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>
13 #include "transaction.h"
14 #include "btrfs_inode.h"
15 #include "extent_io.h"
17 #include "compression.h"
18 #include "delalloc-space.h"
24 static struct kmem_cache *btrfs_ordered_extent_cache;
26 static u64 entry_end(struct btrfs_ordered_extent *entry)
28 if (entry->file_offset + entry->num_bytes < entry->file_offset)
30 return entry->file_offset + entry->num_bytes;
33 /* returns NULL if the insertion worked, or it returns the node it did find
36 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
39 struct rb_node **p = &root->rb_node;
40 struct rb_node *parent = NULL;
41 struct btrfs_ordered_extent *entry;
45 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
47 if (file_offset < entry->file_offset)
49 else if (file_offset >= entry_end(entry))
55 rb_link_node(node, parent, p);
56 rb_insert_color(node, root);
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,
114 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
117 if (file_offset + len <= entry->file_offset ||
118 entry->file_offset + entry->num_bytes <= file_offset)
124 * look find the first ordered struct that has this offset, otherwise
125 * the first one less than this offset
127 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
130 struct rb_root *root = &tree->tree;
131 struct rb_node *prev = NULL;
133 struct btrfs_ordered_extent *entry;
136 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
138 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
141 ret = __tree_search(root, file_offset, &prev);
150 * Add an ordered extent to the per-inode tree.
152 * @inode: Inode that this extent is for.
153 * @file_offset: Logical offset in file where the extent starts.
154 * @num_bytes: Logical length of extent in file.
155 * @ram_bytes: Full length of unencoded data.
156 * @disk_bytenr: Offset of extent on disk.
157 * @disk_num_bytes: Size of extent on disk.
158 * @offset: Offset into unencoded data where file data starts.
159 * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*).
160 * @compress_type: Compression algorithm used for data.
162 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
163 * tree is given a single reference on the ordered extent that was inserted.
165 * Return: 0 or -ENOMEM.
167 int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
168 u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
169 u64 disk_num_bytes, u64 offset, unsigned flags,
172 struct btrfs_root *root = inode->root;
173 struct btrfs_fs_info *fs_info = root->fs_info;
174 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
175 struct rb_node *node;
176 struct btrfs_ordered_extent *entry;
180 ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
181 /* For nocow write, we can release the qgroup rsv right now */
182 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
188 * The ordered extent has reserved qgroup space, release now
189 * and pass the reserved number for qgroup_record to free.
191 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
195 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
199 entry->file_offset = file_offset;
200 entry->num_bytes = num_bytes;
201 entry->ram_bytes = ram_bytes;
202 entry->disk_bytenr = disk_bytenr;
203 entry->disk_num_bytes = disk_num_bytes;
204 entry->offset = offset;
205 entry->bytes_left = num_bytes;
206 entry->inode = igrab(&inode->vfs_inode);
207 entry->compress_type = compress_type;
208 entry->truncated_len = (u64)-1;
209 entry->qgroup_rsv = ret;
210 entry->physical = (u64)-1;
212 ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
213 entry->flags = flags;
215 percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
216 fs_info->delalloc_batch);
218 /* one ref for the tree */
219 refcount_set(&entry->refs, 1);
220 init_waitqueue_head(&entry->wait);
221 INIT_LIST_HEAD(&entry->list);
222 INIT_LIST_HEAD(&entry->log_list);
223 INIT_LIST_HEAD(&entry->root_extent_list);
224 INIT_LIST_HEAD(&entry->work_list);
225 init_completion(&entry->completion);
227 trace_btrfs_ordered_extent_add(inode, entry);
229 spin_lock_irq(&tree->lock);
230 node = tree_insert(&tree->tree, file_offset,
233 btrfs_panic(fs_info, -EEXIST,
234 "inconsistency in ordered tree at offset %llu",
236 spin_unlock_irq(&tree->lock);
238 spin_lock(&root->ordered_extent_lock);
239 list_add_tail(&entry->root_extent_list,
240 &root->ordered_extents);
241 root->nr_ordered_extents++;
242 if (root->nr_ordered_extents == 1) {
243 spin_lock(&fs_info->ordered_root_lock);
244 BUG_ON(!list_empty(&root->ordered_root));
245 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
246 spin_unlock(&fs_info->ordered_root_lock);
248 spin_unlock(&root->ordered_extent_lock);
251 * We don't need the count_max_extents here, we can assume that all of
252 * that work has been done at higher layers, so this is truly the
253 * smallest the extent is going to get.
