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.
19 #include <linux/slab.h>
20 #include <linux/blkdev.h>
21 #include <linux/writeback.h>
22 #include <linux/pagevec.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "extent_io.h"
28 static struct kmem_cache
*btrfs_ordered_extent_cache
;
30 static u64
entry_end(struct btrfs_ordered_extent
*entry
)
32 if (entry
->file_offset
+ entry
->len
< entry
->file_offset
)
34 return entry
->file_offset
+ entry
->len
;
37 /* returns NULL if the insertion worked, or it returns the node it did find
40 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 file_offset
,
43 struct rb_node
**p
= &root
->rb_node
;
44 struct rb_node
*parent
= NULL
;
45 struct btrfs_ordered_extent
*entry
;
49 entry
= rb_entry(parent
, struct btrfs_ordered_extent
, rb_node
);
51 if (file_offset
< entry
->file_offset
)
53 else if (file_offset
>= entry_end(entry
))
59 rb_link_node(node
, parent
, p
);
60 rb_insert_color(node
, root
);
64 static void ordered_data_tree_panic(struct inode
*inode
, int errno
,
67 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
68 btrfs_panic(fs_info
, errno
, "Inconsistency in ordered tree at offset "
69 "%llu\n", (unsigned long long)offset
);
73 * look for a given offset in the tree, and if it can't be found return the
76 static struct rb_node
*__tree_search(struct rb_root
*root
, u64 file_offset
,
77 struct rb_node
**prev_ret
)
79 struct rb_node
*n
= root
->rb_node
;
80 struct rb_node
*prev
= NULL
;
82 struct btrfs_ordered_extent
*entry
;
83 struct btrfs_ordered_extent
*prev_entry
= NULL
;
86 entry
= rb_entry(n
, struct btrfs_ordered_extent
, rb_node
);
90 if (file_offset
< entry
->file_offset
)
92 else if (file_offset
>= entry_end(entry
))
100 while (prev
&& file_offset
>= entry_end(prev_entry
)) {
101 test
= rb_next(prev
);
104 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
106 if (file_offset
< entry_end(prev_entry
))
112 prev_entry
= rb_entry(prev
, struct btrfs_ordered_extent
,
114 while (prev
&& file_offset
< entry_end(prev_entry
)) {
115 test
= rb_prev(prev
);
118 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
127 * helper to check if a given offset is inside a given entry
129 static int offset_in_entry(struct btrfs_ordered_extent
*entry
, u64 file_offset
)
131 if (file_offset
< entry
->file_offset
||
132 entry
->file_offset
+ entry
->len
<= file_offset
)
137 static int range_overlaps(struct btrfs_ordered_extent
*entry
, u64 file_offset
,
140 if (file_offset
+ len
<= entry
->file_offset
||
141 entry
->file_offset
+ entry
->len
<= file_offset
)
147 * look find the first ordered struct that has this offset, otherwise
148 * the first one less than this offset
150 static inline struct rb_node
*tree_search(struct btrfs_ordered_inode_tree
*tree
,
153 struct rb_root
*root
= &tree
->tree
;
154 struct rb_node
*prev
= NULL
;
156 struct btrfs_ordered_extent
*entry
;
159 entry
= rb_entry(tree
->last
, struct btrfs_ordered_extent
,
161 if (offset_in_entry(entry
, file_offset
))
164 ret
= __tree_search(root
, file_offset
, &prev
);
172 /* allocate and add a new ordered_extent into the per-inode tree.
