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;
463 bool dec_pending_ordered = false;
465 /* This is paired with btrfs_add_ordered_extent. */
466 spin_lock(&btrfs_inode->lock);
467 btrfs_mod_outstanding_extents(btrfs_inode, -1);
468 spin_unlock(&btrfs_inode->lock);
469 if (root != fs_info->tree_root)
470 btrfs_delalloc_release_metadata(btrfs_inode, entry->len, false);
472 tree = &btrfs_inode->ordered_tree;
473 spin_lock_irq(&tree->lock);
474 node = &entry->rb_node;
475 rb_erase(node, &tree->tree);
477 if (tree->last == node)
479 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
480 if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
481 dec_pending_ordered = true;
482 spin_unlock_irq(&tree->lock);
485 * The current running transaction is waiting on us, we need to let it
486 * know that we're complete and wake it up.
488 if (dec_pending_ordered) {
489 struct btrfs_transaction *trans;
492 * The checks for trans are just a formality, it should be set,
493 * but if it isn't we don't want to deref/assert under the spin
494 * lock, so be nice and check if trans is set, but ASSERT() so
495 * if it isn't set a developer will notice.
497 spin_lock(&fs_info->trans_lock);
498 trans = fs_info->running_transaction;
500 refcount_inc(&trans->use_count);
501 spin_unlock(&fs_info->trans_lock);
505 if (atomic_dec_and_test(&trans->pending_ordered))
506 wake_up(&trans->pending_wait);
507 btrfs_put_transaction(trans);
511 spin_lock(&root->ordered_extent_lock);
512 list_del_init(&entry->root_extent_list);
513 root->nr_ordered_extents--;
515 trace_btrfs_ordered_extent_remove(inode, entry);
517 if (!root->nr_ordered_extents) {
518 spin_lock(&fs_info->ordered_root_lock);
519 BUG_ON(list_empty(&root->ordered_root));
520 list_del_init(&root->ordered_root);
521 spin_unlock(&fs_info->ordered_root_lock);
523 spin_unlock(&root->ordered_extent_lock);
524 wake_up(&entry->wait);
527 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
529 struct btrfs_ordered_extent *ordered;
531 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
532 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
533 complete(&ordered->completion);
537 * wait for all the ordered extents in a root. This is done when balancing
538 * space between drives.
540 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
541 const u64 range_start, const u64 range_len)
543 struct btrfs_fs_info *fs_info = root->fs_info;
547 struct btrfs_ordered_extent *ordered, *next;
549 const u64 range_end = range_start + range_len;
551 mutex_lock(&root->ordered_extent_mutex);
552 spin_lock(&root->ordered_extent_lock);
553 list_splice_init(&root->ordered_extents, &splice);
554 while (!list_empty(&splice) && nr) {
555 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
558 if (range_end <= ordered->start ||
559 ordered->start + ordered->disk_len <= range_start) {
560 list_move_tail(&ordered->root_extent_list, &skipped);
561 cond_resched_lock(&root->ordered_extent_lock);
565 list_move_tail(&ordered->root_extent_list,
566 &root->ordered_extents);
567 refcount_inc(&ordered->refs);
568 spin_unlock(&root->ordered_extent_lock);
570 btrfs_init_work(&ordered->flush_work,
571 btrfs_flush_delalloc_helper,
572 btrfs_run_ordered_extent_work, NULL, NULL);
573 list_add_tail(&ordered->work_list, &works);
574 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
577 spin_lock(&root->ordered_extent_lock);
582 list_splice_tail(&skipped, &root->ordered_extents);
583 list_splice_tail(&splice, &root->ordered_extents);
584 spin_unlock(&root->ordered_extent_lock);
586 list_for_each_entry_safe(ordered, next, &works, work_list) {
587 list_del_init(&ordered->work_list);
588 wait_for_completion(&ordered->completion);
589 btrfs_put_ordered_extent(ordered);
592 mutex_unlock(&root->ordered_extent_mutex);
597 u64 btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
598 const u64 range_start, const u64 range_len)
600 struct btrfs_root *root;
601 struct list_head splice;
605 INIT_LIST_HEAD(&splice);
607 mutex_lock(&fs_info->ordered_operations_mutex);
608 spin_lock(&fs_info->ordered_root_lock);
609 list_splice_init(&fs_info->ordered_roots, &splice);
610 while (!list_empty(&splice) && nr) {
611 root = list_first_entry(&splice, struct btrfs_root,
613 root = btrfs_grab_fs_root(root);
615 list_move_tail(&root->ordered_root,
616 &fs_info->ordered_roots);
617 spin_unlock(&fs_info->ordered_root_lock);
619 done = btrfs_wait_ordered_extents(root, nr,
620 range_start, range_len);
621 btrfs_put_fs_root(root);
624 spin_lock(&fs_info->ordered_root_lock);
629 list_splice_tail(&splice, &fs_info->ordered_roots);
630 spin_unlock(&fs_info->ordered_root_lock);
631 mutex_unlock(&fs_info->ordered_operations_mutex);
637 * Used to start IO or wait for a given ordered extent to finish.
