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 printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n",
342 to_dec = dec_end - dec_start;
343 if (to_dec > entry->bytes_left) {
344 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
345 entry->bytes_left, to_dec);
347 entry->bytes_left -= to_dec;
349 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
351 if (entry->bytes_left == 0)
352 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
356 if (!ret && cached && entry) {
358 atomic_inc(&entry->refs);
360 spin_unlock_irqrestore(&tree->lock, flags);
365 * this is used to account for finished IO across a given range
366 * of the file. The IO should not span ordered extents. If
367 * a given ordered_extent is completely done, 1 is returned, otherwise
370 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
371 * to make sure this function only returns 1 once for a given ordered extent.
373 int btrfs_dec_test_ordered_pending(struct inode *inode,
374 struct btrfs_ordered_extent **cached,
375 u64 file_offset, u64 io_size, int uptodate)
377 struct btrfs_ordered_inode_tree *tree;
378 struct rb_node *node;
379 struct btrfs_ordered_extent *entry = NULL;
383 tree = &BTRFS_I(inode)->ordered_tree;
384 spin_lock_irqsave(&tree->lock, flags);
385 if (cached && *cached) {
390 node = tree_search(tree, file_offset);
396 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
398 if (!offset_in_entry(entry, file_offset)) {
403 if (io_size > entry->bytes_left) {
404 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
405 entry->bytes_left, io_size);
407 entry->bytes_left -= io_size;
409 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
411 if (entry->bytes_left == 0)
412 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
416 if (!ret && cached && entry) {
418 atomic_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_root *log, struct inode *inode)
427 struct btrfs_ordered_inode_tree *tree;
428 struct btrfs_ordered_extent *ordered;
430 int index = log->log_transid % 2;
432 tree = &BTRFS_I(inode)->ordered_tree;
433 spin_lock_irq(&tree->lock);
434 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
435 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
436 spin_lock(&log->log_extents_lock[index]);
437 if (list_empty(&ordered->log_list)) {
438 list_add_tail(&ordered->log_list, &log->logged_list[index]);
439 atomic_inc(&ordered->refs);
441 spin_unlock(&log->log_extents_lock[index]);
443 spin_unlock_irq(&tree->lock);
446 void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
448 struct btrfs_ordered_extent *ordered;
449 int index = transid % 2;
451 spin_lock_irq(&log->log_extents_lock[index]);
452 while (!list_empty(&log->logged_list[index])) {
453 ordered = list_first_entry(&log->logged_list[index],
454 struct btrfs_ordered_extent,
456 list_del_init(&ordered->log_list);
457 spin_unlock_irq(&log->log_extents_lock[index]);
458 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
460 btrfs_put_ordered_extent(ordered);
461 spin_lock_irq(&log->log_extents_lock[index]);
463 spin_unlock_irq(&log->log_extents_lock[index]);
466 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
468 struct btrfs_ordered_extent *ordered;
469 int index = transid % 2;
471 spin_lock_irq(&log->log_extents_lock[index]);
472 while (!list_empty(&log->logged_list[index])) {
473 ordered = list_first_entry(&log->logged_list[index],
474 struct btrfs_ordered_extent,
476 list_del_init(&ordered->log_list);
477 spin_unlock_irq(&log->log_extents_lock[index]);
478 btrfs_put_ordered_extent(ordered);
479 spin_lock_irq(&log->log_extents_lock[index]);
481 spin_unlock_irq(&log->log_extents_lock[index]);
485 * used to drop a reference on an ordered extent. This will free
486 * the extent if the last reference is dropped
488 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
490 struct list_head *cur;
491 struct btrfs_ordered_sum *sum;
493 trace_btrfs_ordered_extent_put(entry->inode, entry);
495 if (atomic_dec_and_test(&entry->refs)) {
497 btrfs_add_delayed_iput(entry->inode);
498 while (!list_empty(&entry->list)) {
499 cur = entry->list.next;
500 sum = list_entry(cur, struct btrfs_ordered_sum, list);
501 list_del(&sum->list);
504 kmem_cache_free(btrfs_ordered_extent_cache, entry);
509 * remove an ordered extent from the tree. No references are dropped
510 * and waiters are woken up.
