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>
10 #include "transaction.h"
11 #include "btrfs_inode.h"
12 #include "extent_io.h"
14 #include "compression.h"
16 static struct kmem_cache
*btrfs_ordered_extent_cache
;
18 static u64
entry_end(struct btrfs_ordered_extent
*entry
)
20 if (entry
->file_offset
+ entry
->len
< entry
->file_offset
)
22 return entry
->file_offset
+ entry
->len
;
25 /* returns NULL if the insertion worked, or it returns the node it did find
28 static struct rb_node
*tree_insert(struct rb_root
*root
, u64 file_offset
,
31 struct rb_node
**p
= &root
->rb_node
;
32 struct rb_node
*parent
= NULL
;
33 struct btrfs_ordered_extent
*entry
;
37 entry
= rb_entry(parent
, struct btrfs_ordered_extent
, rb_node
);
39 if (file_offset
< entry
->file_offset
)
41 else if (file_offset
>= entry_end(entry
))
47 rb_link_node(node
, parent
, p
);
48 rb_insert_color(node
, root
);
52 static void ordered_data_tree_panic(struct inode
*inode
, int errno
,
55 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
56 btrfs_panic(fs_info
, errno
,
57 "Inconsistency in ordered tree at offset %llu", offset
);
61 * look for a given offset in the tree, and if it can't be found return the
64 static struct rb_node
*__tree_search(struct rb_root
*root
, u64 file_offset
,
65 struct rb_node
**prev_ret
)
67 struct rb_node
*n
= root
->rb_node
;
68 struct rb_node
*prev
= NULL
;
70 struct btrfs_ordered_extent
*entry
;
71 struct btrfs_ordered_extent
*prev_entry
= NULL
;
74 entry
= rb_entry(n
, struct btrfs_ordered_extent
, rb_node
);
78 if (file_offset
< entry
->file_offset
)
80 else if (file_offset
>= entry_end(entry
))
88 while (prev
&& file_offset
>= entry_end(prev_entry
)) {
92 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
94 if (file_offset
< entry_end(prev_entry
))
100 prev_entry
= rb_entry(prev
, struct btrfs_ordered_extent
,
102 while (prev
&& file_offset
< entry_end(prev_entry
)) {
103 test
= rb_prev(prev
);
106 prev_entry
= rb_entry(test
, struct btrfs_ordered_extent
,
115 * helper to check if a given offset is inside a given entry
117 static int offset_in_entry(struct btrfs_ordered_extent
*entry
, u64 file_offset
)
119 if (file_offset
< entry
->file_offset
||
120 entry
->file_offset
+ entry
->len
<= file_offset
)
125 static int range_overlaps(struct btrfs_ordered_extent
*entry
, u64 file_offset
,
128 if (file_offset
+ len
<= entry
->file_offset
||
129 entry
->file_offset
+ entry
->len
<= file_offset
)
135 * look find the first ordered struct that has this offset, otherwise
136 * the first one less than this offset
138 static inline struct rb_node
*tree_search(struct btrfs_ordered_inode_tree
*tree
,
141 struct rb_root
*root
= &tree
->tree
;
142 struct rb_node
*prev
= NULL
;
144 struct btrfs_ordered_extent
*entry
;
147 entry
= rb_entry(tree
->last
, struct btrfs_ordered_extent
,
149 if (offset_in_entry(entry
, file_offset
))
152 ret
= __tree_search(root
, file_offset
, &prev
);
160 /* allocate and add a new ordered_extent into the per-inode tree.
