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 #include "compression.h"
30 static struct kmem_cache
*btrfs_ordered_extent_cache
;
32 static u64
entry_end(struct btrfs_ordered_extent
*entry
)
34 if (entry
->file_offset
+ entry
->len
< entry
->file_offset
)
36 return entry
->file_offset
+ entry
->len
;
39 /* returns NULL if the insertion worked, or it returns the node it did find
42 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 file_offset
,
45 struct rb_node
**p
= &root
->rb_node
;
46 struct rb_node
*parent
= NULL
;
47 struct btrfs_ordered_extent
*entry
;
51 entry
= rb_entry(parent
, struct btrfs_ordered_extent
, rb_node
);
53 if (file_offset
< entry
->file_offset
)
55 else if (file_offset
>= entry_end(entry
))
61 rb_link_node(node
, parent
, p
);
62 rb_insert_color(node
, root
);
66 static void ordered_data_tree_panic(struct inode
*inode
, int errno
,
69 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
70 btrfs_panic(fs_info
, errno
, "Inconsistency in ordered tree at offset "
75 * look for a given offset in the tree, and if it can't be found return the
78 static struct rb_node
*__tree_search(struct rb_root
*root
, u64 file_offset
,
79 struct rb_node
**prev_ret
)
81 struct rb_node
*n
= root
->rb_node
;
82 struct rb_node
*prev
= NULL
;
84 struct btrfs_ordered_extent
*entry
;
85 struct btrfs_ordered_extent
*prev_entry
= NULL
;
88 entry
= rb_entry(n
, struct btrfs_ordered_extent
, rb_node
);
92 if (file_offset
< entry
->file_offset
)
94 else if (file_offset
>= entry_end(entry
))
102 while (prev
&& file_offset
>= entry_end(prev_entry
)) {
103 test
= rb_next(prev
);
106 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
108 if (file_offset
< entry_end(prev_entry
))
114 prev_entry
= rb_entry(prev
, struct btrfs_ordered_extent
,
116 while (prev
&& file_offset
< entry_end(prev_entry
)) {
117 test
= rb_prev(prev
);
120 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
129 * helper to check if a given offset is inside a given entry
131 static int offset_in_entry(struct btrfs_ordered_extent
*entry
, u64 file_offset
)
133 if (file_offset
< entry
->file_offset
||
134 entry
->file_offset
+ entry
->len
<= file_offset
)
139 static int range_overlaps(struct btrfs_ordered_extent
*entry
, u64 file_offset
,
142 if (file_offset
+ len
<= entry
->file_offset
||
143 entry
->file_offset
+ entry
->len
<= file_offset
)
149 * look find the first ordered struct that has this offset, otherwise
150 * the first one less than this offset
152 static inline struct rb_node
*tree_search(struct btrfs_ordered_inode_tree
*tree
,
155 struct rb_root
*root
= &tree
->tree
;
156 struct rb_node
*prev
= NULL
;
158 struct btrfs_ordered_extent
*entry
;
161 entry
= rb_entry(tree
->last
, struct btrfs_ordered_extent
,
163 if (offset_in_entry(entry
, file_offset
))
166 ret
= __tree_search(root
, file_offset
, &prev
);
174 /* allocate and add a new ordered_extent into the per-inode tree.