255 spin_lock(&inode->lock);
256 btrfs_mod_outstanding_extents(inode, 1);
257 spin_unlock(&inode->lock);
263 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
264 * when an ordered extent is finished. If the list covers more than one
265 * ordered extent, it is split across multiples.
267 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
268 struct btrfs_ordered_sum *sum)
270 struct btrfs_ordered_inode_tree *tree;
272 tree = &BTRFS_I(entry->inode)->ordered_tree;
273 spin_lock_irq(&tree->lock);
274 list_add_tail(&sum->list, &entry->list);
275 spin_unlock_irq(&tree->lock);
278 static void finish_ordered_fn(struct btrfs_work *work)
280 struct btrfs_ordered_extent *ordered_extent;
282 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
283 btrfs_finish_ordered_io(ordered_extent);
287 * Mark all ordered extents io inside the specified range finished.
289 * @page: The involved page for the operation.
290 * For uncompressed buffered IO, the page status also needs to be
291 * updated to indicate whether the pending ordered io is finished.
292 * Can be NULL for direct IO and compressed write.
293 * For these cases, callers are ensured they won't execute the
294 * endio function twice.
296 * This function is called for endio, thus the range must have ordered
297 * extent(s) covering it.
299 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
300 struct page *page, u64 file_offset,
301 u64 num_bytes, bool uptodate)
303 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
304 struct btrfs_fs_info *fs_info = inode->root->fs_info;
305 struct btrfs_workqueue *wq;
306 struct rb_node *node;
307 struct btrfs_ordered_extent *entry = NULL;
309 u64 cur = file_offset;
311 if (btrfs_is_free_space_inode(inode))
312 wq = fs_info->endio_freespace_worker;
314 wq = fs_info->endio_write_workers;
317 ASSERT(page->mapping && page_offset(page) <= file_offset &&
318 file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
320 spin_lock_irqsave(&tree->lock, flags);
321 while (cur < file_offset + num_bytes) {
326 node = tree_search(tree, cur);
327 /* No ordered extents at all */
331 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
332 entry_end = entry->file_offset + entry->num_bytes;
338 if (cur >= entry_end) {
339 node = rb_next(node);
340 /* No more ordered extents, exit */
343 entry = rb_entry(node, struct btrfs_ordered_extent,
346 /* Go to next ordered extent and continue */
347 cur = entry->file_offset;
353 * Go to the start of OE.
355 if (cur < entry->file_offset) {
356 cur = entry->file_offset;
361 * Now we are definitely inside one ordered extent.
367 end = min(entry->file_offset + entry->num_bytes,
368 file_offset + num_bytes) - 1;
369 ASSERT(end + 1 - cur < U32_MAX);
374 * Ordered (Private2) bit indicates whether we still
375 * have pending io unfinished for the ordered extent.
377 * If there's no such bit, we need to skip to next range.
379 if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
383 btrfs_page_clear_ordered(fs_info, page, cur, len);
386 /* Now we're fine to update the accounting */
387 if (unlikely(len > entry->bytes_left)) {
390 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
391 inode->root->root_key.objectid,
395 len, entry->bytes_left);
396 entry->bytes_left = 0;
398 entry->bytes_left -= len;
402 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
405 * All the IO of the ordered extent is finished, we need to queue
406 * the finish_func to be executed.
408 if (entry->bytes_left == 0) {
409 set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
410 cond_wake_up(&entry->wait);
411 refcount_inc(&entry->refs);
412 trace_btrfs_ordered_extent_mark_finished(inode, entry);
413 spin_unlock_irqrestore(&tree->lock, flags);
414 btrfs_init_work(&entry->work, finish_ordered_fn, NULL, NULL);
415 btrfs_queue_work(wq, &entry->work);
416 spin_lock_irqsave(&tree->lock, flags);
420 spin_unlock_irqrestore(&tree->lock, flags);
424 * Finish IO for one ordered extent across a given range. The range can only
425 * contain one ordered extent.
427 * @cached: The cached ordered extent. If not NULL, we can skip the tree
428 * search and use the ordered extent directly.
429 * Will be also used to store the finished ordered extent.
430 * @file_offset: File offset for the finished IO
431 * @io_size: Length of the finish IO range
433 * Return true if the ordered extent is finished in the range, and update
435 * Return false otherwise.
437 * NOTE: The range can NOT cross multiple ordered extents.
