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);
63 * look for a given offset in the tree, and if it can't be found return the
66 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
67 struct rb_node **prev_ret)
69 struct rb_node *n = root->rb_node;
70 struct rb_node *prev = NULL;
72 struct btrfs_ordered_extent *entry;
73 struct btrfs_ordered_extent *prev_entry = NULL;
76 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
80 if (file_offset < entry->file_offset)
82 else if (file_offset >= entry_end(entry))
90 while (prev && file_offset >= entry_end(prev_entry)) {
94 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
96 if (file_offset < entry_end(prev_entry))
102 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
104 while (prev && file_offset < entry_end(prev_entry)) {
105 test = rb_prev(prev);
108 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
117 * helper to check if a given offset is inside a given entry
119 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
121 if (file_offset < entry->file_offset ||
122 entry->file_offset + entry->len <= file_offset)
128 * look find the first ordered struct that has this offset, otherwise
129 * the first one less than this offset
131 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
134 struct rb_root *root = &tree->tree;
135 struct rb_node *prev;
137 struct btrfs_ordered_extent *entry;
140 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
142 if (offset_in_entry(entry, file_offset))
145 ret = __tree_search(root, file_offset, &prev);
153 /* allocate and add a new ordered_extent into the per-inode tree.
154 * file_offset is the logical offset in the file
156 * start is the disk block number of an extent already reserved in the
157 * extent allocation tree
159 * len is the length of the extent
161 * The tree is given a single reference on the ordered extent that was
164 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
165 u64 start, u64 len, u64 disk_len, int type)
167 struct btrfs_ordered_inode_tree *tree;
168 struct rb_node *node;
169 struct btrfs_ordered_extent *entry;
171 tree = &BTRFS_I(inode)->ordered_tree;
172 entry = kzalloc(sizeof(*entry), GFP_NOFS);
176 entry->file_offset = file_offset;
177 entry->start = start;
179 entry->disk_len = disk_len;
180 entry->bytes_left = len;
181 entry->inode = inode;
182 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
183 set_bit(type, &entry->flags);
185 /* one ref for the tree */
186 atomic_set(&entry->refs, 1);
187 init_waitqueue_head(&entry->wait);
188 INIT_LIST_HEAD(&entry->list);
189 INIT_LIST_HEAD(&entry->root_extent_list);
191 spin_lock(&tree->lock);
192 node = tree_insert(&tree->tree, file_offset,
195 spin_unlock(&tree->lock);
197 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
198 list_add_tail(&entry->root_extent_list,
199 &BTRFS_I(inode)->root->fs_info->ordered_extents);
200 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
207 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
208 * when an ordered extent is finished. If the list covers more than one
209 * ordered extent, it is split across multiples.
211 int btrfs_add_ordered_sum(struct inode *inode,
212 struct btrfs_ordered_extent *entry,
213 struct btrfs_ordered_sum *sum)
215 struct btrfs_ordered_inode_tree *tree;
217 tree = &BTRFS_I(inode)->ordered_tree;
218 spin_lock(&tree->lock);
219 list_add_tail(&sum->list, &entry->list);
220 spin_unlock(&tree->lock);
225 * this is used to account for finished IO across a given range
226 * of the file. The IO should not span ordered extents. If
227 * a given ordered_extent is completely done, 1 is returned, otherwise
230 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
231 * to make sure this function only returns 1 once for a given ordered extent.
233 int btrfs_dec_test_ordered_pending(struct inode *inode,
234 struct btrfs_ordered_extent **cached,
235 u64 file_offset, u64 io_size)
237 struct btrfs_ordered_inode_tree *tree;
238 struct rb_node *node;
239 struct btrfs_ordered_extent *entry = NULL;
242 tree = &BTRFS_I(inode)->ordered_tree;
243 spin_lock(&tree->lock);
244 node = tree_search(tree, file_offset);
250 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
251 if (!offset_in_entry(entry, file_offset)) {
256 if (io_size > entry->bytes_left) {
257 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
258 (unsigned long long)entry->bytes_left,
259 (unsigned long long)io_size);
261 entry->bytes_left -= io_size;
262 if (entry->bytes_left == 0)
263 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
267 if (!ret && cached && entry) {
269 atomic_inc(&entry->refs);
271 spin_unlock(&tree->lock);
276 * used to drop a reference on an ordered extent. This will free
277 * the extent if the last reference is dropped
279 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
281 struct list_head *cur;
282 struct btrfs_ordered_sum *sum;
284 if (atomic_dec_and_test(&entry->refs)) {
285 while (!list_empty(&entry->list)) {
286 cur = entry->list.next;
287 sum = list_entry(cur, struct btrfs_ordered_sum, list);
288 list_del(&sum->list);
297 * remove an ordered extent from the tree. No references are dropped
298 * and you must wake_up entry->wait. You must hold the tree lock
299 * while you call this function.
