2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/backing-dev.h>
26 #include <linux/mpage.h>
27 #include <linux/falloc.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/slab.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
38 #include "print-tree.h"
44 /* simple helper to fault in pages and copy. This should go away
45 * and be replaced with calls into generic code.
47 static noinline
int btrfs_copy_from_user(loff_t pos
, int num_pages
,
49 struct page
**prepared_pages
,
53 size_t total_copied
= 0;
55 int offset
= pos
& (PAGE_CACHE_SIZE
- 1);
57 while (write_bytes
> 0) {
58 size_t count
= min_t(size_t,
59 PAGE_CACHE_SIZE
- offset
, write_bytes
);
60 struct page
*page
= prepared_pages
[pg
];
62 * Copy data from userspace to the current page
64 * Disable pagefault to avoid recursive lock since
65 * the pages are already locked
68 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, count
);
71 /* Flush processor's dcache for this page */
72 flush_dcache_page(page
);
75 * if we get a partial write, we can end up with
76 * partially up to date pages. These add
77 * a lot of complexity, so make sure they don't
78 * happen by forcing this copy to be retried.
80 * The rest of the btrfs_file_write code will fall
81 * back to page at a time copies after we return 0.
83 if (!PageUptodate(page
) && copied
< count
)
86 iov_iter_advance(i
, copied
);
87 write_bytes
-= copied
;
88 total_copied
+= copied
;
90 /* Return to btrfs_file_aio_write to fault page */
91 if (unlikely(copied
== 0))
94 if (unlikely(copied
< PAGE_CACHE_SIZE
- offset
)) {
105 * unlocks pages after btrfs_file_write is done with them
107 static noinline
void btrfs_drop_pages(struct page
**pages
, size_t num_pages
)
110 for (i
= 0; i
< num_pages
; i
++) {
111 /* page checked is some magic around finding pages that
112 * have been modified without going through btrfs_set_page_dirty
115 ClearPageChecked(pages
[i
]);
116 unlock_page(pages
[i
]);
117 mark_page_accessed(pages
[i
]);
118 page_cache_release(pages
[i
]);
123 * after copy_from_user, pages need to be dirtied and we need to make
124 * sure holes are created between the current EOF and the start of
125 * any next extents (if required).
127 * this also makes the decision about creating an inline extent vs
128 * doing real data extents, marking pages dirty and delalloc as required.
130 static noinline
int dirty_and_release_pages(struct btrfs_root
*root
,
139 struct inode
*inode
= fdentry(file
)->d_inode
;
142 u64 end_of_last_block
;
143 u64 end_pos
= pos
+ write_bytes
;
144 loff_t isize
= i_size_read(inode
);
146 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
147 num_bytes
= (write_bytes
+ pos
- start_pos
+
148 root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
150 end_of_last_block
= start_pos
+ num_bytes
- 1;
151 err
= btrfs_set_extent_delalloc(inode
, start_pos
, end_of_last_block
,
156 for (i
= 0; i
< num_pages
; i
++) {
157 struct page
*p
= pages
[i
];
164 * we've only changed i_size in ram, and we haven't updated
165 * the disk i_size. There is no need to log the inode
169 i_size_write(inode
, end_pos
);
174 * this drops all the extents in the cache that intersect the range
175 * [start, end]. Existing extents are split as required.
