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"
29 static struct kmem_cache
*btrfs_ordered_extent_cache
;
31 static u64
entry_end(struct btrfs_ordered_extent
*entry
)
33 if (entry
->file_offset
+ entry
->len
< entry
->file_offset
)
35 return entry
->file_offset
+ entry
->len
;
38 /* returns NULL if the insertion worked, or it returns the node it did find
41 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 file_offset
,
44 struct rb_node
**p
= &root
->rb_node
;
45 struct rb_node
*parent
= NULL
;
46 struct btrfs_ordered_extent
*entry
;
50 entry
= rb_entry(parent
, struct btrfs_ordered_extent
, rb_node
);
52 if (file_offset
< entry
->file_offset
)
54 else if (file_offset
>= entry_end(entry
))
60 rb_link_node(node
, parent
, p
);
61 rb_insert_color(node
, root
);
65 static void ordered_data_tree_panic(struct inode
*inode
, int errno
,
68 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
69 btrfs_panic(fs_info
, errno
, "Inconsistency in ordered tree at offset "
74 * look for a given offset in the tree, and if it can't be found return the
77 static struct rb_node
*__tree_search(struct rb_root
*root
, u64 file_offset
,
78 struct rb_node
**prev_ret
)
80 struct rb_node
*n
= root
->rb_node
;
81 struct rb_node
*prev
= NULL
;
83 struct btrfs_ordered_extent
*entry
;
84 struct btrfs_ordered_extent
*prev_entry
= NULL
;
87 entry
= rb_entry(n
, struct btrfs_ordered_extent
, rb_node
);
91 if (file_offset
< entry
->file_offset
)
93 else if (file_offset
>= entry_end(entry
))
101 while (prev
&& file_offset
>= entry_end(prev_entry
)) {
102 test
= rb_next(prev
);
105 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
107 if (file_offset
< entry_end(prev_entry
))
113 prev_entry
= rb_entry(prev
, struct btrfs_ordered_extent
,
115 while (prev
&& file_offset
< entry_end(prev_entry
)) {
116 test
= rb_prev(prev
);
119 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
128 * helper to check if a given offset is inside a given entry
130 static int offset_in_entry(struct btrfs_ordered_extent
*entry
, u64 file_offset
)
132 if (file_offset
< entry
->file_offset
||
133 entry
->file_offset
+ entry
->len
<= file_offset
)
138 static int range_overlaps(struct btrfs_ordered_extent
*entry
, u64 file_offset
,
141 if (file_offset
+ len
<= entry
->file_offset
||
142 entry
->file_offset
+ entry
->len
<= file_offset
)
148 * look find the first ordered struct that has this offset, otherwise
149 * the first one less than this offset
151 static inline struct rb_node
*tree_search(struct btrfs_ordered_inode_tree
*tree
,
154 struct rb_root
*root
= &tree
->tree
;
155 struct rb_node
*prev
= NULL
;
157 struct btrfs_ordered_extent
*entry
;
160 entry
= rb_entry(tree
->last
, struct btrfs_ordered_extent
,
162 if (offset_in_entry(entry
, file_offset
))
165 ret
= __tree_search(root
, file_offset
, &prev
);
173 /* allocate and add a new ordered_extent into the per-inode tree.
