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);
149 static struct btrfs_ordered_extent *alloc_ordered_extent(
150 struct btrfs_inode *inode, u64 file_offset, u64 num_bytes,
151 u64 ram_bytes, u64 disk_bytenr, u64 disk_num_bytes,
152 u64 offset, unsigned long flags, int compress_type)
154 struct btrfs_ordered_extent *entry;
158 ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
159 /* For nocow write, we can release the qgroup rsv right now */
160 ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
165 * The ordered extent has reserved qgroup space, release now
166 * and pass the reserved number for qgroup_record to free.
168 ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
172 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
174 return ERR_PTR(-ENOMEM);
176 entry->file_offset = file_offset;
177 entry->num_bytes = num_bytes;
178 entry->ram_bytes = ram_bytes;
179 entry->disk_bytenr = disk_bytenr;
180 entry->disk_num_bytes = disk_num_bytes;
181 entry->offset = offset;
182 entry->bytes_left = num_bytes;
183 entry->inode = igrab(&inode->vfs_inode);
184 entry->compress_type = compress_type;
185 entry->truncated_len = (u64)-1;
186 entry->qgroup_rsv = ret;
187 entry->flags = flags;
188 refcount_set(&entry->refs, 1);
189 init_waitqueue_head(&entry->wait);
190 INIT_LIST_HEAD(&entry->list);
191 INIT_LIST_HEAD(&entry->log_list);
192 INIT_LIST_HEAD(&entry->root_extent_list);
193 INIT_LIST_HEAD(&entry->work_list);
194 init_completion(&entry->completion);
197 * We don't need the count_max_extents here, we can assume that all of
198 * that work has been done at higher layers, so this is truly the
199 * smallest the extent is going to get.
201 spin_lock(&inode->lock);
202 btrfs_mod_outstanding_extents(inode, 1);
203 spin_unlock(&inode->lock);
208 static void insert_ordered_extent(struct btrfs_ordered_extent *entry)
210 struct btrfs_inode *inode = BTRFS_I(entry->inode);
211 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
212 struct btrfs_root *root = inode->root;
213 struct btrfs_fs_info *fs_info = root->fs_info;
214 struct rb_node *node;
216 trace_btrfs_ordered_extent_add(inode, entry);
218 percpu_counter_add_batch(&fs_info->ordered_bytes, entry->num_bytes,
219 fs_info->delalloc_batch);
221 /* One ref for the tree. */
222 refcount_inc(&entry->refs);
224 spin_lock_irq(&tree->lock);
225 node = tree_insert(&tree->tree, entry->file_offset, &entry->rb_node);
227 btrfs_panic(fs_info, -EEXIST,
228 "inconsistency in ordered tree at offset %llu",
230 spin_unlock_irq(&tree->lock);
232 spin_lock(&root->ordered_extent_lock);
233 list_add_tail(&entry->root_extent_list,
234 &root->ordered_extents);
235 root->nr_ordered_extents++;
236 if (root->nr_ordered_extents == 1) {
237 spin_lock(&fs_info->ordered_root_lock);
238 BUG_ON(!list_empty(&root->ordered_root));
239 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
240 spin_unlock(&fs_info->ordered_root_lock);
242 spin_unlock(&root->ordered_extent_lock);
246 * Add an ordered extent to the per-inode tree.
248 * @inode: Inode that this extent is for.
249 * @file_offset: Logical offset in file where the extent starts.
250 * @num_bytes: Logical length of extent in file.
251 * @ram_bytes: Full length of unencoded data.
252 * @disk_bytenr: Offset of extent on disk.
253 * @disk_num_bytes: Size of extent on disk.
254 * @offset: Offset into unencoded data where file data starts.
255 * @flags: Flags specifying type of extent (1 << BTRFS_ORDERED_*).
256 * @compress_type: Compression algorithm used for data.
258 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
259 * tree is given a single reference on the ordered extent that was inserted, and
260 * the returned pointer is given a second reference.
262 * Return: the new ordered extent or error pointer.
