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>
32 #include <linux/version.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 /* simple helper to fault in pages and copy. This should go away
45 * and be replaced with calls into generic code.
47 static int noinline
btrfs_copy_from_user(loff_t pos
, int num_pages
,
49 struct page
**prepared_pages
,
50 const char __user
* buf
)
54 int offset
= pos
& (PAGE_CACHE_SIZE
- 1);
56 for (i
= 0; i
< num_pages
&& write_bytes
> 0; i
++, offset
= 0) {
57 size_t count
= min_t(size_t,
58 PAGE_CACHE_SIZE
- offset
, write_bytes
);
59 struct page
*page
= prepared_pages
[i
];
60 fault_in_pages_readable(buf
, count
);
62 /* Copy data from userspace to the current page */
64 page_fault
= __copy_from_user(page_address(page
) + offset
,
66 /* Flush processor's dcache for this page */
67 flush_dcache_page(page
);
75 return page_fault
? -EFAULT
: 0;
79 * unlocks pages after btrfs_file_write is done with them
81 static void noinline
btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
84 for (i
= 0; i
< num_pages
; i
++) {
87 /* page checked is some magic around finding pages that
88 * have been modified without going through btrfs_set_page_dirty
91 ClearPageChecked(pages
[i
]);
92 unlock_page(pages
[i
]);
93 mark_page_accessed(pages
[i
]);
94 page_cache_release(pages
[i
]);
99 * after copy_from_user, pages need to be dirtied and we need to make
100 * sure holes are created between the current EOF and the start of
101 * any next extents (if required).
103 * this also makes the decision about creating an inline extent vs
104 * doing real data extents, marking pages dirty and delalloc as required.
106 static int noinline
dirty_and_release_pages(struct btrfs_trans_handle
*trans
,
107 struct btrfs_root
*root
,
116 struct inode
*inode
= fdentry(file
)->d_inode
;
117 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
121 u64 end_of_last_block
;
122 u64 end_pos
= pos
+ write_bytes
;
123 loff_t isize
= i_size_read(inode
);
125 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
126 num_bytes
= (write_bytes
+ pos
- start_pos
+
127 root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
129 end_of_last_block
= start_pos
+ num_bytes
- 1;
131 lock_extent(io_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
132 trans
= btrfs_join_transaction(root
, 1);
137 btrfs_set_trans_block_group(trans
, inode
);
140 if ((end_of_last_block
& 4095) == 0) {
141 printk("strange end of last %Lu %zu %Lu\n", start_pos
, write_bytes
, end_of_last_block
);
143 set_extent_uptodate(io_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
145 /* check for reserved extents on each page, we don't want
146 * to reset the delalloc bit on things that already have
149 btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
);
150 for (i
= 0; i
< num_pages
; i
++) {
151 struct page
*p
= pages
[i
];
156 if (end_pos
> isize
) {
157 i_size_write(inode
, end_pos
);
158 btrfs_update_inode(trans
, root
, inode
);
160 err
= btrfs_end_transaction(trans
, root
);
162 unlock_extent(io_tree
, start_pos
, end_of_last_block
, GFP_NOFS
);
167 * this drops all the extents in the cache that intersect the range
168 * [start, end]. Existing extents are split as required.