438 * Thus caller should ensure the range doesn't cross ordered extents.
440 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
441 struct btrfs_ordered_extent **cached,
442 u64 file_offset, u64 io_size)
444 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
445 struct rb_node *node;
446 struct btrfs_ordered_extent *entry = NULL;
448 bool finished = false;
450 spin_lock_irqsave(&tree->lock, flags);
451 if (cached && *cached) {
456 node = tree_search(tree, file_offset);
460 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
462 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
465 if (io_size > entry->bytes_left)
466 btrfs_crit(inode->root->fs_info,
467 "bad ordered accounting left %llu size %llu",
468 entry->bytes_left, io_size);
470 entry->bytes_left -= io_size;
472 if (entry->bytes_left == 0) {
474 * Ensure only one caller can set the flag and finished_ret
477 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
478 /* test_and_set_bit implies a barrier */
479 cond_wake_up_nomb(&entry->wait);
482 if (finished && cached && entry) {
484 refcount_inc(&entry->refs);
485 trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
487 spin_unlock_irqrestore(&tree->lock, flags);
492 * used to drop a reference on an ordered extent. This will free
493 * the extent if the last reference is dropped
495 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
497 struct list_head *cur;
498 struct btrfs_ordered_sum *sum;
500 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
502 if (refcount_dec_and_test(&entry->refs)) {
503 ASSERT(list_empty(&entry->root_extent_list));
504 ASSERT(list_empty(&entry->log_list));
505 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
507 btrfs_add_delayed_iput(entry->inode);
508 while (!list_empty(&entry->list)) {
509 cur = entry->list.next;
510 sum = list_entry(cur, struct btrfs_ordered_sum, list);
511 list_del(&sum->list);
514 kmem_cache_free(btrfs_ordered_extent_cache, entry);
519 * remove an ordered extent from the tree. No references are dropped
520 * and waiters are woken up.
522 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
523 struct btrfs_ordered_extent *entry)
525 struct btrfs_ordered_inode_tree *tree;
526 struct btrfs_root *root = btrfs_inode->root;
527 struct btrfs_fs_info *fs_info = root->fs_info;
528 struct rb_node *node;
530 bool freespace_inode;
533 * If this is a free space inode the thread has not acquired the ordered
534 * extents lockdep map.
536 freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
538 btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
539 /* This is paired with btrfs_add_ordered_extent. */
540 spin_lock(&btrfs_inode->lock);
541 btrfs_mod_outstanding_extents(btrfs_inode, -1);
542 spin_unlock(&btrfs_inode->lock);
543 if (root != fs_info->tree_root) {
546 if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
547 release = entry->disk_num_bytes;
549 release = entry->num_bytes;
550 btrfs_delalloc_release_metadata(btrfs_inode, release, false);
553 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
554 fs_info->delalloc_batch);
556 tree = &btrfs_inode->ordered_tree;
557 spin_lock_irq(&tree->lock);
558 node = &entry->rb_node;
559 rb_erase(node, &tree->tree);
561 if (tree->last == node)
563 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
564 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
565 spin_unlock_irq(&tree->lock);
568 * The current running transaction is waiting on us, we need to let it
569 * know that we're complete and wake it up.
572 struct btrfs_transaction *trans;
575 * The checks for trans are just a formality, it should be set,
576 * but if it isn't we don't want to deref/assert under the spin
577 * lock, so be nice and check if trans is set, but ASSERT() so
578 * if it isn't set a developer will notice.
580 spin_lock(&fs_info->trans_lock);
581 trans = fs_info->running_transaction;
583 refcount_inc(&trans->use_count);
584 spin_unlock(&fs_info->trans_lock);
588 if (atomic_dec_and_test(&trans->pending_ordered))
589 wake_up(&trans->pending_wait);
590 btrfs_put_transaction(trans);
594 btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
596 spin_lock(&root->ordered_extent_lock);
597 list_del_init(&entry->root_extent_list);
598 root->nr_ordered_extents--;
600 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
602 if (!root->nr_ordered_extents) {
603 spin_lock(&fs_info->ordered_root_lock);
604 BUG_ON(list_empty(&root->ordered_root));
605 list_del_init(&root->ordered_root);
606 spin_unlock(&fs_info->ordered_root_lock);
608 spin_unlock(&root->ordered_extent_lock);
609 wake_up(&entry->wait);
610 if (!freespace_inode)
611 btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
614 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
616 struct btrfs_ordered_extent *ordered;
618 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
619 btrfs_start_ordered_extent(ordered, 1);
620 complete(&ordered->completion);
624 * wait for all the ordered extents in a root. This is done when balancing
625 * space between drives.
