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 * This also sets the EXTENT_ORDERED bit on the range in the inode.
164 * The tree is given a single reference on the ordered extent that was
167 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
168 u64 start, u64 len, u64 disk_len, int type)
170 struct btrfs_ordered_inode_tree *tree;
171 struct rb_node *node;
172 struct btrfs_ordered_extent *entry;
174 tree = &BTRFS_I(inode)->ordered_tree;
175 entry = kzalloc(sizeof(*entry), GFP_NOFS);
179 mutex_lock(&tree->mutex);
180 entry->file_offset = file_offset;
181 entry->start = start;
183 entry->disk_len = disk_len;
184 entry->inode = inode;
185 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
186 set_bit(type, &entry->flags);
188 /* one ref for the tree */
189 atomic_set(&entry->refs, 1);
190 init_waitqueue_head(&entry->wait);
191 INIT_LIST_HEAD(&entry->list);
192 INIT_LIST_HEAD(&entry->root_extent_list);
194 node = tree_insert(&tree->tree, file_offset,
198 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
199 entry_end(entry) - 1, GFP_NOFS);
201 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
202 list_add_tail(&entry->root_extent_list,
203 &BTRFS_I(inode)->root->fs_info->ordered_extents);
204 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
206 mutex_unlock(&tree->mutex);
212 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
213 * when an ordered extent is finished. If the list covers more than one
214 * ordered extent, it is split across multiples.
216 int btrfs_add_ordered_sum(struct inode *inode,
217 struct btrfs_ordered_extent *entry,
218 struct btrfs_ordered_sum *sum)
220 struct btrfs_ordered_inode_tree *tree;
222 tree = &BTRFS_I(inode)->ordered_tree;
223 mutex_lock(&tree->mutex);
224 list_add_tail(&sum->list, &entry->list);
225 mutex_unlock(&tree->mutex);
230 * this is used to account for finished IO across a given range
231 * of the file. The IO should not span ordered extents. If
232 * a given ordered_extent is completely done, 1 is returned, otherwise
235 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
236 * to make sure this function only returns 1 once for a given ordered extent.
238 int btrfs_dec_test_ordered_pending(struct inode *inode,
239 u64 file_offset, u64 io_size)
241 struct btrfs_ordered_inode_tree *tree;
242 struct rb_node *node;
243 struct btrfs_ordered_extent *entry;
244 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
247 tree = &BTRFS_I(inode)->ordered_tree;
248 mutex_lock(&tree->mutex);
249 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
251 node = tree_search(tree, file_offset);
257 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
258 if (!offset_in_entry(entry, file_offset)) {
263 ret = test_range_bit(io_tree, entry->file_offset,
264 entry->file_offset + entry->len - 1,
267 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
269 mutex_unlock(&tree->mutex);
274 * used to drop a reference on an ordered extent. This will free
275 * the extent if the last reference is dropped
277 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
279 struct list_head *cur;
280 struct btrfs_ordered_sum *sum;
282 if (atomic_dec_and_test(&entry->refs)) {
283 while (!list_empty(&entry->list)) {
284 cur = entry->list.next;
285 sum = list_entry(cur, struct btrfs_ordered_sum, list);
286 list_del(&sum->list);
295 * remove an ordered extent from the tree. No references are dropped
296 * but, anyone waiting on this extent is woken up.
298 int btrfs_remove_ordered_extent(struct inode *inode,
299 struct btrfs_ordered_extent *entry)
301 struct btrfs_ordered_inode_tree *tree;
302 struct rb_node *node;
304 tree = &BTRFS_I(inode)->ordered_tree;
305 mutex_lock(&tree->mutex);
306 node = &entry->rb_node;
307 rb_erase(node, &tree->tree);
309 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
311 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
312 list_del_init(&entry->root_extent_list);
313 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
315 mutex_unlock(&tree->mutex);
316 wake_up(&entry->wait);
321 * wait for all the ordered extents in a root. This is done when balancing
322 * space between drives.
324 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
326 struct list_head splice;
327 struct list_head *cur;
328 struct btrfs_ordered_extent *ordered;
331 INIT_LIST_HEAD(&splice);
333 spin_lock(&root->fs_info->ordered_extent_lock);
334 list_splice_init(&root->fs_info->ordered_extents, &splice);
335 while (!list_empty(&splice)) {
337 ordered = list_entry(cur, struct btrfs_ordered_extent,
340 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) &&
341 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
342 list_move(&ordered->root_extent_list,
343 &root->fs_info->ordered_extents);
344 cond_resched_lock(&root->fs_info->ordered_extent_lock);
348 list_del_init(&ordered->root_extent_list);
349 atomic_inc(&ordered->refs);
352 * the inode may be getting freed (in sys_unlink path).
