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/aio.h>
28 #include <linux/falloc.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/statfs.h>
32 #include <linux/compat.h>
33 #include <linux/slab.h>
34 #include <linux/btrfs.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.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 inode
*inode
,
96 struct inode_defrag
*defrag
)
98 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
99 struct inode_defrag
*entry
;
101 struct rb_node
*parent
= NULL
;
104 p
= &root
->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
, &BTRFS_I(inode
)->runtime_flags
);
127 rb_link_node(&defrag
->rb_node
, parent
, p
);
128 rb_insert_color(&defrag
->rb_node
, &root
->fs_info
->defrag_inodes
);
132 static inline int __need_auto_defrag(struct btrfs_root
*root
)
134 if (!btrfs_test_opt(root
, AUTO_DEFRAG
))
137 if (btrfs_fs_closing(root
->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
,
150 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
151 struct inode_defrag
*defrag
;
155 if (!__need_auto_defrag(root
))
158 if (test_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
))
162 transid
= trans
->transid
;
164 transid
= BTRFS_I(inode
)->root
->last_trans
;
166 defrag
= kmem_cache_zalloc(btrfs_inode_defrag_cachep
, GFP_NOFS
);
170 defrag
->ino
= btrfs_ino(inode
);
171 defrag
->transid
= transid
;
172 defrag
->root
= root
->root_key
.objectid
;
174 spin_lock(&root
->fs_info
->defrag_inodes_lock
);
175 if (!test_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
)) {
177 * If we set IN_DEFRAG flag and evict the inode from memory,
178 * and then re-read this inode, this new inode doesn't have
179 * IN_DEFRAG flag. At the case, we may find the existed defrag.
181 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
183 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
185 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
187 spin_unlock(&root
->fs_info
->defrag_inodes_lock
);
192 * Requeue the defrag object. If there is a defrag object that points to
193 * the same inode in the tree, we will merge them together (by
194 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
196 static void btrfs_requeue_inode_defrag(struct inode
*inode
,
197 struct inode_defrag
*defrag
)
199 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
202 if (!__need_auto_defrag(root
))
206 * Here we don't check the IN_DEFRAG flag, because we need merge
209 spin_lock(&root
->fs_info
->defrag_inodes_lock
);
210 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
211 spin_unlock(&root
->fs_info
->defrag_inodes_lock
);
216 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
220 * pick the defragable inode that we want, if it doesn't exist, we will get
223 static struct inode_defrag
*
224 btrfs_pick_defrag_inode(struct btrfs_fs_info
*fs_info
, u64 root
, u64 ino
)
226 struct inode_defrag
*entry
= NULL
;
227 struct inode_defrag tmp
;
229 struct rb_node
*parent
= NULL
;
235 spin_lock(&fs_info
->defrag_inodes_lock
);
236 p
= fs_info
->defrag_inodes
.rb_node
;
239 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
241 ret
= __compare_inode_defrag(&tmp
, entry
);
245 p
= parent
->rb_right
;
250 if (parent
&& __compare_inode_defrag(&tmp
, entry
) > 0) {
251 parent
= rb_next(parent
);
253 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
259 rb_erase(parent
, &fs_info
->defrag_inodes
);
260 spin_unlock(&fs_info
->defrag_inodes_lock
);
264 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info
*fs_info
)
266 struct inode_defrag
*defrag
;
267 struct rb_node
*node
;
269 spin_lock(&fs_info
->defrag_inodes_lock
);
270 node
= rb_first(&fs_info
->defrag_inodes
);
272 rb_erase(node
, &fs_info
->defrag_inodes
);
273 defrag
= rb_entry(node
, struct inode_defrag
, rb_node
);
274 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
276 if (need_resched()) {
277 spin_unlock(&fs_info
->defrag_inodes_lock
);
279 spin_lock(&fs_info
->defrag_inodes_lock
);
282 node
= rb_first(&fs_info
->defrag_inodes
);
284 spin_unlock(&fs_info
->defrag_inodes_lock
);
287 #define BTRFS_DEFRAG_BATCH 1024
289 static int __btrfs_run_defrag_inode(struct btrfs_fs_info
*fs_info
,
290 struct inode_defrag
*defrag
)
292 struct btrfs_root
*inode_root
;
294 struct btrfs_key key
;
295 struct btrfs_ioctl_defrag_range_args range
;
301 key
.objectid
= defrag
->root
;
302 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
303 key
.offset
= (u64
)-1;
305 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
307 inode_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
308 if (IS_ERR(inode_root
)) {
309 ret
= PTR_ERR(inode_root
);
313 key
.objectid
= defrag
->ino
;
314 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
316 inode
= btrfs_iget(fs_info
->sb
, &key
, inode_root
, NULL
);
318 ret
= PTR_ERR(inode
);
321 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
323 /* do a chunk of defrag */
324 clear_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
);
325 memset(&range
, 0, sizeof(range
));
327 range
.start
= defrag
->last_offset
;
329 sb_start_write(fs_info
->sb
);
330 num_defrag
= btrfs_defrag_file(inode
, NULL
, &range
, defrag
->transid
,
332 sb_end_write(fs_info
->sb
);
334 * if we filled the whole defrag batch, there
335 * must be more work to do. Queue this defrag
338 if (num_defrag
== BTRFS_DEFRAG_BATCH
) {
339 defrag
->last_offset
= range
.start
;
340 btrfs_requeue_inode_defrag(inode
, defrag
);
341 } else if (defrag
->last_offset
&& !defrag
->cycled
) {
343 * we didn't fill our defrag batch, but
344 * we didn't start at zero. Make sure we loop
345 * around to the start of the file.
347 defrag
->last_offset
= 0;
349 btrfs_requeue_inode_defrag(inode
, defrag
);
351 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
357 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
358 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
363 * run through the list of inodes in the FS that need
366 int btrfs_run_defrag_inodes(struct btrfs_fs_info
*fs_info
)
368 struct inode_defrag
*defrag
;
370 u64 root_objectid
= 0;
372 atomic_inc(&fs_info
->defrag_running
);
374 /* Pause the auto defragger. */
375 if (test_bit(BTRFS_FS_STATE_REMOUNTING
,
379 if (!__need_auto_defrag(fs_info
->tree_root
))
382 /* find an inode to defrag */
383 defrag
= btrfs_pick_defrag_inode(fs_info
, root_objectid
,
386 if (root_objectid
|| first_ino
) {
395 first_ino
= defrag
->ino
+ 1;
396 root_objectid
= defrag
->root
;
398 __btrfs_run_defrag_inode(fs_info
, defrag
);
400 atomic_dec(&fs_info
->defrag_running
);
403 * during unmount, we use the transaction_wait queue to
404 * wait for the defragger to stop
406 wake_up(&fs_info
->transaction_wait
);
410 /* simple helper to fault in pages and copy. This should go away
411 * and be replaced with calls into generic code.
413 static noinline
int btrfs_copy_from_user(loff_t pos
, int num_pages
,
415 struct page
**prepared_pages
,
419 size_t total_copied
= 0;
421 int offset
= pos
& (PAGE_CACHE_SIZE
- 1);
423 while (write_bytes
> 0) {
424 size_t count
= min_t(size_t,
425 PAGE_CACHE_SIZE
- offset
, write_bytes
);
426 struct page
*page
= prepared_pages
[pg
];
428 * Copy data from userspace to the current page
430 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, count
);
432 /* Flush processor's dcache for this page */
433 flush_dcache_page(page
);
436 * if we get a partial write, we can end up with
437 * partially up to date pages. These add
438 * a lot of complexity, so make sure they don't
439 * happen by forcing this copy to be retried.
441 * The rest of the btrfs_file_write code will fall
442 * back to page at a time copies after we return 0.
444 if (!PageUptodate(page
) && copied
< count
)
447 iov_iter_advance(i
, copied
);
448 write_bytes
-= copied
;
449 total_copied
+= copied
;
451 /* Return to btrfs_file_write_iter to fault page */
452 if (unlikely(copied
== 0))
455 if (unlikely(copied
< PAGE_CACHE_SIZE
- offset
)) {
466 * unlocks pages after btrfs_file_write is done with them
468 static void btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
471 for (i
= 0; i
< num_pages
; i
++) {
472 /* page checked is some magic around finding pages that
473 * have been modified without going through btrfs_set_page_dirty
474 * clear it here. There should be no need to mark the pages
475 * accessed as prepare_pages should have marked them accessed
476 * in prepare_pages via find_or_create_page()
478 ClearPageChecked(pages
[i
]);
479 unlock_page(pages
[i
]);
480 page_cache_release(pages
[i
]);
485 * after copy_from_user, pages need to be dirtied and we need to make
486 * sure holes are created between the current EOF and the start of
487 * any next extents (if required).
