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 btrfs_root *root = BTRFS_I(inode)->root;
306 struct rb_node *node;
308 tree = &BTRFS_I(inode)->ordered_tree;
309 node = &entry->rb_node;
310 rb_erase(node, &tree->tree);
312 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
314 spin_lock(&BTRFS_I(inode)->accounting_lock);
315 WARN_ON(!BTRFS_I(inode)->outstanding_extents);
316 BTRFS_I(inode)->outstanding_extents--;
317 spin_unlock(&BTRFS_I(inode)->accounting_lock);
318 btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
321 spin_lock(&root->fs_info->ordered_extent_lock);
322 list_del_init(&entry->root_extent_list);
325 * we have no more ordered extents for this inode and
326 * no dirty pages. We can safely remove it from the
327 * list of ordered extents
329 if (RB_EMPTY_ROOT(&tree->tree) &&
330 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
331 list_del_init(&BTRFS_I(inode)->ordered_operations);
333 spin_unlock(&root->fs_info->ordered_extent_lock);
339 * remove an ordered extent from the tree. No references are dropped
340 * but any waiters are woken.
342 int btrfs_remove_ordered_extent(struct inode *inode,
343 struct btrfs_ordered_extent *entry)
345 struct btrfs_ordered_inode_tree *tree;
348 tree = &BTRFS_I(inode)->ordered_tree;
349 spin_lock(&tree->lock);
350 ret = __btrfs_remove_ordered_extent(inode, entry);
351 spin_unlock(&tree->lock);
352 wake_up(&entry->wait);
358 * wait for all the ordered extents in a root. This is done when balancing
359 * space between drives.
361 int btrfs_wait_ordered_extents(struct btrfs_root *root,
362 int nocow_only, int delay_iput)
364 struct list_head splice;
365 struct list_head *cur;
366 struct btrfs_ordered_extent *ordered;
369 INIT_LIST_HEAD(&splice);
371 spin_lock(&root->fs_info->ordered_extent_lock);
372 list_splice_init(&root->fs_info->ordered_extents, &splice);
373 while (!list_empty(&splice)) {
375 ordered = list_entry(cur, struct btrfs_ordered_extent,
378 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
379 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
380 list_move(&ordered->root_extent_list,
381 &root->fs_info->ordered_extents);
382 cond_resched_lock(&root->fs_info->ordered_extent_lock);
386 list_del_init(&ordered->root_extent_list);
387 atomic_inc(&ordered->refs);
390 * the inode may be getting freed (in sys_unlink path).
392 inode = igrab(ordered->inode);
394 spin_unlock(&root->fs_info->ordered_extent_lock);
397 btrfs_start_ordered_extent(inode, ordered, 1);
398 btrfs_put_ordered_extent(ordered);
400 btrfs_add_delayed_iput(inode);
404 btrfs_put_ordered_extent(ordered);
407 spin_lock(&root->fs_info->ordered_extent_lock);
409 spin_unlock(&root->fs_info->ordered_extent_lock);
414 * this is used during transaction commit to write all the inodes
415 * added to the ordered operation list. These files must be fully on
416 * disk before the transaction commits.
418 * we have two modes here, one is to just start the IO via filemap_flush
419 * and the other is to wait for all the io. When we wait, we have an
420 * extra check to make sure the ordered operation list really is empty
423 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
425 struct btrfs_inode *btrfs_inode;
427 struct list_head splice;
429 INIT_LIST_HEAD(&splice);
431 mutex_lock(&root->fs_info->ordered_operations_mutex);
432 spin_lock(&root->fs_info->ordered_extent_lock);
434 list_splice_init(&root->fs_info->ordered_operations, &splice);
436 while (!list_empty(&splice)) {
437 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
440 inode = &btrfs_inode->vfs_inode;
442 list_del_init(&btrfs_inode->ordered_operations);
445 * the inode may be getting freed (in sys_unlink path).
