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/smp_lock.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mpage.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.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
;
116 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
120 u64 end_of_last_block
;
121 u64 end_pos
= pos
+ write_bytes
;
122 loff_t isize
= i_size_read(inode
);
124 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
125 num_bytes
= (write_bytes
+ pos
- start_pos
+
126 root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
128 end_of_last_block
= start_pos
+ num_bytes
- 1;
130 lock_extent(io_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
131 trans
= btrfs_join_transaction(root
, 1);
136 btrfs_set_trans_block_group(trans
, inode
);
139 set_extent_uptodate(io_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
141 /* check for reserved extents on each page, we don't want
142 * to reset the delalloc bit on things that already have
145 btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
);
146 for (i
= 0; i
< num_pages
; i
++) {
147 struct page
*p
= pages
[i
];
152 if (end_pos
> isize
) {
153 i_size_write(inode
, end_pos
);
154 btrfs_update_inode(trans
, root
, inode
);
156 err
= btrfs_end_transaction(trans
, root
);
158 unlock_extent(io_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
163 * this drops all the extents in the cache that intersect the range
164 * [start, end]. Existing extents are split as required.
166 int btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
,
169 struct extent_map
*em
;
170 struct extent_map
*split
= NULL
;
171 struct extent_map
*split2
= NULL
;
172 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
173 u64 len
= end
- start
+ 1;
179 WARN_ON(end
< start
);
180 if (end
== (u64
)-1) {
186 split
= alloc_extent_map(GFP_NOFS
);
188 split2
= alloc_extent_map(GFP_NOFS
);
190 spin_lock(&em_tree
->lock
);
191 em
= lookup_extent_mapping(em_tree
, start
, len
);
193 spin_unlock(&em_tree
->lock
);
197 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
198 spin_unlock(&em_tree
->lock
);
199 if (em
->start
<= start
&&
200 (!testend
|| em
->start
+ em
->len
>= start
+ len
)) {
204 if (start
< em
->start
) {
205 len
= em
->start
- start
;
207 len
= start
+ len
- (em
->start
+ em
->len
);
208 start
= em
->start
+ em
->len
;
213 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
214 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
215 remove_extent_mapping(em_tree
, em
);
217 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
219 split
->start
= em
->start
;
220 split
->len
= start
- em
->start
;
221 split
->orig_start
= em
->orig_start
;
222 split
->block_start
= em
->block_start
;
225 split
->block_len
= em
->block_len
;
227 split
->block_len
= split
->len
;
229 split
->bdev
= em
->bdev
;
230 split
->flags
= flags
;
231 ret
= add_extent_mapping(em_tree
, split
);
233 free_extent_map(split
);
237 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
238 testend
&& em
->start
+ em
->len
> start
+ len
) {
239 u64 diff
= start
+ len
- em
->start
;
241 split
->start
= start
+ len
;
242 split
->len
= em
->start
+ em
->len
- (start
+ len
);
243 split
->bdev
= em
->bdev
;
244 split
->flags
= flags
;
247 split
->block_len
= em
->block_len
;
248 split
->block_start
= em
->block_start
;
249 split
->orig_start
= em
->orig_start
;
251 split
->block_len
= split
->len
;
252 split
->block_start
= em
->block_start
+ diff
;
253 split
->orig_start
= split
->start
;
256 ret
= add_extent_mapping(em_tree
, split
);
258 free_extent_map(split
);
261 spin_unlock(&em_tree
->lock
);
265 /* once for the tree*/
269 free_extent_map(split
);
271 free_extent_map(split2
);
275 int btrfs_check_file(struct btrfs_root
