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 ASSERT(type == BTRFS_ORDERED_REGULAR ||
195 type == BTRFS_ORDERED_NOCOW ||
196 type == BTRFS_ORDERED_PREALLOC ||
197 type == BTRFS_ORDERED_COMPRESSED);
198 set_bit(type, &entry->flags);
200 percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
201 fs_info->delalloc_batch);
204 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
206 /* one ref for the tree */
207 refcount_set(&entry->refs, 1);
208 init_waitqueue_head(&entry->wait);
209 INIT_LIST_HEAD(&entry->list);
210 INIT_LIST_HEAD(&entry->log_list);
211 INIT_LIST_HEAD(&entry->root_extent_list);
212 INIT_LIST_HEAD(&entry->work_list);
213 init_completion(&entry->completion);
215 trace_btrfs_ordered_extent_add(inode, entry);
217 spin_lock_irq(&tree->lock);
218 node = tree_insert(&tree->tree, file_offset,
221 btrfs_panic(fs_info, -EEXIST,
222 "inconsistency in ordered tree at offset %llu",
224 spin_unlock_irq(&tree->lock);
226 spin_lock(&root->ordered_extent_lock);
227 list_add_tail(&entry->root_extent_list,
228 &root->ordered_extents);
229 root->nr_ordered_extents++;
230 if (root->nr_ordered_extents == 1) {
231 spin_lock(&fs_info->ordered_root_lock);
232 BUG_ON(!list_empty(&root->ordered_root));
233 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
234 spin_unlock(&fs_info->ordered_root_lock);
236 spin_unlock(&root->ordered_extent_lock);
239 * We don't need the count_max_extents here, we can assume that all of
240 * that work has been done at higher layers, so this is truly the
241 * smallest the extent is going to get.
243 spin_lock(&inode->lock);
244 btrfs_mod_outstanding_extents(inode, 1);
245 spin_unlock(&inode->lock);
250 int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
251 u64 disk_bytenr, u64 num_bytes, u64 disk_num_bytes,
254 ASSERT(type == BTRFS_ORDERED_REGULAR ||
255 type == BTRFS_ORDERED_NOCOW ||
256 type == BTRFS_ORDERED_PREALLOC);
257 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
258 num_bytes, disk_num_bytes, type, 0,
259 BTRFS_COMPRESS_NONE);
262 int btrfs_add_ordered_extent_dio(struct btrfs_inode *inode, u64 file_offset,
263 u64 disk_bytenr, u64 num_bytes,
264 u64 disk_num_bytes, int type)
266 ASSERT(type == BTRFS_ORDERED_REGULAR ||
267 type == BTRFS_ORDERED_NOCOW ||
268 type == BTRFS_ORDERED_PREALLOC);
269 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
270 num_bytes, disk_num_bytes, type, 1,
271 BTRFS_COMPRESS_NONE);
274 int btrfs_add_ordered_extent_compress(struct btrfs_inode *inode, u64 file_offset,
275 u64 disk_bytenr, u64 num_bytes,
276 u64 disk_num_bytes, int compress_type)
278 ASSERT(compress_type != BTRFS_COMPRESS_NONE);
279 return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
280 num_bytes, disk_num_bytes,
281 BTRFS_ORDERED_COMPRESSED, 0,
286 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
287 * when an ordered extent is finished. If the list covers more than one
288 * ordered extent, it is split across multiples.
290 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
291 struct btrfs_ordered_sum *sum)
293 struct btrfs_ordered_inode_tree *tree;
295 tree = &BTRFS_I(entry->inode)->ordered_tree;
296 spin_lock_irq(&tree->lock);
297 list_add_tail(&sum->list, &entry->list);
298 spin_unlock_irq(&tree->lock);
302 * Mark all ordered extents io inside the specified range finished.
304 * @page: The invovled page for the opeartion.
305 * For uncompressed buffered IO, the page status also needs to be
306 * updated to indicate whether the pending ordered io is finished.
307 * Can be NULL for direct IO and compressed write.