639 * If wait is one, this effectively waits on page writeback for all the pages
640 * in the extent, and it waits on the io completion code to insert
641 * metadata into the btree corresponding to the extent
643 void btrfs_start_ordered_extent(struct inode *inode,
644 struct btrfs_ordered_extent *entry,
647 u64 start = entry->file_offset;
648 u64 end = start + entry->len - 1;
650 trace_btrfs_ordered_extent_start(inode, entry);
653 * pages in the range can be dirty, clean or writeback. We
654 * start IO on any dirty ones so the wait doesn't stall waiting
655 * for the flusher thread to find them
657 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
658 filemap_fdatawrite_range(inode->i_mapping, start, end);
660 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
666 * Used to wait on ordered extents across a large range of bytes.
668 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
674 struct btrfs_ordered_extent *ordered;
676 if (start + len < start) {
677 orig_end = INT_LIMIT(loff_t);
679 orig_end = start + len - 1;
680 if (orig_end > INT_LIMIT(loff_t))
681 orig_end = INT_LIMIT(loff_t);
684 /* start IO across the range first to instantiate any delalloc
687 ret = btrfs_fdatawrite_range(inode, start, orig_end);
692 * If we have a writeback error don't return immediately. Wait first
693 * for any ordered extents that haven't completed yet. This is to make
694 * sure no one can dirty the same page ranges and call writepages()
695 * before the ordered extents complete - to avoid failures (-EEXIST)
696 * when adding the new ordered extents to the ordered tree.
698 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
702 ordered = btrfs_lookup_first_ordered_extent(inode, end);
705 if (ordered->file_offset > orig_end) {
706 btrfs_put_ordered_extent(ordered);
709 if (ordered->file_offset + ordered->len <= start) {
710 btrfs_put_ordered_extent(ordered);
713 btrfs_start_ordered_extent(inode, ordered, 1);
714 end = ordered->file_offset;
716 * If the ordered extent had an error save the error but don't
717 * exit without waiting first for all other ordered extents in
718 * the range to complete.
720 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
722 btrfs_put_ordered_extent(ordered);
723 if (end == 0 || end == start)
727 return ret_wb ? ret_wb : ret;
731 * find an ordered extent corresponding to file_offset. return NULL if
732 * nothing is found, otherwise take a reference on the extent and return it
734 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
737 struct btrfs_ordered_inode_tree *tree;
738 struct rb_node *node;
739 struct btrfs_ordered_extent *entry = NULL;
741 tree = &BTRFS_I(inode)->ordered_tree;
742 spin_lock_irq(&tree->lock);
743 node = tree_search(tree, file_offset);
747 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
748 if (!offset_in_entry(entry, file_offset))
751 refcount_inc(&entry->refs);
753 spin_unlock_irq(&tree->lock);
757 /* Since the DIO code tries to lock a wide area we need to look for any ordered
758 * extents that exist in the range, rather than just the start of the range.
760 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
761 struct btrfs_inode *inode, u64 file_offset, u64 len)
763 struct btrfs_ordered_inode_tree *tree;
764 struct rb_node *node;
765 struct btrfs_ordered_extent *entry = NULL;
767 tree = &inode->ordered_tree;
768 spin_lock_irq(&tree->lock);
769 node = tree_search(tree, file_offset);
771 node = tree_search(tree, file_offset + len);
777 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
778 if (range_overlaps(entry, file_offset, len))
781 if (entry->file_offset >= file_offset + len) {
786 node = rb_next(node);
792 refcount_inc(&entry->refs);
793 spin_unlock_irq(&tree->lock);
798 * lookup and return any extent before 'file_offset'. NULL is returned
801 struct btrfs_ordered_extent *
802 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
804 struct btrfs_ordered_inode_tree *tree;
805 struct rb_node *node;
806 struct btrfs_ordered_extent *entry = NULL;
808 tree = &BTRFS_I(inode)->ordered_tree;
809 spin_lock_irq(&tree->lock);
810 node = tree_search(tree, file_offset);
814 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
815 refcount_inc(&entry->refs);
817 spin_unlock_irq(&tree->lock);
822 * After an extent is done, call this to conditionally update the on disk
823 * i_size. i_size is updated to cover any fully written part of the file.
