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"
21 static struct kmem_cache *btrfs_ordered_extent_cache;
23 static u64 entry_end(struct btrfs_ordered_extent *entry)
25 if (entry->file_offset + entry->num_bytes < entry->file_offset)
27 return entry->file_offset + entry->num_bytes;
30 /* returns NULL if the insertion worked, or it returns the node it did find
33 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
36 struct rb_node **p = &root->rb_node;
37 struct rb_node *parent = NULL;
38 struct btrfs_ordered_extent *entry;
42 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
44 if (file_offset < entry->file_offset)
46 else if (file_offset >= entry_end(entry))
52 rb_link_node(node, parent, p);
53 rb_insert_color(node, root);
58 * look for a given offset in the tree, and if it can't be found return the
61 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
62 struct rb_node **prev_ret)
64 struct rb_node *n = root->rb_node;
65 struct rb_node *prev = NULL;
67 struct btrfs_ordered_extent *entry;
68 struct btrfs_ordered_extent *prev_entry = NULL;
71 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
75 if (file_offset < entry->file_offset)
77 else if (file_offset >= entry_end(entry))
85 while (prev && file_offset >= entry_end(prev_entry)) {
89 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
91 if (file_offset < entry_end(prev_entry))
97 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
99 while (prev && file_offset < entry_end(prev_entry)) {
100 test = rb_prev(prev);
103 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
111 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
114 if (file_offset + len <= entry->file_offset ||
115 entry->file_offset + entry->num_bytes <= file_offset)
121 * look find the first ordered struct that has this offset, otherwise
122 * the first one less than this offset
124 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
127 struct rb_root *root = &tree->tree;
128 struct rb_node *prev = NULL;
130 struct btrfs_ordered_extent *entry;
133 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
135 if (in_range(file_offset, entry->file_offset, entry->num_bytes))
138 ret = __tree_search(root, file_offset, &prev);
147 * Allocate and add a new ordered_extent into the per-inode tree.
149 * The tree is given a single reference on the ordered extent that was
152 static int __btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
153 u64 disk_bytenr, u64 num_bytes,
154 u64 disk_num_bytes, int type, int dio,
157 struct btrfs_root *root = inode->root;
158 struct btrfs_fs_info *fs_info = root->fs_info;
159 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
160 struct rb_node *node;
161 struct btrfs_ordered_extent *entry;
164 if (type == BTRFS_ORDERED_NOCOW || type == BTRFS_ORDERED_PREALLOC) {
165 /* For nocow write, we can release the qgroup rsv right now */
166 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
172 * The ordered extent has reserved qgroup space, release now
173 * and pass the reserved number for qgroup_record to free.
175 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
179 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
183 entry->file_offset = file_offset;
184 entry->disk_bytenr = disk_bytenr;
185 entry->num_bytes = num_bytes;
186 entry->disk_num_bytes = disk_num_bytes;
187 entry->bytes_left = num_bytes;
188 entry->inode = igrab(&inode->vfs_inode);
189 entry->compress_type = compress_type;
190 entry->truncated_len = (u64)-1;
191 entry->qgroup_rsv = ret;
192 entry->physical = (u64)-1;
194 entry->partno = (u8)-1;
196 ASSERT(type == BTRFS_ORDERED_REGULAR ||
197 type == BTRFS_ORDERED_NOCOW ||
198 type == BTRFS_ORDERED_PREALLOC ||
199 type == BTRFS_ORDERED_COMPRESSED);
200 set_bit(type, &entry->flags);
202 percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
203 fs_info->delalloc_batch);
206 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
208 /* one ref for the tree */
209 refcount_set(&entry->refs, 1);
210 init_waitqueue_head(&entry->wait);
211 INIT_LIST_HEAD(&entry->list);
212 INIT_LIST_HEAD(&entry->log_list);
213 INIT_LIST_HEAD(&entry->root_extent_list);
214 INIT_LIST_HEAD(&entry->work_list);
215 init_completion(&entry->completion);
217 trace_btrfs_ordered_extent_add(inode, entry);
219 spin_lock_irq(&tree->lock);
220 node = tree_insert(&tree->tree, file_offset,
223 btrfs_panic(fs_info, -EEXIST,
224 "inconsistency in ordered tree at offset %llu",
226 spin_unlock_irq(&tree->lock);
228 spin_lock(&root->ordered_extent_lock);
229 list_add_tail(&entry->root_extent_list,
230 &root->ordered_extents);
231 root->nr_ordered_extents++;
232 if (root->nr_ordered_extents == 1) {
233 spin_lock(&fs_info->ordered_root_lock);
234 BUG_ON(!list_empty(&root->ordered_root));
235 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
236 spin_unlock(&fs_info->ordered_root_lock);
238 spin_unlock(&root->ordered_extent_lock);
241 * We don't need the count_max_extents here, we can assume that all of
242 * that work has been done at higher layers, so this is truly the
243 * smallest the extent is going to get.
