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"
29 static struct kmem_cache *btrfs_ordered_extent_cache;
31 static u64 entry_end(struct btrfs_ordered_extent *entry)
33 if (entry->file_offset + entry->len < entry->file_offset)
35 return entry->file_offset + entry->len;
38 /* returns NULL if the insertion worked, or it returns the node it did find
41 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
44 struct rb_node **p = &root->rb_node;
45 struct rb_node *parent = NULL;
46 struct btrfs_ordered_extent *entry;
50 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
52 if (file_offset < entry->file_offset)
54 else if (file_offset >= entry_end(entry))
60 rb_link_node(node, parent, p);
61 rb_insert_color(node, root);
65 static void ordered_data_tree_panic(struct inode *inode, int errno,
68 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
69 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
74 * look for a given offset in the tree, and if it can't be found return the
77 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
78 struct rb_node **prev_ret)
80 struct rb_node *n = root->rb_node;
81 struct rb_node *prev = NULL;
83 struct btrfs_ordered_extent *entry;
84 struct btrfs_ordered_extent *prev_entry = NULL;
87 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
91 if (file_offset < entry->file_offset)
93 else if (file_offset >= entry_end(entry))
101 while (prev && file_offset >= entry_end(prev_entry)) {
102 test = rb_next(prev);
105 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
107 if (file_offset < entry_end(prev_entry))
113 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
115 while (prev && file_offset < entry_end(prev_entry)) {
116 test = rb_prev(prev);
119 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
128 * helper to check if a given offset is inside a given entry
130 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
132 if (file_offset < entry->file_offset ||
133 entry->file_offset + entry->len <= file_offset)
138 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
141 if (file_offset + len <= entry->file_offset ||
142 entry->file_offset + entry->len <= file_offset)
148 * look find the first ordered struct that has this offset, otherwise
149 * the first one less than this offset
151 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
154 struct rb_root *root = &tree->tree;
155 struct rb_node *prev = NULL;
157 struct btrfs_ordered_extent *entry;
160 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
162 if (offset_in_entry(entry, file_offset))
165 ret = __tree_search(root, file_offset, &prev);
173 /* allocate and add a new ordered_extent into the per-inode tree.
174 * file_offset is the logical offset in the file
176 * start is the disk block number of an extent already reserved in the
177 * extent allocation tree
179 * len is the length of the extent
181 * The tree is given a single reference on the ordered extent that was
184 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
185 u64 start, u64 len, u64 disk_len,
186 int type, int dio, int compress_type)
188 struct btrfs_root *root = BTRFS_I(inode)->root;
189 struct btrfs_ordered_inode_tree *tree;
190 struct rb_node *node;
191 struct btrfs_ordered_extent *entry;
193 tree = &BTRFS_I(inode)->ordered_tree;
194 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
198 entry->file_offset = file_offset;
199 entry->start = start;
201 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) &&
202 !(type == BTRFS_ORDERED_NOCOW))
203 entry->csum_bytes_left = disk_len;
204 entry->disk_len = disk_len;
205 entry->bytes_left = len;
206 entry->inode = igrab(inode);
207 entry->compress_type = compress_type;
208 entry->truncated_len = (u64)-1;
209 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
210 set_bit(type, &entry->flags);
213 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
215 /* one ref for the tree */
216 atomic_set(&entry->refs, 1);
217 init_waitqueue_head(&entry->wait);
218 INIT_LIST_HEAD(&entry->list);
219 INIT_LIST_HEAD(&entry->root_extent_list);
220 INIT_LIST_HEAD(&entry->work_list);
221 init_completion(&entry->completion);
222 INIT_LIST_HEAD(&entry->log_list);
224 trace_btrfs_ordered_extent_add(inode, entry);
226 spin_lock_irq(&tree->lock);
227 node = tree_insert(&tree->tree, file_offset,
230 ordered_data_tree_panic(inode, -EEXIST, file_offset);
231 spin_unlock_irq(&tree->lock);
233 spin_lock(&root->ordered_extent_lock);
234 list_add_tail(&entry->root_extent_list,
235 &root->ordered_extents);
236 root->nr_ordered_extents++;
237 if (root->nr_ordered_extents == 1) {
238 spin_lock(&root->fs_info->ordered_root_lock);
239 BUG_ON(!list_empty(&root->ordered_root));
240 list_add_tail(&root->ordered_root,
241 &root->fs_info->ordered_roots);
242 spin_unlock(&root->fs_info->ordered_root_lock);
244 spin_unlock(&root->ordered_extent_lock);
249 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
250 u64 start, u64 len, u64 disk_len, int type)
252 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
254 BTRFS_COMPRESS_NONE);
257 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
258 u64 start, u64 len, u64 disk_len, int type)
260 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
262 BTRFS_COMPRESS_NONE);
265 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
266 u64 start, u64 len, u64 disk_len,
267 int type, int compress_type)
269 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
275 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
276 * when an ordered extent is finished. If the list covers more than one
277 * ordered extent, it is split across multiples.