161 * file_offset is the logical offset in the file
163 * start is the disk block number of an extent already reserved in the
164 * extent allocation tree
166 * len is the length of the extent
168 * The tree is given a single reference on the ordered extent that was
171 static int __btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
172 u64 start
, u64 len
, u64 disk_len
,
173 int type
, int dio
, int compress_type
)
175 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
176 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
177 struct btrfs_ordered_inode_tree
*tree
;
178 struct rb_node
*node
;
179 struct btrfs_ordered_extent
*entry
;
181 tree
= &BTRFS_I(inode
)->ordered_tree
;
182 entry
= kmem_cache_zalloc(btrfs_ordered_extent_cache
, GFP_NOFS
);
186 entry
->file_offset
= file_offset
;
187 entry
->start
= start
;
189 entry
->disk_len
= disk_len
;
190 entry
->bytes_left
= len
;
191 entry
->inode
= igrab(inode
);
192 entry
->compress_type
= compress_type
;
193 entry
->truncated_len
= (u64
)-1;
194 if (type
!= BTRFS_ORDERED_IO_DONE
&& type
!= BTRFS_ORDERED_COMPLETE
)
195 set_bit(type
, &entry
->flags
);
198 set_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
);
200 /* one ref for the tree */
201 refcount_set(&entry
->refs
, 1);
202 init_waitqueue_head(&entry
->wait
);
203 INIT_LIST_HEAD(&entry
->list
);
204 INIT_LIST_HEAD(&entry
->root_extent_list
);
205 INIT_LIST_HEAD(&entry
->work_list
);
206 init_completion(&entry
->completion
);
207 INIT_LIST_HEAD(&entry
->log_list
);
208 INIT_LIST_HEAD(&entry
->trans_list
);
210 trace_btrfs_ordered_extent_add(inode
, entry
);
212 spin_lock_irq(&tree
->lock
);
213 node
= tree_insert(&tree
->tree
, file_offset
,
216 ordered_data_tree_panic(inode
, -EEXIST
, file_offset
);
217 spin_unlock_irq(&tree
->lock
);
219 spin_lock(&root
->ordered_extent_lock
);
220 list_add_tail(&entry
->root_extent_list
,
221 &root
->ordered_extents
);
222 root
->nr_ordered_extents
++;
223 if (root
->nr_ordered_extents
== 1) {
224 spin_lock(&fs_info
->ordered_root_lock
);
225 BUG_ON(!list_empty(&root
->ordered_root
));
226 list_add_tail(&root
->ordered_root
, &fs_info
->ordered_roots
);
227 spin_unlock(&fs_info
->ordered_root_lock
);
229 spin_unlock(&root
->ordered_extent_lock
);
232 * We don't need the count_max_extents here, we can assume that all of
233 * that work has been done at higher layers, so this is truly the
234 * smallest the extent is going to get.
236 spin_lock(&BTRFS_I(inode
)->lock
);
237 btrfs_mod_outstanding_extents(BTRFS_I(inode
), 1);
238 spin_unlock(&BTRFS_I(inode
)->lock
);
243 int btrfs_add_ordered_extent(struct inode
*inode
, u64 file_offset
,
244 u64 start
, u64 len
, u64 disk_len
, int type
)
246 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
248 BTRFS_COMPRESS_NONE
);
251 int btrfs_add_ordered_extent_dio(struct inode
*inode
, u64 file_offset
,
252 u64 start
, u64 len
, u64 disk_len
, int type
)
254 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
256 BTRFS_COMPRESS_NONE
);
259 int btrfs_add_ordered_extent_compress(struct inode
*inode
, u64 file_offset
,
260 u64 start
, u64 len
, u64 disk_len
,
261 int type
, int compress_type
)
263 return __btrfs_add_ordered_extent(inode
, file_offset
, start
, len
,
269 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
270 * when an ordered extent is finished. If the list covers more than one
271 * ordered extent, it is split across multiples.
273 void btrfs_add_ordered_sum(struct inode
*inode
,
274 struct btrfs_ordered_extent
*entry
,
275 struct btrfs_ordered_sum
*sum
)
277 struct btrfs_ordered_inode_tree
*tree
;
279 tree
= &BTRFS_I(inode
)->ordered_tree
;
280 spin_lock_irq(&tree
->lock
);
281 list_add_tail(&sum
->list
, &entry
->list
);
282 spin_unlock_irq(&tree
->lock
);
286 * this is used to account for finished IO across a given range
287 * of the file. The IO may span ordered extents. If
288 * a given ordered_extent is completely done, 1 is returned, otherwise
291 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
292 * to make sure this function only returns 1 once for a given ordered extent.