177 int btrfs_drop_extent_cache(struct inode
*inode
, u64 start
, u64 end
,
180 struct extent_map
*em
;
181 struct extent_map
*split
= NULL
;
182 struct extent_map
*split2
= NULL
;
183 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
184 u64 len
= end
- start
+ 1;
190 WARN_ON(end
< start
);
191 if (end
== (u64
)-1) {
197 split
= alloc_extent_map(GFP_NOFS
);
199 split2
= alloc_extent_map(GFP_NOFS
);
200 BUG_ON(!split
|| !split2
);
202 write_lock(&em_tree
->lock
);
203 em
= lookup_extent_mapping(em_tree
, start
, len
);
205 write_unlock(&em_tree
->lock
);
209 if (skip_pinned
&& test_bit(EXTENT_FLAG_PINNED
, &em
->flags
)) {
210 if (testend
&& em
->start
+ em
->len
>= start
+ len
) {
212 write_unlock(&em_tree
->lock
);
215 start
= em
->start
+ em
->len
;
217 len
= start
+ len
- (em
->start
+ em
->len
);
219 write_unlock(&em_tree
->lock
);
222 compressed
= test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
223 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
224 remove_extent_mapping(em_tree
, em
);
226 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
228 split
->start
= em
->start
;
229 split
->len
= start
- em
->start
;
230 split
->orig_start
= em
->orig_start
;
231 split
->block_start
= em
->block_start
;
234 split
->block_len
= em
->block_len
;
236 split
->block_len
= split
->len
;
238 split
->bdev
= em
->bdev
;
239 split
->flags
= flags
;
240 split
->compress_type
= em
->compress_type
;
241 ret
= add_extent_mapping(em_tree
, split
);
243 free_extent_map(split
);
247 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
248 testend
&& em
->start
+ em
->len
> start
+ len
) {
249 u64 diff
= start
+ len
- em
->start
;
251 split
->start
= start
+ len
;
252 split
->len
= em
->start
+ em
->len
- (start
+ len
);
253 split
->bdev
= em
->bdev
;
254 split
->flags
= flags
;
255 split
->compress_type
= em
->compress_type
;
258 split
->block_len
= em
->block_len
;
259 split
->block_start
= em
->block_start
;
260 split
->orig_start
= em
->orig_start
;
262 split
->block_len
= split
->len
;
263 split
->block_start
= em
->block_start
+ diff
;
264 split
->orig_start
= split
->start
;
267 ret
= add_extent_mapping(em_tree
, split
);
269 free_extent_map(split
);
272 write_unlock(&em_tree
->lock
);
276 /* once for the tree*/
280 free_extent_map(split
);
282 free_extent_map(split2
);
287 * this is very complex, but the basic idea is to drop all extents
288 * in the range start - end. hint_block is filled in with a block number
289 * that would be a good hint to the block allocator for this file.
291 * If an extent intersects the range but is not entirely inside the range
292 * it is either truncated or split. Anything entirely inside the range
293 * is deleted from the tree.
295 int btrfs_drop_extents(struct btrfs_trans_handle
*trans
, struct inode
*inode
,
296 u64 start
, u64 end
, u64
*hint_byte
, int drop_cache
)
298 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
299 struct extent_buffer
*leaf
;
300 struct btrfs_file_extent_item
*fi
;
301 struct btrfs_path
*path
;
302 struct btrfs_key key
;
303 struct btrfs_key new_key
;
304 u64 search_start
= start
;
307 u64 extent_offset
= 0;
316 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
318 path
= btrfs_alloc_path();
324 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
328 if (ret
> 0 && path
->slots
[0] > 0 && search_start
== start
) {
329 leaf
= path
->nodes
[0];
330 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0] - 1);
331 if (key
.objectid
== inode
->i_ino
&&
332 key
.type
== BTRFS_EXTENT_DATA_KEY
)
337 leaf
= path
->nodes
[0];
338 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
340 ret
= btrfs_next_leaf(root
, path
);
347 leaf
= path
->nodes
[0];
351 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
352 if (key
.objectid
> inode
->i_ino
||
353 key
.type
> BTRFS_EXTENT_DATA_KEY
|| key
.offset
>= end
)
356 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
357 struct btrfs_file_extent_item
);
358 extent_type
= btrfs_file_extent_type(leaf
, fi
);
360 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
361 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
362 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
363 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
364 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
365 extent_end