245 spin_lock(&inode->lock);
246 btrfs_mod_outstanding_extents(inode, 1);
247 spin_unlock(&inode->lock);
252 int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
253 u64 disk_bytenr, u64 num_bytes, u64 disk_num_bytes,
256 ASSERT(type == BTRFS_ORDERED_REGULAR ||
257 type == BTRFS_ORDERED_NOCOW ||
258 type == BTRFS_ORDERED_PREALLOC);
259 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
260 num_bytes, disk_num_bytes, type, 0,
261 BTRFS_COMPRESS_NONE);
264 int btrfs_add_ordered_extent_dio(struct btrfs_inode *inode, u64 file_offset,
265 u64 disk_bytenr, u64 num_bytes,
266 u64 disk_num_bytes, int type)
268 ASSERT(type == BTRFS_ORDERED_REGULAR ||
269 type == BTRFS_ORDERED_NOCOW ||
270 type == BTRFS_ORDERED_PREALLOC);
271 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
272 num_bytes, disk_num_bytes, type, 1,
273 BTRFS_COMPRESS_NONE);
276 int btrfs_add_ordered_extent_compress(struct btrfs_inode *inode, u64 file_offset,
277 u64 disk_bytenr, u64 num_bytes,
278 u64 disk_num_bytes, int compress_type)
280 ASSERT(compress_type != BTRFS_COMPRESS_NONE);
281 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
282 num_bytes, disk_num_bytes,
283 BTRFS_ORDERED_COMPRESSED, 0,
288 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
289 * when an ordered extent is finished. If the list covers more than one
290 * ordered extent, it is split across multiples.
292 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
293 struct btrfs_ordered_sum *sum)
295 struct btrfs_ordered_inode_tree *tree;
297 tree = &BTRFS_I(entry->inode)->ordered_tree;
298 spin_lock_irq(&tree->lock);
299 list_add_tail(&sum->list, &entry->list);
300 spin_unlock_irq(&tree->lock);
304 * Mark all ordered extents io inside the specified range finished.
306 * @page: The invovled page for the opeartion.
307 * For uncompressed buffered IO, the page status also needs to be
308 * updated to indicate whether the pending ordered io is finished.
309 * Can be NULL for direct IO and compressed write.
310 * For these cases, callers are ensured they won't execute the
311 * endio function twice.
312 * @finish_func: The function to be executed when all the IO of an ordered
313 * extent are finished.
315 * This function is called for endio, thus the range must have ordered
316 * extent(s) coveri it.