174 * file_offset is the logical offset in the file
176 * start is the disk block number of an extent already reserved in the
177 * extent allocation tree
179 * len is the length of the extent
181 * The tree is given a single reference on the ordered extent that was
184 static int __btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
185 u64 start
, u64 len
, u64 disk_len
,
186 int type
, int dio
, int compress_type
)
188 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
189 struct btrfs_ordered_inode_tree
*tree
;
190 struct rb_node
*node
;
191 struct btrfs_ordered_extent
*entry
;
193 tree
= &BTRFS_I(inode
)->ordered_tree
;
194 entry
= kmem_cache_zalloc(btrfs_ordered_extent_cache
, GFP_NOFS
);
198 entry
->file_offset
= file_offset
;
199 entry
->start
= start
;
201 entry
->disk_len
= disk_len
;
202 entry
->bytes_left
= len
;
203 entry
->inode
= igrab(inode
);
204 entry
->compress_type
= compress_type
;
205 entry
->truncated_len
= (u64
)-1;
206 if (type
!= BTRFS_ORDERED_IO_DONE
&& type
!= BTRFS_ORDERED_COMPLETE
)
207 set_bit(type
, &entry
->flags
);
210 set_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
);
212 /* one ref for the tree */
213 atomic_set(&entry
->refs
, 1);
214 init_waitqueue_head(&entry
->wait
);
215 INIT_LIST_HEAD(&entry
->list
);
216 INIT_LIST_HEAD(&entry
->root_extent_list
);
217 INIT_LIST_HEAD(&entry
->work_list
);
218 init_completion(&entry
->completion
);
219 INIT_LIST_HEAD(&entry
->log_list
);
220 INIT_LIST_HEAD(&entry
->trans_list
);
222 trace_btrfs_ordered_extent_add(inode
, entry
);
224 spin_lock_irq(&tree
->lock
);
225 node
= tree_insert(&tree
->tree
, file_offset
,
228 ordered_data_tree_panic(inode
, -EEXIST
, file_offset
);
229 spin_unlock_irq(&tree
->lock
);
231 spin_lock(&root
->ordered_extent_lock
);
232 list_add_tail(&entry
->root_extent_list
,
233 &root
->ordered_extents
);
234 root
->nr_ordered_extents
++;
235 if (root
->nr_ordered_extents
== 1) {
236 spin_lock(&root
->fs_info
->ordered_root_lock
);
237 BUG_ON(!list_empty(&root
->ordered_root
));
238 list_add_tail(&root
->ordered_root
,
239 &root
->fs_info
->ordered_roots
);
240 spin_unlock(&root
->fs_info
->ordered_root_lock
);
242 spin_unlock(&root
->ordered_extent_lock
);
247 int btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
248 u64 start
, u64 len
, u64 disk_len
, int type
)
250 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
252 BTRFS_COMPRESS_NONE
);
255 int btrfs_add_ordered_extent_dio(struct inode
*inode
, u64 file_offset
,
256 u64 start
, u64 len
, u64 disk_len
, int type
)
258 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
260 BTRFS_COMPRESS_NONE
);
263 int btrfs_add_ordered_extent_compress(struct inode
*inode
, u64 file_offset
,
264 u64 start
, u64 len
, u64 disk_len
,
265 int type
, int compress_type
)
267 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
273 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
274 * when an ordered extent is finished. If the list covers more than one
275 * ordered extent, it is split across multiples.
277 void btrfs_add_ordered_sum(struct inode
*inode
,
278 struct btrfs_ordered_extent
*entry
,
279 struct btrfs_ordered_sum
*sum
)
281 struct btrfs_ordered_inode_tree
*tree
;
283 tree
= &BTRFS_I(inode
)->ordered_tree
;
284 spin_lock_irq(&tree
->lock
);
285 list_add_tail(&sum
->list
, &entry
->list
);
286 spin_unlock_irq(&tree
->lock
);
290 * this is used to account for finished IO across a given range
291 * of the file. The IO may span ordered extents. If
292 * a given ordered_extent is completely done, 1 is returned, otherwise
295 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
296 * to make sure this function only returns 1 once for a given ordered extent.
298 * file_offset is updated to one byte past the range that is recorded as
299 * complete. This allows you to walk forward in the file.
