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 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
) &&
202 !(type
== BTRFS_ORDERED_NOCOW
))
203 entry
->csum_bytes_left
= disk_len
;
204 entry
->disk_len
= disk_len
;
205 entry
->bytes_left
= len
;
206 entry
->inode
= igrab(inode
);
207 entry
->compress_type
= compress_type
;
208 entry
->truncated_len
= (u64
)-1;
209 if (type
!= BTRFS_ORDERED_IO_DONE
&& type
!= BTRFS_ORDERED_COMPLETE
)
210 set_bit(type
, &entry
->flags
);
213 set_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
);
215 /* one ref for the tree */
216 atomic_set(&entry
->refs
, 1);
217 init_waitqueue_head(&entry
->wait
);
218 INIT_LIST_HEAD(&entry
->list
);
219 INIT_LIST_HEAD(&entry
->root_extent_list
);
220 INIT_LIST_HEAD(&entry
->work_list
);
221 init_completion(&entry
->completion
);
222 INIT_LIST_HEAD(&entry
->log_list
);
223 INIT_LIST_HEAD(&entry
->trans_list
);
225 trace_btrfs_ordered_extent_add(inode
, entry
);
227 spin_lock_irq(&tree
->lock
);
228 node
= tree_insert(&tree
->tree
, file_offset
,
231 ordered_data_tree_panic(inode
, -EEXIST
, file_offset
);
232 spin_unlock_irq(&tree
->lock
);
234 spin_lock(&root
->ordered_extent_lock
);
235 list_add_tail(&entry
->root_extent_list
,
236 &root
->ordered_extents
);
237 root
->nr_ordered_extents
++;
238 if (root
->nr_ordered_extents
== 1) {
239 spin_lock(&root
->fs_info
->ordered_root_lock
);
240 BUG_ON(!list_empty(&root
->ordered_root
));
241 list_add_tail(&root
->ordered_root
,
242 &root
->fs_info
->ordered_roots
);
243 spin_unlock(&root
->fs_info
->ordered_root_lock
);
245 spin_unlock(&root
->ordered_extent_lock
);
250 int btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
251 u64 start
, u64 len
, u64 disk_len
, int type
)
253 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
255 BTRFS_COMPRESS_NONE
);
258 int btrfs_add_ordered_extent_dio(struct inode
*inode
, u64 file_offset
,
259 u64 start
, u64 len
, u64 disk_len
, int type
)
261 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
263 BTRFS_COMPRESS_NONE
);
266 int btrfs_add_ordered_extent_compress(struct inode
*inode
, u64 file_offset
,
267 u64 start
, u64 len
, u64 disk_len
,
268 int type
, int compress_type
)
270 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
276 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
277 * when an ordered extent is finished. If the list covers more than one
278 * ordered extent, it is split across multiples.
280 void btrfs_add_ordered_sum(struct inode
*inode
,
281 struct btrfs_ordered_extent
*entry
,
282 struct btrfs_ordered_sum
*sum
)
284 struct btrfs_ordered_inode_tree
*tree
;
286 tree
= &BTRFS_I(inode
)->ordered_tree
;
287 spin_lock_irq(&tree
->lock
);
288 list_add_tail(&sum
->list
, &entry
->list
);
289 WARN_ON(entry
->csum_bytes_left
< sum
->len
);
290 entry
->csum_bytes_left
-= sum
->len
;
291 if (entry
->csum_bytes_left
== 0)
292 wake_up(&entry
->wait
);
293 spin_unlock_irq(&tree
->lock
);
297 * this is used to account for finished IO across a given range
298 * of the file. The IO may span ordered extents. If
299 * a given ordered_extent is completely done, 1 is returned, otherwise
302 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
303 * to make sure this function only returns 1 once for a given ordered extent.
305 * file_offset is updated to one byte past the range that is recorded as
306 * complete. This allows you to walk forward in the file.