318 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
319 struct page *page, u64 file_offset,
320 u64 num_bytes, btrfs_func_t finish_func,
323 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
324 struct btrfs_fs_info *fs_info = inode->root->fs_info;
325 struct btrfs_workqueue *wq;
326 struct rb_node *node;
327 struct btrfs_ordered_extent *entry = NULL;
329 u64 cur = file_offset;
331 if (btrfs_is_free_space_inode(inode))
332 wq = fs_info->endio_freespace_worker;
334 wq = fs_info->endio_write_workers;
337 ASSERT(page->mapping && page_offset(page) <= file_offset &&
338 file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
340 spin_lock_irqsave(&tree->lock, flags);
341 while (cur < file_offset + num_bytes) {
346 node = tree_search(tree, cur);
347 /* No ordered extents at all */
351 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
352 entry_end = entry->file_offset + entry->num_bytes;
358 if (cur >= entry_end) {
359 node = rb_next(node);
360 /* No more ordered extents, exit */
363 entry = rb_entry(node, struct btrfs_ordered_extent,
366 /* Go to next ordered extent and continue */
367 cur = entry->file_offset;
373 * Go to the start of OE.
375 if (cur < entry->file_offset) {
376 cur = entry->file_offset;
381 * Now we are definitely inside one ordered extent.
387 end = min(entry->file_offset + entry->num_bytes,
388 file_offset + num_bytes) - 1;
389 ASSERT(end + 1 - cur < U32_MAX);
394 * Ordered (Private2) bit indicates whether we still
395 * have pending io unfinished for the ordered extent.
397 * If there's no such bit, we need to skip to next range.
399 if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
403 btrfs_page_clear_ordered(fs_info, page, cur, len);
406 /* Now we're fine to update the accounting */
407 if (unlikely(len > entry->bytes_left)) {
410 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
411 inode->root->root_key.objectid,
415 len, entry->bytes_left);
416 entry->bytes_left = 0;
418 entry->bytes_left -= len;
422 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
425 * All the IO of the ordered extent is finished, we need to queue
426 * the finish_func to be executed.
428 if (entry->bytes_left == 0) {
429 set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
430 cond_wake_up(&entry->wait);
431 refcount_inc(&entry->refs);
432 spin_unlock_irqrestore(&tree->lock, flags);
433 btrfs_init_work(&entry->work, finish_func, NULL, NULL);
434 btrfs_queue_work(wq, &entry->work);
435 spin_lock_irqsave(&tree->lock, flags);
439 spin_unlock_irqrestore(&tree->lock, flags);
443 * Finish IO for one ordered extent across a given range. The range can only
444 * contain one ordered extent.
446 * @cached: The cached ordered extent. If not NULL, we can skip the tree
447 * search and use the ordered extent directly.
448 * Will be also used to store the finished ordered extent.
449 * @file_offset: File offset for the finished IO
450 * @io_size: Length of the finish IO range
451 * @uptodate: If the IO finishes without problem
453 * Return true if the ordered extent is finished in the range, and update
455 * Return false otherwise.
457 * NOTE: The range can NOT cross multiple ordered extents.
458 * Thus caller should ensure the range doesn't cross ordered extents.
460 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
461 struct btrfs_ordered_extent **cached,
462 u64 file_offset, u64 io_size, int uptodate)
464 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
465 struct rb_node *node;
466 struct btrfs_ordered_extent *entry = NULL;
468 bool finished = false;
470 spin_lock_irqsave(&tree->lock, flags);
471 if (cached && *cached) {
476 node = tree_search(tree, file_offset);
480 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
482 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
485 if (io_size > entry->bytes_left)
486 btrfs_crit(inode->root->fs_info,
487 "bad ordered accounting left %llu size %llu",
488 entry->bytes_left, io_size);
490 entry->bytes_left -= io_size;
492 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
494 if (entry->bytes_left == 0) {
496 * Ensure only one caller can set the flag and finished_ret
499 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
500 /* test_and_set_bit implies a barrier */
501 cond_wake_up_nomb(&entry->wait);
504 if (finished && cached && entry) {
506 refcount_inc(&entry->refs);
508 spin_unlock_irqrestore(&tree->lock, flags);
513 * used to drop a reference on an ordered extent. This will free
514 * the extent if the last reference is dropped
516 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
518 struct list_head *cur;
519 struct btrfs_ordered_sum *sum;
521 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
523 if (refcount_dec_and_test(&entry->refs)) {
524 ASSERT(list_empty(&entry->root_extent_list));
525 ASSERT(list_empty(&entry->log_list));
526 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
528 btrfs_add_delayed_iput(entry->inode);
529 while (!list_empty(&entry->list)) {
530 cur = entry->list.next;
531 sum = list_entry(cur, struct btrfs_ordered_sum, list);
532 list_del(&sum->list);
535 kmem_cache_free(btrfs_ordered_extent_cache, entry);
540 * remove an ordered extent from the tree. No references are dropped
541 * and waiters are woken up.
