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
->block_start
= em
->block_start
;
228 split
->block_len
= em
->block_len
;
230 split
->block_len
= split
->len
;
232 split
->bdev
= em
->bdev
;
233 split
->flags
= flags
;
234 ret
= add_extent_mapping(em_tree
, split
);
236 free_extent_map(split
);
240 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
241 testend
&& em
->start
+ em
->len
> start
+ len
) {
242 u64 diff
= start
+ len
- em
->start
;
244 split
->start
= start
+ len
;
245 split
->len
= em
->start
+ em
->len
- (start
+ len
);
246 split
->bdev
= em
->bdev
;
247 split
->flags
= flags
;
250 split
->block_len
= em
->block_len
;
251 split
->block_start
= em
->block_start
;
253 split
->block_len
= split
->len
;
254 split
->block_start
= em
->block_start
+ diff
;
257 ret
= add_extent_mapping(em_tree
, split
);
259 free_extent_map(split
);
262 spin_unlock(&em_tree
->lock
);
266 /* once for the tree*/
270 free_extent_map(split
);
272 free_extent_map(split2
);
276 int btrfs_check_file(struct btrfs_root
*root
, struct inode
*inode
)
280 struct btrfs_path
*path
;
281 struct btrfs_key found_key
;
282 struct extent_buffer
*leaf
;
283 struct btrfs_file_extent_item
*extent
;
292 path
= btrfs_alloc_path();
293 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, inode
->i_ino
,
296 nritems
= btrfs_header_nritems(path
->nodes
[0]);
297 if (path
->slots
[0] >= nritems
) {
298 ret
= btrfs_next_leaf(root
, path
);
301 nritems
= btrfs_header_nritems(path
->nodes
[0]);
303 slot
= path
->slots
[0];
304 leaf
= path
->nodes
[0];
305 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
306 if (found_key
.objectid
!= inode
->i_ino
)
308 if (found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
311 if (found_key
.offset
< last_offset
) {
313 btrfs_print_leaf(root
, leaf
);
314 printk("inode %lu found offset %Lu expected %Lu\n",
315 inode
->i_ino
, found_key
.offset
, last_offset
);
319 extent
= btrfs_item_ptr(leaf
, slot
,
320 struct btrfs_file_extent_item
);
321 found_type
= btrfs_file_extent_type(leaf
, extent
);
322 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
323 extent_end
= found_key
.offset
+
324 btrfs_file_extent_num_bytes(leaf
, extent
);
325 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
326 struct btrfs_item
*item
;
327 item
= btrfs_item_nr(leaf
, slot
);
328 extent_end
= found_key
.offset
+
329 btrfs_file_extent_inline_len(leaf
, extent
);
330 extent_end
= (extent_end
+ root
->sectorsize
- 1) &
331 ~((u64
)root
->sectorsize
-1 );
333 last_offset
= extent_end
;
336 if (0 && last_offset
< inode
->i_size
) {
338 btrfs_print_leaf(root
, leaf
);
339 printk("inode %lu found offset %Lu size %Lu\n", inode
->i_ino
,
340 last_offset
, inode
->i_size
);
345 btrfs_free_path(path
);
351 * this is very complex, but the basic idea is to drop all extents
352 * in the range start - end. hint_block is filled in with a block number
353 * that would be a good hint to the block allocator for this file.
355 * If an extent intersects the range but is not entirely inside the range
356 * it is either truncated or split. Anything entirely inside the range
357 * is deleted from the tree.
359 * inline_limit is used to tell this code which offsets in the file to keep
360 * if they contain inline extents.