301 int btrfs_dec_test_first_ordered_pending(struct inode
*inode
,
302 struct btrfs_ordered_extent
**cached
,
303 u64
*file_offset
, u64 io_size
, int uptodate
)
305 struct btrfs_ordered_inode_tree
*tree
;
306 struct rb_node
*node
;
307 struct btrfs_ordered_extent
*entry
= NULL
;
314 tree
= &BTRFS_I(inode
)->ordered_tree
;
315 spin_lock_irqsave(&tree
->lock
, flags
);
316 node
= tree_search(tree
, *file_offset
);
322 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
323 if (!offset_in_entry(entry
, *file_offset
)) {
328 dec_start
= max(*file_offset
, entry
->file_offset
);
329 dec_end
= min(*file_offset
+ io_size
, entry
->file_offset
+
331 *file_offset
= dec_end
;
332 if (dec_start
> dec_end
) {
333 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
334 "bad ordering dec_start %llu end %llu", dec_start
, dec_end
);
336 to_dec
= dec_end
- dec_start
;
337 if (to_dec
> entry
->bytes_left
) {
338 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
339 "bad ordered accounting left %llu size %llu",
340 entry
->bytes_left
, to_dec
);
342 entry
->bytes_left
-= to_dec
;
344 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
346 if (entry
->bytes_left
== 0) {
347 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
349 * Implicit memory barrier after test_and_set_bit
351 if (waitqueue_active(&entry
->wait
))
352 wake_up(&entry
->wait
);
357 if (!ret
&& cached
&& entry
) {
359 atomic_inc(&entry
->refs
);
361 spin_unlock_irqrestore(&tree
->lock
, flags
);
366 * this is used to account for finished IO across a given range
367 * of the file. The IO should not span ordered extents. If
368 * a given ordered_extent is completely done, 1 is returned, otherwise
371 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
372 * to make sure this function only returns 1 once for a given ordered extent.
374 int btrfs_dec_test_ordered_pending(struct inode
*inode
,
375 struct btrfs_ordered_extent
**cached
,
376 u64 file_offset
, u64 io_size
, int uptodate
)
378 struct btrfs_ordered_inode_tree
*tree
;
379 struct rb_node
*node
;
380 struct btrfs_ordered_extent
*entry
= NULL
;
384 tree
= &BTRFS_I(inode
)->ordered_tree
;
385 spin_lock_irqsave(&tree
->lock
, flags
);
386 if (cached
&& *cached
) {
391 node
= tree_search(tree
, file_offset
);
397 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
399 if (!offset_in_entry(entry
, file_offset
)) {
404 if (io_size
> entry
->bytes_left
) {
405 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
406 "bad ordered accounting left %llu size %llu",
407 entry
->bytes_left
, io_size
);
409 entry
->bytes_left
-= io_size
;
411 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
413 if (entry
->bytes_left
== 0) {
414 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
416 * Implicit memory barrier after test_and_set_bit
418 if (waitqueue_active(&entry
->wait
))
419 wake_up(&entry
->wait
);
424 if (!ret
&& cached
&& entry
) {
426 atomic_inc(&entry
->refs
);
428 spin_unlock_irqrestore(&tree
->lock
, flags
);
432 /* Needs to either be called under a log transaction or the log_mutex */
433 void btrfs_get_logged_extents(struct inode
*inode
,
434 struct list_head
*logged_list
,
438 struct btrfs_ordered_inode_tree
*tree
;
439 struct btrfs_ordered_extent
*ordered
;
441 struct rb_node
*prev
;
443 tree
= &BTRFS_I(inode
)->ordered_tree
;
444 spin_lock_irq(&tree
->lock
);
445 n
= __tree_search(&tree
->tree
, end
, &prev
);
448 for (; n
; n
= rb_prev(n
)) {
449 ordered
= rb_entry(n
, struct btrfs_ordered_extent
, rb_node
);
450 if (ordered
->file_offset
> end
)
452 if (entry_end(ordered
) <= start
)
454 if (test_and_set_bit(BTRFS_ORDERED_LOGGED
, &ordered
->flags
))
456 list_add(&ordered
->log_list
, logged_list
);
457 atomic_inc(&ordered
->refs
);
459 spin_unlock_irq(&tree
->lock
);
462 void btrfs_put_logged_extents(struct list_head
*logged_list