543 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
544 struct btrfs_ordered_extent *entry)
546 struct btrfs_ordered_inode_tree *tree;
547 struct btrfs_root *root = btrfs_inode->root;
548 struct btrfs_fs_info *fs_info = root->fs_info;
549 struct rb_node *node;
552 /* This is paired with btrfs_add_ordered_extent. */
553 spin_lock(&btrfs_inode->lock);
554 btrfs_mod_outstanding_extents(btrfs_inode, -1);
555 spin_unlock(&btrfs_inode->lock);
556 if (root != fs_info->tree_root)
557 btrfs_delalloc_release_metadata(btrfs_inode, entry->num_bytes,
560 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
561 fs_info->delalloc_batch);
563 tree = &btrfs_inode->ordered_tree;
564 spin_lock_irq(&tree->lock);
565 node = &entry->rb_node;
566 rb_erase(node, &tree->tree);
568 if (tree->last == node)
570 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
571 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
572 spin_unlock_irq(&tree->lock);
575 * The current running transaction is waiting on us, we need to let it
576 * know that we're complete and wake it up.
579 struct btrfs_transaction *trans;
582 * The checks for trans are just a formality, it should be set,
583 * but if it isn't we don't want to deref/assert under the spin
584 * lock, so be nice and check if trans is set, but ASSERT() so
585 * if it isn't set a developer will notice.
587 spin_lock(&fs_info->trans_lock);
588 trans = fs_info->running_transaction;
590 refcount_inc(&trans->use_count);
591 spin_unlock(&fs_info->trans_lock);
595 if (atomic_dec_and_test(&trans->pending_ordered))
596 wake_up(&trans->pending_wait);
597 btrfs_put_transaction(trans);
601 spin_lock(&root->ordered_extent_lock);
602 list_del_init(&entry->root_extent_list);
603 root->nr_ordered_extents--;
605 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
607 if (!root->nr_ordered_extents) {
608 spin_lock(&fs_info->ordered_root_lock);
609 BUG_ON(list_empty(&root->ordered_root));
610 list_del_init(&root->ordered_root);
611 spin_unlock(&fs_info->ordered_root_lock);
613 spin_unlock(&root->ordered_extent_lock);
614 wake_up(&entry->wait);
617 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
619 struct btrfs_ordered_extent *ordered;
621 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
622 btrfs_start_ordered_extent(ordered, 1);
623 complete(&ordered->completion);
627 * wait for all the ordered extents in a root. This is done when balancing
628 * space between drives.