175 * file_offset is the logical offset in the file
177 * start is the disk block number of an extent already reserved in the
178 * extent allocation tree
180 * len is the length of the extent
182 * The tree is given a single reference on the ordered extent that was
185 static int __btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
186 u64 start
, u64 len
, u64 disk_len
,
187 int type
, int dio
, int compress_type
)
189 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
190 struct btrfs_ordered_inode_tree
*tree
;
191 struct rb_node
*node
;
192 struct btrfs_ordered_extent
*entry
;
194 tree
= &BTRFS_I(inode
)->ordered_tree
;
195 entry
= kmem_cache_zalloc(btrfs_ordered_extent_cache
, GFP_NOFS
);
199 entry
->file_offset
= file_offset
;
200 entry
->start
= start
;
202 entry
->disk_len
= disk_len
;
203 entry
->bytes_left
= len
;
204 entry
->inode
= igrab(inode
);
205 entry
->compress_type
= compress_type
;
206 entry
->truncated_len
= (u64
)-1;
207 if (type
!= BTRFS_ORDERED_IO_DONE
&& type
!= BTRFS_ORDERED_COMPLETE
)
208 set_bit(type
, &entry
->flags
);
211 set_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
);
213 /* one ref for the tree */
214 atomic_set(&entry
->refs
, 1);
215 init_waitqueue_head(&entry
->wait
);
216 INIT_LIST_HEAD(&entry
->list
);
217 INIT_LIST_HEAD(&entry
->root_extent_list
);
218 INIT_LIST_HEAD(&entry
->work_list
);
219 init_completion(&entry
->completion
);
220 INIT_LIST_HEAD(&entry
->log_list
);
221 INIT_LIST_HEAD(&entry
->trans_list
);
223 trace_btrfs_ordered_extent_add(inode
, entry
);
225 spin_lock_irq(&tree
->lock
);
226 node
= tree_insert(&tree
->tree
, file_offset
,
229 ordered_data_tree_panic(inode
, -EEXIST
, file_offset
);
230 spin_unlock_irq(&tree
->lock
);
232 spin_lock(&root
->ordered_extent_lock
);
233 list_add_tail(&entry
->root_extent_list
,
234 &root
->ordered_extents
);
235 root
->nr_ordered_extents
++;
236 if (root
->nr_ordered_extents
== 1) {
237 spin_lock(&root
->fs_info
->ordered_root_lock
);
238 BUG_ON(!list_empty(&root
->ordered_root
));
239 list_add_tail(&root
->ordered_root
,
240 &root
->fs_info
->ordered_roots
);
241 spin_unlock(&root
->fs_info
->ordered_root_lock
);
243 spin_unlock(&root
->ordered_extent_lock
);
248 int btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
249 u64 start
, u64 len
, u64 disk_len
, int type
)
251 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
253 BTRFS_COMPRESS_NONE
);
256 int btrfs_add_ordered_extent_dio(struct inode
*inode
, u64 file_offset
,
257 u64 start
, u64 len
, u64 disk_len
, int type
)
259 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
261 BTRFS_COMPRESS_NONE
);
264 int btrfs_add_ordered_extent_compress(struct inode
*inode
, u64 file_offset
,
265 u64 start
, u64 len
, u64 disk_len
,
266 int type
, int compress_type
)
268 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
274 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
275 * when an ordered extent is finished. If the list covers more than one
276 * ordered extent, it is split across multiples.
278 void btrfs_add_ordered_sum(struct inode
*inode
,
279 struct btrfs_ordered_extent
*entry
,
280 struct btrfs_ordered_sum
*sum
)
282 struct btrfs_ordered_inode_tree
*tree
;
284 tree
= &BTRFS_I(inode
)->ordered_tree
;
285 spin_lock_irq(&tree
->lock
);
286 list_add_tail(&sum
->list
, &entry
->list
);
287 spin_unlock_irq(&tree
->lock
);
291 * this is used to account for finished IO across a given range
292 * of the file. The IO may span ordered extents. If
293 * a given ordered_extent is completely done, 1 is returned, otherwise
296 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
297 * to make sure this function only returns 1 once for a given ordered extent.
299 * file_offset is updated to one byte past the range that is recorded as
300 * complete. This allows you to walk forward in the file.