)
464 struct btrfs_ordered_extent
*ordered
;
466 while (!list_empty(logged_list
)) {
467 ordered
= list_first_entry(logged_list
,
468 struct btrfs_ordered_extent
,
470 list_del_init(&ordered
->log_list
);
471 btrfs_put_ordered_extent(ordered
);
475 void btrfs_submit_logged_extents(struct list_head
*logged_list
,
476 struct btrfs_root
*log
)
478 int index
= log
->log_transid
% 2;
480 spin_lock_irq(&log
->log_extents_lock
[index
]);
481 list_splice_tail(logged_list
, &log
->logged_list
[index
]);
482 spin_unlock_irq(&log
->log_extents_lock
[index
]);
485 void btrfs_wait_logged_extents(struct btrfs_trans_handle
*trans
,
486 struct btrfs_root
*log
, u64 transid
)
488 struct btrfs_ordered_extent
*ordered
;
489 int index
= transid
% 2;
491 spin_lock_irq(&log
->log_extents_lock
[index
]);
492 while (!list_empty(&log
->logged_list
[index
])) {
494 ordered
= list_first_entry(&log
->logged_list
[index
],
495 struct btrfs_ordered_extent
,
497 list_del_init(&ordered
->log_list
);
498 inode
= ordered
->inode
;
499 spin_unlock_irq(&log
->log_extents_lock
[index
]);
501 if (!test_bit(BTRFS_ORDERED_IO_DONE
, &ordered
->flags
) &&
502 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
)) {
503 u64 start
= ordered
->file_offset
;
504 u64 end
= ordered
->file_offset
+ ordered
->len
- 1;
507 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
509 wait_event(ordered
->wait
, test_bit(BTRFS_ORDERED_IO_DONE
,
513 * In order to keep us from losing our ordered extent
514 * information when committing the transaction we have to make
515 * sure that any logged extents are completed when we go to
516 * commit the transaction. To do this we simply increase the
517 * current transactions pending_ordered counter and decrement it
518 * when the ordered extent completes.
520 if (!test_bit(BTRFS_ORDERED_COMPLETE
, &ordered
->flags
)) {
521 struct btrfs_ordered_inode_tree
*tree
;
523 tree
= &BTRFS_I(inode
)->ordered_tree
;
524 spin_lock_irq(&tree
->lock
);
525 if (!test_bit(BTRFS_ORDERED_COMPLETE
, &ordered
->flags
)) {
526 set_bit(BTRFS_ORDERED_PENDING
, &ordered
->flags
);
527 atomic_inc(&trans
->transaction
->pending_ordered
);
529 spin_unlock_irq(&tree
->lock
);
531 btrfs_put_ordered_extent(ordered
);
532 spin_lock_irq(&log
->log_extents_lock
[index
]);
534 spin_unlock_irq(&log
->log_extents_lock
[index
]);
537 void btrfs_free_logged_extents(struct btrfs_root
*log
, u64 transid
)
539 struct btrfs_ordered_extent
*ordered
;
540 int index
= transid
% 2;
542 spin_lock_irq(&log
->log_extents_lock
[index
]);
543 while (!list_empty(&log
->logged_list
[index
])) {
544 ordered
= list_first_entry(&log
->logged_list
[index
],
545 struct btrfs_ordered_extent
,
547 list_del_init(&ordered
->log_list
);
548 spin_unlock_irq(&log
->log_extents_lock
[index
]);
549 btrfs_put_ordered_extent(ordered
);
550 spin_lock_irq(&log
->log_extents_lock
[index
]);
552 spin_unlock_irq(&log
->log_extents_lock
[index
]);
556 * used to drop a reference on an ordered extent. This will free
557 * the extent if the last reference is dropped
559 void btrfs_put_ordered_extent(struct btrfs_ordered_extent
*entry
)
561 struct list_head
*cur
;
562 struct btrfs_ordered_sum
*sum
;
564 trace_btrfs_ordered_extent_put(entry
->inode
, entry
);
566 if (atomic_dec_and_test(&entry
->refs
)) {
567 ASSERT(list_empty(&entry
->log_list
));
568 ASSERT(list_empty(&entry
->trans_list
));
569 ASSERT(list_empty(&entry
->root_extent_list
));
570 ASSERT(RB_EMPTY_NODE(&entry
->rb_node
));
572 btrfs_add_delayed_iput(entry
->inode
);
573 while (!list_empty(&entry
->list
)) {
574 cur
= entry
->list
.next
;
575 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
576 list_del(&sum
->list
);
579 kmem_cache_free(btrfs_ordered_extent_cache
, entry
);
584 * remove an ordered extent from the tree. No references are dropped
585 * and waiters are woken up.
