1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/slab.h>
7 #include <linux/blkdev.h>
8 #include <linux/writeback.h>
9 #include <linux/sched/mm.h>
12 #include "transaction.h"
13 #include "btrfs_inode.h"
14 #include "extent_io.h"
16 #include "compression.h"
17 #include "delalloc-space.h"
19 static struct kmem_cache
*btrfs_ordered_extent_cache
;
21 static u64
entry_end(struct btrfs_ordered_extent
*entry
)
23 if (entry
->file_offset
+ entry
->num_bytes
< entry
->file_offset
)
25 return entry
->file_offset
+ entry
->num_bytes
;
28 /* returns NULL if the insertion worked, or it returns the node it did find
31 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 file_offset
,
34 struct rb_node
**p
= &root
->rb_node
;
35 struct rb_node
*parent
= NULL
;
36 struct btrfs_ordered_extent
*entry
;
40 entry
= rb_entry(parent
, struct btrfs_ordered_extent
, rb_node
);
42 if (file_offset
< entry
->file_offset
)
44 else if (file_offset
>= entry_end(entry
))
50 rb_link_node(node
, parent
, p
);
51 rb_insert_color(node
, root
);
56 * look for a given offset in the tree, and if it can't be found return the
59 static struct rb_node
*__tree_search(struct rb_root
*root
, u64 file_offset
,
60 struct rb_node
**prev_ret
)
62 struct rb_node
*n
= root
->rb_node
;
63 struct rb_node
*prev
= NULL
;
65 struct btrfs_ordered_extent
*entry
;
66 struct btrfs_ordered_extent
*prev_entry
= NULL
;
69 entry
= rb_entry(n
, struct btrfs_ordered_extent
, rb_node
);
73 if (file_offset
< entry
->file_offset
)
75 else if (file_offset
>= entry_end(entry
))
83 while (prev
&& file_offset
>= entry_end(prev_entry
)) {
87 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
89 if (file_offset
< entry_end(prev_entry
))
95 prev_entry
= rb_entry(prev
, struct btrfs_ordered_extent
,
97 while (prev
&& file_offset
< entry_end(prev_entry
)) {
101 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
110 * helper to check if a given offset is inside a given entry
112 static int offset_in_entry(struct btrfs_ordered_extent
*entry
, u64 file_offset
)
114 if (file_offset
< entry
->file_offset
||
115 entry
->file_offset
+ entry
->num_bytes
<= file_offset
)
120 static int range_overlaps(struct btrfs_ordered_extent
*entry
, u64 file_offset
,
123 if (file_offset
+ len
<= entry
->file_offset
||
124 entry
->file_offset
+ entry
->num_bytes
<= file_offset
)
130 * look find the first ordered struct that has this offset, otherwise
131 * the first one less than this offset
133 static inline struct rb_node
*tree_search(struct btrfs_ordered_inode_tree
*tree
,
136 struct rb_root
*root
= &tree
->tree
;
137 struct rb_node
*prev
= NULL
;
139 struct btrfs_ordered_extent
*entry
;
142 entry
= rb_entry(tree
->last
, struct btrfs_ordered_extent
,
144 if (offset_in_entry(entry
, file_offset
))
147 ret
= __tree_search(root
, file_offset
, &prev
);
155 /* allocate and add a new ordered_extent into the per-inode tree.