308 * For these cases, callers are ensured they won't execute the
309 * endio function twice.
310 * @finish_func: The function to be executed when all the IO of an ordered
311 * extent are finished.
313 * This function is called for endio, thus the range must have ordered
314 * extent(s) coveri it.
316 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
317 struct page *page, u64 file_offset,
318 u64 num_bytes, btrfs_func_t finish_func,
321 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
322 struct btrfs_fs_info *fs_info = inode->root->fs_info;
323 struct btrfs_workqueue *wq;
324 struct rb_node *node;
325 struct btrfs_ordered_extent *entry = NULL;
327 u64 cur = file_offset;
329 if (btrfs_is_free_space_inode(inode))
330 wq = fs_info->endio_freespace_worker;
332 wq = fs_info->endio_write_workers;
335 ASSERT(page->mapping && page_offset(page) <= file_offset &&
336 file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
338 spin_lock_irqsave(&tree->lock, flags);
339 while (cur < file_offset + num_bytes) {
344 node = tree_search(tree, cur);
345 /* No ordered extents at all */
349 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
350 entry_end = entry->file_offset + entry->num_bytes;
356 if (cur >= entry_end) {
357 node = rb_next(node);
358 /* No more ordered extents, exit */
361 entry = rb_entry(node, struct btrfs_ordered_extent,
364 /* Go to next ordered extent and continue */
365 cur = entry->file_offset;
371 * Go to the start of OE.
373 if (cur < entry->file_offset) {
374 cur = entry->file_offset;
379 * Now we are definitely inside one ordered extent.
385 end = min(entry->file_offset + entry->num_bytes,
386 file_offset + num_bytes) - 1;
387 ASSERT(end + 1 - cur < U32_MAX);
392 * Ordered (Private2) bit indicates whether we still
393 * have pending io unfinished for the ordered extent.
395 * If there's no such bit, we need to skip to next range.
397 if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
401 btrfs_page_clear_ordered(fs_info, page, cur, len);
404 /* Now we're fine to update the accounting */
405 if (unlikely(len > entry->bytes_left)) {
408 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
409 inode->root->root_key.objectid,
413 len, entry->bytes_left);
414 entry->bytes_left = 0;
416 entry->bytes_left -= len;
420 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
423 * All the IO of the ordered extent is finished, we need to queue
424 * the finish_func to be executed.
426 if (entry->bytes_left == 0) {
427 set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
428 cond_wake_up(&entry->wait);
429 refcount_inc(&entry->refs);
430 spin_unlock_irqrestore(&tree->lock, flags);
431 btrfs_init_work(&entry->work, finish_func, NULL, NULL);
432 btrfs_queue_work(wq, &entry->work);
433 spin_lock_irqsave(&tree->lock, flags);
437 spin_unlock_irqrestore(&tree->lock, flags);
441 * Finish IO for one ordered extent across a given range. The range can only
442 * contain one ordered extent.
444 * @cached: The cached ordered extent. If not NULL, we can skip the tree
445 * search and use the ordered extent directly.
446 * Will be also used to store the finished ordered extent.
447 * @file_offset: File offset for the finished IO
448 * @io_size: Length of the finish IO range
449 * @uptodate: If the IO finishes without problem
451 * Return true if the ordered extent is finished in the range, and update
453 * Return false otherwise.
455 * NOTE: The range can NOT cross multiple ordered extents.
456 * Thus caller should ensure the range doesn't cross ordered extents.