294 * file_offset is updated to one byte past the range that is recorded as
295 * complete. This allows you to walk forward in the file.
297 int btrfs_dec_test_first_ordered_pending(struct inode
*inode
,
298 struct btrfs_ordered_extent
**cached
,
299 u64
*file_offset
, u64 io_size
, int uptodate
)
301 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
302 struct btrfs_ordered_inode_tree
*tree
;
303 struct rb_node
*node
;
304 struct btrfs_ordered_extent
*entry
= NULL
;
311 tree
= &BTRFS_I(inode
)->ordered_tree
;
312 spin_lock_irqsave(&tree
->lock
, flags
);
313 node
= tree_search(tree
, *file_offset
);
319 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
320 if (!offset_in_entry(entry
, *file_offset
)) {
325 dec_start
= max(*file_offset
, entry
->file_offset
);
326 dec_end
= min(*file_offset
+ io_size
, entry
->file_offset
+
328 *file_offset
= dec_end
;
329 if (dec_start
> dec_end
) {
330 btrfs_crit(fs_info
, "bad ordering dec_start %llu end %llu",
333 to_dec
= dec_end
- dec_start
;
334 if (to_dec
> entry
->bytes_left
) {
336 "bad ordered accounting left %llu size %llu",
337 entry
->bytes_left
, to_dec
);
339 entry
->bytes_left
-= to_dec
;
341 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
343 if (entry
->bytes_left
== 0) {
344 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
345 /* test_and_set_bit implies a barrier */
346 cond_wake_up_nomb(&entry
->wait
);
351 if (!ret
&& cached
&& entry
) {
353 refcount_inc(&entry
->refs
);
355 spin_unlock_irqrestore(&tree
->lock
, flags
);
360 * this is used to account for finished IO across a given range
361 * of the file. The IO should not span ordered extents. If
362 * a given ordered_extent is completely done, 1 is returned, otherwise
365 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
366 * to make sure this function only returns 1 once for a given ordered extent.
368 int btrfs_dec_test_ordered_pending(struct inode
*inode
,
369 struct btrfs_ordered_extent
**cached
,
370 u64 file_offset
, u64 io_size
, int uptodate
)
372 struct btrfs_ordered_inode_tree
*tree
;
373 struct rb_node
*node
;
374 struct btrfs_ordered_extent
*entry
= NULL
;
378 tree
= &BTRFS_I(inode
)->ordered_tree
;
379 spin_lock_irqsave(&tree
->lock
, flags
);
380 if (cached
&& *cached
) {
385 node
= tree_search(tree
, file_offset
);
391 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
393 if (!offset_in_entry(entry
, file_offset
)) {
398 if (io_size
> entry
->bytes_left
) {
399 btrfs_crit(BTRFS_I(inode
)->root
->fs_info
,
400 "bad ordered accounting left %llu size %llu",
401 entry
->bytes_left
, io_size
);
403 entry
->bytes_left
-= io_size
;
405 set_bit(BTRFS_ORDERED_IOERR
, &entry
->flags
);
407 if (entry
->bytes_left
== 0) {
408 ret
= test_and_set_bit(BTRFS_ORDERED_IO_DONE
, &entry
->flags
);
409 /* test_and_set_bit implies a barrier */
410 cond_wake_up_nomb(&entry
->wait
);
415 if (!ret
&& cached
&& entry
) {
417 refcount_inc(&entry
->refs
);
419 spin_unlock_irqrestore(&tree
->lock
, flags
);
424 * used to drop a reference on an ordered extent. This will free
425 * the extent if the last reference is dropped
427 void btrfs_put_ordered_extent(struct btrfs_ordered_extent
*entry
)
429 struct list_head
*cur
;
430 struct btrfs_ordered_sum
*sum
;
432 trace_btrfs_ordered_extent_put(entry
->inode
, entry
);
434 if (refcount_dec_and_test(&entry
->refs
)) {
435 ASSERT(list_empty(&entry
->log_list
));
436 ASSERT(list_empty(&entry
->trans_list
));
437 ASSERT(list_empty(&entry
->root_extent_list
));
438 ASSERT(RB_EMPTY_NODE(&entry
->rb_node
));
440 btrfs_add_delayed_iput(entry
->inode
);
441 while (!list_empty(&entry
->list
)) {
442 cur
= entry
->list
.next
;
443 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
444 list_del(&sum
->list
);
447 kmem_cache_free(btrfs_ordered_extent_cache
, entry
);
452 * remove an ordered extent from the tree. No references are dropped
453 * and waiters are woken up.