264 struct btrfs_ordered_extent *btrfs_alloc_ordered_extent(
265 struct btrfs_inode *inode, u64 file_offset,
266 u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
267 u64 disk_num_bytes, u64 offset, unsigned long flags,
270 struct btrfs_ordered_extent *entry;
272 ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
274 entry = alloc_ordered_extent(inode, file_offset, num_bytes, ram_bytes,
275 disk_bytenr, disk_num_bytes, offset, flags,
278 insert_ordered_extent(entry);
283 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
284 * when an ordered extent is finished. If the list covers more than one
285 * ordered extent, it is split across multiples.
287 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
288 struct btrfs_ordered_sum *sum)
290 struct btrfs_ordered_inode_tree *tree;
292 tree = &BTRFS_I(entry->inode)->ordered_tree;
293 spin_lock_irq(&tree->lock);
294 list_add_tail(&sum->list, &entry->list);
295 spin_unlock_irq(&tree->lock);
298 static void finish_ordered_fn(struct btrfs_work *work)
300 struct btrfs_ordered_extent *ordered_extent;
302 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
303 btrfs_finish_ordered_io(ordered_extent);
306 static bool can_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
307 struct page *page, u64 file_offset,
308 u64 len, bool uptodate)
310 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
311 struct btrfs_fs_info *fs_info = inode->root->fs_info;
313 lockdep_assert_held(&inode->ordered_tree.lock);
316 ASSERT(page->mapping);
317 ASSERT(page_offset(page) <= file_offset);
318 ASSERT(file_offset + len <= page_offset(page) + PAGE_SIZE);
321 * Ordered (Private2) bit indicates whether we still have
322 * pending io unfinished for the ordered extent.
324 * If there's no such bit, we need to skip to next range.
326 if (!btrfs_page_test_ordered(fs_info, page, file_offset, len))
328 btrfs_page_clear_ordered(fs_info, page, file_offset, len);
331 /* Now we're fine to update the accounting. */
332 if (WARN_ON_ONCE(len > ordered->bytes_left)) {
334 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%llu left=%llu",
335 inode->root->root_key.objectid, btrfs_ino(inode),
336 ordered->file_offset, ordered->num_bytes,
337 len, ordered->bytes_left);
338 ordered->bytes_left = 0;
340 ordered->bytes_left -= len;
344 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
346 if (ordered->bytes_left)
350 * All the IO of the ordered extent is finished, we need to queue
351 * the finish_func to be executed.
353 set_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags);
354 cond_wake_up(&ordered->wait);
355 refcount_inc(&ordered->refs);
356 trace_btrfs_ordered_extent_mark_finished(inode, ordered);
360 static void btrfs_queue_ordered_fn(struct btrfs_ordered_extent *ordered)
362 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
363 struct btrfs_fs_info *fs_info = inode->root->fs_info;
364 struct btrfs_workqueue *wq = btrfs_is_free_space_inode(inode) ?
365 fs_info->endio_freespace_worker : fs_info->endio_write_workers;
367 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
368 btrfs_queue_work(wq, &ordered->work);
371 bool btrfs_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
372 struct page *page, u64 file_offset, u64 len,
375 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
379 trace_btrfs_finish_ordered_extent(inode, file_offset, len, uptodate);
381 spin_lock_irqsave(&inode->ordered_tree.lock, flags);
382 ret = can_finish_ordered_extent(ordered, page, file_offset, len, uptodate);
383 spin_unlock_irqrestore(&inode->ordered_tree.lock, flags);
386 btrfs_queue_ordered_fn(ordered);
391 * Mark all ordered extents io inside the specified range finished.
393 * @page: The involved page for the operation.
394 * For uncompressed buffered IO, the page status also needs to be
395 * updated to indicate whether the pending ordered io is finished.
396 * Can be NULL for direct IO and compressed write.
397 * For these cases, callers are ensured they won't execute the
398 * endio function twice.
400 * This function is called for endio, thus the range must have ordered
401 * extent(s) covering it.
