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 mutex_lock(&tree->mutex);
178 entry->file_offset = file_offset;
179 entry->start = start;
181 entry->disk_len = disk_len;
182 entry->bytes_left = len;
183 entry->inode = inode;
184 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
185 set_bit(type, &entry->flags);
187 /* one ref for the tree */
188 atomic_set(&entry->refs, 1);
189 init_waitqueue_head(&entry->wait);
190 INIT_LIST_HEAD(&entry->list);
191 INIT_LIST_HEAD(&entry->root_extent_list);
193 node = tree_insert(&tree->tree, file_offset,
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);
202 mutex_unlock(&tree->mutex);
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 mutex_lock(&tree->mutex);
220 list_add_tail(&sum->list, &entry->list);
221 mutex_unlock(&tree->mutex);
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 u64 file_offset, u64 io_size)
237 struct btrfs_ordered_inode_tree *tree;
238 struct rb_node *node;
239 struct btrfs_ordered_extent *entry;
242 tree = &BTRFS_I(inode)->ordered_tree;
243 mutex_lock(&tree->mutex);
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 mutex_unlock(&tree->mutex);
272 * used to drop a reference on an ordered extent. This will free
273 * the extent if the last reference is dropped
275 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
277 struct list_head *cur;
278 struct btrfs_ordered_sum *sum;
280 if (atomic_dec_and_test(&entry->refs)) {
281 while (!list_empty(&entry->list)) {
282 cur = entry->list.next;
283 sum = list_entry(cur, struct btrfs_ordered_sum, list);
284 list_del(&sum->list);
293 * remove an ordered extent from the tree. No references are dropped
294 * but, anyone waiting on this extent is woken up.
296 int btrfs_remove_ordered_extent(struct inode *inode,
297 struct btrfs_ordered_extent *entry)
299 struct btrfs_ordered_inode_tree *tree;
300 struct rb_node *node;
302 tree = &BTRFS_I(inode)->ordered_tree;
303 mutex_lock(&tree->mutex);
304 node = &entry->rb_node;
305 rb_erase(node, &tree->tree);
307 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
309 spin_lock(&BTRFS_I(inode)->accounting_lock);
310 BTRFS_I(inode)->outstanding_extents--;
311 spin_unlock(&BTRFS_I(inode)->accounting_lock);
312 btrfs_unreserve_metadata_for_delalloc(BTRFS_I(inode)->root,
315 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
316 list_del_init(&entry->root_extent_list);
319 * we have no more ordered extents for this inode and
320 * no dirty pages. We can safely remove it from the
321 * list of ordered extents
323 if (RB_EMPTY_ROOT(&tree->tree) &&
324 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
325 list_del_init(&BTRFS_I(inode)->ordered_operations);
327 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
329 mutex_unlock(&tree->mutex);
330 wake_up(&entry->wait);
335 * wait for all the ordered extents in a root. This is done when balancing
336 * space between drives.
338 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
340 struct list_head splice;
341 struct list_head *cur;
342 struct btrfs_ordered_extent *ordered;
345 INIT_LIST_HEAD(&splice);
347 spin_lock(&root->fs_info->ordered_extent_lock);
348 list_splice_init(&root->fs_info->ordered_extents, &splice);
349 while (!list_empty(&splice)) {
351 ordered = list_entry(cur, struct btrfs_ordered_extent,
354 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
355 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
356 list_move(&ordered->root_extent_list,
357 &root->fs_info->ordered_extents);
358 cond_resched_lock(&root->fs_info->ordered_extent_lock);
362 list_del_init(&ordered->root_extent_list);
363 atomic_inc(&ordered->refs);
366 * the inode may be getting freed (in sys_unlink path).
368 inode = igrab(ordered->inode);
370 spin_unlock(&root->fs_info->ordered_extent_lock);
373 btrfs_start_ordered_extent(inode, ordered, 1);
374 btrfs_put_ordered_extent(ordered);
377 btrfs_put_ordered_extent(ordered);
380 spin_lock(&root->fs_info->ordered_extent_lock);
382 spin_unlock(&root->fs_info->ordered_extent_lock);
387 * this is used during transaction commit to write all the inodes
388 * added to the ordered operation list. These files must be fully on
389 * disk before the transaction commits.
391 * we have two modes here, one is to just start the IO via filemap_flush
392 * and the other is to wait for all the io. When we wait, we have an
393 * extra check to make sure the ordered operation list really is empty
396 int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
398 struct btrfs_inode *btrfs_inode;
400 struct list_head splice;
402 INIT_LIST_HEAD(&splice);
404 mutex_lock(&root->fs_info->ordered_operations_mutex);
405 spin_lock(&root->fs_info->ordered_extent_lock);
407 list_splice_init(&root->fs_info->ordered_operations, &splice);
409 while (!list_empty(&splice)) {
410 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
413 inode = &btrfs_inode->vfs_inode;
415 list_del_init(&btrfs_inode->ordered_operations);
418 * the inode may be getting freed (in sys_unlink path).