587 void btrfs_remove_ordered_extent(struct inode
*inode
,
588 struct btrfs_ordered_extent
*entry
)
590 struct btrfs_ordered_inode_tree
*tree
;
591 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
592 struct rb_node
*node
;
593 bool dec_pending_ordered
= false;
595 tree
= &BTRFS_I(inode
)->ordered_tree
;
596 spin_lock_irq(&tree
->lock
);
597 node
= &entry
->rb_node
;
598 rb_erase(node
, &tree
->tree
);
600 if (tree
->last
== node
)
602 set_bit(BTRFS_ORDERED_COMPLETE
, &entry
->flags
);
603 if (test_and_clear_bit(BTRFS_ORDERED_PENDING
, &entry
->flags
))
604 dec_pending_ordered
= true;
605 spin_unlock_irq(&tree
->lock
);
608 * The current running transaction is waiting on us, we need to let it
609 * know that we're complete and wake it up.
611 if (dec_pending_ordered
) {
612 struct btrfs_transaction
*trans
;
615 * The checks for trans are just a formality, it should be set,
616 * but if it isn't we don't want to deref/assert under the spin
617 * lock, so be nice and check if trans is set, but ASSERT() so
618 * if it isn't set a developer will notice.
620 spin_lock(&root
->fs_info
->trans_lock
);
621 trans
= root
->fs_info
->running_transaction
;
623 atomic_inc(&trans
->use_count
);
624 spin_unlock(&root
->fs_info
->trans_lock
);
628 if (atomic_dec_and_test(&trans
->pending_ordered
))
629 wake_up(&trans
->pending_wait
);
630 btrfs_put_transaction(trans
);
634 spin_lock(&root
->ordered_extent_lock
);
635 list_del_init(&entry
->root_extent_list
);
636 root
->nr_ordered_extents
--;
638 trace_btrfs_ordered_extent_remove(inode
, entry
);
640 if (!root
->nr_ordered_extents
) {
641 spin_lock(&root
->fs_info
->ordered_root_lock
);
642 BUG_ON(list_empty(&root
->ordered_root
));
643 list_del_init(&root
->ordered_root
);
644 spin_unlock(&root
->fs_info
->ordered_root_lock
);
646 spin_unlock(&root
->ordered_extent_lock
);
647 wake_up(&entry
->wait
);
650 static void btrfs_run_ordered_extent_work(struct btrfs_work
*work
)
652 struct btrfs_ordered_extent
*ordered
;
654 ordered
= container_of(work
, struct btrfs_ordered_extent
, flush_work
);
655 btrfs_start_ordered_extent(ordered
->inode
, ordered
, 1);
656 complete(&ordered
->completion
);
660 * wait for all the ordered extents in a root. This is done when balancing
661 * space between drives.