403 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
404 struct page *page, u64 file_offset,
405 u64 num_bytes, bool uptodate)
407 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
408 struct rb_node *node;
409 struct btrfs_ordered_extent *entry = NULL;
411 u64 cur = file_offset;
413 spin_lock_irqsave(&tree->lock, flags);
414 while (cur < file_offset + num_bytes) {
419 node = tree_search(tree, cur);
420 /* No ordered extents at all */
424 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
425 entry_end = entry->file_offset + entry->num_bytes;
431 if (cur >= entry_end) {
432 node = rb_next(node);
433 /* No more ordered extents, exit */
436 entry = rb_entry(node, struct btrfs_ordered_extent,
439 /* Go to next ordered extent and continue */
440 cur = entry->file_offset;
446 * Go to the start of OE.
448 if (cur < entry->file_offset) {
449 cur = entry->file_offset;
454 * Now we are definitely inside one ordered extent.
460 end = min(entry->file_offset + entry->num_bytes,
461 file_offset + num_bytes) - 1;
462 ASSERT(end + 1 - cur < U32_MAX);
465 if (can_finish_ordered_extent(entry, page, cur, len, uptodate)) {
466 spin_unlock_irqrestore(&tree->lock, flags);
467 btrfs_queue_ordered_fn(entry);
468 spin_lock_irqsave(&tree->lock, flags);
472 spin_unlock_irqrestore(&tree->lock, flags);
476 * Finish IO for one ordered extent across a given range. The range can only
477 * contain one ordered extent.
479 * @cached: The cached ordered extent. If not NULL, we can skip the tree
480 * search and use the ordered extent directly.
481 * Will be also used to store the finished ordered extent.
482 * @file_offset: File offset for the finished IO
483 * @io_size: Length of the finish IO range
485 * Return true if the ordered extent is finished in the range, and update
487 * Return false otherwise.
489 * NOTE: The range can NOT cross multiple ordered extents.
490 * Thus caller should ensure the range doesn't cross ordered extents.
492 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
493 struct btrfs_ordered_extent **cached,
494 u64 file_offset, u64 io_size)
496 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
497 struct rb_node *node;
498 struct btrfs_ordered_extent *entry = NULL;
500 bool finished = false;
502 spin_lock_irqsave(&tree->lock, flags);
503 if (cached && *cached) {
508 node = tree_search(tree, file_offset);
512 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
514 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
517 if (io_size > entry->bytes_left)
518 btrfs_crit(inode->root->fs_info,
519 "bad ordered accounting left %llu size %llu",
520 entry->bytes_left, io_size);
522 entry->bytes_left -= io_size;
524 if (entry->bytes_left == 0) {
526 * Ensure only one caller can set the flag and finished_ret
529 finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
530 /* test_and_set_bit implies a barrier */
531 cond_wake_up_nomb(&entry->wait);
534 if (finished && cached && entry) {
536 refcount_inc(&entry->refs);
537 trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
539 spin_unlock_irqrestore(&tree->lock, flags);
544 * used to drop a reference on an ordered extent. This will free
545 * the extent if the last reference is dropped
547 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
549 struct list_head *cur;
550 struct btrfs_ordered_sum *sum;
552 trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
554 if (refcount_dec_and_test(&entry->refs)) {
555 ASSERT(list_empty(&entry->root_extent_list));
556 ASSERT(list_empty(&entry->log_list));
557 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
559 btrfs_add_delayed_iput(BTRFS_I(entry->inode));
560 while (!list_empty(&entry->list)) {
561 cur = entry->list.next;
562 sum = list_entry(cur, struct btrfs_ordered_sum, list);
563 list_del(&sum->list);
566 kmem_cache_free(btrfs_ordered_extent_cache, entry);
571 * remove an ordered extent from the tree. No references are dropped
572 * and waiters are woken up.
574 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
575 struct btrfs_ordered_extent *entry)
577 struct btrfs_ordered_inode_tree *tree;
578 struct btrfs_root *root = btrfs_inode->root;
579 struct btrfs_fs_info *fs_info = root->fs_info;
580 struct rb_node *node;
582 bool freespace_inode;
585 * If this is a free space inode the thread has not acquired the ordered
586 * extents lockdep map.
