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 nocow,
171 struct btrfs_ordered_inode_tree *tree;
172 struct rb_node *node;
173 struct btrfs_ordered_extent *entry;
175 tree = &BTRFS_I(inode)->ordered_tree;
176 entry = kzalloc(sizeof(*entry), GFP_NOFS);
180 mutex_lock(&tree->mutex);
181 entry->file_offset = file_offset;
182 entry->start = start;
184 entry->disk_len = disk_len;
185 entry->inode = inode;
187 set_bit(BTRFS_ORDERED_NOCOW, &entry->flags);
189 set_bit(BTRFS_ORDERED_COMPRESSED, &entry->flags);
191 /* one ref for the tree */
192 atomic_set(&entry->refs, 1);
193 init_waitqueue_head(&entry->wait);
194 INIT_LIST_HEAD(&entry->list);
195 INIT_LIST_HEAD(&entry->root_extent_list);
197 node = tree_insert(&tree->tree, file_offset,
200 printk("warning dup entry from add_ordered_extent\n");
203 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
204 entry_end(entry) - 1, GFP_NOFS);
206 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
207 list_add_tail(&entry->root_extent_list,
208 &BTRFS_I(inode)->root->fs_info->ordered_extents);
209 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
211 mutex_unlock(&tree->mutex);
217 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
218 * when an ordered extent is finished. If the list covers more than one
219 * ordered extent, it is split across multiples.
221 int btrfs_add_ordered_sum(struct inode *inode,
222 struct btrfs_ordered_extent *entry,
223 struct btrfs_ordered_sum *sum)
225 struct btrfs_ordered_inode_tree *tree;
227 tree = &BTRFS_I(inode)->ordered_tree;
228 mutex_lock(&tree->mutex);
229 list_add_tail(&sum->list, &entry->list);
230 mutex_unlock(&tree->mutex);
235 * this is used to account for finished IO across a given range
236 * of the file. The IO should not span ordered extents. If
237 * a given ordered_extent is completely done, 1 is returned, otherwise
240 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
241 * to make sure this function only returns 1 once for a given ordered extent.
243 int btrfs_dec_test_ordered_pending(struct inode *inode,
244 u64 file_offset, u64 io_size)
246 struct btrfs_ordered_inode_tree *tree;
247 struct rb_node *node;
248 struct btrfs_ordered_extent *entry;
249 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
252 tree = &BTRFS_I(inode)->ordered_tree;
253 mutex_lock(&tree->mutex);
254 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
256 node = tree_search(tree, file_offset);
262 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
263 if (!offset_in_entry(entry, file_offset)) {
268 ret = test_range_bit(io_tree, entry->file_offset,
269 entry->file_offset + entry->len - 1,
272 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
274 mutex_unlock(&tree->mutex);
279 * used to drop a reference on an ordered extent. This will free
280 * the extent if the last reference is dropped
282 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
284 struct list_head *cur;
285 struct btrfs_ordered_sum *sum;
287 if (atomic_dec_and_test(&entry->refs)) {
288 while(!list_empty(&entry->list)) {
289 cur = entry->list.next;
290 sum = list_entry(cur, struct btrfs_ordered_sum, list);
291 list_del(&sum->list);
300 * remove an ordered extent from the tree. No references are dropped
301 * but, anyone waiting on this extent is woken up.
303 int btrfs_remove_ordered_extent(struct inode *inode,
304 struct btrfs_ordered_extent *entry)
306 struct btrfs_ordered_inode_tree *tree;
307 struct rb_node *node;
309 tree = &BTRFS_I(inode)->ordered_tree;
310 mutex_lock(&tree->mutex);
311 node = &entry->rb_node;
312 rb_erase(node, &tree->tree);
314 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
316 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
317 list_del_init(&entry->root_extent_list);
318 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
320 mutex_unlock(&tree->mutex);
321 wake_up(&entry->wait);
326 * wait for all the ordered extents in a root. This is done when balancing
327 * space between drives.
329 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
331 struct list_head splice;
332 struct list_head *cur;
333 struct btrfs_ordered_extent *ordered;
336 INIT_LIST_HEAD(&splice);
338 spin_lock(&root->fs_info->ordered_extent_lock);
339 list_splice_init(&root->fs_info->ordered_extents, &splice);
340 while (!list_empty(&splice)) {
342 ordered = list_entry(cur, struct btrfs_ordered_extent,
345 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
346 list_move(&ordered->root_extent_list,
347 &root->fs_info->ordered_extents);
348 cond_resched_lock(&root->fs_info->ordered_extent_lock);
352 list_del_init(&ordered->root_extent_list);
353 atomic_inc(&ordered->refs);
356 * the inode may be getting freed (in sys_unlink path).