173 * file_offset is the logical offset in the file
175 * start is the disk block number of an extent already reserved in the
176 * extent allocation tree
178 * len is the length of the extent
180 * The tree is given a single reference on the ordered extent that was
183 static int __btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
184 u64 start
, u64 len
, u64 disk_len
,
185 int type
, int dio
, int compress_type
)
187 struct btrfs_ordered_inode_tree
*tree
;
188 struct rb_node
*node
;
189 struct btrfs_ordered_extent
*entry
;
191 tree
= &BTRFS_I(inode
)->ordered_tree
;
192 entry
= kmem_cache_zalloc(btrfs_ordered_extent_cache
, GFP_NOFS
);
196 entry
->file_offset
= file_offset
;
197 entry
->start
= start
;
199 entry
->disk_len
= disk_len
;
200 entry
->bytes_left
= len
;
201 entry
->inode
= igrab(inode
);
202 entry
->compress_type
= compress_type
;
203 if (type
!= BTRFS_ORDERED_IO_DONE
&& type
!= BTRFS_ORDERED_COMPLETE
)
204 set_bit(type
, &entry
->flags
);
207 set_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
);
209 /* one ref for the tree */
210 atomic_set(&entry
->refs
, 1);
211 init_waitqueue_head(&entry
->wait
);
212 INIT_LIST_HEAD(&entry
->list
);
213 INIT_LIST_HEAD(&entry
->root_extent_list
);
215 trace_btrfs_ordered_extent_add(inode
, entry
);
217 spin_lock_irq(&tree
->lock
);
218 node
= tree_insert(&tree
->tree
, file_offset
,
221 ordered_data_tree_panic(inode
, -EEXIST
, file_offset
);
222 spin_unlock_irq(&tree
->lock
);
224 spin_lock(&BTRFS_I(inode
)->root
->fs_info
->ordered_extent_lock
);
225 list_add_tail(&entry
->root_extent_list
,
226 &BTRFS_I(inode
)->root
->fs_info
->ordered_extents
);
227 spin_unlock(&BTRFS_I(inode
)->root
->fs_info
->ordered_extent_lock
);
232 int btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
233 u64 start
, u64 len
, u64 disk_len
, int type
)
235 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
237 BTRFS_COMPRESS_NONE
);
240 int btrfs_add_ordered_extent_dio(struct inode
*inode
, u64 file_offset
,
241 u64 start
, u64 len
, u64 disk_len
, int type
)
243 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
245 BTRFS_COMPRESS_NONE
);
248 int btrfs_add_ordered_extent_compress(struct inode
*inode
, u64 file_offset
,
249 u64 start
, u64 len
, u64 disk_len
,
250 int type
, int compress_type
)
252 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
258 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
259 * when an ordered extent is finished. If the list covers more than one
260 * ordered extent, it is split across multiples.
262 void btrfs_add_ordered_sum(struct inode
*inode
,
263 struct btrfs_ordered_extent
*entry
,
264 struct btrfs_ordered_sum
*sum
)
266 struct btrfs_ordered_inode_tree
*tree
;
268 tree
= &BTRFS_I(inode
)->ordered_tree
;
269 spin_lock_irq(&tree
->lock
);
270 list_add_tail(&sum
->list
, &entry
->list
);
271 spin_unlock_irq(&tree
->lock
);
275 * this is used to account for finished IO across a given range
276 * of the file. The IO may span ordered extents. If
277 * a given ordered_extent is completely done, 1 is returned, otherwise
280 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
281 * to make sure this function only returns 1 once for a given ordered extent.
283 * file_offset is updated to one byte past the range that is recorded as
284 * complete. This allows you to walk forward in the file.
286 int btrfs_dec_test_first_ordered_pending(struct inode
*inode
,
287 struct btrfs_ordered_extent
**cached
,
288 u64
*file_offset
, u64 io_size
, int uptodate
)
290 struct btrfs_ordered_inode_tree
*tree
;
291 struct rb_node
*node
;
292 struct btrfs_ordered_extent
*entry
= NULL
;
299 tree
= &BTRFS_I(inode
)->ordered_tree
;
300 spin_lock_irqsave(&tree
->lock
, flags
);
301 node
= tree_search(tree
, *file_offset
);
307 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
308 if (!offset_in_entry(entry
, *file_offset
)) {
313 dec_start
= max(*file_offset
, entry
->file_offset
);
314 dec_end
= min(*file_offset
+ io_size
, entry
->file_offset
+
316 *file_offset
= dec_end
;
317 if (dec_start
> dec_end
) {
318 printk(KERN_CRIT
"bad ordering dec_start %llu end %llu\n",
319 (unsigned long long)dec_start
,
320 (unsigned long long)dec_end
);
322 to_dec
= dec_end
- dec_start
;
323 if (to_dec
> entry
->bytes_left
) {
324 printk(KERN_CRIT
"bad ordered accounting left %llu size %llu\n",
325 (unsigned long long)entry
->bytes_left
,
326 (unsigned long long)to_dec
);
328 entry
->bytes_left
-= to_dec
;
330 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
332 if (entry
->bytes_left
== 0)
333 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
337 if (!ret
&& cached
&& entry
) {
339 atomic_inc(&entry
->refs
);
341 spin_unlock_irqrestore(&tree
->lock
, flags
);
346 * this is used to account for finished IO across a given range
347 * of the file. The IO should not span ordered extents. If
348 * a given ordered_extent is completely done, 1 is returned, otherwise
351 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
352 * to make sure this function only returns 1 once for a given ordered extent.