588 freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
590 btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
591 /* This is paired with btrfs_alloc_ordered_extent. */
592 spin_lock(&btrfs_inode->lock);
593 btrfs_mod_outstanding_extents(btrfs_inode, -1);
594 spin_unlock(&btrfs_inode->lock);
595 if (root != fs_info->tree_root) {
598 if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
599 release = entry->disk_num_bytes;
601 release = entry->num_bytes;
602 btrfs_delalloc_release_metadata(btrfs_inode, release, false);
605 percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
606 fs_info->delalloc_batch);
608 tree = &btrfs_inode->ordered_tree;
609 spin_lock_irq(&tree->lock);
610 node = &entry->rb_node;
611 rb_erase(node, &tree->tree);
613 if (tree->last == node)
615 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
616 pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
617 spin_unlock_irq(&tree->lock);
620 * The current running transaction is waiting on us, we need to let it
621 * know that we're complete and wake it up.
624 struct btrfs_transaction *trans;
627 * The checks for trans are just a formality, it should be set,
628 * but if it isn't we don't want to deref/assert under the spin
629 * lock, so be nice and check if trans is set, but ASSERT() so
630 * if it isn't set a developer will notice.
632 spin_lock(&fs_info->trans_lock);
633 trans = fs_info->running_transaction;
635 refcount_inc(&trans->use_count);
636 spin_unlock(&fs_info->trans_lock);
640 if (atomic_dec_and_test(&trans->pending_ordered))
641 wake_up(&trans->pending_wait);
642 btrfs_put_transaction(trans);
646 btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
648 spin_lock(&root->ordered_extent_lock);
649 list_del_init(&entry->root_extent_list);
650 root->nr_ordered_extents--;
652 trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
654 if (!root->nr_ordered_extents) {
655 spin_lock(&fs_info->ordered_root_lock);
656 BUG_ON(list_empty(&root->ordered_root));
657 list_del_init(&root->ordered_root);
658 spin_unlock(&fs_info->ordered_root_lock);
660 spin_unlock(&root->ordered_extent_lock);
661 wake_up(&entry->wait);
662 if (!freespace_inode)
663 btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
666 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
668 struct btrfs_ordered_extent *ordered;
670 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
671 btrfs_start_ordered_extent(ordered);
672 complete(&ordered->completion);
676 * wait for all the ordered extents in a root. This is done when balancing
677 * space between drives.
679 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
680 const u64 range_start, const u64 range_len)
682 struct btrfs_fs_info *fs_info = root->fs_info;
686 struct btrfs_ordered_extent *ordered, *next;
688 const u64 range_end = range_start + range_len;
690 mutex_lock(&root->ordered_extent_mutex);
691 spin_lock(&root->ordered_extent_lock);
692 list_splice_init(&root->ordered_extents, &splice);
693 while (!list_empty(&splice) && nr) {
694 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
697 if (range_end <= ordered->disk_bytenr ||
698 ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
699 list_move_tail(&ordered->root_extent_list, &skipped);
700 cond_resched_lock(&root->ordered_extent_lock);
704 list_move_tail(&ordered->root_extent_list,
705 &root->ordered_extents);
706 refcount_inc(&ordered->refs);
707 spin_unlock(&root->ordered_extent_lock);
709 btrfs_init_work(&ordered->flush_work,
710 btrfs_run_ordered_extent_work, NULL, NULL);
711 list_add_tail(&ordered->work_list, &works);
712 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
715 spin_lock(&root->ordered_extent_lock);
720 list_splice_tail(&skipped, &root->ordered_extents);
721 list_splice_tail(&splice, &root->ordered_extents);
722 spin_unlock(&root->ordered_extent_lock);
724 list_for_each_entry_safe(ordered, next, &works, work_list) {
725 list_del_init(&ordered->work_list);
726 wait_for_completion(&ordered->completion);
727 btrfs_put_ordered_extent(ordered);
730 mutex_unlock(&root->ordered_extent_mutex);
735 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
736 const u64 range_start, const u64 range_len)
738 struct btrfs_root *root;
739 struct list_head splice;
742 INIT_LIST_HEAD(&splice);
744 mutex_lock(&fs_info->ordered_operations_mutex);
745 spin_lock(&fs_info->ordered_root_lock);
746 list_splice_init(&fs_info->ordered_roots, &splice);
747 while (!list_empty(&splice) && nr) {
748 root = list_first_entry(&splice, struct btrfs_root,
750 root = btrfs_grab_root(root);
752 list_move_tail(&root->ordered_root,
753 &fs_info->ordered_roots);
754 spin_unlock(&fs_info->ordered_root_lock);
756 done = btrfs_wait_ordered_extents(root, nr,
757 range_start, range_len);
758 btrfs_put_root(root);
760 spin_lock(&fs_info->ordered_root_lock);
765 list_splice_tail(&splice, &fs_info->ordered_roots);
766 spin_unlock(&fs_info->ordered_root_lock);
767 mutex_unlock(&fs_info->ordered_operations_mutex);
771 * Start IO and wait for a given ordered extent to finish.