489 * this also makes the decision about creating an inline extent vs
490 * doing real data extents, marking pages dirty and delalloc as required.
492 int btrfs_dirty_pages(struct btrfs_root
*root
, struct inode
*inode
,
493 struct page
**pages
, size_t num_pages
,
494 loff_t pos
, size_t write_bytes
,
495 struct extent_state
**cached
)
501 u64 end_of_last_block
;
502 u64 end_pos
= pos
+ write_bytes
;
503 loff_t isize
= i_size_read(inode
);
505 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
506 num_bytes
= ALIGN(write_bytes
+ pos
- start_pos
, root
->sectorsize
);
508 end_of_last_block
= start_pos
+ num_bytes
- 1;
509 err
= btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
,
514 for (i
= 0; i
< num_pages
; i
++) {
515 struct page
*p
= pages
[i
];
522 * we've only changed i_size in ram, and we haven't updated
523 * the disk i_size. There is no need to log the inode
527 i_size_write(inode
, end_pos
);
532 * this drops all the extents in the cache that intersect the range
533 * [start, end]. Existing extents are split as required.
535 void btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
,
538 struct extent_map
*em
;
539 struct extent_map
*split
= NULL
;
540 struct extent_map
*split2
= NULL
;
541 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
542 u64 len
= end
- start
+ 1;
550 WARN_ON(end
< start
);
551 if (end
== (u64
)-1) {
560 split
= alloc_extent_map();
562 split2
= alloc_extent_map();
563 if (!split
|| !split2
)
566 write_lock(&em_tree
->lock
);
567 em
= lookup_extent_mapping(em_tree
, start
, len
);
569 write_unlock(&em_tree
->lock
);
573 gen
= em
->generation
;
574 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
575 if (testend
&& em
->start
+ em
->len
>= start
+ len
) {
577 write_unlock(&em_tree
->lock
);
580 start
= em
->start
+ em
->len
;
582 len
= start
+ len
- (em
->start
+ em
->len
);
584 write_unlock(&em_tree
->lock
);
587 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
588 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
589 clear_bit(EXTENT_FLAG_LOGGING
, &flags
);
590 modified
= !list_empty(&em
->list
);
594 if (em
->start
< start
) {
595 split
->start
= em
->start
;
596 split
->len
= start
- em
->start
;
598 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
) {
599 split
->orig_start
= em
->orig_start
;
600 split
->block_start
= em
->block_start
;
603 split
->block_len
= em
->block_len
;
605 split
->block_len
= split
->len
;
606 split
->orig_block_len
= max(split
->block_len
,
608 split
->ram_bytes
= em
->ram_bytes
;
610 split
->orig_start
= split
->start
;
611 split
->block_len
= 0;
612 split
->block_start
= em
->block_start
;
613 split
->orig_block_len
= 0;
614 split
->ram_bytes
= split
->len
;
617 split
->generation
= gen
;
618 split
->bdev
= em
->bdev
;
619 split
->flags
= flags
;
620 split
->compress_type
= em
->compress_type
;
621 replace_extent_mapping(em_tree
, em
, split
, modified
);
622 free_extent_map(split
);
626 if (testend
&& em
->start
+ em
->len
> start
+ len
) {
627 u64 diff
= start
+ len
- em
->start
;
629 split
->start
= start
+ len
;
630 split
->len
= em
->start
+ em
->len
- (start
+ len
);
631 split
->bdev
= em
->bdev
;
632 split
->flags
= flags
;
633 split
->compress_type
= em
->compress_type
;
634 split
->generation
= gen
;
636 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
) {
637 split
->orig_block_len
= max(em
->block_len
,
640 split
->ram_bytes
= em
->ram_bytes
;
642 split
->block_len
= em
->block_len
;
643 split
->block_start
= em
->block_start
;
644 split
->orig_start
= em
->orig_start
;
646 split
->block_len
= split
->len
;
647 split
->block_start
= em
->block_start
649 split
->orig_start
= em
->orig_start
;
652 split
->ram_bytes
= split
->len
;
653 split
->orig_start
= split
->start
;
654 split
->block_len
= 0;
655 split
->block_start
= em
->block_start
;
656 split
->orig_block_len
= 0;
659 if (extent_map_in_tree(em
)) {
660 replace_extent_mapping(em_tree
, em
, split
,
663 ret
= add_extent_mapping(em_tree
, split
,
665 ASSERT(ret
== 0); /* Logic error */
667 free_extent_map(split
);
671 if (extent_map_in_tree(em
))
672 remove_extent_mapping(em_tree
, em
);
673 write_unlock(&em_tree
->lock
);
677 /* once for the tree*/
681 free_extent_map(split
);
683 free_extent_map(split2
);
687 * this is very complex, but the basic idea is to drop all extents
688 * in the range start - end. hint_block is filled in with a block number
689 * that would be a good hint to the block allocator for this file.
691 * If an extent intersects the range but is not entirely inside the range
692 * it is either truncated or split. Anything entirely inside the range
693 * is deleted from the tree.
695 int __btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
696 struct btrfs_root
*root
, struct inode
*inode
,
697 struct btrfs_path
*path
, u64 start
, u64 end
,
698 u64
*drop_end
, int drop_cache
,
700 u32 extent_item_size
,
703 struct extent_buffer
*leaf
;
704 struct btrfs_file_extent_item
*fi
;
705 struct btrfs_key key
;
706 struct btrfs_key new_key
;
707 u64 ino
= btrfs_ino(inode
);
708 u64 search_start
= start
;
711 u64 extent_offset
= 0;
718 int modify_tree
= -1;
721 int leafs_visited
= 0;
724 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
726 if (start
>= BTRFS_I(inode
)->disk_i_size
&& !replace_extent
)
729 update_refs
= (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
730 root
== root
->fs_info
->tree_root
);
733 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
734 search_start
, modify_tree
);
737 if (ret
> 0 && path
->slots
[0] > 0 && search_start
== start
) {
738 leaf
= path
->nodes
[0];
739 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0] - 1);
740 if (key
.objectid
== ino
&&
741 key
.type
== BTRFS_EXTENT_DATA_KEY
)
747 leaf
= path
->nodes
[0];
748 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
750 ret
= btrfs_next_leaf(root
, path
);
758 leaf
= path
->nodes
[0];
762 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
763 if (key
.objectid
> ino
||
764 key
.type
> BTRFS_EXTENT_DATA_KEY
|| key
.offset
>= end
)
767 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
768 struct btrfs_file_extent_item
);
769 extent_type
= btrfs_file_extent_type(leaf
, fi
);
771 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
772 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
773 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
774 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
775 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
776 extent_end
= key
.offset
+
777 btrfs_file_extent_num_bytes(leaf
, fi
);
778 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
779 extent_end
= key
.offset
+
780 btrfs_file_extent_inline_len(leaf
,
784 extent_end
= search_start
;
788 * Don't skip extent items representing 0 byte lengths. They
789 * used to be created (bug) if while punching holes we hit
790 * -ENOSPC condition. So if we find one here, just ensure we
791 * delete it, otherwise we would insert a new file extent item
792 * with the same key (offset) as that 0 bytes length file
793 * extent item in the call to setup_items_for_insert() later
796 if (extent_end
== key
.offset
&& extent_end
>= search_start
)
797 goto delete_extent_item
;
799 if (extent_end
<= search_start
) {
805 search_start
= max(key
.offset
, start
);
806 if (recow
|| !modify_tree
) {
808 btrfs_release_path(path
);
813 * | - range to drop - |
814 * | -------- extent -------- |
816 if (start
> key
.offset
&& end
< extent_end
) {
818 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
823 memcpy(&new_key
, &key
, sizeof(new_key
));
824 new_key
.offset
= start
;
825 ret
= btrfs_duplicate_item(trans
, root
, path
,
827 if (ret
== -EAGAIN
) {
828 btrfs_release_path(path
);
834 leaf
= path
->nodes
[0];
835 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
836 struct btrfs_file_extent_item
);
837 btrfs_set_file_extent_num_bytes(leaf
, fi
,
840 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
841 struct btrfs_file_extent_item
);
843 extent_offset
+= start
- key
.offset
;
844 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
845 btrfs_set_file_extent_num_bytes(leaf
, fi
,
847 btrfs_mark_buffer_dirty(leaf