157 * The tree is given a single reference on the ordered extent that was
160 static int __btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
161 u64 disk_bytenr
, u64 num_bytes
,
162 u64 disk_num_bytes
, int type
, int dio
,
165 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
166 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
167 struct btrfs_ordered_inode_tree
*tree
;
168 struct rb_node
*node
;
169 struct btrfs_ordered_extent
*entry
;
171 tree
= &BTRFS_I(inode
)->ordered_tree
;
172 entry
= kmem_cache_zalloc(btrfs_ordered_extent_cache
, GFP_NOFS
);
176 entry
->file_offset
= file_offset
;
177 entry
->disk_bytenr
= disk_bytenr
;
178 entry
->num_bytes
= num_bytes
;
179 entry
->disk_num_bytes
= disk_num_bytes
;
180 entry
->bytes_left
= num_bytes
;
181 entry
->inode
= igrab(inode
);
182 entry
->compress_type
= compress_type
;
183 entry
->truncated_len
= (u64
)-1;
184 if (type
!= BTRFS_ORDERED_IO_DONE
&& type
!= BTRFS_ORDERED_COMPLETE
)
185 set_bit(type
, &entry
->flags
);
188 percpu_counter_add_batch(&fs_info
->dio_bytes
, num_bytes
,
189 fs_info
->delalloc_batch
);
190 set_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
);
193 /* one ref for the tree */
194 refcount_set(&entry
->refs
, 1);
195 init_waitqueue_head(&entry
->wait
);
196 INIT_LIST_HEAD(&entry
->list
);
197 INIT_LIST_HEAD(&entry
->root_extent_list
);
198 INIT_LIST_HEAD(&entry
->work_list
);
199 init_completion(&entry
->completion
);
200 INIT_LIST_HEAD(&entry
->log_list
);
201 INIT_LIST_HEAD(&entry
->trans_list
);
203 trace_btrfs_ordered_extent_add(inode
, entry
);
205 spin_lock_irq(&tree
->lock
);
206 node
= tree_insert(&tree
->tree
, file_offset
,
209 btrfs_panic(fs_info
, -EEXIST
,
210 "inconsistency in ordered tree at offset %llu",
212 spin_unlock_irq(&tree
->lock
);
214 spin_lock(&root
->ordered_extent_lock
);
215 list_add_tail(&entry
->root_extent_list
,
216 &root
->ordered_extents
);
217 root
->nr_ordered_extents
++;
218 if (root
->nr_ordered_extents
== 1) {
219 spin_lock(&fs_info
->ordered_root_lock
);
220 BUG_ON(!list_empty(&root
->ordered_root
));
221 list_add_tail(&root
->ordered_root
, &fs_info
->ordered_roots
);
222 spin_unlock(&fs_info
->ordered_root_lock
);
224 spin_unlock(&root
->ordered_extent_lock
);
227 * We don't need the count_max_extents here, we can assume that all of
228 * that work has been done at higher layers, so this is truly the
229 * smallest the extent is going to get.
231 spin_lock(&BTRFS_I(inode
)->lock
);
232 btrfs_mod_outstanding_extents(BTRFS_I(inode
), 1);
233 spin_unlock(&BTRFS_I(inode
)->lock
);
238 int btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
239 u64 disk_bytenr
, u64 num_bytes
, u64 disk_num_bytes
,
242 return __btrfs_add_ordered_extent(inode
, file_offset
, disk_bytenr
,
243 num_bytes
, disk_num_bytes
, type
, 0,
244 BTRFS_COMPRESS_NONE
);
247 int btrfs_add_ordered_extent_dio(struct inode
*inode
, u64 file_offset
,
248 u64 disk_bytenr
, u64 num_bytes
,
249 u64 disk_num_bytes
, int type
)
251 return __btrfs_add_ordered_extent(inode
, file_offset
, disk_bytenr
,
252 num_bytes
, disk_num_bytes
, type
, 1,
253 BTRFS_COMPRESS_NONE
);
256 int btrfs_add_ordered_extent_compress(struct inode
*inode
, u64 file_offset
,
257 u64 disk_bytenr
, u64 num_bytes
,
258 u64 disk_num_bytes
, int type
,
261 return __btrfs_add_ordered_extent(inode
, file_offset
, disk_bytenr
,
262 num_bytes
, disk_num_bytes
, type
, 0,
267 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
268 * when an ordered extent is finished. If the list covers more than one
269 * ordered extent, it is split across multiples.
271 void btrfs_add_ordered_sum(struct btrfs_ordered_extent
*entry
,
272 struct btrfs_ordered_sum
*sum
)
274 struct btrfs_ordered_inode_tree
*tree
;
276 tree
= &BTRFS_I(entry
->inode
)->ordered_tree
;
277 spin_lock_irq(&tree
->lock
);
278 list_add_tail(&sum
->list
, &entry
->list
);
279 spin_unlock_irq(&tree
->lock
);
283 * this is used to account for finished IO across a given range
284 * of the file. The IO may span ordered extents. If
285 * a given ordered_extent is completely done, 1 is returned, otherwise
288 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
289 * to make sure this function only returns 1 once for a given ordered extent.
291 * file_offset is updated to one byte past the range that is recorded as
292 * complete. This allows you to walk forward in the file.