455 void btrfs_remove_ordered_extent(struct inode
*inode
,
456 struct btrfs_ordered_extent
*entry
)
458 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
459 struct btrfs_ordered_inode_tree
*tree
;
460 struct btrfs_inode
*btrfs_inode
= BTRFS_I(inode
);
461 struct btrfs_root
*root
= btrfs_inode
->root
;
462 struct rb_node
*node
;
464 /* This is paired with btrfs_add_ordered_extent. */
465 spin_lock(&btrfs_inode
->lock
);
466 btrfs_mod_outstanding_extents(btrfs_inode
, -1);
467 spin_unlock(&btrfs_inode
->lock
);
468 if (root
!= fs_info
->tree_root
)
469 btrfs_delalloc_release_metadata(btrfs_inode
, entry
->len
, false);
471 tree
= &btrfs_inode
->ordered_tree
;
472 spin_lock_irq(&tree
->lock
);
473 node
= &entry
->rb_node
;
474 rb_erase(node
, &tree
->tree
);
476 if (tree
->last
== node
)
478 set_bit(BTRFS_ORDERED_COMPLETE
, &entry
->flags
);
479 spin_unlock_irq(&tree
->lock
);
481 spin_lock(&root
->ordered_extent_lock
);
482 list_del_init(&entry
->root_extent_list
);
483 root
->nr_ordered_extents
--;
485 trace_btrfs_ordered_extent_remove(inode
, entry
);
487 if (!root
->nr_ordered_extents
) {
488 spin_lock(&fs_info
->ordered_root_lock
);
489 BUG_ON(list_empty(&root
->ordered_root
));
490 list_del_init(&root
->ordered_root
);
491 spin_unlock(&fs_info
->ordered_root_lock
);
493 spin_unlock(&root
->ordered_extent_lock
);
494 wake_up(&entry
->wait
);
497 static void btrfs_run_ordered_extent_work(struct btrfs_work
*work
)
499 struct btrfs_ordered_extent
*ordered
;
501 ordered
= container_of(work
, struct btrfs_ordered_extent
, flush_work
);
502 btrfs_start_ordered_extent(ordered
->inode
, ordered
, 1);
503 complete(&ordered
->completion
);
507 * wait for all the ordered extents in a root. This is done when balancing
508 * space between drives.
510 u64
btrfs_wait_ordered_extents(struct btrfs_root
*root
, u64 nr
,
511 const u64 range_start
, const u64 range_len
)
513 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
517 struct btrfs_ordered_extent
*ordered
, *next
;
519 const u64 range_end
= range_start
+ range_len
;
521 mutex_lock(&root
->ordered_extent_mutex
);
522 spin_lock(&root
->ordered_extent_lock
);
523 list_splice_init(&root
->ordered_extents
, &splice
);
524 while (!list_empty(&splice
) && nr
) {
525 ordered
= list_first_entry(&splice
, struct btrfs_ordered_extent
,
528 if (range_end
<= ordered
->start
||
529 ordered
->start
+ ordered
->disk_len
<= range_start
) {
530 list_move_tail(&ordered
->root_extent_list
, &skipped
);
531 cond_resched_lock(&root
->ordered_extent_lock
);
535 list_move_tail(&ordered
->root_extent_list
,
536 &root
->ordered_extents
);
537 refcount_inc(&ordered
->refs
);
538 spin_unlock(&root
->ordered_extent_lock
);
540 btrfs_init_work(&ordered
->flush_work
,
541 btrfs_flush_delalloc_helper
,
542 btrfs_run_ordered_extent_work
, NULL
, NULL
);
543 list_add_tail(&ordered
->work_list
, &works
);
544 btrfs_queue_work(fs_info
->flush_workers
, &ordered
->flush_work
);
547 spin_lock(&root
->ordered_extent_lock
);
552 list_splice_tail(&skipped
, &root
->ordered_extents
);
553 list_splice_tail(&splice
, &root
->ordered_extents
);
554 spin_unlock(&root
->ordered_extent_lock
);
556 list_for_each_entry_safe(ordered
, next
, &works
, work_list
) {
557 list_del_init(&ordered
->work_list
);
558 wait_for_completion(&ordered
->completion
);
559 btrfs_put_ordered_extent(ordered
);
562 mutex_unlock(&root
->ordered_extent_mutex
);
567 u64
btrfs_wait_ordered_roots(struct btrfs_fs_info
*fs_info
, u64 nr
,
568 const u64 range_start