458 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
459 struct btrfs_ordered_extent **cached,
460 u64 file_offset, u64 io_size, int uptodate)
462 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
463 struct rb_node *node;
464 struct btrfs_ordered_extent *entry = NULL;
466 bool finished = false;
468 spin_lock_irqsave(&tree->lock, flags);
469 if (cached && *cached) {
474 node = tree_search(tree, file_offset);
478 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
480 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
483 if (io_size > entry->bytes_left)
484 btrfs_crit(inode->root->fs_info,
485 "bad ordered accounting left %llu size %llu",
486 entry->bytes_left, io_size);
488 entry->bytes_left -= io_size;
490 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
492 if (entry->bytes_left == 0) {
494 * Ensure only one caller can set the flag and finished_ret
497 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
498 /* test_and_set_bit implies a barrier */
499 cond_wake_up_nomb(&entry->wait);
502 if (finished && cached && entry) {
504 refcount_inc(&entry->refs);
506 spin_unlock_irqrestore(&tree->lock, flags);
511 * used to drop a reference on an ordered extent. This will free
512 * the extent if the last reference is dropped
514 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
516 struct list_head *cur;
517 struct btrfs_ordered_sum *sum;
519 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
521 if (refcount_dec_and_test(&entry->refs)) {
522 ASSERT(list_empty(&entry->root_extent_list));
523 ASSERT(list_empty(&entry->log_list));
524 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
526 btrfs_add_delayed_iput(entry->inode);
527 while (!list_empty(&entry->list)) {
528 cur = entry->list.next;
529 sum = list_entry(cur, struct btrfs_ordered_sum, list);
530 list_del(&sum->list);
533 kmem_cache_free(btrfs_ordered_extent_cache, entry);
538 * remove an ordered extent from the tree. No references are dropped
539 * and waiters are woken up.
541 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
542 struct btrfs_ordered_extent *entry)
544 struct btrfs_ordered_inode_tree *tree;
545 struct btrfs_root *root = btrfs_inode->root;
546 struct btrfs_fs_info *fs_info = root->fs_info;
547 struct rb_node *node;
550 /* This is paired with btrfs_add_ordered_extent. */
551 spin_lock(&btrfs_inode->lock);
552 btrfs_mod_outstanding_extents(btrfs_inode, -1);
553 spin_unlock(&btrfs_inode->lock);
554 if (root != fs_info->tree_root)
555 btrfs_delalloc_release_metadata(btrfs_inode, entry->num_bytes,
558 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
559 fs_info->delalloc_batch);
561 tree = &btrfs_inode->ordered_tree;
562 spin_lock_irq(&tree->lock);
563 node = &entry->rb_node;
564 rb_erase(node, &tree->tree);
566 if (tree->last == node)
568 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
569 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
570 spin_unlock_irq(&tree->lock);
573 * The current running transaction is waiting on us, we need to let it
574 * know that we're complete and wake it up.
577 struct btrfs_transaction *trans;
580 * The checks for trans are just a formality, it should be set,
581 * but if it isn't we don't want to deref/assert under the spin
582 * lock, so be nice and check if trans is set, but ASSERT() so
583 * if it isn't set a developer will notice.
585 spin_lock(&fs_info->trans_lock);
586 trans = fs_info->running_transaction;
588 refcount_inc(&trans->use_count);
589 spin_unlock(&fs_info->trans_lock);
593 if (atomic_dec_and_test(&trans->pending_ordered))
594 wake_up(&trans->pending_wait);
595 btrfs_put_transaction(trans);
599 spin_lock(&root->ordered_extent_lock);
600 list_del_init(&entry->root_extent_list);
601 root->nr_ordered_extents--;
603 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
605 if (!root->nr_ordered_extents) {
606 spin_lock(&fs_info->ordered_root_lock);
607 BUG_ON(list_empty(&root->ordered_root));
608 list_del_init(&root->ordered_root);
609 spin_unlock(&fs_info->ordered_root_lock);
611 spin_unlock(&root->ordered_extent_lock);
612 wake_up(&entry->wait);
615 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
617 struct btrfs_ordered_extent *ordered;
619 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
620 btrfs_start_ordered_extent(ordered, 1);
621 complete(&ordered->completion);
625 * wait for all the ordered extents in a root. This is done when balancing
626 * space between drives.