354 int btrfs_dec_test_ordered_pending(struct inode
*inode
,
355 struct btrfs_ordered_extent
**cached
,
356 u64 file_offset
, u64 io_size
, int uptodate
)
358 struct btrfs_ordered_inode_tree
*tree
;
359 struct rb_node
*node
;
360 struct btrfs_ordered_extent
*entry
= NULL
;
364 tree
= &BTRFS_I(inode
)->ordered_tree
;
365 spin_lock_irqsave(&tree
->lock
, flags
);
366 if (cached
&& *cached
) {
371 node
= tree_search(tree
, file_offset
);
377 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
379 if (!offset_in_entry(entry
, file_offset
)) {
384 if (io_size
> entry
->bytes_left
) {
385 printk(KERN_CRIT
"bad ordered accounting left %llu size %llu\n",
386 (unsigned long long)entry
->bytes_left
,
387 (unsigned long long)io_size
);
389 entry
->bytes_left
-= io_size
;
391 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
393 if (entry
->bytes_left
== 0)
394 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
398 if (!ret
&& cached
&& entry
) {
400 atomic_inc(&entry
->refs
);
402 spin_unlock_irqrestore(&tree
->lock
, flags
);
407 * used to drop a reference on an ordered extent. This will free
408 * the extent if the last reference is dropped
410 void btrfs_put_ordered_extent(struct btrfs_ordered_extent
*entry
)
412 struct list_head
*cur
;
413 struct btrfs_ordered_sum
*sum
;
415 trace_btrfs_ordered_extent_put(entry
->inode
, entry
);
417 if (atomic_dec_and_test(&entry
->refs
)) {
419 btrfs_add_delayed_iput(entry
->inode
);
420 while (!list_empty(&entry
->list
)) {
421 cur
= entry
->list
.next
;
422 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
423 list_del(&sum
->list
);
426 kmem_cache_free(btrfs_ordered_extent_cache
, entry
);
431 * remove an ordered extent from the tree. No references are dropped
432 * and waiters are woken up.
434 void btrfs_remove_ordered_extent(struct inode
*inode
,
435 struct btrfs_ordered_extent
*entry
)
437 struct btrfs_ordered_inode_tree
*tree
;
438 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
439 struct rb_node
*node
;
441 tree
= &BTRFS_I(inode
)->ordered_tree
;
442 spin_lock_irq(&tree
->lock
);
443 node
= &entry
->rb_node
;
444 rb_erase(node
, &tree
->tree
);
446 set_bit(BTRFS_ORDERED_COMPLETE
, &entry
->flags
);
447 spin_unlock_irq(&tree
->lock
);
449 spin_lock(&root
->fs_info
->ordered_extent_lock
);
450 list_del_init(&entry
->root_extent_list
);
452 trace_btrfs_ordered_extent_remove(inode
, entry
);
455 * we have no more ordered extents for this inode and
456 * no dirty pages. We can safely remove it from the
457 * list of ordered extents
459 if (RB_EMPTY_ROOT(&tree
->tree
) &&
460 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
461 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
463 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
464 wake_up(&entry
->wait
);
468 * wait for all the ordered extents in a root. This is done when balancing
469 * space between drives.
471 void btrfs_wait_ordered_extents(struct btrfs_root
*root
, int delay_iput
)
473 struct list_head splice
;
474 struct list_head
*cur
;
475 struct btrfs_ordered_extent
*ordered
;
478 INIT_LIST_HEAD(&splice
);
480 spin_lock(&root
->fs_info
->ordered_extent_lock
);
481 list_splice_init(&root
->fs_info
->ordered_extents
, &splice
);
482 while (!list_empty(&splice
)) {
484 ordered
= list_entry(cur
, struct btrfs_ordered_extent
,
486 list_del_init(&ordered
->root_extent_list
);
487 atomic_inc(&ordered
->refs
);
490 * the inode may be getting freed (in sys_unlink path).