302 int btrfs_dec_test_first_ordered_pending(struct inode
*inode
,
303 struct btrfs_ordered_extent
**cached
,
304 u64
*file_offset
, u64 io_size
, int uptodate
)
306 struct btrfs_ordered_inode_tree
*tree
;
307 struct rb_node
*node
;
308 struct btrfs_ordered_extent
*entry
= NULL
;
315 tree
= &BTRFS_I(inode
)->ordered_tree
;
316 spin_lock_irqsave(&tree
->lock
, flags
);
317 node
= tree_search(tree
, *file_offset
);
323 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
324 if (!offset_in_entry(entry
, *file_offset
)) {
329 dec_start
= max(*file_offset
, entry
->file_offset
);
330 dec_end
= min(*file_offset
+ io_size
, entry
->file_offset
+
332 *file_offset
= dec_end
;
333 if (dec_start
> dec_end
) {
334 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
335 "bad ordering dec_start %llu end %llu", dec_start
, dec_end
);
337 to_dec
= dec_end
- dec_start
;
338 if (to_dec
> entry
->bytes_left
) {
339 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
340 "bad ordered accounting left %llu size %llu",
341 entry
->bytes_left
, to_dec
);
343 entry
->bytes_left
-= to_dec
;
345 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
347 if (entry
->bytes_left
== 0) {
348 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
350 * Implicit memory barrier after test_and_set_bit
352 if (waitqueue_active(&entry
->wait
))
353 wake_up(&entry
->wait
);
358 if (!ret
&& cached
&& entry
) {
360 atomic_inc(&entry
->refs
);
362 spin_unlock_irqrestore(&tree
->lock
, flags
);
367 * this is used to account for finished IO across a given range
368 * of the file. The IO should not span ordered extents. If
369 * a given ordered_extent is completely done, 1 is returned, otherwise
372 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
373 * to make sure this function only returns 1 once for a given ordered extent.
375 int btrfs_dec_test_ordered_pending(struct inode
*inode
,
376 struct btrfs_ordered_extent
**cached
,
377 u64 file_offset
, u64 io_size
, int uptodate
)
379 struct btrfs_ordered_inode_tree
*tree
;
380 struct rb_node
*node
;
381 struct btrfs_ordered_extent
*entry
= NULL
;
385 tree
= &BTRFS_I(inode
)->ordered_tree
;
386 spin_lock_irqsave(&tree
->lock
, flags
);
387 if (cached
&& *cached
) {
392 node
= tree_search(tree
, file_offset
);
398 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
400 if (!offset_in_entry(entry
, file_offset
)) {
405 if (io_size
> entry
->bytes_left
) {
406 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
407 "bad ordered accounting left %llu size %llu",
408 entry
->bytes_left
, io_size
);
410 entry
->bytes_left
-= io_size
;
412 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
414 if (entry
->bytes_left
== 0) {
415 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
417 * Implicit memory barrier after test_and_set_bit
419 if (waitqueue_active(&entry
->wait
))
420 wake_up(&entry
->wait
);
425 if (!ret
&& cached
&& entry
) {
427 atomic_inc(&entry
->refs
);
429 spin_unlock_irqrestore(&tree
->lock
, flags
);
433 /* Needs to either be called under a log transaction or the log_mutex */
434 void btrfs_get_logged_extents(struct inode
*inode
,
435 struct list_head
*logged_list
,
439 struct btrfs_ordered_inode_tree
*tree
;
440 struct btrfs_ordered_extent
*ordered
;
442 struct rb_node
*prev
;
444 tree
= &BTRFS_I(inode
)->ordered_tree
;
445 spin_lock_irq(&tree
->lock
);
446 n
= __tree_search(&tree
->tree
, end
, &prev
);
449 for (; n
; n
= rb_prev(n
)) {
450 ordered
= rb_entry(n
, struct btrfs_ordered_extent
, rb_node
);
451 if (ordered
->file_offset
> end
)
453 if (entry_end(ordered
) <= start
)
455 if (test_and_set_bit(BTRFS_ORDERED_LOGGED
, &ordered
->flags
))
457 list_add(&ordered
->log_list
, logged_list
);
458 atomic_inc(&ordered
->refs
);
460 spin_unlock_irq(&tree
->lock
);
463 void btrfs_put_logged_extents(struct list_head
*logged_list