308 int btrfs_dec_test_first_ordered_pending(struct inode
*inode
,
309 struct btrfs_ordered_extent
**cached
,
310 u64
*file_offset
, u64 io_size
, int uptodate
)
312 struct btrfs_ordered_inode_tree
*tree
;
313 struct rb_node
*node
;
314 struct btrfs_ordered_extent
*entry
= NULL
;
321 tree
= &BTRFS_I(inode
)->ordered_tree
;
322 spin_lock_irqsave(&tree
->lock
, flags
);
323 node
= tree_search(tree
, *file_offset
);
329 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
330 if (!offset_in_entry(entry
, *file_offset
)) {
335 dec_start
= max(*file_offset
, entry
->file_offset
);
336 dec_end
= min(*file_offset
+ io_size
, entry
->file_offset
+
338 *file_offset
= dec_end
;
339 if (dec_start
> dec_end
) {
340 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
341 "bad ordering dec_start %llu end %llu", dec_start
, dec_end
);
343 to_dec
= dec_end
- dec_start
;
344 if (to_dec
> entry
->bytes_left
) {
345 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
346 "bad ordered accounting left %llu size %llu",
347 entry
->bytes_left
, to_dec
);
349 entry
->bytes_left
-= to_dec
;
351 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
353 if (entry
->bytes_left
== 0) {
354 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
355 if (waitqueue_active(&entry
->wait
))
356 wake_up(&entry
->wait
);
361 if (!ret
&& cached
&& entry
) {
363 atomic_inc(&entry
->refs
);
365 spin_unlock_irqrestore(&tree
->lock
, flags
);
370 * this is used to account for finished IO across a given range
371 * of the file. The IO should not span ordered extents. If
372 * a given ordered_extent is completely done, 1 is returned, otherwise
375 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
376 * to make sure this function only returns 1 once for a given ordered extent.
378 int btrfs_dec_test_ordered_pending(struct inode
*inode
,
379 struct btrfs_ordered_extent
**cached
,
380 u64 file_offset
, u64 io_size
, int uptodate
)
382 struct btrfs_ordered_inode_tree
*tree
;
383 struct rb_node
*node
;
384 struct btrfs_ordered_extent
*entry
= NULL
;
388 tree
= &BTRFS_I(inode
)->ordered_tree
;
389 spin_lock_irqsave(&tree
->lock
, flags
);
390 if (cached
&& *cached
) {
395 node
= tree_search(tree
, file_offset
);
401 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
403 if (!offset_in_entry(entry
, file_offset
)) {
408 if (io_size
> entry
->bytes_left
) {
409 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
410 "bad ordered accounting left %llu size %llu",
411 entry
->bytes_left
, io_size
);
413 entry
->bytes_left
-= io_size
;
415 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
417 if (entry
->bytes_left
== 0) {
418 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
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 (!list_empty(&ordered
->log_list
))
457 if (test_bit(BTRFS_ORDERED_LOGGED
, &ordered
->flags
))
459 list_add(&ordered
->log_list
, logged_list
);
460 atomic_inc(&ordered
->refs
);
462 spin_unlock_irq(&tree
->lock
);
465 void btrfs_put_logged_extents(struct list_head
*logged_list
)
467 struct btrfs_ordered_extent
*ordered
;
469 while (!list_empty(logged_list
)) {
470 ordered
= list_first_entry(logged_list
,
471 struct btrfs_ordered_extent
,
473 list_del_init(&ordered
->log_list
);
474 btrfs_put_ordered_extent(ordered
);
478 void btrfs_submit_logged_extents(struct list_head
*logged_list
,
479 struct btrfs_root
*log
)
481 int index
= log
->log_transid
% 2;
483 spin_lock_irq(&log
->log_extents_lock
[index
]);
484 list_splice_tail(logged_list
, &log
->logged_list
[index
]);
485 spin_unlock_irq(&log
->log_extents_lock
[index
]);
488 void btrfs_wait_logged_extents(struct btrfs_trans_handle
