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/pagevec.h>
11 #include "transaction.h"
12 #include "btrfs_inode.h"
13 #include "extent_io.h"
15 #include "compression.h"
17 static struct kmem_cache *btrfs_ordered_extent_cache;
19 static u64 entry_end(struct btrfs_ordered_extent *entry)
21 if (entry->file_offset + entry->len < entry->file_offset)
23 return entry->file_offset + entry->len;
26 /* returns NULL if the insertion worked, or it returns the node it did find
29 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
32 struct rb_node **p = &root->rb_node;
33 struct rb_node *parent = NULL;
34 struct btrfs_ordered_extent *entry;
38 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
40 if (file_offset < entry->file_offset)
42 else if (file_offset >= entry_end(entry))
48 rb_link_node(node, parent, p);
49 rb_insert_color(node, root);
53 static void ordered_data_tree_panic(struct inode *inode, int errno,
56 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
57 btrfs_panic(fs_info, errno,
58 "Inconsistency in ordered tree at offset %llu", offset);
62 * look for a given offset in the tree, and if it can't be found return the
65 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
66 struct rb_node **prev_ret)
68 struct rb_node *n = root->rb_node;
69 struct rb_node *prev = NULL;
71 struct btrfs_ordered_extent *entry;
72 struct btrfs_ordered_extent *prev_entry = NULL;
75 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
79 if (file_offset < entry->file_offset)
81 else if (file_offset >= entry_end(entry))
89 while (prev && file_offset >= entry_end(prev_entry)) {
93 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
95 if (file_offset < entry_end(prev_entry))
101 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
103 while (prev && file_offset < entry_end(prev_entry)) {
104 test = rb_prev(prev);
107 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
116 * helper to check if a given offset is inside a given entry
118 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
120 if (file_offset < entry->file_offset ||
121 entry->file_offset + entry->len <= file_offset)
126 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
129 if (file_offset + len <= entry->file_offset ||
130 entry->file_offset + entry->len <= file_offset)
136 * look find the first ordered struct that has this offset, otherwise
137 * the first one less than this offset
139 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
142 struct rb_root *root = &tree->tree;
143 struct rb_node *prev = NULL;
145 struct btrfs_ordered_extent *entry;
148 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
150 if (offset_in_entry(entry, file_offset))
153 ret = __tree_search(root, file_offset, &prev);
161 /* allocate and add a new ordered_extent into the per-inode tree.
162 * file_offset is the logical offset in the file
164 * start is the disk block number of an extent already reserved in the
165 * extent allocation tree
167 * len is the length of the extent
169 * The tree is given a single reference on the ordered extent that was
172 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
173 u64 start, u64 len, u64 disk_len,
174 int type, int dio, int compress_type)
176 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
177 struct btrfs_root *root = BTRFS_I(inode)->root;
178 struct btrfs_ordered_inode_tree *tree;
179 struct rb_node *node;
180 struct btrfs_ordered_extent *entry;
182 tree = &BTRFS_I(inode)->ordered_tree;
183 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
187 entry->file_offset = file_offset;
188 entry->start = start;
190 entry->disk_len = disk_len;
191 entry->bytes_left = len;
192 entry->inode = igrab(inode);
193 entry->compress_type = compress_type;
194 entry->truncated_len = (u64)-1;
195 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
196 set_bit(type, &entry->flags);
199 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
201 /* one ref for the tree */
202 refcount_set(&entry->refs, 1);
203 init_waitqueue_head(&entry->wait);
204 INIT_LIST_HEAD(&entry->list);
205 INIT_LIST_HEAD(&entry->root_extent_list);
206 INIT_LIST_HEAD(&entry->work_list);
207 init_completion(&entry->completion);
208 INIT_LIST_HEAD(&entry->log_list);
209 INIT_LIST_HEAD(&entry->trans_list);
211 trace_btrfs_ordered_extent_add(inode, entry);
213 spin_lock_irq(&tree->lock);
214 node = tree_insert(&tree->tree, file_offset,
217 ordered_data_tree_panic(inode, -EEXIST, file_offset);
218 spin_unlock_irq(&tree->lock);
220 spin_lock(&root->ordered_extent_lock);
221 list_add_tail(&entry->root_extent_list,
222 &root->ordered_extents);
223 root->nr_ordered_extents++;
224 if (root->nr_ordered_extents == 1) {
225 spin_lock(&fs_info->ordered_root_lock);
226 BUG_ON(!list_empty(&root->ordered_root));
227 list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
228 spin_unlock(&fs_info->ordered_root_lock);
230 spin_unlock(&root->ordered_extent_lock);
233 * We don't need the count_max_extents here, we can assume that all of
234 * that work has been done at higher layers, so this is truly the
235 * smallest the extent is going to get.