294 int btrfs_dec_test_first_ordered_pending(struct inode
*inode
,
295 struct btrfs_ordered_extent
**cached
,
296 u64
*file_offset
, u64 io_size
, int uptodate
)
298 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
299 struct btrfs_ordered_inode_tree
*tree
;
300 struct rb_node
*node
;
301 struct btrfs_ordered_extent
*entry
= NULL
;
308 tree
= &BTRFS_I(inode
)->ordered_tree
;
309 spin_lock_irqsave(&tree
->lock
, flags
);
310 node
= tree_search(tree
, *file_offset
);
316 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
317 if (!offset_in_entry(entry
, *file_offset
)) {
322 dec_start
= max(*file_offset
, entry
->file_offset
);
323 dec_end
= min(*file_offset
+ io_size
,
324 entry
->file_offset
+ entry
->num_bytes
);
325 *file_offset
= dec_end
;
326 if (dec_start
> dec_end
) {
327 btrfs_crit(fs_info
, "bad ordering dec_start %llu end %llu",
330 to_dec
= dec_end
- dec_start
;
331 if (to_dec
> entry
->bytes_left
) {
333 "bad ordered accounting left %llu size %llu",
334 entry
->bytes_left
, to_dec
);
336 entry
->bytes_left
-= to_dec
;
338 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
340 if (entry
->bytes_left
== 0) {
341 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
342 /* test_and_set_bit implies a barrier */
343 cond_wake_up_nomb(&entry
->wait
);
348 if (!ret
&& cached
&& entry
) {
350 refcount_inc(&entry
->refs
);
352 spin_unlock_irqrestore(&tree
->lock
, flags
);
357 * this is used to account for finished IO across a given range
358 * of the file. The IO should not span ordered extents. If
359 * a given ordered_extent is completely done, 1 is returned, otherwise
362 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
363 * to make sure this function only returns 1 once for a given ordered extent.
365 int btrfs_dec_test_ordered_pending(struct inode
*inode
,
366 struct btrfs_ordered_extent
**cached
,
367 u64 file_offset
, u64 io_size
, int uptodate
)
369 struct btrfs_ordered_inode_tree
*tree
;
370 struct rb_node
*node
;
371 struct btrfs_ordered_extent
*entry
= NULL
;
375 tree
= &BTRFS_I(inode
)->ordered_tree
;
376 spin_lock_irqsave(&tree
->lock
, flags
);
377 if (cached
&& *cached
) {
382 node
= tree_search(tree
, file_offset
);
388 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
390 if (!offset_in_entry(entry
, file_offset
)) {
395 if (io_size
> entry
->bytes_left
) {
396 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
397 "bad ordered accounting left %llu size %llu",
398 entry
->bytes_left
, io_size
);
400 entry
->bytes_left
-= io_size
;
402 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
404 if (entry
->bytes_left
== 0) {
405 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
406 /* test_and_set_bit implies a barrier */
407 cond_wake_up_nomb(&entry
->wait
);
412 if (!ret
&& cached
&& entry
) {
414 refcount_inc(&entry
->refs
);
416 spin_unlock_irqrestore(&tree
->lock
, flags
);
421 * used to drop a reference on an ordered extent. This will free
422 * the extent if the last reference is dropped
424 void btrfs_put_ordered_extent(struct btrfs_ordered_extent
*entry
)
426 struct list_head
*cur
;
427 struct btrfs_ordered_sum
*sum
;
429 trace_btrfs_ordered_extent_put(entry
->inode
, entry
);
431 if (refcount_dec_and_test(&entry
->refs
)) {
432 ASSERT(list_empty(&entry
->log_list
));
433 ASSERT(list_empty(&entry
->trans_list
));
434 ASSERT(list_empty(&entry
->root_extent_list
));
435 ASSERT(RB_EMPTY_NODE(&entry
->rb_node
));
437 btrfs_add_delayed_iput(entry
->inode
);
438 while (!list_empty(&entry
->list
)) {
439 cur
= entry
->list
.next
;
440 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
441 list_del(&sum
->list
);
444 kmem_cache_free(btrfs_ordered_extent_cache
, entry
);
449 * remove an ordered extent from the tree. No references are dropped
450 * and waiters are woken up.