362 int noinline
btrfs_drop_extents(struct btrfs_trans_handle
*trans
,
363 struct btrfs_root
*root
, struct inode
*inode
,
364 u64 start
, u64 end
, u64 inline_limit
, u64
*hint_byte
)
367 u64 locked_end
= end
;
368 u64 search_start
= start
;
375 u16 other_encoding
= 0;
378 struct extent_buffer
*leaf
;
379 struct btrfs_file_extent_item
*extent
;
380 struct btrfs_path
*path
;
381 struct btrfs_key key
;
382 struct btrfs_file_extent_item old
;
393 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
395 path
= btrfs_alloc_path();
400 btrfs_release_path(root
, path
);
401 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
406 if (path
->slots
[0] == 0) {
423 leaf
= path
->nodes
[0];
424 slot
= path
->slots
[0];
426 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
427 if (btrfs_key_type(&key
) == BTRFS_EXTENT_DATA_KEY
&&
431 if (btrfs_key_type(&key
) > BTRFS_EXTENT_DATA_KEY
||
432 key
.objectid
!= inode
->i_ino
) {
436 search_start
= key
.offset
;
439 if (btrfs_key_type(&key
) == BTRFS_EXTENT_DATA_KEY
) {
440 extent
= btrfs_item_ptr(leaf
, slot
,
441 struct btrfs_file_extent_item
);
442 found_type
= btrfs_file_extent_type(leaf
, extent
);
443 compression
= btrfs_file_extent_compression(leaf
,
445 encryption
= btrfs_file_extent_encryption(leaf
,
447 other_encoding
= btrfs_file_extent_other_encoding(leaf
,
449 if (found_type
== BTRFS_FILE_EXTENT_REG
||
450 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
452 btrfs_file_extent_disk_bytenr(leaf
,
455 *hint_byte
= extent_end
;
457 extent_end
= key
.offset
+
458 btrfs_file_extent_num_bytes(leaf
, extent
);
459 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
,
462 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
464 extent_end
= key
.offset
+
465 btrfs_file_extent_inline_len(leaf
, extent
);
468 extent_end
= search_start
;
471 /* we found nothing we can drop */
472 if ((!found_extent
&& !found_inline
) ||
473 search_start
>= extent_end
) {
476 nritems
= btrfs_header_nritems(leaf
);
477 if (slot
>= nritems
- 1) {
478 nextret
= btrfs_next_leaf(root
, path
);
488 if (end
<= extent_end
&& start
>= key
.offset
&& found_inline
)
489 *hint_byte
= EXTENT_MAP_INLINE
;
492 read_extent_buffer(leaf
, &old
, (unsigned long)extent
,
494 root_gen
= btrfs_header_generation(leaf
);
495 root_owner
= btrfs_header_owner(leaf
);
496 leaf_start
= leaf
->start
;
499 if (end
< extent_end
&& end
>= key
.offset
) {
501 if (found_inline
&& start
<= key
.offset
)
505 if (bookend
&& found_extent
) {
506 if (locked_end
< extent_end
) {
507 ret
= try_lock_extent(&BTRFS_I(inode
)->io_tree
,
508 locked_end
, extent_end
- 1,
511 btrfs_release_path(root
, path
);
512 lock_extent(&BTRFS_I(inode
)->io_tree
,
513 locked_end
, extent_end
- 1,
515 locked_end
= extent_end
;
518 locked_end
= extent_end
;
520 orig_parent
= path
->nodes
[0]->start
;
521 disk_bytenr
= le64_to_cpu(old
.disk_bytenr
);
522 if (disk_bytenr
!= 0) {
523 ret
= btrfs_inc_extent_ref(trans
, root
,
525 le64_to_cpu(old
.disk_num_bytes
),
526 orig_parent
, root
->root_key
.objectid
,
527 trans
->transid
, inode
->i_ino
);
533 u64 mask
= root
->sectorsize
- 1;
534 search_start
= (extent_end
+ mask
) & ~mask
;
536 search_start
= extent_end
;
538 /* truncate existing extent */
539 if (start
> key
.offset
) {
543 WARN_ON(start
& (root
->sectorsize
- 1));
545 new_num
= start
- key
.offset
;
546 old_num
= btrfs_file_extent_num_bytes(leaf
,
549 btrfs_file_extent_disk_bytenr(leaf
,
551 if (btrfs_file_extent_disk_bytenr(leaf
,
553 inode_sub_bytes(inode
, old_num
-
556 if (!compression
&& !encryption
) {
557 btrfs_set_file_extent_ram_bytes(leaf
,
560 btrfs_set_file_extent_num_bytes(leaf
,
562 btrfs_mark_buffer_dirty(leaf
);
563 } else if (key
.offset
< inline_limit
&&
564 (end
> extent_end
) &&
565 (inline_limit
