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);
223 INIT_LIST_HEAD(&entry->trans_list);
225 trace_btrfs_ordered_extent_add(inode, entry);
227 spin_lock_irq(&tree->lock);
228 node = tree_insert(&tree->tree, file_offset,
231 ordered_data_tree_panic(inode, -EEXIST, file_offset);
232 spin_unlock_irq(&tree->lock);
234 spin_lock(&root->ordered_extent_lock);
235 list_add_tail(&entry->root_extent_list,
236 &root->ordered_extents);
237 root->nr_ordered_extents++;
238 if (root->nr_ordered_extents == 1) {
239 spin_lock(&root->fs_info->ordered_root_lock);
240 BUG_ON(!list_empty(&root->ordered_root));
241 list_add_tail(&root->ordered_root,
242 &root->fs_info->ordered_roots);
243 spin_unlock(&root->fs_info->ordered_root_lock);
245 spin_unlock(&root->ordered_extent_lock);
250 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
251 u64 start, u64 len, u64 disk_len, int type)
253 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
255 BTRFS_COMPRESS_NONE);
258 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
259 u64 start, u64 len, u64 disk_len, int type)
261 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
263 BTRFS_COMPRESS_NONE);
266 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
267 u64 start, u64 len, u64 disk_len,
268 int type, int compress_type)
270 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
276 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
277 * when an ordered extent is finished. If the list covers more than one
278 * ordered extent, it is split across multiples.
280 void btrfs_add_ordered_sum(struct inode *inode,
281 struct btrfs_ordered_extent *entry,
282 struct btrfs_ordered_sum *sum)
284 struct btrfs_ordered_inode_tree *tree;
286 tree = &BTRFS_I(inode)->ordered_tree;
287 spin_lock_irq(&tree->lock);
288 list_add_tail(&sum->list, &entry->list);
289 WARN_ON(entry->csum_bytes_left < sum->len);
290 entry->csum_bytes_left -= sum->len;
291 if (entry->csum_bytes_left == 0)
292 wake_up(&entry->wait);
293 spin_unlock_irq(&tree->lock);
297 * this is used to account for finished IO across a given range
298 * of the file. The IO may span ordered extents. If
299 * a given ordered_extent is completely done, 1 is returned, otherwise
302 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
303 * to make sure this function only returns 1 once for a given ordered extent.
305 * file_offset is updated to one byte past the range that is recorded as
306 * complete. This allows you to walk forward in the file.
308 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
309 struct btrfs_ordered_extent **cached,
310 u64 *file_offset, u64 io_size, int uptodate)
312 struct btrfs_ordered_inode_tree *tree;
313 struct rb_node *node;
314 struct btrfs_ordered_extent *entry = NULL;
321 tree = &BTRFS_I(inode)->ordered_tree;
322 spin_lock_irqsave(&tree->lock, flags);
323 node = tree_search(tree, *file_offset);
329 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
330 if (!offset_in_entry(entry, *file_offset)) {
335 dec_start = max(*file_offset, entry->file_offset);
336 dec_end = min(*file_offset + io_size, entry->file_offset +
338 *file_offset = dec_end;
339 if (dec_start > dec_end) {
340 btrfs_crit(BTRFS_I(inode)->root->fs_info,
341 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
343 to_dec = dec_end - dec_start;
344 if (to_dec > entry->bytes_left) {
345 btrfs_crit(BTRFS_I(inode)->root->fs_info,
346 "bad ordered accounting left %llu size %llu",
347 entry->bytes_left, to_dec);
349 entry->bytes_left -= to_dec;
351 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
353 if (entry->bytes_left == 0) {
354 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
355 if (waitqueue_active(&entry->wait))
356 wake_up(&entry->wait);
361 if (!ret && cached && entry) {
363 atomic_inc(&entry->refs);
365 spin_unlock_irqrestore(&tree->lock, flags);
370 * this is used to account for finished IO across a given range
371 * of the file. The IO should not span ordered extents. If
372 * a given ordered_extent is completely done, 1 is returned, otherwise
375 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
376 * to make sure this function only returns 1 once for a given ordered extent.