627 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
628 const u64 range_start, const u64 range_len)
630 struct btrfs_fs_info *fs_info = root->fs_info;
634 struct btrfs_ordered_extent *ordered, *next;
636 const u64 range_end = range_start + range_len;
638 mutex_lock(&root->ordered_extent_mutex);
639 spin_lock(&root->ordered_extent_lock);
640 list_splice_init(&root->ordered_extents, &splice);
641 while (!list_empty(&splice) && nr) {
642 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
645 if (range_end <= ordered->disk_bytenr ||
646 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
647 list_move_tail(&ordered->root_extent_list, &skipped);
648 cond_resched_lock(&root->ordered_extent_lock);
652 list_move_tail(&ordered->root_extent_list,
653 &root->ordered_extents);
654 refcount_inc(&ordered->refs);
655 spin_unlock(&root->ordered_extent_lock);
657 btrfs_init_work(&ordered->flush_work,
658 btrfs_run_ordered_extent_work, NULL, NULL);
659 list_add_tail(&ordered->work_list, &works);
660 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
663 spin_lock(&root->ordered_extent_lock);
668 list_splice_tail(&skipped, &root->ordered_extents);
669 list_splice_tail(&splice, &root->ordered_extents);
670 spin_unlock(&root->ordered_extent_lock);
672 list_for_each_entry_safe(ordered, next, &works, work_list) {
673 list_del_init(&ordered->work_list);
674 wait_for_completion(&ordered->completion);
675 btrfs_put_ordered_extent(ordered);
678 mutex_unlock(&root->ordered_extent_mutex);
683 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
684 const u64 range_start, const u64 range_len)
686 struct btrfs_root *root;
687 struct list_head splice;
690 INIT_LIST_HEAD(&splice);
692 mutex_lock(&fs_info->ordered_operations_mutex);
693 spin_lock(&fs_info->ordered_root_lock);
694 list_splice_init(&fs_info->ordered_roots, &splice);
695 while (!list_empty(&splice) && nr) {
696 root = list_first_entry(&splice, struct btrfs_root,
698 root = btrfs_grab_root(root);
700 list_move_tail(&root->ordered_root,
701 &fs_info->ordered_roots);
702 spin_unlock(&fs_info->ordered_root_lock);
704 done = btrfs_wait_ordered_extents(root, nr,
705 range_start, range_len);
706 btrfs_put_root(root);
708 spin_lock(&fs_info->ordered_root_lock);
713 list_splice_tail(&splice, &fs_info->ordered_roots);
714 spin_unlock(&fs_info->ordered_root_lock);
715 mutex_unlock(&fs_info->ordered_operations_mutex);
719 * Used to start IO or wait for a given ordered extent to finish.
721 * If wait is one, this effectively waits on page writeback for all the pages
722 * in the extent, and it waits on the io completion code to insert
723 * metadata into the btree corresponding to the extent
725 void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
727 u64 start = entry->file_offset;
728 u64 end = start + entry->num_bytes - 1;
729 struct btrfs_inode *inode = BTRFS_I(entry->inode);
730 bool freespace_inode;
732 trace_btrfs_ordered_extent_start(inode, entry);
735 * If this is a free space inode do not take the ordered extents lockdep
738 freespace_inode = btrfs_is_free_space_inode(inode);
741 * pages in the range can be dirty, clean or writeback. We
742 * start IO on any dirty ones so the wait doesn't stall waiting
743 * for the flusher thread to find them
745 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
746 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
748 if (!freespace_inode)
749 btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
750 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
756 * Used to wait on ordered extents across a large range of bytes.
758 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
764 struct btrfs_ordered_extent *ordered;
766 if (start + len < start) {
767 orig_end = INT_LIMIT(loff_t);
769 orig_end = start + len - 1;
770 if (orig_end > INT_LIMIT(loff_t))
771 orig_end = INT_LIMIT(loff_t);
774 /* start IO across the range first to instantiate any delalloc
777 ret = btrfs_fdatawrite_range(inode, start, orig_end);
782 * If we have a writeback error don't return immediately. Wait first
783 * for any ordered extents that haven't completed yet. This is to make
784 * sure no one can dirty the same page ranges and call writepages()
785 * before the ordered extents complete - to avoid failures (-EEXIST)
786 * when adding the new ordered extents to the ordered tree.