452 void btrfs_remove_ordered_extent(struct inode
*inode
,
453 struct btrfs_ordered_extent
*entry
)
455 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
456 struct btrfs_ordered_inode_tree
*tree
;
457 struct btrfs_inode
*btrfs_inode
= BTRFS_I(inode
);
458 struct btrfs_root
*root
= btrfs_inode
->root
;
459 struct rb_node
*node
;
461 /* This is paired with btrfs_add_ordered_extent. */
462 spin_lock(&btrfs_inode
->lock
);
463 btrfs_mod_outstanding_extents(btrfs_inode
, -1);
464 spin_unlock(&btrfs_inode
->lock
);
465 if (root
!= fs_info
->tree_root
)
466 btrfs_delalloc_release_metadata(btrfs_inode
, entry
->num_bytes
,
469 if (test_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
))
470 percpu_counter_add_batch(&fs_info
->dio_bytes
, -entry
->num_bytes
,
471 fs_info
->delalloc_batch
);
473 tree
= &btrfs_inode
->ordered_tree
;
474 spin_lock_irq(&tree
->lock
);
475 node
= &entry
->rb_node
;
476 rb_erase(node
, &tree
->tree
);
478 if (tree
->last
== node
)
480 set_bit(BTRFS_ORDERED_COMPLETE
, &entry
->flags
);
481 spin_unlock_irq(&tree
->lock
);
483 spin_lock(&root
->ordered_extent_lock
);
484 list_del_init(&entry
->root_extent_list
);
485 root
->nr_ordered_extents
--;
487 trace_btrfs_ordered_extent_remove(inode
, entry
);
489 if (!root
->nr_ordered_extents
) {
490 spin_lock(&fs_info
->ordered_root_lock
);
491 BUG_ON(list_empty(&root
->ordered_root
));
492 list_del_init(&root
->ordered_root
);
493 spin_unlock(&fs_info
->ordered_root_lock
);
495 spin_unlock(&root
->ordered_extent_lock
);
496 wake_up(&entry
->wait
);
499 static void btrfs_run_ordered_extent_work(struct btrfs_work
*work
)
501 struct btrfs_ordered_extent
*ordered
;
503 ordered
= container_of(work
, struct btrfs_ordered_extent
, flush_work
);
504 btrfs_start_ordered_extent(ordered
->inode
, ordered
, 1);
505 complete(&ordered
->completion
);
509 * wait for all the ordered extents in a root. This is done when balancing
510 * space between drives.
512 u64
btrfs_wait_ordered_extents(struct btrfs_root
*root
, u64 nr
,
513 const u64 range_start
, const u64 range_len
)
515 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
519 struct btrfs_ordered_extent
*ordered
, *next
;
521 const u64 range_end
= range_start
+ range_len
;
523 mutex_lock(&root
->ordered_extent_mutex
);
524 spin_lock(&root
->ordered_extent_lock
);
525 list_splice_init(&root
->ordered_extents
, &splice
);
526 while (!list_empty(&splice
) && nr
) {
527 ordered
= list_first_entry(&splice
, struct btrfs_ordered_extent
,
530 if (range_end
<= ordered
->disk_bytenr
||
531 ordered
->disk_bytenr
+ ordered
->disk_num_bytes
<= range_start
) {
532 list_move_tail(&ordered
->root_extent_list
, &skipped
);
533 cond_resched_lock(&root
->ordered_extent_lock
);
537 list_move_tail(&ordered
->root_extent_list
,
538 &root
->ordered_extents
);
539 refcount_inc(&ordered
->refs
);
540 spin_unlock(&root
->ordered_extent_lock
);
542 btrfs_init_work(&ordered
->flush_work
,
543 btrfs_run_ordered_extent_work
, NULL
, NULL
);
544 list_add_tail(&ordered
->work_list
, &works
);
545 btrfs_queue_work(fs_info
->flush_workers
, &ordered
->flush_work
);
548 spin_lock(&root
->ordered_extent_lock
);
553 list_splice_tail(&skipped
, &root
->ordered_extents
);
554 list_splice_tail(&splice
, &root
->ordered_extents
);
555 spin_unlock(&root
->ordered_extent_lock
);
557 list_for_each_entry_safe(ordered
, next
, &works
, work_list
) {
558 list_del_init(&ordered
->work_list
);
559 wait_for_completion(&ordered
->completion
);
560 btrfs_put_ordered_extent(ordered
);
563 mutex_unlock(&root
->ordered_extent_mutex
);
568 void btrfs_wait_ordered_roots(struct btrfs_fs_info
*fs_info
, u64 nr
,
569 const u64 range_start
, const u64 range_len
)
571 struct btrfs_root
*root
;
572 struct list_head splice
;
575 INIT_LIST_HEAD(&splice
);
577 mutex_lock(&fs_info
->ordered_operations_mutex
);
578 spin_lock(&fs_info
->ordered_root_lock
);
579 list_splice_init(&fs_info
->ordered_roots
, &splice
);
580 while (!list_empty(&splice
) && nr
) {
581 root
= list_first_entry(&splice
, struct btrfs_root
,
583 root
= btrfs_grab_root(root
);
585 list_move_tail(&root
->ordered_root
,
586 &fs_info
->ordered_roots
);
587 spin_unlock(&fs_info
->ordered_root_lock
);
589 done
= btrfs_wait_ordered_extents(root
, nr
,
590 range_start
, range_len
);
591 btrfs_put_root(root
);
593 spin_lock(&fs_info
->ordered_root_lock
);
598 list_splice_tail(&splice
, &fs_info
->ordered_roots
);
599 spin_unlock(&fs_info
->ordered_root_lock
);
600 mutex_unlock(&fs_info
->ordered_operations_mutex
);
604 * Used to start IO or wait for a given ordered extent to finish.