, const u64 range_len
)
570 struct btrfs_root
*root
;
571 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_fs_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_fs_root(root
);
594 spin_lock(&fs_info
->ordered_root_lock
);
599 list_splice_tail(&splice
, &fs_info
->ordered_roots
);
600 spin_unlock(&fs_info
->ordered_root_lock
);
601 mutex_unlock(&fs_info
->ordered_operations_mutex
);
607 * Used to start IO or wait for a given ordered extent to finish.
609 * If wait is one, this effectively waits on page writeback for all the pages
610 * in the extent, and it waits on the io completion code to insert
611 * metadata into the btree corresponding to the extent
613 void btrfs_start_ordered_extent(struct inode
*inode
,
614 struct btrfs_ordered_extent
*entry
,
617 u64 start
= entry
->file_offset
;
618 u64 end
= start
+ entry
->len
- 1;
620 trace_btrfs_ordered_extent_start(inode
, entry
);
623 * pages in the range can be dirty, clean or writeback. We
624 * start IO on any dirty ones so the wait doesn't stall waiting
625 * for the flusher thread to find them
627 if (!test_bit(BTRFS_ORDERED_DIRECT
, &entry
->flags
))
628 filemap_fdatawrite_range(inode
->i_mapping
, start
, end
);
630 wait_event(entry
->wait
, test_bit(BTRFS_ORDERED_COMPLETE
,
636 * Used to wait on ordered extents across a large range of bytes.
638 int btrfs_wait_ordered_range(struct inode
*inode
, u64 start
, u64 len
)
644 struct btrfs_ordered_extent
*ordered
;
646 if (start
+ len
< start
) {
647 orig_end
= INT_LIMIT(loff_t
);
649 orig_end
= start
+ len
- 1;
650 if (orig_end
> INT_LIMIT(loff_t
))
651 orig_end
= INT_LIMIT(loff_t
);
654 /* start IO across the range first to instantiate any delalloc
657 ret
= btrfs_fdatawrite_range(inode
, start
, orig_end
);
662 * If we have a writeback error don't return immediately. Wait first
663 * for any ordered extents that haven't completed yet. This is to make
664 * sure no one can dirty the same page ranges and call writepages()
665 * before the ordered extents complete - to avoid failures (-EEXIST)
666 * when adding the new ordered extents to the ordered tree.
668 ret_wb
= filemap_fdatawait_range(inode
->i_mapping
, start
, orig_end
);
672 ordered
= btrfs_lookup_first_ordered_extent(inode
, end
);
675 if (ordered
->file_offset
> orig_end
) {
676 btrfs_put_ordered_extent(ordered
);
679 if (ordered
->file_offset
+ ordered
->len
<= start
) {
680 btrfs_put_ordered_extent(ordered
);
683 btrfs_start_ordered_extent(inode
, ordered
, 1);
684 end
= ordered
->file_offset
;
685 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
))
687 btrfs_put_ordered_extent(ordered
);
688 if (ret
|| end
== 0 || end
== start
)
692 return ret_wb
? ret_wb
: ret
;
696 * find an ordered extent corresponding to file_offset. return NULL if
697 * nothing is found, otherwise take a reference on the extent and return it
699 struct btrfs_ordered_extent
*btrfs_lookup_ordered_extent(struct inode
*inode
,
702 struct btrfs_ordered_inode_tree
*tree
;
703 struct rb_node
*node
;
704 struct btrfs_ordered_extent
*entry
= NULL
;
706 tree
= &BTRFS_I(inode
)->ordered_tree
;
707 spin_lock_irq(&tree
->lock
);
708 node
= tree_search(tree
, file_offset
);
712 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
713 if (!offset_in_entry(entry
, file_offset
))
716 refcount_inc(&entry
->refs
);
718 spin_unlock_irq(&tree
->lock
);
722 /* Since the DIO code tries to lock a wide area we need to look for any ordered
723 * extents that exist in the range, rather than just the start of the range.