354 inode = igrab(ordered->inode);
356 spin_unlock(&root->fs_info->ordered_extent_lock);
359 btrfs_start_ordered_extent(inode, ordered, 1);
360 btrfs_put_ordered_extent(ordered);
363 btrfs_put_ordered_extent(ordered);
366 spin_lock(&root->fs_info->ordered_extent_lock);
368 spin_unlock(&root->fs_info->ordered_extent_lock);
373 * Used to start IO or wait for a given ordered extent to finish.
375 * If wait is one, this effectively waits on page writeback for all the pages
376 * in the extent, and it waits on the io completion code to insert
377 * metadata into the btree corresponding to the extent
379 void btrfs_start_ordered_extent(struct inode *inode,
380 struct btrfs_ordered_extent *entry,
383 u64 start = entry->file_offset;
384 u64 end = start + entry->len - 1;
387 * pages in the range can be dirty, clean or writeback. We
388 * start IO on any dirty ones so the wait doesn't stall waiting
389 * for pdflush to find them
391 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_ALL);
393 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
399 * Used to wait on ordered extents across a large range of bytes.
401 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
406 struct btrfs_ordered_extent *ordered;
408 if (start + len < start) {
409 orig_end = INT_LIMIT(loff_t);
411 orig_end = start + len - 1;
412 if (orig_end > INT_LIMIT(loff_t))
413 orig_end = INT_LIMIT(loff_t);
417 /* start IO across the range first to instantiate any delalloc
420 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
422 /* The compression code will leave pages locked but return from
423 * writepage without setting the page writeback. Starting again
424 * with WB_SYNC_ALL will end up waiting for the IO to actually start.
426 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_ALL);
428 btrfs_wait_on_page_writeback_range(inode->i_mapping,
429 start >> PAGE_CACHE_SHIFT,
430 orig_end >> PAGE_CACHE_SHIFT);
434 ordered = btrfs_lookup_first_ordered_extent(inode, end);
437 if (ordered->file_offset > orig_end) {
438 btrfs_put_ordered_extent(ordered);
441 if (ordered->file_offset + ordered->len < start) {
442 btrfs_put_ordered_extent(ordered);
445 btrfs_start_ordered_extent(inode, ordered, 1);
446 end = ordered->file_offset;
447 btrfs_put_ordered_extent(ordered);
448 if (end == 0 || end == start)
452 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
453 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
461 * find an ordered extent corresponding to file_offset. return NULL if
462 * nothing is found, otherwise take a reference on the extent and return it
464 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
467 struct btrfs_ordered_inode_tree *tree;
468 struct rb_node *node;
469 struct btrfs_ordered_extent *entry = NULL;
471 tree = &BTRFS_I(inode)->ordered_tree;
472 mutex_lock(&tree->mutex);
473 node = tree_search(tree, file_offset);
477 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
478 if (!offset_in_entry(entry, file_offset))
481 atomic_inc(&entry->refs);
483 mutex_unlock(&tree->mutex);
488 * lookup and return any extent before 'file_offset'. NULL is returned
491 struct btrfs_ordered_extent *
492 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
494 struct btrfs_ordered_inode_tree *tree;
495 struct rb_node *node;
496 struct btrfs_ordered_extent *entry = NULL;
498 tree = &BTRFS_I(inode)->ordered_tree;
499 mutex_lock(&tree->mutex);
500 node = tree_search(tree, file_offset);
504 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
505 atomic_inc(&entry->refs);
507 mutex_unlock(&tree->mutex);
512 * After an extent is done, call this to conditionally update the on disk
513 * i_size. i_size is updated to cover any fully written part of the file.