663 int btrfs_wait_ordered_extents(struct btrfs_root
*root
, int nr
)
665 struct list_head splice
, works
;
666 struct btrfs_ordered_extent
*ordered
, *next
;
669 INIT_LIST_HEAD(&splice
);
670 INIT_LIST_HEAD(&works
);
672 mutex_lock(&root
->ordered_extent_mutex
);
673 spin_lock(&root
->ordered_extent_lock
);
674 list_splice_init(&root
->ordered_extents
, &splice
);
675 while (!list_empty(&splice
) && nr
) {
676 ordered
= list_first_entry(&splice
, struct btrfs_ordered_extent
,
678 list_move_tail(&ordered
->root_extent_list
,
679 &root
->ordered_extents
);
680 atomic_inc(&ordered
->refs
);
681 spin_unlock(&root
->ordered_extent_lock
);
683 btrfs_init_work(&ordered
->flush_work
,
684 btrfs_flush_delalloc_helper
,
685 btrfs_run_ordered_extent_work
, NULL
, NULL
);
686 list_add_tail(&ordered
->work_list
, &works
);
687 btrfs_queue_work(root
->fs_info
->flush_workers
,
688 &ordered
->flush_work
);
691 spin_lock(&root
->ordered_extent_lock
);
696 list_splice_tail(&splice
, &root
->ordered_extents
);
697 spin_unlock(&root
->ordered_extent_lock
);
699 list_for_each_entry_safe(ordered
, next
, &works
, work_list
) {
700 list_del_init(&ordered
->work_list
);
701 wait_for_completion(&ordered
->completion
);
702 btrfs_put_ordered_extent(ordered
);
705 mutex_unlock(&root
->ordered_extent_mutex
);
710 void btrfs_wait_ordered_roots(struct btrfs_fs_info
*fs_info
, int nr
)
712 struct btrfs_root
*root
;
713 struct list_head splice
;
716 INIT_LIST_HEAD(&splice
);
718 mutex_lock(&fs_info
->ordered_operations_mutex
);
719 spin_lock(&fs_info
->ordered_root_lock
);
720 list_splice_init(&fs_info
->ordered_roots
, &splice
);
721 while (!list_empty(&splice
) && nr
) {
722 root
= list_first_entry(&splice
, struct btrfs_root
,
724 root
= btrfs_grab_fs_root(root
);
726 list_move_tail(&root
->ordered_root
,
727 &fs_info
->ordered_roots
);
728 spin_unlock(&fs_info
->ordered_root_lock
);
730 done
= btrfs_wait_ordered_extents(root
, nr
);
731 btrfs_put_fs_root(root
);
733 spin_lock(&fs_info
->ordered_root_lock
);
739 list_splice_tail(&splice
, &fs_info
->ordered_roots
);
740 spin_unlock(&fs_info
->ordered_root_lock
);
741 mutex_unlock(&fs_info
->ordered_operations_mutex
);
745 * Used to start IO or wait for a given ordered extent to finish.
747 * If wait is one, this effectively waits on page writeback for all the pages
748 * in the extent, and it waits on the io completion code to insert
749 * metadata into the btree corresponding to the extent
751 void btrfs_start_ordered_extent(struct inode
*inode
,
752 struct btrfs_ordered_extent
*entry
,
755 u64 start
= entry
->file_offset
;
756 u64 end
= start
+ entry
->len
- 1;
758 trace_btrfs_ordered_extent_start(inode
, entry
);
761 * pages in the range can be dirty, clean or writeback. We
762 * start IO on any dirty ones so the wait doesn't stall waiting
763 * for the flusher thread to find them
765 if (!test_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
))
766 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
768 wait_event(entry
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
774 * Used to wait on ordered extents across a large range of bytes.
776 int btrfs_wait_ordered_range(struct inode
*inode
, u64 start
, u64 len
)
782 struct btrfs_ordered_extent
*ordered
;
784 if (start
+ len
< start
) {
785 orig_end
= INT_LIMIT(loff_t
);
787 orig_end
= start
+ len
- 1;
788 if (orig_end
> INT_LIMIT(loff_t
))
789 orig_end
= INT_LIMIT(loff_t
);
792 /* start IO across the range first to instantiate any delalloc
795 ret
= btrfs_fdatawrite_range(inode
, start
, orig_end
);
800 * If we have a writeback error don't return immediately. Wait first
801 * for any ordered extents that haven't completed yet. This is to make
802 * sure no one can dirty the same page ranges and call writepages()
803 * before the ordered extents complete - to avoid failures (-EEXIST)
804 * when adding the new ordered extents to the ordered tree.