279 void btrfs_add_ordered_sum(struct inode *inode,
280 struct btrfs_ordered_extent *entry,
281 struct btrfs_ordered_sum *sum)
283 struct btrfs_ordered_inode_tree *tree;
285 tree = &BTRFS_I(inode)->ordered_tree;
286 spin_lock_irq(&tree->lock);
287 list_add_tail(&sum->list, &entry->list);
288 WARN_ON(entry->csum_bytes_left < sum->len);
289 entry->csum_bytes_left -= sum->len;
290 if (entry->csum_bytes_left == 0)
291 wake_up(&entry->wait);
292 spin_unlock_irq(&tree->lock);
296 * this is used to account for finished IO across a given range
297 * of the file. The IO may span ordered extents. If
298 * a given ordered_extent is completely done, 1 is returned, otherwise
301 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
302 * to make sure this function only returns 1 once for a given ordered extent.
304 * file_offset is updated to one byte past the range that is recorded as
305 * complete. This allows you to walk forward in the file.
307 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
308 struct btrfs_ordered_extent **cached,
309 u64 *file_offset, u64 io_size, int uptodate)
311 struct btrfs_ordered_inode_tree *tree;
312 struct rb_node *node;
313 struct btrfs_ordered_extent *entry = NULL;
320 tree = &BTRFS_I(inode)->ordered_tree;
321 spin_lock_irqsave(&tree->lock, flags);
322 node = tree_search(tree, *file_offset);
328 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
329 if (!offset_in_entry(entry, *file_offset)) {
334 dec_start = max(*file_offset, entry->file_offset);
335 dec_end = min(*file_offset + io_size, entry->file_offset +
337 *file_offset = dec_end;
338 if (dec_start > dec_end) {
339 btrfs_crit(BTRFS_I(inode)->root->fs_info,
340 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
342 to_dec = dec_end - dec_start;
343 if (to_dec > entry->bytes_left) {
344 btrfs_crit(BTRFS_I(inode)->root->fs_info,
345 "bad ordered accounting left %llu size %llu",
346 entry->bytes_left, to_dec);
348 entry->bytes_left -= to_dec;
350 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
352 if (entry->bytes_left == 0) {
353 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
354 if (waitqueue_active(&entry->wait))
355 wake_up(&entry->wait);
360 if (!ret && cached && entry) {
362 atomic_inc(&entry->refs);
364 spin_unlock_irqrestore(&tree->lock, flags);
369 * this is used to account for finished IO across a given range
370 * of the file. The IO should not span ordered extents. If
371 * a given ordered_extent is completely done, 1 is returned, otherwise
374 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
375 * to make sure this function only returns 1 once for a given ordered extent.
