2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/falloc.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/slab.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 static struct kmem_cache
*btrfs_inode_defrag_cachep
;
46 * when auto defrag is enabled we
47 * queue up these defrag structs to remember which
48 * inodes need defragging passes
51 struct rb_node rb_node
;
55 * transid where the defrag was added, we search for
56 * extents newer than this
63 /* last offset we were able to defrag */
66 /* if we've wrapped around back to zero once already */
70 static int __compare_inode_defrag(struct inode_defrag
*defrag1
,
71 struct inode_defrag
*defrag2
)
73 if (defrag1
->root
> defrag2
->root
)
75 else if (defrag1
->root
< defrag2
->root
)
77 else if (defrag1
->ino
> defrag2
->ino
)
79 else if (defrag1
->ino
< defrag2
->ino
)
85 /* pop a record for an inode into the defrag tree. The lock
86 * must be held already
88 * If you're inserting a record for an older transid than an
89 * existing record, the transid already in the tree is lowered
91 * If an existing record is found the defrag item you
94 static int __btrfs_add_inode_defrag(struct inode
*inode
,
95 struct inode_defrag
*defrag
)
97 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
98 struct inode_defrag
*entry
;
100 struct rb_node
*parent
= NULL
;
103 p
= &root
->fs_info
->defrag_inodes
.rb_node
;
106 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
108 ret
= __compare_inode_defrag(defrag
, entry
);
110 p
= &parent
->rb_left
;
112 p
= &parent
->rb_right
;
114 /* if we're reinserting an entry for
115 * an old defrag run, make sure to
116 * lower the transid of our existing record
118 if (defrag
->transid
< entry
->transid
)
119 entry
->transid
= defrag
->transid
;
120 if (defrag
->last_offset
> entry
->last_offset
)
121 entry
->last_offset
= defrag
->last_offset
;
125 set_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
);
126 rb_link_node(&defrag
->rb_node
, parent
, p
);
127 rb_insert_color(&defrag
->rb_node
, &root
->fs_info
->defrag_inodes
);
131 static inline int __need_auto_defrag(struct btrfs_root
*root
)
133 if (!btrfs_test_opt(root
, AUTO_DEFRAG
))
136 if (btrfs_fs_closing(root
->fs_info
))
143 * insert a defrag record for this inode if auto defrag is
146 int btrfs_add_inode_defrag(struct btrfs_trans_handle
*trans
,
149 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
150 struct inode_defrag
*defrag
;
154 if (!__need_auto_defrag(root
))
157 if (test_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
))
161 transid
= trans
->transid
;
163 transid
= BTRFS_I(inode
)->root
->last_trans
;
165 defrag
= kmem_cache_zalloc(btrfs_inode_defrag_cachep
, GFP_NOFS
);
169 defrag
->ino
= btrfs_ino(inode
);
170 defrag
->transid
= transid
;
171 defrag
->root
= root
->root_key
.objectid
;
173 spin_lock(&root
->fs_info
->defrag_inodes_lock
);
174 if (!test_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
)) {
176 * If we set IN_DEFRAG flag and evict the inode from memory,
177 * and then re-read this inode, this new inode doesn't have
178 * IN_DEFRAG flag. At the case, we may find the existed defrag.
180 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
182 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
184 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
186 spin_unlock(&root
->fs_info
->defrag_inodes_lock
);
191 * Requeue the defrag object. If there is a defrag object that points to
192 * the same inode in the tree, we will merge them together (by
193 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
195 void btrfs_requeue_inode_defrag(struct inode
*inode
,
196 struct inode_defrag
*defrag
)
198 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
201 if (!__need_auto_defrag(root
))
205 * Here we don't check the IN_DEFRAG flag, because we need merge
208 spin_lock(&root
->fs_info
->defrag_inodes_lock
);
209 ret
= __btrfs_add_inode_defrag(inode
, defrag
);
210 spin_unlock(&root
->fs_info
->defrag_inodes_lock
);
215 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
219 * pick the defragable inode that we want, if it doesn't exist, we will get
222 static struct inode_defrag
*
223 btrfs_pick_defrag_inode(struct btrfs_fs_info
*fs_info
, u64 root
, u64 ino
)
225 struct inode_defrag
*entry
= NULL
;
226 struct inode_defrag tmp
;
228 struct rb_node
*parent
= NULL
;
234 spin_lock(&fs_info
->defrag_inodes_lock
);
235 p
= fs_info
->defrag_inodes
.rb_node
;
238 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
240 ret
= __compare_inode_defrag(&tmp
, entry
);
244 p
= parent
->rb_right
;
249 if (parent
&& __compare_inode_defrag(&tmp
, entry
) > 0) {
250 parent
= rb_next(parent
);
252 entry
= rb_entry(parent
, struct inode_defrag
, rb_node
);
258 rb_erase(parent
, &fs_info
->defrag_inodes
);
259 spin_unlock(&fs_info
->defrag_inodes_lock
);
263 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info
*fs_info
)
265 struct inode_defrag
*defrag
;
266 struct rb_node
*node
;
268 spin_lock(&fs_info
->defrag_inodes_lock
);
269 node
= rb_first(&fs_info
->defrag_inodes
);
271 rb_erase(node
, &fs_info
->defrag_inodes
);
272 defrag
= rb_entry(node
, struct inode_defrag
, rb_node
);
273 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
275 if (need_resched()) {
276 spin_unlock(&fs_info
->defrag_inodes_lock
);
278 spin_lock(&fs_info
->defrag_inodes_lock
);
281 node
= rb_first(&fs_info
->defrag_inodes
);
283 spin_unlock(&fs_info
->defrag_inodes_lock
);
286 #define BTRFS_DEFRAG_BATCH 1024
288 static int __btrfs_run_defrag_inode(struct btrfs_fs_info
*fs_info
,
289 struct inode_defrag
*defrag
)
291 struct btrfs_root
*inode_root
;
293 struct btrfs_key key
;
294 struct btrfs_ioctl_defrag_range_args range
;
300 key
.objectid
= defrag
->root
;
301 btrfs_set_key_type(&key
, BTRFS_ROOT_ITEM_KEY
);
302 key
.offset
= (u64
)-1;
304 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
306 inode_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
307 if (IS_ERR(inode_root
)) {
308 ret
= PTR_ERR(inode_root
);
311 if (btrfs_root_refs(&inode_root
->root_item
) == 0) {
316 key
.objectid
= defrag
->ino
;
317 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
319 inode
= btrfs_iget(fs_info
->sb
, &key
, inode_root
, NULL
);
321 ret
= PTR_ERR(inode
);
324 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
326 /* do a chunk of defrag */
327 clear_bit(BTRFS_INODE_IN_DEFRAG
, &BTRFS_I(inode
)->runtime_flags
);
328 memset(&range
, 0, sizeof(range
));
330 range
.start
= defrag
->last_offset
;
332 sb_start_write(fs_info
->sb
);
333 num_defrag
= btrfs_defrag_file(inode
, NULL
, &range
, defrag
->transid
,
335 sb_end_write(fs_info
->sb
);
337 * if we filled the whole defrag batch, there
338 * must be more work to do. Queue this defrag
341 if (num_defrag
== BTRFS_DEFRAG_BATCH
) {
342 defrag
->last_offset
= range
.start
;
343 btrfs_requeue_inode_defrag(inode
, defrag
);
344 } else if (defrag
->last_offset
&& !defrag
->cycled
) {
346 * we didn't fill our defrag batch, but
347 * we didn't start at zero. Make sure we loop
348 * around to the start of the file.