606 * If wait is one, this effectively waits on page writeback for all the pages
607 * in the extent, and it waits on the io completion code to insert
608 * metadata into the btree corresponding to the extent
610 void btrfs_start_ordered_extent(struct inode
*inode
,
611 struct btrfs_ordered_extent
*entry
,
614 u64 start
= entry
->file_offset
;
615 u64 end
= start
+ entry
->num_bytes
- 1;
617 trace_btrfs_ordered_extent_start(inode
, entry
);
620 * pages in the range can be dirty, clean or writeback. We
621 * start IO on any dirty ones so the wait doesn't stall waiting
622 * for the flusher thread to find them
624 if (!test_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
))
625 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
627 wait_event(entry
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
633 * Used to wait on ordered extents across a large range of bytes.
635 int btrfs_wait_ordered_range(struct inode
*inode
, u64 start
, u64 len
)
641 struct btrfs_ordered_extent
*ordered
;
643 if (start
+ len
< start
) {
644 orig_end
= INT_LIMIT(loff_t
);
646 orig_end
= start
+ len
- 1;
647 if (orig_end
> INT_LIMIT(loff_t
))
648 orig_end
= INT_LIMIT(loff_t
);
651 /* start IO across the range first to instantiate any delalloc
654 ret
= btrfs_fdatawrite_range(inode
, start
, orig_end
);
659 * If we have a writeback error don't return immediately. Wait first
660 * for any ordered extents that haven't completed yet. This is to make
661 * sure no one can dirty the same page ranges and call writepages()
662 * before the ordered extents complete - to avoid failures (-EEXIST)
663 * when adding the new ordered extents to the ordered tree.
665 ret_wb
= filemap_fdatawait_range(inode
->i_mapping
, start
, orig_end
);
669 ordered
= btrfs_lookup_first_ordered_extent(inode
, end
);
672 if (ordered
->file_offset
> orig_end
) {
673 btrfs_put_ordered_extent(ordered
);
676 if (ordered
->file_offset
+ ordered
->num_bytes
<= start
) {
677 btrfs_put_ordered_extent(ordered
);
680 btrfs_start_ordered_extent(inode
, ordered
, 1);
681 end
= ordered
->file_offset
;
683 * If the ordered extent had an error save the error but don't
684 * exit without waiting first for all other ordered extents in
685 * the range to complete.
687 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
))
689 btrfs_put_ordered_extent(ordered
);
690 if (end
== 0 || end
== start
)
694 return ret_wb
? ret_wb
: ret
;
698 * find an ordered extent corresponding to file_offset. return NULL if
699 * nothing is found, otherwise take a reference on the extent and return it
701 struct btrfs_ordered_extent
*btrfs_lookup_ordered_extent(struct inode
*inode
,
704 struct btrfs_ordered_inode_tree
*tree
;
705 struct rb_node
*node
;
706 struct btrfs_ordered_extent
*entry
= NULL
;
708 tree
= &BTRFS_I(inode
)->ordered_tree
;
709 spin_lock_irq(&tree
->lock
);
710 node
= tree_search(tree
, file_offset
);
714 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
715 if (!offset_in_entry(entry
, file_offset
))
718 refcount_inc(&entry
->refs
);
720 spin_unlock_irq(&tree
->lock
);
724 /* Since the DIO code tries to lock a wide area we need to look for any ordered
725 * extents that exist in the range, rather than just the start of the range.
