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 static struct kmem_cache *btrfs_ordered_extent_cache;
30 static u64 entry_end(struct btrfs_ordered_extent *entry)
32 if (entry->file_offset + entry->len < entry->file_offset)
34 return entry->file_offset + entry->len;
37 /* returns NULL if the insertion worked, or it returns the node it did find
40 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
43 struct rb_node **p = &root->rb_node;
44 struct rb_node *parent = NULL;
45 struct btrfs_ordered_extent *entry;
49 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
51 if (file_offset < entry->file_offset)
53 else if (file_offset >= entry_end(entry))
59 rb_link_node(node, parent, p);
60 rb_insert_color(node, root);
64 static void ordered_data_tree_panic(struct inode *inode, int errno,
67 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
68 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
69 "%llu\n", (unsigned long long)offset);
73 * look for a given offset in the tree, and if it can't be found return the
76 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
77 struct rb_node **prev_ret)
79 struct rb_node *n = root->rb_node;
80 struct rb_node *prev = NULL;
82 struct btrfs_ordered_extent *entry;
83 struct btrfs_ordered_extent *prev_entry = NULL;
86 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
90 if (file_offset < entry->file_offset)
92 else if (file_offset >= entry_end(entry))
100 while (prev && file_offset >= entry_end(prev_entry)) {
101 test = rb_next(prev);
104 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
106 if (file_offset < entry_end(prev_entry))
112 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
114 while (prev && file_offset < entry_end(prev_entry)) {
115 test = rb_prev(prev);
118 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
127 * helper to check if a given offset is inside a given entry
129 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
131 if (file_offset < entry->file_offset ||
132 entry->file_offset + entry->len <= file_offset)
137 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
140 if (file_offset + len <= entry->file_offset ||
141 entry->file_offset + entry->len <= file_offset)
147 * look find the first ordered struct that has this offset, otherwise
148 * the first one less than this offset
150 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
153 struct rb_root *root = &tree->tree;
154 struct rb_node *prev = NULL;
156 struct btrfs_ordered_extent *entry;
159 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
161 if (offset_in_entry(entry, file_offset))
164 ret = __tree_search(root, file_offset, &prev);
172 /* allocate and add a new ordered_extent into the per-inode tree.
173 * file_offset is the logical offset in the file
175 * start is the disk block number of an extent already reserved in the
176 * extent allocation tree
178 * len is the length of the extent
180 * The tree is given a single reference on the ordered extent that was
183 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
184 u64 start, u64 len, u64 disk_len,
185 int type, int dio, int compress_type)
187 struct btrfs_ordered_inode_tree *tree;
188 struct rb_node *node;
189 struct btrfs_ordered_extent *entry;
191 tree = &BTRFS_I(inode)->ordered_tree;
192 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
196 entry->file_offset = file_offset;
197 entry->start = start;
199 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) &&
200 !(type == BTRFS_ORDERED_NOCOW))
201 entry->csum_bytes_left = disk_len;
202 entry->disk_len = disk_len;
203 entry->bytes_left = len;
204 entry->inode = igrab(inode);
205 entry->compress_type = compress_type;
206 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
207 set_bit(type, &entry->flags);
210 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
212 /* one ref for the tree */
213 atomic_set(&entry->refs, 1);
214 init_waitqueue_head(&entry->wait);
215 INIT_LIST_HEAD(&entry->list);
216 INIT_LIST_HEAD(&entry->root_extent_list);
217 INIT_LIST_HEAD(&entry->work_list);
218 init_completion(&entry->completion);
219 INIT_LIST_HEAD(&entry->log_list);
221 trace_btrfs_ordered_extent_add(inode, entry);
223 spin_lock_irq(&tree->lock);
224 node = tree_insert(&tree->tree, file_offset,
227 ordered_data_tree_panic(inode, -EEXIST, file_offset);
228 spin_unlock_irq(&tree->lock);
230 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
231 list_add_tail(&entry->root_extent_list,
232 &BTRFS_I(inode)->root->fs_info->ordered_extents);
233 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
238 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
239 u64 start, u64 len, u64 disk_len, int type)
241 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
243 BTRFS_COMPRESS_NONE);
246 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
247 u64 start, u64 len, u64 disk_len, int type)
249 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
251 BTRFS_COMPRESS_NONE);
254 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
255 u64 start, u64 len, u64 disk_len,
256 int type, int compress_type)
258 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
264 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
265 * when an ordered extent is finished. If the list covers more than one
266 * ordered extent, it is split across multiples.