)
465 struct btrfs_ordered_extent
*ordered
;
467 while (!list_empty(logged_list
)) {
468 ordered
= list_first_entry(logged_list
,
469 struct btrfs_ordered_extent
,
471 list_del_init(&ordered
->log_list
);
472 btrfs_put_ordered_extent(ordered
);
476 void btrfs_submit_logged_extents(struct list_head
*logged_list
,
477 struct btrfs_root
*log
)
479 int index
= log
->log_transid
% 2;
481 spin_lock_irq(&log
->log_extents_lock
[index
]);
482 list_splice_tail(logged_list
, &log
->logged_list
[index
]);
483 spin_unlock_irq(&log
->log_extents_lock
[index
]);
486 void btrfs_wait_logged_extents(struct btrfs_trans_handle
*trans
,
487 struct btrfs_root
*log
, u64 transid
)
489 struct btrfs_ordered_extent
*ordered
;
490 int index
= transid
% 2;
492 spin_lock_irq(&log
->log_extents_lock
[index
]);
493 while (!list_empty(&log
->logged_list
[index
])) {
495 ordered
= list_first_entry(&log
->logged_list
[index
],
496 struct btrfs_ordered_extent
,
498 list_del_init(&ordered
->log_list
);
499 inode
= ordered
->inode
;
500 spin_unlock_irq(&log
->log_extents_lock
[index
]);
502 if (!test_bit(BTRFS_ORDERED_IO_DONE
, &ordered
->flags
) &&
503 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
)) {
504 u64 start
= ordered
->file_offset
;
505 u64 end
= ordered
->file_offset
+ ordered
->len
- 1;
508 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
510 wait_event(ordered
->wait
, test_bit(BTRFS_ORDERED_IO_DONE
,
514 * In order to keep us from losing our ordered extent
515 * information when committing the transaction we have to make
516 * sure that any logged extents are completed when we go to
517 * commit the transaction. To do this we simply increase the
518 * current transactions pending_ordered counter and decrement it
519 * when the ordered extent completes.
521 if (!test_bit(BTRFS_ORDERED_COMPLETE
, &ordered
->flags
)) {
522 struct btrfs_ordered_inode_tree
*tree
;
524 tree
= &BTRFS_I(inode
)->ordered_tree
;
525 spin_lock_irq(&tree
->lock
);
526 if (!test_bit(BTRFS_ORDERED_COMPLETE
, &ordered
->flags
)) {
527 set_bit(BTRFS_ORDERED_PENDING
, &ordered
->flags
);
528 atomic_inc(&trans
->transaction
->pending_ordered
);
530 spin_unlock_irq(&tree
->lock
);
532 btrfs_put_ordered_extent(ordered
);
533 spin_lock_irq(&log
->log_extents_lock
[index
]);
535 spin_unlock_irq(&log
->log_extents_lock
[index
]);
538 void btrfs_free_logged_extents(struct btrfs_root
*log
, u64 transid
)
540 struct btrfs_ordered_extent
*ordered
;
541 int index
= transid
% 2;
543 spin_lock_irq(&log
->log_extents_lock
[index
]);
544 while (!list_empty(&log
->logged_list
[index
])) {
545 ordered
= list_first_entry(&log
->logged_list
[index
],
546 struct btrfs_ordered_extent
,
548 list_del_init(&ordered
->log_list
);
549 spin_unlock_irq(&log
->log_extents_lock
[index
]);
550 btrfs_put_ordered_extent(ordered
);
551 spin_lock_irq(&log
->log_extents_lock
[index
]);
553 spin_unlock_irq(&log
->log_extents_lock
[index
]);
557 * used to drop a reference on an ordered extent. This will free
558 * the extent if the last reference is dropped
560 void btrfs_put_ordered_extent(struct btrfs_ordered_extent
*entry
)
562 struct list_head
*cur
;
563 struct btrfs_ordered_sum
*sum
;
565 trace_btrfs_ordered_extent_put(entry
->inode
, entry
);
567 if (atomic_dec_and_test(&entry
->refs
)) {
568 ASSERT(list_empty(&entry
->log_list
));
569 ASSERT(list_empty(&entry
->trans_list
));
570 ASSERT(list_empty(&entry
->root_extent_list
));
571 ASSERT(RB_EMPTY_NODE(&entry
->rb_node
));
573 btrfs_add_delayed_iput(entry
->inode
);
574 while (!list_empty(&entry
->list
)) {
575 cur
= entry
->list
.next
;
576 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
577 list_del(&sum
->list
);
580 kmem_cache_free(btrfs_ordered_extent_cache
, entry
);
585 * remove an ordered extent from the tree. No references are dropped
586 * and waiters are woken up.
