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 filemap_fdatawrite_range(inode->i_mapping, start, end);
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 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
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 filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
505 filemap_fdatawait_range(inode->i_mapping, start, orig_end);
510 ordered = btrfs_lookup_first_ordered_extent(inode, end);
513 if (ordered->file_offset > orig_end) {
514 btrfs_put_ordered_extent(ordered);
517 if (ordered->file_offset + ordered->len < start) {
518 btrfs_put_ordered_extent(ordered);
522 btrfs_start_ordered_extent(inode, ordered, 1);
523 end = ordered->file_offset;
524 btrfs_put_ordered_extent(ordered);
525 if (end == 0 || end == start)
529 if (found || test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
530 EXTENT_DELALLOC, 0, NULL)) {
538 * find an ordered extent corresponding to file_offset. return NULL if
539 * nothing is found, otherwise take a reference on the extent and return it
541 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
544 struct btrfs_ordered_inode_tree *tree;
545 struct rb_node *node;
546 struct btrfs_ordered_extent *entry = NULL;
548 tree = &BTRFS_I(inode)->ordered_tree;
549 mutex_lock(&tree->mutex);
550 node = tree_search(tree, file_offset);
554 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
555 if (!offset_in_entry(entry, file_offset))
558 atomic_inc(&entry->refs);
560 mutex_unlock(&tree->mutex);
565 * lookup and return any extent before 'file_offset'. NULL is returned
568 struct btrfs_ordered_extent *
569 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
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 mutex_lock(&tree->mutex);
577 node = tree_search(tree, file_offset);
581 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
582 atomic_inc(&entry->refs);
584 mutex_unlock(&tree->mutex);
589 * After an extent is done, call this to conditionally update the on disk
590 * i_size. i_size is updated to cover any fully written part of the file.
592 int btrfs_ordered_update_i_size(struct inode *inode,
593 struct btrfs_ordered_extent *ordered)
595 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
596 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
600 struct rb_node *node;
601 struct btrfs_ordered_extent *test;
603 mutex_lock(&tree->mutex);
604 disk_i_size = BTRFS_I(inode)->disk_i_size;
607 * if the disk i_size is already at the inode->i_size, or
608 * this ordered extent is inside the disk i_size, we're done
610 if (disk_i_size >= inode->i_size ||
611 ordered->file_offset + ordered->len <= disk_i_size) {
616 * we can't update the disk_isize if there are delalloc bytes
617 * between disk_i_size and this ordered extent
619 if (test_range_bit(io_tree, disk_i_size,
620 ordered->file_offset + ordered->len - 1,
621 EXTENT_DELALLOC, 0, NULL)) {
625 * walk backward from this ordered extent to disk_i_size.
626 * if we find an ordered extent then we can't update disk i_size
629 node = &ordered->rb_node;
631 node = rb_prev(node);
634 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
635 if (test->file_offset + test->len <= disk_i_size)
637 if (test->file_offset >= inode->i_size)
639 if (test->file_offset >= disk_i_size)
642 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
645 * at this point, we know we can safely update i_size to at least
646 * the offset from this ordered extent. But, we need to
647 * walk forward and see if ios from higher up in the file have
650 node = rb_next(&ordered->rb_node);
654 * do we have an area where IO might have finished
655 * between our ordered extent and the next one.
657 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
658 if (test->file_offset > entry_end(ordered))
659 i_size_test = test->file_offset;
661 i_size_test = i_size_read(inode);
665 * i_size_test is the end of a region after this ordered
666 * extent where there are no ordered extents. As long as there
667 * are no delalloc bytes in this area, it is safe to update
668 * disk_i_size to the end of the region.
670 if (i_size_test > entry_end(ordered) &&
671 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
672 EXTENT_DELALLOC, 0, NULL)) {
673 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
675 BTRFS_I(inode)->disk_i_size = new_i_size;
677 mutex_unlock(&tree->mutex);
682 * search the ordered extents for one corresponding to 'offset' and
683 * try to find a checksum. This is used because we allow pages to
684 * be reclaimed before their checksum is actually put into the btree
686 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
689 struct btrfs_ordered_sum *ordered_sum;
690 struct btrfs_sector_sum *sector_sums;
691 struct btrfs_ordered_extent *ordered;
692 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
693 unsigned long num_sectors;
695 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
698 ordered = btrfs_lookup_ordered_extent(inode, offset);
702 mutex_lock(&tree->mutex);
703 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
704 if (disk_bytenr >= ordered_sum->bytenr) {
705 num_sectors = ordered_sum->len / sectorsize;
706 sector_sums = ordered_sum->sums;
707 for (i = 0; i < num_sectors; i++) {
708 if (sector_sums[i].bytenr == disk_bytenr) {
709 *sum = sector_sums[i].sum;
717 mutex_unlock(&tree->mutex);
718 btrfs_put_ordered_extent(ordered);
724 * add a given inode to the list of inodes that must be fully on
725 * disk before a transaction commit finishes.
727 * This basically gives us the ext3 style data=ordered mode, and it is mostly
728 * used to make sure renamed files are fully on disk.
730 * It is a noop if the inode is already fully on disk.
732 * If trans is not null, we'll do a friendly check for a transaction that
733 * is already flushing things and force the IO down ourselves.
735 int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
736 struct btrfs_root *root,
741 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
744 * if this file hasn't been changed since the last transaction
745 * commit, we can safely return without doing anything
747 if (last_mod < root->fs_info->last_trans_committed)
751 * the transaction is already committing. Just start the IO and
752 * don't bother with all of this list nonsense
754 if (trans && root->fs_info->running_transaction->blocked) {
755 btrfs_wait_ordered_range(inode, 0, (u64)-1);
759 spin_lock(&root->fs_info->ordered_extent_lock);
760 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
761 list_add_tail(&BTRFS_I(inode)->ordered_operations,
762 &root->fs_info->ordered_operations);
764 spin_unlock(&root->fs_info->ordered_extent_lock);