773 * Wait on page writeback for all the pages in the extent and the IO completion
774 * code to insert metadata into the btree corresponding to the extent.
776 void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry)
778 u64 start = entry->file_offset;
779 u64 end = start + entry->num_bytes - 1;
780 struct btrfs_inode *inode = BTRFS_I(entry->inode);
781 bool freespace_inode;
783 trace_btrfs_ordered_extent_start(inode, entry);
786 * If this is a free space inode do not take the ordered extents lockdep
789 freespace_inode = btrfs_is_free_space_inode(inode);
792 * pages in the range can be dirty, clean or writeback. We
793 * start IO on any dirty ones so the wait doesn't stall waiting
794 * for the flusher thread to find them
796 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
797 filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
799 if (!freespace_inode)
800 btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
801 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, &entry->flags));
805 * Used to wait on ordered extents across a large range of bytes.
807 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
813 struct btrfs_ordered_extent *ordered;
815 if (start + len < start) {
816 orig_end = OFFSET_MAX;
818 orig_end = start + len - 1;
819 if (orig_end > OFFSET_MAX)
820 orig_end = OFFSET_MAX;
823 /* start IO across the range first to instantiate any delalloc
826 ret = btrfs_fdatawrite_range(inode, start, orig_end);
831 * If we have a writeback error don't return immediately. Wait first
832 * for any ordered extents that haven't completed yet. This is to make
833 * sure no one can dirty the same page ranges and call writepages()
834 * before the ordered extents complete - to avoid failures (-EEXIST)
835 * when adding the new ordered extents to the ordered tree.
837 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
841 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
844 if (ordered->file_offset > orig_end) {
845 btrfs_put_ordered_extent(ordered);
848 if (ordered->file_offset + ordered->num_bytes <= start) {
849 btrfs_put_ordered_extent(ordered);
852 btrfs_start_ordered_extent(ordered);
853 end = ordered->file_offset;
855 * If the ordered extent had an error save the error but don't
856 * exit without waiting first for all other ordered extents in
857 * the range to complete.
859 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
861 btrfs_put_ordered_extent(ordered);
862 if (end == 0 || end == start)
866 return ret_wb ? ret_wb : ret;
870 * find an ordered extent corresponding to file_offset. return NULL if
871 * nothing is found, otherwise take a reference on the extent and return it
873 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
876 struct btrfs_ordered_inode_tree *tree;
877 struct rb_node *node;
878 struct btrfs_ordered_extent *entry = NULL;
881 tree = &inode->ordered_tree;
882 spin_lock_irqsave(&tree->lock, flags);
883 node = tree_search(tree, file_offset);
887 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
888 if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
891 refcount_inc(&entry->refs);
892 trace_btrfs_ordered_extent_lookup(inode, entry);
895 spin_unlock_irqrestore(&tree->lock, flags);
899 /* Since the DIO code tries to lock a wide area we need to look for any ordered
900 * extents that exist in the range, rather than just the start of the range.
902 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
903 struct btrfs_inode *inode, u64 file_offset, u64 len)
905 struct btrfs_ordered_inode_tree *tree;
906 struct rb_node *node;
907 struct btrfs_ordered_extent *entry = NULL;
909 tree = &inode->ordered_tree;
910 spin_lock_irq(&tree->lock);
911 node = tree_search(tree, file_offset);
913 node = tree_search(tree, file_offset + len);
919 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
920 if (range_overlaps(entry, file_offset, len))
923 if (entry->file_offset >= file_offset + len) {
928 node = rb_next(node);
934 refcount_inc(&entry->refs);
935 trace_btrfs_ordered_extent_lookup_range(inode, entry);
937 spin_unlock_irq(&tree->lock);
942 * Adds all ordered extents to the given list. The list ends up sorted by the
943 * file_offset of the ordered extents.