= key
.offset
+
366 btrfs_file_extent_num_bytes(leaf
, fi
);
367 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
368 extent_end
= key
.offset
+
369 btrfs_file_extent_inline_len(leaf
, fi
);
372 extent_end
= search_start
;
375 if (extent_end
<= search_start
) {
380 search_start
= max(key
.offset
, start
);
382 btrfs_release_path(root
, path
);
387 * | - range to drop - |
388 * | -------- extent -------- |
390 if (start
> key
.offset
&& end
< extent_end
) {
392 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
394 memcpy(&new_key
, &key
, sizeof(new_key
));
395 new_key
.offset
= start
;
396 ret
= btrfs_duplicate_item(trans
, root
, path
,
398 if (ret
== -EAGAIN
) {
399 btrfs_release_path(root
, path
);
405 leaf
= path
->nodes
[0];
406 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
407 struct btrfs_file_extent_item
);
408 btrfs_set_file_extent_num_bytes(leaf
, fi
,
411 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
412 struct btrfs_file_extent_item
);
414 extent_offset
+= start
- key
.offset
;
415 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
416 btrfs_set_file_extent_num_bytes(leaf
, fi
,
418 btrfs_mark_buffer_dirty(leaf
);
420 if (disk_bytenr
> 0) {
421 ret
= btrfs_inc_extent_ref(trans
, root
,
422 disk_bytenr
, num_bytes
, 0,
423 root
->root_key
.objectid
,
425 start
- extent_offset
);
427 *hint_byte
= disk_bytenr
;
432 * | ---- range to drop ----- |
433 * | -------- extent -------- |
435 if (start
<= key
.offset
&& end
< extent_end
) {
436 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
438 memcpy(&new_key
, &key
, sizeof(new_key
));
439 new_key
.offset
= end
;
440 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
442 extent_offset
+= end
- key
.offset
;
443 btrfs_set_file_extent_offset(leaf
, fi
, extent_offset
);
444 btrfs_set_file_extent_num_bytes(leaf
, fi
,
446 btrfs_mark_buffer_dirty(leaf
);
447 if (disk_bytenr
> 0) {
448 inode_sub_bytes(inode
, end
- key
.offset
);
449 *hint_byte
= disk_bytenr
;
454 search_start
= extent_end
;
456 * | ---- range to drop ----- |
457 * | -------- extent -------- |
459 if (start
> key
.offset
&& end
>= extent_end
) {
461 BUG_ON(extent_type
== BTRFS_FILE_EXTENT_INLINE
);
463 btrfs_set_file_extent_num_bytes(leaf
, fi
,
465 btrfs_mark_buffer_dirty(leaf
);
466 if (disk_bytenr
> 0) {
467 inode_sub_bytes(inode
, extent_end
- start
);
468 *hint_byte
= disk_bytenr
;
470 if (end
== extent_end
)
478 * | ---- range to drop ----- |
479 * | ------ extent ------ |
481 if (start
<= key
.offset
&& end
>= extent_end
) {
483 del_slot
= path
->slots
[0];
486 BUG_ON(del_slot
+ del_nr
!= path
->slots
[0]);
490 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
491 inode_sub_bytes(inode
,
492 extent_end
- key
.offset
);
493 extent_end
= ALIGN(extent_end
,
495 } else if (disk_bytenr
> 0) {
496 ret
= btrfs_free_extent(trans
, root
,
497 disk_bytenr
, num_bytes
, 0,
498 root
->root_key
.objectid
,
499 key
.objectid
, key
.offset
-
502 inode_sub_bytes(inode
,
503 extent_end
- key
.offset
);
504 *hint_byte
= disk_bytenr
;
507 if (end
== extent_end
)
510 if (path
->slots
[0] + 1 < btrfs_header_nritems(leaf
)) {
515 ret
= btrfs_del_items(trans
, root
, path
, del_slot
,
522 btrfs_release_path(root
, path
);
530 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
534 btrfs_free_path(path
);
538 static int extent_mergeable(struct extent_buffer
*leaf
, int slot
,
539 u64 objectid
, u64 bytenr
, u64 orig_offset
,
540 u64
*start
, u64
*end
)
542 struct btrfs_file_extent_item
*fi
;
543 struct btrfs_key key
;
546 if (slot
< 0 || slot
>= btrfs_header_nritems(leaf
))
549 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
550 if (key
.objectid
!= objectid
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
553 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
554 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
||
555 btrfs_file_extent_disk_bytenr(leaf
, fi
) != bytenr
||
556 btrfs_file_extent_offset(leaf
, fi
) != key
.offset
- orig_offset
||
557 btrfs_file_extent_compression(leaf
, fi
) ||
558 btrfs_file_extent_encryption(leaf
, fi
) ||
559 btrfs_file_extent_other_encoding(leaf
, fi
))
562 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
563 if ((*start
&& *start
!= key
.offset
) || (*end
&& *end
!= extent_end
))
572 * Mark extent in the range start - end as written.