170 int btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
,
173 struct extent_map
*em
;
174 struct extent_map
*split
= NULL
;
175 struct extent_map
*split2
= NULL
;
176 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
177 u64 len
= end
- start
+ 1;
183 WARN_ON(end
< start
);
184 if (end
== (u64
)-1) {
190 split
= alloc_extent_map(GFP_NOFS
);
192 split2
= alloc_extent_map(GFP_NOFS
);
194 spin_lock(&em_tree
->lock
);
195 em
= lookup_extent_mapping(em_tree
, start
, len
);
197 spin_unlock(&em_tree
->lock
);
201 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
202 spin_unlock(&em_tree
->lock
);
203 if (em
->start
<= start
&&
204 (!testend
|| em
->start
+ em
->len
>= start
+ len
)) {
208 if (start
< em
->start
) {
209 len
= em
->start
- start
;
211 len
= start
+ len
- (em
->start
+ em
->len
);
212 start
= em
->start
+ em
->len
;
217 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
218 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
219 remove_extent_mapping(em_tree
, em
);
221 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
223 split
->start
= em
->start
;
224 split
->len
= start
- em
->start
;
225 split
->orig_start
= em
->orig_start
;
226 split
->block_start
= em
->block_start
;
229 split
->block_len
= em
->block_len
;
231 split
->block_len
= split
->len
;
233 split
->bdev
= em
->bdev
;
234 split
->flags
= flags
;
235 ret
= add_extent_mapping(em_tree
, split
);
237 free_extent_map(split
);
241 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
242 testend
&& em
->start
+ em
->len
> start
+ len
) {
243 u64 diff
= start
+ len
- em
->start
;
245 split
->start
= start
+ len
;
246 split
->len
= em
->start
+ em
->len
- (start
+ len
);
247 split
->bdev
= em
->bdev
;
248 split
->flags
= flags
;
251 split
->block_len
= em
->block_len
;
252 split
->block_start
= em
->block_start
;
253 split
->orig_start
= em
->orig_start
;
255 split
->block_len
= split
->len
;
256 split
->block_start
= em
->block_start
+ diff
;
257 split
->orig_start
= split
->start
;
260 ret
= add_extent_mapping(em_tree
, split
);
262 free_extent_map(split
);
265 spin_unlock(&em_tree
->lock
);
269 /* once for the tree*/
273 free_extent_map(split
);
275 free_extent_map(split2
);
279 int btrfs_check_file(struct btrfs_root
*root
, struct inode
*inode
)
283 struct btrfs_path
*path
;
284 struct btrfs_key found_key
;
285 struct extent_buffer
*leaf
;
286 struct btrfs_file_extent_item
*extent
;
295 path
= btrfs_alloc_path();
296 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, inode
->i_ino
,
299 nritems
= btrfs_header_nritems(path
->nodes
[0]);
300 if (path
->slots
[0] >= nritems
) {
301 ret
= btrfs_next_leaf(root
, path
);
304 nritems
= btrfs_header_nritems(path
->nodes
[0]);
306 slot
= path
->slots
[0];
307 leaf
= path
->nodes
[0];
308 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
309 if (found_key
.objectid
!= inode
->i_ino
)
311 if (found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
314 if (found_key
.offset
< last_offset
) {
316 btrfs_print_leaf(root
, leaf
);
317 printk("inode %lu found offset %Lu expected %Lu\n",
318 inode
->i_ino
, found_key
.offset
, last_offset
);
322 extent
= btrfs_item_ptr(leaf
, slot
,
323 struct btrfs_file_extent_item
);
324 found_type
= btrfs_file_extent_type(leaf
, extent
);
325 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
326 extent_end
= found_key
.offset
+
327 btrfs_file_extent_num_bytes(leaf
, extent
);
328 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
329 struct btrfs_item
*item
;
330 item
= btrfs_item_nr(leaf
, slot
);
331 extent_end
= found_key
.offset
+
332 btrfs_file_extent_inline_len(leaf
, extent
);
333 extent_end
= (extent_end
+ root
->sectorsize
- 1) &
334 ~((u64
)root
->sectorsize
-1 );
336 last_offset
= extent_end
;
339 if (0 && last_offset
< inode
->i_size
) {
341 btrfs_print_leaf(root
, leaf
);
342 printk("inode %lu found offset %Lu size %Lu\n", inode
->i_ino
,
343 last_offset
, inode
->i_size
);
348 btrfs_free_path(path
);
354 * this is very complex, but the basic idea is to drop all extents
355 * in the range start - end. hint_block is filled in with a block number
356 * that would be a good hint to the block allocator for this file.
358 * If an extent intersects the range but is not entirely inside the range
359 * it is either truncated or split. Anything entirely inside the range
360 * is deleted from the tree.
362 * inline_limit is used to tell this code which offsets in the file to keep
363 * if they contain inline extents.