588 void btrfs_remove_ordered_extent(struct inode
*inode
,
589 struct btrfs_ordered_extent
*entry
)
591 struct btrfs_ordered_inode_tree
*tree
;
592 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
593 struct rb_node
*node
;
594 bool dec_pending_ordered
= false;
596 tree
= &BTRFS_I(inode
)->ordered_tree
;
597 spin_lock_irq(&tree
->lock
);
598 node
= &entry
->rb_node
;
599 rb_erase(node
, &tree
->tree
);
601 if (tree
->last
== node
)
603 set_bit(BTRFS_ORDERED_COMPLETE
, &entry
->flags
);
604 if (test_and_clear_bit(BTRFS_ORDERED_PENDING
, &entry
->flags
))
605 dec_pending_ordered
= true;
606 spin_unlock_irq(&tree
->lock
);
609 * The current running transaction is waiting on us, we need to let it
610 * know that we're complete and wake it up.
612 if (dec_pending_ordered
) {
613 struct btrfs_transaction
*trans
;
616 * The checks for trans are just a formality, it should be set,
617 * but if it isn't we don't want to deref/assert under the spin
618 * lock, so be nice and check if trans is set, but ASSERT() so
619 * if it isn't set a developer will notice.
621 spin_lock(&root
->fs_info
->trans_lock
);
622 trans
= root
->fs_info
->running_transaction
;
624 atomic_inc(&trans
->use_count
);
625 spin_unlock(&root
->fs_info
->trans_lock
);
629 if (atomic_dec_and_test(&trans
->pending_ordered
))
630 wake_up(&trans
->pending_wait
);
631 btrfs_put_transaction(trans
);
635 spin_lock(&root
->ordered_extent_lock
);
636 list_del_init(&entry
->root_extent_list
);
637 root
->nr_ordered_extents
--;
639 trace_btrfs_ordered_extent_remove(inode
, entry
);
641 if (!root
->nr_ordered_extents
) {
642 spin_lock(&root
->fs_info
->ordered_root_lock
);
643 BUG_ON(list_empty(&root
->ordered_root
));
644 list_del_init(&root
->ordered_root
);
645 spin_unlock(&root
->fs_info
->ordered_root_lock
);
647 spin_unlock(&root
->ordered_extent_lock
);
648 wake_up(&entry
->wait
);
651 static void btrfs_run_ordered_extent_work(struct btrfs_work
*work
)
653 struct btrfs_ordered_extent
*ordered
;
655 ordered
= container_of(work
, struct btrfs_ordered_extent
, flush_work
);
656 btrfs_start_ordered_extent(ordered
->inode
, ordered
, 1);
657 complete(&ordered
->completion
);
661 * wait for all the ordered extents in a root. This is done when balancing
662 * space between drives.