301 static int __btrfs_remove_ordered_extent(struct inode *inode,
302 struct btrfs_ordered_extent *entry)
304 struct btrfs_ordered_inode_tree *tree;
305 struct rb_node *node;
307 tree = &BTRFS_I(inode)->ordered_tree;
308 node = &entry->rb_node;
309 rb_erase(node, &tree->tree);
311 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
313 spin_lock(&BTRFS_I(inode)->accounting_lock);
314 BTRFS_I(inode)->outstanding_extents--;
315 spin_unlock(&BTRFS_I(inode)->accounting_lock);
316 btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
319 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
320 list_del_init(&entry->root_extent_list);
323 * we have no more ordered extents for this inode and
324 * no dirty pages. We can safely remove it from the
325 * list of ordered extents
327 if (RB_EMPTY_ROOT(&tree->tree) &&
328 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
329 list_del_init(&BTRFS_I(inode)->ordered_operations);
331 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
337 * remove an ordered extent from the tree. No references are dropped
338 * but any waiters are woken.
340 int btrfs_remove_ordered_extent(struct inode *inode,
341 struct btrfs_ordered_extent *entry)
343 struct btrfs_ordered_inode_tree *tree;
346 tree = &BTRFS_I(inode)->ordered_tree;
347 spin_lock(&tree->lock);
348 ret = __btrfs_remove_ordered_extent(inode, entry);
349 spin_unlock(&tree->lock);
350 wake_up(&entry->wait);
356 * wait for all the ordered extents in a root. This is done when balancing
357 * space between drives.
359 int btrfs_wait_ordered_extents(struct btrfs_root *root,
360 int nocow_only, int delay_iput)
362 struct list_head splice;
363 struct list_head *cur;
364 struct btrfs_ordered_extent *ordered;
367 INIT_LIST_HEAD(&splice);
369 spin_lock(&root->fs_info->ordered_extent_lock);
370 list_splice_init(&root->fs_info->ordered_extents, &splice);
371 while (!list_empty(&splice)) {
373 ordered = list_entry(cur, struct btrfs_ordered_extent,
376 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
377 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
378 list_move(&ordered->root_extent_list,
379 &root->fs_info->ordered_extents);
380 cond_resched_lock(&root->fs_info->ordered_extent_lock);
384 list_del_init(&ordered->root_extent_list);
385 atomic_inc(&ordered->refs);
388 * the inode may be getting freed (in sys_unlink path).
390 inode = igrab(ordered->inode);
392 spin_unlock(&root->fs_info->ordered_extent_lock);
395 btrfs_start_ordered_extent(inode, ordered, 1);
396 btrfs_put_ordered_extent(ordered);
398 btrfs_add_delayed_iput(inode);
402 btrfs_put_ordered_extent(ordered);
405 spin_lock(&root->fs_info->ordered_extent_lock);
407 spin_unlock(&root->fs_info->ordered_extent_lock);
412 * this is used during transaction commit to write all the inodes
413 * added to the ordered operation list. These files must be fully on
414 * disk before the transaction commits.
416 * we have two modes here, one is to just start the IO via filemap_flush
417 * and the other is to wait for all the io. When we wait, we have an
418 * extra check to make sure the ordered operation list really is empty
421 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
423 struct btrfs_inode *btrfs_inode;
425 struct list_head splice;
427 INIT_LIST_HEAD(&splice);
429 mutex_lock(&root->fs_info->ordered_operations_mutex);
430 spin_lock(&root->fs_info->ordered_extent_lock);
432 list_splice_init(&root->fs_info->ordered_operations, &splice);
434 while (!list_empty(&splice)) {
435 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
438 inode = &btrfs_inode->vfs_inode;
440 list_del_init(&btrfs_inode->ordered_operations);
443 * the inode may be getting freed (in sys_unlink path).