574 * This changes extent type from 'pre-allocated' to 'regular'. If only
575 * part of extent is marked as written, the extent will be split into
578 int btrfs_mark_extent_written(struct btrfs_trans_handle
*trans
,
579 struct inode
*inode
, u64 start
, u64 end
)
581 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
582 struct extent_buffer
*leaf
;
583 struct btrfs_path
*path
;
584 struct btrfs_file_extent_item
*fi
;
585 struct btrfs_key key
;
586 struct btrfs_key new_key
;
599 btrfs_drop_extent_cache(inode
, start
, end
- 1, 0);
601 path
= btrfs_alloc_path();
606 key
.objectid
= inode
->i_ino
;
607 key
.type
= BTRFS_EXTENT_DATA_KEY
;
610 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
611 if (ret
> 0 && path
->slots
[0] > 0)
614 leaf
= path
->nodes
[0];
615 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
616 BUG_ON(key
.objectid
!= inode
->i_ino
||
617 key
.type
!= BTRFS_EXTENT_DATA_KEY
);
618 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
619 struct btrfs_file_extent_item
);
620 BUG_ON(btrfs_file_extent_type(leaf
, fi
) !=
621 BTRFS_FILE_EXTENT_PREALLOC
);
622 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
623 BUG_ON(key
.offset
> start
|| extent_end
< end
);
625 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
626 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
627 orig_offset
= key
.offset
- btrfs_file_extent_offset(leaf
, fi
);
628 memcpy(&new_key
, &key
, sizeof(new_key
));
630 if (start
== key
.offset
&& end
< extent_end
) {
633 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
634 inode
->i_ino
, bytenr
, orig_offset
,
635 &other_start
, &other_end
)) {
636 new_key
.offset
= end
;
637 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
638 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
639 struct btrfs_file_extent_item
);
640 btrfs_set_file_extent_num_bytes(leaf
, fi
,
642 btrfs_set_file_extent_offset(leaf
, fi
,
644 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
645 struct btrfs_file_extent_item
);
646 btrfs_set_file_extent_num_bytes(leaf
, fi
,
648 btrfs_mark_buffer_dirty(leaf
);
653 if (start
> key
.offset
&& end
== extent_end
) {
656 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
657 inode
->i_ino
, bytenr
, orig_offset
,
658 &other_start
, &other_end
)) {
659 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
660 struct btrfs_file_extent_item
);
661 btrfs_set_file_extent_num_bytes(leaf
, fi
,
664 new_key
.offset
= start
;
665 btrfs_set_item_key_safe(trans
, root
, path
, &new_key
);
667 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
668 struct btrfs_file_extent_item
);
669 btrfs_set_file_extent_num_bytes(leaf
, fi
,
671 btrfs_set_file_extent_offset(leaf
, fi
,
672 start
- orig_offset
);
673 btrfs_mark_buffer_dirty(leaf
);
678 while (start
> key
.offset
|| end
< extent_end
) {
679 if (key
.offset
== start
)
682 new_key
.offset
= split
;
683 ret
= btrfs_duplicate_item(trans
, root
, path
, &new_key
);
684 if (ret
== -EAGAIN
) {
685 btrfs_release_path(root
, path
);
690 leaf
= path
->nodes
[0];
691 fi
= btrfs_item_ptr(leaf
, path
->slots
[0] - 1,
692 struct btrfs_file_extent_item
);
693 btrfs_set_file_extent_num_bytes(leaf
, fi
,
696 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
697 struct btrfs_file_extent_item
);
699 btrfs_set_file_extent_offset(leaf
, fi
, split
- orig_offset
);
700 btrfs_set_file_extent_num_bytes(leaf
, fi
,
702 btrfs_mark_buffer_dirty(leaf
);
704 ret
= btrfs_inc_extent_ref(trans
, root
, bytenr
, num_bytes
, 0,
705 root
->root_key
.objectid
,
706 inode
->i_ino
, orig_offset
);
709 if (split
== start
) {
712 BUG_ON(start
!= key
.offset
);
721 if (extent_mergeable(leaf
, path
->slots
[0] + 1,
722 inode
->i_ino
, bytenr
, orig_offset
,
723 &other_start
, &other_end
)) {
725 btrfs_release_path(root
, path
);
728 extent_end
= other_end
;
729 del_slot
= path
->slots
[0] + 1;
731 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
732 0, root
->root_key
.objectid
,
733 inode
->i_ino
, orig_offset
);
738 if (extent_mergeable(leaf
, path
->slots
[0] - 1,
739 inode
->i_ino
, bytenr
, orig_offset
,
740 &other_start
, &other_end
)) {
742 btrfs_release_path(root
, path
);
745 key
.offset
= other_start
;
746 del_slot
= path
->slots
[0];
748 ret
= btrfs_free_extent(trans