725 struct btrfs_ordered_extent
*btrfs_lookup_ordered_range(
726 struct btrfs_inode
*inode
, u64 file_offset
, u64 len
)
728 struct btrfs_ordered_inode_tree
*tree
;
729 struct rb_node
*node
;
730 struct btrfs_ordered_extent
*entry
= NULL
;
732 tree
= &inode
->ordered_tree
;
733 spin_lock_irq(&tree
->lock
);
734 node
= tree_search(tree
, file_offset
);
736 node
= tree_search(tree
, file_offset
+ len
);
742 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
743 if (range_overlaps(entry
, file_offset
, len
))
746 if (entry
->file_offset
>= file_offset
+ len
) {
751 node
= rb_next(node
);
757 refcount_inc(&entry
->refs
);
758 spin_unlock_irq(&tree
->lock
);
763 * lookup and return any extent before 'file_offset'. NULL is returned
766 struct btrfs_ordered_extent
*
767 btrfs_lookup_first_ordered_extent(struct inode
*inode
, u64 file_offset
)
769 struct btrfs_ordered_inode_tree
*tree
;
770 struct rb_node
*node
;
771 struct btrfs_ordered_extent
*entry
= NULL
;
773 tree
= &BTRFS_I(inode
)->ordered_tree
;
774 spin_lock_irq(&tree
->lock
);
775 node
= tree_search(tree
, file_offset
);
779 entry
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
780 refcount_inc(&entry
->refs
);
782 spin_unlock_irq(&tree
->lock
);
787 * After an extent is done, call this to conditionally update the on disk
788 * i_size. i_size is updated to cover any fully written part of the file.
790 int btrfs_ordered_update_i_size(struct inode
*inode
, u64 offset
,
791 struct btrfs_ordered_extent
*ordered
)
793 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
796 u64 i_size
= i_size_read(inode
);
797 struct rb_node
*node
;
798 struct rb_node
*prev
= NULL
;
799 struct btrfs_ordered_extent
*test
;
801 u64 orig_offset
= offset
;
803 spin_lock_irq(&tree
->lock
);
805 offset
= entry_end(ordered
);
806 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
))
808 ordered
->file_offset
+
809 ordered
->truncated_len
);
811 offset
= ALIGN(offset
, btrfs_inode_sectorsize(inode
));
813 disk_i_size
= BTRFS_I(inode
)->disk_i_size
;
817 * If ordered is not NULL, then this is called from endio and
818 * disk_i_size will be updated by either truncate itself or any
819 * in-flight IOs which are inside the disk_i_size.
821 * Because btrfs_setsize() may set i_size with disk_i_size if truncate
822 * fails somehow, we need to make sure we have a precise disk_i_size by
823 * updating it as usual.
826 if (!ordered
&& disk_i_size
> i_size
) {
827 BTRFS_I(inode
)->disk_i_size
= orig_offset
;
833 * if the disk i_size is already at the inode->i_size, or
834 * this ordered extent is inside the disk i_size, we're done
836 if (disk_i_size
== i_size
)
840 * We still need to update disk_i_size if outstanding_isize is greater
843 if (offset
<= disk_i_size
&&
844 (!ordered
|| ordered
->outstanding_isize
<= disk_i_size
))
848 * walk backward from this ordered extent to disk_i_size.
