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 u64 entry_end(struct btrfs_ordered_extent *entry)
30 if (entry->file_offset + entry->len < entry->file_offset)
32 return entry->file_offset + entry->len;
35 /* returns NULL if the insertion worked, or it returns the node it did find
38 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
41 struct rb_node **p = &root->rb_node;
42 struct rb_node *parent = NULL;
43 struct btrfs_ordered_extent *entry;
47 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
49 if (file_offset < entry->file_offset)
51 else if (file_offset >= entry_end(entry))
57 rb_link_node(node, parent, p);
58 rb_insert_color(node, root);
62 static void ordered_data_tree_panic(struct inode *inode, int errno,
65 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
66 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
67 "%llu\n", (unsigned long long)offset);
71 * look for a given offset in the tree, and if it can't be found return the
74 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
75 struct rb_node **prev_ret)
77 struct rb_node *n = root->rb_node;
78 struct rb_node *prev = NULL;
80 struct btrfs_ordered_extent *entry;
81 struct btrfs_ordered_extent *prev_entry = NULL;
84 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
88 if (file_offset < entry->file_offset)
90 else if (file_offset >= entry_end(entry))
98 while (prev && file_offset >= entry_end(prev_entry)) {
102 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
104 if (file_offset < entry_end(prev_entry))
110 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
112 while (prev && file_offset < entry_end(prev_entry)) {
113 test = rb_prev(prev);
116 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
125 * helper to check if a given offset is inside a given entry
127 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
129 if (file_offset < entry->file_offset ||
130 entry->file_offset + entry->len <= file_offset)
135 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
138 if (file_offset + len <= entry->file_offset ||
139 entry->file_offset + entry->len <= file_offset)
145 * look find the first ordered struct that has this offset, otherwise
146 * the first one less than this offset
148 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
151 struct rb_root *root = &tree->tree;
152 struct rb_node *prev = NULL;
154 struct btrfs_ordered_extent *entry;
157 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
159 if (offset_in_entry(entry, file_offset))
162 ret = __tree_search(root, file_offset, &prev);
170 /* allocate and add a new ordered_extent into the per-inode tree.
171 * file_offset is the logical offset in the file
173 * start is the disk block number of an extent already reserved in the
174 * extent allocation tree
176 * len is the length of the extent
178 * The tree is given a single reference on the ordered extent that was
181 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
182 u64 start, u64 len, u64 disk_len,
183 int type, int dio, int compress_type)
185 struct btrfs_ordered_inode_tree *tree;
186 struct rb_node *node;
187 struct btrfs_ordered_extent *entry;
189 tree = &BTRFS_I(inode)->ordered_tree;
190 entry = kzalloc(sizeof(*entry), GFP_NOFS);
194 entry->file_offset = file_offset;
195 entry->start = start;
197 entry->disk_len = disk_len;
198 entry->bytes_left = len;
199 entry->inode = inode;
200 entry->compress_type = compress_type;
201 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
202 set_bit(type, &entry->flags);
205 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
207 /* one ref for the tree */
208 atomic_set(&entry->refs, 1);
209 init_waitqueue_head(&entry->wait);
210 INIT_LIST_HEAD(&entry->list);
211 INIT_LIST_HEAD(&entry->root_extent_list);
213 trace_btrfs_ordered_extent_add(inode, entry);
215 spin_lock(&tree->lock);
216 node = tree_insert(&tree->tree, file_offset,
219 ordered_data_tree_panic(inode, -EEXIST, file_offset);
220 spin_unlock(&tree->lock);
222 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
223 list_add_tail(&entry->root_extent_list,
224 &BTRFS_I(inode)->root->fs_info->ordered_extents);
225 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
230 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
231 u64 start, u64 len, u64 disk_len, int type)
233 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
235 BTRFS_COMPRESS_NONE);
238 int btrfs_add_ordered_extent_dio(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_compress(struct inode *inode, u64 file_offset,
247 u64 start, u64 len, u64 disk_len,
248 int type, int compress_type)
250 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
256 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
257 * when an ordered extent is finished. If the list covers more than one
258 * ordered extent, it is split across multiples.