, root
, bytenr
, num_bytes
,
749 0, root
->root_key
.objectid
,
750 inode
->i_ino
, orig_offset
);
754 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
755 struct btrfs_file_extent_item
);
756 btrfs_set_file_extent_type(leaf
, fi
,
757 BTRFS_FILE_EXTENT_REG
);
758 btrfs_mark_buffer_dirty(leaf
);
760 fi
= btrfs_item_ptr(leaf
, del_slot
- 1,
761 struct btrfs_file_extent_item
);
762 btrfs_set_file_extent_type(leaf
, fi
,
763 BTRFS_FILE_EXTENT_REG
);
764 btrfs_set_file_extent_num_bytes(leaf
, fi
,
765 extent_end
- key
.offset
);
766 btrfs_mark_buffer_dirty(leaf
);
768 ret
= btrfs_del_items(trans
, root
, path
, del_slot
, del_nr
);
772 btrfs_free_path(path
);
777 * on error we return an unlocked page and the error value
778 * on success we return a locked page and 0
780 static int prepare_uptodate_page(struct page
*page
, u64 pos
)
784 if ((pos
& (PAGE_CACHE_SIZE
- 1)) && !PageUptodate(page
)) {
785 ret
= btrfs_readpage(NULL
, page
);
789 if (!PageUptodate(page
)) {
798 * this gets pages into the page cache and locks them down, it also properly
799 * waits for data=ordered extents to finish before allowing the pages to be
802 static noinline
int prepare_pages(struct btrfs_root
*root
, struct file
*file
,
803 struct page
**pages
, size_t num_pages
,
804 loff_t pos
, unsigned long first_index
,
805 unsigned long last_index
, size_t write_bytes
)
807 struct extent_state
*cached_state
= NULL
;
809 unsigned long index
= pos
>> PAGE_CACHE_SHIFT
;
810 struct inode
*inode
= fdentry(file
)->d_inode
;
816 start_pos
= pos
& ~((u64
)root
->sectorsize
- 1);
817 last_pos
= ((u64
)index
+ num_pages
) << PAGE_CACHE_SHIFT
;
819 if (start_pos
> inode
->i_size
) {
820 err
= btrfs_cont_expand(inode
, i_size_read(inode
), start_pos
);
826 for (i
= 0; i
< num_pages
; i
++) {
827 pages
[i
] = grab_cache_page(inode
->i_mapping
, index
+ i
);
835 err
= prepare_uptodate_page(pages
[i
], pos
);
836 if (i
== num_pages
- 1)
837 err
= prepare_uptodate_page(pages
[i
],
840 page_cache_release(pages
[i
]);
844 wait_on_page_writeback(pages
[i
]);
847 if (start_pos
< inode
->i_size
) {
848 struct btrfs_ordered_extent
*ordered
;
849 lock_extent_bits(&BTRFS_I(inode
)->io_tree
,
850 start_pos
, last_pos
- 1, 0, &cached_state
,
852 ordered
= btrfs_lookup_first_ordered_extent(inode
,
855 ordered
->file_offset
+ ordered
->len
> start_pos
&&
856 ordered
->file_offset
< last_pos
) {
857 btrfs_put_ordered_extent(ordered
);
858 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
859 start_pos
, last_pos
- 1,
860 &cached_state
, GFP_NOFS
);
861 for (i
= 0; i
< num_pages
; i
++) {
862 unlock_page(pages
[i
]);
863 page_cache_release(pages
[i
]);
865 btrfs_wait_ordered_range(inode
, start_pos
,
866 last_pos
- start_pos
);
870 btrfs_put_ordered_extent(ordered
);
872 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start_pos
,
873 last_pos
- 1, EXTENT_DIRTY
| EXTENT_DELALLOC
|
874 EXTENT_DO_ACCOUNTING
, 0, 0, &cached_state
,
876 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
877 start_pos
, last_pos
- 1, &cached_state
,
880 for (i
= 0; i
< num_pages
; i
++) {
881 clear_page_dirty_for_io(pages
[i
]);
882 set_page_extent_mapped(pages
[i
]);
883 WARN_ON(!PageLocked(pages
[i
]));
888 unlock_page(pages
[faili
]);
889 page_cache_release(pages
[faili
]);
896 static noinline ssize_t
__btrfs_buffered_write(struct file
*file
,
900 struct inode
*inode
= fdentry(file
)->d_inode
;
901 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
902 struct page
**pages
= NULL
;
903 unsigned long first_index
;
904 unsigned long last_index
;
905 size_t num_written
= 0;
909 nrptrs
= min((iov_iter_count(i
) + PAGE_CACHE_SIZE
- 1) /
910 PAGE_CACHE_SIZE
, PAGE_CACHE_SIZE
/
911 (sizeof(struct page
*)));
912 pages
= kmalloc(nrptrs
* sizeof(struct page
*), GFP_KERNEL
);
916 first_index
= pos
>> PAGE_CACHE_SHIFT
;
917 last_index
= (pos
+ iov_iter_count(i
)) >> PAGE_CACHE_SHIFT
;
919 while (iov_iter_count(i
) > 0) {
920 size_t offset
= pos
& (PAGE_CACHE_SIZE
- 1);
921 size_t write_bytes
= min(iov_iter_count(i
),
922 nrptrs
* (size_t)PAGE_CACHE_SIZE
-
924 size_t num_pages
= (write_bytes
+ offset
+
925 PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
929 WARN_ON(num_pages
> nrptrs
);
932 * Fault pages before locking them in prepare_pages