);
849 if (update_refs
&& disk_bytenr
> 0) {
850 ret
= btrfs_inc_extent_ref(trans
, root
,
851 disk_bytenr
, num_bytes
, 0,
852 root
->root_key
.objectid
,
854 start
- extent_offset
, 1);
855 BUG_ON(ret
); /* -ENOMEM */
860 * | ---- range to drop ----- |
861 * | -------- extent -------- |
863 if (start
<= key
.offset
&& end
< extent_end
) {
864 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
869 memcpy(&new_key
, &key
, sizeof(new_key
));
870 new_key
.offset
= end
;
871 btrfs_set_item_key_safe(root
, path
, &new_key
);
873 extent_offset
+= end
- key
.offset
;
874 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
875 btrfs_set_file_extent_num_bytes(leaf
, fi
,
877 btrfs_mark_buffer_dirty(leaf
);
878 if (update_refs
&& disk_bytenr
> 0)
879 inode_sub_bytes(inode
, end
- key
.offset
);
883 search_start
= extent_end
;
885 * | ---- range to drop ----- |
886 * | -------- extent -------- |
888 if (start
> key
.offset
&& end
>= extent_end
) {
890 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
895 btrfs_set_file_extent_num_bytes(leaf
, fi
,
897 btrfs_mark_buffer_dirty(leaf
);
898 if (update_refs
&& disk_bytenr
> 0)
899 inode_sub_bytes(inode
, extent_end
- start
);
900 if (end
== extent_end
)
908 * | ---- range to drop ----- |
909 * | ------ extent ------ |
911 if (start
<= key
.offset
&& end
>= extent_end
) {
914 del_slot
= path
->slots
[0];
917 BUG_ON(del_slot
+ del_nr
!= path
->slots
[0]);
922 extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
923 inode_sub_bytes(inode
,
924 extent_end
- key
.offset
);
925 extent_end
= ALIGN(extent_end
,
927 } else if (update_refs
&& disk_bytenr
> 0) {
928 ret
= btrfs_free_extent(trans
, root
,
929 disk_bytenr
, num_bytes
, 0,
930 root
->root_key
.objectid
,
931 key
.objectid
, key
.offset
-
933 BUG_ON(ret
); /* -ENOMEM */
934 inode_sub_bytes(inode
,
935 extent_end
- key
.offset
);
938 if (end
== extent_end
)
941 if (path
->slots
[0] + 1 < btrfs_header_nritems(leaf
)) {
946 ret
= btrfs_del_items(trans
, root
, path
, del_slot
,
949 btrfs_abort_transaction(trans
, root
, ret
);
956 btrfs_release_path(path
);
963 if (!ret
&& del_nr
> 0) {
965 * Set path->slots[0] to first slot, so that after the delete
966 * if items are move off from our leaf to its immediate left or
967 * right neighbor leafs, we end up with a correct and adjusted
968 * path->slots[0] for our insertion (if replace_extent != 0).
970 path
->slots
[0] = del_slot
;
971 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
973 btrfs_abort_transaction(trans
, root
, ret
);
976 leaf
= path
->nodes
[0];
978 * If btrfs_del_items() was called, it might have deleted a leaf, in
979 * which case it unlocked our path, so check path->locks[0] matches a
982 if (!ret
&& replace_extent
&& leafs_visited
== 1 &&
983 (path
->locks
[0] == BTRFS_WRITE_LOCK_BLOCKING
||
984 path
->locks
[0] == BTRFS_WRITE_LOCK
) &&
985 btrfs_leaf_free_space(root
, leaf
) >=
986 sizeof(struct btrfs_item
) + extent_item_size
) {
989 key
.type
= BTRFS_EXTENT_DATA_KEY
;
991 if (!del_nr
&& path
->slots
[0] < btrfs_header_nritems(leaf
)) {
992 struct btrfs_key slot_key
;
994 btrfs_item_key_to_cpu(leaf
, &slot_key
, path
->slots
[0]);
995 if (btrfs_comp_cpu_keys(&key
, &slot_key
) > 0)
998 setup_items_for_insert(root
, path
, &key
,
1001 sizeof(struct btrfs_item
) +
1002 extent_item_size
, 1);
1006 if (!replace_extent
|| !(*key_inserted
))
1007 btrfs_release_path(path
);
1009 *drop_end
= found
? min(end
, extent_end
) : end
;
1013 int btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
1014 struct btrfs_root
*root
, struct inode
*inode
, u64 start
,
1015 u64 end
, int drop_cache
)
1017 struct btrfs_path
*path
;
1020 path
= btrfs_alloc_path();
1023 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, start
, end
, NULL
,
1024 drop_cache
, 0, 0, NULL
);
1025 btrfs_free_path(path
);
1029 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
1030 u64 objectid
, u64 bytenr
, u64 orig_offset
,
1031 u64
*start
, u64
*end
)
1033 struct btrfs_file_extent_item
*fi
;
1034 struct btrfs_key key
;
1037 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
1040 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
1041 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
1044 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
1045 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
1046 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
1047 btrfs_file_extent_offset(leaf
, fi
) != key
.offset
- orig_offset
||
1048 btrfs_file_extent_compression(leaf
, fi
) ||
1049 btrfs_file_extent_encryption(leaf
, fi
) ||
1050 btrfs_file_extent_other_encoding(leaf
, fi
))
1053 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
1054 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
1057 *start
= key
.offset
;
1063 * Mark extent in the range start - end as written.
1065 * This changes extent type from 'pre-allocated' to 'regular'. If only
1066 * part of extent is marked as written, the extent will be split into
1069 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
1070 struct inode
*inode
, u64 start
, u64 end
)
1072 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1073 struct extent_buffer
*leaf
;
1074 struct btrfs_path
*path
;
1075 struct btrfs_file_extent_item
*fi
;
1076 struct btrfs_key key
;
1077 struct btrfs_key new_key
;
1089 u64 ino
= btrfs_ino(inode
);
1091 path
= btrfs_alloc_path();
1098 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1101 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1104 if (ret
> 0 && path
->slots
[0] > 0)
1107 leaf
= path
->nodes
[0];
1108 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1109 BUG_ON(key
.objectid
!= ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
);
1110 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1111 struct btrfs_file_extent_item
);
1112 BUG_ON(btrfs_file_extent_type(leaf
, fi
) !=
1113 BTRFS_FILE_EXTENT_PREALLOC
);
1114 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
1115 BUG_ON(key
.offset
> start
|| extent_end
< end
);
1117 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1118 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1119 orig_offset
= key
.offset
- btrfs_file_extent_offset(leaf
, fi
);
1120 memcpy(&new_key
, &key
, sizeof(new_key
));
1122 if (start
== key
.offset
&& end
< extent_end
) {
1125 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1126 ino
, bytenr
, orig_offset
,
1127 &other_start
, &other_end
)) {
1128 new_key
.offset
= end
;
1129 btrfs_set_item_key_safe(root
, path
, &new_key
);
1130 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1131 struct btrfs_file_extent_item
);
1132 btrfs_set_file_extent_generation(leaf
, fi
,
1134 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1136 btrfs_set_file_extent_offset(leaf
, fi
,
1138 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1139 struct btrfs_file_extent_item
);
1140 btrfs_set_file_extent_generation(leaf
, fi
,
1142 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1144 btrfs_mark_buffer_dirty(leaf
);
1149 if (start
> key
.offset
&& end
== extent_end
) {
1152 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1153 ino
, bytenr
, orig_offset
,
1154 &other_start
, &other_end
)) {
1155 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1156 struct btrfs_file_extent_item
);
1157 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1158 start
- key
.offset
);
1159 btrfs_set_file_extent_generation(leaf
, fi
,
1162 new_key
.offset
= start
;
1163 btrfs_set_item_key_safe(root
, path
, &new_key
);
1165 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1166 struct btrfs_file_extent_item
);
1167 btrfs_set_file_extent_generation(leaf
, fi
,
1169 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1171 btrfs_set_file_extent_offset(leaf
, fi
,
1172 start
- orig_offset
);
1173 btrfs_mark_buffer_dirty(leaf
);
1178 while (start
> key
.offset
|| end
< extent_end
) {
1179 if (key
.offset
== start
)
1182 new_key
.offset
= split
;
1183 ret
= btrfs_duplicate_item(trans
, root
, path
, &new_key
);
1184 if (ret
== -EAGAIN
) {
1185 btrfs_release_path(path
);
1189 btrfs_abort_transaction(trans
, root
, ret
);
1193 leaf
= path
->nodes
[0];
