2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/slab.h>
20 #include <linux/blkdev.h>
21 #include <linux/writeback.h>
22 #include <linux/pagevec.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "extent_io.h"
28 #include "compression.h"
30 static struct kmem_cache *btrfs_ordered_extent_cache;
32 static u64 entry_end(struct btrfs_ordered_extent *entry)
34 if (entry->file_offset + entry->len < entry->file_offset)
36 return entry->file_offset + entry->len;
39 /* returns NULL if the insertion worked, or it returns the node it did find
42 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
45 struct rb_node **p = &root->rb_node;
46 struct rb_node *parent = NULL;
47 struct btrfs_ordered_extent *entry;
51 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
53 if (file_offset < entry->file_offset)
55 else if (file_offset >= entry_end(entry))
61 rb_link_node(node, parent, p);
62 rb_insert_color(node, root);
66 static void ordered_data_tree_panic(struct inode *inode, int errno,
69 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
70 btrfs_panic(fs_info, errno,
71 "Inconsistency in ordered tree at offset %llu", offset);
75 * look for a given offset in the tree, and if it can't be found return the
78 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
79 struct rb_node **prev_ret)
81 struct rb_node *n = root->rb_node;
82 struct rb_node *prev = NULL;
84 struct btrfs_ordered_extent *entry;
85 struct btrfs_ordered_extent *prev_entry = NULL;
88 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
92 if (file_offset < entry->file_offset)
94 else if (file_offset >= entry_end(entry))
102 while (prev && file_offset >= entry_end(prev_entry)) {
103 test = rb_next(prev);
106 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
108 if (file_offset < entry_end(prev_entry))
114 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
116 while (prev && file_offset < entry_end(prev_entry)) {
117 test = rb_prev(prev);
120 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
129 * helper to check if a given offset is inside a given entry
131 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
133 if (file_offset < entry->file_offset ||
134 entry->file_offset + entry->len <= file_offset)
139 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
142 if (file_offset + len <= entry->file_offset ||
143 entry->file_offset + entry->len <= file_offset)
149 * look find the first ordered struct that has this offset, otherwise
150 * the first one less than this offset
152 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
155 struct rb_root *root = &tree->tree;
156 struct rb_node *prev = NULL;
158 struct btrfs_ordered_extent *entry;
161 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
163 if (offset_in_entry(entry, file_offset))
166 ret = __tree_search(root, file_offset, &prev);
174 /* allocate and add a new ordered_extent into the per-inode tree.
175 * file_offset is the logical offset in the file
177 * start is the disk block number of an extent already reserved in the
178 * extent allocation tree
180 * len is the length of the extent
182 * The tree is given a single reference on the ordered extent that was
185 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
186 u64 start, u64 len, u64 disk_len,
187 int type, int dio, int compress_type)
189 struct btrfs_root *root = BTRFS_I(inode)->root;
190 struct btrfs_ordered_inode_tree *tree;
191 struct rb_node *node;
192 struct btrfs_ordered_extent *entry;
194 tree = &BTRFS_I(inode)->ordered_tree;
195 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
199 entry->file_offset = file_offset;
200 entry->start = start;
202 entry->disk_len = disk_len;
203 entry->bytes_left = len;
204 entry->inode = igrab(inode);
205 entry->compress_type = compress_type;
206 entry->truncated_len = (u64)-1;
207 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
208 set_bit(type, &entry->flags);
211 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
213 /* one ref for the tree */
214 atomic_set(&entry->refs, 1);
215 init_waitqueue_head(&entry->wait);
216 INIT_LIST_HEAD(&entry->list);
217 INIT_LIST_HEAD(&entry->root_extent_list);
218 INIT_LIST_HEAD(&entry->work_list);
219 init_completion(&entry->completion);
220 INIT_LIST_HEAD(&entry->log_list);
221 INIT_LIST_HEAD(&entry->trans_list);
223 trace_btrfs_ordered_extent_add(inode, entry);
225 spin_lock_irq(&tree->lock);
226 node = tree_insert(&tree->tree, file_offset,
229 ordered_data_tree_panic(inode, -EEXIST, file_offset);
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(&root->fs_info->ordered_root_lock);
238 BUG_ON(!list_empty(&root->ordered_root));
239 list_add_tail(&root->ordered_root,
240 &root->fs_info->ordered_roots);
241 spin_unlock(&root->fs_info->ordered_root_lock);
243 spin_unlock(&root->ordered_extent_lock);
248 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
249 u64 start, u64 len, u64 disk_len, int type)
251 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
253 BTRFS_COMPRESS_NONE);
256 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
257 u64 start, u64 len, u64 disk_len, int type)
259 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
261 BTRFS_COMPRESS_NONE);
264 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
265 u64 start, u64 len, u64 disk_len,
266 int type, int compress_type)
268 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
274 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
275 * when an ordered extent is finished. If the list covers more than one
276 * ordered extent, it is split across multiples.