420 inode = igrab(inode);
422 if (!wait && inode) {
423 list_add_tail(&BTRFS_I(inode)->ordered_operations,
424 &root->fs_info->ordered_operations);
426 spin_unlock(&root->fs_info->ordered_extent_lock);
430 btrfs_wait_ordered_range(inode, 0, (u64)-1);
432 filemap_flush(inode->i_mapping);
437 spin_lock(&root->fs_info->ordered_extent_lock);
439 if (wait && !list_empty(&root->fs_info->ordered_operations))
442 spin_unlock(&root->fs_info->ordered_extent_lock);
443 mutex_unlock(&root->fs_info->ordered_operations_mutex);
449 * Used to start IO or wait for a given ordered extent to finish.
451 * If wait is one, this effectively waits on page writeback for all the pages
452 * in the extent, and it waits on the io completion code to insert
453 * metadata into the btree corresponding to the extent
455 void btrfs_start_ordered_extent(struct inode *inode,
456 struct btrfs_ordered_extent *entry,
459 u64 start = entry->file_offset;
460 u64 end = start + entry->len - 1;
463 * pages in the range can be dirty, clean or writeback. We
464 * start IO on any dirty ones so the wait doesn't stall waiting
465 * for pdflush to find them
467 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_ALL);
469 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
475 * Used to wait on ordered extents across a large range of bytes.
477 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
482 struct btrfs_ordered_extent *ordered;
485 if (start + len < start) {
486 orig_end = INT_LIMIT(loff_t);
488 orig_end = start + len - 1;
489 if (orig_end > INT_LIMIT(loff_t))
490 orig_end = INT_LIMIT(loff_t);
494 /* start IO across the range first to instantiate any delalloc
497 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
499 /* The compression code will leave pages locked but return from
500 * writepage without setting the page writeback. Starting again
501 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
503 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
505 btrfs_wait_on_page_writeback_range(inode->i_mapping,
506 start >> PAGE_CACHE_SHIFT,
507 orig_end >> PAGE_CACHE_SHIFT);
512 ordered = btrfs_lookup_first_ordered_extent(inode, end);
515 if (ordered->file_offset > orig_end) {
516 btrfs_put_ordered_extent(ordered);
519 if (ordered->file_offset + ordered->len < start) {
520 btrfs_put_ordered_extent(ordered);
524 btrfs_start_ordered_extent(inode, ordered, 1);
525 end = ordered->file_offset;
526 btrfs_put_ordered_extent(ordered);
527 if (end == 0 || end == start)
531 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
532 EXTENT_DELALLOC, 0, NULL)) {
540 * find an ordered extent corresponding to file_offset. return NULL if
541 * nothing is found, otherwise take a reference on the extent and return it
543 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
546 struct btrfs_ordered_inode_tree *tree;
547 struct rb_node *node;
548 struct btrfs_ordered_extent *entry = NULL;
550 tree = &BTRFS_I(inode)->ordered_tree;
551 mutex_lock(&tree->mutex);
552 node = tree_search(tree, file_offset);
556 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
557 if (!offset_in_entry(entry, file_offset))
560 atomic_inc(&entry->refs);
562 mutex_unlock(&tree->mutex);
567 * lookup and return any extent before 'file_offset'. NULL is returned
570 struct btrfs_ordered_extent *
571 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
573 struct btrfs_ordered_inode_tree *tree;
574 struct rb_node *node;
575 struct btrfs_ordered_extent *entry = NULL;
577 tree = &BTRFS_I(inode)->ordered_tree;
578 mutex_lock(&tree->mutex);
579 node = tree_search(tree, file_offset);
583 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
584 atomic_inc(&entry->refs);
586 mutex_unlock(&tree->mutex);
591 * After an extent is done, call this to conditionally update the on disk
592 * i_size. i_size is updated to cover any fully written part of the file.
594 int btrfs_ordered_update_i_size(struct inode *inode,
595 struct btrfs_ordered_extent *ordered)
597 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
598 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
602 struct rb_node *node;
603 struct btrfs_ordered_extent *test;
605 mutex_lock(&tree->mutex);
606 disk_i_size = BTRFS_I(inode)->disk_i_size;
609 * if the disk i_size is already at the inode->i_size, or
610 * this ordered extent is inside the disk i_size, we're done
612 if (disk_i_size >= inode->i_size ||
613 ordered->file_offset + ordered->len <= disk_i_size) {
618 * we can't update the disk_isize if there are delalloc bytes
619 * between disk_i_size and this ordered extent
621 if (test_range_bit(io_tree, disk_i_size,
622 ordered->file_offset + ordered->len - 1,
623 EXTENT_DELALLOC, 0, NULL)) {
627 * walk backward from this ordered extent to disk_i_size.