1194 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1195 struct btrfs_file_extent_item
);
1196 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1197 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1198 split
- key
.offset
);
1200 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1201 struct btrfs_file_extent_item
);
1203 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1204 btrfs_set_file_extent_offset(leaf
, fi
, split
- orig_offset
);
1205 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1206 extent_end
- split
);
1207 btrfs_mark_buffer_dirty(leaf
);
1209 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
, 0,
1210 root
->root_key
.objectid
,
1211 ino
, orig_offset
, 1);
1212 BUG_ON(ret
); /* -ENOMEM */
1214 if (split
== start
) {
1217 BUG_ON(start
!= key
.offset
);
1226 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1227 ino
, bytenr
, orig_offset
,
1228 &other_start
, &other_end
)) {
1230 btrfs_release_path(path
);
1233 extent_end
= other_end
;
1234 del_slot
= path
->slots
[0] + 1;
1236 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1237 0, root
->root_key
.objectid
,
1238 ino
, orig_offset
, 0);
1239 BUG_ON(ret
); /* -ENOMEM */
1243 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1244 ino
, bytenr
, orig_offset
,
1245 &other_start
, &other_end
)) {
1247 btrfs_release_path(path
);
1250 key
.offset
= other_start
;
1251 del_slot
= path
->slots
[0];
1253 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1254 0, root
->root_key
.objectid
,
1255 ino
, orig_offset
, 0);
1256 BUG_ON(ret
); /* -ENOMEM */
1259 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1260 struct btrfs_file_extent_item
);
1261 btrfs_set_file_extent_type(leaf
, fi
,
1262 BTRFS_FILE_EXTENT_REG
);
1263 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1264 btrfs_mark_buffer_dirty(leaf
);
1266 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
1267 struct btrfs_file_extent_item
);
1268 btrfs_set_file_extent_type(leaf
, fi
,
1269 BTRFS_FILE_EXTENT_REG
);
1270 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1271 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1272 extent_end
- key
.offset
);
1273 btrfs_mark_buffer_dirty(leaf
);
1275 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
1277 btrfs_abort_transaction(trans
, root
, ret
);
1282 btrfs_free_path(path
);
1287 * on error we return an unlocked page and the error value
1288 * on success we return a locked page and 0
1290 static int prepare_uptodate_page(struct page
*page
, u64 pos
,
1291 bool force_uptodate
)
1295 if (((pos
& (PAGE_CACHE_SIZE
- 1)) || force_uptodate
) &&
1296 !PageUptodate(page
)) {
1297 ret
= btrfs_readpage(NULL
, page
);
1301 if (!PageUptodate(page
)) {
1310 * this just gets pages into the page cache and locks them down.
1312 static noinline
int prepare_pages(struct inode
*inode
, struct page
**pages
,
1313 size_t num_pages
, loff_t pos
,
1314 size_t write_bytes
, bool force_uptodate
)
1317 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
1318 gfp_t mask
= btrfs_alloc_write_mask(inode
->i_mapping
);
1322 for (i
= 0; i
< num_pages
; i
++) {
1323 pages
[i
] = find_or_create_page(inode
->i_mapping
, index
+ i
,
1324 mask
| __GFP_WRITE
);
1332 err
= prepare_uptodate_page(pages
[i
], pos
,
1334 if (i
== num_pages
- 1)
1335 err
= prepare_uptodate_page(pages
[i
],
1336 pos
+ write_bytes
, false);
1338 page_cache_release(pages
[i
]);
1342 wait_on_page_writeback(pages
[i
]);
1347 while (faili
>= 0) {
1348 unlock_page(pages
[faili
]);
1349 page_cache_release(pages
[faili
]);
1357 * This function locks the extent and properly waits for data=ordered extents
1358 * to finish before allowing the pages to be modified if need.
1361 * 1 - the extent is locked
1362 * 0 - the extent is not locked, and everything is OK
1363 * -EAGAIN - need re-prepare the pages
1364 * the other < 0 number - Something wrong happens
1367 lock_and_cleanup_extent_if_need(struct inode
*inode
, struct page
**pages
,
1368 size_t num_pages
, loff_t pos
,
1369 u64
*lockstart
, u64
*lockend
,
1370 struct extent_state
**cached_state
)
1377 start_pos
= pos
& ~((u64
)PAGE_CACHE_SIZE
- 1);
1378 last_pos
= start_pos
+ ((u64
)num_pages
<< PAGE_CACHE_SHIFT
) - 1;
1380 if (start_pos
< inode
->i_size
) {
1381 struct btrfs_ordered_extent
*ordered
;
1382 lock_extent_bits(&BTRFS_I(inode
)->io_tree
,
1383 start_pos
, last_pos
, 0, cached_state
);
1384 ordered
= btrfs_lookup_ordered_range(inode
, start_pos
,
1385 last_pos
- start_pos
+ 1);
1387 ordered
->file_offset
+ ordered
->len
> start_pos
&&
1388 ordered
->file_offset
<= last_pos
) {
1389 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1390 start_pos
, last_pos
,
1391 cached_state
, GFP_NOFS
);
1392 for (i
= 0; i
< num_pages
; i
++) {
1393 unlock_page(pages
[i
]);
1394 page_cache_release(pages
[i
]);
1396 btrfs_start_ordered_extent(inode
, ordered
, 1);
1397 btrfs_put_ordered_extent(ordered
);
1401 btrfs_put_ordered_extent(ordered
);
1403 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start_pos
,
1404 last_pos
, EXTENT_DIRTY
| EXTENT_DELALLOC
|
1405 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
1406 0, 0, cached_state
, GFP_NOFS
);
1407 *lockstart
= start_pos
;
1408 *lockend
= last_pos
;
1412 for (i
= 0; i
< num_pages
; i
++) {
1413 if (clear_page_dirty_for_io(pages
[i
]))
1414 account_page_redirty(pages
[i
]);
1415 set_page_extent_mapped(pages
[i
]);
1416 WARN_ON(!PageLocked(pages
[i
]));
1422 static noinline
int check_can_nocow(struct inode
*inode
, loff_t pos
,
1423 size_t *write_bytes
)
1425 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1426 struct btrfs_ordered_extent
*ordered
;
1427 u64 lockstart
, lockend
;
1431 ret
= btrfs_start_nocow_write(root
);
1435 lockstart
= round_down(pos
, root
->sectorsize
);
1436 lockend
= round_up(pos
+ *write_bytes
, root
->sectorsize
) - 1;
1439 lock_extent(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
);
1440 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
1441 lockend
- lockstart
+ 1);
1445 unlock_extent(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
);
1446 btrfs_start_ordered_extent(inode
, ordered
, 1);
1447 btrfs_put_ordered_extent(ordered
);
1450 num_bytes
= lockend
- lockstart
+ 1;
1451 ret
= can_nocow_extent(inode
, lockstart
, &num_bytes
, NULL
, NULL
, NULL
);
1454 btrfs_end_nocow_write(root
);
1456 *write_bytes
= min_t(size_t, *write_bytes
,
1457 num_bytes
- pos
+ lockstart
);
1460 unlock_extent(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
);
1465 static noinline ssize_t
__btrfs_buffered_write(struct file
*file
,
1469 struct inode
*inode
= file_inode(file
);
1470 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1471 struct page
**pages
= NULL
;
1472 struct extent_state
*cached_state
= NULL
;
1473 u64 release_bytes
= 0;
1476 unsigned long first_index
;
1477 size_t num_written
= 0;
1480 bool only_release_metadata
= false;
1481 bool force_page_uptodate
= false;
1484 nrptrs
= min((iov_iter_count(i
) + PAGE_CACHE_SIZE
- 1) /
1485 PAGE_CACHE_SIZE
, PAGE_CACHE_SIZE
/
1486 (sizeof(struct page
*)));
1487 nrptrs
= min(nrptrs
, current
->nr_dirtied_pause
- current
->nr_dirtied
);
1488 nrptrs
= max(nrptrs
, 8);
1489 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
1493 first_index
= pos
>> PAGE_CACHE_SHIFT
;
1495 while (iov_iter_count(i
) > 0) {
1496 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1497 size_t write_bytes
= min(iov_iter_count(i
),
1498 nrptrs
* (size_t)PAGE_CACHE_SIZE
-
1500 size_t num_pages
= (write_bytes
+ offset
+
1501 PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1502 size_t reserve_bytes
;
1506 WARN_ON(num_pages
> nrptrs
);
1509 * Fault pages before locking them in prepare_pages
1510 * to avoid recursive lock
1512 if (unlikely(iov_iter_fault_in_readable(i
, write_bytes
))) {
1517 reserve_bytes
= num_pages
<< PAGE_CACHE_SHIFT
;
1518 ret
= btrfs_check_data_free_space(inode
, reserve_bytes
);
1519 if (ret
== -ENOSPC
&&
1520 (BTRFS_I(inode
)->flags
& (BTRFS_INODE_NODATACOW
|
1521 BTRFS_INODE_PREALLOC
))) {
1522 ret
= check_can_nocow(inode
, pos
, &write_bytes
);
1524 only_release_metadata
= true;
1526 * our prealloc extent may be smaller than
1527 * write_bytes, so scale down.