358 inode = igrab(ordered->inode);
360 spin_unlock(&root->fs_info->ordered_extent_lock);
363 btrfs_start_ordered_extent(inode, ordered, 1);
364 btrfs_put_ordered_extent(ordered);
367 btrfs_put_ordered_extent(ordered);
370 spin_lock(&root->fs_info->ordered_extent_lock);
372 spin_unlock(&root->fs_info->ordered_extent_lock);
377 * Used to start IO or wait for a given ordered extent to finish.
379 * If wait is one, this effectively waits on page writeback for all the pages
380 * in the extent, and it waits on the io completion code to insert
381 * metadata into the btree corresponding to the extent
383 void btrfs_start_ordered_extent(struct inode *inode,
384 struct btrfs_ordered_extent *entry,
387 u64 start = entry->file_offset;
388 u64 end = start + entry->len - 1;
391 * pages in the range can be dirty, clean or writeback. We
392 * start IO on any dirty ones so the wait doesn't stall waiting
393 * for pdflush to find them
395 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_NONE);
397 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
403 * Used to wait on ordered extents across a large range of bytes.
405 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
410 struct btrfs_ordered_extent *ordered;
412 if (start + len < start) {
413 orig_end = INT_LIMIT(loff_t);
415 orig_end = start + len - 1;
416 if (orig_end > INT_LIMIT(loff_t))
417 orig_end = INT_LIMIT(loff_t);
421 /* start IO across the range first to instantiate any delalloc
424 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
426 btrfs_wait_on_page_writeback_range(inode->i_mapping,
427 start >> PAGE_CACHE_SHIFT,
428 orig_end >> PAGE_CACHE_SHIFT);
432 ordered = btrfs_lookup_first_ordered_extent(inode, end);
436 if (ordered->file_offset > orig_end) {
437 btrfs_put_ordered_extent(ordered);
440 if (ordered->file_offset + ordered->len < start) {
441 btrfs_put_ordered_extent(ordered);
444 btrfs_start_ordered_extent(inode, ordered, 1);
445 end = ordered->file_offset;
446 btrfs_put_ordered_extent(ordered);
447 if (end == 0 || end == start)
451 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
452 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
453 printk("inode %lu still ordered or delalloc after wait "
454 "%llu %llu\n", inode->i_ino,
455 (unsigned long long)start,
456 (unsigned long long)orig_end);
463 * find an ordered extent corresponding to file_offset. return NULL if
464 * nothing is found, otherwise take a reference on the extent and return it
466 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
469 struct btrfs_ordered_inode_tree *tree;
470 struct rb_node *node;
471 struct btrfs_ordered_extent *entry = NULL;
473 tree = &BTRFS_I(inode)->ordered_tree;
474 mutex_lock(&tree->mutex);
475 node = tree_search(tree, file_offset);
479 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
480 if (!offset_in_entry(entry, file_offset))
483 atomic_inc(&entry->refs);
485 mutex_unlock(&tree->mutex);
490 * lookup and return any extent before 'file_offset'. NULL is returned
493 struct btrfs_ordered_extent *
494 btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset)
496 struct btrfs_ordered_inode_tree *tree;
497 struct rb_node *node;
498 struct btrfs_ordered_extent *entry = NULL;
500 tree = &BTRFS_I(inode)->ordered_tree;
501 mutex_lock(&tree->mutex);
502 node = tree_search(tree, file_offset);
506 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
507 atomic_inc(&entry->refs);
509 mutex_unlock(&tree->mutex);
514 * After an extent is done, call this to conditionally update the on disk
515 * i_size. i_size is updated to cover any fully written part of the file.
517 int btrfs_ordered_update_i_size(struct inode *inode,
518 struct btrfs_ordered_extent *ordered)
520 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
521 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
525 struct rb_node *node;
526 struct btrfs_ordered_extent *test;
528 mutex_lock(&tree->mutex);
529 disk_i_size = BTRFS_I(inode)->disk_i_size;
532 * if the disk i_size is already at the inode->i_size, or
533 * this ordered extent is inside the disk i_size, we're done
535 if (disk_i_size >= inode->i_size ||
536 ordered->file_offset + ordered->len <= disk_i_size) {
541 * we can't update the disk_isize if there are delalloc bytes
542 * between disk_i_size and this ordered extent
544 if (test_range_bit(io_tree, disk_i_size,
545 ordered->file_offset + ordered->len - 1,
546 EXTENT_DELALLOC, 0)) {
550 * walk backward from this ordered extent to disk_i_size.
