
,
2686 cur_offset
, drop_end
);
2689 * If we failed then we didn't insert our hole
2690 * entries for the area we dropped, so now the
2691 * fs is corrupted, so we must abort the
2694 btrfs_abort_transaction(trans
, ret
);
2700 cur_offset
= drop_end
;
2702 ret
= btrfs_update_inode(trans
, root
, inode
);
2708 btrfs_end_transaction(trans
);
2709 btrfs_btree_balance_dirty(fs_info
);
2711 trans
= btrfs_start_transaction(root
, rsv_count
);
2712 if (IS_ERR(trans
)) {
2713 ret
= PTR_ERR(trans
);
2718 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
,
2720 BUG_ON(ret
); /* shouldn't happen */
2721 trans
->block_rsv
= rsv
;
2723 ret
= find_first_non_hole(inode
, &cur_offset
, &len
);
2724 if (unlikely(ret
< 0))
2737 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
2739 * If we are using the NO_HOLES feature we might have had already an
2740 * hole that overlaps a part of the region [lockstart, lockend] and
2741 * ends at (or beyond) lockend. Since we have no file extent items to
2742 * represent holes, drop_end can be less than lockend and so we must
2743 * make sure we have an extent map representing the existing hole (the
2744 * call to __btrfs_drop_extents() might have dropped the existing extent
2745 * map representing the existing hole), otherwise the fast fsync path
2746 * will not record the existence of the hole region
2747 * [existing_hole_start, lockend].
2749 if (drop_end
<= lockend
)
2750 drop_end
= lockend
+ 1;
2752 * Don't insert file hole extent item if it's for a range beyond eof
2753 * (because it's useless) or if it represents a 0 bytes range (when
2754 * cur_offset == drop_end).
2756 if (cur_offset
< ino_size
&& cur_offset
< drop_end
) {
2757 ret
= fill_holes(trans
, BTRFS_I(inode
), path
,
2758 cur_offset
, drop_end
);
2760 /* Same comment as above. */
2761 btrfs_abort_transaction(trans
, ret
);
2771 inode_inc_iversion(inode
);
2772 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
2774 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
2775 ret
= btrfs_update_inode(trans
, root
, inode
);
2776 updated_inode
= true;
2777 btrfs_end_transaction(trans
);
2778 btrfs_btree_balance_dirty(fs_info
);
2780 btrfs_free_path(path
);
2781 btrfs_free_block_rsv(fs_info
, rsv
);
2783 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2784 &cached_state
, GFP_NOFS
);
2786 if (!updated_inode
&& truncated_block
&& !ret
&& !err
) {
2788 * If we only end up zeroing part of a page, we still need to
2789 * update the inode item, so that all the time fields are
2790 * updated as well as the necessary btrfs inode in memory fields
2791 * for detecting, at fsync time, if the inode isn't yet in the
2792 * log tree or it's there but not up to date.
2794 trans
= btrfs_start_transaction(root
, 1);
2795 if (IS_ERR(trans
)) {
2796 err
= PTR_ERR(trans
);
2798 err
= btrfs_update_inode(trans
, root
, inode
);
2799 ret
= btrfs_end_transaction(trans
);
2802 inode_unlock(inode
);
2808 /* Helper structure to record which range is already reserved */
2809 struct falloc_range
{
2810 struct list_head list
;
2816 * Helper function to add falloc range
2818 * Caller should have locked the larger range of extent containing
2821 static int add_falloc_range(struct list_head
*head
, u64 start
, u64 len
)
2823 struct falloc_range
*prev
= NULL
;
2824 struct falloc_range
*range
= NULL
;
2826 if (list_empty(head
))
2830 * As fallocate iterate by bytenr order, we only need to check
2833 prev
= list_entry(head
->prev
, struct falloc_range
, list
);
2834 if (prev
->start
+ prev
->len
== start
) {
2839 range
= kmalloc(sizeof(*range
), GFP_KERNEL
);
2842 range
->start
= start
;
2844 list_add_tail(&range
->list
, head
);
2848 static long btrfs_fallocate(struct file
*file
, int mode
,
2849 loff_t offset
, loff_t len
)
2851 struct inode
*inode
= file_inode(file
);
2852 struct extent_state
*cached_state
= NULL
;
2853 struct extent_changeset
*data_reserved
= NULL
;
2854 struct falloc_range
*range
;
2855 struct falloc_range
*tmp
;
2856 struct list_head reserve_list
;
2864 struct extent_map
*em
;
2865 int blocksize
= btrfs_inode_sectorsize(inode
);
2868 alloc_start
= round_down(offset
, blocksize
);
2869 alloc_end
= round_up(offset
+ len
, blocksize
);
2870 cur_offset
= alloc_start
;
2872 /* Make sure we aren't being give some crap mode */
2873 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2876 if (mode
& FALLOC_FL_PUNCH_HOLE
)
2877 return btrfs_punch_hole(inode
, offset
, len
);
2880 * Only trigger disk allocation, don't trigger qgroup reserve
2882 * For qgroup space, it will be checked later.