378 int btrfs_dec_test_ordered_pending(struct inode *inode,
379 struct btrfs_ordered_extent **cached,
380 u64 file_offset, u64 io_size, int uptodate)
382 struct btrfs_ordered_inode_tree *tree;
383 struct rb_node *node;
384 struct btrfs_ordered_extent *entry = NULL;
388 tree = &BTRFS_I(inode)->ordered_tree;
389 spin_lock_irqsave(&tree->lock, flags);
390 if (cached && *cached) {
395 node = tree_search(tree, file_offset);
401 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
403 if (!offset_in_entry(entry, file_offset)) {
408 if (io_size > entry->bytes_left) {
409 btrfs_crit(BTRFS_I(inode)->root->fs_info,
410 "bad ordered accounting left %llu size %llu",
411 entry->bytes_left, io_size);
413 entry->bytes_left -= io_size;
415 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
417 if (entry->bytes_left == 0) {
418 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
419 if (waitqueue_active(&entry->wait))
420 wake_up(&entry->wait);
425 if (!ret && cached && entry) {
427 atomic_inc(&entry->refs);
429 spin_unlock_irqrestore(&tree->lock, flags);
433 /* Needs to either be called under a log transaction or the log_mutex */
434 void btrfs_get_logged_extents(struct inode *inode,
435 struct list_head *logged_list,
439 struct btrfs_ordered_inode_tree *tree;
440 struct btrfs_ordered_extent *ordered;
442 struct rb_node *prev;
444 tree = &BTRFS_I(inode)->ordered_tree;
445 spin_lock_irq(&tree->lock);
446 n = __tree_search(&tree->tree, end, &prev);
449 for (; n; n = rb_prev(n)) {
450 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
451 if (ordered->file_offset > end)
453 if (entry_end(ordered) <= start)
455 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
457 list_add(&ordered->log_list, logged_list);
458 atomic_inc(&ordered->refs);
460 spin_unlock_irq(&tree->lock);
463 void btrfs_put_logged_extents(struct list_head *logged_list)
465 struct btrfs_ordered_extent *ordered;
467 while (!list_empty(logged_list)) {
468 ordered = list_first_entry(logged_list,
469 struct btrfs_ordered_extent,
471 list_del_init(&ordered->log_list);
472 btrfs_put_ordered_extent(ordered);
476 void btrfs_submit_logged_extents(struct list_head *logged_list,
477 struct btrfs_root *log)
479 int index = log->log_transid % 2;
481 spin_lock_irq(&log->log_extents_lock[index]);
482 list_splice_tail(logged_list, &log->logged_list[index]);
483 spin_unlock_irq(&log->log_extents_lock[index]);
486 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
487 struct btrfs_root *log, u64 transid)
489 struct btrfs_ordered_extent *ordered;
490 int index = transid % 2;
492 spin_lock_irq(&log->log_extents_lock[index]);
493 while (!list_empty(&log->logged_list[index])) {
494 ordered = list_first_entry(&log->logged_list[index],
495 struct btrfs_ordered_extent,
497 list_del_init(&ordered->log_list);
498 spin_unlock_irq(&log->log_extents_lock[index]);
500 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
501 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
502 struct inode *inode = ordered->inode;
503 u64 start = ordered->file_offset;
504 u64 end = ordered->file_offset + ordered->len - 1;
507 filemap_fdatawrite_range(inode->i_mapping, start, end);
509 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
512 list_add_tail(&ordered->trans_list, &trans->ordered);
513 spin_lock_irq(&log->log_extents_lock[index]);
515 spin_unlock_irq(&log->log_extents_lock[index]);
518 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
520 struct btrfs_ordered_extent *ordered;
521 int index = transid % 2;
523 spin_lock_irq(&log->log_extents_lock[index]);
524 while (!list_empty(&log->logged_list[index])) {
525 ordered = list_first_entry(&log->logged_list[index],
526 struct btrfs_ordered_extent,
528 list_del_init(&ordered->log_list);
529 spin_unlock_irq(&log->log_extents_lock[index]);
530 btrfs_put_ordered_extent(ordered);
531 spin_lock_irq(&log->log_extents_lock[index]);
533 spin_unlock_irq(&log->log_extents_lock[index]);
537 * used to drop a reference on an ordered extent. This will free
538 * the extent if the last reference is dropped
540 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
542 struct list_head *cur;
543 struct btrfs_ordered_sum *sum;
545 trace_btrfs_ordered_extent_put(entry->inode, entry);
547 if (atomic_dec_and_test(&entry->refs)) {
549 btrfs_add_delayed_iput(entry->inode);
550 while (!list_empty(&entry->list)) {
551 cur = entry->list.next;
552 sum = list_entry(cur, struct btrfs_ordered_sum, list);
553 list_del(&sum->list);
556 kmem_cache_free(btrfs_ordered_extent_cache, entry);
561 * remove an ordered extent from the tree. No references are dropped
562 * and waiters are woken up.