268 void btrfs_add_ordered_sum(struct inode *inode,
269 struct btrfs_ordered_extent *entry,
270 struct btrfs_ordered_sum *sum)
272 struct btrfs_ordered_inode_tree *tree;
274 tree = &BTRFS_I(inode)->ordered_tree;
275 spin_lock_irq(&tree->lock);
276 list_add_tail(&sum->list, &entry->list);
277 WARN_ON(entry->csum_bytes_left < sum->len);
278 entry->csum_bytes_left -= sum->len;
279 if (entry->csum_bytes_left == 0)
280 wake_up(&entry->wait);
281 spin_unlock_irq(&tree->lock);
285 * this is used to account for finished IO across a given range
286 * of the file. The IO may span ordered extents. If
287 * a given ordered_extent is completely done, 1 is returned, otherwise
290 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
291 * to make sure this function only returns 1 once for a given ordered extent.
293 * file_offset is updated to one byte past the range that is recorded as
294 * complete. This allows you to walk forward in the file.
296 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
297 struct btrfs_ordered_extent **cached,
298 u64 *file_offset, u64 io_size, int uptodate)
300 struct btrfs_ordered_inode_tree *tree;
301 struct rb_node *node;
302 struct btrfs_ordered_extent *entry = NULL;
309 tree = &BTRFS_I(inode)->ordered_tree;
310 spin_lock_irqsave(&tree->lock, flags);
311 node = tree_search(tree, *file_offset);
317 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
318 if (!offset_in_entry(entry, *file_offset)) {
323 dec_start = max(*file_offset, entry->file_offset);
324 dec_end = min(*file_offset + io_size, entry->file_offset +
326 *file_offset = dec_end;
327 if (dec_start > dec_end) {
328 printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n",
329 (unsigned long long)dec_start,
330 (unsigned long long)dec_end);
332 to_dec = dec_end - dec_start;
333 if (to_dec > entry->bytes_left) {
334 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
335 (unsigned long long)entry->bytes_left,
336 (unsigned long long)to_dec);
338 entry->bytes_left -= to_dec;
340 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
342 if (entry->bytes_left == 0)
343 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
347 if (!ret && cached && entry) {
349 atomic_inc(&entry->refs);
351 spin_unlock_irqrestore(&tree->lock, flags);
356 * this is used to account for finished IO across a given range
357 * of the file. The IO should not span ordered extents. If
358 * a given ordered_extent is completely done, 1 is returned, otherwise
361 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
362 * to make sure this function only returns 1 once for a given ordered extent.
364 int btrfs_dec_test_ordered_pending(struct inode *inode,
365 struct btrfs_ordered_extent **cached,
366 u64 file_offset, u64 io_size, int uptodate)
368 struct btrfs_ordered_inode_tree *tree;
369 struct rb_node *node;
370 struct btrfs_ordered_extent *entry = NULL;
374 tree = &BTRFS_I(inode)->ordered_tree;
375 spin_lock_irqsave(&tree->lock, flags);
376 if (cached && *cached) {
381 node = tree_search(tree, file_offset);
387 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
389 if (!offset_in_entry(entry, file_offset)) {
394 if (io_size > entry->bytes_left) {
395 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
396 (unsigned long long)entry->bytes_left,
397 (unsigned long long)io_size);
399 entry->bytes_left -= io_size;
401 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
403 if (entry->bytes_left == 0)
404 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
408 if (!ret && cached && entry) {
410 atomic_inc(&entry->refs);
412 spin_unlock_irqrestore(&tree->lock, flags);
416 /* Needs to either be called under a log transaction or the log_mutex */
417 void btrfs_get_logged_extents(struct btrfs_root *log, struct inode *inode)
419 struct btrfs_ordered_inode_tree *tree;
420 struct btrfs_ordered_extent *ordered;
422 int index = log->log_transid % 2;
424 tree = &BTRFS_I(inode)->ordered_tree;
425 spin_lock_irq(&tree->lock);
426 for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
427 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