237 spin_lock(&BTRFS_I(inode)->lock);
238 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
239 spin_unlock(&BTRFS_I(inode)->lock);
244 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
245 u64 start, u64 len, u64 disk_len, int type)
247 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
249 BTRFS_COMPRESS_NONE);
252 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
253 u64 start, u64 len, u64 disk_len, int type)
255 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
257 BTRFS_COMPRESS_NONE);
260 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
261 u64 start, u64 len, u64 disk_len,
262 int type, int compress_type)
264 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
270 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
271 * when an ordered extent is finished. If the list covers more than one
272 * ordered extent, it is split across multiples.
274 void btrfs_add_ordered_sum(struct inode *inode,
275 struct btrfs_ordered_extent *entry,
276 struct btrfs_ordered_sum *sum)
278 struct btrfs_ordered_inode_tree *tree;
280 tree = &BTRFS_I(inode)->ordered_tree;
281 spin_lock_irq(&tree->lock);
282 list_add_tail(&sum->list, &entry->list);
283 spin_unlock_irq(&tree->lock);
287 * this is used to account for finished IO across a given range
288 * of the file. The IO may span ordered extents. If
289 * a given ordered_extent is completely done, 1 is returned, otherwise
292 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
293 * to make sure this function only returns 1 once for a given ordered extent.
295 * file_offset is updated to one byte past the range that is recorded as
296 * complete. This allows you to walk forward in the file.
298 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
299 struct btrfs_ordered_extent **cached,
300 u64 *file_offset, u64 io_size, int uptodate)
302 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
303 struct btrfs_ordered_inode_tree *tree;
304 struct rb_node *node;
305 struct btrfs_ordered_extent *entry = NULL;
312 tree = &BTRFS_I(inode)->ordered_tree;
313 spin_lock_irqsave(&tree->lock, flags);
314 node = tree_search(tree, *file_offset);
320 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
321 if (!offset_in_entry(entry, *file_offset)) {
326 dec_start = max(*file_offset, entry->file_offset);
327 dec_end = min(*file_offset + io_size, entry->file_offset +
329 *file_offset = dec_end;
330 if (dec_start > dec_end) {
331 btrfs_crit(fs_info, "bad ordering dec_start %llu end %llu",
334 to_dec = dec_end - dec_start;
335 if (to_dec > entry->bytes_left) {
337 "bad ordered accounting left %llu size %llu",
338 entry->bytes_left, to_dec);
340 entry->bytes_left -= to_dec;
342 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
344 if (entry->bytes_left == 0) {
345 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
346 /* test_and_set_bit implies a barrier */
347 cond_wake_up_nomb(&entry->wait);
352 if (!ret && cached && entry) {
354 refcount_inc(&entry->refs);
356 spin_unlock_irqrestore(&tree->lock, flags);
361 * this is used to account for finished IO across a given range
362 * of the file. The IO should not span ordered extents. If
363 * a given ordered_extent is completely done, 1 is returned, otherwise
366 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
367 * to make sure this function only returns 1 once for a given ordered extent.