377 int btrfs_dec_test_ordered_pending(struct inode *inode,
378 struct btrfs_ordered_extent **cached,
379 u64 file_offset, u64 io_size, int uptodate)
381 struct btrfs_ordered_inode_tree *tree;
382 struct rb_node *node;
383 struct btrfs_ordered_extent *entry = NULL;
387 tree = &BTRFS_I(inode)->ordered_tree;
388 spin_lock_irqsave(&tree->lock, flags);
389 if (cached && *cached) {
394 node = tree_search(tree, file_offset);
400 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
402 if (!offset_in_entry(entry, file_offset)) {
407 if (io_size > entry->bytes_left) {
408 btrfs_crit(BTRFS_I(inode)->root->fs_info,
409 "bad ordered accounting left %llu size %llu",
410 entry->bytes_left, io_size);
412 entry->bytes_left -= io_size;
414 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
416 if (entry->bytes_left == 0) {
417 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
418 if (waitqueue_active(&entry->wait))
419 wake_up(&entry->wait);
424 if (!ret && cached && entry) {
426 atomic_inc(&entry->refs);
428 spin_unlock_irqrestore(&tree->lock, flags);
432 /* Needs to either be called under a log transaction or the log_mutex */
433 void btrfs_get_logged_extents(struct inode *inode,
434 struct list_head *logged_list)
436 struct btrfs_ordered_inode_tree *tree;
437 struct btrfs_ordered_extent *ordered;
440 tree = &BTRFS_I(inode)->ordered_tree;
441 spin_lock_irq(&tree->lock);
442 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
443 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
444 if (!list_empty(&ordered->log_list))
446 list_add_tail(&ordered->log_list, logged_list);
447 atomic_inc(&ordered->refs);
449 spin_unlock_irq(&tree->lock);
452 void btrfs_put_logged_extents(struct list_head *logged_list)
454 struct btrfs_ordered_extent *ordered;
456 while (!list_empty(logged_list)) {
457 ordered = list_first_entry(logged_list,
458 struct btrfs_ordered_extent,
460 list_del_init(&ordered->log_list);
461 btrfs_put_ordered_extent(ordered);
465 void btrfs_submit_logged_extents(struct list_head *logged_list,
466 struct btrfs_root *log)
468 int index = log->log_transid % 2;
470 spin_lock_irq(&log->log_extents_lock[index]);
471 list_splice_tail(logged_list, &log->logged_list[index]);
472 spin_unlock_irq(&log->log_extents_lock[index]);
475 void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
477 struct btrfs_ordered_extent *ordered;
478 int index = transid % 2;
480 spin_lock_irq(&log->log_extents_lock[index]);
481 while (!list_empty(&log->logged_list[index])) {
482 ordered = list_first_entry(&log->logged_list[index],
483 struct btrfs_ordered_extent,
485 list_del_init(&ordered->log_list);
486 spin_unlock_irq(&log->log_extents_lock[index]);
487 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
489 btrfs_put_ordered_extent(ordered);
490 spin_lock_irq(&log->log_extents_lock[index]);
492 spin_unlock_irq(&log->log_extents_lock[index]);
495 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
497 struct btrfs_ordered_extent *ordered;
498 int index = transid % 2;
500 spin_lock_irq(&log->log_extents_lock[index]);
501 while (!list_empty(&log->logged_list[index])) {
502 ordered = list_first_entry(&log->logged_list[index],
503 struct btrfs_ordered_extent,
505 list_del_init(&ordered->log_list);
506 spin_unlock_irq(&log->log_extents_lock[index]);
507 btrfs_put_ordered_extent(ordered);
508 spin_lock_irq(&log->log_extents_lock[index]);
510 spin_unlock_irq(&log->log_extents_lock[index]);
514 * used to drop a reference on an ordered extent. This will free
515 * the extent if the last reference is dropped
517 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
519 struct list_head *cur;
520 struct btrfs_ordered_sum *sum;
522 trace_btrfs_ordered_extent_put(entry->inode, entry);
524 if (atomic_dec_and_test(&entry->refs)) {
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 inode *inode,