*root
, struct inode
*inode
)
279 struct btrfs_path
*path
;
280 struct btrfs_key found_key
;
281 struct extent_buffer
*leaf
;
282 struct btrfs_file_extent_item
*extent
;
291 path
= btrfs_alloc_path();
292 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, inode
->i_ino
,
295 nritems
= btrfs_header_nritems(path
->nodes
[0]);
296 if (path
->slots
[0] >= nritems
) {
297 ret
= btrfs_next_leaf(root
, path
);
300 nritems
= btrfs_header_nritems(path
->nodes
[0]);
302 slot
= path
->slots
[0];
303 leaf
= path
->nodes
[0];
304 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
305 if (found_key
.objectid
!= inode
->i_ino
)
307 if (found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
310 if (found_key
.offset
< last_offset
) {
312 btrfs_print_leaf(root
, leaf
);
313 printk(KERN_ERR
"inode %lu found offset %llu "
314 "expected %llu\n", inode
->i_ino
,
315 (unsigned long long)found_key
.offset
,
316 (unsigned long long)last_offset
);
320 extent
= btrfs_item_ptr(leaf
, slot
,
321 struct btrfs_file_extent_item
);
322 found_type
= btrfs_file_extent_type(leaf
, extent
);
323 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
324 extent_end
= found_key
.offset
+
325 btrfs_file_extent_num_bytes(leaf
, extent
);
326 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
327 struct btrfs_item
*item
;
328 item
= btrfs_item_nr(leaf
, slot
);
329 extent_end
= found_key
.offset
+
330 btrfs_file_extent_inline_len(leaf
, extent
);
331 extent_end
= (extent_end
+ root
->sectorsize
- 1) &
332 ~((u64
)root
->sectorsize
- 1);
334 last_offset
= extent_end
;
337 if (0 && last_offset
< inode
->i_size
) {
339 btrfs_print_leaf(root
, leaf
);
340 printk(KERN_ERR
"inode %lu found offset %llu size %llu\n",
341 inode
->i_ino
, (unsigned long long)last_offset
,
342 (unsigned long long)inode
->i_size
);
347 btrfs_free_path(path
);
353 * this is very complex, but the basic idea is to drop all extents
354 * in the range start - end. hint_block is filled in with a block number
355 * that would be a good hint to the block allocator for this file.
357 * If an extent intersects the range but is not entirely inside the range
358 * it is either truncated or split. Anything entirely inside the range
359 * is deleted from the tree.
361 * inline_limit is used to tell this code which offsets in the file to keep
362 * if they contain inline extents.
364 noinline
int btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
365 struct btrfs_root
*root
, struct inode
*inode
,
366 u64 start
, u64 end
, u64 inline_limit
, u64
*hint_byte
)
369 u64 locked_end
= end
;
370 u64 search_start
= start
;
377 u16 other_encoding
= 0;
380 struct extent_buffer
*leaf
;
381 struct btrfs_file_extent_item
*extent
;
382 struct btrfs_path
*path
;
383 struct btrfs_key key
;
384 struct btrfs_file_extent_item old
;
395 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
397 path
= btrfs_alloc_path();
402 btrfs_release_path(root
, path
);
403 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
408 if (path
->slots
[0] == 0) {
425 leaf
= path
->nodes
[0];
426 slot
= path
->slots
[0];
428 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
429 if (btrfs_key_type(&key
) == BTRFS_EXTENT_DATA_KEY
&&
433 if (btrfs_key_type(&key
) > BTRFS_EXTENT_DATA_KEY
||
434 key
.objectid
!= inode
->i_ino
) {
438 search_start
= max(key
.offset
, start
);
441 if (btrfs_key_type(&key
) == BTRFS_EXTENT_DATA_KEY
) {
442 extent
= btrfs_item_ptr(leaf
, slot
,
443 struct btrfs_file_extent_item
);
444 found_type
= btrfs_file_extent_type(leaf
, extent
);
445 compression
= btrfs_file_extent_compression(leaf
,
447 encryption
= btrfs_file_extent_encryption(leaf
,
449 other_encoding
= btrfs_file_extent_other_encoding(leaf
,