278 void btrfs_add_ordered_sum(struct inode *inode,
279 struct btrfs_ordered_extent *entry,
280 struct btrfs_ordered_sum *sum)
282 struct btrfs_ordered_inode_tree *tree;
284 tree = &BTRFS_I(inode)->ordered_tree;
285 spin_lock_irq(&tree->lock);
286 list_add_tail(&sum->list, &entry->list);
287 spin_unlock_irq(&tree->lock);
291 * this is used to account for finished IO across a given range
292 * of the file. The IO may span ordered extents. If
293 * a given ordered_extent is completely done, 1 is returned, otherwise
296 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
297 * to make sure this function only returns 1 once for a given ordered extent.
299 * file_offset is updated to one byte past the range that is recorded as
300 * complete. This allows you to walk forward in the file.
302 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
303 struct btrfs_ordered_extent **cached,
304 u64 *file_offset, u64 io_size, int uptodate)
306 struct btrfs_ordered_inode_tree *tree;
307 struct rb_node *node;
308 struct btrfs_ordered_extent *entry = NULL;
315 tree = &BTRFS_I(inode)->ordered_tree;
316 spin_lock_irqsave(&tree->lock, flags);
317 node = tree_search(tree, *file_offset);
323 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
324 if (!offset_in_entry(entry, *file_offset)) {
329 dec_start = max(*file_offset, entry->file_offset);
330 dec_end = min(*file_offset + io_size, entry->file_offset +
332 *file_offset = dec_end;
333 if (dec_start > dec_end) {
334 btrfs_crit(BTRFS_I(inode)->root->fs_info,
335 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
337 to_dec = dec_end - dec_start;
338 if (to_dec > entry->bytes_left) {
339 btrfs_crit(BTRFS_I(inode)->root->fs_info,
340 "bad ordered accounting left %llu size %llu",
341 entry->bytes_left, to_dec);
343 entry->bytes_left -= to_dec;
345 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
347 if (entry->bytes_left == 0) {
348 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
350 * Implicit memory barrier after test_and_set_bit
352 if (waitqueue_active(&entry->wait))
353 wake_up(&entry->wait);
358 if (!ret && cached && entry) {
360 atomic_inc(&entry->refs);
362 spin_unlock_irqrestore(&tree->lock, flags);
367 * this is used to account for finished IO across a given range
368 * of the file. The IO should not span ordered extents. If
369 * a given ordered_extent is completely done, 1 is returned, otherwise
372 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
373 * to make sure this function only returns 1 once for a given ordered extent.
375 int btrfs_dec_test_ordered_pending(struct inode *inode,
376 struct btrfs_ordered_extent **cached,
377 u64 file_offset, u64 io_size, int uptodate)
379 struct btrfs_ordered_inode_tree *tree;
380 struct rb_node *node;
381 struct btrfs_ordered_extent *entry = NULL;
385 tree = &BTRFS_I(inode)->ordered_tree;
386 spin_lock_irqsave(&tree->lock, flags);
387 if (cached && *cached) {
392 node = tree_search(tree, file_offset);
398 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
400 if (!offset_in_entry(entry, file_offset)) {
405 if (io_size > entry->bytes_left) {
406 btrfs_crit(BTRFS_I(inode)->root->fs_info,
407 "bad ordered accounting left %llu size %llu",
408 entry->bytes_left, io_size);
410 entry->bytes_left -= io_size;
412 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
414 if (entry->bytes_left == 0) {
415 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
417 * Implicit memory barrier after test_and_set_bit
419 if (waitqueue_active(&entry->wait))
420 wake_up(&entry->wait);
425 if (!ret && cached && entry) {
427 atomic_inc(&entry->refs);
429 spin_unlock_irqrestore(&tree->lock, flags);
433 /* Needs to either be called under a log transaction or the log_mutex */
434 void btrfs_get_logged_extents(struct inode *inode,
435 struct list_head *logged_list,
439 struct btrfs_ordered_inode_tree *tree;
440 struct btrfs_ordered_extent *ordered;
442 struct rb_node *prev;
444 tree = &BTRFS_I(inode)->ordered_tree;
445 spin_lock_irq(&tree->lock);
446 n = __tree_search(&tree->tree, end, &prev);