542 struct btrfs_ordered_extent *entry)
544 struct btrfs_ordered_inode_tree *tree;
545 struct btrfs_root *root = BTRFS_I(inode)->root;
546 struct rb_node *node;
548 tree = &BTRFS_I(inode)->ordered_tree;
549 spin_lock_irq(&tree->lock);
550 node = &entry->rb_node;
551 rb_erase(node, &tree->tree);
552 if (tree->last == node)
554 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
555 spin_unlock_irq(&tree->lock);
557 spin_lock(&root->ordered_extent_lock);
558 list_del_init(&entry->root_extent_list);
559 root->nr_ordered_extents--;
561 trace_btrfs_ordered_extent_remove(inode, entry);
564 * we have no more ordered extents for this inode and
565 * no dirty pages. We can safely remove it from the
566 * list of ordered extents
568 if (RB_EMPTY_ROOT(&tree->tree) &&
569 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
570 spin_lock(&root->fs_info->ordered_root_lock);
571 list_del_init(&BTRFS_I(inode)->ordered_operations);
572 spin_unlock(&root->fs_info->ordered_root_lock);
575 if (!root->nr_ordered_extents) {
576 spin_lock(&root->fs_info->ordered_root_lock);
577 BUG_ON(list_empty(&root->ordered_root));
578 list_del_init(&root->ordered_root);
579 spin_unlock(&root->fs_info->ordered_root_lock);
581 spin_unlock(&root->ordered_extent_lock);
582 wake_up(&entry->wait);
585 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
587 struct btrfs_ordered_extent *ordered;
589 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
590 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
591 complete(&ordered->completion);
595 * wait for all the ordered extents in a root. This is done when balancing
596 * space between drives.
598 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
600 struct list_head splice, works;
601 struct btrfs_ordered_extent *ordered, *next;
604 INIT_LIST_HEAD(&splice);
605 INIT_LIST_HEAD(&works);
607 mutex_lock(&root->ordered_extent_mutex);
608 spin_lock(&root->ordered_extent_lock);
609 list_splice_init(&root->ordered_extents, &splice);
610 while (!list_empty(&splice) && nr) {
611 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
613 list_move_tail(&ordered->root_extent_list,
614 &root->ordered_extents);
615 atomic_inc(&ordered->refs);
616 spin_unlock(&root->ordered_extent_lock);
618 btrfs_init_work(&ordered->flush_work,
619 btrfs_run_ordered_extent_work, NULL, NULL);
620 list_add_tail(&ordered->work_list, &works);
621 btrfs_queue_work(root->fs_info->flush_workers,
622 &ordered->flush_work);
625 spin_lock(&root->ordered_extent_lock);
630 list_splice_tail(&splice, &root->ordered_extents);
631 spin_unlock(&root->ordered_extent_lock);
633 list_for_each_entry_safe(ordered, next, &works, work_list) {
634 list_del_init(&ordered->work_list);
635 wait_for_completion(&ordered->completion);
636 btrfs_put_ordered_extent(ordered);
639 mutex_unlock(&root->ordered_extent_mutex);
644 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
646 struct btrfs_root *root;
647 struct list_head splice;
650 INIT_LIST_HEAD(&splice);
652 mutex_lock(&fs_info->ordered_operations_mutex);
653 spin_lock(&fs_info->ordered_root_lock);
654 list_splice_init(&fs_info->ordered_roots, &splice);
655 while (!list_empty(&splice) && nr) {
656 root = list_first_entry(&splice, struct btrfs_root,
658 root = btrfs_grab_fs_root(root);
660 list_move_tail(&root->ordered_root,
661 &fs_info->ordered_roots);
662 spin_unlock(&fs_info->ordered_root_lock);
664 done = btrfs_wait_ordered_extents(root, nr);
665 btrfs_put_fs_root(root);
667 spin_lock(&fs_info->ordered_root_lock);
673 list_splice_tail(&splice, &fs_info->ordered_roots);
674 spin_unlock(&fs_info->ordered_root_lock);
675 mutex_unlock(&fs_info->ordered_operations_mutex);
679 * this is used during transaction commit to write all the inodes
680 * added to the ordered operation list. These files must be fully on
681 * disk before the transaction commits.