1529 num_pages
= (write_bytes
+ offset
+
1530 PAGE_CACHE_SIZE
- 1) >>
1532 reserve_bytes
= num_pages
<< PAGE_CACHE_SHIFT
;
1542 ret
= btrfs_delalloc_reserve_metadata(inode
, reserve_bytes
);
1544 if (!only_release_metadata
)
1545 btrfs_free_reserved_data_space(inode
,
1548 btrfs_end_nocow_write(root
);
1552 release_bytes
= reserve_bytes
;
1553 need_unlock
= false;
1556 * This is going to setup the pages array with the number of
1557 * pages we want, so we don't really need to worry about the
1558 * contents of pages from loop to loop
1560 ret
= prepare_pages(inode
, pages
, num_pages
,
1562 force_page_uptodate
);
1566 ret
= lock_and_cleanup_extent_if_need(inode
, pages
, num_pages
,
1567 pos
, &lockstart
, &lockend
,
1573 } else if (ret
> 0) {
1578 copied
= btrfs_copy_from_user(pos
, num_pages
,
1579 write_bytes
, pages
, i
);
1582 * if we have trouble faulting in the pages, fall
1583 * back to one page at a time
1585 if (copied
< write_bytes
)
1589 force_page_uptodate
= true;
1592 force_page_uptodate
= false;
1593 dirty_pages
= (copied
+ offset
+
1594 PAGE_CACHE_SIZE
- 1) >>
1599 * If we had a short copy we need to release the excess delaloc
1600 * bytes we reserved. We need to increment outstanding_extents
1601 * because btrfs_delalloc_release_space will decrement it, but
1602 * we still have an outstanding extent for the chunk we actually
1605 if (num_pages
> dirty_pages
) {
1606 release_bytes
= (num_pages
- dirty_pages
) <<
1609 spin_lock(&BTRFS_I(inode
)->lock
);
1610 BTRFS_I(inode
)->outstanding_extents
++;
1611 spin_unlock(&BTRFS_I(inode
)->lock
);
1613 if (only_release_metadata
)
1614 btrfs_delalloc_release_metadata(inode
,
1617 btrfs_delalloc_release_space(inode
,
1621 release_bytes
= dirty_pages
<< PAGE_CACHE_SHIFT
;
1624 ret
= btrfs_dirty_pages(root
, inode
, pages
,
1625 dirty_pages
, pos
, copied
,
1628 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1629 lockstart
, lockend
, &cached_state
,
1632 btrfs_drop_pages(pages
, num_pages
);
1637 if (only_release_metadata
)
1638 btrfs_end_nocow_write(root
);
1640 if (only_release_metadata
&& copied
> 0) {
1641 u64 lockstart
= round_down(pos
, root
->sectorsize
);
1642 u64 lockend
= lockstart
+
1643 (dirty_pages
<< PAGE_CACHE_SHIFT
) - 1;
1645 set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
1646 lockend
, EXTENT_NORESERVE
, NULL
,
1648 only_release_metadata
= false;
1651 btrfs_drop_pages(pages
, num_pages
);
1655 balance_dirty_pages_ratelimited(inode
->i_mapping
);
1656 if (dirty_pages
< (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
1657 btrfs_btree_balance_dirty(root
);
1660 num_written
+= copied
;
1665 if (release_bytes
) {
1666 if (only_release_metadata
) {
1667 btrfs_end_nocow_write(root
);
1668 btrfs_delalloc_release_metadata(inode
, release_bytes
);
1670 btrfs_delalloc_release_space(inode
, release_bytes
);
1674 return num_written
? num_written
: ret
;
1677 static ssize_t
__btrfs_direct_write(struct kiocb
*iocb
,
1678 struct iov_iter
*from
,
1681 struct file
*file
= iocb
->ki_filp
;
1683 ssize_t written_buffered
;
1687 written
= generic_file_direct_write(iocb
, from
, pos
);
1689 if (written
< 0 || !iov_iter_count(from
))
1693 written_buffered
= __btrfs_buffered_write(file
, from
, pos
);
1694 if (written_buffered
< 0) {
1695 err
= written_buffered
;
1698 endbyte
= pos
+ written_buffered
- 1;
1699 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
1702 written
+= written_buffered
;
1703 iocb
->ki_pos
= pos
+ written_buffered
;
1704 invalidate_mapping_pages(file
->f_mapping
, pos
>> PAGE_CACHE_SHIFT
,
1705 endbyte
>> PAGE_CACHE_SHIFT
);
1707 return written
? written
: err
;
1710 static void update_time_for_write(struct inode
*inode
)
1712 struct timespec now
;
1714 if (IS_NOCMTIME(inode
))
1717 now
= current_fs_time(inode
->i_sb
);
1718 if (!timespec_equal(&inode
->i_mtime
, &now
))
1719 inode
->i_mtime
= now
;
1721 if (!timespec_equal(&inode
->i_ctime
, &now
))
1722 inode
->i_ctime
= now
;
1724 if (IS_I_VERSION(inode
))
1725 inode_inc_iversion(inode
);
1728 static ssize_t
btrfs_file_write_iter(struct kiocb
*iocb
,
1729 struct iov_iter
*from
)
1731 struct file
*file
= iocb
->ki_filp
;
1732 struct inode
*inode
= file_inode(file
);
1733 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1736 ssize_t num_written
= 0;
1738 size_t count
= iov_iter_count(from
);
1739 bool sync
= (file
->f_flags
& O_DSYNC
) || IS_SYNC(file
->f_mapping
->host
);
1740 loff_t pos
= iocb
->ki_pos
;
1742 mutex_lock(&inode
->i_mutex
);
1744 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
1745 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1747 mutex_unlock(&inode
->i_mutex
);
1752 mutex_unlock(&inode
->i_mutex
);
1756 iov_iter_truncate(from
, count
);
1758 err
= file_remove_suid(file
);
1760 mutex_unlock(&inode
->i_mutex
);
1765 * If BTRFS flips readonly due to some impossible error
1766 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1767 * although we have opened a file as writable, we have
1768 * to stop this write operation to ensure FS consistency.
1770 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
)) {
1771 mutex_unlock(&inode
->i_mutex
);
1777 * We reserve space for updating the inode when we reserve space for the
1778 * extent we are going to write, so we will enospc out there. We don't
1779 * need to start yet another transaction to update the inode as we will
1780 * update the inode when we finish writing whatever data we write.
1782 update_time_for_write(inode
);
1784 start_pos
= round_down(pos
, root
->sectorsize
);
1785 if (start_pos
> i_size_read(inode
)) {
1786 /* Expand hole size to cover write data, preventing empty gap */
1787 end_pos
= round_up(pos
+ count
, root
->sectorsize
);
1788 err
= btrfs_cont_expand(inode
, i_size_read(inode
), end_pos
);
1790 mutex_unlock(&inode
->i_mutex
);
1796 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
1798 if (unlikely(file
->f_flags
& O_DIRECT
)) {
1799 num_written
= __btrfs_direct_write(iocb
, from
, pos
);
1801 num_written
= __btrfs_buffered_write(file
, from
, pos
);
1802 if (num_written
> 0)
1803 iocb
->ki_pos
= pos
+ num_written
;
1806 mutex_unlock(&inode
->i_mutex
);
1809 * we want to make sure fsync finds this change
1810 * but we haven't joined a transaction running right now.