451 if (found_type
== BTRFS_FILE_EXTENT_REG
||
452 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
454 btrfs_file_extent_disk_bytenr(leaf
,
457 *hint_byte
= extent_end
;
459 extent_end
= key
.offset
+
460 btrfs_file_extent_num_bytes(leaf
, extent
);
461 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
,
464 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
466 extent_end
= key
.offset
+
467 btrfs_file_extent_inline_len(leaf
, extent
);
470 extent_end
= search_start
;
473 /* we found nothing we can drop */
474 if ((!found_extent
&& !found_inline
) ||
475 search_start
>= extent_end
) {
478 nritems
= btrfs_header_nritems(leaf
);
479 if (slot
>= nritems
- 1) {
480 nextret
= btrfs_next_leaf(root
, path
);
490 if (end
<= extent_end
&& start
>= key
.offset
&& found_inline
)
491 *hint_byte
= EXTENT_MAP_INLINE
;
494 read_extent_buffer(leaf
, &old
, (unsigned long)extent
,
496 root_gen
= btrfs_header_generation(leaf
);
497 root_owner
= btrfs_header_owner(leaf
);
498 leaf_start
= leaf
->start
;
501 if (end
< extent_end
&& end
>= key
.offset
) {
503 if (found_inline
&& start
<= key
.offset
)
507 if (bookend
&& found_extent
) {
508 if (locked_end
< extent_end
) {
509 ret
= try_lock_extent(&BTRFS_I(inode
)->io_tree
,
510 locked_end
, extent_end
- 1,
513 btrfs_release_path(root
, path
);
514 lock_extent(&BTRFS_I(inode
)->io_tree
,
515 locked_end
, extent_end
- 1,
517 locked_end
= extent_end
;
520 locked_end
= extent_end
;
522 orig_parent
= path
->nodes
[0]->start
;
523 disk_bytenr
= le64_to_cpu(old
.disk_bytenr
);
524 if (disk_bytenr
!= 0) {
525 ret
= btrfs_inc_extent_ref(trans
, root
,
527 le64_to_cpu(old
.disk_num_bytes
),
528 orig_parent
, root
->root_key
.objectid
,
529 trans
->transid
, inode
->i_ino
);
535 u64 mask
= root
->sectorsize
- 1;
536 search_start
= (extent_end
+ mask
) & ~mask
;
538 search_start
= extent_end
;
540 /* truncate existing extent */
541 if (start
> key
.offset
) {
545 WARN_ON(start
& (root
->sectorsize
- 1));
547 new_num
= start
- key
.offset
;
548 old_num
= btrfs_file_extent_num_bytes(leaf
,
551 btrfs_file_extent_disk_bytenr(leaf
,
553 if (btrfs_file_extent_disk_bytenr(leaf
,
555 inode_sub_bytes(inode
, old_num
-
558 btrfs_set_file_extent_num_bytes(leaf
,
560 btrfs_mark_buffer_dirty(leaf
);
561 } else if (key
.offset
< inline_limit
&&
562 (end
> extent_end
) &&
563 (inline_limit
< extent_end
)) {
565 new_size
= btrfs_file_extent_calc_inline_size(
566 inline_limit
- key
.offset
);
567 inode_sub_bytes(inode
, extent_end
-
569 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
571 if (!compression
&& !encryption
) {
572 btrfs_truncate_item(trans
, root
, path
,
577 /* delete the entire extent */
580 inode_sub_bytes(inode
, extent_end
-
582 ret
= btrfs_del_item(trans
, root
, path
);
583 /* TODO update progress marker and return */
586 btrfs_release_path(root
, path
);
587 /* the extent will be freed later */
589 if (bookend
&& found_inline
&& start
<= key
.offset
) {
591 new_size
= btrfs_file_extent_calc_inline_size(
593 inode_sub_bytes(inode
, end
- key
.offset
);
594 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
596 if (!compression
&& !encryption
)
597 ret
= btrfs_truncate_item(trans
, root
, path
,
601 /* create bookend, splitting the extent in two */
602 if (bookend
&& found_extent
) {
603 struct btrfs_key ins
;
604 ins
.objectid
= inode
->i_ino
;
606 btrfs_set_key_type(&ins
, BTRFS_EXTENT_DATA_KEY
);
608 btrfs_release_path(root
, path
);
609 path
->leave_spinning
= 1;
610 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
614 leaf
= path
->nodes
[0];
615 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
616 struct btrfs_file_extent_item
);
617 write_extent_buffer(leaf
, &old
,
618 (unsigned long)extent