< extent_end
)) {
567 new_size
= btrfs_file_extent_calc_inline_size(
568 inline_limit
- key
.offset
);
569 inode_sub_bytes(inode
, extent_end
-
571 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
573 if (!compression
&& !encryption
) {
574 btrfs_truncate_item(trans
, root
, path
,
579 /* delete the entire extent */
582 inode_sub_bytes(inode
, extent_end
-
584 ret
= btrfs_del_item(trans
, root
, path
);
585 /* TODO update progress marker and return */
588 btrfs_release_path(root
, path
);
589 /* the extent will be freed later */
591 if (bookend
&& found_inline
&& start
<= key
.offset
) {
593 new_size
= btrfs_file_extent_calc_inline_size(
595 inode_sub_bytes(inode
, end
- key
.offset
);
596 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
598 if (!compression
&& !encryption
)
599 ret
= btrfs_truncate_item(trans
, root
, path
,
603 /* create bookend, splitting the extent in two */
604 if (bookend
&& found_extent
) {
605 struct btrfs_key ins
;
606 ins
.objectid
= inode
->i_ino
;
608 btrfs_set_key_type(&ins
, BTRFS_EXTENT_DATA_KEY
);
610 btrfs_release_path(root
, path
);
611 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
615 leaf
= path
->nodes
[0];
616 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
617 struct btrfs_file_extent_item
);
618 write_extent_buffer(leaf
, &old
,
619 (unsigned long)extent
, sizeof(old
));
621 btrfs_set_file_extent_compression(leaf
, extent
,
623 btrfs_set_file_extent_encryption(leaf
, extent
,
625 btrfs_set_file_extent_other_encoding(leaf
, extent
,
627 btrfs_set_file_extent_offset(leaf
, extent
,
628 le64_to_cpu(old
.offset
) + end
- key
.offset
);
629 WARN_ON(le64_to_cpu(old
.num_bytes
) <
631 btrfs_set_file_extent_num_bytes(leaf
, extent
,
635 * set the ram bytes to the size of the full extent
636 * before splitting. This is a worst case flag,
637 * but its the best we can do because we don't know
638 * how splitting affects compression
640 btrfs_set_file_extent_ram_bytes(leaf
, extent
,
642 btrfs_set_file_extent_type(leaf
, extent
, found_type
);
644 btrfs_mark_buffer_dirty(path
->nodes
[0]);
646 if (disk_bytenr
!= 0) {
647 ret
= btrfs_update_extent_ref(trans
, root
,
648 disk_bytenr
, orig_parent
,
650 root
->root_key
.objectid
,
651 trans
->transid
, ins
.objectid
);
655 btrfs_release_path(root
, path
);
656 if (disk_bytenr
!= 0) {
657 inode_add_bytes(inode
, extent_end
- end
);
661 if (found_extent
&& !keep
) {
662 u64 disk_bytenr
= le64_to_cpu(old
.disk_bytenr
);
664 if (disk_bytenr
!= 0) {
665 inode_sub_bytes(inode
,
666 le64_to_cpu(old
.num_bytes
));
667 ret
= btrfs_free_extent(trans
, root
,
669 le64_to_cpu(old
.disk_num_bytes
),
670 leaf_start
, root_owner
,
671 root_gen
, key
.objectid
, 0);
673 *hint_byte
= disk_bytenr
;
677 if (search_start
>= end
) {
683 btrfs_free_path(path
);
684 if (locked_end
> end
) {
685 unlock_extent(&BTRFS_I(inode
)->io_tree
, end
, locked_end
- 1,
688 btrfs_check_file(root
, inode
);
692 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
693 u64 objectid
, u64 bytenr
, u64
*start
, u64
*end
)
695 struct btrfs_file_extent_item
*fi
;
696 struct btrfs_key key
;
699 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
702 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
703 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
706 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
707 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
708 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
709 btrfs_file_extent_compression(leaf
, fi
) ||
710 btrfs_file_extent_encryption(leaf
, fi
) ||
711 btrfs_file_extent_other_encoding(leaf
, fi
))
714 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
715 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
724 * Mark extent in the range start - end as written.