492 inode
= igrab(ordered
->inode
);
494 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
497 btrfs_start_ordered_extent(inode
, ordered
, 1);
498 btrfs_put_ordered_extent(ordered
);
500 btrfs_add_delayed_iput(inode
);
504 btrfs_put_ordered_extent(ordered
);
507 spin_lock(&root
->fs_info
->ordered_extent_lock
);
509 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
513 * this is used during transaction commit to write all the inodes
514 * added to the ordered operation list. These files must be fully on
515 * disk before the transaction commits.
517 * we have two modes here, one is to just start the IO via filemap_flush
518 * and the other is to wait for all the io. When we wait, we have an
519 * extra check to make sure the ordered operation list really is empty
522 void btrfs_run_ordered_operations(struct btrfs_root
*root
, int wait
)
524 struct btrfs_inode
*btrfs_inode
;
526 struct list_head splice
;
528 INIT_LIST_HEAD(&splice
);
530 mutex_lock(&root
->fs_info
->ordered_operations_mutex
);
531 spin_lock(&root
->fs_info
->ordered_extent_lock
);
533 list_splice_init(&root
->fs_info
->ordered_operations
, &splice
);
535 while (!list_empty(&splice
)) {
536 btrfs_inode
= list_entry(splice
.next
, struct btrfs_inode
,
539 inode
= &btrfs_inode
->vfs_inode
;
541 list_del_init(&btrfs_inode
->ordered_operations
);
544 * the inode may be getting freed (in sys_unlink path).
546 inode
= igrab(inode
);
548 if (!wait
&& inode
) {
549 list_add_tail(&BTRFS_I(inode
)->ordered_operations
,
550 &root
->fs_info
->ordered_operations
);
552 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
556 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
558 filemap_flush(inode
->i_mapping
);
559 btrfs_add_delayed_iput(inode
);
563 spin_lock(&root
->fs_info
->ordered_extent_lock
);
565 if (wait
&& !list_empty(&root
->fs_info
->ordered_operations
))
568 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
569 mutex_unlock(&root
->fs_info
->ordered_operations_mutex
);
573 * Used to start IO or wait for a given ordered extent to finish.
575 * If wait is one, this effectively waits on page writeback for all the pages
576 * in the extent, and it waits on the io completion code to insert
577 * metadata into the btree corresponding to the extent
579 void btrfs_start_ordered_extent(struct inode
*inode
,
580 struct btrfs_ordered_extent
*entry
,
583 u64 start
= entry
->file_offset
;
584 u64 end
= start
+ entry
->len
- 1;
586 trace_btrfs_ordered_extent_start(inode
, entry
);
589 * pages in the range can be dirty, clean or writeback. We
590 * start IO on any dirty ones so the wait doesn't stall waiting
591 * for the flusher thread to find them
593 if (!test_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
))
594 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
596 wait_event(entry
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
602 * Used to wait on ordered extents across a large range of bytes.
604 void btrfs_wait_ordered_range(struct inode
*inode
, u64 start
, u64 len
)
608 struct btrfs_ordered_extent
*ordered
;
611 if (start
+ len
< start
) {
612 orig_end
= INT_LIMIT(loff_t
);
614 orig_end
= start
+ len
- 1;
615 if (orig_end
> INT_LIMIT(loff_t
))
616 orig_end
= INT_LIMIT(loff_t
);
619 /* start IO across the range first to instantiate any delalloc
622 filemap_fdatawrite_range(inode
->i_mapping
, start
, orig_end
);
625 * So with compression we will find and lock a dirty page and clear the
626 * first one as dirty, setup an async extent, and immediately return
627 * with the entire range locked but with nobody actually marked with
628 * writeback. So we can't just filemap_write_and_wait_range() and
629 * expect it to work since it will just kick off a thread to do the
630 * actual work. So we need to call filemap_fdatawrite_range _again_
631 * since it will wait on the page lock, which won't be unlocked until
632 * after the pages have been marked as writeback and so we're good to go
633 * from there. We have to do this otherwise we'll miss the ordered
634 * extents and that results in badness. Please Josef, do not think you
635 * know better and pull this out at some point in the future, it is
636 * right and you are wrong.