369 int btrfs_dec_test_ordered_pending(struct inode *inode,
370 struct btrfs_ordered_extent **cached,
371 u64 file_offset, u64 io_size, int uptodate)
373 struct btrfs_ordered_inode_tree *tree;
374 struct rb_node *node;
375 struct btrfs_ordered_extent *entry = NULL;
379 tree = &BTRFS_I(inode)->ordered_tree;
380 spin_lock_irqsave(&tree->lock, flags);
381 if (cached && *cached) {
386 node = tree_search(tree, file_offset);
392 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
394 if (!offset_in_entry(entry, file_offset)) {
399 if (io_size > entry->bytes_left) {
400 btrfs_crit(BTRFS_I(inode)->root->fs_info,
401 "bad ordered accounting left %llu size %llu",
402 entry->bytes_left, io_size);
404 entry->bytes_left -= io_size;
406 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
408 if (entry->bytes_left == 0) {
409 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
410 /* test_and_set_bit implies a barrier */
411 cond_wake_up_nomb(&entry->wait);
416 if (!ret && cached && entry) {
418 refcount_inc(&entry->refs);
420 spin_unlock_irqrestore(&tree->lock, flags);
424 /* Needs to either be called under a log transaction or the log_mutex */
425 void btrfs_get_logged_extents(struct btrfs_inode *inode,
426 struct list_head *logged_list,
430 struct btrfs_ordered_inode_tree *tree;
431 struct btrfs_ordered_extent *ordered;
433 struct rb_node *prev;
435 tree = &inode->ordered_tree;
436 spin_lock_irq(&tree->lock);
437 n = __tree_search(&tree->tree, end, &prev);
440 for (; n; n = rb_prev(n)) {
441 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
442 if (ordered->file_offset > end)
444 if (entry_end(ordered) <= start)
446 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
448 list_add(&ordered->log_list, logged_list);
449 refcount_inc(&ordered->refs);
451 spin_unlock_irq(&tree->lock);
454 void btrfs_put_logged_extents(struct list_head *logged_list)
456 struct btrfs_ordered_extent *ordered;
458 while (!list_empty(logged_list)) {
459 ordered = list_first_entry(logged_list,
460 struct btrfs_ordered_extent,
462 list_del_init(&ordered->log_list);
463 btrfs_put_ordered_extent(ordered);
467 void btrfs_submit_logged_extents(struct list_head *logged_list,
468 struct btrfs_root *log)
470 int index = log->log_transid % 2;
472 spin_lock_irq(&log->log_extents_lock[index]);
473 list_splice_tail(logged_list, &log->logged_list[index]);
474 spin_unlock_irq(&log->log_extents_lock[index]);
477 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
478 struct btrfs_root *log, u64 transid)
480 struct btrfs_ordered_extent *ordered;
481 int index = transid % 2;
483 spin_lock_irq(&log->log_extents_lock[index]);
484 while (!list_empty(&log->logged_list[index])) {
486 ordered = list_first_entry(&log->logged_list[index],
487 struct btrfs_ordered_extent,
489 list_del_init(&ordered->log_list);
490 inode = ordered->inode;
491 spin_unlock_irq(&log->log_extents_lock[index]);
493 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
494 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
495 u64 start = ordered->file_offset;
496 u64 end = ordered->file_offset + ordered->len - 1;
499 filemap_fdatawrite_range(inode->i_mapping, start, end);
501 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
505 * In order to keep us from losing our ordered extent
506 * information when committing the transaction we have to make
507 * sure that any logged extents are completed when we go to
508 * commit the transaction. To do this we simply increase the
509 * current transactions pending_ordered counter and decrement it
510 * when the ordered extent completes.