788 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
792 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
795 if (ordered->file_offset > orig_end) {
796 btrfs_put_ordered_extent(ordered);
799 if (ordered->file_offset + ordered->num_bytes <= start) {
800 btrfs_put_ordered_extent(ordered);
803 btrfs_start_ordered_extent(ordered, 1);
804 end = ordered->file_offset;
806 * If the ordered extent had an error save the error but don't
807 * exit without waiting first for all other ordered extents in
808 * the range to complete.
810 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
812 btrfs_put_ordered_extent(ordered);
813 if (end == 0 || end == start)
817 return ret_wb ? ret_wb : ret;
821 * find an ordered extent corresponding to file_offset. return NULL if
822 * nothing is found, otherwise take a reference on the extent and return it
824 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
827 struct btrfs_ordered_inode_tree *tree;
828 struct rb_node *node;
829 struct btrfs_ordered_extent *entry = NULL;
832 tree = &inode->ordered_tree;
833 spin_lock_irqsave(&tree->lock, flags);
834 node = tree_search(tree, file_offset);
838 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
839 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
842 refcount_inc(&entry->refs);
843 trace_btrfs_ordered_extent_lookup(inode, entry);
846 spin_unlock_irqrestore(&tree->lock, flags);
850 /* Since the DIO code tries to lock a wide area we need to look for any ordered
851 * extents that exist in the range, rather than just the start of the range.
853 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
854 struct btrfs_inode *inode, u64 file_offset, u64 len)
856 struct btrfs_ordered_inode_tree *tree;
857 struct rb_node *node;
858 struct btrfs_ordered_extent *entry = NULL;
860 tree = &inode->ordered_tree;
861 spin_lock_irq(&tree->lock);
862 node = tree_search(tree, file_offset);
864 node = tree_search(tree, file_offset + len);
870 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
871 if (range_overlaps(entry, file_offset, len))
874 if (entry->file_offset >= file_offset + len) {
879 node = rb_next(node);
885 refcount_inc(&entry->refs);
886 trace_btrfs_ordered_extent_lookup_range(inode, entry);
888 spin_unlock_irq(&tree->lock);
893 * Adds all ordered extents to the given list. The list ends up sorted by the
894 * file_offset of the ordered extents.
896 void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
897 struct list_head *list)
899 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
902 ASSERT(inode_is_locked(&inode->vfs_inode));
904 spin_lock_irq(&tree->lock);
905 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
906 struct btrfs_ordered_extent *ordered;
908 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
910 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
913 ASSERT(list_empty(&ordered->log_list));
914 list_add_tail(&ordered->log_list, list);
915 refcount_inc(&ordered->refs);
916 trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
918 spin_unlock_irq(&tree->lock);
922 * lookup and return any extent before 'file_offset'. NULL is returned
925 struct btrfs_ordered_extent *
926 btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
928 struct btrfs_ordered_inode_tree *tree;
929 struct rb_node *node;
930 struct btrfs_ordered_extent *entry = NULL;
932 tree = &inode->ordered_tree;
933 spin_lock_irq(&tree->lock);
934 node = tree_search(tree, file_offset);
938 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
939 refcount_inc(&entry->refs);
940 trace_btrfs_ordered_extent_lookup_first(inode, entry);
942 spin_unlock_irq(&tree->lock);
947 * Lookup the first ordered extent that overlaps the range
948 * [@file_offset, @file_offset + @len).
950 * The difference between this and btrfs_lookup_first_ordered_extent() is
951 * that this one won't return any ordered extent that does not overlap the range.
952 * And the difference against btrfs_lookup_ordered_extent() is, this function
953 * ensures the first ordered extent gets returned.
955 struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
956 struct btrfs_inode *inode, u64 file_offset, u64 len)
958 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
959 struct rb_node *node;
961 struct rb_node *prev;
962 struct rb_node *next;
963 struct btrfs_ordered_extent *entry = NULL;
965 spin_lock_irq(&tree->lock);
966 node = tree->tree.rb_node;
968 * Here we don't want to use tree_search() which will use tree->last
969 * and screw up the search order.
970 * And __tree_search() can't return the adjacent ordered extents
971 * either, thus here we do our own search.
