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"
30 static u64 entry_end(struct btrfs_ordered_extent *entry)
32 if (entry->file_offset + entry->len < entry->file_offset)
34 return entry->file_offset + entry->len;
37 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
40 struct rb_node ** p = &root->rb_node;
41 struct rb_node * parent = NULL;
42 struct btrfs_ordered_extent *entry;
46 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
48 if (file_offset < entry->file_offset)
50 else if (file_offset >= entry_end(entry))
56 rb_link_node(node, parent, p);
57 rb_insert_color(node, root);
61 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
62 struct rb_node **prev_ret)
64 struct rb_node * n = root->rb_node;
65 struct rb_node *prev = NULL;
67 struct btrfs_ordered_extent *entry;
68 struct btrfs_ordered_extent *prev_entry = NULL;
71 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
75 if (file_offset < entry->file_offset)
77 else if (file_offset >= entry_end(entry))
85 while(prev && file_offset >= entry_end(prev_entry)) {
89 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
91 if (file_offset < entry_end(prev_entry))
97 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
99 while(prev && file_offset < entry_end(prev_entry)) {
100 test = rb_prev(prev);
103 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
111 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
113 if (file_offset < entry->file_offset ||
114 entry->file_offset + entry->len <= file_offset)
119 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
122 struct rb_root *root = &tree->tree;
123 struct rb_node *prev;
125 struct btrfs_ordered_extent *entry;
128 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
130 if (offset_in_entry(entry, file_offset))
133 ret = __tree_search(root, file_offset, &prev);
141 /* allocate and add a new ordered_extent into the per-inode tree.
142 * file_offset is the logical offset in the file
144 * start is the disk block number of an extent already reserved in the
145 * extent allocation tree
147 * len is the length of the extent
149 * This also sets the EXTENT_ORDERED bit on the range in the inode.
151 * The tree is given a single reference on the ordered extent that was
154 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
155 u64 start, u64 len, int nocow)
157 struct btrfs_ordered_inode_tree *tree;
158 struct rb_node *node;
159 struct btrfs_ordered_extent *entry;
161 tree = &BTRFS_I(inode)->ordered_tree;
162 entry = kzalloc(sizeof(*entry), GFP_NOFS);
166 mutex_lock(&tree->mutex);
167 entry->file_offset = file_offset;
168 entry->start = start;
170 entry->inode = inode;
172 set_bit(BTRFS_ORDERED_NOCOW, &entry->flags);
174 /* one ref for the tree */
175 atomic_set(&entry->refs, 1);
176 init_waitqueue_head(&entry->wait);
177 INIT_LIST_HEAD(&entry->list);
178 INIT_LIST_HEAD(&entry->root_extent_list);
180 node = tree_insert(&tree->tree, file_offset,
183 printk("warning dup entry from add_ordered_extent\n");
186 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
187 entry_end(entry) - 1, GFP_NOFS);
189 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
190 list_add_tail(&entry->root_extent_list,
191 &BTRFS_I(inode)->root->fs_info->ordered_extents);
192 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
194 mutex_unlock(&tree->mutex);
200 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
201 * when an ordered extent is finished. If the list covers more than one
202 * ordered extent, it is split across multiples.
204 int btrfs_add_ordered_sum(struct inode *inode,
205 struct btrfs_ordered_extent *entry,
206 struct btrfs_ordered_sum *sum)
208 struct btrfs_ordered_inode_tree *tree;
210 tree = &BTRFS_I(inode)->ordered_tree;
211 mutex_lock(&tree->mutex);
212 list_add_tail(&sum->list, &entry->list);
213 mutex_unlock(&tree->mutex);
218 * this is used to account for finished IO across a given range
219 * of the file. The IO should not span ordered extents. If
220 * a given ordered_extent is completely done, 1 is returned, otherwise
223 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
224 * to make sure this function only returns 1 once for a given ordered extent.
226 int btrfs_dec_test_ordered_pending(struct inode *inode,
227 u64 file_offset, u64 io_size)
229 struct btrfs_ordered_inode_tree *tree;
230 struct rb_node *node;
231 struct btrfs_ordered_extent *entry;
232 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
235 tree = &BTRFS_I(inode)->ordered_tree;
236 mutex_lock(&tree->mutex);
237 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
239 node = tree_search(tree, file_offset);
245 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
246 if (!offset_in_entry(entry, file_offset)) {
251 ret = test_range_bit(io_tree, entry->file_offset,
252 entry->file_offset + entry->len - 1,
255 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
257 mutex_unlock(&tree->mutex);
262 * used to drop a reference on an ordered extent. This will free
263 * the extent if the last reference is dropped
265 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
267 struct list_head *cur;
268 struct btrfs_ordered_sum *sum;
270 if (atomic_dec_and_test(&entry->refs)) {
271 while(!list_empty(&entry->list)) {
272 cur = entry->list.next;
273 sum = list_entry(cur, struct btrfs_ordered_sum, list);
274 list_del(&sum->list);
283 * remove an ordered extent from the tree. No references are dropped
284 * but, anyone waiting on this extent is woken up.
