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/gfp.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/writeback.h>
23 #include <linux/pagevec.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
27 #include "extent_io.h"
29 static u64 entry_end(struct btrfs_ordered_extent *entry)
31 if (entry->file_offset + entry->len < entry->file_offset)
33 return entry->file_offset + entry->len;
36 /* returns NULL if the insertion worked, or it returns the node it did find
39 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
42 struct rb_node **p = &root->rb_node;
43 struct rb_node *parent = NULL;
44 struct btrfs_ordered_extent *entry;
48 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
50 if (file_offset < entry->file_offset)
52 else if (file_offset >= entry_end(entry))
58 rb_link_node(node, parent, p);
59 rb_insert_color(node, root);
64 * look for a given offset in the tree, and if it can't be found return the
67 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
68 struct rb_node **prev_ret)
70 struct rb_node *n = root->rb_node;
71 struct rb_node *prev = NULL;
73 struct btrfs_ordered_extent *entry;
74 struct btrfs_ordered_extent *prev_entry = NULL;
77 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
81 if (file_offset < entry->file_offset)
83 else if (file_offset >= entry_end(entry))
91 while (prev && file_offset >= entry_end(prev_entry)) {
95 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
97 if (file_offset < entry_end(prev_entry))
103 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
105 while (prev && file_offset < entry_end(prev_entry)) {
106 test = rb_prev(prev);
109 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
118 * helper to check if a given offset is inside a given entry
120 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
122 if (file_offset < entry->file_offset ||
123 entry->file_offset + entry->len <= file_offset)
129 * look find the first ordered struct that has this offset, otherwise
130 * the first one less than this offset
132 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
135 struct rb_root *root = &tree->tree;
136 struct rb_node *prev;
138 struct btrfs_ordered_extent *entry;
141 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
143 if (offset_in_entry(entry, file_offset))
146 ret = __tree_search(root, file_offset, &prev);
154 /* allocate and add a new ordered_extent into the per-inode tree.
155 * file_offset is the logical offset in the file
157 * start is the disk block number of an extent already reserved in the
158 * extent allocation tree
160 * len is the length of the extent
162 * The tree is given a single reference on the ordered extent that was
165 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
166 u64 start, u64 len, u64 disk_len, int type)
168 struct btrfs_ordered_inode_tree *tree;
169 struct rb_node *node;
170 struct btrfs_ordered_extent *entry;
172 tree = &BTRFS_I(inode)->ordered_tree;
173 entry = kzalloc(sizeof(*entry), GFP_NOFS);
177 entry->file_offset = file_offset;
178 entry->start = start;
180 entry->disk_len = disk_len;
181 entry->bytes_left = len;
182 entry->inode = inode;
183 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
184 set_bit(type, &entry->flags);
186 /* one ref for the tree */
187 atomic_set(&entry->refs, 1);
188 init_waitqueue_head(&entry->wait);
189 INIT_LIST_HEAD(&entry->list);
190 INIT_LIST_HEAD(&entry->root_extent_list);
192 spin_lock(&tree->lock);
193 node = tree_insert(&tree->tree, file_offset,
196 spin_unlock(&tree->lock);
198 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
199 list_add_tail(&entry->root_extent_list,
200 &BTRFS_I(inode)->root->fs_info->ordered_extents);
201 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
208 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
209 * when an ordered extent is finished. If the list covers more than one
210 * ordered extent, it is split across multiples.
212 int btrfs_add_ordered_sum(struct inode *inode,
213 struct btrfs_ordered_extent *entry,
214 struct btrfs_ordered_sum *sum)
216 struct btrfs_ordered_inode_tree *tree;
218 tree = &BTRFS_I(inode)->ordered_tree;
219 spin_lock(&tree->lock);
220 list_add_tail(&sum->list, &entry->list);
221 spin_unlock(&tree->lock);
226 * this is used to account for finished IO across a given range
227 * of the file. The IO should not span ordered extents. If
228 * a given ordered_extent is completely done, 1 is returned, otherwise
231 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
232 * to make sure this function only returns 1 once for a given ordered extent.