683 * we have two modes here, one is to just start the IO via filemap_flush
684 * and the other is to wait for all the io. When we wait, we have an
685 * extra check to make sure the ordered operation list really is empty
688 int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans,
689 struct btrfs_root *root, int wait)
691 struct btrfs_inode *btrfs_inode;
693 struct btrfs_transaction *cur_trans = trans->transaction;
694 struct list_head splice;
695 struct list_head works;
696 struct btrfs_delalloc_work *work, *next;
699 INIT_LIST_HEAD(&splice);
700 INIT_LIST_HEAD(&works);
702 mutex_lock(&root->fs_info->ordered_extent_flush_mutex);
703 spin_lock(&root->fs_info->ordered_root_lock);
704 list_splice_init(&cur_trans->ordered_operations, &splice);
705 while (!list_empty(&splice)) {
706 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
708 inode = &btrfs_inode->vfs_inode;
710 list_del_init(&btrfs_inode->ordered_operations);
713 * the inode may be getting freed (in sys_unlink path).
715 inode = igrab(inode);
720 list_add_tail(&BTRFS_I(inode)->ordered_operations,
721 &cur_trans->ordered_operations);
722 spin_unlock(&root->fs_info->ordered_root_lock);
724 work = btrfs_alloc_delalloc_work(inode, wait, 1);
726 spin_lock(&root->fs_info->ordered_root_lock);
727 if (list_empty(&BTRFS_I(inode)->ordered_operations))
728 list_add_tail(&btrfs_inode->ordered_operations,
730 list_splice_tail(&splice,
731 &cur_trans->ordered_operations);
732 spin_unlock(&root->fs_info->ordered_root_lock);
736 list_add_tail(&work->list, &works);
737 btrfs_queue_work(root->fs_info->flush_workers,
741 spin_lock(&root->fs_info->ordered_root_lock);
743 spin_unlock(&root->fs_info->ordered_root_lock);
745 list_for_each_entry_safe(work, next, &works, list) {
746 list_del_init(&work->list);
747 btrfs_wait_and_free_delalloc_work(work);
749 mutex_unlock(&root->fs_info->ordered_extent_flush_mutex);
754 * Used to start IO or wait for a given ordered extent to finish.
756 * If wait is one, this effectively waits on page writeback for all the pages
757 * in the extent, and it waits on the io completion code to insert
758 * metadata into the btree corresponding to the extent
760 void btrfs_start_ordered_extent(struct inode *inode,
761 struct btrfs_ordered_extent *entry,
764 u64 start = entry->file_offset;
765 u64 end = start + entry->len - 1;
767 trace_btrfs_ordered_extent_start(inode, entry);
770 * pages in the range can be dirty, clean or writeback. We
771 * start IO on any dirty ones so the wait doesn't stall waiting
772 * for the flusher thread to find them
774 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
775 filemap_fdatawrite_range(inode->i_mapping, start, end);
777 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
783 * Used to wait on ordered extents across a large range of bytes.
785 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
790 struct btrfs_ordered_extent *ordered;
792 if (start + len < start) {
793 orig_end = INT_LIMIT(loff_t);
795 orig_end = start + len - 1;
796 if (orig_end > INT_LIMIT(loff_t))
797 orig_end = INT_LIMIT(loff_t);
800 /* start IO across the range first to instantiate any delalloc
803 ret = filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
807 * So with compression we will find and lock a dirty page and clear the
808 * first one as dirty, setup an async extent, and immediately return
809 * with the entire range locked but with nobody actually marked with
810 * writeback. So we can't just filemap_write_and_wait_range() and
811 * expect it to work since it will just kick off a thread to do the
812 * actual work. So we need to call filemap_fdatawrite_range _again_
813 * since it will wait on the page lock, which won't be unlocked until
814 * after the pages have been marked as writeback and so we're good to go
815 * from there. We have to do this otherwise we'll miss the ordered
816 * extents and that results in badness. Please Josef, do not think you
817 * know better and pull this out at some point in the future, it is
818 * right and you are wrong.
