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/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/statfs.h>
30 #include <linux/compat.h>
31 #include <linux/slab.h>
34 #include "transaction.h"
35 #include "btrfs_inode.h"
37 #include "print-tree.h"
43 /* simple helper to fault in pages and copy. This should go away
44 * and be replaced with calls into generic code.
46 static noinline
int btrfs_copy_from_user(loff_t pos
, int num_pages
,
48 struct page
**prepared_pages
,
49 const char __user
*buf
)
53 int offset
= pos
& (PAGE_CACHE_SIZE
- 1);
55 for (i
= 0; i
< num_pages
&& write_bytes
> 0; i
++, offset
= 0) {
56 size_t count
= min_t(size_t,
57 PAGE_CACHE_SIZE
- offset
, write_bytes
);
58 struct page
*page
= prepared_pages
[i
];
59 fault_in_pages_readable(buf
, count
);
61 /* Copy data from userspace to the current page */
63 page_fault
= __copy_from_user(page_address(page
) + offset
,
65 /* Flush processor's dcache for this page */
66 flush_dcache_page(page
);
74 return page_fault
? -EFAULT
: 0;
78 * unlocks pages after btrfs_file_write is done with them
80 static noinline
void btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
83 for (i
= 0; i
< num_pages
; i
++) {
86 /* page checked is some magic around finding pages that
87 * have been modified without going through btrfs_set_page_dirty
90 ClearPageChecked(pages
[i
]);
91 unlock_page(pages
[i
]);
92 mark_page_accessed(pages
[i
]);
93 page_cache_release(pages
[i
]);
98 * after copy_from_user, pages need to be dirtied and we need to make
99 * sure holes are created between the current EOF and the start of
100 * any next extents (if required).
102 * this also makes the decision about creating an inline extent vs
103 * doing real data extents, marking pages dirty and delalloc as required.
105 static noinline
int dirty_and_release_pages(struct btrfs_trans_handle
*trans
,
106 struct btrfs_root
*root
,
115 struct inode
*inode
= fdentry(file
)->d_inode
;
118 u64 end_of_last_block
;
119 u64 end_pos
= pos
+ write_bytes
;
120 loff_t isize
= i_size_read(inode
);
122 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
123 num_bytes
= (write_bytes
+ pos
- start_pos
+
124 root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
126 end_of_last_block
= start_pos
+ num_bytes
- 1;
127 err
= btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
,
132 for (i
= 0; i
< num_pages
; i
++) {
133 struct page
*p
= pages
[i
];
138 if (end_pos
> isize
) {
139 i_size_write(inode
, end_pos
);
140 /* we've only changed i_size in ram, and we haven't updated
141 * the disk i_size. There is no need to log the inode
149 * this drops all the extents in the cache that intersect the range
150 * [start, end]. Existing extents are split as required.
152 int btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
,
155 struct extent_map
*em
;
156 struct extent_map
*split
= NULL
;
157 struct extent_map
*split2
= NULL
;
158 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
159 u64 len
= end
- start
+ 1;
165 WARN_ON(end
< start
);
166 if (end
== (u64
)-1) {
172 split
= alloc_extent_map(GFP_NOFS
);
174 split2
= alloc_extent_map(GFP_NOFS
);
176 write_lock(&em_tree
->lock
);
177 em
= lookup_extent_mapping(em_tree
, start
, len
);
179 write_unlock(&em_tree
->lock
);
183 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
184 if (testend
&& em
->start
+ em
->len
>= start
+ len
) {
186 write_unlock(&em_tree
->lock
);
189 start
= em
->start
+ em
->len
;
191 len
= start
+ len
- (em
->start
+ em
->len
);
193 write_unlock(&em_tree
->lock
);
196 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
197 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
198 remove_extent_mapping(em_tree
, em
);
200 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
202 split
->start
= em
->start
;
203 split
->len
= start
- em
->start
;
204 split
->orig_start
= em
->orig_start
;
205 split
->block_start
= em
->block_start
;
208 split
->block_len
= em
->block_len
;
210 split
->block_len
= split
->len
;
212 split
->bdev
= em
->bdev
;
213 split
->flags
= flags
;
214 ret
= add_extent_mapping(em_tree
, split
);
216 free_extent_map(split
);
220 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
221 testend
&& em
->start
+ em
->len
> start
+ len
) {
222 u64 diff
= start
+ len
- em
->start
;
224 split
->start
= start
+ len
;
225 split
->len
= em
->start
+ em
->len
- (start
+ len
);
226 split
->bdev
= em
->bdev
;
227 split
->flags
= flags
;
230 split
->block_len
= em
->block_len
;
231 split
->block_start
= em
->block_start
;
232 split
->orig_start
= em
->orig_start
;
234 split
->block_len
= split
->len
;
235 split
->block_start
= em
->block_start
+ diff
;
236 split
->orig_start
= split
->start
;
239 ret
= add_extent_mapping(em_tree
, split
);
241 free_extent_map(split
);
244 write_unlock(&em_tree
->lock
);
248 /* once for the tree*/
252 free_extent_map(split
);
254 free_extent_map(split2
);
259 * this is very complex, but the basic idea is to drop all extents
260 * in the range start - end. hint_block is filled in with a block number
261 * that would be a good hint to the block allocator for this file.