365 int noinline
btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
366 struct btrfs_root
*root
, struct inode
*inode
,
367 u64 start
, u64 end
, u64 inline_limit
, u64
*hint_byte
)
370 u64 locked_end
= end
;
371 u64 search_start
= start
;
378 u16 other_encoding
= 0;
381 struct extent_buffer
*leaf
;
382 struct btrfs_file_extent_item
*extent
;
383 struct btrfs_path
*path
;
384 struct btrfs_key key
;
385 struct btrfs_file_extent_item old
;
396 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
398 path
= btrfs_alloc_path();
403 btrfs_release_path(root
, path
);
404 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
409 if (path
->slots
[0] == 0) {
426 leaf
= path
->nodes
[0];
427 slot
= path
->slots
[0];
429 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
430 if (btrfs_key_type(&key
) == BTRFS_EXTENT_DATA_KEY
&&
434 if (btrfs_key_type(&key
) > BTRFS_EXTENT_DATA_KEY
||
435 key
.objectid
!= inode
->i_ino
) {
439 search_start
= max(key
.offset
, start
);
442 if (btrfs_key_type(&key
) == BTRFS_EXTENT_DATA_KEY
) {
443 extent
= btrfs_item_ptr(leaf
, slot
,
444 struct btrfs_file_extent_item
);
445 found_type
= btrfs_file_extent_type(leaf
, extent
);
446 compression
= btrfs_file_extent_compression(leaf
,
448 encryption
= btrfs_file_extent_encryption(leaf
,
450 other_encoding
= btrfs_file_extent_other_encoding(leaf
,
452 if (found_type
== BTRFS_FILE_EXTENT_REG
||
453 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
455 btrfs_file_extent_disk_bytenr(leaf
,
458 *hint_byte
= extent_end
;
460 extent_end
= key
.offset
+
461 btrfs_file_extent_num_bytes(leaf
, extent
);
462 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
,
465 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
467 extent_end
= key
.offset
+
468 btrfs_file_extent_inline_len(leaf
, extent
);
471 extent_end
= search_start
;
474 /* we found nothing we can drop */
475 if ((!found_extent
&& !found_inline
) ||
476 search_start
>= extent_end
) {
479 nritems
= btrfs_header_nritems(leaf
);
480 if (slot
>= nritems
- 1) {
481 nextret
= btrfs_next_leaf(root
, path
);
491 if (end
<= extent_end
&& start
>= key
.offset
&& found_inline
)
492 *hint_byte
= EXTENT_MAP_INLINE
;
495 read_extent_buffer(leaf
, &old
, (unsigned long)extent
,
497 root_gen
= btrfs_header_generation(leaf
);
498 root_owner
= btrfs_header_owner(leaf
);
499 leaf_start
= leaf
->start
;
502 if (end
< extent_end
&& end
>= key
.offset
) {
504 if (found_inline
&& start
<= key
.offset
)
508 if (bookend
&& found_extent
) {
509 if (locked_end
< extent_end
) {
510 ret
= try_lock_extent(&BTRFS_I(inode
)->io_tree
,
511 locked_end
, extent_end
- 1,
514 btrfs_release_path(root
, path
);
515 lock_extent(&BTRFS_I(inode
)->io_tree
,
516 locked_end
, extent_end
- 1,
518 locked_end
= extent_end
;
521 locked_end
= extent_end
;
523 orig_parent
= path
->nodes
[0]->start
;
524 disk_bytenr
= le64_to_cpu(old
.disk_bytenr
);
525 if (disk_bytenr
!= 0) {
526 ret
= btrfs_inc_extent_ref(trans
, root
,
528 le64_to_cpu(old
.disk_num_bytes
),
529 orig_parent
, root
->root_key
.objectid
,
530 trans
->transid
, inode
->i_ino
);
536 u64 mask
= root
->sectorsize
- 1;
537 search_start
= (extent_end
+ mask
) & ~mask
;
539 search_start
= extent_end
;
541 /* truncate existing extent */
542 if (start
> key
.offset
) {
546 WARN_ON(start
& (root
->sectorsize
- 1));
548 new_num
= start
- key
.offset
;
549 old_num
= btrfs_file_extent_num_bytes(leaf
,
552 btrfs_file_extent_disk_bytenr(leaf
,
554 if (btrfs_file_extent_disk_bytenr(leaf
,
556 inode_sub_bytes(inode
, old_num
-
559 if (!compression
&& !encryption
) {
560 btrfs_set_file_extent_ram_bytes(leaf
,
563 btrfs_set_file_extent_num_bytes(leaf
,
565 btrfs_mark_buffer_dirty(leaf
);
566 } else if (key
.offset
< inline_limit
&&
567 (end
> extent_end
) &&
568 (inline_limit
< extent_end
)) {
570 new_size
= btrfs_file_extent_calc_inline_size(
571 inline_limit
- key
.offset
);
572 inode_sub_bytes(inode
, extent_end
-
574 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
576 if (!compression
&& !encryption
) {
577 btrfs_truncate_item(trans
, root
, path
,
582 /* delete the entire extent */
585 inode_sub_bytes(inode
, extent_end
-
587 ret
= btrfs_del_item(trans
, root
, path
);
