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,
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 /* one ref for the tree */
171 atomic_set(&entry->refs, 1);
172 init_waitqueue_head(&entry->wait);
173 INIT_LIST_HEAD(&entry->list);
175 node = tree_insert(&tree->tree, file_offset,
178 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
179 atomic_inc(&entry->refs);
181 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
182 entry_end(entry) - 1, GFP_NOFS);
184 mutex_unlock(&tree->mutex);
190 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
191 * when an ordered extent is finished. If the list covers more than one
192 * ordered extent, it is split across multiples.
194 int btrfs_add_ordered_sum(struct inode *inode,
195 struct btrfs_ordered_extent *entry,
196 struct btrfs_ordered_sum *sum)
198 struct btrfs_ordered_inode_tree *tree;
200 tree = &BTRFS_I(inode)->ordered_tree;
201 mutex_lock(&tree->mutex);
202 list_add_tail(&sum->list, &entry->list);
203 mutex_unlock(&tree->mutex);
208 * this is used to account for finished IO across a given range
209 * of the file. The IO should not span ordered extents. If
210 * a given ordered_extent is completely done, 1 is returned, otherwise
213 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
214 * to make sure this function only returns 1 once for a given ordered extent.
216 int btrfs_dec_test_ordered_pending(struct inode *inode,
217 u64 file_offset, u64 io_size)
219 struct btrfs_ordered_inode_tree *tree;
220 struct rb_node *node;
221 struct btrfs_ordered_extent *entry;
222 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
225 tree = &BTRFS_I(inode)->ordered_tree;
226 mutex_lock(&tree->mutex);
227 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
229 node = tree_search(tree, file_offset);
235 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
236 if (!offset_in_entry(entry, file_offset)) {
241 ret = test_range_bit(io_tree, entry->file_offset,
242 entry->file_offset + entry->len - 1,
245 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
247 mutex_unlock(&tree->mutex);
252 * used to drop a reference on an ordered extent. This will free
253 * the extent if the last reference is dropped
255 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
257 struct list_head *cur;
258 struct btrfs_ordered_sum *sum;
260 if (atomic_dec_and_test(&entry->refs)) {
261 while(!list_empty(&entry->list)) {
262 cur = entry->list.next;
263 sum = list_entry(cur, struct btrfs_ordered_sum, list);
264 list_del(&sum->list);
273 * remove an ordered extent from the tree. No references are dropped
274 * but, anyone waiting on this extent is woken up.
276 int btrfs_remove_ordered_extent(struct inode *inode,
277 struct btrfs_ordered_extent *entry)
279 struct btrfs_ordered_inode_tree *tree;
280 struct rb_node *node;
282 tree = &BTRFS_I(inode)->ordered_tree;
283 mutex_lock(&tree->mutex);
284 node = &entry->rb_node;
285 rb_erase(node, &tree->tree);
287 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
288 mutex_unlock(&tree->mutex);
289 wake_up(&entry->wait);
294 * Used to start IO or wait for a given ordered extent to finish.
296 * If wait is one, this effectively waits on page writeback for all the pages
297 * in the extent, and it waits on the io completion code to insert
298 * metadata into the btree corresponding to the extent
300 void btrfs_start_ordered_extent(struct inode *inode,
301 struct btrfs_ordered_extent *entry,
304 u64 start = entry->file_offset;
305 u64 end = start + entry->len - 1;
308 * pages in the range can be dirty, clean or writeback. We
309 * start IO on any dirty ones so the wait doesn't stall waiting
310 * for pdflush to find them
312 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_NONE);
314 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
319 * Used to wait on ordered extents across a large range of bytes.
321 void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
326 struct btrfs_ordered_extent *ordered;
328 if (start + len < start) {
329 orig_end = INT_LIMIT(loff_t);
331 orig_end = start + len - 1;
332 if (orig_end > INT_LIMIT(loff_t))
333 orig_end = INT_LIMIT(loff_t);
337 /* start IO across the range first to instantiate any delalloc
340 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
342 btrfs_wait_on_page_writeback_range(inode->i_mapping,
343 start >> PAGE_CACHE_SHIFT,
344 orig_end >> PAGE_CACHE_SHIFT);
348 ordered = btrfs_lookup_first_ordered_extent(inode, end);
352 if (ordered->file_offset > orig_end) {
353 btrfs_put_ordered_extent(ordered);
356 if (ordered->file_offset + ordered->len < start) {
357 btrfs_put_ordered_extent(ordered);
360 btrfs_start_ordered_extent(inode, ordered, 1);
361 end = ordered->file_offset;
362 btrfs_put_ordered_extent(ordered);
363 if (end == 0 || end == start)
367 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
368 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
369 printk("inode %lu still ordered or delalloc after wait "
370 "%llu %llu\n", inode->i_ino,
371 (unsigned long long)start,
372 (unsigned long long)orig_end);
378 * find an ordered extent corresponding to file_offset. return NULL if
379 * nothing is found, otherwise take a reference on the extent and return it
381 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
384 struct btrfs_ordered_inode_tree *tree;
385 struct rb_node *node;
386 struct btrfs_ordered_extent *entry = NULL;
388 tree = &BTRFS_I(inode)->ordered_tree;
389 mutex_lock(&tree->mutex);
390 node = tree_search(tree, file_offset);
394 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
395 if (!offset_in_entry(entry, file_offset))
398 atomic_inc(&entry->refs);
400 mutex_unlock(&tree->mutex);
405 * lookup and return any extent before 'file_offset'. NULL is returned
408 struct btrfs_ordered_extent *
409 btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset)
411 struct btrfs_ordered_inode_tree *tree;
412 struct rb_node *node;
413 struct btrfs_ordered_extent *entry = NULL;
415 tree = &BTRFS_I(inode)->ordered_tree;
416 mutex_lock(&tree->mutex);
417 node = tree_search(tree, file_offset);
421 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
422 atomic_inc(&entry->refs);
424 mutex_unlock(&tree->mutex);
429 * After an extent is done, call this to conditionally update the on disk
430 * i_size. i_size is updated to cover any fully written part of the file.