263 * If an extent intersects the range but is not entirely inside the range
264 * it is either truncated or split. Anything entirely inside the range
265 * is deleted from the tree.
267 int btrfs_drop_extents(struct btrfs_trans_handle
*trans
, struct inode
*inode
,
268 u64 start
, u64 end
, u64
*hint_byte
, int drop_cache
)
270 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
271 struct extent_buffer
*leaf
;
272 struct btrfs_file_extent_item
*fi
;
273 struct btrfs_path
*path
;
274 struct btrfs_key key
;
275 struct btrfs_key new_key
;
276 u64 search_start
= start
;
279 u64 extent_offset
= 0;
288 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
290 path
= btrfs_alloc_path();
296 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
300 if (ret
> 0 && path
->slots
[0] > 0 && search_start
== start
) {
301 leaf
= path
->nodes
[0];
302 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0] - 1);
303 if (key
.objectid
== inode
->i_ino
&&
304 key
.type
== BTRFS_EXTENT_DATA_KEY
)
309 leaf
= path
->nodes
[0];
310 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
312 ret
= btrfs_next_leaf(root
, path
);
319 leaf
= path
->nodes
[0];
323 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
324 if (key
.objectid
> inode
->i_ino
||
325 key
.type
> BTRFS_EXTENT_DATA_KEY
|| key
.offset
>= end
)
328 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
329 struct btrfs_file_extent_item
);
330 extent_type
= btrfs_file_extent_type(leaf
, fi
);
332 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
333 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
334 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
335 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
336 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
337 extent_end
= key
.offset
+
338 btrfs_file_extent_num_bytes(leaf
, fi
);
339 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
340 extent_end
= key
.offset
+
341 btrfs_file_extent_inline_len(leaf
, fi
);
344 extent_end
= search_start
;
347 if (extent_end
<= search_start
) {
352 search_start
= max(key
.offset
, start
);
354 btrfs_release_path(root
, path
);
359 * | - range to drop - |
360 * | -------- extent -------- |
362 if (start
> key
.offset
&& end
< extent_end
) {
364 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
366 memcpy(&new_key
, &key
, sizeof(new_key
));
367 new_key
.offset
= start
;
368 ret
= btrfs_duplicate_item(trans
, root
, path
,
370 if (ret
== -EAGAIN
) {
371 btrfs_release_path(root
, path
);
377 leaf
= path
->nodes
[0];
378 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
379 struct btrfs_file_extent_item
);
380 btrfs_set_file_extent_num_bytes(leaf
, fi
,
383 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
384 struct btrfs_file_extent_item
);
386 extent_offset
+= start
- key
.offset
;
387 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
388 btrfs_set_file_extent_num_bytes(leaf
, fi
,
390 btrfs_mark_buffer_dirty(leaf
);
392 if (disk_bytenr
> 0) {
393 ret
= btrfs_inc_extent_ref(trans