664 int btrfs_wait_ordered_extents(struct btrfs_root
*root
, int nr
)
666 struct list_head splice
, works
;
667 struct btrfs_ordered_extent
*ordered
, *next
;
670 INIT_LIST_HEAD(&splice
);
671 INIT_LIST_HEAD(&works
);
673 mutex_lock(&root
->ordered_extent_mutex
);
674 spin_lock(&root
->ordered_extent_lock
);
675 list_splice_init(&root
->ordered_extents
, &splice
);
676 while (!list_empty(&splice
) && nr
) {
677 ordered
= list_first_entry(&splice
, struct btrfs_ordered_extent
,
679 list_move_tail(&ordered
->root_extent_list
,
680 &root
->ordered_extents
);
681 atomic_inc(&ordered
->refs
);
682 spin_unlock(&root
->ordered_extent_lock
);
684 btrfs_init_work(&ordered
->flush_work
,
685 btrfs_flush_delalloc_helper
,
686 btrfs_run_ordered_extent_work
, NULL
, NULL
);
687 list_add_tail(&ordered
->work_list
, &works
);
688 btrfs_queue_work(root
->fs_info
->flush_workers
,
689 &ordered
->flush_work
);
692 spin_lock(&root
->ordered_extent_lock
);
697 list_splice_tail(&splice
, &root
->ordered_extents
);
698 spin_unlock(&root
->ordered_extent_lock
);
700 list_for_each_entry_safe(ordered
, next
, &works
, work_list
) {
701 list_del_init(&ordered
->work_list
);
702 wait_for_completion(&ordered
->completion
);
703 btrfs_put_ordered_extent(ordered
);
706 mutex_unlock(&root
->ordered_extent_mutex
);
711 void btrfs_wait_ordered_roots(struct btrfs_fs_info
*fs_info
, int nr
)
713 struct btrfs_root
*root
;
714 struct list_head splice
;
717 INIT_LIST_HEAD(&splice
);
719 mutex_lock(&fs_info
->ordered_operations_mutex
);
720 spin_lock(&fs_info
->ordered_root_lock
);
721 list_splice_init(&fs_info
->ordered_roots
, &splice
);
722 while (!list_empty(&splice
) && nr
) {
723 root
= list_first_entry(&splice
, struct btrfs_root
,
725 root
= btrfs_grab_fs_root(root
);
727 list_move_tail(&root
->ordered_root
,
728 &fs_info
->ordered_roots
);
729 spin_unlock(&fs_info
->ordered_root_lock
);
731 done
= btrfs_wait_ordered_extents(root
, nr
);
732 btrfs_put_fs_root(root
);
734 spin_lock(&fs_info
->ordered_root_lock
);
740 list_splice_tail(&splice
, &fs_info
->ordered_roots
);
741 spin_unlock(&fs_info
->ordered_root_lock
);
742 mutex_unlock(&fs_info
->ordered_operations_mutex
);
746 * Used to start IO or wait for a given ordered extent to finish.
748 * If wait is one, this effectively waits on page writeback for all the pages
749 * in the extent, and it waits on the io completion code to insert
750 * metadata into the btree corresponding to the extent
752 void btrfs_start_ordered_extent(struct inode
*inode
,
753 struct btrfs_ordered_extent
*entry
,
756 u64 start
= entry
->file_offset
;
757 u64 end
= start
+ entry
->len
- 1;
759 trace_btrfs_ordered_extent_start(inode
, entry
);
762 * pages in the range can be dirty, clean or writeback. We
763 * start IO on any dirty ones so the wait doesn't stall waiting
764 * for the flusher thread to find them
766 if (!test_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
))
767 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
769 wait_event(entry
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
775 * Used to wait on ordered extents across a large range of bytes.
777 int btrfs_wait_ordered_range(struct inode
*inode
, u64 start
, u64 len
)
783 struct btrfs_ordered_extent
*ordered
;
785 if (start
+ len
< start
) {
786 orig_end
= INT_LIMIT(loff_t
);
788 orig_end
= start
+ len
- 1;
789 if (orig_end
> INT_LIMIT(loff_t
))
790 orig_end
= INT_LIMIT(loff_t
);
793 /* start IO across the range first to instantiate any delalloc
796 ret
= btrfs_fdatawrite_range(inode
, start
, orig_end
);
801 * If we have a writeback error don't return immediately. Wait first
802 * for any ordered extents that haven't completed yet. This is to make
803 * sure no one can dirty the same page ranges and call writepages()
804 * before the ordered extents complete - to avoid failures (-EEXIST)
805 * when adding the new ordered extents to the ordered tree.