432 int btrfs_ordered_update_i_size(struct inode *inode,
433 struct btrfs_ordered_extent *ordered)
435 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
436 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
440 struct rb_node *node;
441 struct btrfs_ordered_extent *test;
443 mutex_lock(&tree->mutex);
444 disk_i_size = BTRFS_I(inode)->disk_i_size;
447 * if the disk i_size is already at the inode->i_size, or
448 * this ordered extent is inside the disk i_size, we're done
450 if (disk_i_size >= inode->i_size ||
451 ordered->file_offset + ordered->len <= disk_i_size) {
456 * we can't update the disk_isize if there are delalloc bytes
457 * between disk_i_size and this ordered extent
459 if (test_range_bit(io_tree, disk_i_size,
460 ordered->file_offset + ordered->len - 1,
461 EXTENT_DELALLOC, 0)) {
465 * walk backward from this ordered extent to disk_i_size.
466 * if we find an ordered extent then we can't update disk i_size
469 node = &ordered->rb_node;
471 node = rb_prev(node);
474 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
475 if (test->file_offset + test->len <= disk_i_size)
477 if (test->file_offset >= inode->i_size)
479 if (test->file_offset >= disk_i_size)
482 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
485 * at this point, we know we can safely update i_size to at least
486 * the offset from this ordered extent. But, we need to
487 * walk forward and see if ios from higher up in the file have
490 node = rb_next(&ordered->rb_node);
494 * do we have an area where IO might have finished
495 * between our ordered extent and the next one.
497 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
498 if (test->file_offset > entry_end(ordered)) {
499 i_size_test = test->file_offset - 1;
502 i_size_test = i_size_read(inode);
506 * i_size_test is the end of a region after this ordered
507 * extent where there are no ordered extents. As long as there
508 * are no delalloc bytes in this area, it is safe to update
509 * disk_i_size to the end of the region.
511 if (i_size_test > entry_end(ordered) &&
512 !test_range_bit(io_tree, entry_end(ordered), i_size_test,
513 EXTENT_DELALLOC, 0)) {
514 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
516 BTRFS_I(inode)->disk_i_size = new_i_size;
518 mutex_unlock(&tree->mutex);
523 * search the ordered extents for one corresponding to 'offset' and
524 * try to find a checksum. This is used because we allow pages to
525 * be reclaimed before their checksum is actually put into the btree
527 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u32 *sum)
529 struct btrfs_ordered_sum *ordered_sum;
530 struct btrfs_sector_sum *sector_sums;
531 struct btrfs_ordered_extent *ordered;
532 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
533 struct list_head *cur;
534 unsigned long num_sectors;
536 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
539 ordered = btrfs_lookup_ordered_extent(inode, offset);
543 mutex_lock(&tree->mutex);
544 list_for_each_prev(cur, &ordered->list) {
545 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
546 if (offset >= ordered_sum->file_offset) {
547 num_sectors = ordered_sum->len / sectorsize;
548 sector_sums = &ordered_sum->sums;
549 for (i = 0; i < num_sectors; i++) {
550 if (sector_sums[i].offset == offset) {
551 *sum = sector_sums[i].sum;
559 mutex_unlock(&tree->mutex);
565 * taken from mm/filemap.c because it isn't exported
567 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
568 * @mapping: address space structure to write
569 * @start: offset in bytes where the range starts
570 * @end: offset in bytes where the range ends (inclusive)
571 * @sync_mode: enable synchronous operation
573 * Start writeback against all of a mapping's dirty pages that lie
574 * within the byte offsets <start, end> inclusive.
576 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
577 * opposed to a regular memory cleansing writeback. The difference between
578 * these two operations is that if a dirty page/buffer is encountered, it must
579 * be waited upon, and not just skipped over.
581 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
582 loff_t end, int sync_mode)
584 struct writeback_control wbc = {
585 .sync_mode = sync_mode,
586 .nr_to_write = mapping->nrpages * 2,
587 .range_start = start,
591 return btrfs_writepages(mapping, &wbc);
595 * taken from mm/filemap.c because it isn't exported
597 * wait_on_page_writeback_range - wait for writeback to complete
598 * @mapping: target address_space
599 * @start: beginning page index
600 * @end: ending page index
602 * Wait for writeback to complete against pages indexed by start->end
605 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
606 pgoff_t start, pgoff_t end)
616 pagevec_init(&pvec, 0);
618 while ((index <= end) &&
619 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
620 PAGECACHE_TAG_WRITEBACK,
621 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
624 for (i = 0; i < nr_pages; i++) {
625 struct page *page = pvec.pages[i];
627 /* until radix tree lookup accepts end_index */
628 if (page->index > end)
631 wait_on_page_writeback(page);
635 pagevec_release(&pvec);
639 /* Check for outstanding write errors */
640 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
642 if (test_and_clear_bit(AS_EIO, &mapping->flags))