428 spin_lock(&log->log_extents_lock[index]);
429 if (list_empty(&ordered->log_list)) {
430 list_add_tail(&ordered->log_list, &log->logged_list[index]);
431 atomic_inc(&ordered->refs);
433 spin_unlock(&log->log_extents_lock[index]);
435 spin_unlock_irq(&tree->lock);
438 void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
440 struct btrfs_ordered_extent *ordered;
441 int index = transid % 2;
443 spin_lock_irq(&log->log_extents_lock[index]);
444 while (!list_empty(&log->logged_list[index])) {
445 ordered = list_first_entry(&log->logged_list[index],
446 struct btrfs_ordered_extent,
448 list_del_init(&ordered->log_list);
449 spin_unlock_irq(&log->log_extents_lock[index]);
450 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
452 btrfs_put_ordered_extent(ordered);
453 spin_lock_irq(&log->log_extents_lock[index]);
455 spin_unlock_irq(&log->log_extents_lock[index]);
458 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
460 struct btrfs_ordered_extent *ordered;
461 int index = transid % 2;
463 spin_lock_irq(&log->log_extents_lock[index]);
464 while (!list_empty(&log->logged_list[index])) {
465 ordered = list_first_entry(&log->logged_list[index],
466 struct btrfs_ordered_extent,
468 list_del_init(&ordered->log_list);
469 spin_unlock_irq(&log->log_extents_lock[index]);
470 btrfs_put_ordered_extent(ordered);
471 spin_lock_irq(&log->log_extents_lock[index]);
473 spin_unlock_irq(&log->log_extents_lock[index]);
477 * used to drop a reference on an ordered extent. This will free
478 * the extent if the last reference is dropped
480 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
482 struct list_head *cur;
483 struct btrfs_ordered_sum *sum;
485 trace_btrfs_ordered_extent_put(entry->inode, entry);
487 if (atomic_dec_and_test(&entry->refs)) {
489 btrfs_add_delayed_iput(entry->inode);
490 while (!list_empty(&entry->list)) {
491 cur = entry->list.next;
492 sum = list_entry(cur, struct btrfs_ordered_sum, list);
493 list_del(&sum->list);
496 kmem_cache_free(btrfs_ordered_extent_cache, entry);
501 * remove an ordered extent from the tree. No references are dropped
502 * and waiters are woken up.
504 void btrfs_remove_ordered_extent(struct inode *inode,
505 struct btrfs_ordered_extent *entry)
507 struct btrfs_ordered_inode_tree *tree;
508 struct btrfs_root *root = BTRFS_I(inode)->root;
509 struct rb_node *node;
511 tree = &BTRFS_I(inode)->ordered_tree;
512 spin_lock_irq(&tree->lock);
513 node = &entry->rb_node;
514 rb_erase(node, &tree->tree);
516 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
517 spin_unlock_irq(&tree->lock);
519 spin_lock(&root->fs_info->ordered_extent_lock);
520 list_del_init(&entry->root_extent_list);
522 trace_btrfs_ordered_extent_remove(inode, entry);
525 * we have no more ordered extents for this inode and
526 * no dirty pages. We can safely remove it from the
527 * list of ordered extents
529 if (RB_EMPTY_ROOT(&tree->tree) &&
530 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
531 list_del_init(&BTRFS_I(inode)->ordered_operations);
533 spin_unlock(&root->fs_info->ordered_extent_lock);
534 wake_up(&entry->wait);
537 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
539 struct btrfs_ordered_extent *ordered;
541 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
542 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
543 complete(&ordered->completion);
547 * wait for all the ordered extents in a root. This is done when balancing
548 * space between drives.
550 void btrfs_wait_ordered_extents(struct btrfs_root *root, int delay_iput)
552 struct list_head splice, works;
553 struct list_head *cur;
554 struct btrfs_ordered_extent *ordered, *next;
557 INIT_LIST_HEAD(&splice);
558 INIT_LIST_HEAD(&works);
560 spin_lock(&root->fs_info->ordered_extent_lock);
561 list_splice_init(&root->fs_info->ordered_extents, &splice);
562 while (!list_empty(&splice)) {
564 ordered = list_entry(cur, struct btrfs_ordered_extent,
566 list_del_init(&ordered->root_extent_list);
567 atomic_inc(&ordered->refs);
570 * the inode may be getting freed (in sys_unlink path).