, root
,
394 disk_bytenr
, num_bytes
, 0,
395 root
->root_key
.objectid
,
397 start
- extent_offset
);
399 *hint_byte
= disk_bytenr
;
404 * | ---- range to drop ----- |
405 * | -------- extent -------- |
407 if (start
<= key
.offset
&& end
< extent_end
) {
408 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
410 memcpy(&new_key
, &key
, sizeof(new_key
));
411 new_key
.offset
= end
;
412 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
414 extent_offset
+= end
- key
.offset
;
415 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
416 btrfs_set_file_extent_num_bytes(leaf
, fi
,
418 btrfs_mark_buffer_dirty(leaf
);
419 if (disk_bytenr
> 0) {
420 inode_sub_bytes(inode
, end
- key
.offset
);
421 *hint_byte
= disk_bytenr
;
426 search_start
= extent_end
;
428 * | ---- range to drop ----- |
429 * | -------- extent -------- |
431 if (start
> key
.offset
&& end
>= extent_end
) {
433 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
435 btrfs_set_file_extent_num_bytes(leaf
, fi
,
437 btrfs_mark_buffer_dirty(leaf
);
438 if (disk_bytenr
> 0) {
439 inode_sub_bytes(inode
, extent_end
- start
);
440 *hint_byte
= disk_bytenr
;
442 if (end
== extent_end
)
450 * | ---- range to drop ----- |
451 * | ------ extent ------ |
453 if (start
<= key
.offset
&& end
>= extent_end
) {
455 del_slot
= path
->slots
[0];
458 BUG_ON(del_slot
+ del_nr
!= path
->slots
[0]);
462 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
463 inode_sub_bytes(inode
,
464 extent_end
- key
.offset
);
465 extent_end
= ALIGN(extent_end
,
467 } else if (disk_bytenr
> 0) {
468 ret
= btrfs_free_extent(trans
, root
,
469 disk_bytenr
, num_bytes
, 0,
470 root
->root_key
.objectid
,
471 key
.objectid
, key
.offset
-
474 inode_sub_bytes(inode
,
475 extent_end
- key
.offset
);
476 *hint_byte
= disk_bytenr
;
479 if (end
== extent_end
)
482 if (path
->slots
[0] + 1 < btrfs_header_nritems(leaf
)) {
487 ret
= btrfs_del_items(trans
, root
, path
, del_slot
,
494 btrfs_release_path(root
, path
);
502 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
506 btrfs_free_path(path
);
510 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
511 u64 objectid
, u64 bytenr
, u64 orig_offset
,
512 u64
*start
, u64
*end
)
514 struct btrfs_file_extent_item
*fi
;
515 struct btrfs_key key
;
518 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
521 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
522 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
525 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
526 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
527 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
528 btrfs_file_extent_offset(leaf
, fi
) != key
.offset
- orig_offset
||
529 btrfs_file_extent_compression(leaf
, fi
) ||
530 btrfs_file_extent_encryption(leaf
, fi
) ||
531 btrfs_file_extent_other_encoding(leaf
, fi
))
534 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
535 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
544 * Mark extent in the range start - end as written.