234 int btrfs_dec_test_ordered_pending(struct inode *inode,
235 struct btrfs_ordered_extent **cached,
236 u64 file_offset, u64 io_size)
238 struct btrfs_ordered_inode_tree *tree;
239 struct rb_node *node;
240 struct btrfs_ordered_extent *entry = NULL;
243 tree = &BTRFS_I(inode)->ordered_tree;
244 spin_lock(&tree->lock);
245 node = tree_search(tree, file_offset);
251 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
252 if (!offset_in_entry(entry, file_offset)) {
257 if (io_size > entry->bytes_left) {
258 printk(KERN_CRIT "bad ordered accounting left %llu size %llu\n",
259 (unsigned long long)entry->bytes_left,
260 (unsigned long long)io_size);
262 entry->bytes_left -= io_size;
263 if (entry->bytes_left == 0)
264 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
268 if (!ret && cached && entry) {
270 atomic_inc(&entry->refs);
272 spin_unlock(&tree->lock);
277 * used to drop a reference on an ordered extent. This will free
278 * the extent if the last reference is dropped
280 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
282 struct list_head *cur;
283 struct btrfs_ordered_sum *sum;
285 if (atomic_dec_and_test(&entry->refs)) {
286 while (!list_empty(&entry->list)) {
287 cur = entry->list.next;
288 sum = list_entry(cur, struct btrfs_ordered_sum, list);
289 list_del(&sum->list);
298 * remove an ordered extent from the tree. No references are dropped
299 * and you must wake_up entry->wait. You must hold the tree lock
300 * while you call this function.
302 static int __btrfs_remove_ordered_extent(struct inode *inode,
303 struct btrfs_ordered_extent *entry)
305 struct btrfs_ordered_inode_tree *tree;
306 struct btrfs_root *root = BTRFS_I(inode)->root;
307 struct rb_node *node;
309 tree = &BTRFS_I(inode)->ordered_tree;
310 node = &entry->rb_node;
311 rb_erase(node, &tree->tree);
313 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
315 spin_lock(&BTRFS_I(inode)->accounting_lock);
316 WARN_ON(!BTRFS_I(inode)->outstanding_extents);
317 BTRFS_I(inode)->outstanding_extents--;
318 spin_unlock(&BTRFS_I(inode)->accounting_lock);
319 btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
322 spin_lock(&root->fs_info->ordered_extent_lock);
323 list_del_init(&entry->root_extent_list);
326 * we have no more ordered extents for this inode and
327 * no dirty pages. We can safely remove it from the
328 * list of ordered extents
330 if (RB_EMPTY_ROOT(&tree->tree) &&
331 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
332 list_del_init(&BTRFS_I(inode)->ordered_operations);
334 spin_unlock(&root->fs_info->ordered_extent_lock);
340 * remove an ordered extent from the tree. No references are dropped
341 * but any waiters are woken.
343 int btrfs_remove_ordered_extent(struct inode *inode,
344 struct btrfs_ordered_extent *entry)
346 struct btrfs_ordered_inode_tree *tree;
349 tree = &BTRFS_I(inode)->ordered_tree;
350 spin_lock(&tree->lock);
351 ret = __btrfs_remove_ordered_extent(inode, entry);
352 spin_unlock(&tree->lock);
353 wake_up(&entry->wait);
359 * wait for all the ordered extents in a root. This is done when balancing
360 * space between drives.
362 int btrfs_wait_ordered_extents(struct btrfs_root *root,
363 int nocow_only, int delay_iput)
365 struct list_head splice;
366 struct list_head *cur;
367 struct btrfs_ordered_extent *ordered;
370 INIT_LIST_HEAD(&splice);
372 spin_lock(&root->fs_info->ordered_extent_lock);
373 list_splice_init(&root->fs_info->ordered_extents, &splice);
374 while (!list_empty(&splice)) {
376 ordered = list_entry(cur, struct btrfs_ordered_extent,
379 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
380 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
381 list_move(&ordered->root_extent_list,
382 &root->fs_info->ordered_extents);
383 cond_resched_lock(&root->fs_info->ordered_extent_lock);
387 list_del_init(&ordered->root_extent_list);
388 atomic_inc(&ordered->refs);
391 * the inode may be getting freed (in sys_unlink path).
393 inode = igrab(ordered->inode);
395 spin_unlock(&root->fs_info->ordered_extent_lock);
398 btrfs_start_ordered_extent(inode, ordered, 1);
399 btrfs_put_ordered_extent(ordered);
401 btrfs_add_delayed_iput(inode);
405 btrfs_put_ordered_extent(ordered);
408 spin_lock(&root->fs_info->ordered_extent_lock);
410 spin_unlock(&root->fs_info->ordered_extent_lock);
415 * this is used during transaction commit to write all the inodes
416 * added to the ordered operation list. These files must be fully on
417 * disk before the transaction commits.