, sizeof(old
));
620 btrfs_set_file_extent_compression(leaf
, extent
,
622 btrfs_set_file_extent_encryption(leaf
, extent
,
624 btrfs_set_file_extent_other_encoding(leaf
, extent
,
626 btrfs_set_file_extent_offset(leaf
, extent
,
627 le64_to_cpu(old
.offset
) + end
- key
.offset
);
628 WARN_ON(le64_to_cpu(old
.num_bytes
) <
630 btrfs_set_file_extent_num_bytes(leaf
, extent
,
634 * set the ram bytes to the size of the full extent
635 * before splitting. This is a worst case flag,
636 * but its the best we can do because we don't know
637 * how splitting affects compression
639 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
641 btrfs_set_file_extent_type(leaf
, extent
, found_type
);
643 btrfs_unlock_up_safe(path
, 1);
644 btrfs_mark_buffer_dirty(path
->nodes
[0]);
645 btrfs_set_lock_blocking(path
->nodes
[0]);
647 if (disk_bytenr
!= 0) {
648 ret
= btrfs_update_extent_ref(trans
, root
,
650 le64_to_cpu(old
.disk_num_bytes
),
653 root
->root_key
.objectid
,
654 trans
->transid
, ins
.objectid
);
658 path
->leave_spinning
= 0;
659 btrfs_release_path(root
, path
);
660 if (disk_bytenr
!= 0)
661 inode_add_bytes(inode
, extent_end
- end
);
664 if (found_extent
&& !keep
) {
665 u64 old_disk_bytenr
= le64_to_cpu(old
.disk_bytenr
);
667 if (old_disk_bytenr
!= 0) {
668 inode_sub_bytes(inode
,
669 le64_to_cpu(old
.num_bytes
));
670 ret
= btrfs_free_extent(trans
, root
,
672 le64_to_cpu(old
.disk_num_bytes
),
673 leaf_start
, root_owner
,
674 root_gen
, key
.objectid
, 0);
676 *hint_byte
= old_disk_bytenr
;
680 if (search_start
>= end
) {
686 btrfs_free_path(path
);
687 if (locked_end
> end
) {
688 unlock_extent(&BTRFS_I(inode
)->io_tree
, end
, locked_end
- 1,
691 btrfs_check_file(root
, inode
);
695 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
696 u64 objectid
, u64 bytenr
, u64
*start
, u64
*end
)
698 struct btrfs_file_extent_item
*fi
;
699 struct btrfs_key key
;
702 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
705 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
706 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
709 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
710 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
711 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
712 btrfs_file_extent_compression(leaf
, fi
) ||
713 btrfs_file_extent_encryption(leaf
, fi
) ||
714 btrfs_file_extent_other_encoding(leaf
, fi
))
717 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
718 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
727 * Mark extent in the range start - end as written.
729 * This changes extent type from 'pre-allocated' to 'regular'. If only
730 * part of extent is marked as written, the extent will be split into
733 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
734 struct btrfs_root
*root
,
735 struct inode
*inode
, u64 start
, u64 end
)
737 struct extent_buffer
*leaf
;
738 struct btrfs_path
*path
;
739 struct btrfs_file_extent_item
*fi
;
740 struct btrfs_key key
;
748 u64 locked_end
= end
;
754 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
756 path
= btrfs_alloc_path();
759 key
.objectid
= inode
->i_ino
;
760 key
.type
= BTRFS_EXTENT_DATA_KEY
;
764 key
.offset
= split
- 1;
766 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
767 if (ret
> 0 && path
->slots
[0] > 0)
770 leaf
= path
->nodes
[0];
771 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
772 BUG_ON(key
.objectid
!= inode
->i_ino
||
773 key
.type
!= BTRFS_EXTENT_DATA_KEY
);
774 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
775 struct btrfs_file_extent_item
);
776 extent_type
= btrfs_file_extent_type(leaf
, fi
);
777 BUG_ON(extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
);