628 * if we find an ordered extent then we can't update disk i_size
631 node = &ordered->rb_node;
633 node = rb_prev(node);
636 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
637 if (test->file_offset + test->len <= disk_i_size)
639 if (test->file_offset >= inode->i_size)
641 if (test->file_offset >= disk_i_size)
644 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
647 * at this point, we know we can safely update i_size to at least
648 * the offset from this ordered extent. But, we need to
649 * walk forward and see if ios from higher up in the file have
652 node = rb_next(&ordered->rb_node);
656 * do we have an area where IO might have finished
657 * between our ordered extent and the next one.
659 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
660 if (test->file_offset > entry_end(ordered))
661 i_size_test = test->file_offset;
663 i_size_test = i_size_read(inode);
667 * i_size_test is the end of a region after this ordered
668 * extent where there are no ordered extents. As long as there
669 * are no delalloc bytes in this area, it is safe to update
670 * disk_i_size to the end of the region.
672 if (i_size_test > entry_end(ordered) &&
673 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
674 EXTENT_DELALLOC, 0, NULL)) {
675 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
677 BTRFS_I(inode)->disk_i_size = new_i_size;
679 mutex_unlock(&tree->mutex);
684 * search the ordered extents for one corresponding to 'offset' and
685 * try to find a checksum. This is used because we allow pages to
686 * be reclaimed before their checksum is actually put into the btree
688 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
691 struct btrfs_ordered_sum *ordered_sum;
692 struct btrfs_sector_sum *sector_sums;
693 struct btrfs_ordered_extent *ordered;
694 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
695 unsigned long num_sectors;
697 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
700 ordered = btrfs_lookup_ordered_extent(inode, offset);
704 mutex_lock(&tree->mutex);
705 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
706 if (disk_bytenr >= ordered_sum->bytenr) {
707 num_sectors = ordered_sum->len / sectorsize;
708 sector_sums = ordered_sum->sums;
709 for (i = 0; i < num_sectors; i++) {
710 if (sector_sums[i].bytenr == disk_bytenr) {
711 *sum = sector_sums[i].sum;
719 mutex_unlock(&tree->mutex);
720 btrfs_put_ordered_extent(ordered);
726 * taken from mm/filemap.c because it isn't exported
728 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
729 * @mapping: address space structure to write
730 * @start: offset in bytes where the range starts
731 * @end: offset in bytes where the range ends (inclusive)
732 * @sync_mode: enable synchronous operation
734 * Start writeback against all of a mapping's dirty pages that lie
735 * within the byte offsets <start, end> inclusive.
737 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
738 * opposed to a regular memory cleansing writeback. The difference between
739 * these two operations is that if a dirty page/buffer is encountered, it must
740 * be waited upon, and not just skipped over.
742 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
743 loff_t end, int sync_mode)
745 struct writeback_control wbc = {
746 .sync_mode = sync_mode,
747 .nr_to_write = mapping->nrpages * 2,
748 .range_start = start,
752 return btrfs_writepages(mapping, &wbc);
756 * taken from mm/filemap.c because it isn't exported
758 * wait_on_page_writeback_range - wait for writeback to complete
759 * @mapping: target address_space
760 * @start: beginning page index
761 * @end: ending page index
763 * Wait for writeback to complete against pages indexed by start->end
766 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
767 pgoff_t start, pgoff_t end)
777 pagevec_init(&pvec, 0);
779 while ((index <= end) &&
780 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
781 PAGECACHE_TAG_WRITEBACK,
782 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
785 for (i = 0; i < nr_pages; i++) {
786 struct page *page = pvec.pages[i];
788 /* until radix tree lookup accepts end_index */
789 if (page->index > end)
792 wait_on_page_writeback(page);
796 pagevec_release(&pvec);
800 /* Check for outstanding write errors */
801 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
803 if (test_and_clear_bit(AS_EIO, &mapping->flags))
810 * add a given inode to the list of inodes that must be fully on
811 * disk before a transaction commit finishes.
813 * This basically gives us the ext3 style data=ordered mode, and it is mostly
814 * used to make sure renamed files are fully on disk.
816 * It is a noop if the inode is already fully on disk.
818 * If trans is not null, we'll do a friendly check for a transaction that
819 * is already flushing things and force the IO down ourselves.
821 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root,
827 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
830 * if this file hasn't been changed since the last transaction
831 * commit, we can safely return without doing anything
833 if (last_mod < root->fs_info->last_trans_committed)
837 * the transaction is already committing. Just start the IO and
838 * don't bother with all of this list nonsense
840 if (trans && root->fs_info->running_transaction->blocked) {
841 btrfs_wait_ordered_range(inode, 0, (u64)-1);
845 spin_lock(&root->fs_info->ordered_extent_lock);
846 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
847 list_add_tail(&BTRFS_I(inode)->ordered_operations,
848 &root->fs_info->ordered_operations);
850 spin_unlock(&root->fs_info->ordered_extent_lock);