726 * This changes extent type from 'pre-allocated' to 'regular'. If only
727 * part of extent is marked as written, the extent will be split into
730 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
731 struct btrfs_root
*root
,
732 struct inode
*inode
, u64 start
, u64 end
)
734 struct extent_buffer
*leaf
;
735 struct btrfs_path
*path
;
736 struct btrfs_file_extent_item
*fi
;
737 struct btrfs_key key
;
745 u64 locked_end
= end
;
750 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
752 path
= btrfs_alloc_path();
755 key
.objectid
= inode
->i_ino
;
756 key
.type
= BTRFS_EXTENT_DATA_KEY
;
760 key
.offset
= split
- 1;
762 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
763 if (ret
> 0 && path
->slots
[0] > 0)
766 leaf
= path
->nodes
[0];
767 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
768 BUG_ON(key
.objectid
!= inode
->i_ino
||
769 key
.type
!= BTRFS_EXTENT_DATA_KEY
);
770 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
771 struct btrfs_file_extent_item
);
772 extent_type
= btrfs_file_extent_type(leaf
, fi
);
773 BUG_ON(extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
);
774 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
775 BUG_ON(key
.offset
> start
|| extent_end
< end
);
777 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
778 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
779 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
781 if (key
.offset
== start
)
784 if (key
.offset
== start
&& extent_end
== end
) {
787 u64 leaf_owner
= btrfs_header_owner(leaf
);
788 u64 leaf_gen
= btrfs_header_generation(leaf
);
791 if (extent_mergeable(leaf
, path
->slots
[0] + 1, inode
->i_ino
,
792 bytenr
, &other_start
, &other_end
)) {
793 extent_end
= other_end
;
794 del_slot
= path
->slots
[0] + 1;
796 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
797 leaf
->start
, leaf_owner
,
798 leaf_gen
, inode
->i_ino
, 0);
803 if (extent_mergeable(leaf
, path
->slots
[0] - 1, inode
->i_ino
,
804 bytenr
, &other_start
, &other_end
)) {
805 key
.offset
= other_start
;
806 del_slot
= path
->slots
[0];
808 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
809 leaf
->start
, leaf_owner
,
810 leaf_gen
, inode
->i_ino
, 0);
815 btrfs_set_file_extent_type(leaf
, fi
,
816 BTRFS_FILE_EXTENT_REG
);
820 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
821 struct btrfs_file_extent_item
);
822 btrfs_set_file_extent_type(leaf
, fi
, BTRFS_FILE_EXTENT_REG
);
823 btrfs_set_file_extent_num_bytes(leaf
, fi
,
824 extent_end
- key
.offset
);
825 btrfs_mark_buffer_dirty(leaf
);
827 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
830 } else if (split
== start
) {
831 if (locked_end
< extent_end
) {
832 ret
= try_lock_extent(&BTRFS_I(inode
)->io_tree
,
833 locked_end
, extent_end
- 1, GFP_NOFS
);
835 btrfs_release_path(root
, path
);
836 lock_extent(&BTRFS_I(inode
)->io_tree
,
837 locked_end
, extent_end
- 1, GFP_NOFS
);
838 locked_end
= extent_end
;
841 locked_end
= extent_end
;
843 btrfs_set_file_extent_num_bytes(leaf
, fi
, split
- key
.offset
);
844 extent_offset
+= split
- key
.offset
;
846 BUG_ON(key
.offset
!= start
);
847 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
+
849 btrfs_set_file_extent_num_bytes(leaf
, fi
, extent_end
- split
);
851 btrfs_set_item_key_safe(trans
, root
, path
, &key
);
855 if (extent_end
== end
) {
857 extent_type
= BTRFS_FILE_EXTENT_REG
;
859 if (extent_end
== end
&& split
== start
) {
862 if (extent_mergeable(leaf
, path
->slots
[0] + 1, inode
->i_ino
,
863 bytenr
, &other_start
, &other_end
)) {
865 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
866 struct btrfs_file_extent_item
);
868 btrfs_set_item_key_safe(trans
, root
, path
, &key
);
869 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
870 btrfs_set_file_extent_num_bytes(leaf
, fi
,
875 if (extent_end