350 defrag
->last_offset
= 0;
352 btrfs_requeue_inode_defrag(inode
, defrag
);
354 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
360 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
361 kmem_cache_free(btrfs_inode_defrag_cachep
, defrag
);
366 * run through the list of inodes in the FS that need
369 int btrfs_run_defrag_inodes(struct btrfs_fs_info
*fs_info
)
371 struct inode_defrag
*defrag
;
373 u64 root_objectid
= 0;
375 atomic_inc(&fs_info
->defrag_running
);
377 if (!__need_auto_defrag(fs_info
->tree_root
))
380 /* find an inode to defrag */
381 defrag
= btrfs_pick_defrag_inode(fs_info
, root_objectid
,
384 if (root_objectid
|| first_ino
) {
393 first_ino
= defrag
->ino
+ 1;
394 root_objectid
= defrag
->root
;
396 __btrfs_run_defrag_inode(fs_info
, defrag
);
398 atomic_dec(&fs_info
->defrag_running
);
401 * during unmount, we use the transaction_wait queue to
402 * wait for the defragger to stop
404 wake_up(&fs_info
->transaction_wait
);
408 /* simple helper to fault in pages and copy. This should go away
409 * and be replaced with calls into generic code.
411 static noinline
int btrfs_copy_from_user(loff_t pos
, int num_pages
,
413 struct page
**prepared_pages
,
417 size_t total_copied
= 0;
419 int offset
= pos
& (PAGE_CACHE_SIZE
- 1);
421 while (write_bytes
> 0) {
422 size_t count
= min_t(size_t,
423 PAGE_CACHE_SIZE
- offset
, write_bytes
);
424 struct page
*page
= prepared_pages
[pg
];
426 * Copy data from userspace to the current page
428 * Disable pagefault to avoid recursive lock since
429 * the pages are already locked
432 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, count
);
435 /* Flush processor's dcache for this page */
436 flush_dcache_page(page
);
439 * if we get a partial write, we can end up with
440 * partially up to date pages. These add
441 * a lot of complexity, so make sure they don't
442 * happen by forcing this copy to be retried.
444 * The rest of the btrfs_file_write code will fall
445 * back to page at a time copies after we return 0.
447 if (!PageUptodate(page
) && copied
< count
)
450 iov_iter_advance(i
, copied
);
451 write_bytes
-= copied
;
452 total_copied
+= copied
;
454 /* Return to btrfs_file_aio_write to fault page */
455 if (unlikely(copied
== 0))
458 if (unlikely(copied
< PAGE_CACHE_SIZE
- offset
)) {
469 * unlocks pages after btrfs_file_write is done with them
471 void btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
474 for (i
= 0; i
< num_pages
; i
++) {
475 /* page checked is some magic around finding pages that
476 * have been modified without going through btrfs_set_page_dirty
479 ClearPageChecked(pages
[i
]);
480 unlock_page(pages
[i
]);
481 mark_page_accessed(pages
[i
]);
482 page_cache_release(pages
[i
]);
487 * after copy_from_user, pages need to be dirtied and we need to make
488 * sure holes are created between the current EOF and the start of
489 * any next extents (if required).
491 * this also makes the decision about creating an inline extent vs
492 * doing real data extents, marking pages dirty and delalloc as required.
494 int btrfs_dirty_pages(struct btrfs_root
*root
, struct inode
*inode
,
495 struct page
**pages
, size_t num_pages
,
496 loff_t pos
, size_t write_bytes
,
497 struct extent_state
**cached
)
503 u64 end_of_last_block
;
504 u64 end_pos
= pos
+ write_bytes
;
505 loff_t isize
= i_size_read(inode
);
507 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
508 num_bytes
= (write_bytes
+ pos
- start_pos
+
509 root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
511 end_of_last_block
= start_pos
+ num_bytes
- 1;
512 err
= btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
,
517 for (i
= 0; i
< num_pages
; i
++) {
518 struct page
*p
= pages
[i
];
525 * we've only changed i_size in ram, and we haven't updated
526 * the disk i_size. There is no need to log the inode
530 i_size_write(inode
, end_pos
);
535 * this drops all the extents in the cache that intersect the range
536 * [start, end]. Existing extents are split as required.
538 void btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
,
541 struct extent_map
*em
;
542 struct extent_map
*split
= NULL
;
543 struct extent_map
*split2
= NULL
;
544 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
545 u64 len
= end
- start
+ 1;
552 WARN_ON(end
< start
);
553 if (end
== (u64
)-1) {
561 split
= alloc_extent_map();
563 split2
= alloc_extent_map();
564 if (!split
|| !split2
)
567 write_lock(&em_tree
->lock
);
568 em
= lookup_extent_mapping(em_tree
, start
, len
);
570 write_unlock(&em_tree
->lock
);
574 gen
= em
->generation
;
575 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
576 if (testend
&& em
->start
+ em
->len
>= start
+ len
) {
578 write_unlock(&em_tree
->lock
);
581 start
= em
->start
+ em
->len
;
583 len
= start
+ len
- (em
->start
+ em
->len
);
585 write_unlock(&em_tree
->lock
);
588 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
589 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
590 remove_extent_mapping(em_tree
, em
);
594 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
596 split
->start
= em
->start
;
597 split
->len
= start
- em
->start
;
598 split
->orig_start
= em
->orig_start
;
599 split
->block_start
= em
->block_start
;
602 split
->block_len
= em
->block_len
;
604 split
->block_len
= split
->len
;
605 split
->orig_block_len
= max(split
->block_len
,
607 split
->generation
= gen
;
608 split
->bdev
= em
->bdev
;
609 split
->flags
= flags
;
610 split
->compress_type
= em
->compress_type
;
611 ret
= add_extent_mapping(em_tree
, split
);
612 BUG_ON(ret
); /* Logic error */
613 list_move(&split
->list
, &em_tree
->modified_extents
);
614 free_extent_map(split
);
618 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
619 testend
&& em
->start
+ em
->len
> start
+ len
) {
620 u64 diff
= start
+ len
- em
->start
;
622 split
->start
= start
+ len
;
623 split
->len
= em
->start
+ em
->len
- (start
+ len
);
624 split
->bdev
= em
->bdev
;
625 split
->flags
= flags
;
626 split
->compress_type
= em
->compress_type
;
627 split
->generation
= gen
;
628 split
->orig_block_len
= max(em
->block_len
,
632 split
->block_len
= em
->block_len
;
633 split
->block_start
= em
->block_start
;
634 split
->orig_start
= em
->orig_start
;
636 split
->block_len
= split
->len
;
637 split
->block_start
= em
->block_start
+ diff
;
638 split
->orig_start
= em
->orig_start
;
641 ret
= add_extent_mapping(em_tree
, split
);
642 BUG_ON(ret
); /* Logic error */
643 list_move(&split
->list
, &em_tree
->modified_extents
);
644 free_extent_map(split
);
648 write_unlock(&em_tree
->lock
);
652 /* once for the tree*/
656 free_extent_map(split
);
658 free_extent_map(split2
);
662 * this is very complex, but the basic idea is to drop all extents
663 * in the range start - end. hint_block is filled in with a block number
664 * that would be a good hint to the block allocator for this file.
666 * If an extent intersects the range but is not entirely inside the range
667 * it is either truncated or split. Anything entirely inside the range
668 * is deleted from the tree.