933 * to avoid recursive lock
935 if (unlikely(iov_iter_fault_in_readable(i
, write_bytes
))) {
940 ret
= btrfs_delalloc_reserve_space(inode
,
941 num_pages
<< PAGE_CACHE_SHIFT
);
946 * This is going to setup the pages array with the number of
947 * pages we want, so we don't really need to worry about the
948 * contents of pages from loop to loop
950 ret
= prepare_pages(root
, file
, pages
, num_pages
,
951 pos
, first_index
, last_index
,
954 btrfs_delalloc_release_space(inode
,
955 num_pages
<< PAGE_CACHE_SHIFT
);
959 copied
= btrfs_copy_from_user(pos
, num_pages
,
960 write_bytes
, pages
, i
);
963 * if we have trouble faulting in the pages, fall
964 * back to one page at a time
966 if (copied
< write_bytes
)
972 dirty_pages
= (copied
+ offset
+
973 PAGE_CACHE_SIZE
- 1) >>
977 * If we had a short copy we need to release the excess delaloc
978 * bytes we reserved. We need to increment outstanding_extents
979 * because btrfs_delalloc_release_space will decrement it, but
980 * we still have an outstanding extent for the chunk we actually
983 if (num_pages
> dirty_pages
) {
986 &BTRFS_I(inode
)->outstanding_extents
);
987 btrfs_delalloc_release_space(inode
,
988 (num_pages
- dirty_pages
) <<
993 ret
= dirty_and_release_pages(root
, file
, pages
,
997 btrfs_delalloc_release_space(inode
,
998 dirty_pages
<< PAGE_CACHE_SHIFT
);
999 btrfs_drop_pages(pages
, num_pages
);
1004 btrfs_drop_pages(pages
, num_pages
);
1008 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
,
1010 if (dirty_pages
< (root
->leafsize
>> PAGE_CACHE_SHIFT
) + 1)
1011 btrfs_btree_balance_dirty(root
, 1);
1012 btrfs_throttle(root
);
1015 num_written
+= copied
;
1020 return num_written
? num_written
: ret
;
1023 static ssize_t
__btrfs_direct_write(struct kiocb
*iocb
,
1024 const struct iovec
*iov
,
1025 unsigned long nr_segs
, loff_t pos
,
1026 loff_t
*ppos
, size_t count
, size_t ocount
)
1028 struct file
*file
= iocb
->ki_filp
;
1029 struct inode
*inode
= fdentry(file
)->d_inode
;
1032 ssize_t written_buffered
;
1036 written
= generic_file_direct_write(iocb
, iov
, &nr_segs
, pos
, ppos
,
1040 * the generic O_DIRECT will update in-memory i_size after the
1041 * DIOs are done. But our endio handlers that update the on
1042 * disk i_size never update past the in memory i_size. So we
1043 * need one more update here to catch any additions to the
1046 if (inode
->i_size
!= BTRFS_I(inode
)->disk_i_size
) {
1047 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
1048 mark_inode_dirty(inode
);
1051 if (written
< 0 || written
== count
)
1056 iov_iter_init(&i
, iov
, nr_segs
, count
, written
);
1057 written_buffered
= __btrfs_buffered_write(file
, &i
, pos
);
1058 if (written_buffered
< 0) {
1059 err
= written_buffered
;
1062 endbyte
= pos
+ written_buffered
- 1;
1063 err
= filemap_write_and_wait_range(file
->f_mapping
, pos
, endbyte
);
1066 written
+= written_buffered
;
1067 *ppos
= pos
+ written_buffered
;
1068 invalidate_mapping_pages(file
->f_mapping
, pos
>> PAGE_CACHE_SHIFT
,
1069 endbyte
>> PAGE_CACHE_SHIFT
);
1071 return written
? written
: err
;
1074 static ssize_t
btrfs_file_aio_write(struct kiocb
*iocb
,
1075 const struct iovec
*iov
,
1076 unsigned long nr_segs
, loff_t pos
)
1078 struct file
*file
= iocb
->ki_filp
;
1079 struct inode
*inode
= fdentry(file
)->d_inode
;
1080 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1081 loff_t
*ppos
= &iocb
->ki_pos
;
1082 ssize_t num_written
= 0;
1084 size_t count
, ocount
;
1086 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
1088 mutex_lock(&inode
->i_mutex
);
1090 err
= generic_segment_checks(iov
, &nr_segs
, &ocount
, VERIFY_READ
);
1092 mutex_unlock(&inode
->i_mutex
);
1097 current
->backing_dev_info
= inode
->i_mapping
->backing_dev_info
;
1098 err
= generic_write_checks(file
, &pos
, &count
, S_ISBLK(inode
->i_mode
));
1100 mutex_unlock(&inode
->i_mutex
);
1105 mutex_unlock(&inode
->i_mutex
);
1109 err
= file_remove_suid(file
);
1111 mutex_unlock(&inode
->i_mutex
);
1116 * If BTRFS flips readonly due to some impossible error
1117 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1118 * although we have opened a file as writable, we have
1119 * to stop this write operation to ensure FS consistency.