630 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
631 const u64 range_start, const u64 range_len)
633 struct btrfs_fs_info *fs_info = root->fs_info;
637 struct btrfs_ordered_extent *ordered, *next;
639 const u64 range_end = range_start + range_len;
641 mutex_lock(&root->ordered_extent_mutex);
642 spin_lock(&root->ordered_extent_lock);
643 list_splice_init(&root->ordered_extents, &splice);
644 while (!list_empty(&splice) && nr) {
645 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
648 if (range_end <= ordered->disk_bytenr ||
649 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
650 list_move_tail(&ordered->root_extent_list, &skipped);
651 cond_resched_lock(&root->ordered_extent_lock);
655 list_move_tail(&ordered->root_extent_list,
656 &root->ordered_extents);
657 refcount_inc(&ordered->refs);
658 spin_unlock(&root->ordered_extent_lock);
660 btrfs_init_work(&ordered->flush_work,
661 btrfs_run_ordered_extent_work, NULL, NULL);
662 list_add_tail(&ordered->work_list, &works);
663 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
666 spin_lock(&root->ordered_extent_lock);
671 list_splice_tail(&skipped, &root->ordered_extents);
672 list_splice_tail(&splice, &root->ordered_extents);
673 spin_unlock(&root->ordered_extent_lock);
675 list_for_each_entry_safe(ordered, next, &works, work_list) {
676 list_del_init(&ordered->work_list);
677 wait_for_completion(&ordered->completion);
678 btrfs_put_ordered_extent(ordered);
681 mutex_unlock(&root->ordered_extent_mutex);
686 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
687 const u64 range_start, const u64 range_len)
689 struct btrfs_root *root;
690 struct list_head splice;
693 INIT_LIST_HEAD(&splice);
695 mutex_lock(&fs_info->ordered_operations_mutex);
696 spin_lock(&fs_info->ordered_root_lock);
697 list_splice_init(&fs_info->ordered_roots, &splice);
698 while (!list_empty(&splice) && nr) {
699 root = list_first_entry(&splice, struct btrfs_root,
701 root = btrfs_grab_root(root);
703 list_move_tail(&root->ordered_root,
704 &fs_info->ordered_roots);
705 spin_unlock(&fs_info->ordered_root_lock);
707 done = btrfs_wait_ordered_extents(root, nr,
708 range_start, range_len);
709 btrfs_put_root(root);
711 spin_lock(&fs_info->ordered_root_lock);
716 list_splice_tail(&splice, &fs_info->ordered_roots);
717 spin_unlock(&fs_info->ordered_root_lock);
718 mutex_unlock(&fs_info->ordered_operations_mutex);
722 * Used to start IO or wait for a given ordered extent to finish.
724 * If wait is one, this effectively waits on page writeback for all the pages
725 * in the extent, and it waits on the io completion code to insert
726 * metadata into the btree corresponding to the extent
728 void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
730 u64 start = entry->file_offset;
731 u64 end = start + entry->num_bytes - 1;
732 struct btrfs_inode *inode = BTRFS_I(entry->inode);
734 trace_btrfs_ordered_extent_start(inode, entry);
737 * pages in the range can be dirty, clean or writeback. We
738 * start IO on any dirty ones so the wait doesn't stall waiting
739 * for the flusher thread to find them
741 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
742 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
744 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
750 * Used to wait on ordered extents across a large range of bytes.
752 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
758 struct btrfs_ordered_extent *ordered;
760 if (start + len < start) {
761 orig_end = INT_LIMIT(loff_t);
763 orig_end = start + len - 1;
764 if (orig_end > INT_LIMIT(loff_t))
765 orig_end = INT_LIMIT(loff_t);
768 /* start IO across the range first to instantiate any delalloc
771 ret = btrfs_fdatawrite_range(inode, start, orig_end);
776 * If we have a writeback error don't return immediately. Wait first
777 * for any ordered extents that haven't completed yet. This is to make
778 * sure no one can dirty the same page ranges and call writepages()
779 * before the ordered extents complete - to avoid failures (-EEXIST)
780 * when adding the new ordered extents to the ordered tree.
782 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
786 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
789 if (ordered->file_offset > orig_end) {
790 btrfs_put_ordered_extent(ordered);
793 if (ordered->file_offset + ordered->num_bytes <= start) {
794 btrfs_put_ordered_extent(ordered);
797 btrfs_start_ordered_extent(ordered, 1);
798 end = ordered->file_offset;
800 * If the ordered extent had an error save the error but don't
801 * exit without waiting first for all other ordered extents in
802 * the range to complete.