945 void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
946 struct list_head *list)
948 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
951 ASSERT(inode_is_locked(&inode->vfs_inode));
953 spin_lock_irq(&tree->lock);
954 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
955 struct btrfs_ordered_extent *ordered;
957 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
959 if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
962 ASSERT(list_empty(&ordered->log_list));
963 list_add_tail(&ordered->log_list, list);
964 refcount_inc(&ordered->refs);
965 trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
967 spin_unlock_irq(&tree->lock);
971 * lookup and return any extent before 'file_offset'. NULL is returned
974 struct btrfs_ordered_extent *
975 btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
977 struct btrfs_ordered_inode_tree *tree;
978 struct rb_node *node;
979 struct btrfs_ordered_extent *entry = NULL;
981 tree = &inode->ordered_tree;
982 spin_lock_irq(&tree->lock);
983 node = tree_search(tree, file_offset);
987 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
988 refcount_inc(&entry->refs);
989 trace_btrfs_ordered_extent_lookup_first(inode, entry);
991 spin_unlock_irq(&tree->lock);
996 * Lookup the first ordered extent that overlaps the range
997 * [@file_offset, @file_offset + @len).
999 * The difference between this and btrfs_lookup_first_ordered_extent() is
1000 * that this one won't return any ordered extent that does not overlap the range.
1001 * And the difference against btrfs_lookup_ordered_extent() is, this function
1002 * ensures the first ordered extent gets returned.
1004 struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
1005 struct btrfs_inode *inode, u64 file_offset, u64 len)
1007 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
1008 struct rb_node *node;
1009 struct rb_node *cur;
1010 struct rb_node *prev;
1011 struct rb_node *next;
1012 struct btrfs_ordered_extent *entry = NULL;
1014 spin_lock_irq(&tree->lock);
1015 node = tree->tree.rb_node;
1017 * Here we don't want to use tree_search() which will use tree->last
1018 * and screw up the search order.
1019 * And __tree_search() can't return the adjacent ordered extents
1020 * either, thus here we do our own search.
1023 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1025 if (file_offset < entry->file_offset) {
1026 node = node->rb_left;
1027 } else if (file_offset >= entry_end(entry)) {
1028 node = node->rb_right;
1031 * Direct hit, got an ordered extent that starts at
1042 cur = &entry->rb_node;
1043 /* We got an entry around @file_offset, check adjacent entries */
1044 if (entry->file_offset < file_offset) {
1046 next = rb_next(cur);
1048 prev = rb_prev(cur);
1052 entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1053 if (range_overlaps(entry, file_offset, len))
1057 entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1058 if (range_overlaps(entry, file_offset, len))
1061 /* No ordered extent in the range */
1065 refcount_inc(&entry->refs);
1066 trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1069 spin_unlock_irq(&tree->lock);
1074 * Lock the passed range and ensures all pending ordered extents in it are run
1077 * @inode: Inode whose ordered tree is to be searched
1078 * @start: Beginning of range to flush
1079 * @end: Last byte of range to lock
1080 * @cached_state: If passed, will return the extent state responsible for the
1081 * locked range. It's the caller's responsibility to free the
1084 * Always return with the given range locked, ensuring after it's called no
1085 * order extent can be pending.
1087 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1089 struct extent_state **cached_state)
1091 struct btrfs_ordered_extent *ordered;
1092 struct extent_state *cache = NULL;
1093 struct extent_state **cachedp = &cache;
1096 cachedp = cached_state;
1099 lock_extent(&inode->io_tree, start, end, cachedp);
1100 ordered = btrfs_lookup_ordered_range(inode, start,
1104 * If no external cached_state has been passed then
1105 * decrement the extra ref taken for cachedp since we
1106 * aren't exposing it outside of this function
1109 refcount_dec(&cache->refs);
1112 unlock_extent(&inode->io_tree, start, end, cachedp);
1113 btrfs_start_ordered_extent(ordered);
1114 btrfs_put_ordered_extent(ordered);
1119 * Lock the passed range and ensure all pending ordered extents in it are run
1120 * to completion in nowait mode.