447 inode = igrab(inode);
449 if (!wait && inode) {
450 list_add_tail(&BTRFS_I(inode)->ordered_operations,
451 &root->fs_info->ordered_operations);
453 spin_unlock(&root->fs_info->ordered_extent_lock);
457 btrfs_wait_ordered_range(inode, 0, (u64)-1);
459 filemap_flush(inode->i_mapping);
460 btrfs_add_delayed_iput(inode);
464 spin_lock(&root->fs_info->ordered_extent_lock);
466 if (wait && !list_empty(&root->fs_info->ordered_operations))
469 spin_unlock(&root->fs_info->ordered_extent_lock);
470 mutex_unlock(&root->fs_info->ordered_operations_mutex);
476 * Used to start IO or wait for a given ordered extent to finish.
478 * If wait is one, this effectively waits on page writeback for all the pages
479 * in the extent, and it waits on the io completion code to insert
480 * metadata into the btree corresponding to the extent
482 void btrfs_start_ordered_extent(struct inode *inode,
483 struct btrfs_ordered_extent *entry,
486 u64 start = entry->file_offset;
487 u64 end = start + entry->len - 1;
490 * pages in the range can be dirty, clean or writeback. We
491 * start IO on any dirty ones so the wait doesn't stall waiting
492 * for pdflush to find them
494 filemap_fdatawrite_range(inode->i_mapping, start, end);
496 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
502 * Used to wait on ordered extents across a large range of bytes.
504 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
509 struct btrfs_ordered_extent *ordered;
512 if (start + len < start) {
513 orig_end = INT_LIMIT(loff_t);
515 orig_end = start + len - 1;
516 if (orig_end > INT_LIMIT(loff_t))
517 orig_end = INT_LIMIT(loff_t);
521 /* start IO across the range first to instantiate any delalloc
524 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
526 /* The compression code will leave pages locked but return from
527 * writepage without setting the page writeback. Starting again
528 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
530 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
532 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
537 ordered = btrfs_lookup_first_ordered_extent(inode, end);
540 if (ordered->file_offset > orig_end) {
541 btrfs_put_ordered_extent(ordered);
544 if (ordered->file_offset + ordered->len < start) {
545 btrfs_put_ordered_extent(ordered);
549 btrfs_start_ordered_extent(inode, ordered, 1);
550 end = ordered->file_offset;
551 btrfs_put_ordered_extent(ordered);
552 if (end == 0 || end == start)
556 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
557 EXTENT_DELALLOC, 0, NULL)) {
565 * find an ordered extent corresponding to file_offset. return NULL if
566 * nothing is found, otherwise take a reference on the extent and return it
568 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
571 struct btrfs_ordered_inode_tree *tree;
572 struct rb_node *node;
573 struct btrfs_ordered_extent *entry = NULL;
575 tree = &BTRFS_I(inode)->ordered_tree;
576 spin_lock(&tree->lock);
577 node = tree_search(tree, file_offset);
581 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
582 if (!offset_in_entry(entry, file_offset))
585 atomic_inc(&entry->refs);
587 spin_unlock(&tree->lock);
592 * lookup and return any extent before 'file_offset'. NULL is returned
595 struct btrfs_ordered_extent *
596 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
598 struct btrfs_ordered_inode_tree *tree;
599 struct rb_node *node;
600 struct btrfs_ordered_extent *entry = NULL;
602 tree = &BTRFS_I(inode)->ordered_tree;
603 spin_lock(&tree->lock);
604 node = tree_search(tree, file_offset);
608 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
609 atomic_inc(&entry->refs);
611 spin_unlock(&tree->lock);
616 * After an extent is done, call this to conditionally update the on disk
617 * i_size. i_size is updated to cover any fully written part of the file.
619 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
620 struct btrfs_ordered_extent *ordered)
622 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
623 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
627 u64 i_size = i_size_read(inode);
628 struct rb_node *node;
629 struct rb_node *prev = NULL;
630 struct btrfs_ordered_extent *test;
634 offset = entry_end(ordered);
636 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
638 spin_lock(&tree->lock);
639 disk_i_size = BTRFS_I(inode)->disk_i_size;
642 if (disk_i_size > i_size) {
643 BTRFS_I(inode)->disk_i_size = i_size;
649 * if the disk i_size is already at the inode->i_size, or
650 * this ordered extent is inside the disk i_size, we're done
652 if (disk_i_size == i_size || offset <= disk_i_size) {
657 * we can't update the disk_isize if there are delalloc bytes
658 * between disk_i_size and this ordered extent
660 if (test_range_bit(io_tree, disk_i_size, offset - 1,
661 EXTENT_DELALLOC, 0, NULL)) {
665 * walk backward from this ordered extent to disk_i_size.