286 int btrfs_remove_ordered_extent(struct inode *inode,
287 struct btrfs_ordered_extent *entry)
289 struct btrfs_ordered_inode_tree *tree;
290 struct rb_node *node;
292 tree = &BTRFS_I(inode)->ordered_tree;
293 mutex_lock(&tree->mutex);
294 node = &entry->rb_node;
295 rb_erase(node, &tree->tree);
297 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
299 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
300 list_del_init(&entry->root_extent_list);
301 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
303 mutex_unlock(&tree->mutex);
304 wake_up(&entry->wait);
308 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
310 struct list_head splice;
311 struct list_head *cur;
312 struct list_head *tmp;
313 struct btrfs_ordered_extent *ordered;
316 INIT_LIST_HEAD(&splice);
318 spin_lock(&root->fs_info->ordered_extent_lock);
319 list_splice_init(&root->fs_info->ordered_extents, &splice);
320 list_for_each_safe(cur, tmp, &splice) {
322 ordered = list_entry(cur, struct btrfs_ordered_extent,
325 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
326 cond_resched_lock(&root->fs_info->ordered_extent_lock);
330 list_del_init(&ordered->root_extent_list);
331 atomic_inc(&ordered->refs);
332 inode = ordered->inode;
335 * the inode can't go away until all the pages are gone
336 * and the pages won't go away while there is still
337 * an ordered extent and the ordered extent won't go
338 * away until it is off this list. So, we can safely
339 * increment i_count here and call iput later
341 atomic_inc(&inode->i_count);
342 spin_unlock(&root->fs_info->ordered_extent_lock);
344 btrfs_start_ordered_extent(inode, ordered, 1);
345 btrfs_put_ordered_extent(ordered);
348 spin_lock(&root->fs_info->ordered_extent_lock);
350 list_splice_init(&splice, &root->fs_info->ordered_extents);
351 spin_unlock(&root->fs_info->ordered_extent_lock);
356 * Used to start IO or wait for a given ordered extent to finish.
358 * If wait is one, this effectively waits on page writeback for all the pages
359 * in the extent, and it waits on the io completion code to insert
360 * metadata into the btree corresponding to the extent
362 void btrfs_start_ordered_extent(struct inode *inode,
363 struct btrfs_ordered_extent *entry,
366 u64 start = entry->file_offset;
367 u64 end = start + entry->len - 1;
370 * pages in the range can be dirty, clean or writeback. We
371 * start IO on any dirty ones so the wait doesn't stall waiting
372 * for pdflush to find them
374 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_NONE);
376 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
381 * Used to wait on ordered extents across a large range of bytes.
383 void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
388 struct btrfs_ordered_extent *ordered;
390 if (start + len < start) {
391 orig_end = INT_LIMIT(loff_t);
393 orig_end = start + len - 1;
394 if (orig_end > INT_LIMIT(loff_t))
395 orig_end = INT_LIMIT(loff_t);
399 /* start IO across the range first to instantiate any delalloc
402 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
404 btrfs_wait_on_page_writeback_range(inode->i_mapping,
405 start >> PAGE_CACHE_SHIFT,
406 orig_end >> PAGE_CACHE_SHIFT);
410 ordered = btrfs_lookup_first_ordered_extent(inode, end);
414 if (ordered->file_offset > orig_end) {
415 btrfs_put_ordered_extent(ordered);
418 if (ordered->file_offset + ordered->len < start) {
419 btrfs_put_ordered_extent(ordered);
422 btrfs_start_ordered_extent(inode, ordered, 1);
423 end = ordered->file_offset;
424 btrfs_put_ordered_extent(ordered);
425 if (end == 0 || end == start)
429 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
430 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
431 printk("inode %lu still ordered or delalloc after wait "
432 "%llu %llu\n", inode->i_ino,
433 (unsigned long long)start,
434 (unsigned long long)orig_end);
440 * find an ordered extent corresponding to file_offset. return NULL if
441 * nothing is found, otherwise take a reference on the extent and return it
443 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
446 struct btrfs_ordered_inode_tree *tree;
447 struct rb_node *node;
448 struct btrfs_ordered_extent *entry = NULL;
450 tree = &BTRFS_I(inode)->ordered_tree;
451 mutex_lock(&tree->mutex);
452 node = tree_search(tree, file_offset);
456 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
457 if (!offset_in_entry(entry, file_offset))
460 atomic_inc(&entry->refs);
462 mutex_unlock(&tree->mutex);
467 * lookup and return any extent before 'file_offset'. NULL is returned
470 struct btrfs_ordered_extent *
471 btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset)
473 struct btrfs_ordered_inode_tree *tree;
474 struct rb_node *node;
475 struct btrfs_ordered_extent *entry = NULL;
477 tree = &BTRFS_I(inode)->ordered_tree;
478 mutex_lock(&tree->mutex);
479 node = tree_search(tree, file_offset);
483 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
484 atomic_inc(&entry->refs);
486 mutex_unlock(&tree->mutex);
491 * After an extent is done, call this to conditionally update the on disk
492 * i_size. i_size is updated to cover any fully written part of the file.