588 /* TODO update progress marker and return */
591 btrfs_release_path(root
, path
);
592 /* the extent will be freed later */
594 if (bookend
&& found_inline
&& start
<= key
.offset
) {
596 new_size
= btrfs_file_extent_calc_inline_size(
598 inode_sub_bytes(inode
, end
- key
.offset
);
599 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
601 if (!compression
&& !encryption
)
602 ret
= btrfs_truncate_item(trans
, root
, path
,
606 /* create bookend, splitting the extent in two */
607 if (bookend
&& found_extent
) {
608 struct btrfs_key ins
;
609 ins
.objectid
= inode
->i_ino
;
611 btrfs_set_key_type(&ins
, BTRFS_EXTENT_DATA_KEY
);
613 btrfs_release_path(root
, path
);
614 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
618 leaf
= path
->nodes
[0];
619 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
620 struct btrfs_file_extent_item
);
621 write_extent_buffer(leaf
, &old
,
622 (unsigned long)extent
, sizeof(old
));
624 btrfs_set_file_extent_compression(leaf
, extent
,
626 btrfs_set_file_extent_encryption(leaf
, extent
,
628 btrfs_set_file_extent_other_encoding(leaf
, extent
,
630 btrfs_set_file_extent_offset(leaf
, extent
,
631 le64_to_cpu(old
.offset
) + end
- key
.offset
);
632 WARN_ON(le64_to_cpu(old
.num_bytes
) <
634 btrfs_set_file_extent_num_bytes(leaf
, extent
,
638 * set the ram bytes to the size of the full extent
639 * before splitting. This is a worst case flag,
640 * but its the best we can do because we don't know
641 * how splitting affects compression
643 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
645 btrfs_set_file_extent_type(leaf
, extent
, found_type
);
647 btrfs_mark_buffer_dirty(path
->nodes
[0]);
649 if (disk_bytenr
!= 0) {
650 ret
= btrfs_update_extent_ref(trans
, root
,
651 disk_bytenr
, orig_parent
,
653 root
->root_key
.objectid
,
654 trans
->transid
, ins
.objectid
);
658 btrfs_release_path(root
, path
);
659 if (disk_bytenr
!= 0) {
660 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
,
922 orig_parent
, leaf
->start
,
923 root
->root_key
.objectid
,
924 trans
->transid
, inode
->i_ino
);
928 btrfs_mark_buffer_dirty(leaf
);
929 btrfs_release_path(root
, path
);
930 if (split_end
&& split
== start
) {
934 if (locked_end
> end
) {
935 unlock_extent(&BTRFS_I(inode
)->io_tree
, end
, locked_end
- 1,
938 btrfs_free_path(path
);
943 * this gets pages into the page cache and locks them down, it also properly
944 * waits for data=ordered extents to finish before allowing the pages to be
947 static int noinline
prepare_pages(struct btrfs_root
*root
, struct file
*file
,
948 struct page
**pages
, size_t num_pages
,
949 loff_t pos
, unsigned long first_index
,
950 unsigned long last_index
, size_t write_bytes
)
953 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
954 struct inode
*inode
= fdentry(file
)->d_inode
;
959 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
960 last_pos
= ((u64
)index
+ num_pages
) << PAGE_CACHE_SHIFT
;
962 if (start_pos
> inode
->i_size
) {
963 err
= btrfs_cont_expand(inode
, start_pos
);
968 memset(pages
, 0, num_pages
* sizeof(struct page
*));
970 for (i
= 0; i
< num_pages
; i
++) {
971 pages
[i
] = grab_cache_page(inode
->i_mapping
, index
+ i
);
976 wait_on_page_writeback(pages
[i
]);
978 if (start_pos
< inode
->i_size
) {
979 struct btrfs_ordered_extent
*ordered
;
980 lock_extent(&BTRFS_I(inode
)->io_tree
,
981 start_pos
, last_pos
- 1, GFP_NOFS
);
982 ordered
= btrfs_lookup_first_ordered_extent(inode
, last_pos
-1);
984 ordered
->file_offset
+ ordered
->len
> start_pos
&&
985 ordered
->file_offset
< last_pos
) {
986 btrfs_put_ordered_extent(ordered
);
987 unlock_extent(&BTRFS_I(inode
)->io_tree
,
988 start_pos
, last_pos
- 1, GFP_NOFS
);
989 for (i
= 0; i
< num_pages
; i
++) {
990 unlock_page(pages
[i
]);
991 page_cache_release(pages
[i
]);
993 btrfs_wait_ordered_range(inode
, start_pos
,
994 last_pos
- start_pos
);
998 btrfs_put_ordered_extent(ordered
);
1000 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, start_pos
,
1001 last_pos
- 1, EXTENT_DIRTY
| EXTENT_DELALLOC
,
1003 unlock_extent(&BTRFS_I(inode
)->io_tree
,
1004 start_pos
, last_pos
- 1, GFP_NOFS
);
1006 for (i
= 0; i
< num_pages
; i
++) {
1007 clear_page_dirty_for_io(pages
[i