419 * we have two modes here, one is to just start the IO via filemap_flush
420 * and the other is to wait for all the io. When we wait, we have an
421 * extra check to make sure the ordered operation list really is empty
424 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
426 struct btrfs_inode *btrfs_inode;
428 struct list_head splice;
430 INIT_LIST_HEAD(&splice);
432 mutex_lock(&root->fs_info->ordered_operations_mutex);
433 spin_lock(&root->fs_info->ordered_extent_lock);
435 list_splice_init(&root->fs_info->ordered_operations, &splice);
437 while (!list_empty(&splice)) {
438 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
441 inode = &btrfs_inode->vfs_inode;
443 list_del_init(&btrfs_inode->ordered_operations);
446 * the inode may be getting freed (in sys_unlink path).
448 inode = igrab(inode);
450 if (!wait && inode) {
451 list_add_tail(&BTRFS_I(inode)->ordered_operations,
452 &root->fs_info->ordered_operations);
454 spin_unlock(&root->fs_info->ordered_extent_lock);
458 btrfs_wait_ordered_range(inode, 0, (u64)-1);
460 filemap_flush(inode->i_mapping);
461 btrfs_add_delayed_iput(inode);
465 spin_lock(&root->fs_info->ordered_extent_lock);
467 if (wait && !list_empty(&root->fs_info->ordered_operations))
470 spin_unlock(&root->fs_info->ordered_extent_lock);
471 mutex_unlock(&root->fs_info->ordered_operations_mutex);
477 * Used to start IO or wait for a given ordered extent to finish.
479 * If wait is one, this effectively waits on page writeback for all the pages
480 * in the extent, and it waits on the io completion code to insert
481 * metadata into the btree corresponding to the extent
483 void btrfs_start_ordered_extent(struct inode *inode,
484 struct btrfs_ordered_extent *entry,
487 u64 start = entry->file_offset;
488 u64 end = start + entry->len - 1;
491 * pages in the range can be dirty, clean or writeback. We
492 * start IO on any dirty ones so the wait doesn't stall waiting
493 * for pdflush to find them
495 filemap_fdatawrite_range(inode->i_mapping, start, end);
497 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
503 * Used to wait on ordered extents across a large range of bytes.
505 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
510 struct btrfs_ordered_extent *ordered;
513 if (start + len < start) {
514 orig_end = INT_LIMIT(loff_t);
516 orig_end = start + len - 1;
517 if (orig_end > INT_LIMIT(loff_t))
518 orig_end = INT_LIMIT(loff_t);
522 /* start IO across the range first to instantiate any delalloc
525 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
527 /* The compression code will leave pages locked but return from
528 * writepage without setting the page writeback. Starting again
529 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
531 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
533 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
538 ordered = btrfs_lookup_first_ordered_extent(inode, end);
541 if (ordered->file_offset > orig_end) {
542 btrfs_put_ordered_extent(ordered);
545 if (ordered->file_offset + ordered->len < start) {
546 btrfs_put_ordered_extent(ordered);
550 btrfs_start_ordered_extent(inode, ordered, 1);
551 end = ordered->file_offset;
552 btrfs_put_ordered_extent(ordered);
553 if (end == 0 || end == start)
557 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
558 EXTENT_DELALLOC, 0, NULL)) {
566 * find an ordered extent corresponding to file_offset. return NULL if
567 * nothing is found, otherwise take a reference on the extent and return it
569 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
572 struct btrfs_ordered_inode_tree *tree;
573 struct rb_node *node;
574 struct btrfs_ordered_extent *entry = NULL;
576 tree = &BTRFS_I(inode)->ordered_tree;
577 spin_lock(&tree->lock);
578 node = tree_search(tree, file_offset);
582 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
583 if (!offset_in_entry(entry, file_offset))
586 atomic_inc(&entry->refs);
588 spin_unlock(&tree->lock);
593 * lookup and return any extent before 'file_offset'. NULL is returned
596 struct btrfs_ordered_extent *
597 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
599 struct btrfs_ordered_inode_tree *tree;
600 struct rb_node *node;
601 struct btrfs_ordered_extent *entry = NULL;
603 tree = &BTRFS_I(inode)->ordered_tree;
604 spin_lock(&tree->lock);
605 node = tree_search(tree, file_offset);
609 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
610 atomic_inc(&entry->refs);
612 spin_unlock(&tree->lock);
617 * After an extent is done, call this to conditionally update the on disk
618 * i_size. i_size is updated to cover any fully written part of the file.