449 for (; n; n = rb_prev(n)) {
450 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
451 if (ordered->file_offset > end)
453 if (entry_end(ordered) <= start)
455 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
457 list_add(&ordered->log_list, logged_list);
458 atomic_inc(&ordered->refs);
460 spin_unlock_irq(&tree->lock);
463 void btrfs_put_logged_extents(struct list_head *logged_list)
465 struct btrfs_ordered_extent *ordered;
467 while (!list_empty(logged_list)) {
468 ordered = list_first_entry(logged_list,
469 struct btrfs_ordered_extent,
471 list_del_init(&ordered->log_list);
472 btrfs_put_ordered_extent(ordered);
476 void btrfs_submit_logged_extents(struct list_head *logged_list,
477 struct btrfs_root *log)
479 int index = log->log_transid % 2;
481 spin_lock_irq(&log->log_extents_lock[index]);
482 list_splice_tail(logged_list, &log->logged_list[index]);
483 spin_unlock_irq(&log->log_extents_lock[index]);
486 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
487 struct btrfs_root *log, u64 transid)
489 struct btrfs_ordered_extent *ordered;
490 int index = transid % 2;
492 spin_lock_irq(&log->log_extents_lock[index]);
493 while (!list_empty(&log->logged_list[index])) {
495 ordered = list_first_entry(&log->logged_list[index],
496 struct btrfs_ordered_extent,
498 list_del_init(&ordered->log_list);
499 inode = ordered->inode;
500 spin_unlock_irq(&log->log_extents_lock[index]);
502 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
503 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
504 u64 start = ordered->file_offset;
505 u64 end = ordered->file_offset + ordered->len - 1;
508 filemap_fdatawrite_range(inode->i_mapping, start, end);
510 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
514 * In order to keep us from losing our ordered extent
515 * information when committing the transaction we have to make
516 * sure that any logged extents are completed when we go to
517 * commit the transaction. To do this we simply increase the
518 * current transactions pending_ordered counter and decrement it
519 * when the ordered extent completes.
521 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
522 struct btrfs_ordered_inode_tree *tree;
524 tree = &BTRFS_I(inode)->ordered_tree;
525 spin_lock_irq(&tree->lock);
526 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
527 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
528 atomic_inc(&trans->transaction->pending_ordered);
530 spin_unlock_irq(&tree->lock);
532 btrfs_put_ordered_extent(ordered);
533 spin_lock_irq(&log->log_extents_lock[index]);
535 spin_unlock_irq(&log->log_extents_lock[index]);
538 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
540 struct btrfs_ordered_extent *ordered;
541 int index = transid % 2;
543 spin_lock_irq(&log->log_extents_lock[index]);
544 while (!list_empty(&log->logged_list[index])) {
545 ordered = list_first_entry(&log->logged_list[index],
546 struct btrfs_ordered_extent,
548 list_del_init(&ordered->log_list);
549 spin_unlock_irq(&log->log_extents_lock[index]);
550 btrfs_put_ordered_extent(ordered);
551 spin_lock_irq(&log->log_extents_lock[index]);
553 spin_unlock_irq(&log->log_extents_lock[index]);
557 * used to drop a reference on an ordered extent. This will free
558 * the extent if the last reference is dropped
560 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
562 struct list_head *cur;
563 struct btrfs_ordered_sum *sum;
565 trace_btrfs_ordered_extent_put(entry->inode, entry);
567 if (atomic_dec_and_test(&entry->refs)) {
568 ASSERT(list_empty(&entry->log_list));
569 ASSERT(list_empty(&entry->trans_list));
570 ASSERT(list_empty(&entry->root_extent_list));
571 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
573 btrfs_add_delayed_iput(entry->inode);
574 while (!list_empty(&entry->list)) {
575 cur = entry->list.next;
576 sum = list_entry(cur, struct btrfs_ordered_sum, list);
577 list_del(&sum->list);
580 kmem_cache_free(btrfs_ordered_extent_cache, entry);
585 * remove an ordered extent from the tree. No references are dropped
586 * and waiters are woken up.