670 int __btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
671 struct btrfs_root
*root
, struct inode
*inode
,
672 struct btrfs_path
*path
, u64 start
, u64 end
,
673 u64
*drop_end
, int drop_cache
)
675 struct extent_buffer
*leaf
;
676 struct btrfs_file_extent_item
*fi
;
677 struct btrfs_key key
;
678 struct btrfs_key new_key
;
679 u64 ino
= btrfs_ino(inode
);
680 u64 search_start
= start
;
683 u64 extent_offset
= 0;
690 int modify_tree
= -1;
691 int update_refs
= (root
->ref_cows
|| root
== root
->fs_info
->tree_root
);
695 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
697 if (start
>= BTRFS_I(inode
)->disk_i_size
)
702 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
703 search_start
, modify_tree
);
706 if (ret
> 0 && path
->slots
[0] > 0 && search_start
== start
) {
707 leaf
= path
->nodes
[0];
708 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0] - 1);
709 if (key
.objectid
== ino
&&
710 key
.type
== BTRFS_EXTENT_DATA_KEY
)
715 leaf
= path
->nodes
[0];
716 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
718 ret
= btrfs_next_leaf(root
, path
);
725 leaf
= path
->nodes
[0];
729 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
730 if (key
.objectid
> ino
||
731 key
.type
> BTRFS_EXTENT_DATA_KEY
|| key
.offset
>= end
)
734 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
735 struct btrfs_file_extent_item
);
736 extent_type
= btrfs_file_extent_type(leaf
, fi
);
738 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
739 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
740 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
741 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
742 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
743 extent_end
= key
.offset
+
744 btrfs_file_extent_num_bytes(leaf
, fi
);
745 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
746 extent_end
= key
.offset
+
747 btrfs_file_extent_inline_len(leaf
, fi
);
750 extent_end
= search_start
;
753 if (extent_end
<= search_start
) {
759 search_start
= max(key
.offset
, start
);
760 if (recow
|| !modify_tree
) {
762 btrfs_release_path(path
);
767 * | - range to drop - |
768 * | -------- extent -------- |
770 if (start
> key
.offset
&& end
< extent_end
) {
772 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
774 memcpy(&new_key
, &key
, sizeof(new_key
));
775 new_key
.offset
= start
;
776 ret
= btrfs_duplicate_item(trans
, root
, path
,
778 if (ret
== -EAGAIN
) {
779 btrfs_release_path(path
);
785 leaf
= path
->nodes
[0];
786 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
787 struct btrfs_file_extent_item
);
788 btrfs_set_file_extent_num_bytes(leaf
, fi
,
791 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
792 struct btrfs_file_extent_item
);
794 extent_offset
+= start
- key
.offset
;
795 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
796 btrfs_set_file_extent_num_bytes(leaf
, fi
,
798 btrfs_mark_buffer_dirty(leaf
);
800 if (update_refs
&& disk_bytenr
> 0) {
801 ret
= btrfs_inc_extent_ref(trans
, root
,
802 disk_bytenr
, num_bytes
, 0,
803 root
->root_key
.objectid
,
805 start
- extent_offset
, 0);
806 BUG_ON(ret
); /* -ENOMEM */
811 * | ---- range to drop ----- |
812 * | -------- extent -------- |
814 if (start
<= key
.offset
&& end
< extent_end
) {
815 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
817 memcpy(&new_key
, &key
, sizeof(new_key
));
818 new_key
.offset
= end
;
819 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
821 extent_offset
+= end
- key
.offset
;
822 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
823 btrfs_set_file_extent_num_bytes(leaf
, fi
,
825 btrfs_mark_buffer_dirty(leaf
);
826 if (update_refs
&& disk_bytenr
> 0)
827 inode_sub_bytes(inode
, end
- key
.offset
);
831 search_start
= extent_end
;
833 * | ---- range to drop ----- |
834 * | -------- extent -------- |
836 if (start
> key
.offset
&& end
>= extent_end
) {
838 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
840 btrfs_set_file_extent_num_bytes(leaf
, fi
,
842 btrfs_mark_buffer_dirty(leaf
);
843 if (update_refs
&& disk_bytenr
> 0)
844 inode_sub_bytes(inode
, extent_end
- start
);
845 if (end
== extent_end
)
853 * | ---- range to drop ----- |
854 * | ------ extent ------ |
856 if (start
<= key
.offset
&& end
>= extent_end
) {
858 del_slot
= path
->slots
[0];
861 BUG_ON(del_slot
+ del_nr
!= path
->slots
[0]);
866 extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
867 inode_sub_bytes(inode
,
868 extent_end
- key
.offset
);
869 extent_end
= ALIGN(extent_end
,
871 } else if (update_refs
&& disk_bytenr
> 0) {
872 ret
= btrfs_free_extent(trans
, root
,
873 disk_bytenr
, num_bytes
, 0,
874 root
->root_key
.objectid
,
875 key
.objectid
, key
.offset
-
877 BUG_ON(ret
); /* -ENOMEM */
878 inode_sub_bytes(inode
,
879 extent_end
- key
.offset
);
882 if (end
== extent_end
)
885 if (path
->slots
[0] + 1 < btrfs_header_nritems(leaf
)) {
890 ret
= btrfs_del_items(trans
, root
, path
, del_slot
,
893 btrfs_abort_transaction(trans
, root
, ret
);
900 btrfs_release_path(path
);
907 if (!ret
&& del_nr
> 0) {
908 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
910 btrfs_abort_transaction(trans
, root
, ret
);
914 *drop_end
= found
? min(end
, extent_end
) : end
;
915 btrfs_release_path(path
);
919 int btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
920 struct btrfs_root
*root
, struct inode
*inode
, u64 start
,
921 u64 end
, int drop_cache
)
923 struct btrfs_path
*path
;
926 path
= btrfs_alloc_path();
929 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, start
, end
, NULL
,
931 btrfs_free_path(path
);
935 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
936 u64 objectid
, u64 bytenr
, u64 orig_offset
,
937 u64
*start
, u64
*end
)
939 struct btrfs_file_extent_item
*fi
;
940 struct btrfs_key key
;
943 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
946 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
947 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
950 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
951 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
952 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
953 btrfs_file_extent_offset(leaf
, fi
) != key
.offset
- orig_offset
||
954 btrfs_file_extent_compression(leaf
, fi
) ||
955 btrfs_file_extent_encryption(leaf
, fi
) ||
956 btrfs_file_extent_other_encoding(leaf
, fi
))
959 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
960 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
969 * Mark extent in the range start - end as written.