572 inode = igrab(ordered->inode);
574 spin_unlock(&root->fs_info->ordered_extent_lock);
577 ordered->flush_work.func = btrfs_run_ordered_extent_work;
578 list_add_tail(&ordered->work_list, &works);
579 btrfs_queue_worker(&root->fs_info->flush_workers,
580 &ordered->flush_work);
582 btrfs_put_ordered_extent(ordered);
586 spin_lock(&root->fs_info->ordered_extent_lock);
588 spin_unlock(&root->fs_info->ordered_extent_lock);
590 list_for_each_entry_safe(ordered, next, &works, work_list) {
591 list_del_init(&ordered->work_list);
592 wait_for_completion(&ordered->completion);
594 inode = ordered->inode;
595 btrfs_put_ordered_extent(ordered);
597 btrfs_add_delayed_iput(inode);
606 * this is used during transaction commit to write all the inodes
607 * added to the ordered operation list. These files must be fully on
608 * disk before the transaction commits.
610 * we have two modes here, one is to just start the IO via filemap_flush
611 * and the other is to wait for all the io. When we wait, we have an
612 * extra check to make sure the ordered operation list really is empty
615 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
617 struct btrfs_inode *btrfs_inode;
619 struct list_head splice;
620 struct list_head works;
621 struct btrfs_delalloc_work *work, *next;
624 INIT_LIST_HEAD(&splice);
625 INIT_LIST_HEAD(&works);
627 mutex_lock(&root->fs_info->ordered_operations_mutex);
628 spin_lock(&root->fs_info->ordered_extent_lock);
630 list_splice_init(&root->fs_info->ordered_operations, &splice);
632 while (!list_empty(&splice)) {
634 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
637 inode = &btrfs_inode->vfs_inode;
639 list_del_init(&btrfs_inode->ordered_operations);
642 * the inode may be getting freed (in sys_unlink path).
644 inode = igrab(inode);
646 if (!wait && inode) {
647 list_add_tail(&BTRFS_I(inode)->ordered_operations,
648 &root->fs_info->ordered_operations);
653 spin_unlock(&root->fs_info->ordered_extent_lock);
655 work = btrfs_alloc_delalloc_work(inode, wait, 1);
657 if (list_empty(&BTRFS_I(inode)->ordered_operations))
658 list_add_tail(&btrfs_inode->ordered_operations,
660 spin_lock(&root->fs_info->ordered_extent_lock);
661 list_splice_tail(&splice,
662 &root->fs_info->ordered_operations);
663 spin_unlock(&root->fs_info->ordered_extent_lock);
667 list_add_tail(&work->list, &works);
668 btrfs_queue_worker(&root->fs_info->flush_workers,
672 spin_lock(&root->fs_info->ordered_extent_lock);
674 if (wait && !list_empty(&root->fs_info->ordered_operations))
677 spin_unlock(&root->fs_info->ordered_extent_lock);
679 list_for_each_entry_safe(work, next, &works, list) {
680 list_del_init(&work->list);
681 btrfs_wait_and_free_delalloc_work(work);
683 mutex_unlock(&root->fs_info->ordered_operations_mutex);
688 * Used to start IO or wait for a given ordered extent to finish.
690 * If wait is one, this effectively waits on page writeback for all the pages
691 * in the extent, and it waits on the io completion code to insert
692 * metadata into the btree corresponding to the extent
694 void btrfs_start_ordered_extent(struct inode *inode,
695 struct btrfs_ordered_extent *entry,
698 u64 start = entry->file_offset;
699 u64 end = start + entry->len - 1;
701 trace_btrfs_ordered_extent_start(inode, entry);
704 * pages in the range can be dirty, clean or writeback. We
705 * start IO on any dirty ones so the wait doesn't stall waiting
706 * for the flusher thread to find them
708 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
709 filemap_fdatawrite_range(inode->i_mapping, start, end);
711 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
717 * Used to wait on ordered extents across a large range of bytes.