628 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
629 const u64 range_start, const u64 range_len)
631 struct btrfs_fs_info *fs_info = root->fs_info;
635 struct btrfs_ordered_extent *ordered, *next;
637 const u64 range_end = range_start + range_len;
639 mutex_lock(&root->ordered_extent_mutex);
640 spin_lock(&root->ordered_extent_lock);
641 list_splice_init(&root->ordered_extents, &splice);
642 while (!list_empty(&splice) && nr) {
643 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
646 if (range_end <= ordered->disk_bytenr ||
647 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
648 list_move_tail(&ordered->root_extent_list, &skipped);
649 cond_resched_lock(&root->ordered_extent_lock);
653 list_move_tail(&ordered->root_extent_list,
654 &root->ordered_extents);
655 refcount_inc(&ordered->refs);
656 spin_unlock(&root->ordered_extent_lock);
658 btrfs_init_work(&ordered->flush_work,
659 btrfs_run_ordered_extent_work, NULL, NULL);
660 list_add_tail(&ordered->work_list, &works);
661 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
664 spin_lock(&root->ordered_extent_lock);
669 list_splice_tail(&skipped, &root->ordered_extents);
670 list_splice_tail(&splice, &root->ordered_extents);
671 spin_unlock(&root->ordered_extent_lock);
673 list_for_each_entry_safe(ordered, next, &works, work_list) {
674 list_del_init(&ordered->work_list);
675 wait_for_completion(&ordered->completion);
676 btrfs_put_ordered_extent(ordered);
679 mutex_unlock(&root->ordered_extent_mutex);
684 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
685 const u64 range_start, const u64 range_len)
687 struct btrfs_root *root;
688 struct list_head splice;
691 INIT_LIST_HEAD(&splice);
693 mutex_lock(&fs_info->ordered_operations_mutex);
694 spin_lock(&fs_info->ordered_root_lock);
695 list_splice_init(&fs_info->ordered_roots, &splice);
696 while (!list_empty(&splice) && nr) {
697 root = list_first_entry(&splice, struct btrfs_root,
699 root = btrfs_grab_root(root);
701 list_move_tail(&root->ordered_root,
702 &fs_info->ordered_roots);
703 spin_unlock(&fs_info->ordered_root_lock);
705 done = btrfs_wait_ordered_extents(root, nr,
706 range_start, range_len);
707 btrfs_put_root(root);
709 spin_lock(&fs_info->ordered_root_lock);
714 list_splice_tail(&splice, &fs_info->ordered_roots);
715 spin_unlock(&fs_info->ordered_root_lock);
716 mutex_unlock(&fs_info->ordered_operations_mutex);
720 * Used to start IO or wait for a given ordered extent to finish.
722 * If wait is one, this effectively waits on page writeback for all the pages
723 * in the extent, and it waits on the io completion code to insert
724 * metadata into the btree corresponding to the extent
726 void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
728 u64 start = entry->file_offset;
729 u64 end = start + entry->num_bytes - 1;
730 struct btrfs_inode *inode = BTRFS_I(entry->inode);
732 trace_btrfs_ordered_extent_start(inode, entry);
735 * pages in the range can be dirty, clean or writeback. We
736 * start IO on any dirty ones so the wait doesn't stall waiting
737 * for the flusher thread to find them
739 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
740 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
742 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
748 * Used to wait on ordered extents across a large range of bytes.
750 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
756 struct btrfs_ordered_extent *ordered;
758 if (start + len < start) {
759 orig_end = INT_LIMIT(loff_t);
761 orig_end = start + len - 1;
762 if (orig_end > INT_LIMIT(loff_t))
763 orig_end = INT_LIMIT(loff_t);
766 /* start IO across the range first to instantiate any delalloc
769 ret = btrfs_fdatawrite_range(inode, start, orig_end);
774 * If we have a writeback error don't return immediately. Wait first
775 * for any ordered extents that haven't completed yet. This is to make
776 * sure no one can dirty the same page ranges and call writepages()
777 * before the ordered extents complete - to avoid failures (-EEXIST)
778 * when adding the new ordered extents to the ordered tree.
780 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
784 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
787 if (ordered->file_offset > orig_end) {
788 btrfs_put_ordered_extent(ordered);
791 if (ordered->file_offset + ordered->num_bytes <= start) {
792 btrfs_put_ordered_extent(ordered);
795 btrfs_start_ordered_extent(ordered, 1);
796 end = ordered->file_offset;
798 * If the ordered extent had an error save the error but don't
799 * exit without waiting first for all other ordered extents in
800 * the range to complete.