512 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
513 struct btrfs_ordered_inode_tree *tree;
515 tree = &BTRFS_I(inode)->ordered_tree;
516 spin_lock_irq(&tree->lock);
517 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
518 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
519 atomic_inc(&trans->transaction->pending_ordered);
521 spin_unlock_irq(&tree->lock);
523 btrfs_put_ordered_extent(ordered);
524 spin_lock_irq(&log->log_extents_lock[index]);
526 spin_unlock_irq(&log->log_extents_lock[index]);
529 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
531 struct btrfs_ordered_extent *ordered;
532 int index = transid % 2;
534 spin_lock_irq(&log->log_extents_lock[index]);
535 while (!list_empty(&log->logged_list[index])) {
536 ordered = list_first_entry(&log->logged_list[index],
537 struct btrfs_ordered_extent,
539 list_del_init(&ordered->log_list);
540 spin_unlock_irq(&log->log_extents_lock[index]);
541 btrfs_put_ordered_extent(ordered);
542 spin_lock_irq(&log->log_extents_lock[index]);
544 spin_unlock_irq(&log->log_extents_lock[index]);
548 * used to drop a reference on an ordered extent. This will free
549 * the extent if the last reference is dropped
551 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
553 struct list_head *cur;
554 struct btrfs_ordered_sum *sum;
556 trace_btrfs_ordered_extent_put(entry->inode, entry);
558 if (refcount_dec_and_test(&entry->refs)) {
559 ASSERT(list_empty(&entry->log_list));
560 ASSERT(list_empty(&entry->trans_list));
561 ASSERT(list_empty(&entry->root_extent_list));
562 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
564 btrfs_add_delayed_iput(entry->inode);
565 while (!list_empty(&entry->list)) {
566 cur = entry->list.next;
567 sum = list_entry(cur, struct btrfs_ordered_sum, list);
568 list_del(&sum->list);
571 kmem_cache_free(btrfs_ordered_extent_cache, entry);
576 * remove an ordered extent from the tree. No references are dropped
577 * and waiters are woken up.
579 void btrfs_remove_ordered_extent(struct inode *inode,
580 struct btrfs_ordered_extent *entry)
582 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
583 struct btrfs_ordered_inode_tree *tree;
584 struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
585 struct btrfs_root *root = btrfs_inode->root;
586 struct rb_node *node;
587 bool dec_pending_ordered = false;
589 /* This is paired with btrfs_add_ordered_extent. */
590 spin_lock(&btrfs_inode->lock);
591 btrfs_mod_outstanding_extents(btrfs_inode, -1);
592 spin_unlock(&btrfs_inode->lock);
593 if (root != fs_info->tree_root)
594 btrfs_delalloc_release_metadata(btrfs_inode, entry->len, false);
596 tree = &btrfs_inode->ordered_tree;
597 spin_lock_irq(&tree->lock);
598 node = &entry->rb_node;
599 rb_erase(node, &tree->tree);
601 if (tree->last == node)
603 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
604 if (test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags))
605 dec_pending_ordered = true;
606 spin_unlock_irq(&tree->lock);
609 * The current running transaction is waiting on us, we need to let it
610 * know that we're complete and wake it up.
612 if (dec_pending_ordered) {
613 struct btrfs_transaction *trans;
616 * The checks for trans are just a formality, it should be set,
617 * but if it isn't we don't want to deref/assert under the spin
618 * lock, so be nice and check if trans is set, but ASSERT() so
619 * if it isn't set a developer will notice.
621 spin_lock(&fs_info->trans_lock);
622 trans = fs_info->running_transaction;
624 refcount_inc(&trans->use_count);
625 spin_unlock(&fs_info->trans_lock);
629 if (atomic_dec_and_test(&trans->pending_ordered))
630 wake_up(&trans->pending_wait);
631 btrfs_put_transaction(trans);
635 spin_lock(&root->ordered_extent_lock);
636 list_del_init(&entry->root_extent_list);
637 root->nr_ordered_extents--;
639 trace_btrfs_ordered_extent_remove(inode, entry);
641 if (!root->nr_ordered_extents) {
642 spin_lock(&fs_info->ordered_root_lock);
643 BUG_ON(list_empty(&root->ordered_root));
644 list_del_init(&root->ordered_root);
645 spin_unlock(&fs_info->ordered_root_lock);
647 spin_unlock(&root->ordered_extent_lock);
648 wake_up(&entry->wait);
651 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
653 struct btrfs_ordered_extent *ordered;
655 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
656 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
657 complete(&ordered->completion);
661 * wait for all the ordered extents in a root. This is done when balancing
662 * space between drives.