778 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
779 BUG_ON(key
.offset
> start
|| extent_end
< end
);
781 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
782 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
783 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
785 if (key
.offset
== start
)
788 if (key
.offset
== start
&& extent_end
== end
) {
791 u64 leaf_owner
= btrfs_header_owner(leaf
);
792 u64 leaf_gen
= btrfs_header_generation(leaf
);
795 if (extent_mergeable(leaf
, path
->slots
[0] + 1, inode
->i_ino
,
796 bytenr
, &other_start
, &other_end
)) {
797 extent_end
= other_end
;
798 del_slot
= path
->slots
[0] + 1;
800 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
801 leaf
->start
, leaf_owner
,
802 leaf_gen
, inode
->i_ino
, 0);
807 if (extent_mergeable(leaf
, path
->slots
[0] - 1, inode
->i_ino
,
808 bytenr
, &other_start
, &other_end
)) {
809 key
.offset
= other_start
;
810 del_slot
= path
->slots
[0];
812 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
813 leaf
->start
, leaf_owner
,
814 leaf_gen
, inode
->i_ino
, 0);
819 btrfs_set_file_extent_type(leaf
, fi
,
820 BTRFS_FILE_EXTENT_REG
);
824 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
825 struct btrfs_file_extent_item
);
826 btrfs_set_file_extent_type(leaf
, fi
, BTRFS_FILE_EXTENT_REG
);
827 btrfs_set_file_extent_num_bytes(leaf
, fi
,
828 extent_end
- key
.offset
);
829 btrfs_mark_buffer_dirty(leaf
);
831 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
834 } else if (split
== start
) {
835 if (locked_end
< extent_end
) {
836 ret
= try_lock_extent(&BTRFS_I(inode
)->io_tree
,
837 locked_end
, extent_end
- 1, GFP_NOFS
);
839 btrfs_release_path(root
, path
);
840 lock_extent(&BTRFS_I(inode
)->io_tree
,
841 locked_end
, extent_end
- 1, GFP_NOFS
);
842 locked_end
= extent_end
;
845 locked_end
= extent_end
;
847 btrfs_set_file_extent_num_bytes(leaf
, fi
, split
- key
.offset
);
848 extent_offset
+= split
- key
.offset
;
850 BUG_ON(key
.offset
!= start
);
851 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
+
853 btrfs_set_file_extent_num_bytes(leaf
, fi
, extent_end
- split
);
855 btrfs_set_item_key_safe(trans
, root
, path
, &key
);
859 if (extent_end
== end
) {
861 extent_type
= BTRFS_FILE_EXTENT_REG
;
863 if (extent_end
== end
&& split
== start
) {
866 if (extent_mergeable(leaf
, path
->slots
[0] + 1, inode
->i_ino
,
867 bytenr
, &other_start
, &other_end
)) {
869 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
870 struct btrfs_file_extent_item
);
872 btrfs_set_item_key_safe(trans
, root
, path
, &key
);
873 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
874 btrfs_set_file_extent_num_bytes(leaf
, fi
,
879 if (extent_end
== end
&& split
== end
) {
882 if (extent_mergeable(leaf
, path
->slots
[0] - 1 , inode
->i_ino
,
883 bytenr
, &other_start
, &other_end
)) {
885 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
886 struct btrfs_file_extent_item
);
887 btrfs_set_file_extent_num_bytes(leaf
, fi
, extent_end
-
893 btrfs_mark_buffer_dirty(leaf
);
895 orig_parent
= leaf
->start
;
896 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
,
897 orig_parent
, root
->root_key
.objectid
,
898 trans
->transid
, inode
->i_ino
);
900 btrfs_release_path(root
, path
);
903 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(*fi
));
906 leaf
= path
->nodes
[0];
907 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
908 struct btrfs_file_extent_item
);
909 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
910 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
911 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, bytenr
);
912 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, num_bytes
);
913 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