638 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
639 &BTRFS_I(inode
)->runtime_flags
))
640 filemap_fdatawrite_range(inode
->i_mapping
, start
, orig_end
);
642 filemap_fdatawait_range(inode
->i_mapping
, start
, orig_end
);
647 ordered
= btrfs_lookup_first_ordered_extent(inode
, end
);
650 if (ordered
->file_offset
> orig_end
) {
651 btrfs_put_ordered_extent(ordered
);
654 if (ordered
->file_offset
+ ordered
->len
< start
) {
655 btrfs_put_ordered_extent(ordered
);
659 btrfs_start_ordered_extent(inode
, ordered
, 1);
660 end
= ordered
->file_offset
;
661 btrfs_put_ordered_extent(ordered
);
662 if (end
== 0 || end
== start
)
669 * find an ordered extent corresponding to file_offset. return NULL if
670 * nothing is found, otherwise take a reference on the extent and return it
672 struct btrfs_ordered_extent
*btrfs_lookup_ordered_extent(struct inode
*inode
,
675 struct btrfs_ordered_inode_tree
*tree
;
676 struct rb_node
*node
;
677 struct btrfs_ordered_extent
*entry
= NULL
;
679 tree
= &BTRFS_I(inode
)->ordered_tree
;
680 spin_lock_irq(&tree
->lock
);
681 node
= tree_search(tree
, file_offset
);
685 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
686 if (!offset_in_entry(entry
, file_offset
))
689 atomic_inc(&entry
->refs
);
691 spin_unlock_irq(&tree
->lock
);
695 /* Since the DIO code tries to lock a wide area we need to look for any ordered
696 * extents that exist in the range, rather than just the start of the range.
698 struct btrfs_ordered_extent
*btrfs_lookup_ordered_range(struct inode
*inode
,
702 struct btrfs_ordered_inode_tree
*tree
;
703 struct rb_node
*node
;
704 struct btrfs_ordered_extent
*entry
= NULL
;
706 tree
= &BTRFS_I(inode
)->ordered_tree
;
707 spin_lock_irq(&tree
->lock
);
708 node
= tree_search(tree
, file_offset
);
710 node
= tree_search(tree
, file_offset
+ len
);
716 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
717 if (range_overlaps(entry
, file_offset
, len
))
720 if (entry
->file_offset
>= file_offset
+ len
) {
725 node
= rb_next(node
);
731 atomic_inc(&entry
->refs
);
732 spin_unlock_irq(&tree
->lock
);
737 * lookup and return any extent before 'file_offset'. NULL is returned
740 struct btrfs_ordered_extent
*
741 btrfs_lookup_first_ordered_extent(struct inode
*inode
, u64 file_offset
)
743 struct btrfs_ordered_inode_tree
*tree
;
744 struct rb_node
*node
;
745 struct btrfs_ordered_extent
*entry
= NULL
;
747 tree
= &BTRFS_I(inode
)->ordered_tree
;
748 spin_lock_irq(&tree
->lock
);
749 node
= tree_search(tree
, file_offset
);
753 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
754 atomic_inc(&entry
->refs
);
756 spin_unlock_irq(&tree
->lock
);
761 * After an extent is done, call this to conditionally update the on disk
762 * i_size. i_size is updated to cover any fully written part of the file.
764 int btrfs_ordered_update_i_size(struct inode
*inode
, u64 offset
,
765 struct btrfs_ordered_extent
*ordered
)
767 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
770 u64 i_size
= i_size_read(inode
);
771 struct rb_node
*node
;
772 struct rb_node
*prev
= NULL
;
773 struct btrfs_ordered_extent
*test
;
777 offset
= entry_end(ordered
);
779 offset
= ALIGN(offset
, BTRFS_I(inode
)->root
->sectorsize
);
781 spin_lock_irq(&tree
->lock
);
782 disk_i_size
= BTRFS_I(inode
)->disk_i_size
;
785 if (disk_i_size
> i_size
) {
786 BTRFS_I(inode
)->disk_i_size
= i_size
;
792 * if the disk i_size is already at the inode->i_size, or
793 * this ordered extent is inside the disk i_size, we're done
795 if (disk_i_size
== i_size
|| offset
<= disk_i_size
) {
800 * walk backward from this ordered extent to disk_i_size.