806 ret_wb
= filemap_fdatawait_range(inode
->i_mapping
, start
, orig_end
);
810 ordered
= btrfs_lookup_first_ordered_extent(inode
, end
);
813 if (ordered
->file_offset
> orig_end
) {
814 btrfs_put_ordered_extent(ordered
);
817 if (ordered
->file_offset
+ ordered
->len
<= start
) {
818 btrfs_put_ordered_extent(ordered
);
821 btrfs_start_ordered_extent(inode
, ordered
, 1);
822 end
= ordered
->file_offset
;
823 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
))
825 btrfs_put_ordered_extent(ordered
);
826 if (ret
|| end
== 0 || end
== start
)
830 return ret_wb
? ret_wb
: ret
;
834 * find an ordered extent corresponding to file_offset. return NULL if
835 * nothing is found, otherwise take a reference on the extent and return it
837 struct btrfs_ordered_extent
*btrfs_lookup_ordered_extent(struct inode
*inode
,
840 struct btrfs_ordered_inode_tree
*tree
;
841 struct rb_node
*node
;
842 struct btrfs_ordered_extent
*entry
= NULL
;
844 tree
= &BTRFS_I(inode
)->ordered_tree
;
845 spin_lock_irq(&tree
->lock
);
846 node
= tree_search(tree
, file_offset
);
850 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
851 if (!offset_in_entry(entry
, file_offset
))
854 atomic_inc(&entry
->refs
);
856 spin_unlock_irq(&tree
->lock
);
860 /* Since the DIO code tries to lock a wide area we need to look for any ordered
861 * extents that exist in the range, rather than just the start of the range.
863 struct btrfs_ordered_extent
*btrfs_lookup_ordered_range(struct inode
*inode
,
867 struct btrfs_ordered_inode_tree
*tree
;
868 struct rb_node
*node
;
869 struct btrfs_ordered_extent
*entry
= NULL
;
871 tree
= &BTRFS_I(inode
)->ordered_tree
;
872 spin_lock_irq(&tree
->lock
);
873 node
= tree_search(tree
, file_offset
);
875 node
= tree_search(tree
, file_offset
+ len
);
881 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
882 if (range_overlaps(entry
, file_offset
, len
))
885 if (entry
->file_offset
>= file_offset
+ len
) {
890 node
= rb_next(node
);
896 atomic_inc(&entry
->refs
);
897 spin_unlock_irq(&tree
->lock
);
901 bool btrfs_have_ordered_extents_in_range(struct inode
*inode
,
905 struct btrfs_ordered_extent
*oe
;
907 oe
= btrfs_lookup_ordered_range(inode
, file_offset
, len
);
909 btrfs_put_ordered_extent(oe
);
916 * lookup and return any extent before 'file_offset'. NULL is returned
919 struct btrfs_ordered_extent
*
920 btrfs_lookup_first_ordered_extent(struct inode
*inode
, u64 file_offset
)
922 struct btrfs_ordered_inode_tree
*tree
;
923 struct rb_node
*node
;
924 struct btrfs_ordered_extent
*entry
= NULL
;
926 tree
= &BTRFS_I(inode
)->ordered_tree
;
927 spin_lock_irq(&tree
->lock
);
928 node
= tree_search(tree
, file_offset
);
932 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
933 atomic_inc(&entry
->refs
);
935 spin_unlock_irq(&tree
->lock
);
940 * After an extent is done, call this to conditionally update the on disk
941 * i_size. i_size is updated to cover any fully written part of the file.
943 int btrfs_ordered_update_i_size(struct inode
*inode
, u64 offset
,
944 struct btrfs_ordered_extent
*ordered
)
946 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
949 u64 i_size
= i_size_read(inode
);
950 struct rb_node
*node
;
951 struct rb_node
*prev
= NULL
;
952 struct btrfs_ordered_extent
*test
;
955 spin_lock_irq(&tree
->lock
);
957 offset
= entry_end(ordered
);
958 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
))
960 ordered
->file_offset
+
961 ordered
->truncated_len
);
963 offset
= ALIGN(offset
, BTRFS_I(inode
)->root
->sectorsize
);
965 disk_i_size
= BTRFS_I(inode
)->disk_i_size
;
968 if (disk_i_size
> i_size
) {
969 BTRFS_I(inode
)->disk_i_size
= i_size
;
975 * if the disk i_size is already at the inode->i_size, or
976 * this ordered extent is inside the disk i_size, we're done
978 if (disk_i_size
== i_size
)
982 * We still need to update disk_i_size if outstanding_isize is greater
985 if (offset
<= disk_i_size
&&
986 (!ordered
|| ordered
->outstanding_isize
<= disk_i_size
))
990 * walk backward from this ordered extent to disk_i_size.