1812 * Later on, someone is sure to update the inode and get the
1813 * real transid recorded.
1815 * We set last_trans now to the fs_info generation + 1,
1816 * this will either be one more than the running transaction
1817 * or the generation used for the next transaction if there isn't
1818 * one running right now.
1820 * We also have to set last_sub_trans to the current log transid,
1821 * otherwise subsequent syncs to a file that's been synced in this
1822 * transaction will appear to have already occured.
1824 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
+ 1;
1825 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
1826 if (num_written
> 0) {
1827 err
= generic_write_sync(file
, pos
, num_written
);
1833 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
1835 current
->backing_dev_info
= NULL
;
1836 return num_written
? num_written
: err
;
1839 int btrfs_release_file(struct inode
*inode
, struct file
*filp
)
1842 * ordered_data_close is set by settattr when we are about to truncate
1843 * a file from a non-zero size to a zero size. This tries to
1844 * flush down new bytes that may have been written if the
1845 * application were using truncate to replace a file in place.
1847 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
1848 &BTRFS_I(inode
)->runtime_flags
)) {
1849 struct btrfs_trans_handle
*trans
;
1850 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1853 * We need to block on a committing transaction to keep us from
1854 * throwing a ordered operation on to the list and causing
1855 * something like sync to deadlock trying to flush out this
1858 trans
= btrfs_start_transaction(root
, 0);
1860 return PTR_ERR(trans
);
1861 btrfs_add_ordered_operation(trans
, BTRFS_I(inode
)->root
, inode
);
1862 btrfs_end_transaction(trans
, root
);
1863 if (inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
1864 filemap_flush(inode
->i_mapping
);
1866 if (filp
->private_data
)
1867 btrfs_ioctl_trans_end(filp
);
1872 * fsync call for both files and directories. This logs the inode into
1873 * the tree log instead of forcing full commits whenever possible.
1875 * It needs to call filemap_fdatawait so that all ordered extent updates are
1876 * in the metadata btree are up to date for copying to the log.
1878 * It drops the inode mutex before doing the tree log commit. This is an
1879 * important optimization for directories because holding the mutex prevents
1880 * new operations on the dir while we write to disk.
1882 int btrfs_sync_file(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1884 struct dentry
*dentry
= file
->f_path
.dentry
;
1885 struct inode
*inode
= dentry
->d_inode
;
1886 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1887 struct btrfs_trans_handle
*trans
;
1888 struct btrfs_log_ctx ctx
;
1892 trace_btrfs_sync_file(file
, datasync
);
1895 * We write the dirty pages in the range and wait until they complete
1896 * out of the ->i_mutex. If so, we can flush the dirty pages by
1897 * multi-task, and make the performance up. See
1898 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1900 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
1901 ret
= filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
1902 if (!ret
&& test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1903 &BTRFS_I(inode
)->runtime_flags
))
1904 ret
= filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
1905 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
1909 mutex_lock(&inode
->i_mutex
);
1912 * We flush the dirty pages again to avoid some dirty pages in the
1915 atomic_inc(&root
->log_batch
);
1916 full_sync
= test_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1917 &BTRFS_I(inode
)->runtime_flags
);
1919 ret
= btrfs_wait_ordered_range(inode
, start
, end
- start
+ 1);
1921 mutex_unlock(&inode
->i_mutex
);
1925 atomic_inc(&root
->log_batch
);
1928 * check the transaction that last modified this inode
1929 * and see if its already been committed
1931 if (!BTRFS_I(inode
)->last_trans
) {
1932 mutex_unlock(&inode
->i_mutex
);
1937 * if the last transaction that changed this file was before
1938 * the current transaction, we can bail out now without any
1942 if (btrfs_inode_in_log(inode
, root
->fs_info
->generation
) ||
1943 BTRFS_I(inode
)->last_trans
<=
1944 root
->fs_info
->last_trans_committed
) {
1945 BTRFS_I(inode
)->last_trans
= 0;
1948 * We'v had everything committed since the last time we were
1949 * modified so clear this flag in case it was set for whatever
1950 * reason, it's no longer relevant.
1952 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1953 &BTRFS_I(inode
)->runtime_flags
);
1954 mutex_unlock(&inode
->i_mutex
);
1959 * ok we haven't committed the transaction yet, lets do a commit
1961 if (file
->private_data
)
1962 btrfs_ioctl_trans_end(file
);
1965 * We use start here because we will need to wait on the IO to complete
1966 * in btrfs_sync_log, which could require joining a transaction (for
1967 * example checking cross references in the nocow path). If we use join
1968 * here we could get into a situation where we're waiting on IO to
1969 * happen that is blocked on a transaction trying to commit. With start
1970 * we inc the extwriter counter, so we wait for all extwriters to exit
1971 * before we start blocking join'ers. This comment is to keep somebody
1972 * from thinking they are super smart and changing this to
1973 * btrfs_join_transaction *cough*Josef*cough*.
1975 trans
= btrfs_start_transaction(root
, 0);
1976 if (IS_ERR(trans
)) {
1977 ret
= PTR_ERR(trans
);
1978 mutex_unlock(&inode
->i_mutex
);
1983 btrfs_init_log_ctx(&ctx
);
1985 ret
= btrfs_log_dentry_safe(trans
, root
, dentry
, &ctx
);
1987 /* Fallthrough and commit/free transaction. */
1991 /* we've logged all the items and now have a consistent
1992 * version of the file in the log. It is possible that
1993 * someone will come in and modify the file, but that's
1994 * fine because the log is consistent on disk, and we
1995 * have references to all of the file's extents
1997 * It is possible that someone will come in and log the
1998 * file again, but that will end up using the synchronization
1999 * inside btrfs_sync_log to keep things safe.
2001 mutex_unlock(&inode
->i_mutex
);
2003 if (ret
!= BTRFS_NO_LOG_SYNC
) {
2005 ret
= btrfs_sync_log(trans
, root
, &ctx
);
2007 ret
= btrfs_end_transaction(trans
, root
);
2012 ret
= btrfs_wait_ordered_range(inode
, start
,
2015 btrfs_end_transaction(trans
, root
);
2019 ret
= btrfs_commit_transaction(trans
, root
);
2021 ret
= btrfs_end_transaction(trans
, root
);
2024 return ret
> 0 ? -EIO
: ret
;
2027 static const struct vm_operations_struct btrfs_file_vm_ops
= {
2028 .fault
= filemap_fault
,
2029 .map_pages
= filemap_map_pages
,
2030 .page_mkwrite
= btrfs_page_mkwrite
,
2031 .remap_pages
= generic_file_remap_pages
,
2034 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
2036 struct address_space
*mapping
= filp
->f_mapping
;
2038 if (!mapping
->a_ops
->readpage
)
2041 file_accessed(filp
);
2042 vma
->vm_ops
= &btrfs_file_vm_ops
;
2047 static int hole_mergeable(struct inode
*inode
, struct extent_buffer
*leaf
,
2048 int slot
, u64 start
, u64 end
)
2050 struct btrfs_file_extent_item
*fi
;
2051 struct btrfs_key key
;
2053 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
2056 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2057 if (key
.objectid
!= btrfs_ino(inode
) ||
2058 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2061 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
2063 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2066 if (btrfs_file_extent_disk_bytenr(leaf
, fi
))
2069 if (key
.offset
== end
)
2071 if (key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
) == start
)
2076 static int fill_holes(struct btrfs_trans_handle
*trans
, struct inode
*inode
,
2077 struct btrfs_path
*path
, u64 offset
, u64 end
)
2079 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2080 struct extent_buffer
*leaf
;
2081 struct btrfs_file_extent_item
*fi
;
2082 struct extent_map
*hole_em
;
2083 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
2084 struct btrfs_key key
;
2087 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
))
2090 key
.objectid
= btrfs_ino(inode
);
2091 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2092 key
.offset
= offset
;
2094 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2099 leaf
= path
->nodes
[0];
2100 if (hole_mergeable(inode
, leaf
, path
->slots
[0]-1, offset
, end
)) {
2104 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2105 struct btrfs_file_extent_item
);
2106 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) +
2108 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2109 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
2110 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2111 btrfs_mark_buffer_dirty(leaf
);
2115 if (hole_mergeable(inode
, leaf
, path
->slots
[0]+1, offset
, end
)) {
2119 key
.offset
= offset
;
2120 btrfs_set_item_key_safe(root
, path
, &key
);
2121 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2122 struct btrfs_file_extent_item
);
2123 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) + end
-
2125 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2126 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
2127 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2128 btrfs_mark_buffer_dirty(leaf
);
2131 btrfs_release_path(path
);
2133 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
2134 0, 0, end
- offset
, 0, end
- offset
,
2140 btrfs_release_path(path
);
2142 hole_em
= alloc_extent_map();
2144 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
2145 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
2146 &BTRFS_I(inode
)->runtime_flags
);
2148 hole_em
->start
= offset
;
2149 hole_em
->len
= end
- offset
;
2150 hole_em
->ram_bytes
= hole_em
->len
;
2151 hole_em
->orig_start
= offset
;
2153 hole_em
->block_start
= EXTENT_MAP_HOLE
;
2154 hole_em
->block_len
= 0;
2155 hole_em
->orig_block_len
= 0;
2156 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
2157 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
2158 hole_em
->generation
= trans
->transid
;
2161 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
2162 write_lock(&em_tree
->lock
);
2163 ret
= add_extent_mapping(em_tree
, hole_em
, 1);
2164 write_unlock(&em_tree
->lock
);
2165 } while (ret
== -EEXIST
);
2166 free_extent_map(hole_em
);
2168 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
2169 &BTRFS_I(inode
)->runtime_flags
);
2176 * Find a hole extent on given inode and change start/len to the end of hole
2177 * extent.(hole/vacuum extent whose em->start <= start &&
2178 * em->start + em->len > start)
2179 * When a hole extent is found, return 1 and modify start/len.