801 * if we find an ordered extent then we can't update disk i_size
805 node
= rb_prev(&ordered
->rb_node
);
807 prev
= tree_search(tree
, offset
);
809 * we insert file extents without involving ordered struct,
810 * so there should be no ordered struct cover this offset
813 test
= rb_entry(prev
, struct btrfs_ordered_extent
,
815 BUG_ON(offset_in_entry(test
, offset
));
819 for (; node
; node
= rb_prev(node
)) {
820 test
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
822 /* We treat this entry as if it doesnt exist */
823 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &test
->flags
))
825 if (test
->file_offset
+ test
->len
<= disk_i_size
)
827 if (test
->file_offset
>= i_size
)
829 if (test
->file_offset
>= disk_i_size
) {
831 * we don't update disk_i_size now, so record this
832 * undealt i_size. Or we will not know the real
835 if (test
->outstanding_isize
< offset
)
836 test
->outstanding_isize
= offset
;
838 ordered
->outstanding_isize
>
839 test
->outstanding_isize
)
840 test
->outstanding_isize
=
841 ordered
->outstanding_isize
;
845 new_i_size
= min_t(u64
, offset
, i_size
);
848 * Some ordered extents may completed before the current one, and
849 * we hold the real i_size in ->outstanding_isize.
851 if (ordered
&& ordered
->outstanding_isize
> new_i_size
)
852 new_i_size
= min_t(u64
, ordered
->outstanding_isize
, i_size
);
853 BTRFS_I(inode
)->disk_i_size
= new_i_size
;
857 * We need to do this because we can't remove ordered extents until
858 * after the i_disk_size has been updated and then the inode has been
859 * updated to reflect the change, so we need to tell anybody who finds
860 * this ordered extent that we've already done all the real work, we
861 * just haven't completed all the other work.
864 set_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &ordered
->flags
);
865 spin_unlock_irq(&tree
->lock
);
870 * search the ordered extents for one corresponding to 'offset' and
871 * try to find a checksum. This is used because we allow pages to
872 * be reclaimed before their checksum is actually put into the btree
874 int btrfs_find_ordered_sum(struct inode
*inode
, u64 offset
, u64 disk_bytenr
,
877 struct btrfs_ordered_sum
*ordered_sum
;
878 struct btrfs_sector_sum
*sector_sums
;
879 struct btrfs_ordered_extent
*ordered
;
880 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
881 unsigned long num_sectors
;
883 u32 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
886 ordered
= btrfs_lookup_ordered_extent(inode
, offset
);
890 spin_lock_irq(&tree
->lock
);
891 list_for_each_entry_reverse(ordered_sum
, &ordered
->list
, list
) {
892 if (disk_bytenr
>= ordered_sum
->bytenr
) {
893 num_sectors
= ordered_sum
->len
/ sectorsize
;
894 sector_sums
= ordered_sum
->sums
;
895 for (i
= 0; i
< num_sectors
; i
++) {
896 if (sector_sums
[i
].bytenr
== disk_bytenr
) {
897 *sum
= sector_sums
[i
].sum
;
905 spin_unlock_irq(&tree
->lock
);
906 btrfs_put_ordered_extent(ordered
);
912 * add a given inode to the list of inodes that must be fully on
913 * disk before a transaction commit finishes.
915 * This basically gives us the ext3 style data=ordered mode, and it is mostly
916 * used to make sure renamed files are fully on disk.
918 * It is a noop if the inode is already fully on disk.
920 * If trans is not null, we'll do a friendly check for a transaction that
921 * is already flushing things and force the IO down ourselves.
923 void btrfs_add_ordered_operation(struct btrfs_trans_handle
*trans
,
924 struct btrfs_root
*root
, struct inode
*inode
)
928 last_mod
= max(BTRFS_I(inode
)->generation
, BTRFS_I(inode
)->last_trans
);
931 * if this file hasn't been changed since the last transaction
932 * commit, we can safely return without doing anything
934 if (last_mod
< root
->fs_info
->last_trans_committed
)
938 * the transaction is already committing. Just start the IO and
939 * don't bother with all of this list nonsense
941 if (trans
&& root
->fs_info
->running_transaction
->blocked
) {
942 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
946 spin_lock(&root
->fs_info
->ordered_extent_lock
);
947 if (list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
948 list_add_tail(&BTRFS_I(inode
)->ordered_operations
,
949 &root
->fs_info
->ordered_operations
);
951 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
954 int __init
ordered_data_init(void)
956 btrfs_ordered_extent_cache
= kmem_cache_create("btrfs_ordered_extent",
957 sizeof(struct btrfs_ordered_extent
), 0,
958 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
960 if (!btrfs_ordered_extent_cache
)
965 void ordered_data_exit(void)
967 if (btrfs_ordered_extent_cache
)
968 kmem_cache_destroy(btrfs_ordered_extent_cache
);