719 void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
723 struct btrfs_ordered_extent *ordered;
725 if (start + len < start) {
726 orig_end = INT_LIMIT(loff_t);
728 orig_end = start + len - 1;
729 if (orig_end > INT_LIMIT(loff_t))
730 orig_end = INT_LIMIT(loff_t);
733 /* start IO across the range first to instantiate any delalloc
736 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
739 * So with compression we will find and lock a dirty page and clear the
740 * first one as dirty, setup an async extent, and immediately return
741 * with the entire range locked but with nobody actually marked with
742 * writeback. So we can't just filemap_write_and_wait_range() and
743 * expect it to work since it will just kick off a thread to do the
744 * actual work. So we need to call filemap_fdatawrite_range _again_
745 * since it will wait on the page lock, which won't be unlocked until
746 * after the pages have been marked as writeback and so we're good to go
747 * from there. We have to do this otherwise we'll miss the ordered
748 * extents and that results in badness. Please Josef, do not think you
749 * know better and pull this out at some point in the future, it is
750 * right and you are wrong.
752 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
753 &BTRFS_I(inode)->runtime_flags))
754 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
756 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
760 ordered = btrfs_lookup_first_ordered_extent(inode, end);
763 if (ordered->file_offset > orig_end) {
764 btrfs_put_ordered_extent(ordered);
767 if (ordered->file_offset + ordered->len < start) {
768 btrfs_put_ordered_extent(ordered);
771 btrfs_start_ordered_extent(inode, ordered, 1);
772 end = ordered->file_offset;
773 btrfs_put_ordered_extent(ordered);
774 if (end == 0 || end == start)
781 * find an ordered extent corresponding to file_offset. return NULL if
782 * nothing is found, otherwise take a reference on the extent and return it
784 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
787 struct btrfs_ordered_inode_tree *tree;
788 struct rb_node *node;
789 struct btrfs_ordered_extent *entry = NULL;
791 tree = &BTRFS_I(inode)->ordered_tree;
792 spin_lock_irq(&tree->lock);
793 node = tree_search(tree, file_offset);
797 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
798 if (!offset_in_entry(entry, file_offset))
801 atomic_inc(&entry->refs);
803 spin_unlock_irq(&tree->lock);
807 /* Since the DIO code tries to lock a wide area we need to look for any ordered
808 * extents that exist in the range, rather than just the start of the range.
810 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
814 struct btrfs_ordered_inode_tree *tree;
815 struct rb_node *node;
816 struct btrfs_ordered_extent *entry = NULL;
818 tree = &BTRFS_I(inode)->ordered_tree;
819 spin_lock_irq(&tree->lock);
820 node = tree_search(tree, file_offset);
822 node = tree_search(tree, file_offset + len);
828 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
829 if (range_overlaps(entry, file_offset, len))
832 if (entry->file_offset >= file_offset + len) {
837 node = rb_next(node);
843 atomic_inc(&entry->refs);
844 spin_unlock_irq(&tree->lock);
849 * lookup and return any extent before 'file_offset'. NULL is returned
852 struct btrfs_ordered_extent *
853 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
855 struct btrfs_ordered_inode_tree *tree;
856 struct rb_node *node;
857 struct btrfs_ordered_extent *entry = NULL;
859 tree = &BTRFS_I(inode)->ordered_tree;
860 spin_lock_irq(&tree->lock);
861 node = tree_search(tree, file_offset);
865 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
866 atomic_inc(&entry->refs);
868 spin_unlock_irq(&tree->lock);
873 * After an extent is done, call this to conditionally update the on disk
874 * i_size. i_size is updated to cover any fully written part of the file.
876 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
877 struct btrfs_ordered_extent *ordered)
879 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
882 u64 i_size = i_size_read(inode);
883 struct rb_node *node;
884 struct rb_node *prev = NULL;
885 struct btrfs_ordered_extent *test;
889 offset = entry_end(ordered);
891 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
893 spin_lock_irq(&tree->lock);
894 disk_i_size = BTRFS_I(inode)->disk_i_size;
897 if (disk_i_size > i_size) {
898 BTRFS_I(inode)->disk_i_size = i_size;
904 * if the disk i_size is already at the inode->i_size, or
905 * this ordered extent is inside the disk i_size, we're done
907 if (disk_i_size == i_size)
911 * We still need to update disk_i_size if outstanding_isize is greater
914 if (offset <= disk_i_size &&
915 (!ordered || ordered->outstanding_isize <= disk_i_size))
919 * walk backward from this ordered extent to disk_i_size.