564 void btrfs_remove_ordered_extent(struct inode *inode,
565 struct btrfs_ordered_extent *entry)
567 struct btrfs_ordered_inode_tree *tree;
568 struct btrfs_root *root = BTRFS_I(inode)->root;
569 struct rb_node *node;
571 tree = &BTRFS_I(inode)->ordered_tree;
572 spin_lock_irq(&tree->lock);
573 node = &entry->rb_node;
574 rb_erase(node, &tree->tree);
575 if (tree->last == node)
577 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
578 spin_unlock_irq(&tree->lock);
580 spin_lock(&root->ordered_extent_lock);
581 list_del_init(&entry->root_extent_list);
582 root->nr_ordered_extents--;
584 trace_btrfs_ordered_extent_remove(inode, entry);
586 if (!root->nr_ordered_extents) {
587 spin_lock(&root->fs_info->ordered_root_lock);
588 BUG_ON(list_empty(&root->ordered_root));
589 list_del_init(&root->ordered_root);
590 spin_unlock(&root->fs_info->ordered_root_lock);
592 spin_unlock(&root->ordered_extent_lock);
593 wake_up(&entry->wait);
596 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
598 struct btrfs_ordered_extent *ordered;
600 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
601 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
602 complete(&ordered->completion);
606 * wait for all the ordered extents in a root. This is done when balancing
607 * space between drives.
609 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
611 struct list_head splice, works;
612 struct btrfs_ordered_extent *ordered, *next;
615 INIT_LIST_HEAD(&splice);
616 INIT_LIST_HEAD(&works);
618 mutex_lock(&root->ordered_extent_mutex);
619 spin_lock(&root->ordered_extent_lock);
620 list_splice_init(&root->ordered_extents, &splice);
621 while (!list_empty(&splice) && nr) {
622 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
624 list_move_tail(&ordered->root_extent_list,
625 &root->ordered_extents);
626 atomic_inc(&ordered->refs);
627 spin_unlock(&root->ordered_extent_lock);
629 btrfs_init_work(&ordered->flush_work,
630 btrfs_flush_delalloc_helper,
631 btrfs_run_ordered_extent_work, NULL, NULL);
632 list_add_tail(&ordered->work_list, &works);
633 btrfs_queue_work(root->fs_info->flush_workers,
634 &ordered->flush_work);
637 spin_lock(&root->ordered_extent_lock);
642 list_splice_tail(&splice, &root->ordered_extents);
643 spin_unlock(&root->ordered_extent_lock);
645 list_for_each_entry_safe(ordered, next, &works, work_list) {
646 list_del_init(&ordered->work_list);
647 wait_for_completion(&ordered->completion);
648 btrfs_put_ordered_extent(ordered);
651 mutex_unlock(&root->ordered_extent_mutex);
656 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
658 struct btrfs_root *root;
659 struct list_head splice;
662 INIT_LIST_HEAD(&splice);
664 mutex_lock(&fs_info->ordered_operations_mutex);
665 spin_lock(&fs_info->ordered_root_lock);
666 list_splice_init(&fs_info->ordered_roots, &splice);
667 while (!list_empty(&splice) && nr) {
668 root = list_first_entry(&splice, struct btrfs_root,
670 root = btrfs_grab_fs_root(root);
672 list_move_tail(&root->ordered_root,
673 &fs_info->ordered_roots);
674 spin_unlock(&fs_info->ordered_root_lock);
676 done = btrfs_wait_ordered_extents(root, nr);
677 btrfs_put_fs_root(root);
679 spin_lock(&fs_info->ordered_root_lock);
685 list_splice_tail(&splice, &fs_info->ordered_roots);
686 spin_unlock(&fs_info->ordered_root_lock);
687 mutex_unlock(&fs_info->ordered_operations_mutex);
691 * Used to start IO or wait for a given ordered extent to finish.