974 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
976 if (file_offset < entry->file_offset) {
977 node = node->rb_left;
978 } else if (file_offset >= entry_end(entry)) {
979 node = node->rb_right;
982 * Direct hit, got an ordered extent that starts at
993 cur = &entry->rb_node;
994 /* We got an entry around @file_offset, check adjacent entries */
995 if (entry->file_offset < file_offset) {
1003 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1004 if (range_overlaps(entry, file_offset, len))
1008 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1009 if (range_overlaps(entry, file_offset, len))
1012 /* No ordered extent in the range */
1016 refcount_inc(&entry->refs);
1017 trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1020 spin_unlock_irq(&tree->lock);
1025 * Lock the passed range and ensures all pending ordered extents in it are run
1028 * @inode: Inode whose ordered tree is to be searched
1029 * @start: Beginning of range to flush
1030 * @end: Last byte of range to lock
1031 * @cached_state: If passed, will return the extent state responsible for the
1032 * locked range. It's the caller's responsibility to free the
1035 * Always return with the given range locked, ensuring after it's called no
1036 * order extent can be pending.
1038 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1040 struct extent_state **cached_state)
1042 struct btrfs_ordered_extent *ordered;
1043 struct extent_state *cache = NULL;
1044 struct extent_state **cachedp = &cache;
1047 cachedp = cached_state;
1050 lock_extent(&inode->io_tree, start, end, cachedp);
1051 ordered = btrfs_lookup_ordered_range(inode, start,
1055 * If no external cached_state has been passed then
1056 * decrement the extra ref taken for cachedp since we
1057 * aren't exposing it outside of this function
1060 refcount_dec(&cache->refs);
1063 unlock_extent(&inode->io_tree, start, end, cachedp);
1064 btrfs_start_ordered_extent(ordered, 1);
1065 btrfs_put_ordered_extent(ordered);
1070 * Lock the passed range and ensure all pending ordered extents in it are run
1071 * to completion in nowait mode.
1073 * Return true if btrfs_lock_ordered_range does not return any extents,
1076 bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1077 struct extent_state **cached_state)
1079 struct btrfs_ordered_extent *ordered;
1081 if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1084 ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1088 btrfs_put_ordered_extent(ordered);
1089 unlock_extent(&inode->io_tree, start, end, cached_state);
1095 static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1098 struct inode *inode = ordered->inode;
1099 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1100 u64 file_offset = ordered->file_offset + pos;
1101 u64 disk_bytenr = ordered->disk_bytenr + pos;
1102 unsigned long flags = ordered->flags & BTRFS_ORDERED_TYPE_FLAGS;
1105 * The splitting extent is already counted and will be added again in
1106 * btrfs_add_ordered_extent_*(). Subtract len to avoid double counting.
1108 percpu_counter_add_batch(&fs_info->ordered_bytes, -len,
1109 fs_info->delalloc_batch);
1110 WARN_ON_ONCE(flags & (1 << BTRFS_ORDERED_COMPRESSED));
1111 return btrfs_add_ordered_extent(BTRFS_I(inode), file_offset, len, len,
1112 disk_bytenr, len, 0, flags,
1113 ordered->compress_type);
1116 int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1119 struct inode *inode = ordered->inode;
1120 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1121 struct rb_node *node;
1122 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1125 trace_btrfs_ordered_extent_split(BTRFS_I(inode), ordered);
1127 spin_lock_irq(&tree->lock);
1128 /* Remove from tree once */
1129 node = &ordered->rb_node;
1130 rb_erase(node, &tree->tree);
1131 RB_CLEAR_NODE(node);
1132 if (tree->last == node)
1135 ordered->file_offset += pre;
1136 ordered->disk_bytenr += pre;
1137 ordered->num_bytes -= (pre + post);
1138 ordered->disk_num_bytes -= (pre + post);
1139 ordered->bytes_left -= (pre + post);
1141 /* Re-insert the node */
1142 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1144 btrfs_panic(fs_info, -EEXIST,
1145 "zoned: inconsistency in ordered tree at offset %llu",
1146 ordered->file_offset);
1148 spin_unlock_irq(&tree->lock);
1151 ret = clone_ordered_extent(ordered, 0, pre);
1152 if (ret == 0 && post)
1153 ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1159 int __init ordered_data_init(void)
1161 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1162 sizeof(struct btrfs_ordered_extent), 0,
1165 if (!btrfs_ordered_extent_cache)
1171 void __cold ordered_data_exit(void)
1173 kmem_cache_destroy(btrfs_ordered_extent_cache);