515 int btrfs_ordered_update_i_size(struct inode *inode,
516 struct btrfs_ordered_extent *ordered)
518 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
519 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
523 struct rb_node *node;
524 struct btrfs_ordered_extent *test;
526 mutex_lock(&tree->mutex);
527 disk_i_size = BTRFS_I(inode)->disk_i_size;
530 * if the disk i_size is already at the inode->i_size, or
531 * this ordered extent is inside the disk i_size, we're done
533 if (disk_i_size >= inode->i_size ||
534 ordered->file_offset + ordered->len <= disk_i_size) {
539 * we can't update the disk_isize if there are delalloc bytes
540 * between disk_i_size and this ordered extent
542 if (test_range_bit(io_tree, disk_i_size,
543 ordered->file_offset + ordered->len - 1,
544 EXTENT_DELALLOC, 0)) {
548 * walk backward from this ordered extent to disk_i_size.
549 * if we find an ordered extent then we can't update disk i_size
552 node = &ordered->rb_node;
554 node = rb_prev(node);
557 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
558 if (test->file_offset + test->len <= disk_i_size)
560 if (test->file_offset >= inode->i_size)
562 if (test->file_offset >= disk_i_size)
565 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
568 * at this point, we know we can safely update i_size to at least
569 * the offset from this ordered extent. But, we need to
570 * walk forward and see if ios from higher up in the file have
573 node = rb_next(&ordered->rb_node);
577 * do we have an area where IO might have finished
578 * between our ordered extent and the next one.
580 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
581 if (test->file_offset > entry_end(ordered))
582 i_size_test = test->file_offset;
584 i_size_test = i_size_read(inode);
588 * i_size_test is the end of a region after this ordered
589 * extent where there are no ordered extents. As long as there
590 * are no delalloc bytes in this area, it is safe to update
591 * disk_i_size to the end of the region.
593 if (i_size_test > entry_end(ordered) &&
594 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
595 EXTENT_DELALLOC, 0)) {
596 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
598 BTRFS_I(inode)->disk_i_size = new_i_size;
600 mutex_unlock(&tree->mutex);
605 * search the ordered extents for one corresponding to 'offset' and
606 * try to find a checksum. This is used because we allow pages to
607 * be reclaimed before their checksum is actually put into the btree
609 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
612 struct btrfs_ordered_sum *ordered_sum;
613 struct btrfs_sector_sum *sector_sums;
614 struct btrfs_ordered_extent *ordered;
615 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
616 struct list_head *cur;
617 unsigned long num_sectors;
619 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
622 ordered = btrfs_lookup_ordered_extent(inode, offset);
626 mutex_lock(&tree->mutex);
627 list_for_each_prev(cur, &ordered->list) {
628 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
629 if (disk_bytenr >= ordered_sum->bytenr) {
630 num_sectors = ordered_sum->len / sectorsize;
631 sector_sums = ordered_sum->sums;
632 for (i = 0; i < num_sectors; i++) {
633 if (sector_sums[i].bytenr == disk_bytenr) {
634 *sum = sector_sums[i].sum;
642 mutex_unlock(&tree->mutex);
643 btrfs_put_ordered_extent(ordered);
649 * taken from mm/filemap.c because it isn't exported
651 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
652 * @mapping: address space structure to write
653 * @start: offset in bytes where the range starts
654 * @end: offset in bytes where the range ends (inclusive)
655 * @sync_mode: enable synchronous operation
657 * Start writeback against all of a mapping's dirty pages that lie
658 * within the byte offsets <start, end> inclusive.
660 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
661 * opposed to a regular memory cleansing writeback. The difference between
662 * these two operations is that if a dirty page/buffer is encountered, it must
663 * be waited upon, and not just skipped over.
665 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
666 loff_t end, int sync_mode)
668 struct writeback_control wbc = {
669 .sync_mode = sync_mode,
670 .nr_to_write = mapping->nrpages * 2,
671 .range_start = start,
675 return btrfs_writepages(mapping, &wbc);
679 * taken from mm/filemap.c because it isn't exported
681 * wait_on_page_writeback_range - wait for writeback to complete
682 * @mapping: target address_space
683 * @start: beginning page index
684 * @end: ending page index
686 * Wait for writeback to complete against pages indexed by start->end
689 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
690 pgoff_t start, pgoff_t end)
700 pagevec_init(&pvec, 0);
702 while ((index <= end) &&
703 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
704 PAGECACHE_TAG_WRITEBACK,
705 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
708 for (i = 0; i < nr_pages; i++) {
709 struct page *page = pvec.pages[i];
711 /* until radix tree lookup accepts end_index */
712 if (page->index > end)
715 wait_on_page_writeback(page);
719 pagevec_release(&pvec);
723 /* Check for outstanding write errors */
724 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
726 if (test_and_clear_bit(AS_EIO, &mapping->flags))