666 * if we find an ordered extent then we can't update disk i_size
670 node = rb_prev(&ordered->rb_node);
672 prev = tree_search(tree, offset);
674 * we insert file extents without involving ordered struct,
675 * so there should be no ordered struct cover this offset
678 test = rb_entry(prev, struct btrfs_ordered_extent,
680 BUG_ON(offset_in_entry(test, offset));
685 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
686 if (test->file_offset + test->len <= disk_i_size)
688 if (test->file_offset >= i_size)
690 if (test->file_offset >= disk_i_size)
692 node = rb_prev(node);
694 new_i_size = min_t(u64, offset, i_size);
697 * at this point, we know we can safely update i_size to at least
698 * the offset from this ordered extent. But, we need to
699 * walk forward and see if ios from higher up in the file have
703 node = rb_next(&ordered->rb_node);
706 node = rb_next(prev);
708 node = rb_first(&tree->tree);
713 * do we have an area where IO might have finished
714 * between our ordered extent and the next one.
716 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
717 if (test->file_offset > offset)
718 i_size_test = test->file_offset;
720 i_size_test = i_size;
724 * i_size_test is the end of a region after this ordered
725 * extent where there are no ordered extents. As long as there
726 * are no delalloc bytes in this area, it is safe to update
727 * disk_i_size to the end of the region.
729 if (i_size_test > offset &&
730 !test_range_bit(io_tree, offset, i_size_test - 1,
731 EXTENT_DELALLOC, 0, NULL)) {
732 new_i_size = min_t(u64, i_size_test, i_size);
734 BTRFS_I(inode)->disk_i_size = new_i_size;
738 * we need to remove the ordered extent with the tree lock held
739 * so that other people calling this function don't find our fully
740 * processed ordered entry and skip updating the i_size
743 __btrfs_remove_ordered_extent(inode, ordered);
744 spin_unlock(&tree->lock);
746 wake_up(&ordered->wait);
751 * search the ordered extents for one corresponding to 'offset' and
752 * try to find a checksum. This is used because we allow pages to
753 * be reclaimed before their checksum is actually put into the btree
755 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
758 struct btrfs_ordered_sum *ordered_sum;
759 struct btrfs_sector_sum *sector_sums;
760 struct btrfs_ordered_extent *ordered;
761 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
762 unsigned long num_sectors;
764 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
767 ordered = btrfs_lookup_ordered_extent(inode, offset);
771 spin_lock(&tree->lock);
772 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
773 if (disk_bytenr >= ordered_sum->bytenr) {
774 num_sectors = ordered_sum->len / sectorsize;
775 sector_sums = ordered_sum->sums;
776 for (i = 0; i < num_sectors; i++) {
777 if (sector_sums[i].bytenr == disk_bytenr) {
778 *sum = sector_sums[i].sum;
786 spin_unlock(&tree->lock);
787 btrfs_put_ordered_extent(ordered);
793 * add a given inode to the list of inodes that must be fully on
794 * disk before a transaction commit finishes.
796 * This basically gives us the ext3 style data=ordered mode, and it is mostly
797 * used to make sure renamed files are fully on disk.
799 * It is a noop if the inode is already fully on disk.
801 * If trans is not null, we'll do a friendly check for a transaction that
802 * is already flushing things and force the IO down ourselves.
804 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
805 struct btrfs_root *root,
810 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
813 * if this file hasn't been changed since the last transaction
814 * commit, we can safely return without doing anything
816 if (last_mod < root->fs_info->last_trans_committed)
820 * the transaction is already committing. Just start the IO and
821 * don't bother with all of this list nonsense
823 if (trans && root->fs_info->running_transaction->blocked) {
824 btrfs_wait_ordered_range(inode, 0, (u64)-1);
828 spin_lock(&root->fs_info->ordered_extent_lock);
829 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
830 list_add_tail(&BTRFS_I(inode)->ordered_operations,
831 &root->fs_info->ordered_operations);
833 spin_unlock(&root->fs_info->ordered_extent_lock);