*trans
,
489 struct btrfs_root
*log
, u64 transid
)
491 struct btrfs_ordered_extent
*ordered
;
492 int index
= transid
% 2;
494 spin_lock_irq(&log
->log_extents_lock
[index
]);
495 while (!list_empty(&log
->logged_list
[index
])) {
496 ordered
= list_first_entry(&log
->logged_list
[index
],
497 struct btrfs_ordered_extent
,
499 list_del_init(&ordered
->log_list
);
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 struct inode
*inode
= ordered
->inode
;
505 u64 start
= ordered
->file_offset
;
506 u64 end
= ordered
->file_offset
+ ordered
->len
- 1;
509 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
511 wait_event(ordered
->wait
, test_bit(BTRFS_ORDERED_IO_DONE
,
514 if (!test_and_set_bit(BTRFS_ORDERED_LOGGED
, &ordered
->flags
))
515 list_add_tail(&ordered
->trans_list
, &trans
->ordered
);
516 spin_lock_irq(&log
->log_extents_lock
[index
]);
518 spin_unlock_irq(&log
->log_extents_lock
[index
]);
521 void btrfs_free_logged_extents(struct btrfs_root
*log
, u64 transid
)
523 struct btrfs_ordered_extent
*ordered
;
524 int index
= transid
% 2;
526 spin_lock_irq(&log
->log_extents_lock
[index
]);
527 while (!list_empty(&log
->logged_list
[index
])) {
528 ordered
= list_first_entry(&log
->logged_list
[index
],
529 struct btrfs_ordered_extent
,
531 list_del_init(&ordered
->log_list
);
532 spin_unlock_irq(&log
->log_extents_lock
[index
]);
533 btrfs_put_ordered_extent(ordered
);
534 spin_lock_irq(&log
->log_extents_lock
[index
]);
536 spin_unlock_irq(&log
->log_extents_lock
[index
]);
540 * used to drop a reference on an ordered extent. This will free
541 * the extent if the last reference is dropped
543 void btrfs_put_ordered_extent(struct btrfs_ordered_extent
*entry
)
545 struct list_head
*cur
;
546 struct btrfs_ordered_sum
*sum
;
548 trace_btrfs_ordered_extent_put(entry
->inode
, entry
);
550 if (atomic_dec_and_test(&entry
->refs
)) {
552 btrfs_add_delayed_iput(entry
->inode
);
553 while (!list_empty(&entry
->list
)) {
554 cur
= entry
->list
.next
;
555 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
556 list_del(&sum
->list
);
559 kmem_cache_free(btrfs_ordered_extent_cache
, entry
);
564 * remove an ordered extent from the tree. No references are dropped
565 * and waiters are woken up.
567 void btrfs_remove_ordered_extent(struct inode
*inode
,
568 struct btrfs_ordered_extent
*entry
)
570 struct btrfs_ordered_inode_tree
*tree
;
571 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
572 struct rb_node
*node
;
574 tree
= &BTRFS_I(inode
)->ordered_tree
;
575 spin_lock_irq(&tree
->lock
);
576 node
= &entry
->rb_node
;
577 rb_erase(node
, &tree
->tree
);
578 if (tree
->last
== node
)
580 set_bit(BTRFS_ORDERED_COMPLETE
, &entry
->flags
);
581 spin_unlock_irq(&tree
->lock
);
583 spin_lock(&root
->ordered_extent_lock
);
584 list_del_init(&entry
->root_extent_list
);
585 root
->nr_ordered_extents
--;
587 trace_btrfs_ordered_extent_remove(inode
, entry
);
589 if (!root
->nr_ordered_extents
) {
590 spin_lock(&root
->fs_info
->ordered_root_lock
);
591 BUG_ON(list_empty(&root
->ordered_root
));
592 list_del_init(&root
->ordered_root
);
593 spin_unlock(&root
->fs_info
->ordered_root_lock
);
595 spin_unlock(&root
->ordered_extent_lock
);
596 wake_up(&entry
->wait
);
599 static void btrfs_run_ordered_extent_work(struct btrfs_work
*work
)
601 struct btrfs_ordered_extent
*ordered
;
603 ordered
= container_of(work
, struct btrfs_ordered_extent
, flush_work
);
604 btrfs_start_ordered_extent(ordered
->inode
, ordered
, 1);
605 complete(&ordered
->completion
);
609 * wait for all the ordered extents in a root. This is done when balancing
610 * space between drives.