807 ret_wb
= filemap_fdatawait_range(inode
->i_mapping
, start
, orig_end
);
811 ordered
= btrfs_lookup_first_ordered_extent(inode
, end
);
814 if (ordered
->file_offset
> orig_end
) {
815 btrfs_put_ordered_extent(ordered
);
818 if (ordered
->file_offset
+ ordered
->len
<= start
) {
819 btrfs_put_ordered_extent(ordered
);
822 btrfs_start_ordered_extent(inode
, ordered
, 1);
823 end
= ordered
->file_offset
;
824 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
))
826 btrfs_put_ordered_extent(ordered
);
827 if (ret
|| end
== 0 || end
== start
)
831 return ret_wb
? ret_wb
: ret
;
835 * find an ordered extent corresponding to file_offset. return NULL if
836 * nothing is found, otherwise take a reference on the extent and return it
838 struct btrfs_ordered_extent
*btrfs_lookup_ordered_extent(struct inode
*inode
,
841 struct btrfs_ordered_inode_tree
*tree
;
842 struct rb_node
*node
;
843 struct btrfs_ordered_extent
*entry
= NULL
;
845 tree
= &BTRFS_I(inode
)->ordered_tree
;
846 spin_lock_irq(&tree
->lock
);
847 node
= tree_search(tree
, file_offset
);
851 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
852 if (!offset_in_entry(entry
, file_offset
))
855 atomic_inc(&entry
->refs
);
857 spin_unlock_irq(&tree
->lock
);
861 /* Since the DIO code tries to lock a wide area we need to look for any ordered
862 * extents that exist in the range, rather than just the start of the range.
864 struct btrfs_ordered_extent
*btrfs_lookup_ordered_range(struct inode
*inode
,
868 struct btrfs_ordered_inode_tree
*tree
;
869 struct rb_node
*node
;
870 struct btrfs_ordered_extent
*entry
= NULL
;
872 tree
= &BTRFS_I(inode
)->ordered_tree
;
873 spin_lock_irq(&tree
->lock
);
874 node
= tree_search(tree
, file_offset
);
876 node
= tree_search(tree
, file_offset
+ len
);
882 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
883 if (range_overlaps(entry
, file_offset
, len
))
886 if (entry
->file_offset
>= file_offset
+ len
) {
891 node
= rb_next(node
);
897 atomic_inc(&entry
->refs
);
898 spin_unlock_irq(&tree
->lock
);
902 bool btrfs_have_ordered_extents_in_range(struct inode
*inode
,
906 struct btrfs_ordered_extent
*oe
;
908 oe
= btrfs_lookup_ordered_range(inode
, file_offset
, len
);
910 btrfs_put_ordered_extent(oe
);
917 * lookup and return any extent before 'file_offset'. NULL is returned
920 struct btrfs_ordered_extent
*
921 btrfs_lookup_first_ordered_extent(struct inode
*inode
, u64 file_offset
)
923 struct btrfs_ordered_inode_tree
*tree
;
924 struct rb_node
*node
;
925 struct btrfs_ordered_extent
*entry
= NULL
;
927 tree
= &BTRFS_I(inode
)->ordered_tree
;
928 spin_lock_irq(&tree
->lock
);
929 node
= tree_search(tree
, file_offset
);
933 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
934 atomic_inc(&entry
->refs
);
936 spin_unlock_irq(&tree
->lock
);
941 * After an extent is done, call this to conditionally update the on disk
942 * i_size. i_size is updated to cover any fully written part of the file.
944 int btrfs_ordered_update_i_size(struct inode
*inode
, u64 offset
,
945 struct btrfs_ordered_extent
*ordered
)
947 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
950 u64 i_size
= i_size_read(inode
);
951 struct rb_node
*node
;
952 struct rb_node
*prev
= NULL
;
953 struct btrfs_ordered_extent
*test
;
956 spin_lock_irq(&tree
->lock
);
958 offset
= entry_end(ordered
);
959 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
))
961 ordered
->file_offset
+
962 ordered
->truncated_len
);
964 offset
= ALIGN(offset
, BTRFS_I(inode
)->root
->sectorsize
);
966 disk_i_size
= BTRFS_I(inode
)->disk_i_size
;
969 if (disk_i_size
> i_size
) {
970 BTRFS_I(inode
)->disk_i_size
= i_size
;
976 * if the disk i_size is already at the inode->i_size, or
977 * this ordered extent is inside the disk i_size, we're done
979 if (disk_i_size
== i_size
)
983 * We still need to update disk_i_size if outstanding_isize is greater
986 if (offset
<= disk_i_size
&&
987 (!ordered
|| ordered
->outstanding_isize
<= disk_i_size
))
991 * walk backward from this ordered extent to disk_i_size.