802 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
804 btrfs_put_ordered_extent(ordered);
805 if (end == 0 || end == start)
809 return ret_wb ? ret_wb : ret;
813 * find an ordered extent corresponding to file_offset. return NULL if
814 * nothing is found, otherwise take a reference on the extent and return it
816 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
819 struct btrfs_ordered_inode_tree *tree;
820 struct rb_node *node;
821 struct btrfs_ordered_extent *entry = NULL;
824 tree = &inode->ordered_tree;
825 spin_lock_irqsave(&tree->lock, flags);
826 node = tree_search(tree, file_offset);
830 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
831 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
834 refcount_inc(&entry->refs);
836 spin_unlock_irqrestore(&tree->lock, flags);
840 /* Since the DIO code tries to lock a wide area we need to look for any ordered
841 * extents that exist in the range, rather than just the start of the range.
843 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
844 struct btrfs_inode *inode, u64 file_offset, u64 len)
846 struct btrfs_ordered_inode_tree *tree;
847 struct rb_node *node;
848 struct btrfs_ordered_extent *entry = NULL;
850 tree = &inode->ordered_tree;
851 spin_lock_irq(&tree->lock);
852 node = tree_search(tree, file_offset);
854 node = tree_search(tree, file_offset + len);
860 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
861 if (range_overlaps(entry, file_offset, len))
864 if (entry->file_offset >= file_offset + len) {
869 node = rb_next(node);
875 refcount_inc(&entry->refs);
876 spin_unlock_irq(&tree->lock);
881 * Adds all ordered extents to the given list. The list ends up sorted by the
882 * file_offset of the ordered extents.
884 void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
885 struct list_head *list)
887 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
890 ASSERT(inode_is_locked(&inode->vfs_inode));
892 spin_lock_irq(&tree->lock);
893 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
894 struct btrfs_ordered_extent *ordered;
896 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
898 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
901 ASSERT(list_empty(&ordered->log_list));
902 list_add_tail(&ordered->log_list, list);
903 refcount_inc(&ordered->refs);
905 spin_unlock_irq(&tree->lock);
909 * lookup and return any extent before 'file_offset'. NULL is returned
912 struct btrfs_ordered_extent *
913 btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
915 struct btrfs_ordered_inode_tree *tree;
916 struct rb_node *node;
917 struct btrfs_ordered_extent *entry = NULL;
919 tree = &inode->ordered_tree;
920 spin_lock_irq(&tree->lock);
921 node = tree_search(tree, file_offset);
925 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
926 refcount_inc(&entry->refs);
928 spin_unlock_irq(&tree->lock);
933 * Lookup the first ordered extent that overlaps the range
934 * [@file_offset, @file_offset + @len).
936 * The difference between this and btrfs_lookup_first_ordered_extent() is
937 * that this one won't return any ordered extent that does not overlap the range.
938 * And the difference against btrfs_lookup_ordered_extent() is, this function
939 * ensures the first ordered extent gets returned.
941 struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
942 struct btrfs_inode *inode, u64 file_offset, u64 len)
944 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
945 struct rb_node *node;
947 struct rb_node *prev;
948 struct rb_node *next;
949 struct btrfs_ordered_extent *entry = NULL;
951 spin_lock_irq(&tree->lock);
952 node = tree->tree.rb_node;
954 * Here we don't want to use tree_search() which will use tree->last
955 * and screw up the search order.
956 * And __tree_search() can't return the adjacent ordered extents
957 * either, thus here we do our own search.