260 void btrfs_add_ordered_sum(struct inode *inode,
261 struct btrfs_ordered_extent *entry,
262 struct btrfs_ordered_sum *sum)
264 struct btrfs_ordered_inode_tree *tree;
266 tree = &BTRFS_I(inode)->ordered_tree;
267 spin_lock(&tree->lock);
268 list_add_tail(&sum->list, &entry->list);
269 spin_unlock(&tree->lock);
273 * this is used to account for finished IO across a given range
274 * of the file. The IO may span ordered extents. If
275 * a given ordered_extent is completely done, 1 is returned, otherwise
278 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
279 * to make sure this function only returns 1 once for a given ordered extent.
281 * file_offset is updated to one byte past the range that is recorded as
282 * complete. This allows you to walk forward in the file.
284 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
285 struct btrfs_ordered_extent **cached,
286 u64 *file_offset, u64 io_size)
288 struct btrfs_ordered_inode_tree *tree;
289 struct rb_node *node;
290 struct btrfs_ordered_extent *entry = NULL;
296 tree = &BTRFS_I(inode)->ordered_tree;
297 spin_lock(&tree->lock);
298 node = tree_search(tree, *file_offset);
304 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
305 if (!offset_in_entry(entry, *file_offset)) {
310 dec_start = max(*file_offset, entry->file_offset);
311 dec_end = min(*file_offset + io_size, entry->file_offset +
313 *file_offset = dec_end;
314 if (dec_start > dec_end) {
315 printk(KERN_CRIT "bad ordering dec_start %llu end %llu\n",
316 (unsigned long long)dec_start,
317 (unsigned long long)dec_end);
319 to_dec = dec_end - dec_start;
320 if (to_dec > entry->bytes_left) {
321 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
322 (unsigned long long)entry->bytes_left,
323 (unsigned long long)to_dec);
325 entry->bytes_left -= to_dec;
326 if (entry->bytes_left == 0)
327 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
331 if (!ret && cached && entry) {
333 atomic_inc(&entry->refs);
335 spin_unlock(&tree->lock);
340 * this is used to account for finished IO across a given range
341 * of the file. The IO should not span ordered extents. If
342 * a given ordered_extent is completely done, 1 is returned, otherwise
345 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
346 * to make sure this function only returns 1 once for a given ordered extent.
348 int btrfs_dec_test_ordered_pending(struct inode *inode,
349 struct btrfs_ordered_extent **cached,
350 u64 file_offset, u64 io_size)
352 struct btrfs_ordered_inode_tree *tree;
353 struct rb_node *node;
354 struct btrfs_ordered_extent *entry = NULL;
357 tree = &BTRFS_I(inode)->ordered_tree;
358 spin_lock(&tree->lock);
359 node = tree_search(tree, file_offset);
365 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
366 if (!offset_in_entry(entry, file_offset)) {
371 if (io_size > entry->bytes_left) {
372 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
373 (unsigned long long)entry->bytes_left,
374 (unsigned long long)io_size);
376 entry->bytes_left -= io_size;
377 if (entry->bytes_left == 0)
378 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
382 if (!ret && cached && entry) {
384 atomic_inc(&entry->refs);
386 spin_unlock(&tree->lock);
391 * used to drop a reference on an ordered extent. This will free
392 * the extent if the last reference is dropped
394 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
396 struct list_head *cur;
397 struct btrfs_ordered_sum *sum;
399 trace_btrfs_ordered_extent_put(entry->inode, entry);
401 if (atomic_dec_and_test(&entry->refs)) {
402 while (!list_empty(&entry->list)) {
403 cur = entry->list.next;
404 sum = list_entry(cur, struct btrfs_ordered_sum, list);
405 list_del(&sum->list);
413 * remove an ordered extent from the tree. No references are dropped
414 * and you must wake_up entry->wait. You must hold the tree lock
415 * while you call this function.
417 static void __btrfs_remove_ordered_extent(struct inode *inode,
418 struct btrfs_ordered_extent *entry)
420 struct btrfs_ordered_inode_tree *tree;
421 struct btrfs_root *root = BTRFS_I(inode)->root;
422 struct rb_node *node;
424 tree = &BTRFS_I(inode)->ordered_tree;
425 node = &entry->rb_node;
426 rb_erase(node, &tree->tree);
428 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
430 spin_lock(&root->fs_info->ordered_extent_lock);
431 list_del_init(&entry->root_extent_list);
433 trace_btrfs_ordered_extent_remove(inode, entry);
436 * we have no more ordered extents for this inode and
437 * no dirty pages. We can safely remove it from the
438 * list of ordered extents
440 if (RB_EMPTY_ROOT(&tree->tree) &&
441 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
442 list_del_init(&BTRFS_I(inode)->ordered_operations);
444 spin_unlock(&root->fs_info->ordered_extent_lock);
448 * remove an ordered extent from the tree. No references are dropped
449 * but any waiters are woken.