551 * if we find an ordered extent then we can't update disk i_size
554 node = &ordered->rb_node;
556 node = rb_prev(node);
559 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
560 if (test->file_offset + test->len <= disk_i_size)
562 if (test->file_offset >= inode->i_size)
564 if (test->file_offset >= disk_i_size)
567 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
570 * at this point, we know we can safely update i_size to at least
571 * the offset from this ordered extent. But, we need to
572 * walk forward and see if ios from higher up in the file have
575 node = rb_next(&ordered->rb_node);
579 * do we have an area where IO might have finished
580 * between our ordered extent and the next one.
582 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
583 if (test->file_offset > entry_end(ordered)) {
584 i_size_test = test->file_offset;
587 i_size_test = i_size_read(inode);
591 * i_size_test is the end of a region after this ordered
592 * extent where there are no ordered extents. As long as there
593 * are no delalloc bytes in this area, it is safe to update
594 * disk_i_size to the end of the region.
596 if (i_size_test > entry_end(ordered) &&
597 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
598 EXTENT_DELALLOC, 0)) {
599 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
601 BTRFS_I(inode)->disk_i_size = new_i_size;
603 mutex_unlock(&tree->mutex);
608 * search the ordered extents for one corresponding to 'offset' and
609 * try to find a checksum. This is used because we allow pages to
610 * be reclaimed before their checksum is actually put into the btree
612 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u32 *sum)
614 struct btrfs_ordered_sum *ordered_sum;
615 struct btrfs_sector_sum *sector_sums;
616 struct btrfs_ordered_extent *ordered;
617 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
618 struct list_head *cur;
619 unsigned long num_sectors;
621 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
624 ordered = btrfs_lookup_ordered_extent(inode, offset);
628 mutex_lock(&tree->mutex);
629 list_for_each_prev(cur, &ordered->list) {
630 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
631 if (offset >= ordered_sum->file_offset) {
632 num_sectors = ordered_sum->len / sectorsize;
633 sector_sums = ordered_sum->sums;
634 for (i = 0; i < num_sectors; i++) {
635 if (sector_sums[i].offset == offset) {
636 *sum = sector_sums[i].sum;
644 mutex_unlock(&tree->mutex);
645 btrfs_put_ordered_extent(ordered);
651 * taken from mm/filemap.c because it isn't exported
653 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
654 * @mapping: address space structure to write
655 * @start: offset in bytes where the range starts
656 * @end: offset in bytes where the range ends (inclusive)
657 * @sync_mode: enable synchronous operation
659 * Start writeback against all of a mapping's dirty pages that lie
660 * within the byte offsets <start, end> inclusive.
662 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
663 * opposed to a regular memory cleansing writeback. The difference between
664 * these two operations is that if a dirty page/buffer is encountered, it must
665 * be waited upon, and not just skipped over.
667 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
668 loff_t end, int sync_mode)
670 struct writeback_control wbc = {
671 .sync_mode = sync_mode,
672 .nr_to_write = mapping->nrpages * 2,
673 .range_start = start,
677 return btrfs_writepages(mapping, &wbc);
681 * taken from mm/filemap.c because it isn't exported
683 * wait_on_page_writeback_range - wait for writeback to complete
684 * @mapping: target address_space
685 * @start: beginning page index
686 * @end: ending page index
688 * Wait for writeback to complete against pages indexed by start->end
691 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
692 pgoff_t start, pgoff_t end)
702 pagevec_init(&pvec, 0);
704 while ((index <= end) &&
705 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
706 PAGECACHE_TAG_WRITEBACK,
707 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
710 for (i = 0; i < nr_pages; i++) {
711 struct page *page = pvec.pages[i];
713 /* until radix tree lookup accepts end_index */
714 if (page->index > end)
717 wait_on_page_writeback(page);
721 pagevec_release(&pvec);
725 /* Check for outstanding write errors */
726 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
728 if (test_and_clear_bit(AS_EIO, &mapping->flags))