727 struct btrfs_ordered_extent
*btrfs_lookup_ordered_range(
728 struct btrfs_inode
*inode
, u64 file_offset
, u64 len
)
730 struct btrfs_ordered_inode_tree
*tree
;
731 struct rb_node
*node
;
732 struct btrfs_ordered_extent
*entry
= NULL
;
734 tree
= &inode
->ordered_tree
;
735 spin_lock_irq(&tree
->lock
);
736 node
= tree_search(tree
, file_offset
);
738 node
= tree_search(tree
, file_offset
+ len
);
744 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
745 if (range_overlaps(entry
, file_offset
, len
))
748 if (entry
->file_offset
>= file_offset
+ len
) {
753 node
= rb_next(node
);
759 refcount_inc(&entry
->refs
);
760 spin_unlock_irq(&tree
->lock
);
765 * lookup and return any extent before 'file_offset'. NULL is returned
768 struct btrfs_ordered_extent
*
769 btrfs_lookup_first_ordered_extent(struct inode
*inode
, u64 file_offset
)
771 struct btrfs_ordered_inode_tree
*tree
;
772 struct rb_node
*node
;
773 struct btrfs_ordered_extent
*entry
= NULL
;
775 tree
= &BTRFS_I(inode
)->ordered_tree
;
776 spin_lock_irq(&tree
->lock
);
777 node
= tree_search(tree
, file_offset
);
781 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
782 refcount_inc(&entry
->refs
);
784 spin_unlock_irq(&tree
->lock
);
789 * search the ordered extents for one corresponding to 'offset' and
790 * try to find a checksum. This is used because we allow pages to
791 * be reclaimed before their checksum is actually put into the btree
793 int btrfs_find_ordered_sum(struct inode
*inode
, u64 offset
, u64 disk_bytenr
,
796 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
797 struct btrfs_ordered_sum
*ordered_sum
;
798 struct btrfs_ordered_extent
*ordered
;
799 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
800 unsigned long num_sectors
;
802 u32 sectorsize
= btrfs_inode_sectorsize(inode
);
803 const u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
806 ordered
= btrfs_lookup_ordered_extent(inode
, offset
);
810 spin_lock_irq(&tree
->lock
);
811 list_for_each_entry_reverse(ordered_sum
, &ordered
->list
, list
) {
812 if (disk_bytenr
>= ordered_sum
->bytenr
&&
813 disk_bytenr
< ordered_sum
->bytenr
+ ordered_sum
->len
) {
814 i
= (disk_bytenr
- ordered_sum
->bytenr
) >>
815 inode
->i_sb
->s_blocksize_bits
;
816 num_sectors
= ordered_sum
->len
>>
817 inode
->i_sb
->s_blocksize_bits
;
818 num_sectors
= min_t(int, len
- index
, num_sectors
- i
);
819 memcpy(sum
+ index
, ordered_sum
->sums
+ i
* csum_size
,
820 num_sectors
* csum_size
);
822 index
+= (int)num_sectors
* csum_size
;
825 disk_bytenr
+= num_sectors
* sectorsize
;
829 spin_unlock_irq(&tree
->lock
);
830 btrfs_put_ordered_extent(ordered
);
835 * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
836 * ordered extents in it are run to completion.
838 * @inode: Inode whose ordered tree is to be searched
839 * @start: Beginning of range to flush
840 * @end: Last byte of range to lock
841 * @cached_state: If passed, will return the extent state responsible for the
842 * locked range. It's the caller's responsibility to free the cached state.
844 * This function always returns with the given range locked, ensuring after it's
845 * called no order extent can be pending.
847 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode
*inode
, u64 start
,
849 struct extent_state
**cached_state
)
851 struct btrfs_ordered_extent
*ordered
;
852 struct extent_state
*cache
= NULL
;
853 struct extent_state
**cachedp
= &cache
;
856 cachedp
= cached_state
;
859 lock_extent_bits(&inode
->io_tree
, start
, end
, cachedp
);
860 ordered
= btrfs_lookup_ordered_range(inode
, start
,
864 * If no external cached_state has been passed then
865 * decrement the extra ref taken for cachedp since we
866 * aren't exposing it outside of this function
869 refcount_dec(&cache
->refs
);
872 unlock_extent_cached(&inode
->io_tree
, start
, end
, cachedp
);
873 btrfs_start_ordered_extent(&inode
->vfs_inode
, ordered
, 1);
874 btrfs_put_ordered_extent(ordered
);
878 int __init
ordered_data_init(void)
880 btrfs_ordered_extent_cache
= kmem_cache_create("btrfs_ordered_extent",
881 sizeof(struct btrfs_ordered_extent
), 0,
884 if (!btrfs_ordered_extent_cache
)
890 void __cold
ordered_data_exit(void)
892 kmem_cache_destroy(btrfs_ordered_extent_cache
);