820 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
821 &BTRFS_I(inode)->runtime_flags)) {
822 ret = filemap_fdatawrite_range(inode->i_mapping, start,
827 ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
833 ordered = btrfs_lookup_first_ordered_extent(inode, end);
836 if (ordered->file_offset > orig_end) {
837 btrfs_put_ordered_extent(ordered);
840 if (ordered->file_offset + ordered->len <= start) {
841 btrfs_put_ordered_extent(ordered);
844 btrfs_start_ordered_extent(inode, ordered, 1);
845 end = ordered->file_offset;
846 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
848 btrfs_put_ordered_extent(ordered);
849 if (ret || end == 0 || end == start)
857 * find an ordered extent corresponding to file_offset. return NULL if
858 * nothing is found, otherwise take a reference on the extent and return it
860 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
863 struct btrfs_ordered_inode_tree *tree;
864 struct rb_node *node;
865 struct btrfs_ordered_extent *entry = NULL;
867 tree = &BTRFS_I(inode)->ordered_tree;
868 spin_lock_irq(&tree->lock);
869 node = tree_search(tree, file_offset);
873 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
874 if (!offset_in_entry(entry, file_offset))
877 atomic_inc(&entry->refs);
879 spin_unlock_irq(&tree->lock);
883 /* Since the DIO code tries to lock a wide area we need to look for any ordered
884 * extents that exist in the range, rather than just the start of the range.
886 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
890 struct btrfs_ordered_inode_tree *tree;
891 struct rb_node *node;
892 struct btrfs_ordered_extent *entry = NULL;
894 tree = &BTRFS_I(inode)->ordered_tree;
895 spin_lock_irq(&tree->lock);
896 node = tree_search(tree, file_offset);
898 node = tree_search(tree, file_offset + len);
904 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
905 if (range_overlaps(entry, file_offset, len))
908 if (entry->file_offset >= file_offset + len) {
913 node = rb_next(node);
919 atomic_inc(&entry->refs);
920 spin_unlock_irq(&tree->lock);
925 * lookup and return any extent before 'file_offset'. NULL is returned
928 struct btrfs_ordered_extent *
929 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
931 struct btrfs_ordered_inode_tree *tree;
932 struct rb_node *node;
933 struct btrfs_ordered_extent *entry = NULL;
935 tree = &BTRFS_I(inode)->ordered_tree;
936 spin_lock_irq(&tree->lock);
937 node = tree_search(tree, file_offset);
941 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
942 atomic_inc(&entry->refs);
944 spin_unlock_irq(&tree->lock);
949 * After an extent is done, call this to conditionally update the on disk
950 * i_size. i_size is updated to cover any fully written part of the file.
952 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
953 struct btrfs_ordered_extent *ordered)
955 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
958 u64 i_size = i_size_read(inode);
959 struct rb_node *node;
960 struct rb_node *prev = NULL;
961 struct btrfs_ordered_extent *test;
964 spin_lock_irq(&tree->lock);
966 offset = entry_end(ordered);
967 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
969 ordered->file_offset +
970 ordered->truncated_len);
972 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
974 disk_i_size = BTRFS_I(inode)->disk_i_size;
977 if (disk_i_size > i_size) {
978 BTRFS_I(inode)->disk_i_size = i_size;
984 * if the disk i_size is already at the inode->i_size, or
985 * this ordered extent is inside the disk i_size, we're done
987 if (disk_i_size == i_size)
991 * We still need to update disk_i_size if outstanding_isize is greater
994 if (offset <= disk_i_size &&
995 (!ordered || ordered->outstanding_isize <= disk_i_size))
999 * walk backward from this ordered extent to disk_i_size.