620 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
621 struct btrfs_ordered_extent *ordered)
623 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
624 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
628 u64 i_size = i_size_read(inode);
629 struct rb_node *node;
630 struct rb_node *prev = NULL;
631 struct btrfs_ordered_extent *test;
635 offset = entry_end(ordered);
637 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
639 spin_lock(&tree->lock);
640 disk_i_size = BTRFS_I(inode)->disk_i_size;
643 if (disk_i_size > i_size) {
644 BTRFS_I(inode)->disk_i_size = i_size;
650 * if the disk i_size is already at the inode->i_size, or
651 * this ordered extent is inside the disk i_size, we're done
653 if (disk_i_size == i_size || offset <= disk_i_size) {
658 * we can't update the disk_isize if there are delalloc bytes
659 * between disk_i_size and this ordered extent
661 if (test_range_bit(io_tree, disk_i_size, offset - 1,
662 EXTENT_DELALLOC, 0, NULL)) {
666 * walk backward from this ordered extent to disk_i_size.
667 * if we find an ordered extent then we can't update disk i_size
671 node = rb_prev(&ordered->rb_node);
673 prev = tree_search(tree, offset);
675 * we insert file extents without involving ordered struct,
676 * so there should be no ordered struct cover this offset
679 test = rb_entry(prev, struct btrfs_ordered_extent,
681 BUG_ON(offset_in_entry(test, offset));
686 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
687 if (test->file_offset + test->len <= disk_i_size)
689 if (test->file_offset >= i_size)
691 if (test->file_offset >= disk_i_size)
693 node = rb_prev(node);
695 new_i_size = min_t(u64, offset, i_size);
698 * at this point, we know we can safely update i_size to at least
699 * the offset from this ordered extent. But, we need to
700 * walk forward and see if ios from higher up in the file have
704 node = rb_next(&ordered->rb_node);
707 node = rb_next(prev);
709 node = rb_first(&tree->tree);
714 * do we have an area where IO might have finished
715 * between our ordered extent and the next one.
717 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
718 if (test->file_offset > offset)
719 i_size_test = test->file_offset;
721 i_size_test = i_size;
725 * i_size_test is the end of a region after this ordered
726 * extent where there are no ordered extents. As long as there
727 * are no delalloc bytes in this area, it is safe to update
728 * disk_i_size to the end of the region.
730 if (i_size_test > offset &&
731 !test_range_bit(io_tree, offset, i_size_test - 1,
732 EXTENT_DELALLOC, 0, NULL)) {
733 new_i_size = min_t(u64, i_size_test, i_size);
735 BTRFS_I(inode)->disk_i_size = new_i_size;
739 * we need to remove the ordered extent with the tree lock held
740 * so that other people calling this function don't find our fully
741 * processed ordered entry and skip updating the i_size
744 __btrfs_remove_ordered_extent(inode, ordered);
745 spin_unlock(&tree->lock);
747 wake_up(&ordered->wait);
752 * search the ordered extents for one corresponding to 'offset' and
753 * try to find a checksum. This is used because we allow pages to
754 * be reclaimed before their checksum is actually put into the btree
756 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
759 struct btrfs_ordered_sum *ordered_sum;
760 struct btrfs_sector_sum *sector_sums;
761 struct btrfs_ordered_extent *ordered;
762 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
763 unsigned long num_sectors;
765 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
768 ordered = btrfs_lookup_ordered_extent(inode, offset);
772 spin_lock(&tree->lock);
773 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
774 if (disk_bytenr >= ordered_sum->bytenr) {
775 num_sectors = ordered_sum->len / sectorsize;
776 sector_sums = ordered_sum->sums;
777 for (i = 0; i < num_sectors; i++) {
778 if (sector_sums[i].bytenr == disk_bytenr) {
779 *sum = sector_sums[i].sum;
787 spin_unlock(&tree->lock);
788 btrfs_put_ordered_extent(ordered);
794 * add a given inode to the list of inodes that must be fully on
795 * disk before a transaction commit finishes.
797 * This basically gives us the ext3 style data=ordered mode, and it is mostly
798 * used to make sure renamed files are fully on disk.
800 * It is a noop if the inode is already fully on disk.
802 * If trans is not null, we'll do a friendly check for a transaction that
803 * is already flushing things and force the IO down ourselves.
805 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
806 struct btrfs_root *root,
811 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
814 * if this file hasn't been changed since the last transaction
815 * commit, we can safely return without doing anything
817 if (last_mod < root->fs_info->last_trans_committed)
821 * the transaction is already committing. Just start the IO and
822 * don't bother with all of this list nonsense
824 if (trans && root->fs_info->running_transaction->blocked) {
825 btrfs_wait_ordered_range(inode, 0, (u64)-1);
829 spin_lock(&root->fs_info->ordered_extent_lock);
830 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
831 list_add_tail(&BTRFS_I(inode)->ordered_operations,
832 &root->fs_info->ordered_operations);
834 spin_unlock(&root->fs_info->ordered_extent_lock);