612 int btrfs_wait_ordered_extents(struct btrfs_root
*root
, int nr
)
614 struct list_head splice
, works
;
615 struct btrfs_ordered_extent
*ordered
, *next
;
618 INIT_LIST_HEAD(&splice
);
619 INIT_LIST_HEAD(&works
);
621 mutex_lock(&root
->ordered_extent_mutex
);
622 spin_lock(&root
->ordered_extent_lock
);
623 list_splice_init(&root
->ordered_extents
, &splice
);
624 while (!list_empty(&splice
) && nr
) {
625 ordered
= list_first_entry(&splice
, struct btrfs_ordered_extent
,
627 list_move_tail(&ordered
->root_extent_list
,
628 &root
->ordered_extents
);
629 atomic_inc(&ordered
->refs
);
630 spin_unlock(&root
->ordered_extent_lock
);
632 btrfs_init_work(&ordered
->flush_work
,
633 btrfs_flush_delalloc_helper
,
634 btrfs_run_ordered_extent_work
, NULL
, NULL
);
635 list_add_tail(&ordered
->work_list
, &works
);
636 btrfs_queue_work(root
->fs_info
->flush_workers
,
637 &ordered
->flush_work
);
640 spin_lock(&root
->ordered_extent_lock
);
645 list_splice_tail(&splice
, &root
->ordered_extents
);
646 spin_unlock(&root
->ordered_extent_lock
);
648 list_for_each_entry_safe(ordered
, next
, &works
, work_list
) {
649 list_del_init(&ordered
->work_list
);
650 wait_for_completion(&ordered
->completion
);
651 btrfs_put_ordered_extent(ordered
);
654 mutex_unlock(&root
->ordered_extent_mutex
);
659 void btrfs_wait_ordered_roots(struct btrfs_fs_info
*fs_info
, int nr
)
661 struct btrfs_root
*root
;
662 struct list_head splice
;
665 INIT_LIST_HEAD(&splice
);
667 mutex_lock(&fs_info
->ordered_operations_mutex
);
668 spin_lock(&fs_info
->ordered_root_lock
);
669 list_splice_init(&fs_info
->ordered_roots
, &splice
);
670 while (!list_empty(&splice
) && nr
) {
671 root
= list_first_entry(&splice
, struct btrfs_root
,
673 root
= btrfs_grab_fs_root(root
);
675 list_move_tail(&root
->ordered_root
,
676 &fs_info
->ordered_roots
);
677 spin_unlock(&fs_info
->ordered_root_lock
);
679 done
= btrfs_wait_ordered_extents(root
, nr
);
680 btrfs_put_fs_root(root
);
682 spin_lock(&fs_info
->ordered_root_lock
);
688 list_splice_tail(&splice
, &fs_info
->ordered_roots
);
689 spin_unlock(&fs_info
->ordered_root_lock
);
690 mutex_unlock(&fs_info
->ordered_operations_mutex
);
694 * Used to start IO or wait for a given ordered extent to finish.
696 * If wait is one, this effectively waits on page writeback for all the pages
697 * in the extent, and it waits on the io completion code to insert
698 * metadata into the btree corresponding to the extent
700 void btrfs_start_ordered_extent(struct inode
*inode
,
701 struct btrfs_ordered_extent
*entry
,
704 u64 start
= entry
->file_offset
;
705 u64 end
= start
+ entry
->len
- 1;
707 trace_btrfs_ordered_extent_start(inode
, entry
);
710 * pages in the range can be dirty, clean or writeback. We
711 * start IO on any dirty ones so the wait doesn't stall waiting
712 * for the flusher thread to find them
714 if (!test_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
))
715 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
717 wait_event(entry
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
723 * Used to wait on ordered extents across a large range of bytes.