546 * This changes extent type from 'pre-allocated' to 'regular'. If only
547 * part of extent is marked as written, the extent will be split into
550 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
551 struct inode
*inode
, u64 start
, u64 end
)
553 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
554 struct extent_buffer
*leaf
;
555 struct btrfs_path
*path
;
556 struct btrfs_file_extent_item
*fi
;
557 struct btrfs_key key
;
558 struct btrfs_key new_key
;
571 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
573 path
= btrfs_alloc_path();
578 key
.objectid
= inode
->i_ino
;
579 key
.type
= BTRFS_EXTENT_DATA_KEY
;
582 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
583 if (ret
> 0 && path
->slots
[0] > 0)
586 leaf
= path
->nodes
[0];
587 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
588 BUG_ON(key
.objectid
!= inode
->i_ino
||
589 key
.type
!= BTRFS_EXTENT_DATA_KEY
);
590 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
591 struct btrfs_file_extent_item
);
592 BUG_ON(btrfs_file_extent_type(leaf
, fi
) !=
593 BTRFS_FILE_EXTENT_PREALLOC
);
594 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
595 BUG_ON(key
.offset
> start
|| extent_end
< end
);
597 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
598 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
599 orig_offset
= key
.offset
- btrfs_file_extent_offset(leaf
, fi
);
600 memcpy(&new_key
, &key
, sizeof(new_key
));
602 if (start
== key
.offset
&& end
< extent_end
) {
605 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
606 inode
->i_ino
, bytenr
, orig_offset
,
607 &other_start
, &other_end
)) {
608 new_key
.offset
= end
;
609 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
610 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
611 struct btrfs_file_extent_item
);
612 btrfs_set_file_extent_num_bytes(leaf
, fi
,
614 btrfs_set_file_extent_offset(leaf
, fi
,
616 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
617 struct btrfs_file_extent_item
);
618 btrfs_set_file_extent_num_bytes(leaf
, fi
,
620 btrfs_mark_buffer_dirty(leaf
);
625 if (start
> key
.offset
&& end
== extent_end
) {
628 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
629 inode
->i_ino
, bytenr
, orig_offset
,
630 &other_start
, &other_end
)) {
631 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
632 struct btrfs_file_extent_item
);
633 btrfs_set_file_extent_num_bytes(leaf
, fi
,
636 new_key
.offset
= start
;
637 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
639 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
640 struct btrfs_file_extent_item
);
641 btrfs_set_file_extent_num_bytes(leaf
, fi
,
643 btrfs_set_file_extent_offset(leaf
, fi
,
644 start
- orig_offset
);
645 btrfs_mark_buffer_dirty(leaf
);
650 while (start
> key
.offset
|| end
< extent_end
) {
651 if (key
.offset
== start
)
654 new_key
.offset
= split
;
655 ret
= btrfs_duplicate_item(trans
, root
, path
, &new_key
);
656 if (ret
== -EAGAIN
) {
657 btrfs_release_path(root
, path
);
662 leaf
= path
->nodes
[0];
663 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
664 struct btrfs_file_extent_item
);
665 btrfs_set_file_extent_num_bytes(leaf
, fi
,
668 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
669 struct btrfs_file_extent_item
);
671 btrfs_set_file_extent_offset(leaf
, fi
, split
- orig_offset
);
672 btrfs_set_file_extent_num_bytes(leaf
, fi
,
674 btrfs_mark_buffer_dirty(leaf
);
676 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
, 0,
677 root
->root_key
.objectid
,
678 inode
->i_ino
, orig_offset
);
681 if (split
== start
) {
684 BUG_ON(start
!= key
.offset
);
693 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
694 inode
->i_ino
, bytenr
, orig_offset
,
695 &other_start
, &other_end
)) {
697 btrfs_release_path(root
, path
);
700 extent_end
= other_end
;
701 del_slot
= path
->slots
[0] + 1;
703 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