2181 static int find_first_non_hole(struct inode
*inode
, u64
*start
, u64
*len
)
2183 struct extent_map
*em
;
2186 em
= btrfs_get_extent(inode
, NULL
, 0, *start
, *len
, 0);
2187 if (IS_ERR_OR_NULL(em
)) {
2195 /* Hole or vacuum extent(only exists in no-hole mode) */
2196 if (em
->block_start
== EXTENT_MAP_HOLE
) {
2198 *len
= em
->start
+ em
->len
> *start
+ *len
?
2199 0 : *start
+ *len
- em
->start
- em
->len
;
2200 *start
= em
->start
+ em
->len
;
2202 free_extent_map(em
);
2206 static int btrfs_punch_hole(struct inode
*inode
, loff_t offset
, loff_t len
)
2208 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2209 struct extent_state
*cached_state
= NULL
;
2210 struct btrfs_path
*path
;
2211 struct btrfs_block_rsv
*rsv
;
2212 struct btrfs_trans_handle
*trans
;
2217 u64 orig_start
= offset
;
2219 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
2225 bool no_holes
= btrfs_fs_incompat(root
->fs_info
, NO_HOLES
);
2228 ret
= btrfs_wait_ordered_range(inode
, offset
, len
);
2232 mutex_lock(&inode
->i_mutex
);
2233 ino_size
= round_up(inode
->i_size
, PAGE_CACHE_SIZE
);
2234 ret
= find_first_non_hole(inode
, &offset
, &len
);
2236 goto out_only_mutex
;
2238 /* Already in a large hole */
2240 goto out_only_mutex
;
2243 lockstart
= round_up(offset
, BTRFS_I(inode
)->root
->sectorsize
);
2244 lockend
= round_down(offset
+ len
,
2245 BTRFS_I(inode
)->root
->sectorsize
) - 1;
2246 same_page
= ((offset
>> PAGE_CACHE_SHIFT
) ==
2247 ((offset
+ len
- 1) >> PAGE_CACHE_SHIFT
));
2250 * We needn't truncate any page which is beyond the end of the file
2251 * because we are sure there is no data there.
2254 * Only do this if we are in the same page and we aren't doing the
2257 if (same_page
&& len
< PAGE_CACHE_SIZE
) {
2258 if (offset
< ino_size
)
2259 ret
= btrfs_truncate_page(inode
, offset
, len
, 0);
2260 goto out_only_mutex
;
2263 /* zero back part of the first page */
2264 if (offset
< ino_size
) {
2265 ret
= btrfs_truncate_page(inode
, offset
, 0, 0);
2267 mutex_unlock(&inode
->i_mutex
);
2272 /* Check the aligned pages after the first unaligned page,
2273 * if offset != orig_start, which means the first unaligned page
2274 * including serveral following pages are already in holes,
2275 * the extra check can be skipped */
2276 if (offset
== orig_start
) {
2277 /* after truncate page, check hole again */
2278 len
= offset
+ len
- lockstart
;
2280 ret
= find_first_non_hole(inode
, &offset
, &len
);
2282 goto out_only_mutex
;
2285 goto out_only_mutex
;
2290 /* Check the tail unaligned part is in a hole */
2291 tail_start
= lockend
+ 1;
2292 tail_len
= offset
+ len
- tail_start
;
2294 ret
= find_first_non_hole(inode
, &tail_start
, &tail_len
);
2295 if (unlikely(ret
< 0))
2296 goto out_only_mutex
;
2298 /* zero the front end of the last page */
2299 if (tail_start
+ tail_len
< ino_size
) {
2300 ret
= btrfs_truncate_page(inode
,
2301 tail_start
+ tail_len
, 0, 1);
2303 goto out_only_mutex
;
2308 if (lockend
< lockstart
) {
2309 mutex_unlock(&inode
->i_mutex
);
2314 struct btrfs_ordered_extent
*ordered
;
2316 truncate_pagecache_range(inode
, lockstart
, lockend
);
2318 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2320 ordered
= btrfs_lookup_first_ordered_extent(inode
, lockend
);
2323 * We need to make sure we have no ordered extents in this range
2324 * and nobody raced in and read a page in this range, if we did
2325 * we need to try again.
2328 (ordered
->file_offset
+ ordered
->len
<= lockstart
||
2329 ordered
->file_offset
> lockend
)) &&
2330 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)) {
2332 btrfs_put_ordered_extent(ordered
);
2336 btrfs_put_ordered_extent(ordered
);
2337 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
,
2338 lockend
, &cached_state
, GFP_NOFS
);
2339 ret
= btrfs_wait_ordered_range(inode
, lockstart
,
2340 lockend
- lockstart
+ 1);
2342 mutex_unlock(&inode
->i_mutex
);
2347 path
= btrfs_alloc_path();
2353 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2358 rsv
->size
= btrfs_calc_trunc_metadata_size(root
, 1);
2362 * 1 - update the inode
2363 * 1 - removing the extents in the range
2364 * 1 - adding the hole extent if no_holes isn't set
2366 rsv_count
= no_holes
? 2 : 3;
2367 trans
= btrfs_start_transaction(root
, rsv_count
);
2368 if (IS_ERR(trans
)) {
2369 err
= PTR_ERR(trans
);
2373 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
2376 trans
->block_rsv
= rsv
;
2378 cur_offset
= lockstart
;
2379 len
= lockend
- cur_offset
;
2380 while (cur_offset
< lockend
) {
2381 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
2382 cur_offset
, lockend
+ 1,
2383 &drop_end
, 1, 0, 0, NULL
);
2387 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2389 if (cur_offset
< ino_size
) {
2390 ret
= fill_holes(trans
, inode
, path
, cur_offset
,
2398 cur_offset
= drop_end
;
2400 ret
= btrfs_update_inode(trans
, root
, inode
);
2406 btrfs_end_transaction(trans
, root
);
2407 btrfs_btree_balance_dirty(root
);
2409 trans
= btrfs_start_transaction(root
, rsv_count
);
2410 if (IS_ERR(trans
)) {
2411 ret
= PTR_ERR(trans
);
2416 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
2418 BUG_ON(ret
); /* shouldn't happen */
2419 trans
->block_rsv
= rsv
;
2421 ret
= find_first_non_hole(inode
, &cur_offset
, &len
);
2422 if (unlikely(ret
< 0))
2435 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2437 * Don't insert file hole extent item if it's for a range beyond eof
2438 * (because it's useless) or if it represents a 0 bytes range (when
2439 * cur_offset == drop_end).