== end
&& split
== end
) {
878 if (extent_mergeable(leaf
, path
->slots
[0] - 1 , inode
->i_ino
,
879 bytenr
, &other_start
, &other_end
)) {
881 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
882 struct btrfs_file_extent_item
);
883 btrfs_set_file_extent_num_bytes(leaf
, fi
, extent_end
-
889 btrfs_mark_buffer_dirty(leaf
);
890 btrfs_release_path(root
, path
);
893 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, sizeof(*fi
));
896 leaf
= path
->nodes
[0];
897 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
898 struct btrfs_file_extent_item
);
899 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
900 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
901 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, bytenr
);
902 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, num_bytes
);
903 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
904 btrfs_set_file_extent_num_bytes(leaf
, fi
, extent_end
- key
.offset
);
905 btrfs_set_file_extent_ram_bytes(leaf
, fi
, num_bytes
);
906 btrfs_set_file_extent_compression(leaf
, fi
, 0);
907 btrfs_set_file_extent_encryption(leaf
, fi
, 0);
908 btrfs_set_file_extent_other_encoding(leaf
, fi
, 0);
910 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
,
911 leaf
->start
, root
->root_key
.objectid
,
912 trans
->transid
, inode
->i_ino
);
915 btrfs_mark_buffer_dirty(leaf
);
916 btrfs_release_path(root
, path
);
917 if (split_end
&& split
== start
) {
921 if (locked_end
> end
) {
922 unlock_extent(&BTRFS_I(inode
)->io_tree
, end
, locked_end
- 1,
925 btrfs_free_path(path
);
930 * this gets pages into the page cache and locks them down, it also properly
931 * waits for data=ordered extents to finish before allowing the pages to be
934 static int noinline
prepare_pages(struct btrfs_root
*root
, struct file
*file
,
935 struct page
**pages
, size_t num_pages
,
936 loff_t pos
, unsigned long first_index
,
937 unsigned long last_index
, size_t write_bytes
)
940 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
941 struct inode
*inode
= fdentry(file
)->d_inode
;
946 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
947 last_pos
= ((u64
)index
+ num_pages
) << PAGE_CACHE_SHIFT
;
949 if (start_pos
> inode
->i_size
) {
950 err
= btrfs_cont_expand(inode
, start_pos
);
955 memset(pages
, 0, num_pages
* sizeof(struct page
*));
957 for (i
= 0; i
< num_pages
; i
++) {
958 pages
[i
] = grab_cache_page(inode
->i_mapping
, index
+ i
);
963 wait_on_page_writeback(pages
[i
]);
965 if (start_pos
< inode
->i_size
) {
966 struct btrfs_ordered_extent
*ordered
;
967 lock_extent(&BTRFS_I(inode
)->io_tree
,
968 start_pos
, last_pos
- 1, GFP_NOFS
);
969 ordered
= btrfs_lookup_first_ordered_extent(inode
, last_pos
-1);
971 ordered
->file_offset
+ ordered
->len
> start_pos
&&
972 ordered
->file_offset
< last_pos
) {
973 btrfs_put_ordered_extent(ordered
);
974 unlock_extent(&BTRFS_I(inode
)->io_tree
,
975 start_pos
, last_pos
- 1, GFP_NOFS
);
976 for (i
= 0; i
< num_pages
; i
++) {
977 unlock_page(pages
[i
]);
978 page_cache_release(pages
[i
]);
980 btrfs_wait_ordered_range(inode
, start_pos
,
981 last_pos
- start_pos
);
985 btrfs_put_ordered_extent(ordered
);
987 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, start_pos
,
988 last_pos
- 1, EXTENT_DIRTY
| EXTENT_DELALLOC
,
990 unlock_extent(&BTRFS_I(inode
)->io_tree
,
991 start_pos
, last_pos
- 1, GFP_NOFS
);
993 for (i
= 0; i
< num_pages
; i
++) {
994 clear_page_dirty_for_io(pages
[i
]);
995 set_page_extent_mapped(pages
[i
]);
996 WARN_ON(!PageLocked(pages
[i
]));
1001 static ssize_t
btrfs_file_write(struct file
*file
, const char __user
*buf
,
1002 size_t count
, loff_t
*ppos
)
1006 ssize_t num_written
= 0;
1009 struct inode
*inode
= fdentry(file
)->d_inode
;
1010 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1011 struct page
**pages
= NULL