704 0, root
->root_key
.objectid
,
705 inode
->i_ino
, orig_offset
);
710 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
711 inode
->i_ino
, bytenr
, orig_offset
,
712 &other_start
, &other_end
)) {
714 btrfs_release_path(root
, path
);
717 key
.offset
= other_start
;
718 del_slot
= path
->slots
[0];
720 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
721 0, root
->root_key
.objectid
,
722 inode
->i_ino
, orig_offset
);
726 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
727 struct btrfs_file_extent_item
);
728 btrfs_set_file_extent_type(leaf
, fi
,
729 BTRFS_FILE_EXTENT_REG
);
730 btrfs_mark_buffer_dirty(leaf
);
732 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
733 struct btrfs_file_extent_item
);
734 btrfs_set_file_extent_type(leaf
, fi
,
735 BTRFS_FILE_EXTENT_REG
);
736 btrfs_set_file_extent_num_bytes(leaf
, fi
,
737 extent_end
- key
.offset
);
738 btrfs_mark_buffer_dirty(leaf
);
740 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
744 btrfs_free_path(path
);
749 * this gets pages into the page cache and locks them down, it also properly
750 * waits for data=ordered extents to finish before allowing the pages to be
753 static noinline
int prepare_pages(struct btrfs_root
*root
, struct file
*file
,
754 struct page
**pages
, size_t num_pages
,
755 loff_t pos
, unsigned long first_index
,
756 unsigned long last_index
, size_t write_bytes
)
758 struct extent_state
*cached_state
= NULL
;
760 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
761 struct inode
*inode
= fdentry(file
)->d_inode
;
766 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
767 last_pos
= ((u64
)index
+ num_pages
) << PAGE_CACHE_SHIFT
;
769 if (start_pos
> inode
->i_size
) {
770 err
= btrfs_cont_expand(inode
, start_pos
);
775 memset(pages
, 0, num_pages
* sizeof(struct page
*));
777 for (i
= 0; i
< num_pages
; i
++) {
778 pages
[i
] = grab_cache_page(inode
->i_mapping
, index
+ i
);
783 wait_on_page_writeback(pages
[i
]);
785 if (start_pos
< inode
->i_size
) {
786 struct btrfs_ordered_extent
*ordered
;
787 lock_extent_bits(&BTRFS_I(inode
)->io_tree
,
788 start_pos
, last_pos
- 1, 0, &cached_state
,
790 ordered
= btrfs_lookup_first_ordered_extent(inode
,
793 ordered
->file_offset
+ ordered
->len
> start_pos
&&
794 ordered
->file_offset
< last_pos
) {
795 btrfs_put_ordered_extent(ordered
);
796 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
797 start_pos
, last_pos
- 1,
798 &cached_state
, GFP_NOFS
);
799 for (i
= 0; i
< num_pages
; i
++) {
800 unlock_page(pages
[i
]);
801 page_cache_release(pages
[i
]);
803 btrfs_wait_ordered_range(inode
, start_pos
,
804 last_pos
- start_pos
);
808 btrfs_put_ordered_extent(ordered
);
810 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start_pos
,
811 last_pos
- 1, EXTENT_DIRTY
| EXTENT_DELALLOC
|
812 EXTENT_DO_ACCOUNTING
, 0, 0, &cached_state
,
814 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
815 start_pos
, last_pos
- 1, &cached_state
,
818 for (i
= 0; i
< num_pages
; i
++) {
819 clear_page_dirty_for_io(pages
[i
]);
820 set_page_extent_mapped(pages
[i
]);
821 WARN_ON(!PageLocked(pages
[i
]));
826 static ssize_t
btrfs_file_write(struct file
*file
, const char __user
*buf
,
827 size_t count
, loff_t
*ppos
)
831 ssize_t num_written
= 0;
834 struct inode
*inode
= fdentry(file
)->d_inode
;
835 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
836 struct page
**pages
= NULL
;
838 struct page
*pinned
[2];
839 unsigned long first_index
;
840 unsigned long last_index
;
843 will_write
= ((file
->f_flags
& O_DSYNC
) || IS_SYNC(inode
) ||
844 (file
->f_flags
& O_DIRECT
));
846 nrptrs
= min((count
+ PAGE_CACHE_SIZE
- 1) / PAGE_CACHE_SIZE
,
847 PAGE_CACHE_SIZE
/ (sizeof(struct page
*)));
854 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
856 /* do the reserve before the mutex lock in case we have to do some
857 * flushing. We wouldn't deadlock, but this is more polite.