971 * This changes extent type from 'pre-allocated' to 'regular'. If only
972 * part of extent is marked as written, the extent will be split into
975 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
976 struct inode
*inode
, u64 start
, u64 end
)
978 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
979 struct extent_buffer
*leaf
;
980 struct btrfs_path
*path
;
981 struct btrfs_file_extent_item
*fi
;
982 struct btrfs_key key
;
983 struct btrfs_key new_key
;
995 u64 ino
= btrfs_ino(inode
);
997 path
= btrfs_alloc_path();
1004 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1007 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1010 if (ret
> 0 && path
->slots
[0] > 0)
1013 leaf
= path
->nodes
[0];
1014 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1015 BUG_ON(key
.objectid
!= ino
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
);
1016 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1017 struct btrfs_file_extent_item
);
1018 BUG_ON(btrfs_file_extent_type(leaf
, fi
) !=
1019 BTRFS_FILE_EXTENT_PREALLOC
);
1020 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
1021 BUG_ON(key
.offset
> start
|| extent_end
< end
);
1023 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1024 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1025 orig_offset
= key
.offset
- btrfs_file_extent_offset(leaf
, fi
);
1026 memcpy(&new_key
, &key
, sizeof(new_key
));
1028 if (start
== key
.offset
&& end
< extent_end
) {
1031 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1032 ino
, bytenr
, orig_offset
,
1033 &other_start
, &other_end
)) {
1034 new_key
.offset
= end
;
1035 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
1036 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1037 struct btrfs_file_extent_item
);
1038 btrfs_set_file_extent_generation(leaf
, fi
,
1040 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1042 btrfs_set_file_extent_offset(leaf
, fi
,
1044 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1045 struct btrfs_file_extent_item
);
1046 btrfs_set_file_extent_generation(leaf
, fi
,
1048 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1050 btrfs_mark_buffer_dirty(leaf
);
1055 if (start
> key
.offset
&& end
== extent_end
) {
1058 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1059 ino
, bytenr
, orig_offset
,
1060 &other_start
, &other_end
)) {
1061 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1062 struct btrfs_file_extent_item
);
1063 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1064 start
- key
.offset
);
1065 btrfs_set_file_extent_generation(leaf
, fi
,
1068 new_key
.offset
= start
;
1069 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
1071 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1072 struct btrfs_file_extent_item
);
1073 btrfs_set_file_extent_generation(leaf
, fi
,
1075 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1077 btrfs_set_file_extent_offset(leaf
, fi
,
1078 start
- orig_offset
);
1079 btrfs_mark_buffer_dirty(leaf
);
1084 while (start
> key
.offset
|| end
< extent_end
) {
1085 if (key
.offset
== start
)
1088 new_key
.offset
= split
;
1089 ret
= btrfs_duplicate_item(trans
, root
, path
, &new_key
);
1090 if (ret
== -EAGAIN
) {
1091 btrfs_release_path(path
);
1095 btrfs_abort_transaction(trans
, root
, ret
);
1099 leaf
= path
->nodes
[0];
1100 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
1101 struct btrfs_file_extent_item
);
1102 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1103 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1104 split
- key
.offset
);
1106 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1107 struct btrfs_file_extent_item
);
1109 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1110 btrfs_set_file_extent_offset(leaf
, fi
, split
- orig_offset
);
1111 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1112 extent_end
- split
);
1113 btrfs_mark_buffer_dirty(leaf
);
1115 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
, 0,
1116 root
->root_key
.objectid
,
1117 ino
, orig_offset
, 0);
1118 BUG_ON(ret
); /* -ENOMEM */
1120 if (split
== start
) {
1123 BUG_ON(start
!= key
.offset
);
1132 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
1133 ino
, bytenr
, orig_offset
,
1134 &other_start
, &other_end
)) {
1136 btrfs_release_path(path
);
1139 extent_end
= other_end
;
1140 del_slot
= path
->slots
[0] + 1;
1142 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1143 0, root
->root_key
.objectid
,
1144 ino
, orig_offset
, 0);
1145 BUG_ON(ret
); /* -ENOMEM */
1149 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
1150 ino
, bytenr
, orig_offset
,
1151 &other_start
, &other_end
)) {
1153 btrfs_release_path(path
);
1156 key
.offset
= other_start
;
1157 del_slot
= path
->slots
[0];
1159 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
1160 0, root
->root_key
.objectid
,
1161 ino
, orig_offset
, 0);
1162 BUG_ON(ret
); /* -ENOMEM */
1165 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1166 struct btrfs_file_extent_item
);
1167 btrfs_set_file_extent_type(leaf
, fi
,
1168 BTRFS_FILE_EXTENT_REG
);
1169 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1170 btrfs_mark_buffer_dirty(leaf
);
1172 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
1173 struct btrfs_file_extent_item
);
1174 btrfs_set_file_extent_type(leaf
, fi
,
1175 BTRFS_FILE_EXTENT_REG
);
1176 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1177 btrfs_set_file_extent_num_bytes(leaf
, fi
,
1178 extent_end
- key
.offset
);
1179 btrfs_mark_buffer_dirty(leaf
);
1181 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
1183 btrfs_abort_transaction(trans
, root
, ret
);
1188 btrfs_free_path(path
);
1193 * on error we return an unlocked page and the error value
1194 * on success we return a locked page and 0
1196 static int prepare_uptodate_page(struct page
*page
, u64 pos
,
1197 bool force_uptodate
)
1201 if (((pos
& (PAGE_CACHE_SIZE
- 1)) || force_uptodate
) &&
1202 !PageUptodate(page
)) {
1203 ret
= btrfs_readpage(NULL
, page
);
1207 if (!PageUptodate(page
)) {
1216 * this gets pages into the page cache and locks them down, it also properly
1217 * waits for data=ordered extents to finish before allowing the pages to be
1220 static noinline
int prepare_pages(struct btrfs_root
*root
, struct file
*file
,
1221 struct page
**pages
, size_t num_pages
,
1222 loff_t pos
, unsigned long first_index
,
1223 size_t write_bytes
, bool force_uptodate
)
1225 struct extent_state
*cached_state
= NULL
;
1227 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
1228 struct inode
*inode
= fdentry(file
)->d_inode
;
1229 gfp_t mask
= btrfs_alloc_write_mask(inode
->i_mapping
);
1235 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
1236 last_pos
= ((u64
)index
+ num_pages
) << PAGE_CACHE_SHIFT
;
1239 for (i
= 0; i
< num_pages
; i
++) {
1240 pages
[i
] = find_or_create_page(inode
->i_mapping
, index
+ i
,
1241 mask
| __GFP_WRITE
);
1249 err
= prepare_uptodate_page(pages
[i
], pos
,
1251 if (i
== num_pages
- 1)
1252 err
= prepare_uptodate_page(pages
[i
],
1253 pos
+ write_bytes
, false);
1255 page_cache_release(pages
[i
]);
1259 wait_on_page_writeback(pages
[i
]);
1262 if (start_pos
< inode
->i_size
) {
1263 struct btrfs_ordered_extent
*ordered
;
1264 lock_extent_bits(&BTRFS_I(inode
)->io_tree
,
1265 start_pos
, last_pos
- 1, 0, &cached_state
);
1266 ordered
= btrfs_lookup_first_ordered_extent(inode
,
1269 ordered
->file_offset
+ ordered
->len
> start_pos
&&
1270 ordered
->file_offset
< last_pos
) {
1271 btrfs_put_ordered_extent(ordered
);