588 void btrfs_remove_ordered_extent(struct inode *inode,
589 struct btrfs_ordered_extent *entry)
591 struct btrfs_ordered_inode_tree *tree;
592 struct btrfs_root *root = BTRFS_I(inode)->root;
593 struct rb_node *node;
594 bool dec_pending_ordered = false;
596 tree = &BTRFS_I(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(&root->fs_info->trans_lock);
622 trans = root->fs_info->running_transaction;
624 atomic_inc(&trans->use_count);
625 spin_unlock(&root->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(&root->fs_info->ordered_root_lock);
643 BUG_ON(list_empty(&root->ordered_root));
644 list_del_init(&root->ordered_root);
645 spin_unlock(&root->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 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr,
665 const u64 range_start, const u64 range_len)
670 struct btrfs_ordered_extent *ordered, *next;
672 const u64 range_end = range_start + range_len;
674 mutex_lock(&root->ordered_extent_mutex);
675 spin_lock(&root->ordered_extent_lock);
676 list_splice_init(&root->ordered_extents, &splice);
677 while (!list_empty(&splice) && nr) {
678 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
681 if (range_end <= ordered->start ||
682 ordered->start + ordered->disk_len <= range_start) {
683 list_move_tail(&ordered->root_extent_list, &skipped);
684 cond_resched_lock(&root->ordered_extent_lock);
688 list_move_tail(&ordered->root_extent_list,
689 &root->ordered_extents);
690 atomic_inc(&ordered->refs);
691 spin_unlock(&root->ordered_extent_lock);
693 btrfs_init_work(&ordered->flush_work,
694 btrfs_flush_delalloc_helper,
695 btrfs_run_ordered_extent_work, NULL, NULL);
696 list_add_tail(&ordered->work_list, &works);
697 btrfs_queue_work(root->fs_info->flush_workers,
698 &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 int btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int 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);
754 list_splice_tail(&splice, &fs_info->ordered_roots);
755 spin_unlock(&fs_info->ordered_root_lock);
756 mutex_unlock(&fs_info->ordered_operations_mutex);
762 * Used to start IO or wait for a given ordered extent to finish.
764 * If wait is one, this effectively waits on page writeback for all the pages
765 * in the extent, and it waits on the io completion code to insert
766 * metadata into the btree corresponding to the extent
768 void btrfs_start_ordered_extent(struct inode *inode,
769 struct btrfs_ordered_extent *entry,
772 u64 start = entry->file_offset;
773 u64 end = start + entry->len - 1;
775 trace_btrfs_ordered_extent_start(inode, entry);
778 * pages in the range can be dirty, clean or writeback. We
779 * start IO on any dirty ones so the wait doesn't stall waiting
780 * for the flusher thread to find them
782 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
783 filemap_fdatawrite_range(inode->i_mapping, start, end);
785 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
791 * Used to wait on ordered extents across a large range of bytes.
793 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
799 struct btrfs_ordered_extent *ordered;
801 if (start + len < start) {
802 orig_end = INT_LIMIT(loff_t);
804 orig_end = start + len - 1;
805 if (orig_end > INT_LIMIT(loff_t))
806 orig_end = INT_LIMIT(loff_t);
809 /* start IO across the range first to instantiate any delalloc
812 ret = btrfs_fdatawrite_range(inode, start, orig_end);
817 * If we have a writeback error don't return immediately. Wait first
818 * for any ordered extents that haven't completed yet. This is to make
819 * sure no one can dirty the same page ranges and call writepages()
820 * before the ordered extents complete - to avoid failures (-EEXIST)
821 * when adding the new ordered extents to the ordered tree.
823 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
827 ordered = btrfs_lookup_first_ordered_extent(inode, end);
830 if (ordered->file_offset > orig_end) {
831 btrfs_put_ordered_extent(ordered);
834 if (ordered->file_offset + ordered->len <= start) {
835 btrfs_put_ordered_extent(ordered);
838 btrfs_start_ordered_extent(inode, ordered, 1);
839 end = ordered->file_offset;
841 * If the ordered extent had an error save the error but don't
842 * exit without waiting first for all other ordered extents in
843 * the range to complete.