]);
1008 set_page_extent_mapped(pages
[i
]);
1009 WARN_ON(!PageLocked(pages
[i
]));
1014 static ssize_t
btrfs_file_write(struct file
*file
, const char __user
*buf
,
1015 size_t count
, loff_t
*ppos
)
1019 ssize_t num_written
= 0;
1022 struct inode
*inode
= fdentry(file
)->d_inode
;
1023 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1024 struct page
**pages
= NULL
;
1026 struct page
*pinned
[2];
1027 unsigned long first_index
;
1028 unsigned long last_index
;
1031 will_write
= ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
) ||
1032 (file
->f_flags
& O_DIRECT
));
1034 nrptrs
= min((count
+ PAGE_CACHE_SIZE
- 1) / PAGE_CACHE_SIZE
,
1035 PAGE_CACHE_SIZE
/ (sizeof(struct page
*)));
1042 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
1043 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
1044 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1050 err
= file_remove_suid(file
);
1053 file_update_time(file
);
1055 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
1057 mutex_lock(&inode
->i_mutex
);
1058 first_index
= pos
>> PAGE_CACHE_SHIFT
;
1059 last_index
= (pos
+ count
) >> PAGE_CACHE_SHIFT
;
1062 * if this is a nodatasum mount, force summing off for the inode
1063 * all the time. That way a later mount with summing on won't
1066 if (btrfs_test_opt(root
, NODATASUM
))
1067 btrfs_set_flag(inode
, NODATASUM
);
1070 * there are lots of better ways to do this, but this code
1071 * makes sure the first and last page in the file range are
1072 * up to date and ready for cow
1074 if ((pos
& (PAGE_CACHE_SIZE
- 1))) {
1075 pinned
[0] = grab_cache_page(inode
->i_mapping
, first_index
);
1076 if (!PageUptodate(pinned
[0])) {
1077 ret
= btrfs_readpage(NULL
, pinned
[0]);
1079 wait_on_page_locked(pinned
[0]);
1081 unlock_page(pinned
[0]);
1084 if ((pos
+ count
) & (PAGE_CACHE_SIZE
- 1)) {
1085 pinned
[1] = grab_cache_page(inode
->i_mapping
, last_index
);
1086 if (!PageUptodate(pinned
[1])) {
1087 ret
= btrfs_readpage(NULL
, pinned
[1]);
1089 wait_on_page_locked(pinned
[1]);
1091 unlock_page(pinned
[1]);
1096 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1097 size_t write_bytes
= min(count
, nrptrs
*
1098 (size_t)PAGE_CACHE_SIZE
-
1100 size_t num_pages
= (write_bytes
+ PAGE_CACHE_SIZE
- 1) >>
1103 WARN_ON(num_pages
> nrptrs
);
1104 memset(pages
, 0, sizeof(pages
));
1106 ret
= btrfs_check_free_space(root
, write_bytes
, 0);
1110 ret
= prepare_pages(root
, file
, pages
, num_pages
,
1111 pos
, first_index
, last_index
,
1116 ret
= btrfs_copy_from_user(pos
, num_pages
,
1117 write_bytes
, pages
, buf
);
1119 btrfs_drop_pages(pages
, num_pages
);
1123 ret
= dirty_and_release_pages(NULL
, root
, file
, pages
,
1124 num_pages
, pos
, write_bytes
);
1125 btrfs_drop_pages(pages
, num_pages
);
1130 btrfs_fdatawrite_range(inode
->i_mapping
, pos
,
1131 pos
+ write_bytes
- 1,
1134 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
1137 (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
1138 btrfs_btree_balance_dirty(root
, 1);
1139 btrfs_throttle(root
);
1143 count
-= write_bytes
;
1145 num_written
+= write_bytes
;
1150 mutex_unlock(&inode
->i_mutex
);
1155 page_cache_release(pinned
[0]);
1157 page_cache_release(pinned
[1]);
1160 if (num_written
> 0 && will_write
) {
1161 struct btrfs_trans_handle
*trans
;
1163 err
= btrfs_wait_ordered_range(inode
, start_pos
, num_written
);
1167 if ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
)) {
1168 trans
= btrfs_start_transaction(root
, 1);
1169 ret
= btrfs_log_dentry_safe(trans
, root
,
1172 btrfs_sync_log(trans
, root
);
1173 btrfs_end_transaction(trans
, root
);
1175 btrfs_commit_transaction(trans
, root
);
1178 if (file
->f_flags
& O_DIRECT
) {
1179 invalidate_mapping_pages(inode
->i_mapping
,
1180 start_pos
>> PAGE_CACHE_SHIFT
,
1181 (start_pos
+ num_written
- 1) >> PAGE_CACHE_SHIFT
);
1184 current
->backing_dev_info
= NULL
;
1185 return num_written
? num_written
: err
;
1188 int btrfs_release_file(struct inode
* inode
, struct file
* filp
)
1190 if (filp
->private_data
)
1191 btrfs_ioctl_trans_end(filp
);
1196 * fsync call for both files and directories. This logs the inode into
1197 * the tree log instead of forcing full commits whenever possible.