849 * if we find an ordered extent then we can't update disk i_size
853 node
= rb_prev(&ordered
->rb_node
);
855 prev
= tree_search(tree
, offset
);
857 * we insert file extents without involving ordered struct,
858 * so there should be no ordered struct cover this offset
861 test
= rb_entry(prev
, struct btrfs_ordered_extent
,
863 BUG_ON(offset_in_entry(test
, offset
));
867 for (; node
; node
= rb_prev(node
)) {
868 test
= rb_entry(node
, struct btrfs_ordered_extent
, rb_node
);
870 /* We treat this entry as if it doesn't exist */
871 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &test
->flags
))
874 if (entry_end(test
) <= disk_i_size
)
876 if (test
->file_offset
>= i_size
)
880 * We don't update disk_i_size now, so record this undealt
881 * i_size. Or we will not know the real i_size.
883 if (test
->outstanding_isize
< offset
)
884 test
->outstanding_isize
= offset
;
886 ordered
->outstanding_isize
> test
->outstanding_isize
)
887 test
->outstanding_isize
= ordered
->outstanding_isize
;
890 new_i_size
= min_t(u64
, offset
, i_size
);
893 * Some ordered extents may completed before the current one, and
894 * we hold the real i_size in ->outstanding_isize.
896 if (ordered
&& ordered
->outstanding_isize
> new_i_size
)
897 new_i_size
= min_t(u64
, ordered
->outstanding_isize
, i_size
);
898 BTRFS_I(inode
)->disk_i_size
= new_i_size
;
902 * We need to do this because we can't remove ordered extents until
903 * after the i_disk_size has been updated and then the inode has been
904 * updated to reflect the change, so we need to tell anybody who finds
905 * this ordered extent that we've already done all the real work, we
906 * just haven't completed all the other work.
909 set_bit(BTRFS_ORDERED_UPDATED_ISIZE
, &ordered
->flags
);
910 spin_unlock_irq(&tree
->lock
);
915 * search the ordered extents for one corresponding to 'offset' and
916 * try to find a checksum. This is used because we allow pages to
917 * be reclaimed before their checksum is actually put into the btree
919 int btrfs_find_ordered_sum(struct inode
*inode
, u64 offset
, u64 disk_bytenr
,
922 struct btrfs_ordered_sum
*ordered_sum
;
923 struct btrfs_ordered_extent
*ordered
;
924 struct btrfs_ordered_inode_tree
*tree
= &BTRFS_I(inode
)->ordered_tree
;
925 unsigned long num_sectors
;
927 u32 sectorsize
= btrfs_inode_sectorsize(inode
);
930 ordered
= btrfs_lookup_ordered_extent(inode
, offset
);
934 spin_lock_irq(&tree
->lock
);
935 list_for_each_entry_reverse(ordered_sum
, &ordered
->list
, list
) {
936 if (disk_bytenr
>= ordered_sum
->bytenr
&&
937 disk_bytenr
< ordered_sum
->bytenr
+ ordered_sum
->len
) {
938 i
= (disk_bytenr
- ordered_sum
->bytenr
) >>
939 inode
->i_sb
->s_blocksize_bits
;
940 num_sectors
= ordered_sum
->len
>>
941 inode
->i_sb
->s_blocksize_bits
;
942 num_sectors
= min_t(int, len
- index
, num_sectors
- i
);
943 memcpy(sum
+ index
, ordered_sum
->sums
+ i
,
946 index
+= (int)num_sectors
;
949 disk_bytenr
+= num_sectors
* sectorsize
;
953 spin_unlock_irq(&tree
->lock
);
954 btrfs_put_ordered_extent(ordered
);
958 int __init
ordered_data_init(void)
960 btrfs_ordered_extent_cache
= kmem_cache_create("btrfs_ordered_extent",
961 sizeof(struct btrfs_ordered_extent
), 0,
964 if (!btrfs_ordered_extent_cache
)
970 void __cold
ordered_data_exit(void)
972 kmem_cache_destroy(btrfs_ordered_extent_cache
);