914 btrfs_set_file_extent_num_bytes(leaf
, fi
, extent_end
- key
.offset
);
915 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
916 btrfs_set_file_extent_compression(leaf
, fi
, 0);
917 btrfs_set_file_extent_encryption(leaf
, fi
, 0);
918 btrfs_set_file_extent_other_encoding(leaf
, fi
, 0);
920 if (orig_parent
!= leaf
->start
) {
921 ret
= btrfs_update_extent_ref(trans
, root
, bytenr
, num_bytes
,
922 orig_parent
, leaf
->start
,
923 root
->root_key
.objectid
,
924 trans
->transid
, inode
->i_ino
);
928 btrfs_mark_buffer_dirty(leaf
);
931 btrfs_release_path(root
, path
);
932 if (split_end
&& split
== start
) {
936 if (locked_end
> end
) {
937 unlock_extent(&BTRFS_I(inode
)->io_tree
, end
, locked_end
- 1,
940 btrfs_free_path(path
);
945 * this gets pages into the page cache and locks them down, it also properly
946 * waits for data=ordered extents to finish before allowing the pages to be
949 static noinline
int prepare_pages(struct btrfs_root
*root
, struct file
*file
,
950 struct page
**pages
, size_t num_pages
,
951 loff_t pos
, unsigned long first_index
,
952 unsigned long last_index
, size_t write_bytes
)
955 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
956 struct inode
*inode
= fdentry(file
)->d_inode
;
961 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
962 last_pos
= ((u64
)index
+ num_pages
) << PAGE_CACHE_SHIFT
;
964 if (start_pos
> inode
->i_size
) {
965 err
= btrfs_cont_expand(inode
, start_pos
);
970 memset(pages
, 0, num_pages
* sizeof(struct page
*));
972 for (i
= 0; i
< num_pages
; i
++) {
973 pages
[i
] = grab_cache_page(inode
->i_mapping
, index
+ i
);
978 wait_on_page_writeback(pages
[i
]);
980 if (start_pos
< inode
->i_size
) {
981 struct btrfs_ordered_extent
*ordered
;
982 lock_extent(&BTRFS_I(inode
)->io_tree
,
983 start_pos
, last_pos
- 1, GFP_NOFS
);
984 ordered
= btrfs_lookup_first_ordered_extent(inode
,
987 ordered
->file_offset
+ ordered
->len
> start_pos
&&
988 ordered
->file_offset
< last_pos
) {
989 btrfs_put_ordered_extent(ordered
);
990 unlock_extent(&BTRFS_I(inode
)->io_tree
,
991 start_pos
, last_pos
- 1, GFP_NOFS
);
992 for (i
= 0; i
< num_pages
; i
++) {
993 unlock_page(pages
[i
]);
994 page_cache_release(pages
[i
]);
996 btrfs_wait_ordered_range(inode
, start_pos
,
997 last_pos
- start_pos
);
1001 btrfs_put_ordered_extent(ordered
);
1003 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, start_pos
,
1004 last_pos
- 1, EXTENT_DIRTY
| EXTENT_DELALLOC
,
1006 unlock_extent(&BTRFS_I(inode
)->io_tree
,
1007 start_pos
, last_pos
- 1, GFP_NOFS
);
1009 for (i
= 0; i
< num_pages
; i
++) {
1010 clear_page_dirty_for_io(pages
[i
]);
1011 set_page_extent_mapped(pages
[i
]);
1012 WARN_ON(!PageLocked(pages
[i
]));
1017 static ssize_t
btrfs_file_write(struct file
*file
, const char __user
*buf
,
1018 size_t count
, loff_t
*ppos
)
1022 ssize_t num_written
= 0;
1025 struct inode
*inode
= fdentry(file
)->d_inode
;
1026 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1027 struct page
**pages
= NULL
;
1029 struct page
*pinned
[2];
1030 unsigned long first_index
;
1031 unsigned long last_index
;
1034 will_write
= ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
) ||
1035 (file
->f_flags
& O_DIRECT
));
1037 nrptrs
= min((count
+ PAGE_CACHE_SIZE
- 1) / PAGE_CACHE_SIZE
,
1038 PAGE_CACHE_SIZE
/ (sizeof(struct page
*)));
1045 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
1046 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
1047 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1053 err
= file_remove_suid(file
);
1056 file_update_time(file
);
1058 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
1060 mutex_lock(&inode
->i_mutex
);
1061 BTRFS_I(inode
)->sequence
++;
1062 first_index