512 void btrfs_remove_ordered_extent(struct inode *inode,
513 struct btrfs_ordered_extent *entry)
515 struct btrfs_ordered_inode_tree *tree;
516 struct btrfs_root *root = BTRFS_I(inode)->root;
517 struct rb_node *node;
519 tree = &BTRFS_I(inode)->ordered_tree;
520 spin_lock_irq(&tree->lock);
521 node = &entry->rb_node;
522 rb_erase(node, &tree->tree);
524 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
525 spin_unlock_irq(&tree->lock);
527 spin_lock(&root->ordered_extent_lock);
528 list_del_init(&entry->root_extent_list);
529 root->nr_ordered_extents--;
531 trace_btrfs_ordered_extent_remove(inode, entry);
534 * we have no more ordered extents for this inode and
535 * no dirty pages. We can safely remove it from the
536 * list of ordered extents
538 if (RB_EMPTY_ROOT(&tree->tree) &&
539 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
540 list_del_init(&BTRFS_I(inode)->ordered_operations);
543 if (!root->nr_ordered_extents) {
544 spin_lock(&root->fs_info->ordered_root_lock);
545 BUG_ON(list_empty(&root->ordered_root));
546 list_del_init(&root->ordered_root);
547 spin_unlock(&root->fs_info->ordered_root_lock);
549 spin_unlock(&root->ordered_extent_lock);
550 wake_up(&entry->wait);
553 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
555 struct btrfs_ordered_extent *ordered;
557 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
558 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
559 complete(&ordered->completion);
563 * wait for all the ordered extents in a root. This is done when balancing
564 * space between drives.
566 void btrfs_wait_ordered_extents(struct btrfs_root *root)
568 struct list_head splice, works;
569 struct btrfs_ordered_extent *ordered, *next;
571 INIT_LIST_HEAD(&splice);
572 INIT_LIST_HEAD(&works);
574 mutex_lock(&root->fs_info->ordered_operations_mutex);
575 spin_lock(&root->ordered_extent_lock);
576 list_splice_init(&root->ordered_extents, &splice);
577 while (!list_empty(&splice)) {
578 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
580 list_move_tail(&ordered->root_extent_list,
581 &root->ordered_extents);
582 atomic_inc(&ordered->refs);
583 spin_unlock(&root->ordered_extent_lock);
585 ordered->flush_work.func = btrfs_run_ordered_extent_work;
586 list_add_tail(&ordered->work_list, &works);
587 btrfs_queue_worker(&root->fs_info->flush_workers,
588 &ordered->flush_work);
591 spin_lock(&root->ordered_extent_lock);
593 spin_unlock(&root->ordered_extent_lock);
595 list_for_each_entry_safe(ordered, next, &works, work_list) {
596 list_del_init(&ordered->work_list);
597 wait_for_completion(&ordered->completion);
598 btrfs_put_ordered_extent(ordered);
601 mutex_unlock(&root->fs_info->ordered_operations_mutex);
604 void btrfs_wait_all_ordered_extents(struct btrfs_fs_info *fs_info)
606 struct btrfs_root *root;
607 struct list_head splice;
609 INIT_LIST_HEAD(&splice);
611 spin_lock(&fs_info->ordered_root_lock);
612 list_splice_init(&fs_info->ordered_roots, &splice);
613 while (!list_empty(&splice)) {
614 root = list_first_entry(&splice, struct btrfs_root,
616 root = btrfs_grab_fs_root(root);
618 list_move_tail(&root->ordered_root,
619 &fs_info->ordered_roots);
620 spin_unlock(&fs_info->ordered_root_lock);
622 btrfs_wait_ordered_extents(root);
623 btrfs_put_fs_root(root);
625 spin_lock(&fs_info->ordered_root_lock);
627 spin_unlock(&fs_info->ordered_root_lock);
631 * this is used during transaction commit to write all the inodes
632 * added to the ordered operation list. These files must be fully on
633 * disk before the transaction commits.