1199 * It needs to call filemap_fdatawait so that all ordered extent updates are
1200 * in the metadata btree are up to date for copying to the log.
1202 * It drops the inode mutex before doing the tree log commit. This is an
1203 * important optimization for directories because holding the mutex prevents
1204 * new operations on the dir while we write to disk.
1206 int btrfs_sync_file(struct file
*file
, struct dentry
*dentry
, int datasync
)
1208 struct inode
*inode
= dentry
->d_inode
;
1209 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1211 struct btrfs_trans_handle
*trans
;
1214 * check the transaction that last modified this inode
1215 * and see if its already been committed
1217 if (!BTRFS_I(inode
)->last_trans
)
1220 mutex_lock(&root
->fs_info
->trans_mutex
);
1221 if (BTRFS_I(inode
)->last_trans
<=
1222 root
->fs_info
->last_trans_committed
) {
1223 BTRFS_I(inode
)->last_trans
= 0;
1224 mutex_unlock(&root
->fs_info
->trans_mutex
);
1227 mutex_unlock(&root
->fs_info
->trans_mutex
);
1229 root
->fs_info
->tree_log_batch
++;
1230 filemap_fdatawait(inode
->i_mapping
);
1231 root
->fs_info
->tree_log_batch
++;
1234 * ok we haven't committed the transaction yet, lets do a commit
1236 if (file
->private_data
)
1237 btrfs_ioctl_trans_end(file
);
1239 trans
= btrfs_start_transaction(root
, 1);
1245 ret
= btrfs_log_dentry_safe(trans
, root
, file
->f_dentry
);
1250 /* we've logged all the items and now have a consistent
1251 * version of the file in the log. It is possible that
1252 * someone will come in and modify the file, but that's
1253 * fine because the log is consistent on disk, and we
1254 * have references to all of the file's extents
1256 * It is possible that someone will come in and log the
1257 * file again, but that will end up using the synchronization
1258 * inside btrfs_sync_log to keep things safe.
1260 mutex_unlock(&file
->f_dentry
->d_inode
->i_mutex
);
1263 ret
= btrfs_commit_transaction(trans
, root
);
1265 btrfs_sync_log(trans
, root
);
1266 ret
= btrfs_end_transaction(trans
, root
);
1268 mutex_lock(&file
->f_dentry
->d_inode
->i_mutex
);
1270 return ret
> 0 ? EIO
: ret
;
1273 static struct vm_operations_struct btrfs_file_vm_ops
= {
1274 .fault
= filemap_fault
,
1275 .page_mkwrite
= btrfs_page_mkwrite
,
1278 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
1280 vma
->vm_ops
= &btrfs_file_vm_ops
;
1281 file_accessed(filp
);
1285 struct file_operations btrfs_file_operations
= {
1286 .llseek
= generic_file_llseek
,
1287 .read
= do_sync_read
,
1288 .aio_read
= generic_file_aio_read
,
1289 .splice_read
= generic_file_splice_read
,
1290 .write
= btrfs_file_write
,
1291 .mmap
= btrfs_file_mmap
,
1292 .open
= generic_file_open
,
1293 .release
= btrfs_release_file
,
1294 .fsync
= btrfs_sync_file
,
1295 .unlocked_ioctl
= btrfs_ioctl
,
1296 #ifdef CONFIG_COMPAT
1297 .compat_ioctl
= btrfs_ioctl
,