664 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
665 const u64 range_start, const u64 range_len)
667 struct btrfs_fs_info *fs_info = root->fs_info;
671 struct btrfs_ordered_extent *ordered, *next;
673 const u64 range_end = range_start + range_len;
675 mutex_lock(&root->ordered_extent_mutex);
676 spin_lock(&root->ordered_extent_lock);
677 list_splice_init(&root->ordered_extents, &splice);
678 while (!list_empty(&splice) && nr) {
679 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
682 if (range_end <= ordered->start ||
683 ordered->start + ordered->disk_len <= range_start) {
684 list_move_tail(&ordered->root_extent_list, &skipped);
685 cond_resched_lock(&root->ordered_extent_lock);
689 list_move_tail(&ordered->root_extent_list,
690 &root->ordered_extents);
691 refcount_inc(&ordered->refs);
692 spin_unlock(&root->ordered_extent_lock);
694 btrfs_init_work(&ordered->flush_work,
695 btrfs_flush_delalloc_helper,
696 btrfs_run_ordered_extent_work, NULL, NULL);
697 list_add_tail(&ordered->work_list, &works);
698 btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
701 spin_lock(&root->ordered_extent_lock);
706 list_splice_tail(&skipped, &root->ordered_extents);
707 list_splice_tail(&splice, &root->ordered_extents);
708 spin_unlock(&root->ordered_extent_lock);
710 list_for_each_entry_safe(ordered, next, &works, work_list) {
711 list_del_init(&ordered->work_list);
712 wait_for_completion(&ordered->completion);
713 btrfs_put_ordered_extent(ordered);
716 mutex_unlock(&root->ordered_extent_mutex);
721 u64 btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
722 const u64 range_start, const u64 range_len)
724 struct btrfs_root *root;
725 struct list_head splice;
729 INIT_LIST_HEAD(&splice);
731 mutex_lock(&fs_info->ordered_operations_mutex);
732 spin_lock(&fs_info->ordered_root_lock);
733 list_splice_init(&fs_info->ordered_roots, &splice);
734 while (!list_empty(&splice) && nr) {
735 root = list_first_entry(&splice, struct btrfs_root,
737 root = btrfs_grab_fs_root(root);
739 list_move_tail(&root->ordered_root,
740 &fs_info->ordered_roots);
741 spin_unlock(&fs_info->ordered_root_lock);
743 done = btrfs_wait_ordered_extents(root, nr,
744 range_start, range_len);
745 btrfs_put_fs_root(root);
748 spin_lock(&fs_info->ordered_root_lock);
753 list_splice_tail(&splice, &fs_info->ordered_roots);
754 spin_unlock(&fs_info->ordered_root_lock);
755 mutex_unlock(&fs_info->ordered_operations_mutex);
761 * Used to start IO or wait for a given ordered extent to finish.
763 * If wait is one, this effectively waits on page writeback for all the pages
764 * in the extent, and it waits on the io completion code to insert
765 * metadata into the btree corresponding to the extent
767 void btrfs_start_ordered_extent(struct inode *inode,
768 struct btrfs_ordered_extent *entry,
771 u64 start = entry->file_offset;
772 u64 end = start + entry->len - 1;
774 trace_btrfs_ordered_extent_start(inode, entry);
777 * pages in the range can be dirty, clean or writeback. We
778 * start IO on any dirty ones so the wait doesn't stall waiting
779 * for the flusher thread to find them
781 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
782 filemap_fdatawrite_range(inode->i_mapping, start, end);
784 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
790 * Used to wait on ordered extents across a large range of bytes.