494 int btrfs_ordered_update_i_size(struct inode *inode,
495 struct btrfs_ordered_extent *ordered)
497 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
498 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
502 struct rb_node *node;
503 struct btrfs_ordered_extent *test;
505 mutex_lock(&tree->mutex);
506 disk_i_size = BTRFS_I(inode)->disk_i_size;
509 * if the disk i_size is already at the inode->i_size, or
510 * this ordered extent is inside the disk i_size, we're done
512 if (disk_i_size >= inode->i_size ||
513 ordered->file_offset + ordered->len <= disk_i_size) {
518 * we can't update the disk_isize if there are delalloc bytes
519 * between disk_i_size and this ordered extent
521 if (test_range_bit(io_tree, disk_i_size,
522 ordered->file_offset + ordered->len - 1,
523 EXTENT_DELALLOC, 0)) {
527 * walk backward from this ordered extent to disk_i_size.
528 * if we find an ordered extent then we can't update disk i_size
531 node = &ordered->rb_node;
533 node = rb_prev(node);
536 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
537 if (test->file_offset + test->len <= disk_i_size)
539 if (test->file_offset >= inode->i_size)
541 if (test->file_offset >= disk_i_size)
544 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
547 * at this point, we know we can safely update i_size to at least
548 * the offset from this ordered extent. But, we need to
549 * walk forward and see if ios from higher up in the file have
552 node = rb_next(&ordered->rb_node);
556 * do we have an area where IO might have finished
557 * between our ordered extent and the next one.
559 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
560 if (test->file_offset > entry_end(ordered)) {
561 i_size_test = test->file_offset;
564 i_size_test = i_size_read(inode);
568 * i_size_test is the end of a region after this ordered
569 * extent where there are no ordered extents. As long as there
570 * are no delalloc bytes in this area, it is safe to update
571 * disk_i_size to the end of the region.
573 if (i_size_test > entry_end(ordered) &&
574 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
575 EXTENT_DELALLOC, 0)) {
576 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
578 BTRFS_I(inode)->disk_i_size = new_i_size;
580 mutex_unlock(&tree->mutex);
585 * search the ordered extents for one corresponding to 'offset' and
586 * try to find a checksum. This is used because we allow pages to
587 * be reclaimed before their checksum is actually put into the btree
589 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u32 *sum)
591 struct btrfs_ordered_sum *ordered_sum;
592 struct btrfs_sector_sum *sector_sums;
593 struct btrfs_ordered_extent *ordered;
594 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
595 struct list_head *cur;
596 unsigned long num_sectors;
598 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
601 ordered = btrfs_lookup_ordered_extent(inode, offset);
605 mutex_lock(&tree->mutex);
606 list_for_each_prev(cur, &ordered->list) {
607 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
608 if (offset >= ordered_sum->file_offset) {
609 num_sectors = ordered_sum->len / sectorsize;
610 sector_sums = ordered_sum->sums;
611 for (i = 0; i < num_sectors; i++) {
612 if (sector_sums[i].offset == offset) {
613 *sum = sector_sums[i].sum;
621 mutex_unlock(&tree->mutex);
622 btrfs_put_ordered_extent(ordered);
628 * taken from mm/filemap.c because it isn't exported
630 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
631 * @mapping: address space structure to write
632 * @start: offset in bytes where the range starts
633 * @end: offset in bytes where the range ends (inclusive)
634 * @sync_mode: enable synchronous operation
636 * Start writeback against all of a mapping's dirty pages that lie
637 * within the byte offsets <start, end> inclusive.
639 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
640 * opposed to a regular memory cleansing writeback. The difference between
641 * these two operations is that if a dirty page/buffer is encountered, it must
642 * be waited upon, and not just skipped over.
644 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
645 loff_t end, int sync_mode)
647 struct writeback_control wbc = {
648 .sync_mode = sync_mode,
649 .nr_to_write = mapping->nrpages * 2,
650 .range_start = start,
654 return btrfs_writepages(mapping, &wbc);
658 * taken from mm/filemap.c because it isn't exported
660 * wait_on_page_writeback_range - wait for writeback to complete
661 * @mapping: target address_space
662 * @start: beginning page index
663 * @end: ending page index
665 * Wait for writeback to complete against pages indexed by start->end
668 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
669 pgoff_t start, pgoff_t end)
679 pagevec_init(&pvec, 0);
681 while ((index <= end) &&
682 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
683 PAGECACHE_TAG_WRITEBACK,
684 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
687 for (i = 0; i < nr_pages; i++) {
688 struct page *page = pvec.pages[i];
690 /* until radix tree lookup accepts end_index */
691 if (page->index > end)
694 wait_on_page_writeback(page);
698 pagevec_release(&pvec);
702 /* Check for outstanding write errors */
703 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
705 if (test_and_clear_bit(AS_EIO, &mapping->flags))