1000 * if we find an ordered extent then we can't update disk i_size
1004 node = rb_prev(&ordered->rb_node);
1006 prev = tree_search(tree, offset);
1008 * we insert file extents without involving ordered struct,
1009 * so there should be no ordered struct cover this offset
1012 test = rb_entry(prev, struct btrfs_ordered_extent,
1014 BUG_ON(offset_in_entry(test, offset));
1018 for (; node; node = rb_prev(node)) {
1019 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1021 /* We treat this entry as if it doesnt exist */
1022 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1024 if (test->file_offset + test->len <= disk_i_size)
1026 if (test->file_offset >= i_size)
1028 if (entry_end(test) > disk_i_size) {
1030 * we don't update disk_i_size now, so record this
1031 * undealt i_size. Or we will not know the real
1034 if (test->outstanding_isize < offset)
1035 test->outstanding_isize = offset;
1037 ordered->outstanding_isize >
1038 test->outstanding_isize)
1039 test->outstanding_isize =
1040 ordered->outstanding_isize;
1044 new_i_size = min_t(u64, offset, i_size);
1047 * Some ordered extents may completed before the current one, and
1048 * we hold the real i_size in ->outstanding_isize.
1050 if (ordered && ordered->outstanding_isize > new_i_size)
1051 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1052 BTRFS_I(inode)->disk_i_size = new_i_size;
1056 * We need to do this because we can't remove ordered extents until
1057 * after the i_disk_size has been updated and then the inode has been
1058 * updated to reflect the change, so we need to tell anybody who finds
1059 * this ordered extent that we've already done all the real work, we
1060 * just haven't completed all the other work.
1063 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1064 spin_unlock_irq(&tree->lock);
1069 * search the ordered extents for one corresponding to 'offset' and
1070 * try to find a checksum. This is used because we allow pages to
1071 * be reclaimed before their checksum is actually put into the btree
1073 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1076 struct btrfs_ordered_sum *ordered_sum;
1077 struct btrfs_ordered_extent *ordered;
1078 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1079 unsigned long num_sectors;
1081 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1084 ordered = btrfs_lookup_ordered_extent(inode, offset);
1088 spin_lock_irq(&tree->lock);
1089 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1090 if (disk_bytenr >= ordered_sum->bytenr &&
1091 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1092 i = (disk_bytenr - ordered_sum->bytenr) >>
1093 inode->i_sb->s_blocksize_bits;
1094 num_sectors = ordered_sum->len >>
1095 inode->i_sb->s_blocksize_bits;
1096 num_sectors = min_t(int, len - index, num_sectors - i);
1097 memcpy(sum + index, ordered_sum->sums + i,
1100 index += (int)num_sectors;
1103 disk_bytenr += num_sectors * sectorsize;
1107 spin_unlock_irq(&tree->lock);
1108 btrfs_put_ordered_extent(ordered);
1114 * add a given inode to the list of inodes that must be fully on
1115 * disk before a transaction commit finishes.
1117 * This basically gives us the ext3 style data=ordered mode, and it is mostly
1118 * used to make sure renamed files are fully on disk.
1120 * It is a noop if the inode is already fully on disk.
1122 * If trans is not null, we'll do a friendly check for a transaction that
1123 * is already flushing things and force the IO down ourselves.
1125 void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
1126 struct btrfs_root *root, struct inode *inode)
1128 struct btrfs_transaction *cur_trans = trans->transaction;
1131 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
1134 * if this file hasn't been changed since the last transaction
1135 * commit, we can safely return without doing anything
1137 if (last_mod <= root->fs_info->last_trans_committed)
1140 spin_lock(&root->fs_info->ordered_root_lock);
1141 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
1142 list_add_tail(&BTRFS_I(inode)->ordered_operations,
1143 &cur_trans->ordered_operations);
1145 spin_unlock(&root->fs_info->ordered_root_lock);
1148 int __init ordered_data_init(void)
1150 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1151 sizeof(struct btrfs_ordered_extent), 0,
1152 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1154 if (!btrfs_ordered_extent_cache)
1160 void ordered_data_exit(void)
1162 if (btrfs_ordered_extent_cache)
1163 kmem_cache_destroy(btrfs_ordered_extent_cache);