825 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
826 struct btrfs_ordered_extent *ordered)
828 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
831 u64 i_size = i_size_read(inode);
832 struct rb_node *node;
833 struct rb_node *prev = NULL;
834 struct btrfs_ordered_extent *test;
836 u64 orig_offset = offset;
838 spin_lock_irq(&tree->lock);
840 offset = entry_end(ordered);
841 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
843 ordered->file_offset +
844 ordered->truncated_len);
846 offset = ALIGN(offset, btrfs_inode_sectorsize(inode));
848 disk_i_size = BTRFS_I(inode)->disk_i_size;
852 * If ordered is not NULL, then this is called from endio and
853 * disk_i_size will be updated by either truncate itself or any
854 * in-flight IOs which are inside the disk_i_size.
856 * Because btrfs_setsize() may set i_size with disk_i_size if truncate
857 * fails somehow, we need to make sure we have a precise disk_i_size by
858 * updating it as usual.
861 if (!ordered && disk_i_size > i_size) {
862 BTRFS_I(inode)->disk_i_size = orig_offset;
868 * if the disk i_size is already at the inode->i_size, or
869 * this ordered extent is inside the disk i_size, we're done
871 if (disk_i_size == i_size)
875 * We still need to update disk_i_size if outstanding_isize is greater
878 if (offset <= disk_i_size &&
879 (!ordered || ordered->outstanding_isize <= disk_i_size))
883 * walk backward from this ordered extent to disk_i_size.
884 * if we find an ordered extent then we can't update disk i_size
888 node = rb_prev(&ordered->rb_node);
890 prev = tree_search(tree, offset);
892 * we insert file extents without involving ordered struct,
893 * so there should be no ordered struct cover this offset
896 test = rb_entry(prev, struct btrfs_ordered_extent,
898 BUG_ON(offset_in_entry(test, offset));
902 for (; node; node = rb_prev(node)) {
903 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
905 /* We treat this entry as if it doesn't exist */
906 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
909 if (entry_end(test) <= disk_i_size)
911 if (test->file_offset >= i_size)
915 * We don't update disk_i_size now, so record this undealt
916 * i_size. Or we will not know the real i_size.
918 if (test->outstanding_isize < offset)
919 test->outstanding_isize = offset;
921 ordered->outstanding_isize > test->outstanding_isize)
922 test->outstanding_isize = ordered->outstanding_isize;
925 new_i_size = min_t(u64, offset, i_size);
928 * Some ordered extents may completed before the current one, and
929 * we hold the real i_size in ->outstanding_isize.
931 if (ordered && ordered->outstanding_isize > new_i_size)
932 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
933 BTRFS_I(inode)->disk_i_size = new_i_size;
937 * We need to do this because we can't remove ordered extents until
938 * after the i_disk_size has been updated and then the inode has been
939 * updated to reflect the change, so we need to tell anybody who finds
940 * this ordered extent that we've already done all the real work, we
941 * just haven't completed all the other work.
944 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
945 spin_unlock_irq(&tree->lock);
950 * search the ordered extents for one corresponding to 'offset' and
951 * try to find a checksum. This is used because we allow pages to
952 * be reclaimed before their checksum is actually put into the btree
954 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
957 struct btrfs_ordered_sum *ordered_sum;
958 struct btrfs_ordered_extent *ordered;
959 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
960 unsigned long num_sectors;
962 u32 sectorsize = btrfs_inode_sectorsize(inode);
965 ordered = btrfs_lookup_ordered_extent(inode, offset);
969 spin_lock_irq(&tree->lock);
970 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
971 if (disk_bytenr >= ordered_sum->bytenr &&
972 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
973 i = (disk_bytenr - ordered_sum->bytenr) >>
974 inode->i_sb->s_blocksize_bits;
975 num_sectors = ordered_sum->len >>
976 inode->i_sb->s_blocksize_bits;
977 num_sectors = min_t(int, len - index, num_sectors - i);
978 memcpy(sum + index, ordered_sum->sums + i,
981 index += (int)num_sectors;
984 disk_bytenr += num_sectors * sectorsize;
988 spin_unlock_irq(&tree->lock);
989 btrfs_put_ordered_extent(ordered);
993 int __init ordered_data_init(void)
995 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
996 sizeof(struct btrfs_ordered_extent), 0,
999 if (!btrfs_ordered_extent_cache)
1005 void __cold ordered_data_exit(void)
1007 kmem_cache_destroy(btrfs_ordered_extent_cache);