2441 if (cur_offset
< ino_size
&& cur_offset
< drop_end
) {
2442 ret
= fill_holes(trans
, inode
, path
, cur_offset
, drop_end
);
2453 inode_inc_iversion(inode
);
2454 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2456 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2457 ret
= btrfs_update_inode(trans
, root
, inode
);
2458 btrfs_end_transaction(trans
, root
);
2459 btrfs_btree_balance_dirty(root
);
2461 btrfs_free_path(path
);
2462 btrfs_free_block_rsv(root
, rsv
);
2464 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2465 &cached_state
, GFP_NOFS
);
2467 mutex_unlock(&inode
->i_mutex
);
2473 static long btrfs_fallocate(struct file
*file
, int mode
,
2474 loff_t offset
, loff_t len
)
2476 struct inode
*inode
= file_inode(file
);
2477 struct extent_state
*cached_state
= NULL
;
2478 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2485 struct extent_map
*em
;
2486 int blocksize
= BTRFS_I(inode
)->root
->sectorsize
;
2489 alloc_start
= round_down(offset
, blocksize
);
2490 alloc_end
= round_up(offset
+ len
, blocksize
);
2492 /* Make sure we aren't being give some crap mode */
2493 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2496 if (mode
& FALLOC_FL_PUNCH_HOLE
)
2497 return btrfs_punch_hole(inode
, offset
, len
);
2500 * Make sure we have enough space before we do the
2503 ret
= btrfs_check_data_free_space(inode
, alloc_end
- alloc_start
);
2506 if (root
->fs_info
->quota_enabled
) {
2507 ret
= btrfs_qgroup_reserve(root
, alloc_end
- alloc_start
);
2509 goto out_reserve_fail
;
2512 mutex_lock(&inode
->i_mutex
);
2513 ret
= inode_newsize_ok(inode
, alloc_end
);
2517 if (alloc_start
> inode
->i_size
) {
2518 ret
= btrfs_cont_expand(inode
, i_size_read(inode
),
2524 * If we are fallocating from the end of the file onward we
2525 * need to zero out the end of the page if i_size lands in the
2528 ret
= btrfs_truncate_page(inode
, inode
->i_size
, 0, 0);
2534 * wait for ordered IO before we have any locks. We'll loop again
2535 * below with the locks held.
2537 ret
= btrfs_wait_ordered_range(inode
, alloc_start
,
2538 alloc_end
- alloc_start
);
2542 locked_end
= alloc_end
- 1;
2544 struct btrfs_ordered_extent
*ordered
;
2546 /* the extent lock is ordered inside the running
2549 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, alloc_start
,
2550 locked_end
, 0, &cached_state
);
2551 ordered
= btrfs_lookup_first_ordered_extent(inode
,
2554 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
2555 ordered
->file_offset
< alloc_end
) {
2556 btrfs_put_ordered_extent(ordered
);
2557 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
2558 alloc_start
, locked_end
,
2559 &cached_state
, GFP_NOFS
);
2561 * we can't wait on the range with the transaction
2562 * running or with the extent lock held
2564 ret
= btrfs_wait_ordered_range(inode
, alloc_start
,
2565 alloc_end
- alloc_start
);
2570 btrfs_put_ordered_extent(ordered
);
2575 cur_offset
= alloc_start
;
2579 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2580 alloc_end
- cur_offset
, 0);
2581 if (IS_ERR_OR_NULL(em
)) {
2588 last_byte
= min(extent_map_end(em
), alloc_end
);
2589 actual_end
= min_t(u64
, extent_map_end(em
), offset
+ len
);
2590 last_byte
= ALIGN(last_byte
, blocksize
);
2592 if (em
->block_start
== EXTENT_MAP_HOLE
||
2593 (cur_offset
>= inode
->i_size
&&
2594 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
2595 ret
= btrfs_prealloc_file_range(inode
, mode
, cur_offset
,
2596 last_byte
- cur_offset
,
2597 1 << inode
->i_blkbits
,
2602 free_extent_map(em
);
2605 } else if (actual_end
> inode
->i_size
&&
2606 !(mode
& FALLOC_FL_KEEP_SIZE
)) {
2608 * We didn't need to allocate any more space, but we
2609 * still extended the size of the file so we need to
2612 inode
->i_ctime
= CURRENT_TIME
;
2613 i_size_write(inode
, actual_end
);
2614 btrfs_ordered_update_i_size(inode
, actual_end
, NULL
);
2616 free_extent_map(em
);
2618 cur_offset
= last_byte
;
2619 if (cur_offset
>= alloc_end
) {
2624 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
2625 &cached_state
, GFP_NOFS
);
2627 mutex_unlock(&inode
->i_mutex
);
2628 if (root
->fs_info
->quota_enabled
)
2629 btrfs_qgroup_free(root
, alloc_end
- alloc_start
);
2631 /* Let go of our reservation. */
2632 btrfs_free_reserved_data_space(inode
, alloc_end
- alloc_start
);
2636 static int find_desired_extent(struct inode
*inode
, loff_t
*offset
, int whence
)
2638 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2639 struct extent_map
*em
= NULL
;
2640 struct extent_state
*cached_state
= NULL
;
2641 u64 lockstart
= *offset
;
2642 u64 lockend
= i_size_read(inode
);
2643 u64 start
= *offset
;
2644 u64 len
= i_size_read(inode
);
2647 lockend
= max_t(u64
, root
->sectorsize
, lockend
);
2648 if (lockend
<= lockstart
)
2649 lockend
= lockstart
+ root
->sectorsize
;
2652 len
= lockend
- lockstart
+ 1;
2654 len
= max_t(u64
, len
, root
->sectorsize
);
2655 if (inode
->i_size
== 0)
2658 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
, 0,
2661 while (start
< inode
->i_size
) {
2662 em
= btrfs_get_extent_fiemap(inode
, NULL
, 0, start
, len
, 0);
2669 if (whence
== SEEK_HOLE
&&
2670 (em
->block_start
== EXTENT_MAP_HOLE
||
2671 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)))
2673 else if (whence
== SEEK_DATA
&&
2674 (em
->block_start
!= EXTENT_MAP_HOLE
&&
2675 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)))
2678 start
= em
->start
+ em
->len
;
2679 free_extent_map(em
);
2683 free_extent_map(em
);
2685 if (whence
== SEEK_DATA
&& start
>= inode
->i_size
)
2688 *offset
= min_t(loff_t
, start
, inode
->i_size
);
2690 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2691 &cached_state
, GFP_NOFS
);
2695 static loff_t
btrfs_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2697 struct inode
*inode
= file
->f_mapping
->host
;
2700 mutex_lock(&inode
->i_mutex
);
2704 offset
= generic_file_llseek(file
, offset
, whence
);
2708 if (offset
>= i_size_read(inode
)) {
2709 mutex_unlock(&inode
->i_mutex
);
2713 ret
= find_desired_extent(inode
, &offset
, whence
);
2715 mutex_unlock(&inode
->i_mutex
);
2720 offset
= vfs_setpos(file
, offset
, inode
->i_sb
->s_maxbytes
);
2722 mutex_unlock(&inode
->i_mutex
);
2726 const struct file_operations btrfs_file_operations
= {
2727 .llseek
= btrfs_file_llseek
,
2728 .read
= new_sync_read
,
2729 .write
= new_sync_write
,
2730 .read_iter
= generic_file_read_iter
,
2731 .splice_read
= generic_file_splice_read
,
2732 .write_iter
= btrfs_file_write_iter
,
2733 .mmap
= btrfs_file_mmap
,
2734 .open
= generic_file_open
,
2735 .release
= btrfs_release_file
,
2736 .fsync
= btrfs_sync_file
,
2737 .fallocate
= btrfs_fallocate
,
2738 .unlocked_ioctl
= btrfs_ioctl
,
2739 #ifdef CONFIG_COMPAT
2740 .compat_ioctl
= btrfs_ioctl
,
2744 void btrfs_auto_defrag_exit(void)
2746 if (btrfs_inode_defrag_cachep
)
2747 kmem_cache_destroy(btrfs_inode_defrag_cachep
);
2750 int btrfs_auto_defrag_init(void)
2752 btrfs_inode_defrag_cachep
= kmem_cache_create("btrfs_inode_defrag",
2753 sizeof(struct inode_defrag
), 0,
2754 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
2756 if (!btrfs_inode_defrag_cachep
)