635 * we have two modes here, one is to just start the IO via filemap_flush
636 * and the other is to wait for all the io. When we wait, we have an
637 * extra check to make sure the ordered operation list really is empty
640 int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans,
641 struct btrfs_root *root, int wait)
643 struct btrfs_inode *btrfs_inode;
645 struct btrfs_transaction *cur_trans = trans->transaction;
646 struct list_head splice;
647 struct list_head works;
648 struct btrfs_delalloc_work *work, *next;
651 INIT_LIST_HEAD(&splice);
652 INIT_LIST_HEAD(&works);
654 mutex_lock(&root->fs_info->ordered_extent_flush_mutex);
655 spin_lock(&root->fs_info->ordered_root_lock);
656 list_splice_init(&cur_trans->ordered_operations, &splice);
657 while (!list_empty(&splice)) {
658 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
660 inode = &btrfs_inode->vfs_inode;
662 list_del_init(&btrfs_inode->ordered_operations);
665 * the inode may be getting freed (in sys_unlink path).
667 inode = igrab(inode);
672 list_add_tail(&BTRFS_I(inode)->ordered_operations,
673 &cur_trans->ordered_operations);
674 spin_unlock(&root->fs_info->ordered_root_lock);
676 work = btrfs_alloc_delalloc_work(inode, wait, 1);
678 spin_lock(&root->fs_info->ordered_root_lock);
679 if (list_empty(&BTRFS_I(inode)->ordered_operations))
680 list_add_tail(&btrfs_inode->ordered_operations,
682 list_splice_tail(&splice,
683 &cur_trans->ordered_operations);
684 spin_unlock(&root->fs_info->ordered_root_lock);
688 list_add_tail(&work->list, &works);
689 btrfs_queue_worker(&root->fs_info->flush_workers,
693 spin_lock(&root->fs_info->ordered_root_lock);
695 spin_unlock(&root->fs_info->ordered_root_lock);
697 list_for_each_entry_safe(work, next, &works, list) {
698 list_del_init(&work->list);
699 btrfs_wait_and_free_delalloc_work(work);
701 mutex_unlock(&root->fs_info->ordered_extent_flush_mutex);
706 * Used to start IO or wait for a given ordered extent to finish.
708 * If wait is one, this effectively waits on page writeback for all the pages
709 * in the extent, and it waits on the io completion code to insert
710 * metadata into the btree corresponding to the extent
712 void btrfs_start_ordered_extent(struct inode *inode,
713 struct btrfs_ordered_extent *entry,
716 u64 start = entry->file_offset;
717 u64 end = start + entry->len - 1;
719 trace_btrfs_ordered_extent_start(inode, entry);
722 * pages in the range can be dirty, clean or writeback. We
723 * start IO on any dirty ones so the wait doesn't stall waiting
724 * for the flusher thread to find them
726 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
727 filemap_fdatawrite_range(inode->i_mapping, start, end);
729 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
735 * Used to wait on ordered extents across a large range of bytes.
737 void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
741 struct btrfs_ordered_extent *ordered;
743 if (start + len < start) {
744 orig_end = INT_LIMIT(loff_t);
746 orig_end = start + len - 1;
747 if (orig_end > INT_LIMIT(loff_t))
748 orig_end = INT_LIMIT(loff_t);
751 /* start IO across the range first to instantiate any delalloc
754 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
757 * So with compression we will find and lock a dirty page and clear the
758 * first one as dirty, setup an async extent, and immediately return
759 * with the entire range locked but with nobody actually marked with
760 * writeback. So we can't just filemap_write_and_wait_range() and
761 * expect it to work since it will just kick off a thread to do the
762 * actual work. So we need to call filemap_fdatawrite_range _again_
763 * since it will wait on the page lock, which won't be unlocked until
764 * after the pages have been marked as writeback and so we're good to go
765 * from there. We have to do this otherwise we'll miss the ordered
766 * extents and that results in badness. Please Josef, do not think you
767 * know better and pull this out at some point in the future, it is
768 * right and you are wrong.