1121 if (root
->fs_info
->fs_state
& BTRFS_SUPER_FLAG_ERROR
) {
1122 mutex_unlock(&inode
->i_mutex
);
1127 file_update_time(file
);
1128 BTRFS_I(inode
)->sequence
++;
1130 if (unlikely(file
->f_flags
& O_DIRECT
)) {
1131 num_written
= __btrfs_direct_write(iocb
, iov
, nr_segs
,
1132 pos
, ppos
, count
, ocount
);
1136 iov_iter_init(&i
, iov
, nr_segs
, count
, num_written
);
1138 num_written
= __btrfs_buffered_write(file
, &i
, pos
);
1139 if (num_written
> 0)
1140 *ppos
= pos
+ num_written
;
1143 mutex_unlock(&inode
->i_mutex
);
1146 * we want to make sure fsync finds this change
1147 * but we haven't joined a transaction running right now.
1149 * Later on, someone is sure to update the inode and get the
1150 * real transid recorded.
1152 * We set last_trans now to the fs_info generation + 1,
1153 * this will either be one more than the running transaction
1154 * or the generation used for the next transaction if there isn't
1155 * one running right now.
1157 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
+ 1;
1158 if (num_written
> 0 || num_written
== -EIOCBQUEUED
) {
1159 err
= generic_write_sync(file
, pos
, num_written
);
1160 if (err
< 0 && num_written
> 0)
1164 current
->backing_dev_info
= NULL
;
1165 return num_written
? num_written
: err
;
1168 int btrfs_release_file(struct inode
*inode
, struct file
*filp
)
1171 * ordered_data_close is set by settattr when we are about to truncate
1172 * a file from a non-zero size to a zero size. This tries to
1173 * flush down new bytes that may have been written if the
1174 * application were using truncate to replace a file in place.
1176 if (BTRFS_I(inode
)->ordered_data_close
) {
1177 BTRFS_I(inode
)->ordered_data_close
= 0;
1178 btrfs_add_ordered_operation(NULL
, BTRFS_I(inode
)->root
, inode
);
1179 if (inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
1180 filemap_flush(inode
->i_mapping
);
1182 if (filp
->private_data
)
1183 btrfs_ioctl_trans_end(filp
);
1188 * fsync call for both files and directories. This logs the inode into
1189 * the tree log instead of forcing full commits whenever possible.
1191 * It needs to call filemap_fdatawait so that all ordered extent updates are
1192 * in the metadata btree are up to date for copying to the log.
1194 * It drops the inode mutex before doing the tree log commit. This is an
1195 * important optimization for directories because holding the mutex prevents
1196 * new operations on the dir while we write to disk.
1198 int btrfs_sync_file(struct file
*file
, int datasync
)
1200 struct dentry
*dentry
= file
->f_path
.dentry
;
1201 struct inode
*inode
= dentry
->d_inode
;
1202 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1204 struct btrfs_trans_handle
*trans
;
1207 /* we wait first, since the writeback may change the inode */
1209 /* the VFS called filemap_fdatawrite for us */
1210 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
1214 * check the transaction that last modified this inode
1215 * and see if its already been committed
1217 if (!BTRFS_I(inode
)->last_trans
)
1221 * if the last transaction that changed this file was before
1222 * the current transaction, we can bail out now without any
1225 mutex_lock(&root
->fs_info
->trans_mutex
);
1226 if (BTRFS_I(inode
)->last_trans
<=
1227 root
->fs_info
->last_trans_committed
) {
1228 BTRFS_I(inode
)->last_trans
= 0;
1229 mutex_unlock(&root
->fs_info
->trans_mutex
);
1232 mutex_unlock(&root
->fs_info
->trans_mutex
);
1235 * ok we haven't committed the transaction yet, lets do a commit
1237 if (file
->private_data
)
1238 btrfs_ioctl_trans_end(file
);
1240 trans
= btrfs_start_transaction(root
, 0);
1241 if (IS_ERR(trans
)) {
1242 ret
= PTR_ERR(trans
);
1246 ret
= btrfs_log_dentry_safe(trans
, root
, 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(&dentry
->d_inode
->i_mutex
);
1262 if (ret
!= BTRFS_NO_LOG_SYNC
) {
1264 ret