845 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
847 btrfs_put_ordered_extent(ordered);
848 if (end == 0 || end == start)
852 return ret_wb ? ret_wb : ret;
856 * find an ordered extent corresponding to file_offset. return NULL if
857 * nothing is found, otherwise take a reference on the extent and return it
859 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
862 struct btrfs_ordered_inode_tree *tree;
863 struct rb_node *node;
864 struct btrfs_ordered_extent *entry = NULL;
866 tree = &BTRFS_I(inode)->ordered_tree;
867 spin_lock_irq(&tree->lock);
868 node = tree_search(tree, file_offset);
872 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
873 if (!offset_in_entry(entry, file_offset))
876 atomic_inc(&entry->refs);
878 spin_unlock_irq(&tree->lock);
882 /* Since the DIO code tries to lock a wide area we need to look for any ordered
883 * extents that exist in the range, rather than just the start of the range.
885 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
889 struct btrfs_ordered_inode_tree *tree;
890 struct rb_node *node;
891 struct btrfs_ordered_extent *entry = NULL;
893 tree = &BTRFS_I(inode)->ordered_tree;
894 spin_lock_irq(&tree->lock);
895 node = tree_search(tree, file_offset);
897 node = tree_search(tree, file_offset + len);
903 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
904 if (range_overlaps(entry, file_offset, len))
907 if (entry->file_offset >= file_offset + len) {
912 node = rb_next(node);
918 atomic_inc(&entry->refs);
919 spin_unlock_irq(&tree->lock);
923 bool btrfs_have_ordered_extents_in_range(struct inode *inode,
927 struct btrfs_ordered_extent *oe;
929 oe = btrfs_lookup_ordered_range(inode, file_offset, len);
931 btrfs_put_ordered_extent(oe);
938 * lookup and return any extent before 'file_offset'. NULL is returned
941 struct btrfs_ordered_extent *
942 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
944 struct btrfs_ordered_inode_tree *tree;
945 struct rb_node *node;
946 struct btrfs_ordered_extent *entry = NULL;
948 tree = &BTRFS_I(inode)->ordered_tree;
949 spin_lock_irq(&tree->lock);
950 node = tree_search(tree, file_offset);
954 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
955 atomic_inc(&entry->refs);
957 spin_unlock_irq(&tree->lock);
962 * After an extent is done, call this to conditionally update the on disk
963 * i_size. i_size is updated to cover any fully written part of the file.
965 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
966 struct btrfs_ordered_extent *ordered)
968 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
971 u64 i_size = i_size_read(inode);
972 struct rb_node *node;
973 struct rb_node *prev = NULL;
974 struct btrfs_ordered_extent *test;
976 u64 orig_offset = offset;
978 spin_lock_irq(&tree->lock);
980 offset = entry_end(ordered);
981 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
983 ordered->file_offset +
984 ordered->truncated_len);
986 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
988 disk_i_size = BTRFS_I(inode)->disk_i_size;
991 if (disk_i_size > i_size) {
992 BTRFS_I(inode)->disk_i_size = orig_offset;
998 * if the disk i_size is already at the inode->i_size, or
999 * this ordered extent is inside the disk i_size, we're done
1001 if (disk_i_size == i_size)
1005 * We still need to update disk_i_size if outstanding_isize is greater
1008 if (offset <= disk_i_size &&
1009 (!ordered || ordered->outstanding_isize <= disk_i_size))
1013 * walk backward from this ordered extent to disk_i_size.
1014 * if we find an ordered extent then we can't update disk i_size
1018 node = rb_prev(&ordered->rb_node);
1020 prev = tree_search(tree, offset);
1022 * we insert file extents without involving ordered struct,
1023 * so there should be no ordered struct cover this offset
1026 test = rb_entry(prev, struct btrfs_ordered_extent,
1028 BUG_ON(offset_in_entry(test, offset));
1032 for (; node; node = rb_prev(node)) {
1033 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1035 /* We treat this entry as if it doesn't exist */
1036 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1038 if (test->file_offset + test->len <= disk_i_size)
1040 if (test->file_offset >= i_size)
1042 if (entry_end(test) > disk_i_size) {
1044 * we don't update disk_i_size now, so record this
1045 * undealt i_size. Or we will not know the real
1048 if (test->outstanding_isize < offset)
1049 test->outstanding_isize = offset;
1051 ordered->outstanding_isize >
1052 test->outstanding_isize)
1053 test->outstanding_isize =
1054 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_I(inode)->root->sectorsize;
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 ordered_data_exit(void)
1140 kmem_cache_destroy(btrfs_ordered_extent_cache);