445 inode = igrab(inode);
447 if (!wait && inode) {
448 list_add_tail(&BTRFS_I(inode)->ordered_operations,
449 &root->fs_info->ordered_operations);
451 spin_unlock(&root->fs_info->ordered_extent_lock);
455 btrfs_wait_ordered_range(inode, 0, (u64)-1);
457 filemap_flush(inode->i_mapping);
458 btrfs_add_delayed_iput(inode);
462 spin_lock(&root->fs_info->ordered_extent_lock);
464 if (wait && !list_empty(&root->fs_info->ordered_operations))
467 spin_unlock(&root->fs_info->ordered_extent_lock);
468 mutex_unlock(&root->fs_info->ordered_operations_mutex);
474 * Used to start IO or wait for a given ordered extent to finish.
476 * If wait is one, this effectively waits on page writeback for all the pages
477 * in the extent, and it waits on the io completion code to insert
478 * metadata into the btree corresponding to the extent
480 void btrfs_start_ordered_extent(struct inode *inode,
481 struct btrfs_ordered_extent *entry,
484 u64 start = entry->file_offset;
485 u64 end = start + entry->len - 1;
488 * pages in the range can be dirty, clean or writeback. We
489 * start IO on any dirty ones so the wait doesn't stall waiting
490 * for pdflush to find them
492 filemap_fdatawrite_range(inode->i_mapping, start, end);
494 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
500 * Used to wait on ordered extents across a large range of bytes.
502 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
507 struct btrfs_ordered_extent *ordered;
510 if (start + len < start) {
511 orig_end = INT_LIMIT(loff_t);
513 orig_end = start + len - 1;
514 if (orig_end > INT_LIMIT(loff_t))
515 orig_end = INT_LIMIT(loff_t);
519 /* start IO across the range first to instantiate any delalloc
522 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
524 /* The compression code will leave pages locked but return from
525 * writepage without setting the page writeback. Starting again
526 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
528 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
530 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
535 ordered = btrfs_lookup_first_ordered_extent(inode, end);
538 if (ordered->file_offset > orig_end) {
539 btrfs_put_ordered_extent(ordered);
542 if (ordered->file_offset + ordered->len < start) {
543 btrfs_put_ordered_extent(ordered);
547 btrfs_start_ordered_extent(inode, ordered, 1);
548 end = ordered->file_offset;
549 btrfs_put_ordered_extent(ordered);
550 if (end == 0 || end == start)
554 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
555 EXTENT_DELALLOC, 0, NULL)) {
563 * find an ordered extent corresponding to file_offset. return NULL if
564 * nothing is found, otherwise take a reference on the extent and return it
566 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
569 struct btrfs_ordered_inode_tree *tree;
570 struct rb_node *node;
571 struct btrfs_ordered_extent *entry = NULL;
573 tree = &BTRFS_I(inode)->ordered_tree;
574 spin_lock(&tree->lock);
575 node = tree_search(tree, file_offset);
579 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
580 if (!offset_in_entry(entry, file_offset))
583 atomic_inc(&entry->refs);
585 spin_unlock(&tree->lock);
590 * lookup and return any extent before 'file_offset'. NULL is returned
593 struct btrfs_ordered_extent *
594 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
596 struct btrfs_ordered_inode_tree *tree;
597 struct rb_node *node;
598 struct btrfs_ordered_extent *entry = NULL;
600 tree = &BTRFS_I(inode)->ordered_tree;
601 spin_lock(&tree->lock);
602 node = tree_search(tree, file_offset);
606 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
607 atomic_inc(&entry->refs);
609 spin_unlock(&tree->lock);
614 * After an extent is done, call this to conditionally update the on disk
615 * i_size. i_size is updated to cover any fully written part of the file.
617 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
618 struct btrfs_ordered_extent *ordered)
620 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
621 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
625 u64 i_size = i_size_read(inode);
626 struct rb_node *node;
627 struct rb_node *prev = NULL;
628 struct btrfs_ordered_extent *test;
632 offset = entry_end(ordered);
634 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
636 spin_lock(&tree->lock);
637 disk_i_size = BTRFS_I(inode)->disk_i_size;
640 if (disk_i_size > i_size) {
641 BTRFS_I(inode)->disk_i_size = i_size;
647 * if the disk i_size is already at the inode->i_size, or
648 * this ordered extent is inside the disk i_size, we're done
650 if (disk_i_size == i_size || offset <= disk_i_size) {
655 * we can't update the disk_isize if there are delalloc bytes
656 * between disk_i_size and this ordered extent
658 if (test_range_bit(io_tree, disk_i_size, offset - 1,
659 EXTENT_DELALLOC, 0, NULL)) {
663 * walk backward from this ordered extent to disk_i_size.