2884 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
),
2885 alloc_end
- alloc_start
);
2891 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && offset
+ len
> inode
->i_size
) {
2892 ret
= inode_newsize_ok(inode
, offset
+ len
);
2898 * TODO: Move these two operations after we have checked
2899 * accurate reserved space, or fallocate can still fail but
2900 * with page truncated or size expanded.
2902 * But that's a minor problem and won't do much harm BTW.
2904 if (alloc_start
> inode
->i_size
) {
2905 ret
= btrfs_cont_expand(inode
, i_size_read(inode
),
2909 } else if (offset
+ len
> inode
->i_size
) {
2911 * If we are fallocating from the end of the file onward we
2912 * need to zero out the end of the block if i_size lands in the
2913 * middle of a block.
2915 ret
= btrfs_truncate_block(inode
, inode
->i_size
, 0, 0);
2921 * wait for ordered IO before we have any locks. We'll loop again
2922 * below with the locks held.
2924 ret
= btrfs_wait_ordered_range(inode
, alloc_start
,
2925 alloc_end
- alloc_start
);
2929 locked_end
= alloc_end
- 1;
2931 struct btrfs_ordered_extent
*ordered
;
2933 /* the extent lock is ordered inside the running
2936 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, alloc_start
,
2937 locked_end
, &cached_state
);
2938 ordered
= btrfs_lookup_first_ordered_extent(inode
,
2941 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
2942 ordered
->file_offset
< alloc_end
) {
2943 btrfs_put_ordered_extent(ordered
);
2944 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
2945 alloc_start
, locked_end
,
2946 &cached_state
, GFP_KERNEL
);
2948 * we can't wait on the range with the transaction
2949 * running or with the extent lock held
2951 ret
= btrfs_wait_ordered_range(inode
, alloc_start
,
2952 alloc_end
- alloc_start
);
2957 btrfs_put_ordered_extent(ordered
);
2962 /* First, check if we exceed the qgroup limit */
2963 INIT_LIST_HEAD(&reserve_list
);
2965 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, cur_offset
,
2966 alloc_end
- cur_offset
, 0);
2971 last_byte
= min(extent_map_end(em
), alloc_end
);
2972 actual_end
= min_t(u64
, extent_map_end(em
), offset
+ len
);
2973 last_byte
= ALIGN(last_byte
, blocksize
);
2974 if (em
->block_start
== EXTENT_MAP_HOLE
||
2975 (cur_offset
>= inode
->i_size
&&
2976 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
2977 ret
= add_falloc_range(&reserve_list
, cur_offset
,
2978 last_byte
- cur_offset
);
2980 free_extent_map(em
);
2983 ret
= btrfs_qgroup_reserve_data(inode
, &data_reserved
,
2984 cur_offset
, last_byte
- cur_offset
);
2986 free_extent_map(em
);
2991 * Do not need to reserve unwritten extent for this
2992 * range, free reserved data space first, otherwise
2993 * it'll result in false ENOSPC error.
2995 btrfs_free_reserved_data_space(inode
, data_reserved
,
2996 cur_offset
, last_byte
- cur_offset
);
2998 free_extent_map(em
);
2999 cur_offset
= last_byte
;
3000 if (cur_offset
>= alloc_end
)
3005 * If ret is still 0, means we're OK to fallocate.
3006 * Or just cleanup the list and exit.
3008 list_for_each_entry_safe(range
, tmp
, &reserve_list
, list
) {
3010 ret
= btrfs_prealloc_file_range(inode
, mode
,
3012 range
->len
, i_blocksize(inode
),
3013 offset
+ len
, &alloc_hint
);
3015 btrfs_free_reserved_data_space(inode
,
3016 data_reserved
, range
->start
,
3018 list_del(&range
->list
);
3024 if (actual_end
> inode
->i_size
&&
3025 !(mode
& FALLOC_FL_KEEP_SIZE
)) {
3026 struct btrfs_trans_handle
*trans
;
3027 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3030 * We didn't need to allocate any more space, but we
3031 * still extended the size of the file so we need to
3032 * update i_size and the inode item.