;
1013 struct page
*pinned
[2];
1014 unsigned long first_index
;
1015 unsigned long last_index
;
1018 will_write
= ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
) ||
1019 (file
->f_flags
& O_DIRECT
));
1021 nrptrs
= min((count
+ PAGE_CACHE_SIZE
- 1) / PAGE_CACHE_SIZE
,
1022 PAGE_CACHE_SIZE
/ (sizeof(struct page
*)));
1029 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
1030 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
1031 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1037 err
= file_remove_suid(file
);
1040 file_update_time(file
);
1042 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
1044 mutex_lock(&inode
->i_mutex
);
1045 first_index
= pos
>> PAGE_CACHE_SHIFT
;
1046 last_index
= (pos
+ count
) >> PAGE_CACHE_SHIFT
;
1049 * if this is a nodatasum mount, force summing off for the inode
1050 * all the time. That way a later mount with summing on won't
1053 if (btrfs_test_opt(root
, NODATASUM
))
1054 btrfs_set_flag(inode
, NODATASUM
);
1057 * there are lots of better ways to do this, but this code
1058 * makes sure the first and last page in the file range are
1059 * up to date and ready for cow
1061 if ((pos
& (PAGE_CACHE_SIZE
- 1))) {
1062 pinned
[0] = grab_cache_page(inode
->i_mapping
, first_index
);
1063 if (!PageUptodate(pinned
[0])) {
1064 ret
= btrfs_readpage(NULL
, pinned
[0]);
1066 wait_on_page_locked(pinned
[0]);
1068 unlock_page(pinned
[0]);
1071 if ((pos
+ count
) & (PAGE_CACHE_SIZE
- 1)) {
1072 pinned
[1] = grab_cache_page(inode
->i_mapping
, last_index
);
1073 if (!PageUptodate(pinned
[1])) {
1074 ret
= btrfs_readpage(NULL
, pinned
[1]);
1076 wait_on_page_locked(pinned
[1]);
1078 unlock_page(pinned
[1]);
1083 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
1084 size_t write_bytes
= min(count
, nrptrs
*
1085 (size_t)PAGE_CACHE_SIZE
-
1087 size_t num_pages
= (write_bytes
+ PAGE_CACHE_SIZE
- 1) >>
1090 WARN_ON(num_pages
> nrptrs
);
1091 memset(pages
, 0, sizeof(pages
));
1093 ret
= btrfs_check_free_space(root
, write_bytes
, 0);
1097 ret
= prepare_pages(root
, file
, pages
, num_pages
,
1098 pos
, first_index
, last_index
,
1103 ret
= btrfs_copy_from_user(pos
, num_pages
,
1104 write_bytes
, pages
, buf
);
1106 btrfs_drop_pages(pages
, num_pages
);
1110 ret
= dirty_and_release_pages(NULL
, root
, file
, pages
,
1111 num_pages
, pos
, write_bytes
);
1112 btrfs_drop_pages(pages
, num_pages
);
1117 btrfs_fdatawrite_range(inode
->i_mapping
, pos
,
1118 pos
+ write_bytes
- 1,
1121 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
1124 (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
1125 btrfs_btree_balance_dirty(root
, 1);
1126 btrfs_throttle(root
);
1130 count
-= write_bytes
;
1132 num_written
+= write_bytes
;
1137 mutex_unlock(&inode
->i_mutex
);
1142 page_cache_release(pinned
[0]);
1144 page_cache_release(pinned
[1]);
1147 if (num_written
> 0 && will_write
) {
1148 struct btrfs_trans_handle
*trans
;
1150 err
= btrfs_wait_ordered_range(inode
, start_pos
, num_written
);
1154 if ((file
->f_flags
& O_SYNC
) || IS_SYNC(inode
)) {
1155 trans
= btrfs_start_transaction(root
, 1);
1156 ret
= btrfs_log_dentry_safe(trans
, root
,
1159 btrfs_sync_log(trans
, root
);
1160 btrfs_end_transaction(trans
, root
);
1162 btrfs_commit_transaction(trans
, root
);
1165 if (file
->f_flags
& O_DIRECT
) {
1166 invalidate_mapping_pages(inode
->i_mapping
,
1167 start_pos
>> PAGE_CACHE_SHIFT
,
1168 (start_pos
+ num_written
- 1) >> PAGE_CACHE_SHIFT
);
1171 current
->backing_dev_info
= NULL
;
1172 return num_written
? num_written
: err
;
1175 int btrfs_release_file(struct inode
* inode
, struct file
* filp
)
1177 if (filp
->private_data
)
1178 btrfs_ioctl_trans_end(filp
);
1183 * fsync call for both files and directories. This logs the inode into
1184 * the tree log instead of forcing full commits whenever possible.