664 * if we find an ordered extent then we can't update disk i_size
668 node = rb_prev(&ordered->rb_node);
670 prev = tree_search(tree, offset);
672 * we insert file extents without involving ordered struct,
673 * so there should be no ordered struct cover this offset
676 test = rb_entry(prev, struct btrfs_ordered_extent,
678 BUG_ON(offset_in_entry(test, offset));
683 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
684 if (test->file_offset + test->len <= disk_i_size)
686 if (test->file_offset >= i_size)
688 if (test->file_offset >= disk_i_size)
690 node = rb_prev(node);
692 new_i_size = min_t(u64, offset, i_size);
695 * at this point, we know we can safely update i_size to at least
696 * the offset from this ordered extent. But, we need to
697 * walk forward and see if ios from higher up in the file have
701 node = rb_next(&ordered->rb_node);
704 node = rb_next(prev);
706 node = rb_first(&tree->tree);
711 * do we have an area where IO might have finished
712 * between our ordered extent and the next one.
714 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
715 if (test->file_offset > offset)
716 i_size_test = test->file_offset;
718 i_size_test = i_size;
722 * i_size_test is the end of a region after this ordered
723 * extent where there are no ordered extents. As long as there
724 * are no delalloc bytes in this area, it is safe to update
725 * disk_i_size to the end of the region.
727 if (i_size_test > offset &&
728 !test_range_bit(io_tree, offset, i_size_test - 1,
729 EXTENT_DELALLOC, 0, NULL)) {
730 new_i_size = min_t(u64, i_size_test, i_size);
732 BTRFS_I(inode)->disk_i_size = new_i_size;
736 * we need to remove the ordered extent with the tree lock held
737 * so that other people calling this function don't find our fully
738 * processed ordered entry and skip updating the i_size
741 __btrfs_remove_ordered_extent(inode, ordered);
742 spin_unlock(&tree->lock);
744 wake_up(&ordered->wait);
749 * search the ordered extents for one corresponding to 'offset' and
750 * try to find a checksum. This is used because we allow pages to
751 * be reclaimed before their checksum is actually put into the btree
753 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
756 struct btrfs_ordered_sum *ordered_sum;
757 struct btrfs_sector_sum *sector_sums;
758 struct btrfs_ordered_extent *ordered;
759 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
760 unsigned long num_sectors;
762 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
765 ordered = btrfs_lookup_ordered_extent(inode, offset);
769 spin_lock(&tree->lock);
770 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
771 if (disk_bytenr >= ordered_sum->bytenr) {
772 num_sectors = ordered_sum->len / sectorsize;
773 sector_sums = ordered_sum->sums;
774 for (i = 0; i < num_sectors; i++) {
775 if (sector_sums[i].bytenr == disk_bytenr) {
776 *sum = sector_sums[i].sum;
784 spin_unlock(&tree->lock);
785 btrfs_put_ordered_extent(ordered);
791 * add a given inode to the list of inodes that must be fully on
792 * disk before a transaction commit finishes.
794 * This basically gives us the ext3 style data=ordered mode, and it is mostly
795 * used to make sure renamed files are fully on disk.
797 * It is a noop if the inode is already fully on disk.
799 * If trans is not null, we'll do a friendly check for a transaction that
800 * is already flushing things and force the IO down ourselves.
802 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
803 struct btrfs_root *root,
808 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
811 * if this file hasn't been changed since the last transaction
812 * commit, we can safely return without doing anything
814 if (last_mod < root->fs_info->last_trans_committed)
818 * the transaction is already committing. Just start the IO and
819 * don't bother with all of this list nonsense
821 if (trans && root->fs_info->running_transaction->blocked) {
822 btrfs_wait_ordered_range(inode, 0, (u64)-1);
826 spin_lock(&root->fs_info->ordered_extent_lock);
827 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
828 list_add_tail(&BTRFS_I(inode)->ordered_operations,
829 &root->fs_info->ordered_operations);
831 spin_unlock(&root->fs_info->ordered_extent_lock);