1272 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1273 start_pos
, last_pos
- 1,
1274 &cached_state
, GFP_NOFS
);
1275 for (i
= 0; i
< num_pages
; i
++) {
1276 unlock_page(pages
[i
]);
1277 page_cache_release(pages
[i
]);
1279 btrfs_wait_ordered_range(inode
, start_pos
,
1280 last_pos
- start_pos
);
1284 btrfs_put_ordered_extent(ordered
);
1286 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start_pos
,
1287 last_pos
- 1, EXTENT_DIRTY
| EXTENT_DELALLOC
|
1288 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
1289 0, 0, &cached_state
, GFP_NOFS
);
1290 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1291 start_pos
, last_pos
- 1, &cached_state
,
1294 for (i
= 0; i
< num_pages
; i
++) {
1295 if (clear_page_dirty_for_io(pages
[i
]))
1296 account_page_redirty(pages
[i
]);
1297 set_page_extent_mapped(pages
[i
]);
1298 WARN_ON(!PageLocked(pages
[i
]));
1302 while (faili
>= 0) {
1303 unlock_page(pages
[faili
]);
1304 page_cache_release(pages
[faili
]);
1311 static noinline ssize_t
__btrfs_buffered_write(struct file
*file
,
1315 struct inode
*inode
= fdentry(file
)->d_inode
;
1316 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1317 struct page
**pages
= NULL
;
1318 unsigned long first_index
;
1319 size_t num_written
= 0;
1322 bool force_page_uptodate
= false;
1324 nrptrs
= min((iov_iter_count(i
) + PAGE_CACHE_SIZE
- 1) /
1325 PAGE_CACHE_SIZE
, PAGE_CACHE_SIZE
/
1326 (sizeof(struct page
*)));
1327 nrptrs
= min(nrptrs
, current
->nr_dirtied_pause
- current
->nr_dirtied
);
1328 nrptrs
= max(nrptrs
, 8);
1329 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
1333 first_index
= pos
>> PAGE_CACHE_SHIFT
;
1335 while (iov_iter_count(i
) > 0) {
1336 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1337 size_t write_bytes
= min(iov_iter_count(i
),
1338 nrptrs
* (size_t)PAGE_CACHE_SIZE
-
1340 size_t num_pages
= (write_bytes
+ offset
+
1341 PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
1345 WARN_ON(num_pages
> nrptrs
);
1348 * Fault pages before locking them in prepare_pages
1349 * to avoid recursive lock
1351 if (unlikely(iov_iter_fault_in_readable(i
, write_bytes
))) {
1356 ret
= btrfs_delalloc_reserve_space(inode
,
1357 num_pages
<< PAGE_CACHE_SHIFT
);
1362 * This is going to setup the pages array with the number of
1363 * pages we want, so we don't really need to worry about the
1364 * contents of pages from loop to loop
1366 ret
= prepare_pages(root
, file
, pages
, num_pages
,
1367 pos
, first_index
, write_bytes
,
1368 force_page_uptodate
);
1370 btrfs_delalloc_release_space(inode
,
1371 num_pages
<< PAGE_CACHE_SHIFT
);
1375 copied
= btrfs_copy_from_user(pos
, num_pages
,
1376 write_bytes
, pages
, i
);
1379 * if we have trouble faulting in the pages, fall
1380 * back to one page at a time
1382 if (copied
< write_bytes
)
1386 force_page_uptodate
= true;
1389 force_page_uptodate
= false;
1390 dirty_pages
= (copied
+ offset
+
1391 PAGE_CACHE_SIZE
- 1) >>
1396 * If we had a short copy we need to release the excess delaloc
1397 * bytes we reserved. We need to increment outstanding_extents
1398 * because btrfs_delalloc_release_space will decrement it, but
1399 * we still have an outstanding extent for the chunk we actually
1402 if (num_pages
> dirty_pages
) {
1404 spin_lock(&BTRFS_I(inode
)->lock
);
1405 BTRFS_I(inode
)->outstanding_extents
++;
1406 spin_unlock(&BTRFS_I(inode
)->lock
);
1408 btrfs_delalloc_release_space(inode
,
1409 (num_pages
- dirty_pages
) <<
1414 ret
= btrfs_dirty_pages(root
, inode
, pages
,
1415 dirty_pages
, pos
, copied
,
1418 btrfs_delalloc_release_space(inode
,
1419 dirty_pages
<< PAGE_CACHE_SHIFT
);
1420 btrfs_drop_pages(pages
, num_pages
);
1425 btrfs_drop_pages(pages
, num_pages
);
1429 balance_dirty_pages_ratelimited(inode
->i_mapping
);
1430 if (dirty_pages
< (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
1431 btrfs_btree_balance_dirty(root
);
1434 num_written
+= copied
;
1439 return num_written
? num_written
: ret
;
1442 static ssize_t
__btrfs_direct_write(struct kiocb
*iocb
,
1443 const struct iovec
*iov
,
1444 unsigned long nr_segs
, loff_t pos
,
1445 loff_t
*ppos
, size_t count
, size_t ocount
)
1447 struct file
*file
= iocb
->ki_filp
;
1450 ssize_t written_buffered
;
1454 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
, ppos
,
1457 if (written
< 0 || written
== count
)
1462 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
1463 written_buffered
= __btrfs_buffered_write(file
, &i
, pos
);
1464 if (written_buffered
< 0) {
1465 err
= written_buffered
;
1468 endbyte
= pos
+ written_buffered
- 1;
1469 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
1472 written
+= written_buffered
;
1473 *ppos
= pos
+ written_buffered
;
1474 invalidate_mapping_pages(file
->f_mapping
, pos
>> PAGE_CACHE_SHIFT
,
1475 endbyte
>> PAGE_CACHE_SHIFT
);
1477 return written
? written
: err
;
1480 static void update_time_for_write(struct inode
*inode
)
1482 struct timespec now
;
1484 if (IS_NOCMTIME(inode
))
1487 now
= current_fs_time(inode
->i_sb
);
1488 if (!timespec_equal(&inode
->i_mtime
, &now
))
1489 inode
->i_mtime
= now
;
1491 if (!timespec_equal(&inode
->i_ctime
, &now
))
1492 inode
->i_ctime
= now
;
1494 if (IS_I_VERSION(inode
))
1495 inode_inc_iversion(inode
);
1498 static ssize_t
btrfs_file_aio_write(struct kiocb
*iocb
,
1499 const struct iovec
*iov
,
1500 unsigned long nr_segs
, loff_t pos
)
1502 struct file
*file
= iocb
->ki_filp
;
1503 struct inode
*inode
= fdentry(file
)->d_inode
;
1504 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1505 loff_t
*ppos
= &iocb
->ki_pos
;
1507 ssize_t num_written
= 0;
1509 size_t count
, ocount
;
1510 bool sync
= (file
->f_flags
& O_DSYNC
) || IS_SYNC(file
->f_mapping
->host
);
1512 sb_start_write(inode
->i_sb
);
1514 mutex_lock(&inode
->i_mutex
);
1516 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
1518 mutex_unlock(&inode
->i_mutex
);
1523 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
1524 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1526 mutex_unlock(&inode
->i_mutex
);
1531 mutex_unlock(&inode
->i_mutex
);
1535 err
= file_remove_suid(file
);
1537 mutex_unlock(&inode
->i_mutex
);
1542 * If BTRFS flips readonly due to some impossible error
1543 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1544 * although we have opened a file as writable, we have
1545 * to stop this write operation to ensure FS consistency.
1547 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
1548 mutex_unlock(&inode
->i_mutex
);
1554 * We reserve space for updating the inode when we reserve space for the
1555 * extent we are going to write, so we will enospc out there. We don't
1556 * need to start yet another transaction to update the inode as we will
1557 * update the inode when we finish writing whatever data we write.
1559 update_time_for_write(inode
);
1561 start_pos
= round_down(pos
, root
->sectorsize
);
1562 if (start_pos
> i_size_read(inode
)) {
1563 err
= btrfs_cont_expand(inode
, i_size_read(inode
), start_pos
);
1565 mutex_unlock(&inode
->i_mutex
);
1571 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
1573 if (unlikely(file
->f_flags
& O_DIRECT
)) {
1574 num_written
= __btrfs_direct_write(iocb
, iov
, nr_segs
,
1575 pos
, ppos
, count
, ocount
);
1579 iov_iter_init(&i
, iov
, nr_segs
, count
, num_written
);
1581 num_written
= __btrfs_buffered_write(file
, &i
, pos
);
1582 if (num_written
> 0)
1583 *ppos
= pos
+ num_written
;
1586 mutex_unlock(&inode
->i_mutex
);
1589 * we want to make sure fsync finds this change
1590 * but we haven't joined a transaction running right now.