451 void btrfs_remove_ordered_extent(struct inode *inode,
452 struct btrfs_ordered_extent *entry)
454 struct btrfs_ordered_inode_tree *tree;
456 tree = &BTRFS_I(inode)->ordered_tree;
457 spin_lock(&tree->lock);
458 __btrfs_remove_ordered_extent(inode, entry);
459 spin_unlock(&tree->lock);
460 wake_up(&entry->wait);
464 * wait for all the ordered extents in a root. This is done when balancing
465 * space between drives.
467 void btrfs_wait_ordered_extents(struct btrfs_root *root,
468 int nocow_only, int delay_iput)
470 struct list_head splice;
471 struct list_head *cur;
472 struct btrfs_ordered_extent *ordered;
475 INIT_LIST_HEAD(&splice);
477 spin_lock(&root->fs_info->ordered_extent_lock);
478 list_splice_init(&root->fs_info->ordered_extents, &splice);
479 while (!list_empty(&splice)) {
481 ordered = list_entry(cur, struct btrfs_ordered_extent,
484 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
485 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
486 list_move(&ordered->root_extent_list,
487 &root->fs_info->ordered_extents);
488 cond_resched_lock(&root->fs_info->ordered_extent_lock);
492 list_del_init(&ordered->root_extent_list);
493 atomic_inc(&ordered->refs);
496 * the inode may be getting freed (in sys_unlink path).
498 inode = igrab(ordered->inode);
500 spin_unlock(&root->fs_info->ordered_extent_lock);
503 btrfs_start_ordered_extent(inode, ordered, 1);
504 btrfs_put_ordered_extent(ordered);
506 btrfs_add_delayed_iput(inode);
510 btrfs_put_ordered_extent(ordered);
513 spin_lock(&root->fs_info->ordered_extent_lock);
515 spin_unlock(&root->fs_info->ordered_extent_lock);
519 * this is used during transaction commit to write all the inodes
520 * added to the ordered operation list. These files must be fully on
521 * disk before the transaction commits.
523 * we have two modes here, one is to just start the IO via filemap_flush
524 * and the other is to wait for all the io. When we wait, we have an
525 * extra check to make sure the ordered operation list really is empty
528 void btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
530 struct btrfs_inode *btrfs_inode;
532 struct list_head splice;
534 INIT_LIST_HEAD(&splice);
536 mutex_lock(&root->fs_info->ordered_operations_mutex);
537 spin_lock(&root->fs_info->ordered_extent_lock);
539 list_splice_init(&root->fs_info->ordered_operations, &splice);
541 while (!list_empty(&splice)) {
542 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
545 inode = &btrfs_inode->vfs_inode;
547 list_del_init(&btrfs_inode->ordered_operations);
550 * the inode may be getting freed (in sys_unlink path).
552 inode = igrab(inode);
554 if (!wait && inode) {
555 list_add_tail(&BTRFS_I(inode)->ordered_operations,
556 &root->fs_info->ordered_operations);
558 spin_unlock(&root->fs_info->ordered_extent_lock);
562 btrfs_wait_ordered_range(inode, 0, (u64)-1);
564 filemap_flush(inode->i_mapping);
565 btrfs_add_delayed_iput(inode);
569 spin_lock(&root->fs_info->ordered_extent_lock);
571 if (wait && !list_empty(&root->fs_info->ordered_operations))
574 spin_unlock(&root->fs_info->ordered_extent_lock);
575 mutex_unlock(&root->fs_info->ordered_operations_mutex);
579 * Used to start IO or wait for a given ordered extent to finish.