960 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
962 if (file_offset < entry->file_offset) {
963 node = node->rb_left;
964 } else if (file_offset >= entry_end(entry)) {
965 node = node->rb_right;
968 * Direct hit, got an ordered extent that starts at
979 cur = &entry->rb_node;
980 /* We got an entry around @file_offset, check adjacent entries */
981 if (entry->file_offset < file_offset) {
989 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
990 if (range_overlaps(entry, file_offset, len))
994 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
995 if (range_overlaps(entry, file_offset, len))
998 /* No ordered extent in the range */
1002 refcount_inc(&entry->refs);
1003 spin_unlock_irq(&tree->lock);
1008 * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
1009 * ordered extents in it are run to completion.
1011 * @inode: Inode whose ordered tree is to be searched
1012 * @start: Beginning of range to flush
1013 * @end: Last byte of range to lock
1014 * @cached_state: If passed, will return the extent state responsible for the
1015 * locked range. It's the caller's responsibility to free the cached state.
1017 * This function always returns with the given range locked, ensuring after it's
1018 * called no order extent can be pending.
1020 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1022 struct extent_state **cached_state)
1024 struct btrfs_ordered_extent *ordered;
1025 struct extent_state *cache = NULL;
1026 struct extent_state **cachedp = &cache;
1029 cachedp = cached_state;
1032 lock_extent_bits(&inode->io_tree, start, end, cachedp);
1033 ordered = btrfs_lookup_ordered_range(inode, start,
1037 * If no external cached_state has been passed then
1038 * decrement the extra ref taken for cachedp since we
1039 * aren't exposing it outside of this function
1042 refcount_dec(&cache->refs);
1045 unlock_extent_cached(&inode->io_tree, start, end, cachedp);
1046 btrfs_start_ordered_extent(ordered, 1);
1047 btrfs_put_ordered_extent(ordered);
1051 static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1054 struct inode *inode = ordered->inode;
1055 u64 file_offset = ordered->file_offset + pos;
1056 u64 disk_bytenr = ordered->disk_bytenr + pos;
1057 u64 num_bytes = len;
1058 u64 disk_num_bytes = len;
1060 unsigned long flags_masked = ordered->flags & ~(1 << BTRFS_ORDERED_DIRECT);
1061 int compress_type = ordered->compress_type;
1062 unsigned long weight;
1065 weight = hweight_long(flags_masked);
1066 WARN_ON_ONCE(weight > 1);
1070 type = __ffs(flags_masked);
1072 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered->flags)) {
1074 ret = btrfs_add_ordered_extent_compress(BTRFS_I(inode),
1075 file_offset, disk_bytenr, num_bytes,
1076 disk_num_bytes, compress_type);
1077 } else if (test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
1078 ret = btrfs_add_ordered_extent_dio(BTRFS_I(inode), file_offset,
1079 disk_bytenr, num_bytes, disk_num_bytes, type);
1081 ret = btrfs_add_ordered_extent(BTRFS_I(inode), file_offset,
1082 disk_bytenr, num_bytes, disk_num_bytes, type);
1088 int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1091 struct inode *inode = ordered->inode;
1092 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1093 struct rb_node *node;
1094 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1097 spin_lock_irq(&tree->lock);
1098 /* Remove from tree once */
1099 node = &ordered->rb_node;
1100 rb_erase(node, &tree->tree);
1101 RB_CLEAR_NODE(node);
1102 if (tree->last == node)
1105 ordered->file_offset += pre;
1106 ordered->disk_bytenr += pre;
1107 ordered->num_bytes -= (pre + post);
1108 ordered->disk_num_bytes -= (pre + post);
1109 ordered->bytes_left -= (pre + post);
1111 /* Re-insert the node */
1112 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1114 btrfs_panic(fs_info, -EEXIST,
1115 "zoned: inconsistency in ordered tree at offset %llu",
1116 ordered->file_offset);
1118 spin_unlock_irq(&tree->lock);
1121 ret = clone_ordered_extent(ordered, 0, pre);
1122 if (ret == 0 && post)
1123 ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1129 int __init ordered_data_init(void)
1131 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1132 sizeof(struct btrfs_ordered_extent), 0,
1135 if (!btrfs_ordered_extent_cache)
1141 void __cold ordered_data_exit(void)
1143 kmem_cache_destroy(btrfs_ordered_extent_cache);