1592 * Later on, someone is sure to update the inode and get the
1593 * real transid recorded.
1595 * We set last_trans now to the fs_info generation + 1,
1596 * this will either be one more than the running transaction
1597 * or the generation used for the next transaction if there isn't
1598 * one running right now.
1600 * We also have to set last_sub_trans to the current log transid,
1601 * otherwise subsequent syncs to a file that's been synced in this
1602 * transaction will appear to have already occured.
1604 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
+ 1;
1605 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
1606 if (num_written
> 0 || num_written
== -EIOCBQUEUED
) {
1607 err
= generic_write_sync(file
, pos
, num_written
);
1608 if (err
< 0 && num_written
> 0)
1613 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
1615 sb_end_write(inode
->i_sb
);
1616 current
->backing_dev_info
= NULL
;
1617 return num_written
? num_written
: err
;
1620 int btrfs_release_file(struct inode
*inode
, struct file
*filp
)
1623 * ordered_data_close is set by settattr when we are about to truncate
1624 * a file from a non-zero size to a zero size. This tries to
1625 * flush down new bytes that may have been written if the
1626 * application were using truncate to replace a file in place.
1628 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
1629 &BTRFS_I(inode
)->runtime_flags
)) {
1630 btrfs_add_ordered_operation(NULL
, BTRFS_I(inode
)->root
, inode
);
1631 if (inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
1632 filemap_flush(inode
->i_mapping
);
1634 if (filp
->private_data
)
1635 btrfs_ioctl_trans_end(filp
);
1640 * fsync call for both files and directories. This logs the inode into
1641 * the tree log instead of forcing full commits whenever possible.
1643 * It needs to call filemap_fdatawait so that all ordered extent updates are
1644 * in the metadata btree are up to date for copying to the log.
1646 * It drops the inode mutex before doing the tree log commit. This is an
1647 * important optimization for directories because holding the mutex prevents
1648 * new operations on the dir while we write to disk.
1650 int btrfs_sync_file(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1652 struct dentry
*dentry
= file
->f_path
.dentry
;
1653 struct inode
*inode
= dentry
->d_inode
;
1654 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1656 struct btrfs_trans_handle
*trans
;
1658 trace_btrfs_sync_file(file
, datasync
);
1661 * We write the dirty pages in the range and wait until they complete
1662 * out of the ->i_mutex. If so, we can flush the dirty pages by
1663 * multi-task, and make the performance up.
1665 atomic_inc(&BTRFS_I(inode
)->sync_writers
);
1666 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1667 atomic_dec(&BTRFS_I(inode
)->sync_writers
);
1671 mutex_lock(&inode
->i_mutex
);
1674 * We flush the dirty pages again to avoid some dirty pages in the
1677 atomic_inc(&root
->log_batch
);
1678 btrfs_wait_ordered_range(inode
, start
, end
- start
+ 1);
1679 atomic_inc(&root
->log_batch
);
1682 * check the transaction that last modified this inode
1683 * and see if its already been committed
1685 if (!BTRFS_I(inode
)->last_trans
) {
1686 mutex_unlock(&inode
->i_mutex
);
1691 * if the last transaction that changed this file was before
1692 * the current transaction, we can bail out now without any
1696 if (btrfs_inode_in_log(inode
, root
->fs_info
->generation
) ||
1697 BTRFS_I(inode
)->last_trans
<=
1698 root
->fs_info
->last_trans_committed
) {
1699 BTRFS_I(inode
)->last_trans
= 0;
1702 * We'v had everything committed since the last time we were
1703 * modified so clear this flag in case it was set for whatever
1704 * reason, it's no longer relevant.
1706 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1707 &BTRFS_I(inode
)->runtime_flags
);
1708 mutex_unlock(&inode
->i_mutex
);
1713 * ok we haven't committed the transaction yet, lets do a commit
1715 if (file
->private_data
)
1716 btrfs_ioctl_trans_end(file
);
1718 trans
= btrfs_start_transaction(root
, 0);
1719 if (IS_ERR(trans
)) {
1720 ret
= PTR_ERR(trans
);
1721 mutex_unlock(&inode
->i_mutex
);
1725 ret
= btrfs_log_dentry_safe(trans
, root
, dentry
);
1727 mutex_unlock(&inode
->i_mutex
);
1731 /* we've logged all the items and now have a consistent
1732 * version of the file in the log. It is possible that
1733 * someone will come in and modify the file, but that's
1734 * fine because the log is consistent on disk, and we
1735 * have references to all of the file's extents
1737 * It is possible that someone will come in and log the
1738 * file again, but that will end up using the synchronization
1739 * inside btrfs_sync_log to keep things safe.
1741 mutex_unlock(&inode
->i_mutex
);
1743 if (ret
!= BTRFS_NO_LOG_SYNC
) {
1745 ret
= btrfs_commit_transaction(trans
, root
);
1747 ret
= btrfs_sync_log(trans
, root
);
1749 ret
= btrfs_end_transaction(trans
, root
);
1751 ret
= btrfs_commit_transaction(trans
, root
);
1754 ret
= btrfs_end_transaction(trans
, root
);
1757 return ret
> 0 ? -EIO
: ret
;
1760 static const struct vm_operations_struct btrfs_file_vm_ops
= {
1761 .fault
= filemap_fault
,
1762 .page_mkwrite
= btrfs_page_mkwrite
,
1763 .remap_pages
= generic_file_remap_pages
,
1766 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
1768 struct address_space
*mapping
= filp
->f_mapping
;
1770 if (!mapping
->a_ops
->readpage
)
1773 file_accessed(filp
);
1774 vma
->vm_ops
= &btrfs_file_vm_ops
;
1779 static int hole_mergeable(struct inode
*inode
, struct extent_buffer
*leaf
,
1780 int slot
, u64 start
, u64 end
)
1782 struct btrfs_file_extent_item
*fi
;
1783 struct btrfs_key key
;
1785 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
1788 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
1789 if (key
.objectid
!= btrfs_ino(inode
) ||
1790 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
1793 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
1795 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
1798 if (btrfs_file_extent_disk_bytenr(leaf
, fi
))
1801 if (key
.offset
== end
)
1803 if (key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
) == start
)
1808 static int fill_holes(struct btrfs_trans_handle
*trans
, struct inode
*inode
,
1809 struct btrfs_path
*path
, u64 offset
, u64 end
)
1811 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1812 struct extent_buffer
*leaf
;
1813 struct btrfs_file_extent_item
*fi
;
1814 struct extent_map
*hole_em
;
1815 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1816 struct btrfs_key key
;
1819 key
.objectid
= btrfs_ino(inode
);
1820 key
.type
= BTRFS_EXTENT_DATA_KEY
;
1821 key
.offset
= offset
;
1824 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1829 leaf
= path
->nodes
[0];
1830 if (hole_mergeable(inode
, leaf
, path
->slots
[0]-1, offset
, end
)) {
1834 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1835 struct btrfs_file_extent_item
);
1836 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) +