693 * If wait is one, this effectively waits on page writeback for all the pages
694 * in the extent, and it waits on the io completion code to insert
695 * metadata into the btree corresponding to the extent
697 void btrfs_start_ordered_extent(struct inode *inode,
698 struct btrfs_ordered_extent *entry,
701 u64 start = entry->file_offset;
702 u64 end = start + entry->len - 1;
704 trace_btrfs_ordered_extent_start(inode, entry);
707 * pages in the range can be dirty, clean or writeback. We
708 * start IO on any dirty ones so the wait doesn't stall waiting
709 * for the flusher thread to find them
711 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
712 filemap_fdatawrite_range(inode->i_mapping, start, end);
714 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
720 * Used to wait on ordered extents across a large range of bytes.
722 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
728 struct btrfs_ordered_extent *ordered;
730 if (start + len < start) {
731 orig_end = INT_LIMIT(loff_t);
733 orig_end = start + len - 1;
734 if (orig_end > INT_LIMIT(loff_t))
735 orig_end = INT_LIMIT(loff_t);
738 /* start IO across the range first to instantiate any delalloc
741 ret = btrfs_fdatawrite_range(inode, start, orig_end);
746 * If we have a writeback error don't return immediately. Wait first
747 * for any ordered extents that haven't completed yet. This is to make
748 * sure no one can dirty the same page ranges and call writepages()
749 * before the ordered extents complete - to avoid failures (-EEXIST)
750 * when adding the new ordered extents to the ordered tree.
752 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
756 ordered = btrfs_lookup_first_ordered_extent(inode, end);
759 if (ordered->file_offset > orig_end) {
760 btrfs_put_ordered_extent(ordered);
763 if (ordered->file_offset + ordered->len <= start) {
764 btrfs_put_ordered_extent(ordered);
767 btrfs_start_ordered_extent(inode, ordered, 1);
768 end = ordered->file_offset;
769 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
771 btrfs_put_ordered_extent(ordered);
772 if (ret || end == 0 || end == start)
776 return ret_wb ? ret_wb : ret;
780 * find an ordered extent corresponding to file_offset. return NULL if
781 * nothing is found, otherwise take a reference on the extent and return it
783 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
786 struct btrfs_ordered_inode_tree *tree;
787 struct rb_node *node;
788 struct btrfs_ordered_extent *entry = NULL;
790 tree = &BTRFS_I(inode)->ordered_tree;
791 spin_lock_irq(&tree->lock);
792 node = tree_search(tree, file_offset);
796 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
797 if (!offset_in_entry(entry, file_offset))
800 atomic_inc(&entry->refs);
802 spin_unlock_irq(&tree->lock);
806 /* Since the DIO code tries to lock a wide area we need to look for any ordered
807 * extents that exist in the range, rather than just the start of the range.
809 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
813 struct btrfs_ordered_inode_tree *tree;
814 struct rb_node *node;
815 struct btrfs_ordered_extent *entry = NULL;
817 tree = &BTRFS_I(inode)->ordered_tree;
818 spin_lock_irq(&tree->lock);
819 node = tree_search(tree, file_offset);
821 node = tree_search(tree, file_offset + len);
827 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
828 if (range_overlaps(entry, file_offset, len))
831 if (entry->file_offset >= file_offset + len) {
836 node = rb_next(node);
842 atomic_inc(&entry->refs);
843 spin_unlock_irq(&tree->lock);
848 * lookup and return any extent before 'file_offset'. NULL is returned
851 struct btrfs_ordered_extent *
852 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
854 struct btrfs_ordered_inode_tree *tree;
855 struct rb_node *node;
856 struct btrfs_ordered_extent *entry = NULL;
858 tree = &BTRFS_I(inode)->ordered_tree;
859 spin_lock_irq(&tree->lock);
860 node = tree_search(tree, file_offset);
864 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
865 atomic_inc(&entry->refs);
867 spin_unlock_irq(&tree->lock);
872 * After an extent is done, call this to conditionally update the on disk
873 * i_size. i_size is updated to cover any fully written part of the file.