581 * If wait is one, this effectively waits on page writeback for all the pages
582 * in the extent, and it waits on the io completion code to insert
583 * metadata into the btree corresponding to the extent
585 void btrfs_start_ordered_extent(struct inode *inode,
586 struct btrfs_ordered_extent *entry,
589 u64 start = entry->file_offset;
590 u64 end = start + entry->len - 1;
592 trace_btrfs_ordered_extent_start(inode, entry);
595 * pages in the range can be dirty, clean or writeback. We
596 * start IO on any dirty ones so the wait doesn't stall waiting
597 * for pdflush to find them
599 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
600 filemap_fdatawrite_range(inode->i_mapping, start, end);
602 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
608 * Used to wait on ordered extents across a large range of bytes.
610 void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
614 struct btrfs_ordered_extent *ordered;
617 if (start + len < start) {
618 orig_end = INT_LIMIT(loff_t);
620 orig_end = start + len - 1;
621 if (orig_end > INT_LIMIT(loff_t))
622 orig_end = INT_LIMIT(loff_t);
625 /* start IO across the range first to instantiate any delalloc
628 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
630 /* The compression code will leave pages locked but return from
631 * writepage without setting the page writeback. Starting again
632 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
634 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
636 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
641 ordered = btrfs_lookup_first_ordered_extent(inode, end);
644 if (ordered->file_offset > orig_end) {
645 btrfs_put_ordered_extent(ordered);
648 if (ordered->file_offset + ordered->len < start) {
649 btrfs_put_ordered_extent(ordered);
653 btrfs_start_ordered_extent(inode, ordered, 1);
654 end = ordered->file_offset;
655 btrfs_put_ordered_extent(ordered);
656 if (end == 0 || end == start)
660 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
661 EXTENT_DELALLOC, 0, NULL)) {
668 * find an ordered extent corresponding to file_offset. return NULL if
669 * nothing is found, otherwise take a reference on the extent and return it
671 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
674 struct btrfs_ordered_inode_tree *tree;
675 struct rb_node *node;
676 struct btrfs_ordered_extent *entry = NULL;
678 tree = &BTRFS_I(inode)->ordered_tree;
679 spin_lock(&tree->lock);
680 node = tree_search(tree, file_offset);
684 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
685 if (!offset_in_entry(entry, file_offset))
688 atomic_inc(&entry->refs);
690 spin_unlock(&tree->lock);
694 /* Since the DIO code tries to lock a wide area we need to look for any ordered
695 * extents that exist in the range, rather than just the start of the range.
697 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
701 struct btrfs_ordered_inode_tree *tree;
702 struct rb_node *node;
703 struct btrfs_ordered_extent *entry = NULL;
705 tree = &BTRFS_I(inode)->ordered_tree;
706 spin_lock(&tree->lock);
707 node = tree_search(tree, file_offset);
709 node = tree_search(tree, file_offset + len);
715 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
716 if (range_overlaps(entry, file_offset, len))
719 if (entry->file_offset >= file_offset + len) {
724 node = rb_next(node);
730 atomic_inc(&entry->refs);
731 spin_unlock(&tree->lock);
736 * lookup and return any extent before 'file_offset'. NULL is returned
739 struct btrfs_ordered_extent *
740 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
742 struct btrfs_ordered_inode_tree *tree;
743 struct rb_node *node;
744 struct btrfs_ordered_extent *entry = NULL;
746 tree = &BTRFS_I(inode)->ordered_tree;
747 spin_lock(&tree->lock);
748 node = tree_search(tree, file_offset);
752 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
753 atomic_inc(&entry->refs);
755 spin_unlock(&tree->lock);
760 * After an extent is done, call this to conditionally update the on disk
761 * i_size. i_size is updated to cover any fully written part of the file.
763 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
764 struct btrfs_ordered_extent *ordered)
766 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
767 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
771 u64 i_size = i_size_read(inode);
772 struct rb_node *node;
773 struct rb_node *prev = NULL;
774 struct btrfs_ordered_extent *test;
778 offset = entry_end(ordered);
780 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
782 spin_lock(&tree->lock);
783 disk_i_size = BTRFS_I(inode)->disk_i_size;
786 if (disk_i_size > i_size) {
787 BTRFS_I(inode)->disk_i_size = i_size;
793 * if the disk i_size is already at the inode->i_size, or
794 * this ordered extent is inside the disk i_size, we're done
796 if (disk_i_size == i_size || offset <= disk_i_size) {
801 * we can't update the disk_isize if there are delalloc bytes
802 * between disk_i_size and this ordered extent
804 if (test_range_bit(io_tree, disk_i_size, offset - 1,
805 EXTENT_DELALLOC, 0, NULL)) {
809 * walk backward from this ordered extent to disk_i_size.