= btrfs_commit_transaction(trans
, root
);
1266 ret
= btrfs_sync_log(trans
, root
);
1268 ret
= btrfs_end_transaction(trans
, root
);
1270 ret
= btrfs_commit_transaction(trans
, root
);
1273 ret
= btrfs_end_transaction(trans
, root
);
1275 mutex_lock(&dentry
->d_inode
->i_mutex
);
1277 return ret
> 0 ? -EIO
: ret
;
1280 static const struct vm_operations_struct btrfs_file_vm_ops
= {
1281 .fault
= filemap_fault
,
1282 .page_mkwrite
= btrfs_page_mkwrite
,
1285 static int btrfs_file_mmap(struct file
*filp
, struct vm_area_struct
*vma
)
1287 struct address_space
*mapping
= filp
->f_mapping
;
1289 if (!mapping
->a_ops
->readpage
)
1292 file_accessed(filp
);
1293 vma
->vm_ops
= &btrfs_file_vm_ops
;
1294 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
1299 static long btrfs_fallocate(struct file
*file
, int mode
,
1300 loff_t offset
, loff_t len
)
1302 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1303 struct extent_state
*cached_state
= NULL
;
1310 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
1311 struct extent_map
*em
;
1314 alloc_start
= offset
& ~mask
;
1315 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
1317 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1318 if (mode
& ~FALLOC_FL_KEEP_SIZE
)
1322 * wait for ordered IO before we have any locks. We'll loop again
1323 * below with the locks held.
1325 btrfs_wait_ordered_range(inode
, alloc_start
, alloc_end
- alloc_start
);
1327 mutex_lock(&inode
->i_mutex
);
1328 ret
= inode_newsize_ok(inode
, alloc_end
);
1332 if (alloc_start
> inode
->i_size
) {
1333 ret
= btrfs_cont_expand(inode
, i_size_read(inode
),
1339 ret
= btrfs_check_data_free_space(inode
, alloc_end
- alloc_start
);
1343 locked_end
= alloc_end
- 1;
1345 struct btrfs_ordered_extent
*ordered
;
1347 /* the extent lock is ordered inside the running
1350 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, alloc_start
,
1351 locked_end
, 0, &cached_state
, GFP_NOFS
);
1352 ordered
= btrfs_lookup_first_ordered_extent(inode
,
1355 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
1356 ordered
->file_offset
< alloc_end
) {
1357 btrfs_put_ordered_extent(ordered
);
1358 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
1359 alloc_start
, locked_end
,
1360 &cached_state
, GFP_NOFS
);
1362 * we can't wait on the range with the transaction
1363 * running or with the extent lock held
1365 btrfs_wait_ordered_range(inode
, alloc_start
,
1366 alloc_end
- alloc_start
);
1369 btrfs_put_ordered_extent(ordered
);
1374 cur_offset
= alloc_start
;
1376 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
1377 alloc_end
- cur_offset
, 0);
1378 BUG_ON(IS_ERR(em
) || !em
);
1379 last_byte
= min(extent_map_end(em
), alloc_end
);
1380 last_byte
= (last_byte
+ mask
) & ~mask
;
1381 if (em
->block_start
== EXTENT_MAP_HOLE
||
1382 (cur_offset
>= inode
->i_size
&&
1383 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
1384 ret
= btrfs_prealloc_file_range(inode
, mode
, cur_offset
,
1385 last_byte
- cur_offset
,
1386 1 << inode
->i_blkbits
,
1390 free_extent_map(em
);
1394 free_extent_map(em
);
1396 cur_offset
= last_byte
;
1397 if (cur_offset
>= alloc_end
) {
1402 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
1403 &cached_state
, GFP_NOFS
);
1405 btrfs_free_reserved_data_space(inode
, alloc_end
- alloc_start
);
1407 mutex_unlock(&inode
->i_mutex
);
1411 const struct file_operations btrfs_file_operations
= {
1412 .llseek
= generic_file_llseek
,
1413 .read
= do_sync_read
,
1414 .write
= do_sync_write
,
1415 .aio_read
= generic_file_aio_read
,
1416 .splice_read
= generic_file_splice_read
,
1417 .aio_write
= btrfs_file_aio_write
,
1418 .mmap
= btrfs_file_mmap
,
1419 .open
= generic_file_open
,
1420 .release
= btrfs_release_file
,
1421 .fsync
= btrfs_sync_file
,
1422 .fallocate
= btrfs_fallocate
,
1423 .unlocked_ioctl
= btrfs_ioctl
,
1424 #ifdef CONFIG_COMPAT
1425 .compat_ioctl
= btrfs_ioctl
,