875 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
876 struct btrfs_ordered_extent *ordered)
878 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
881 u64 i_size = i_size_read(inode);
882 struct rb_node *node;
883 struct rb_node *prev = NULL;
884 struct btrfs_ordered_extent *test;
887 spin_lock_irq(&tree->lock);
889 offset = entry_end(ordered);
890 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
892 ordered->file_offset +
893 ordered->truncated_len);
895 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
897 disk_i_size = BTRFS_I(inode)->disk_i_size;
900 if (disk_i_size > i_size) {
901 BTRFS_I(inode)->disk_i_size = i_size;
907 * if the disk i_size is already at the inode->i_size, or
908 * this ordered extent is inside the disk i_size, we're done
910 if (disk_i_size == i_size)
914 * We still need to update disk_i_size if outstanding_isize is greater
917 if (offset <= disk_i_size &&
918 (!ordered || ordered->outstanding_isize <= disk_i_size))
922 * walk backward from this ordered extent to disk_i_size.
923 * if we find an ordered extent then we can't update disk i_size
927 node = rb_prev(&ordered->rb_node);
929 prev = tree_search(tree, offset);
931 * we insert file extents without involving ordered struct,
932 * so there should be no ordered struct cover this offset
935 test = rb_entry(prev, struct btrfs_ordered_extent,
937 BUG_ON(offset_in_entry(test, offset));
941 for (; node; node = rb_prev(node)) {
942 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
944 /* We treat this entry as if it doesnt exist */
945 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
947 if (test->file_offset + test->len <= disk_i_size)
949 if (test->file_offset >= i_size)
951 if (entry_end(test) > disk_i_size) {
953 * we don't update disk_i_size now, so record this
954 * undealt i_size. Or we will not know the real
957 if (test->outstanding_isize < offset)
958 test->outstanding_isize = offset;
960 ordered->outstanding_isize >
961 test->outstanding_isize)
962 test->outstanding_isize =
963 ordered->outstanding_isize;
967 new_i_size = min_t(u64, offset, i_size);
970 * Some ordered extents may completed before the current one, and
971 * we hold the real i_size in ->outstanding_isize.
973 if (ordered && ordered->outstanding_isize > new_i_size)
974 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
975 BTRFS_I(inode)->disk_i_size = new_i_size;
979 * We need to do this because we can't remove ordered extents until
980 * after the i_disk_size has been updated and then the inode has been
981 * updated to reflect the change, so we need to tell anybody who finds
982 * this ordered extent that we've already done all the real work, we
983 * just haven't completed all the other work.
986 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
987 spin_unlock_irq(&tree->lock);
992 * search the ordered extents for one corresponding to 'offset' and
993 * try to find a checksum. This is used because we allow pages to
994 * be reclaimed before their checksum is actually put into the btree
996 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
999 struct btrfs_ordered_sum *ordered_sum;
1000 struct btrfs_ordered_extent *ordered;
1001 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1002 unsigned long num_sectors;
1004 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1007 ordered = btrfs_lookup_ordered_extent(inode, offset);
1011 spin_lock_irq(&tree->lock);
1012 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1013 if (disk_bytenr >= ordered_sum->bytenr &&
1014 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1015 i = (disk_bytenr - ordered_sum->bytenr) >>
1016 inode->i_sb->s_blocksize_bits;
1017 num_sectors = ordered_sum->len >>
1018 inode->i_sb->s_blocksize_bits;
1019 num_sectors = min_t(int, len - index, num_sectors - i);
1020 memcpy(sum + index, ordered_sum->sums + i,
1023 index += (int)num_sectors;
1026 disk_bytenr += num_sectors * sectorsize;
1030 spin_unlock_irq(&tree->lock);
1031 btrfs_put_ordered_extent(ordered);
1035 int __init ordered_data_init(void)
1037 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1038 sizeof(struct btrfs_ordered_extent), 0,
1039 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1041 if (!btrfs_ordered_extent_cache)
1047 void ordered_data_exit(void)
1049 if (btrfs_ordered_extent_cache)
1050 kmem_cache_destroy(btrfs_ordered_extent_cache);