810 * if we find an ordered extent then we can't update disk i_size
814 node = rb_prev(&ordered->rb_node);
816 prev = tree_search(tree, offset);
818 * we insert file extents without involving ordered struct,
819 * so there should be no ordered struct cover this offset
822 test = rb_entry(prev, struct btrfs_ordered_extent,
824 BUG_ON(offset_in_entry(test, offset));
829 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
830 if (test->file_offset + test->len <= disk_i_size)
832 if (test->file_offset >= i_size)
834 if (test->file_offset >= disk_i_size)
836 node = rb_prev(node);
838 new_i_size = min_t(u64, offset, i_size);
841 * at this point, we know we can safely update i_size to at least
842 * the offset from this ordered extent. But, we need to
843 * walk forward and see if ios from higher up in the file have
847 node = rb_next(&ordered->rb_node);
850 node = rb_next(prev);
852 node = rb_first(&tree->tree);
857 * do we have an area where IO might have finished
858 * between our ordered extent and the next one.
860 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
861 if (test->file_offset > offset)
862 i_size_test = test->file_offset;
864 i_size_test = i_size;
868 * i_size_test is the end of a region after this ordered
869 * extent where there are no ordered extents. As long as there
870 * are no delalloc bytes in this area, it is safe to update
871 * disk_i_size to the end of the region.
873 if (i_size_test > offset &&
874 !test_range_bit(io_tree, offset, i_size_test - 1,
875 EXTENT_DELALLOC, 0, NULL)) {
876 new_i_size = min_t(u64, i_size_test, i_size);
878 BTRFS_I(inode)->disk_i_size = new_i_size;
882 * we need to remove the ordered extent with the tree lock held
883 * so that other people calling this function don't find our fully
884 * processed ordered entry and skip updating the i_size
887 __btrfs_remove_ordered_extent(inode, ordered);
888 spin_unlock(&tree->lock);
890 wake_up(&ordered->wait);
895 * search the ordered extents for one corresponding to 'offset' and
896 * try to find a checksum. This is used because we allow pages to
897 * be reclaimed before their checksum is actually put into the btree
899 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
902 struct btrfs_ordered_sum *ordered_sum;
903 struct btrfs_sector_sum *sector_sums;
904 struct btrfs_ordered_extent *ordered;
905 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
906 unsigned long num_sectors;
908 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
911 ordered = btrfs_lookup_ordered_extent(inode, offset);
915 spin_lock(&tree->lock);
916 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
917 if (disk_bytenr >= ordered_sum->bytenr) {
918 num_sectors = ordered_sum->len / sectorsize;
919 sector_sums = ordered_sum->sums;
920 for (i = 0; i < num_sectors; i++) {
921 if (sector_sums[i].bytenr == disk_bytenr) {
922 *sum = sector_sums[i].sum;
930 spin_unlock(&tree->lock);
931 btrfs_put_ordered_extent(ordered);
937 * add a given inode to the list of inodes that must be fully on
938 * disk before a transaction commit finishes.
940 * This basically gives us the ext3 style data=ordered mode, and it is mostly
941 * used to make sure renamed files are fully on disk.
943 * It is a noop if the inode is already fully on disk.
945 * If trans is not null, we'll do a friendly check for a transaction that
946 * is already flushing things and force the IO down ourselves.
948 void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
949 struct btrfs_root *root, struct inode *inode)
953 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
956 * if this file hasn't been changed since the last transaction
957 * commit, we can safely return without doing anything
959 if (last_mod < root->fs_info->last_trans_committed)
963 * the transaction is already committing. Just start the IO and
964 * don't bother with all of this list nonsense
966 if (trans && root->fs_info->running_transaction->blocked) {
967 btrfs_wait_ordered_range(inode, 0, (u64)-1);
971 spin_lock(&root->fs_info->ordered_extent_lock);
972 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
973 list_add_tail(&BTRFS_I(inode)->ordered_operations,
974 &root->fs_info->ordered_operations);
976 spin_unlock(&root->fs_info->ordered_extent_lock);