920 * if we find an ordered extent then we can't update disk i_size
924 node = rb_prev(&ordered->rb_node);
926 prev = tree_search(tree, offset);
928 * we insert file extents without involving ordered struct,
929 * so there should be no ordered struct cover this offset
932 test = rb_entry(prev, struct btrfs_ordered_extent,
934 BUG_ON(offset_in_entry(test, offset));
938 for (; node; node = rb_prev(node)) {
939 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
941 /* We treat this entry as if it doesnt exist */
942 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
944 if (test->file_offset + test->len <= disk_i_size)
946 if (test->file_offset >= i_size)
948 if (entry_end(test) > disk_i_size) {
950 * we don't update disk_i_size now, so record this
951 * undealt i_size. Or we will not know the real
954 if (test->outstanding_isize < offset)
955 test->outstanding_isize = offset;
957 ordered->outstanding_isize >
958 test->outstanding_isize)
959 test->outstanding_isize =
960 ordered->outstanding_isize;
964 new_i_size = min_t(u64, offset, i_size);
967 * Some ordered extents may completed before the current one, and
968 * we hold the real i_size in ->outstanding_isize.
970 if (ordered && ordered->outstanding_isize > new_i_size)
971 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
972 BTRFS_I(inode)->disk_i_size = new_i_size;
976 * We need to do this because we can't remove ordered extents until
977 * after the i_disk_size has been updated and then the inode has been
978 * updated to reflect the change, so we need to tell anybody who finds
979 * this ordered extent that we've already done all the real work, we
980 * just haven't completed all the other work.
983 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
984 spin_unlock_irq(&tree->lock);
989 * search the ordered extents for one corresponding to 'offset' and
990 * try to find a checksum. This is used because we allow pages to
991 * be reclaimed before their checksum is actually put into the btree
993 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
996 struct btrfs_ordered_sum *ordered_sum;
997 struct btrfs_sector_sum *sector_sums;
998 struct btrfs_ordered_extent *ordered;
999 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1000 unsigned long num_sectors;
1002 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1005 ordered = btrfs_lookup_ordered_extent(inode, offset);
1009 spin_lock_irq(&tree->lock);
1010 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1011 if (disk_bytenr >= ordered_sum->bytenr) {
1012 num_sectors = ordered_sum->len / sectorsize;
1013 sector_sums = ordered_sum->sums;
1014 for (i = 0; i < num_sectors; i++) {
1015 if (sector_sums[i].bytenr == disk_bytenr) {
1016 *sum = sector_sums[i].sum;
1024 spin_unlock_irq(&tree->lock);
1025 btrfs_put_ordered_extent(ordered);
1031 * add a given inode to the list of inodes that must be fully on
1032 * disk before a transaction commit finishes.
1034 * This basically gives us the ext3 style data=ordered mode, and it is mostly
1035 * used to make sure renamed files are fully on disk.
1037 * It is a noop if the inode is already fully on disk.
1039 * If trans is not null, we'll do a friendly check for a transaction that
1040 * is already flushing things and force the IO down ourselves.
1042 void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
1043 struct btrfs_root *root, struct inode *inode)
1047 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
1050 * if this file hasn't been changed since the last transaction
1051 * commit, we can safely return without doing anything
1053 if (last_mod < root->fs_info->last_trans_committed)
1056 spin_lock(&root->fs_info->ordered_extent_lock);
1057 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
1058 list_add_tail(&BTRFS_I(inode)->ordered_operations,
1059 &root->fs_info->ordered_operations);
1061 spin_unlock(&root->fs_info->ordered_extent_lock);
1064 int __init ordered_data_init(void)
1066 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1067 sizeof(struct btrfs_ordered_extent), 0,
1068 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1070 if (!btrfs_ordered_extent_cache)
1076 void ordered_data_exit(void)
1078 if (btrfs_ordered_extent_cache)
1079 kmem_cache_destroy(btrfs_ordered_extent_cache);