3034 trans
= btrfs_start_transaction(root
, 1);
3035 if (IS_ERR(trans
)) {
3036 ret
= PTR_ERR(trans
);
3038 inode
->i_ctime
= current_time(inode
);
3039 i_size_write(inode
, actual_end
);
3040 btrfs_ordered_update_i_size(inode
, actual_end
, NULL
);
3041 ret
= btrfs_update_inode(trans
, root
, inode
);
3043 btrfs_end_transaction(trans
);
3045 ret
= btrfs_end_transaction(trans
);
3049 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
3050 &cached_state
, GFP_KERNEL
);
3052 inode_unlock(inode
);
3053 /* Let go of our reservation. */
3055 btrfs_free_reserved_data_space(inode
, data_reserved
,
3056 alloc_start
, alloc_end
- cur_offset
);
3057 extent_changeset_free(data_reserved
);
3061 static int find_desired_extent(struct inode
*inode
, loff_t
*offset
, int whence
)
3063 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3064 struct extent_map
*em
= NULL
;
3065 struct extent_state
*cached_state
= NULL
;
3072 if (inode
->i_size
== 0)
3076 * *offset can be negative, in this case we start finding DATA/HOLE from
3077 * the very start of the file.
3079 start
= max_t(loff_t
, 0, *offset
);
3081 lockstart
= round_down(start
, fs_info
->sectorsize
);
3082 lockend
= round_up(i_size_read(inode
),
3083 fs_info
->sectorsize
);
3084 if (lockend
<= lockstart
)
3085 lockend
= lockstart
+ fs_info
->sectorsize
;
3087 len
= lockend
- lockstart
+ 1;
3089 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
3092 while (start
< inode
->i_size
) {
3093 em
= btrfs_get_extent_fiemap(BTRFS_I(inode
), NULL
, 0,
3101 if (whence
== SEEK_HOLE
&&
3102 (em
->block_start
== EXTENT_MAP_HOLE
||
3103 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)))
3105 else if (whence
== SEEK_DATA
&&
3106 (em
->block_start
!= EXTENT_MAP_HOLE
&&
3107 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)))
3110 start
= em
->start
+ em
->len
;
3111 free_extent_map(em
);
3115 free_extent_map(em
);
3117 if (whence
== SEEK_DATA
&& start
>= inode
->i_size
)
3120 *offset
= min_t(loff_t
, start
, inode
->i_size
);
3122 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
3123 &cached_state
, GFP_NOFS
);
3127 static loff_t
btrfs_file_llseek(struct file
*file
, loff_t offset
, int whence
)
3129 struct inode
*inode
= file
->f_mapping
->host
;
3136 offset
= generic_file_llseek(file
, offset
, whence
);
3140 if (offset
>= i_size_read(inode
)) {
3141 inode_unlock(inode
);
3145 ret
= find_desired_extent(inode
, &offset
, whence
);
3147 inode_unlock(inode
);
3152 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
3154 inode_unlock(inode
);
3158 static int btrfs_file_open(struct inode
*inode
, struct file
*filp
)
3160 filp
->f_mode
|= FMODE_NOWAIT
;
3161 return generic_file_open(inode
, filp
);
3164 const struct file_operations btrfs_file_operations
= {
3165 .llseek
= btrfs_file_llseek
,
3166 .read_iter
= generic_file_read_iter
,
3167 .splice_read
= generic_file_splice_read
,
3168 .write_iter
= btrfs_file_write_iter
,
3169 .mmap
= btrfs_file_mmap
,
3170 .open
= btrfs_file_open
,
3171 .release
= btrfs_release_file
,
3172 .fsync
= btrfs_sync_file
,
3173 .fallocate
= btrfs_fallocate
,
3174 .unlocked_ioctl
= btrfs_ioctl
,
3175 #ifdef CONFIG_COMPAT
3176 .compat_ioctl
= btrfs_compat_ioctl
,
3178 .clone_file_range
= btrfs_clone_file_range
,
3179 .dedupe_file_range
= btrfs_dedupe_file_range
,
3182 void btrfs_auto_defrag_exit(void)
3184 kmem_cache_destroy(btrfs_inode_defrag_cachep
);
3187 int btrfs_auto_defrag_init(void)
3189 btrfs_inode_defrag_cachep
= kmem_cache_create("btrfs_inode_defrag",
3190 sizeof(struct inode_defrag
), 0,
3193 if (!btrfs_inode_defrag_cachep
)
3199 int btrfs_fdatawrite_range(struct inode
*inode
, loff_t start
, loff_t end
)
3204 * So with compression we will find and lock a dirty page and clear the
3205 * first one as dirty, setup an async extent, and immediately return
3206 * with the entire range locked but with nobody actually marked with
3207 * writeback. So we can't just filemap_write_and_wait_range() and
3208 * expect it to work since it will just kick off a thread to do the
3209 * actual work. So we need to call filemap_fdatawrite_range _again_
3210 * since it will wait on the page lock, which won't be unlocked until
3211 * after the pages have been marked as writeback and so we're good to go
3212 * from there. We have to do this otherwise we'll miss the ordered
3213 * extents and that results in badness. Please Josef, do not think you
3214 * know better and pull this out at some point in the future, it is
3215 * right and you are wrong.
3217 ret
= filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
3218 if (!ret
&& test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
3219 &BTRFS_I(inode
)->runtime_flags
))
3220 ret
= filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);