992 * if we find an ordered extent then we can't update disk i_size
996 node
= rb_prev(&ordered
->rb_node
);
998 prev
= tree_search(tree
, offset
);
1000 * we insert file extents without involving ordered struct,
1001 * so there should be no ordered struct cover this offset
1004 test
= rb_entry(prev
, struct btrfs_ordered_extent
,
1006 BUG_ON(offset_in_entry(test
, offset
));
1010 for (; node
; node
= rb_prev(node
)) {
1011 test
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
1013 /* We treat this entry as if it doesn't exist */
1014 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &test
->flags
))
1016 if (test
->file_offset
+ test
->len
<= disk_i_size
)
1018 if (test
->file_offset
>= i_size
)
1020 if (entry_end(test
) > disk_i_size
) {
1022 * we don't update disk_i_size now, so record this
1023 * undealt i_size. Or we will not know the real
1026 if (test
->outstanding_isize
< offset
)
1027 test
->outstanding_isize
= offset
;
1029 ordered
->outstanding_isize
>
1030 test
->outstanding_isize
)
1031 test
->outstanding_isize
=
1032 ordered
->outstanding_isize
;
1036 new_i_size
= min_t(u64
, offset
, i_size
);
1039 * Some ordered extents may completed before the current one, and
1040 * we hold the real i_size in ->outstanding_isize.
1042 if (ordered
&& ordered
->outstanding_isize
> new_i_size
)
1043 new_i_size
= min_t(u64
, ordered
->outstanding_isize
, i_size
);
1044 BTRFS_I(inode
)->disk_i_size
= new_i_size
;
1048 * We need to do this because we can't remove ordered extents until
1049 * after the i_disk_size has been updated and then the inode has been
1050 * updated to reflect the change, so we need to tell anybody who finds
1051 * this ordered extent that we've already done all the real work, we
1052 * just haven't completed all the other work.
1055 set_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &ordered
->flags
);
1056 spin_unlock_irq(&tree
->lock
);
1061 * search the ordered extents for one corresponding to 'offset' and
1062 * try to find a checksum. This is used because we allow pages to
1063 * be reclaimed before their checksum is actually put into the btree
1065 int btrfs_find_ordered_sum(struct inode
*inode
, u64 offset
, u64 disk_bytenr
,
1068 struct btrfs_ordered_sum
*ordered_sum
;
1069 struct btrfs_ordered_extent
*ordered
;
1070 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
1071 unsigned long num_sectors
;
1073 u32 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
1076 ordered
= btrfs_lookup_ordered_extent(inode
, offset
);
1080 spin_lock_irq(&tree
->lock
);
1081 list_for_each_entry_reverse(ordered_sum
, &ordered
->list
, list
) {
1082 if (disk_bytenr
>= ordered_sum
->bytenr
&&
1083 disk_bytenr
< ordered_sum
->bytenr
+ ordered_sum
->len
) {
1084 i
= (disk_bytenr
- ordered_sum
->bytenr
) >>
1085 inode
->i_sb
->s_blocksize_bits
;
1086 num_sectors
= ordered_sum
->len
>>
1087 inode
->i_sb
->s_blocksize_bits
;
1088 num_sectors
= min_t(int, len
- index
, num_sectors
- i
);
1089 memcpy(sum
+ index
, ordered_sum
->sums
+ i
,
1092 index
+= (int)num_sectors
;
1095 disk_bytenr
+= num_sectors
* sectorsize
;
1099 spin_unlock_irq(&tree
->lock
);
1100 btrfs_put_ordered_extent(ordered
);
1104 int __init
ordered_data_init(void)
1106 btrfs_ordered_extent_cache
= kmem_cache_create("btrfs_ordered_extent",
1107 sizeof(struct btrfs_ordered_extent
), 0,
1108 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
1110 if (!btrfs_ordered_extent_cache
)
1116 void ordered_data_exit(void)
1118 kmem_cache_destroy(btrfs_ordered_extent_cache
);