770 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
771 &BTRFS_I(inode)->runtime_flags))
772 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
774 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
778 ordered = btrfs_lookup_first_ordered_extent(inode, end);
781 if (ordered->file_offset > orig_end) {
782 btrfs_put_ordered_extent(ordered);
785 if (ordered->file_offset + ordered->len < start) {
786 btrfs_put_ordered_extent(ordered);
789 btrfs_start_ordered_extent(inode, ordered, 1);
790 end = ordered->file_offset;
791 btrfs_put_ordered_extent(ordered);
792 if (end == 0 || end == start)
799 * find an ordered extent corresponding to file_offset. return NULL if
800 * nothing is found, otherwise take a reference on the extent and return it
802 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
805 struct btrfs_ordered_inode_tree *tree;
806 struct rb_node *node;
807 struct btrfs_ordered_extent *entry = NULL;
809 tree = &BTRFS_I(inode)->ordered_tree;
810 spin_lock_irq(&tree->lock);
811 node = tree_search(tree, file_offset);
815 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
816 if (!offset_in_entry(entry, file_offset))
819 atomic_inc(&entry->refs);
821 spin_unlock_irq(&tree->lock);
825 /* Since the DIO code tries to lock a wide area we need to look for any ordered
826 * extents that exist in the range, rather than just the start of the range.
828 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
832 struct btrfs_ordered_inode_tree *tree;
833 struct rb_node *node;
834 struct btrfs_ordered_extent *entry = NULL;
836 tree = &BTRFS_I(inode)->ordered_tree;
837 spin_lock_irq(&tree->lock);
838 node = tree_search(tree, file_offset);
840 node = tree_search(tree, file_offset + len);
846 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
847 if (range_overlaps(entry, file_offset, len))
850 if (entry->file_offset >= file_offset + len) {
855 node = rb_next(node);
861 atomic_inc(&entry->refs);
862 spin_unlock_irq(&tree->lock);
867 * lookup and return any extent before 'file_offset'. NULL is returned
870 struct btrfs_ordered_extent *
871 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
873 struct btrfs_ordered_inode_tree *tree;
874 struct rb_node *node;
875 struct btrfs_ordered_extent *entry = NULL;
877 tree = &BTRFS_I(inode)->ordered_tree;
878 spin_lock_irq(&tree->lock);
879 node = tree_search(tree, file_offset);
883 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
884 atomic_inc(&entry->refs);
886 spin_unlock_irq(&tree->lock);
891 * After an extent is done, call this to conditionally update the on disk
892 * i_size. i_size is updated to cover any fully written part of the file.
894 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
895 struct btrfs_ordered_extent *ordered)
897 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
900 u64 i_size = i_size_read(inode);
901 struct rb_node *node;
902 struct rb_node *prev = NULL;
903 struct btrfs_ordered_extent *test;
906 spin_lock_irq(&tree->lock);
908 offset = entry_end(ordered);
909 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
911 ordered->file_offset +
912 ordered->truncated_len);
914 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
916 disk_i_size = BTRFS_I(inode)->disk_i_size;
919 if (disk_i_size > i_size) {
920 BTRFS_I(inode)->disk_i_size = i_size;
926 * if the disk i_size is already at the inode->i_size, or
927 * this ordered extent is inside the disk i_size, we're done
929 if (disk_i_size == i_size)
933 * We still need to update disk_i_size if outstanding_isize is greater
936 if (offset <= disk_i_size &&
937 (!ordered || ordered->outstanding_isize <= disk_i_size))
941 * walk backward from this ordered extent to disk_i_size.