1838 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1839 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
1840 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1841 btrfs_mark_buffer_dirty(leaf
);
1845 if (hole_mergeable(inode
, leaf
, path
->slots
[0]+1, offset
, end
)) {
1849 key
.offset
= offset
;
1850 btrfs_set_item_key_safe(trans
, root
, path
, &key
);
1851 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1852 struct btrfs_file_extent_item
);
1853 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
) + end
-
1855 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1856 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
1857 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1858 btrfs_mark_buffer_dirty(leaf
);
1861 btrfs_release_path(path
);
1863 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
1864 0, 0, end
- offset
, 0, end
- offset
,
1870 btrfs_release_path(path
);
1872 hole_em
= alloc_extent_map();
1874 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
1875 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1876 &BTRFS_I(inode
)->runtime_flags
);
1878 hole_em
->start
= offset
;
1879 hole_em
->len
= end
- offset
;
1880 hole_em
->orig_start
= offset
;
1882 hole_em
->block_start
= EXTENT_MAP_HOLE
;
1883 hole_em
->block_len
= 0;
1884 hole_em
->orig_block_len
= 0;
1885 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1886 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
1887 hole_em
->generation
= trans
->transid
;
1890 btrfs_drop_extent_cache(inode
, offset
, end
- 1, 0);
1891 write_lock(&em_tree
->lock
);
1892 ret
= add_extent_mapping(em_tree
, hole_em
);
1894 list_move(&hole_em
->list
,
1895 &em_tree
->modified_extents
);
1896 write_unlock(&em_tree
->lock
);
1897 } while (ret
== -EEXIST
);
1898 free_extent_map(hole_em
);
1900 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
1901 &BTRFS_I(inode
)->runtime_flags
);
1907 static int btrfs_punch_hole(struct inode
*inode
, loff_t offset
, loff_t len
)
1909 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1910 struct extent_state
*cached_state
= NULL
;
1911 struct btrfs_path
*path
;
1912 struct btrfs_block_rsv
*rsv
;
1913 struct btrfs_trans_handle
*trans
;
1914 u64 lockstart
= round_up(offset
, BTRFS_I(inode
)->root
->sectorsize
);
1915 u64 lockend
= round_down(offset
+ len
,
1916 BTRFS_I(inode
)->root
->sectorsize
) - 1;
1917 u64 cur_offset
= lockstart
;
1918 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
1922 bool same_page
= ((offset
>> PAGE_CACHE_SHIFT
) ==
1923 ((offset
+ len
- 1) >> PAGE_CACHE_SHIFT
));
1925 btrfs_wait_ordered_range(inode
, offset
, len
);
1927 mutex_lock(&inode
->i_mutex
);
1929 * We needn't truncate any page which is beyond the end of the file
1930 * because we are sure there is no data there.
1933 * Only do this if we are in the same page and we aren't doing the
1936 if (same_page
&& len
< PAGE_CACHE_SIZE
) {
1937 if (offset
< round_up(inode
->i_size
, PAGE_CACHE_SIZE
))
1938 ret
= btrfs_truncate_page(inode
, offset
, len
, 0);
1939 mutex_unlock(&inode
->i_mutex
);
1943 /* zero back part of the first page */
1944 if (offset
< round_up(inode
->i_size
, PAGE_CACHE_SIZE
)) {
1945 ret
= btrfs_truncate_page(inode
, offset
, 0, 0);
1947 mutex_unlock(&inode
->i_mutex
);
1952 /* zero the front end of the last page */
1953 if (offset
+ len
< round_up(inode
->i_size
, PAGE_CACHE_SIZE
)) {
1954 ret
= btrfs_truncate_page(inode
, offset
+ len
, 0, 1);
1956 mutex_unlock(&inode
->i_mutex
);
1961 if (lockend
< lockstart
) {
1962 mutex_unlock(&inode
->i_mutex
);
1967 struct btrfs_ordered_extent
*ordered
;
1969 truncate_pagecache_range(inode
, lockstart
, lockend
);
1971 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
1973 ordered
= btrfs_lookup_first_ordered_extent(inode
, lockend
);
1976 * We need to make sure we have no ordered extents in this range
1977 * and nobody raced in and read a page in this range, if we did
1978 * we need to try again.
1981 (ordered
->file_offset
+ ordered
->len
< lockstart
||
1982 ordered
->file_offset
> lockend
)) &&
1983 !test_range_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
1984 lockend
, EXTENT_UPTODATE
, 0,
1987 btrfs_put_ordered_extent(ordered
);
1991 btrfs_put_ordered_extent(ordered
);
1992 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
,
1993 lockend
, &cached_state
, GFP_NOFS
);
1994 btrfs_wait_ordered_range(inode
, lockstart
,
1995 lockend
- lockstart
+ 1);
1998 path
= btrfs_alloc_path();
2004 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2009 rsv
->size
= btrfs_calc_trunc_metadata_size(root
, 1);
2013 * 1 - update the inode
2014 * 1 - removing the extents in the range
2015 * 1 - adding the hole extent
2017 trans
= btrfs_start_transaction(root
, 3);
2018 if (IS_ERR(trans
)) {
2019 err
= PTR_ERR(trans
);
2023 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
2026 trans
->block_rsv
= rsv
;
2028 while (cur_offset
< lockend
) {
2029 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
2030 cur_offset
, lockend
+ 1,
2035 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2037 ret
= fill_holes(trans
, inode
, path
, cur_offset
, drop_end
);
2043 cur_offset
= drop_end
;
2045 ret
= btrfs_update_inode(trans
, root
, inode
);
2051 btrfs_end_transaction(trans
, root
);
2052 btrfs_btree_balance_dirty(root
);
2054 trans
= btrfs_start_transaction(root
, 3);
2055 if (IS_ERR(trans
)) {
2056 ret
= PTR_ERR(trans
);
2061 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
2063 BUG_ON(ret
); /* shouldn't happen */
2064 trans
->block_rsv
= rsv
;
2072 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2073 ret
= fill_holes(trans
, inode
, path
, cur_offset
, drop_end
);
2083 inode_inc_iversion(inode
);
2084 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2086 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2087 ret
= btrfs_update_inode(trans
, root
, inode
);
2088 btrfs_end_transaction(trans
, root
);
2089 btrfs_btree_balance_dirty(root
);
2091 btrfs_free_path(path
);
2092 btrfs_free_block_rsv(root
, rsv
);
2094 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2095 &cached_state
, GFP_NOFS
);
2096 mutex_unlock(&inode
->i_mutex
);
2102 static long btrfs_fallocate(struct file
*file
, int mode
,
2103 loff_t offset
, loff_t len
)
2105 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
2106 struct extent_state
*cached_state
= NULL
;
2113 struct extent_map
*em
;
2114 int blocksize
= BTRFS_I(inode
)->root
->sectorsize
;
2117 alloc_start
= round_down(offset
, blocksize
);
2118 alloc_end
= round_up(offset
+ len
, blocksize
);
2120 /* Make sure we aren't being give some crap mode */
2121 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
2124 if (mode
& FALLOC_FL_PUNCH_HOLE
)
2125 return btrfs_punch_hole(inode
, offset
, len
);
2128 * Make sure we have enough space before we do the
2131 ret
= btrfs_check_data_free_space(inode
, alloc_end
- alloc_start
);
2136 * wait for ordered IO before we have any locks. We'll loop again
2137 * below with the locks held.