= pos
>> PAGE_CACHE_SHIFT
;
1063 last_index
= (pos
+ count
) >> PAGE_CACHE_SHIFT
;
1066 * there are lots of better ways to do this, but this code
1067 * makes sure the first and last page in the file range are
1068 * up to date and ready for cow
1070 if ((pos
& (PAGE_CACHE_SIZE
- 1))) {
1071 pinned
[0] = grab_cache_page(inode
->i_mapping
, first_index
);
1072 if (!PageUptodate(pinned
[0])) {
1073 ret
= btrfs_readpage(NULL
, pinned
[0]);
1075 wait_on_page_locked(pinned
[0]);
1077 unlock_page(pinned
[0]);
1080 if ((pos
+ count
) & (PAGE_CACHE_SIZE
- 1)) {
1081 pinned
[1] = grab_cache_page(inode
->i_mapping
, last_index
);
1082 if (!PageUptodate(pinned
[1])) {
1083 ret
= btrfs_readpage(NULL
, pinned
[1]);
1085 wait_on_page_locked(pinned
[1]);
1087 unlock_page(pinned
[1]);
1092 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1093 size_t write_bytes
= min(count
, nrptrs
*
1094 (size_t)PAGE_CACHE_SIZE
-
1096 size_t num_pages
= (write_bytes
+ PAGE_CACHE_SIZE
- 1) >>
1099 WARN_ON(num_pages
> nrptrs
);
1100 memset(pages
, 0, sizeof(struct page
*) * nrptrs
);
1102 ret
= btrfs_check_data_free_space(root
, inode
, write_bytes
);
1106 ret
= prepare_pages(root
, file
, pages
, num_pages
,
1107 pos
, first_index
, last_index
,
1110 btrfs_free_reserved_data_space(root
, inode
,
1115 ret
= btrfs_copy_from_user(pos
, num_pages
,
1116 write_bytes
, pages
, buf
);
1118 btrfs_free_reserved_data_space(root
, inode
,
1120 btrfs_drop_pages(pages
, num_pages
);
1124 ret
= dirty_and_release_pages(NULL
, root
, file
, pages
,
1125 num_pages
, pos
, write_bytes
);
1126 btrfs_drop_pages(pages
, num_pages
);
1128 btrfs_free_reserved_data_space(root
, inode
,
1134 btrfs_fdatawrite_range(inode
->i_mapping
, pos
,
1135 pos
+ write_bytes
- 1,
1138 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
1141 (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
1142 btrfs_btree_balance_dirty(root
, 1);
1143 btrfs_throttle(root
);
1147 count
-= write_bytes
;
1149 num_written
+= write_bytes
;
1154 mutex_unlock(&inode
->i_mutex
);
1161 page_cache_release(pinned
[0]);
1163 page_cache_release(pinned
[1]);
1167 * we want to make sure fsync finds this change
1168 * but we haven't joined a transaction running right now.
1170 * Later on, someone is sure to update the inode and get the
1171 * real transid recorded.
1173 * We set last_trans now to the fs_info generation + 1,
1174 * this will either be one more than the running transaction
1175 * or the generation used for the next transaction if there isn't
1176 * one running right now.
1178 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
+ 1;
1180 if (num_written
> 0 && will_write
) {
1181 struct btrfs_trans_handle
*trans
;
1183 err
= btrfs_wait_ordered_range(inode
, start_pos
, num_written
);
1187 if ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
)) {
1188 trans
= btrfs_start_transaction(root
, 1);
1189 ret
= btrfs_log_dentry_safe(trans
, root
,
1192 ret
= btrfs_sync_log(trans
, root
);
1194 btrfs_end_transaction(trans
, root
);
1196 btrfs_commit_transaction(trans
, root
);
1198 btrfs_commit_transaction(trans
, root
);
1201 if (file
->f_flags
& O_DIRECT
) {
1202 invalidate_mapping_pages(inode
->i_mapping
,
1203 start_pos
>> PAGE_CACHE_SHIFT
,
1204 (start_pos
+ num_written
- 1) >> PAGE_CACHE_SHIFT
);
1207 current
->backing_dev_info
= NULL
;
1208 return num_written
? num_written
: err
;
1211 int btrfs_release_file(struct inode
*inode
, struct file
*filp
)
1214 * ordered_data_close is set by settattr when we are about to truncate
1215 * a file from a non-zero size to a zero size. This tries to
1216 * flush down new bytes that may have been written if the
1217 * application were using truncate to replace a file in place.