792 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
798 struct btrfs_ordered_extent *ordered;
800 if (start + len < start) {
801 orig_end = INT_LIMIT(loff_t);
803 orig_end = start + len - 1;
804 if (orig_end > INT_LIMIT(loff_t))
805 orig_end = INT_LIMIT(loff_t);
808 /* start IO across the range first to instantiate any delalloc
811 ret = btrfs_fdatawrite_range(inode, start, orig_end);
816 * If we have a writeback error don't return immediately. Wait first
817 * for any ordered extents that haven't completed yet. This is to make
818 * sure no one can dirty the same page ranges and call writepages()
819 * before the ordered extents complete - to avoid failures (-EEXIST)
820 * when adding the new ordered extents to the ordered tree.
822 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
826 ordered = btrfs_lookup_first_ordered_extent(inode, end);
829 if (ordered->file_offset > orig_end) {
830 btrfs_put_ordered_extent(ordered);
833 if (ordered->file_offset + ordered->len <= start) {
834 btrfs_put_ordered_extent(ordered);
837 btrfs_start_ordered_extent(inode, ordered, 1);
838 end = ordered->file_offset;
839 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
841 btrfs_put_ordered_extent(ordered);
842 if (ret || end == 0 || end == start)
846 return ret_wb ? ret_wb : ret;
850 * find an ordered extent corresponding to file_offset. return NULL if
851 * nothing is found, otherwise take a reference on the extent and return it
853 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
856 struct btrfs_ordered_inode_tree *tree;
857 struct rb_node *node;
858 struct btrfs_ordered_extent *entry = NULL;
860 tree = &BTRFS_I(inode)->ordered_tree;
861 spin_lock_irq(&tree->lock);
862 node = tree_search(tree, file_offset);
866 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
867 if (!offset_in_entry(entry, file_offset))
870 refcount_inc(&entry->refs);
872 spin_unlock_irq(&tree->lock);
876 /* Since the DIO code tries to lock a wide area we need to look for any ordered
877 * extents that exist in the range, rather than just the start of the range.
879 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
880 struct btrfs_inode *inode, u64 file_offset, u64 len)
882 struct btrfs_ordered_inode_tree *tree;
883 struct rb_node *node;
884 struct btrfs_ordered_extent *entry = NULL;
886 tree = &inode->ordered_tree;
887 spin_lock_irq(&tree->lock);
888 node = tree_search(tree, file_offset);
890 node = tree_search(tree, file_offset + len);
896 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
897 if (range_overlaps(entry, file_offset, len))
900 if (entry->file_offset >= file_offset + len) {
905 node = rb_next(node);
911 refcount_inc(&entry->refs);
912 spin_unlock_irq(&tree->lock);
916 bool btrfs_have_ordered_extents_in_range(struct inode *inode,
920 struct btrfs_ordered_extent *oe;
922 oe = btrfs_lookup_ordered_range(BTRFS_I(inode), file_offset, len);
924 btrfs_put_ordered_extent(oe);
931 * lookup and return any extent before 'file_offset'. NULL is returned
934 struct btrfs_ordered_extent *
935 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
937 struct btrfs_ordered_inode_tree *tree;
938 struct rb_node *node;
939 struct btrfs_ordered_extent *entry = NULL;
941 tree = &BTRFS_I(inode)->ordered_tree;
942 spin_lock_irq(&tree->lock);
943 node = tree_search(tree, file_offset);
947 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
948 refcount_inc(&entry->refs);
950 spin_unlock_irq(&tree->lock);
955 * After an extent is done, call this to conditionally update the on disk
956 * i_size. i_size is updated to cover any fully written part of the file.
958 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
959 struct btrfs_ordered_extent *ordered)
961 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
964 u64 i_size = i_size_read(inode);
965 struct rb_node *node;
966 struct rb_node *prev = NULL;
967 struct btrfs_ordered_extent *test;
969 u64 orig_offset = offset;
971 spin_lock_irq(&tree->lock);
973 offset = entry_end(ordered);
974 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
976 ordered->file_offset +
977 ordered->truncated_len);
979 offset = ALIGN(offset, btrfs_inode_sectorsize(inode));
981 disk_i_size = BTRFS_I(inode)->disk_i_size;
985 * If ordered is not NULL, then this is called from endio and
986 * disk_i_size will be updated by either truncate itself or any
987 * in-flight IOs which are inside the disk_i_size.