2139 btrfs_wait_ordered_range(inode
, alloc_start
, alloc_end
- alloc_start
);
2141 mutex_lock(&inode
->i_mutex
);
2142 ret
= inode_newsize_ok(inode
, alloc_end
);
2146 if (alloc_start
> inode
->i_size
) {
2147 ret
= btrfs_cont_expand(inode
, i_size_read(inode
),
2153 locked_end
= alloc_end
- 1;
2155 struct btrfs_ordered_extent
*ordered
;
2157 /* the extent lock is ordered inside the running
2160 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, alloc_start
,
2161 locked_end
, 0, &cached_state
);
2162 ordered
= btrfs_lookup_first_ordered_extent(inode
,
2165 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
2166 ordered
->file_offset
< alloc_end
) {
2167 btrfs_put_ordered_extent(ordered
);
2168 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
2169 alloc_start
, locked_end
,
2170 &cached_state
, GFP_NOFS
);
2172 * we can't wait on the range with the transaction
2173 * running or with the extent lock held
2175 btrfs_wait_ordered_range(inode
, alloc_start
,
2176 alloc_end
- alloc_start
);
2179 btrfs_put_ordered_extent(ordered
);
2184 cur_offset
= alloc_start
;
2188 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2189 alloc_end
- cur_offset
, 0);
2190 if (IS_ERR_OR_NULL(em
)) {
2197 last_byte
= min(extent_map_end(em
), alloc_end
);
2198 actual_end
= min_t(u64
, extent_map_end(em
), offset
+ len
);
2199 last_byte
= ALIGN(last_byte
, blocksize
);
2201 if (em
->block_start
== EXTENT_MAP_HOLE
||
2202 (cur_offset
>= inode
->i_size
&&
2203 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
2204 ret
= btrfs_prealloc_file_range(inode
, mode
, cur_offset
,
2205 last_byte
- cur_offset
,
2206 1 << inode
->i_blkbits
,
2211 free_extent_map(em
);
2214 } else if (actual_end
> inode
->i_size
&&
2215 !(mode
& FALLOC_FL_KEEP_SIZE
)) {
2217 * We didn't need to allocate any more space, but we
2218 * still extended the size of the file so we need to
2221 inode
->i_ctime
= CURRENT_TIME
;
2222 i_size_write(inode
, actual_end
);
2223 btrfs_ordered_update_i_size(inode
, actual_end
, NULL
);
2225 free_extent_map(em
);
2227 cur_offset
= last_byte
;
2228 if (cur_offset
>= alloc_end
) {
2233 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
2234 &cached_state
, GFP_NOFS
);
2236 mutex_unlock(&inode
->i_mutex
);
2237 /* Let go of our reservation. */
2238 btrfs_free_reserved_data_space(inode
, alloc_end
- alloc_start
);
2242 static int find_desired_extent(struct inode
*inode
, loff_t
*offset
, int whence
)
2244 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2245 struct extent_map
*em
;
2246 struct extent_state
*cached_state
= NULL
;
2247 u64 lockstart
= *offset
;
2248 u64 lockend
= i_size_read(inode
);
2249 u64 start
= *offset
;
2250 u64 orig_start
= *offset
;
2251 u64 len
= i_size_read(inode
);
2255 lockend
= max_t(u64
, root
->sectorsize
, lockend
);
2256 if (lockend
<= lockstart
)
2257 lockend
= lockstart
+ root
->sectorsize
;
2260 len
= lockend
- lockstart
+ 1;
2262 len
= max_t(u64
, len
, root
->sectorsize
);
2263 if (inode
->i_size
== 0)
2266 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
, 0,
2270 * Delalloc is such a pain. If we have a hole and we have pending
2271 * delalloc for a portion of the hole we will get back a hole that
2272 * exists for the entire range since it hasn't been actually written
2273 * yet. So to take care of this case we need to look for an extent just
2274 * before the position we want in case there is outstanding delalloc
2277 if (whence
== SEEK_HOLE
&& start
!= 0) {
2278 if (start
<= root
->sectorsize
)
2279 em
= btrfs_get_extent_fiemap(inode
, NULL
, 0, 0,
2280 root
->sectorsize
, 0);
2282 em
= btrfs_get_extent_fiemap(inode
, NULL
, 0,
2283 start
- root
->sectorsize
,
2284 root
->sectorsize
, 0);
2289 last_end
= em
->start
+ em
->len
;
2290 if (em
->block_start
== EXTENT_MAP_DELALLOC
)
2291 last_end
= min_t(u64
, last_end
, inode
->i_size
);
2292 free_extent_map(em
);
2296 em
= btrfs_get_extent_fiemap(inode
, NULL
, 0, start
, len
, 0);
2302 if (em
->block_start
== EXTENT_MAP_HOLE
) {
2303 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2304 if (last_end
<= orig_start
) {
2305 free_extent_map(em
);
2311 if (whence
== SEEK_HOLE
) {
2313 free_extent_map(em
);
2317 if (whence
== SEEK_DATA
) {
2318 if (em
->block_start
== EXTENT_MAP_DELALLOC
) {
2319 if (start
>= inode
->i_size
) {
2320 free_extent_map(em
);
2326 if (!test_bit(EXTENT_FLAG_PREALLOC
,
2329 free_extent_map(em
);
2335 start
= em
->start
+ em
->len
;
2336 last_end
= em
->start
+ em
->len
;
2338 if (em
->block_start
== EXTENT_MAP_DELALLOC
)
2339 last_end
= min_t(u64
, last_end
, inode
->i_size
);
2341 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2342 free_extent_map(em
);
2346 free_extent_map(em
);
2350 *offset
= min(*offset
, inode
->i_size
);
2352 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
2353 &cached_state
, GFP_NOFS
);
2357 static loff_t
btrfs_file_llseek(struct file
*file
, loff_t offset
, int whence
)
2359 struct inode
*inode
= file
->f_mapping
->host
;
2362 mutex_lock(&inode
->i_mutex
);
2366 offset
= generic_file_llseek(file
, offset
, whence
);
2370 if (offset
>= i_size_read(inode
)) {
2371 mutex_unlock(&inode
->i_mutex
);
2375 ret
= find_desired_extent(inode
, &offset
, whence
);
2377 mutex_unlock(&inode
->i_mutex
);
2382 if (offset
< 0 && !(file
->f_mode
& FMODE_UNSIGNED_OFFSET
)) {
2386 if (offset
> inode
->i_sb
->s_maxbytes
) {
2391 /* Special lock needed here? */
2392 if (offset
!= file
->f_pos
) {
2393 file
->f_pos
= offset
;
2394 file
->f_version
= 0;
2397 mutex_unlock(&inode
->i_mutex
);
2401 const struct file_operations btrfs_file_operations
= {
2402 .llseek
= btrfs_file_llseek
,
2403 .read
= do_sync_read
,
2404 .write
= do_sync_write
,
2405 .aio_read
= generic_file_aio_read
,
2406 .splice_read
= generic_file_splice_read
,
2407 .aio_write
= btrfs_file_aio_write
,
2408 .mmap
= btrfs_file_mmap
,
2409 .open
= generic_file_open
,
2410 .release
= btrfs_release_file
,
2411 .fsync
= btrfs_sync_file
,
2412 .fallocate
= btrfs_fallocate
,
2413 .unlocked_ioctl
= btrfs_ioctl
,
2414 #ifdef CONFIG_COMPAT
2415 .compat_ioctl
= btrfs_ioctl
,
2419 void btrfs_auto_defrag_exit(void)
2421 if (btrfs_inode_defrag_cachep
)
2422 kmem_cache_destroy(btrfs_inode_defrag_cachep
);
2425 int btrfs_auto_defrag_init(void)
2427 btrfs_inode_defrag_cachep
= kmem_cache_create("btrfs_inode_defrag",
2428 sizeof(struct inode_defrag
), 0,
2429 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
2431 if (!btrfs_inode_defrag_cachep
)