725 int btrfs_wait_ordered_range(struct inode
*inode
, u64 start
, u64 len
)
730 struct btrfs_ordered_extent
*ordered
;
732 if (start
+ len
< start
) {
733 orig_end
= INT_LIMIT(loff_t
);
735 orig_end
= start
+ len
- 1;
736 if (orig_end
> INT_LIMIT(loff_t
))
737 orig_end
= INT_LIMIT(loff_t
);
740 /* start IO across the range first to instantiate any delalloc
743 ret
= btrfs_fdatawrite_range(inode
, start
, orig_end
);
747 ret
= filemap_fdatawait_range(inode
->i_mapping
, start
, orig_end
);
753 ordered
= btrfs_lookup_first_ordered_extent(inode
, end
);
756 if (ordered
->file_offset
> orig_end
) {
757 btrfs_put_ordered_extent(ordered
);
760 if (ordered
->file_offset
+ ordered
->len
<= start
) {
761 btrfs_put_ordered_extent(ordered
);
764 btrfs_start_ordered_extent(inode
, ordered
, 1);
765 end
= ordered
->file_offset
;
766 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
))
768 btrfs_put_ordered_extent(ordered
);
769 if (ret
|| end
== 0 || end
== start
)
777 * find an ordered extent corresponding to file_offset. return NULL if
778 * nothing is found, otherwise take a reference on the extent and return it
780 struct btrfs_ordered_extent
*btrfs_lookup_ordered_extent(struct inode
*inode
,
783 struct btrfs_ordered_inode_tree
*tree
;
784 struct rb_node
*node
;
785 struct btrfs_ordered_extent
*entry
= NULL
;
787 tree
= &BTRFS_I(inode
)->ordered_tree
;
788 spin_lock_irq(&tree
->lock
);
789 node
= tree_search(tree
, file_offset
);
793 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
794 if (!offset_in_entry(entry
, file_offset
))
797 atomic_inc(&entry
->refs
);
799 spin_unlock_irq(&tree
->lock
);
803 /* Since the DIO code tries to lock a wide area we need to look for any ordered
804 * extents that exist in the range, rather than just the start of the range.
806 struct btrfs_ordered_extent
*btrfs_lookup_ordered_range(struct inode
*inode
,
810 struct btrfs_ordered_inode_tree
*tree
;
811 struct rb_node
*node
;
812 struct btrfs_ordered_extent
*entry
= NULL
;
814 tree
= &BTRFS_I(inode
)->ordered_tree
;
815 spin_lock_irq(&tree
->lock
);
816 node
= tree_search(tree
, file_offset
);
818 node
= tree_search(tree
, file_offset
+ len
);
824 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
825 if (range_overlaps(entry
, file_offset
, len
))
828 if (entry
->file_offset
>= file_offset
+ len
) {
833 node
= rb_next(node
);
839 atomic_inc(&entry
->refs
);
840 spin_unlock_irq(&tree
->lock
);
845 * lookup and return any extent before 'file_offset'. NULL is returned
848 struct btrfs_ordered_extent
*
849 btrfs_lookup_first_ordered_extent(struct inode
*inode
, u64 file_offset
)
851 struct btrfs_ordered_inode_tree
*tree
;
852 struct rb_node
*node
;
853 struct btrfs_ordered_extent
*entry
= NULL
;
855 tree
= &BTRFS_I(inode
)->ordered_tree
;
856 spin_lock_irq(&tree
->lock
);
857 node
= tree_search(tree
, file_offset
);
861 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
862 atomic_inc(&entry
->refs
);
864 spin_unlock_irq(&tree
->lock
);
869 * After an extent is done, call this to conditionally update the on disk
870 * i_size. i_size is updated to cover any fully written part of the file.
872 int btrfs_ordered_update_i_size(struct inode
*inode
, u64 offset
,
873 struct btrfs_ordered_extent
*ordered
)
875 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
878 u64 i_size
= i_size_read(inode
);
879 struct rb_node
*node
;
880 struct rb_node
*prev
= NULL
;
881 struct btrfs_ordered_extent
*test
;
884 spin_lock_irq(&tree
->lock
);
886 offset
= entry_end(ordered
);
887 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
))
889 ordered
->file_offset
+
890 ordered
->truncated_len
);
892 offset
= ALIGN(offset
, BTRFS_I(inode
)->root
->sectorsize
);
894 disk_i_size
= BTRFS_I(inode
)->disk_i_size
;
897 if (disk_i_size
> i_size
) {
898 BTRFS_I(inode
)->disk_i_size
= i_size
;
904 * if the disk i_size is already at the inode->i_size, or
905 * this ordered extent is inside the disk i_size, we're done
907 if (disk_i_size
== i_size
)
911 * We still need to update disk_i_size if outstanding_isize is greater
914 if (offset
<= disk_i_size
&&
915 (!ordered
|| ordered
->outstanding_isize
<= disk_i_size
))
919 * walk backward from this ordered extent to disk_i_size.