989 * Because btrfs_setsize() may set i_size with disk_i_size if truncate
990 * fails somehow, we need to make sure we have a precise disk_i_size by
991 * updating it as usual.
994 if (!ordered && disk_i_size > i_size) {
995 BTRFS_I(inode)->disk_i_size = orig_offset;
1001 * if the disk i_size is already at the inode->i_size, or
1002 * this ordered extent is inside the disk i_size, we're done
1004 if (disk_i_size == i_size)
1008 * We still need to update disk_i_size if outstanding_isize is greater
1011 if (offset <= disk_i_size &&
1012 (!ordered || ordered->outstanding_isize <= disk_i_size))
1016 * walk backward from this ordered extent to disk_i_size.
1017 * if we find an ordered extent then we can't update disk i_size
1021 node = rb_prev(&ordered->rb_node);
1023 prev = tree_search(tree, offset);
1025 * we insert file extents without involving ordered struct,
1026 * so there should be no ordered struct cover this offset
1029 test = rb_entry(prev, struct btrfs_ordered_extent,
1031 BUG_ON(offset_in_entry(test, offset));
1035 for (; node; node = rb_prev(node)) {
1036 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1038 /* We treat this entry as if it doesn't exist */
1039 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1042 if (entry_end(test) <= disk_i_size)
1044 if (test->file_offset >= i_size)
1048 * We don't update disk_i_size now, so record this undealt
1049 * i_size. Or we will not know the real i_size.
1051 if (test->outstanding_isize < offset)
1052 test->outstanding_isize = offset;
1054 ordered->outstanding_isize > test->outstanding_isize)
1055 test->outstanding_isize = ordered->outstanding_isize;
1058 new_i_size = min_t(u64, offset, i_size);
1061 * Some ordered extents may completed before the current one, and
1062 * we hold the real i_size in ->outstanding_isize.
1064 if (ordered && ordered->outstanding_isize > new_i_size)
1065 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1066 BTRFS_I(inode)->disk_i_size = new_i_size;
1070 * We need to do this because we can't remove ordered extents until
1071 * after the i_disk_size has been updated and then the inode has been
1072 * updated to reflect the change, so we need to tell anybody who finds
1073 * this ordered extent that we've already done all the real work, we
1074 * just haven't completed all the other work.
1077 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1078 spin_unlock_irq(&tree->lock);
1083 * search the ordered extents for one corresponding to 'offset' and
1084 * try to find a checksum. This is used because we allow pages to
1085 * be reclaimed before their checksum is actually put into the btree
1087 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1090 struct btrfs_ordered_sum *ordered_sum;
1091 struct btrfs_ordered_extent *ordered;
1092 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1093 unsigned long num_sectors;
1095 u32 sectorsize = btrfs_inode_sectorsize(inode);
1098 ordered = btrfs_lookup_ordered_extent(inode, offset);
1102 spin_lock_irq(&tree->lock);
1103 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1104 if (disk_bytenr >= ordered_sum->bytenr &&
1105 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1106 i = (disk_bytenr - ordered_sum->bytenr) >>
1107 inode->i_sb->s_blocksize_bits;
1108 num_sectors = ordered_sum->len >>
1109 inode->i_sb->s_blocksize_bits;
1110 num_sectors = min_t(int, len - index, num_sectors - i);
1111 memcpy(sum + index, ordered_sum->sums + i,
1114 index += (int)num_sectors;
1117 disk_bytenr += num_sectors * sectorsize;
1121 spin_unlock_irq(&tree->lock);
1122 btrfs_put_ordered_extent(ordered);
1126 int __init ordered_data_init(void)
1128 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1129 sizeof(struct btrfs_ordered_extent), 0,
1132 if (!btrfs_ordered_extent_cache)
1138 void __cold ordered_data_exit(void)
1140 kmem_cache_destroy(btrfs_ordered_extent_cache);