942 * if we find an ordered extent then we can't update disk i_size
946 node = rb_prev(&ordered->rb_node);
948 prev = tree_search(tree, offset);
950 * we insert file extents without involving ordered struct,
951 * so there should be no ordered struct cover this offset
954 test = rb_entry(prev, struct btrfs_ordered_extent,
956 BUG_ON(offset_in_entry(test, offset));
960 for (; node; node = rb_prev(node)) {
961 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
963 /* We treat this entry as if it doesnt exist */
964 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
966 if (test->file_offset + test->len <= disk_i_size)
968 if (test->file_offset >= i_size)
970 if (entry_end(test) > disk_i_size) {
972 * we don't update disk_i_size now, so record this
973 * undealt i_size. Or we will not know the real
976 if (test->outstanding_isize < offset)
977 test->outstanding_isize = offset;
979 ordered->outstanding_isize >
980 test->outstanding_isize)
981 test->outstanding_isize =
982 ordered->outstanding_isize;
986 new_i_size = min_t(u64, offset, i_size);
989 * Some ordered extents may completed before the current one, and
990 * we hold the real i_size in ->outstanding_isize.
992 if (ordered && ordered->outstanding_isize > new_i_size)
993 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
994 BTRFS_I(inode)->disk_i_size = new_i_size;
998 * We need to do this because we can't remove ordered extents until
999 * after the i_disk_size has been updated and then the inode has been
1000 * updated to reflect the change, so we need to tell anybody who finds
1001 * this ordered extent that we've already done all the real work, we
1002 * just haven't completed all the other work.
1005 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1006 spin_unlock_irq(&tree->lock);
1011 * search the ordered extents for one corresponding to 'offset' and
1012 * try to find a checksum. This is used because we allow pages to
1013 * be reclaimed before their checksum is actually put into the btree
1015 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1018 struct btrfs_ordered_sum *ordered_sum;
1019 struct btrfs_ordered_extent *ordered;
1020 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1021 unsigned long num_sectors;
1023 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1026 ordered = btrfs_lookup_ordered_extent(inode, offset);
1030 spin_lock_irq(&tree->lock);
1031 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1032 if (disk_bytenr >= ordered_sum->bytenr &&
1033 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1034 i = (disk_bytenr - ordered_sum->bytenr) >>
1035 inode->i_sb->s_blocksize_bits;
1036 num_sectors = ordered_sum->len >>
1037 inode->i_sb->s_blocksize_bits;
1038 num_sectors = min_t(int, len - index, num_sectors - i);
1039 memcpy(sum + index, ordered_sum->sums + i,
1042 index += (int)num_sectors;
1045 disk_bytenr += num_sectors * sectorsize;
1049 spin_unlock_irq(&tree->lock);
1050 btrfs_put_ordered_extent(ordered);
1056 * add a given inode to the list of inodes that must be fully on
1057 * disk before a transaction commit finishes.
1059 * This basically gives us the ext3 style data=ordered mode, and it is mostly
1060 * used to make sure renamed files are fully on disk.
1062 * It is a noop if the inode is already fully on disk.
1064 * If trans is not null, we'll do a friendly check for a transaction that
1065 * is already flushing things and force the IO down ourselves.
1067 void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
1068 struct btrfs_root *root, struct inode *inode)
1070 struct btrfs_transaction *cur_trans = trans->transaction;
1073 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
1076 * if this file hasn't been changed since the last transaction
1077 * commit, we can safely return without doing anything
1079 if (last_mod < root->fs_info->last_trans_committed)
1082 spin_lock(&root->fs_info->ordered_root_lock);
1083 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
1084 list_add_tail(&BTRFS_I(inode)->ordered_operations,
1085 &cur_trans->ordered_operations);
1087 spin_unlock(&root->fs_info->ordered_root_lock);
1090 int __init ordered_data_init(void)
1092 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1093 sizeof(struct btrfs_ordered_extent), 0,
1094 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1096 if (!btrfs_ordered_extent_cache)
1102 void ordered_data_exit(void)
1104 if (btrfs_ordered_extent_cache)
1105 kmem_cache_destroy(btrfs_ordered_extent_cache);