920 * if we find an ordered extent then we can't update disk i_size
924 node
= rb_prev(&ordered
->rb_node
);
926 prev
= tree_search(tree
, offset
);
928 * we insert file extents without involving ordered struct,
929 * so there should be no ordered struct cover this offset
932 test
= rb_entry(prev
, struct btrfs_ordered_extent
,
934 BUG_ON(offset_in_entry(test
, offset
));
938 for (; node
; node
= rb_prev(node
)) {
939 test
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
941 /* We treat this entry as if it doesnt exist */
942 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &test
->flags
))
944 if (test
->file_offset
+ test
->len
<= disk_i_size
)
946 if (test
->file_offset
>= i_size
)
948 if (entry_end(test
) > disk_i_size
) {
950 * we don't update disk_i_size now, so record this
951 * undealt i_size. Or we will not know the real
954 if (test
->outstanding_isize
< offset
)
955 test
->outstanding_isize
= offset
;
957 ordered
->outstanding_isize
>
958 test
->outstanding_isize
)
959 test
->outstanding_isize
=
960 ordered
->outstanding_isize
;
964 new_i_size
= min_t(u64
, offset
, i_size
);
967 * Some ordered extents may completed before the current one, and
968 * we hold the real i_size in ->outstanding_isize.
970 if (ordered
&& ordered
->outstanding_isize
> new_i_size
)
971 new_i_size
= min_t(u64
, ordered
->outstanding_isize
, i_size
);
972 BTRFS_I(inode
)->disk_i_size
= new_i_size
;
976 * We need to do this because we can't remove ordered extents until
977 * after the i_disk_size has been updated and then the inode has been
978 * updated to reflect the change, so we need to tell anybody who finds
979 * this ordered extent that we've already done all the real work, we
980 * just haven't completed all the other work.
983 set_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &ordered
->flags
);
984 spin_unlock_irq(&tree
->lock
);
989 * search the ordered extents for one corresponding to 'offset' and
990 * try to find a checksum. This is used because we allow pages to
991 * be reclaimed before their checksum is actually put into the btree
993 int btrfs_find_ordered_sum(struct inode
*inode
, u64 offset
, u64 disk_bytenr
,
996 struct btrfs_ordered_sum
*ordered_sum
;
997 struct btrfs_ordered_extent
*ordered
;
998 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
999 unsigned long num_sectors
;
1001 u32 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
1004 ordered
= btrfs_lookup_ordered_extent(inode
, offset
);
1008 spin_lock_irq(&tree
->lock
);
1009 list_for_each_entry_reverse(ordered_sum
, &ordered
->list
, list
) {
1010 if (disk_bytenr
>= ordered_sum
->bytenr
&&
1011 disk_bytenr
< ordered_sum
->bytenr
+ ordered_sum
->len
) {
1012 i
= (disk_bytenr
- ordered_sum
->bytenr
) >>
1013 inode
->i_sb
->s_blocksize_bits
;
1014 num_sectors
= ordered_sum
->len
>>
1015 inode
->i_sb
->s_blocksize_bits
;
1016 num_sectors
= min_t(int, len
- index
, num_sectors
- i
);
1017 memcpy(sum
+ index
, ordered_sum
->sums
+ i
,
1020 index
+= (int)num_sectors
;
1023 disk_bytenr
+= num_sectors
* sectorsize
;
1027 spin_unlock_irq(&tree
->lock
);
1028 btrfs_put_ordered_extent(ordered
);
1032 int __init
ordered_data_init(void)
1034 btrfs_ordered_extent_cache
= kmem_cache_create("btrfs_ordered_extent",
1035 sizeof(struct btrfs_ordered_extent
), 0,
1036 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
1038 if (!btrfs_ordered_extent_cache
)
1044 void ordered_data_exit(void)
1046 if (btrfs_ordered_extent_cache
)
1047 kmem_cache_destroy(btrfs_ordered_extent_cache
);