859 err
= btrfs_reserve_metadata_for_delalloc(root
, inode
, 1);
863 mutex_lock(&inode
->i_mutex
);
865 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
866 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
873 err
= file_remove_suid(file
);
877 file_update_time(file
);
879 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
881 /* generic_write_checks can change our pos */
884 BTRFS_I(inode
)->sequence
++;
885 first_index
= pos
>> PAGE_CACHE_SHIFT
;
886 last_index
= (pos
+ count
) >> PAGE_CACHE_SHIFT
;
889 * there are lots of better ways to do this, but this code
890 * makes sure the first and last page in the file range are
891 * up to date and ready for cow
893 if ((pos
& (PAGE_CACHE_SIZE
- 1))) {
894 pinned
[0] = grab_cache_page(inode
->i_mapping
, first_index
);
895 if (!PageUptodate(pinned
[0])) {
896 ret
= btrfs_readpage(NULL
, pinned
[0]);
898 wait_on_page_locked(pinned
[0]);
900 unlock_page(pinned
[0]);
903 if ((pos
+ count
) & (PAGE_CACHE_SIZE
- 1)) {
904 pinned
[1] = grab_cache_page(inode
->i_mapping
, last_index
);
905 if (!PageUptodate(pinned
[1])) {
906 ret
= btrfs_readpage(NULL
, pinned
[1]);
908 wait_on_page_locked(pinned
[1]);
910 unlock_page(pinned
[1]);
915 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
916 size_t write_bytes
= min(count
, nrptrs
*
917 (size_t)PAGE_CACHE_SIZE
-
919 size_t num_pages
= (write_bytes
+ PAGE_CACHE_SIZE
- 1) >>
922 WARN_ON(num_pages
> nrptrs
);
923 memset(pages
, 0, sizeof(struct page
*) * nrptrs
);
925 ret
= btrfs_check_data_free_space(root
, inode
, write_bytes
);
929 ret
= prepare_pages(root
, file
, pages
, num_pages
,
930 pos
, first_index
, last_index
,
933 btrfs_free_reserved_data_space(root
, inode
,
938 ret
= btrfs_copy_from_user(pos
, num_pages
,
939 write_bytes
, pages
, buf
);
941 btrfs_free_reserved_data_space(root
, inode
,
943 btrfs_drop_pages(pages
, num_pages
);
947 ret
= dirty_and_release_pages(NULL
, root
, file
, pages
,
948 num_pages
, pos
, write_bytes
);
949 btrfs_drop_pages(pages
, num_pages
);
951 btrfs_free_reserved_data_space(root
, inode
,
957 filemap_fdatawrite_range(inode
->i_mapping
, pos
,
958 pos
+ write_bytes
- 1);
960 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
963 (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
964 btrfs_btree_balance_dirty(root
, 1);
965 btrfs_throttle(root
);
969 count
-= write_bytes
;
971 num_written
+= write_bytes
;
976 mutex_unlock(&inode
->i_mutex
);
979 btrfs_unreserve_metadata_for_delalloc(root
, inode
, 1);
984 page_cache_release(pinned
[0]);
986 page_cache_release(pinned
[1]);
990 * we want to make sure fsync finds this change
991 * but we haven't joined a transaction running right now.
993 * Later on, someone is sure to update the inode and get the
994 * real transid recorded.
996 * We set last_trans now to the fs_info generation + 1,
997 * this will either be one more than the running transaction
998 * or the generation used for the next transaction if there isn't
999 * one running right now.
1001 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
+ 1;
1003 if (num_written
> 0 && will_write
) {
1004 struct btrfs_trans_handle
*trans
;
1006 err
= btrfs_wait_ordered_range(inode
, start_pos
, num_written
);
1010 if ((file
->f_flags
& O_DSYNC
) || IS_SYNC(inode
)) {
1011 trans
= btrfs_start_transaction(root
, 1);
1012 ret
= btrfs_log_dentry_safe(trans
, root
,
1015 ret
= btrfs_sync_log(trans
, root
);
1017 btrfs_end_transaction(trans
, root
);
1019 btrfs_commit_transaction(trans
, root
);
1020 } else if (ret
!= BTRFS_NO_LOG_SYNC
) {
1021 btrfs_commit_transaction(trans
, root
);
1023 btrfs_end_transaction(trans
, root
);
1026 if (file
->f_flags
& O_DIRECT
) {
1027 invalidate_mapping_pages(inode
->i_mapping
,
1028 start_pos
>> PAGE_CACHE_SHIFT
,
1029 (start_pos
+ num_written
- 1) >> PAGE_CACHE_SHIFT
);
1032 current
->backing_dev_info
= NULL
;
1033 return num_written
? num_written
: err
;
1036 int btrfs_release_file(struct inode
*inode
, struct file
*filp
)
1039 * ordered_data_close is set by settattr when we are about to truncate
1040 * a file from a non-zero size to a zero size. This tries to
1041 * flush down new bytes that may have been written if the
1042 * application were using truncate to replace a file in place.