804 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
806 btrfs_put_ordered_extent(ordered);
807 if (end == 0 || end == start)
811 return ret_wb ? ret_wb : ret;
815 * find an ordered extent corresponding to file_offset. return NULL if
816 * nothing is found, otherwise take a reference on the extent and return it
818 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
821 struct btrfs_ordered_inode_tree *tree;
822 struct rb_node *node;
823 struct btrfs_ordered_extent *entry = NULL;
826 tree = &inode->ordered_tree;
827 spin_lock_irqsave(&tree->lock, flags);
828 node = tree_search(tree, file_offset);
832 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
833 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
836 refcount_inc(&entry->refs);
838 spin_unlock_irqrestore(&tree->lock, flags);
842 /* Since the DIO code tries to lock a wide area we need to look for any ordered
843 * extents that exist in the range, rather than just the start of the range.
845 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
846 struct btrfs_inode *inode, u64 file_offset, u64 len)
848 struct btrfs_ordered_inode_tree *tree;
849 struct rb_node *node;
850 struct btrfs_ordered_extent *entry = NULL;
852 tree = &inode->ordered_tree;
853 spin_lock_irq(&tree->lock);
854 node = tree_search(tree, file_offset);
856 node = tree_search(tree, file_offset + len);
862 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
863 if (range_overlaps(entry, file_offset, len))
866 if (entry->file_offset >= file_offset + len) {
871 node = rb_next(node);
877 refcount_inc(&entry->refs);
878 spin_unlock_irq(&tree->lock);
883 * Adds all ordered extents to the given list. The list ends up sorted by the
884 * file_offset of the ordered extents.
886 void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
887 struct list_head *list)
889 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
892 ASSERT(inode_is_locked(&inode->vfs_inode));
894 spin_lock_irq(&tree->lock);
895 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
896 struct btrfs_ordered_extent *ordered;
898 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
900 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
903 ASSERT(list_empty(&ordered->log_list));
904 list_add_tail(&ordered->log_list, list);
905 refcount_inc(&ordered->refs);
907 spin_unlock_irq(&tree->lock);
911 * lookup and return any extent before 'file_offset'. NULL is returned
914 struct btrfs_ordered_extent *
915 btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
917 struct btrfs_ordered_inode_tree *tree;
918 struct rb_node *node;
919 struct btrfs_ordered_extent *entry = NULL;
921 tree = &inode->ordered_tree;
922 spin_lock_irq(&tree->lock);
923 node = tree_search(tree, file_offset);
927 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
928 refcount_inc(&entry->refs);
930 spin_unlock_irq(&tree->lock);
935 * Lookup the first ordered extent that overlaps the range
936 * [@file_offset, @file_offset + @len).
938 * The difference between this and btrfs_lookup_first_ordered_extent() is
939 * that this one won't return any ordered extent that does not overlap the range.
940 * And the difference against btrfs_lookup_ordered_extent() is, this function
941 * ensures the first ordered extent gets returned.
943 struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
944 struct btrfs_inode *inode, u64 file_offset, u64 len)
946 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
947 struct rb_node *node;
949 struct rb_node *prev;
950 struct rb_node *next;
951 struct btrfs_ordered_extent *entry = NULL;
953 spin_lock_irq(&tree->lock);
954 node = tree->tree.rb_node;
956 * Here we don't want to use tree_search() which will use tree->last
957 * and screw up the search order.
958 * And __tree_search() can't return the adjacent ordered extents
959 * either, thus here we do our own search.
962 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
964 if (file_offset < entry->file_offset) {
965 node = node->rb_left;
966 } else if (file_offset >= entry_end(entry)) {
967 node = node->rb_right;
970 * Direct hit, got an ordered extent that starts at
981 cur = &entry->rb_node;
982 /* We got an entry around @file_offset, check adjacent entries */
983 if (entry->file_offset < file_offset) {
991 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
992 if (range_overlaps(entry, file_offset, len))
996 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
997 if (range_overlaps(entry, file_offset, len))
1000 /* No ordered extent in the range */
1004 refcount_inc(&entry->refs);
1005 spin_unlock_irq(&tree->lock);
1010 * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
1011 * ordered extents in it are run to completion.