1122 * Return true if btrfs_lock_ordered_range does not return any extents,
1125 bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1126 struct extent_state **cached_state)
1128 struct btrfs_ordered_extent *ordered;
1130 if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1133 ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1137 btrfs_put_ordered_extent(ordered);
1138 unlock_extent(&inode->io_tree, start, end, cached_state);
1143 /* Split out a new ordered extent for this first @len bytes of @ordered. */
1144 struct btrfs_ordered_extent *btrfs_split_ordered_extent(
1145 struct btrfs_ordered_extent *ordered, u64 len)
1147 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1148 struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
1149 struct btrfs_root *root = inode->root;
1150 struct btrfs_fs_info *fs_info = root->fs_info;
1151 u64 file_offset = ordered->file_offset;
1152 u64 disk_bytenr = ordered->disk_bytenr;
1153 unsigned long flags = ordered->flags;
1154 struct btrfs_ordered_sum *sum, *tmpsum;
1155 struct btrfs_ordered_extent *new;
1156 struct rb_node *node;
1159 trace_btrfs_ordered_extent_split(inode, ordered);
1161 ASSERT(!(flags & (1U << BTRFS_ORDERED_COMPRESSED)));
1164 * The entire bio must be covered by the ordered extent, but we can't
1165 * reduce the original extent to a zero length either.
1167 if (WARN_ON_ONCE(len >= ordered->num_bytes))
1168 return ERR_PTR(-EINVAL);
1169 /* We cannot split partially completed ordered extents. */
1170 if (ordered->bytes_left) {
1171 ASSERT(!(flags & ~BTRFS_ORDERED_TYPE_FLAGS));
1172 if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes))
1173 return ERR_PTR(-EINVAL);
1175 /* We cannot split a compressed ordered extent. */
1176 if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes))
1177 return ERR_PTR(-EINVAL);
1179 new = alloc_ordered_extent(inode, file_offset, len, len, disk_bytenr,
1180 len, 0, flags, ordered->compress_type);
1184 /* One ref for the tree. */
1185 refcount_inc(&new->refs);
1187 spin_lock_irq(&root->ordered_extent_lock);
1188 spin_lock(&tree->lock);
1189 /* Remove from tree once */
1190 node = &ordered->rb_node;
1191 rb_erase(node, &tree->tree);
1192 RB_CLEAR_NODE(node);
1193 if (tree->last == node)
1196 ordered->file_offset += len;
1197 ordered->disk_bytenr += len;
1198 ordered->num_bytes -= len;
1199 ordered->disk_num_bytes -= len;
1201 if (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags)) {
1202 ASSERT(ordered->bytes_left == 0);
1203 new->bytes_left = 0;
1205 ordered->bytes_left -= len;
1208 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) {
1209 if (ordered->truncated_len > len) {
1210 ordered->truncated_len -= len;
1212 new->truncated_len = ordered->truncated_len;
1213 ordered->truncated_len = 0;
1217 list_for_each_entry_safe(sum, tmpsum, &ordered->list, list) {
1220 list_move_tail(&sum->list, &new->list);
1224 /* Re-insert the node */
1225 node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1227 btrfs_panic(fs_info, -EEXIST,
1228 "zoned: inconsistency in ordered tree at offset %llu",
1229 ordered->file_offset);
1231 node = tree_insert(&tree->tree, new->file_offset, &new->rb_node);
1233 btrfs_panic(fs_info, -EEXIST,
1234 "zoned: inconsistency in ordered tree at offset %llu",
1236 spin_unlock(&tree->lock);
1238 list_add_tail(&new->root_extent_list, &root->ordered_extents);
1239 root->nr_ordered_extents++;
1240 spin_unlock_irq(&root->ordered_extent_lock);
1244 int __init ordered_data_init(void)
1246 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1247 sizeof(struct btrfs_ordered_extent), 0,
1250 if (!btrfs_ordered_extent_cache)
1256 void __cold ordered_data_exit(void)
1258 kmem_cache_destroy(btrfs_ordered_extent_cache);