1044 if (BTRFS_I(inode
)->ordered_data_close
) {
1045 BTRFS_I(inode
)->ordered_data_close
= 0;
1046 btrfs_add_ordered_operation(NULL
, BTRFS_I(inode
)->root
, inode
);
1047 if (inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
1048 filemap_flush(inode
->i_mapping
);
1050 if (filp
->private_data
)
1051 btrfs_ioctl_trans_end(filp
);
1056 * fsync call for both files and directories. This logs the inode into
1057 * the tree log instead of forcing full commits whenever possible.
1059 * It needs to call filemap_fdatawait so that all ordered extent updates are
1060 * in the metadata btree are up to date for copying to the log.
1062 * It drops the inode mutex before doing the tree log commit. This is an
1063 * important optimization for directories because holding the mutex prevents
1064 * new operations on the dir while we write to disk.
1066 int btrfs_sync_file(struct file
*file
, struct dentry
*dentry
, int datasync
)
1068 struct inode
*inode
= dentry
->d_inode
;
1069 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1071 struct btrfs_trans_handle
*trans
;
1074 /* we wait first, since the writeback may change the inode */
1076 /* the VFS called filemap_fdatawrite for us */
1077 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
1081 * check the transaction that last modified this inode
1082 * and see if its already been committed
1084 if (!BTRFS_I(inode
)->last_trans
)
1088 * if the last transaction that changed this file was before
1089 * the current transaction, we can bail out now without any
1092 mutex_lock(&root
->fs_info
->trans_mutex
);
1093 if (BTRFS_I(inode
)->last_trans
<=
1094 root
->fs_info
->last_trans_committed
) {
1095 BTRFS_I(inode
)->last_trans
= 0;
1096 mutex_unlock(&root
->fs_info
->trans_mutex
);
1099 mutex_unlock(&root
->fs_info
->trans_mutex
);
1102 * ok we haven't committed the transaction yet, lets do a commit
1104 if (file
&& file
->private_data
)
1105 btrfs_ioctl_trans_end(file
);
1107 trans
= btrfs_start_transaction(root
, 1);
1113 ret
= btrfs_log_dentry_safe(trans
, root
, dentry
);
1117 /* we've logged all the items and now have a consistent
1118 * version of the file in the log. It is possible that
1119 * someone will come in and modify the file, but that's
1120 * fine because the log is consistent on disk, and we
1121 * have references to all of the file's extents
1123 * It is possible that someone will come in and log the
1124 * file again, but that will end up using the synchronization
1125 * inside btrfs_sync_log to keep things safe.
1127 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1129 if (ret
!= BTRFS_NO_LOG_SYNC
) {
1131 ret
= btrfs_commit_transaction(trans
, root
);
1133 ret
= btrfs_sync_log(trans
, root
);
1135 ret
= btrfs_end_transaction(trans
, root
);
1137 ret
= btrfs_commit_transaction(trans
, root
);
1140 ret
= btrfs_end_transaction(trans
, root
);
1142 mutex_lock(&dentry
->d_inode
->i_mutex
);
1144 return ret
> 0 ? -EIO
: ret
;
1147 static const struct vm_operations_struct btrfs_file_vm_ops
= {
1148 .fault
= filemap_fault
,
1149 .page_mkwrite
= btrfs_page_mkwrite
,
1152 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
1154 vma
->vm_ops
= &btrfs_file_vm_ops
;
1155 file_accessed(filp
);
1159 const struct file_operations btrfs_file_operations
= {
1160 .llseek
= generic_file_llseek
,
1161 .read
= do_sync_read
,
1162 .aio_read
= generic_file_aio_read
,
1163 .splice_read
= generic_file_splice_read
,
1164 .write
= btrfs_file_write
,
1165 .mmap
= btrfs_file_mmap
,
1166 .open
= generic_file_open
,
1167 .release
= btrfs_release_file
,
1168 .fsync
= btrfs_sync_file
,
1169 .unlocked_ioctl
= btrfs_ioctl
,
1170 #ifdef CONFIG_COMPAT
1171 .compat_ioctl
= btrfs_ioctl
,