1013 * @inode: Inode whose ordered tree is to be searched
1014 * @start: Beginning of range to flush
1015 * @end: Last byte of range to lock
1016 * @cached_state: If passed, will return the extent state responsible for the
1017 * locked range. It's the caller's responsibility to free the cached state.
1019 * This function always returns with the given range locked, ensuring after it's
1020 * called no order extent can be pending.
1022 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1024 struct extent_state **cached_state)
1026 struct btrfs_ordered_extent *ordered;
1027 struct extent_state *cache = NULL;
1028 struct extent_state **cachedp = &cache;
1031 cachedp = cached_state;
1034 lock_extent_bits(&inode->io_tree, start, end, cachedp);
1035 ordered = btrfs_lookup_ordered_range(inode, start,
1039 * If no external cached_state has been passed then
1040 * decrement the extra ref taken for cachedp since we
1041 * aren't exposing it outside of this function
1044 refcount_dec(&cache->refs);
1047 unlock_extent_cached(&inode->io_tree, start, end, cachedp);
1048 btrfs_start_ordered_extent(ordered, 1);
1049 btrfs_put_ordered_extent(ordered);
1053 static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1056 struct inode *inode = ordered->inode;
1057 u64 file_offset = ordered->file_offset + pos;
1058 u64 disk_bytenr = ordered->disk_bytenr + pos;
1059 u64 num_bytes = len;
1060 u64 disk_num_bytes = len;
1062 unsigned long flags_masked = ordered->flags & ~(1 << BTRFS_ORDERED_DIRECT);
1063 int compress_type = ordered->compress_type;
1064 unsigned long weight;
1067 weight = hweight_long(flags_masked);
1068 WARN_ON_ONCE(weight > 1);
1072 type = __ffs(flags_masked);
1074 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered->flags)) {
1076 ret = btrfs_add_ordered_extent_compress(BTRFS_I(inode),
1077 file_offset, disk_bytenr, num_bytes,
1078 disk_num_bytes, compress_type);
1079 } else if (test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
1080 ret = btrfs_add_ordered_extent_dio(BTRFS_I(inode), file_offset,
1081 disk_bytenr, num_bytes, disk_num_bytes, type);
1083 ret = btrfs_add_ordered_extent(BTRFS_I(inode), file_offset,
1084 disk_bytenr, num_bytes, disk_num_bytes, type);
1090 int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1093 struct inode *inode = ordered->inode;
1094 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1095 struct rb_node *node;
1096 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1099 spin_lock_irq(&tree->lock);
1100 /* Remove from tree once */
1101 node = &ordered->rb_node;
1102 rb_erase(node, &tree->tree);
1103 RB_CLEAR_NODE(node);
1104 if (tree->last == node)
1107 ordered->file_offset += pre;
1108 ordered->disk_bytenr += pre;
1109 ordered->num_bytes -= (pre + post);
1110 ordered->disk_num_bytes -= (pre + post);
1111 ordered->bytes_left -= (pre + post);
1113 /* Re-insert the node */
1114 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1116 btrfs_panic(fs_info, -EEXIST,
1117 "zoned: inconsistency in ordered tree at offset %llu",
1118 ordered->file_offset);
1120 spin_unlock_irq(&tree->lock);
1123 ret = clone_ordered_extent(ordered, 0, pre);
1124 if (ret == 0 && post)
1125 ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1131 int __init ordered_data_init(void)
1133 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1134 sizeof(struct btrfs_ordered_extent), 0,
1137 if (!btrfs_ordered_extent_cache)
1143 void __cold ordered_data_exit(void)
1145 kmem_cache_destroy(btrfs_ordered_extent_cache);