1219 if (BTRFS_I(inode
)->ordered_data_close
) {
1220 BTRFS_I(inode
)->ordered_data_close
= 0;
1221 btrfs_add_ordered_operation(NULL
, BTRFS_I(inode
)->root
, inode
);
1222 if (inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
1223 filemap_flush(inode
->i_mapping
);
1225 if (filp
->private_data
)
1226 btrfs_ioctl_trans_end(filp
);
1231 * fsync call for both files and directories. This logs the inode into
1232 * the tree log instead of forcing full commits whenever possible.
1234 * It needs to call filemap_fdatawait so that all ordered extent updates are
1235 * in the metadata btree are up to date for copying to the log.
1237 * It drops the inode mutex before doing the tree log commit. This is an
1238 * important optimization for directories because holding the mutex prevents
1239 * new operations on the dir while we write to disk.
1241 int btrfs_sync_file(struct file
*file
, struct dentry
*dentry
, int datasync
)
1243 struct inode
*inode
= dentry
->d_inode
;
1244 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1246 struct btrfs_trans_handle
*trans
;
1249 * check the transaction that last modified this inode
1250 * and see if its already been committed
1252 if (!BTRFS_I(inode
)->last_trans
)
1255 mutex_lock(&root
->fs_info
->trans_mutex
);
1256 if (BTRFS_I(inode
)->last_trans
<=
1257 root
->fs_info
->last_trans_committed
) {
1258 BTRFS_I(inode
)->last_trans
= 0;
1259 mutex_unlock(&root
->fs_info
->trans_mutex
);
1262 mutex_unlock(&root
->fs_info
->trans_mutex
);
1265 filemap_fdatawrite(inode
->i_mapping
);
1266 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
1270 * ok we haven't committed the transaction yet, lets do a commit
1272 if (file
&& file
->private_data
)
1273 btrfs_ioctl_trans_end(file
);
1275 trans
= btrfs_start_transaction(root
, 1);
1281 ret
= btrfs_log_dentry_safe(trans
, root
, dentry
);
1285 /* we've logged all the items and now have a consistent
1286 * version of the file in the log. It is possible that
1287 * someone will come in and modify the file, but that's
1288 * fine because the log is consistent on disk, and we
1289 * have references to all of the file's extents
1291 * It is possible that someone will come in and log the
1292 * file again, but that will end up using the synchronization
1293 * inside btrfs_sync_log to keep things safe.
1295 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1298 ret
= btrfs_commit_transaction(trans
, root
);
1300 ret
= btrfs_sync_log(trans
, root
);
1302 ret
= btrfs_end_transaction(trans
, root
);
1304 ret
= btrfs_commit_transaction(trans
, root
);
1306 mutex_lock(&dentry
->d_inode
->i_mutex
);
1308 return ret
> 0 ? EIO
: ret
;
1311 static struct vm_operations_struct btrfs_file_vm_ops
= {
1312 .fault
= filemap_fault
,
1313 .page_mkwrite
= btrfs_page_mkwrite
,
1316 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
1318 vma
->vm_ops
= &btrfs_file_vm_ops
;
1319 file_accessed(filp
);
1323 struct file_operations btrfs_file_operations
= {
1324 .llseek
= generic_file_llseek
,
1325 .read
= do_sync_read
,
1326 .aio_read
= generic_file_aio_read
,
1327 .splice_read
= generic_file_splice_read
,
1328 .write
= btrfs_file_write
,
1329 .mmap
= btrfs_file_mmap
,
1330 .open
= generic_file_open
,
1331 .release
= btrfs_release_file
,
1332 .fsync
= btrfs_sync_file
,
1333 .unlocked_ioctl
= btrfs_ioctl
,
1334 #ifdef CONFIG_COMPAT
1335 .compat_ioctl
= btrfs_ioctl
,