991 * if we find an ordered extent then we can't update disk i_size
995 node
= rb_prev(&ordered
->rb_node
);
997 prev
= tree_search(tree
, offset
);
999 * we insert file extents without involving ordered struct,
1000 * so there should be no ordered struct cover this offset
1003 test
= rb_entry(prev
, struct btrfs_ordered_extent
,
1005 BUG_ON(offset_in_entry(test
, offset
));
1009 for (; node
; node
= rb_prev(node
)) {
1010 test
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
1012 /* We treat this entry as if it doesnt exist */
1013 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &test
->flags
))
1015 if (test
->file_offset
+ test
->len
<= disk_i_size
)
1017 if (test
->file_offset
>= i_size
)
1019 if (entry_end(test
) > disk_i_size
) {
1021 * we don't update disk_i_size now, so record this
1022 * undealt i_size. Or we will not know the real
1025 if (test
->outstanding_isize
< offset
)
1026 test
->outstanding_isize
= offset
;
1028 ordered
->outstanding_isize
>
1029 test
->outstanding_isize
)
1030 test
->outstanding_isize
=
1031 ordered
->outstanding_isize
;
1035 new_i_size
= min_t(u64
, offset
, i_size
);
1038 * Some ordered extents may completed before the current one, and
1039 * we hold the real i_size in ->outstanding_isize.
1041 if (ordered
&& ordered
->outstanding_isize
> new_i_size
)
1042 new_i_size
= min_t(u64
, ordered
->outstanding_isize
, i_size
);
1043 BTRFS_I(inode
)->disk_i_size
= new_i_size
;
1047 * We need to do this because we can't remove ordered extents until
1048 * after the i_disk_size has been updated and then the inode has been
1049 * updated to reflect the change, so we need to tell anybody who finds
1050 * this ordered extent that we've already done all the real work, we
1051 * just haven't completed all the other work.
1054 set_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &ordered
->flags
);
1055 spin_unlock_irq(&tree
->lock
);
1060 * search the ordered extents for one corresponding to 'offset' and
1061 * try to find a checksum. This is used because we allow pages to
1062 * be reclaimed before their checksum is actually put into the btree
1064 int btrfs_find_ordered_sum(struct inode
*inode
, u64 offset
, u64 disk_bytenr
,
1067 struct btrfs_ordered_sum
*ordered_sum
;
1068 struct btrfs_ordered_extent
*ordered
;
1069 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
1070 unsigned long num_sectors
;
1072 u32 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
1075 ordered
= btrfs_lookup_ordered_extent(inode
, offset
);
1079 spin_lock_irq(&tree
->lock
);
1080 list_for_each_entry_reverse(ordered_sum
, &ordered
->list
, list
) {
1081 if (disk_bytenr
>= ordered_sum
->bytenr
&&
1082 disk_bytenr
< ordered_sum
->bytenr
+ ordered_sum
->len
) {
1083 i
= (disk_bytenr
- ordered_sum
->bytenr
) >>
1084 inode
->i_sb
->s_blocksize_bits
;
1085 num_sectors
= ordered_sum
->len
>>
1086 inode
->i_sb
->s_blocksize_bits
;
1087 num_sectors
= min_t(int, len
- index
, num_sectors
- i
);
1088 memcpy(sum
+ index
, ordered_sum
->sums
+ i
,
1091 index
+= (int)num_sectors
;
1094 disk_bytenr
+= num_sectors
* sectorsize
;
1098 spin_unlock_irq(&tree
->lock
);
1099 btrfs_put_ordered_extent(ordered
);
1103 int __init
ordered_data_init(void)
1105 btrfs_ordered_extent_cache
= kmem_cache_create("btrfs_ordered_extent",
1106 sizeof(struct btrfs_ordered_extent
), 0,
1107 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
1109 if (!btrfs_ordered_extent_cache
)
1115 void ordered_data_exit(void)
1117 if (btrfs_ordered_extent_cache
)
1118 kmem_cache_destroy(btrfs_ordered_extent_cache
);