1186 * It needs to call filemap_fdatawait so that all ordered extent updates are
1187 * in the metadata btree are up to date for copying to the log.
1189 * It drops the inode mutex before doing the tree log commit. This is an
1190 * important optimization for directories because holding the mutex prevents
1191 * new operations on the dir while we write to disk.
1193 int btrfs_sync_file(struct file
*file
, struct dentry
*dentry
, int datasync
)
1195 struct inode
*inode
= dentry
->d_inode
;
1196 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1198 struct btrfs_trans_handle
*trans
;
1201 * check the transaction that last modified this inode
1202 * and see if its already been committed
1204 if (!BTRFS_I(inode
)->last_trans
)
1207 mutex_lock(&root
->fs_info
->trans_mutex
);
1208 if (BTRFS_I(inode
)->last_trans
<=
1209 root
->fs_info
->last_trans_committed
) {
1210 BTRFS_I(inode
)->last_trans
= 0;
1211 mutex_unlock(&root
->fs_info
->trans_mutex
);
1214 mutex_unlock(&root
->fs_info
->trans_mutex
);
1216 root
->fs_info
->tree_log_batch
++;
1217 filemap_fdatawait(inode
->i_mapping
);
1218 root
->fs_info
->tree_log_batch
++;
1221 * ok we haven't committed the transaction yet, lets do a commit
1223 if (file
->private_data
)
1224 btrfs_ioctl_trans_end(file
);
1226 trans
= btrfs_start_transaction(root
, 1);
1232 ret
= btrfs_log_dentry_safe(trans
, root
, file
->f_dentry
);
1237 /* we've logged all the items and now have a consistent
1238 * version of the file in the log. It is possible that
1239 * someone will come in and modify the file, but that's
1240 * fine because the log is consistent on disk, and we
1241 * have references to all of the file's extents
1243 * It is possible that someone will come in and log the
1244 * file again, but that will end up using the synchronization
1245 * inside btrfs_sync_log to keep things safe.
1247 mutex_unlock(&file
->f_dentry
->d_inode
->i_mutex
);
1250 ret
= btrfs_commit_transaction(trans
, root
);
1252 btrfs_sync_log(trans
, root
);
1253 ret
= btrfs_end_transaction(trans
, root
);
1255 mutex_lock(&file
->f_dentry
->d_inode
->i_mutex
);
1257 return ret
> 0 ? EIO
: ret
;
1260 static struct vm_operations_struct btrfs_file_vm_ops
= {
1261 .fault
= filemap_fault
,
1262 .page_mkwrite
= btrfs_page_mkwrite
,
1265 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
1267 vma
->vm_ops
= &btrfs_file_vm_ops
;
1268 file_accessed(filp
);
1272 struct file_operations btrfs_file_operations
= {
1273 .llseek
= generic_file_llseek
,
1274 .read
= do_sync_read
,
1275 .aio_read
= generic_file_aio_read
,
1276 .splice_read
= generic_file_splice_read
,
1277 .write
= btrfs_file_write
,
1278 .mmap
= btrfs_file_mmap
,
1279 .open
= generic_file_open
,
1280 .release
= btrfs_release_file
,
1281 .fsync
= btrfs_sync_file
,
1282 .unlocked_ioctl
= btrfs_ioctl
,
1283 #ifdef CONFIG_COMPAT
1284 .compat_ioctl
= btrfs_ioctl
,