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, int nocow)
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->inode = inode;
185 set_bit(BTRFS_ORDERED_NOCOW, &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,
196 printk("warning dup entry from add_ordered_extent\n");
199 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
200 entry_end(entry) - 1, GFP_NOFS);
202 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
203 list_add_tail(&entry->root_extent_list,
204 &BTRFS_I(inode)->root->fs_info->ordered_extents);
205 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
207 mutex_unlock(&tree->mutex);
213 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
214 * when an ordered extent is finished. If the list covers more than one
215 * ordered extent, it is split across multiples.
217 int btrfs_add_ordered_sum(struct inode *inode,
218 struct btrfs_ordered_extent *entry,
219 struct btrfs_ordered_sum *sum)
221 struct btrfs_ordered_inode_tree *tree;
223 tree = &BTRFS_I(inode)->ordered_tree;
224 mutex_lock(&tree->mutex);
225 list_add_tail(&sum->list, &entry->list);
226 mutex_unlock(&tree->mutex);
231 * this is used to account for finished IO across a given range
232 * of the file. The IO should not span ordered extents. If
233 * a given ordered_extent is completely done, 1 is returned, otherwise
236 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
237 * to make sure this function only returns 1 once for a given ordered extent.
239 int btrfs_dec_test_ordered_pending(struct inode *inode,
240 u64 file_offset, u64 io_size)
242 struct btrfs_ordered_inode_tree *tree;
243 struct rb_node *node;
244 struct btrfs_ordered_extent *entry;
245 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
248 tree = &BTRFS_I(inode)->ordered_tree;
249 mutex_lock(&tree->mutex);
250 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
252 node = tree_search(tree, file_offset);
258 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
259 if (!offset_in_entry(entry, file_offset)) {
264 ret = test_range_bit(io_tree, entry->file_offset,
265 entry->file_offset + entry->len - 1,
268 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
270 mutex_unlock(&tree->mutex);
275 * used to drop a reference on an ordered extent. This will free
276 * the extent if the last reference is dropped
278 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
280 struct list_head *cur;
281 struct btrfs_ordered_sum *sum;
283 if (atomic_dec_and_test(&entry->refs)) {
284 while(!list_empty(&entry->list)) {
285 cur = entry->list.next;
286 sum = list_entry(cur, struct btrfs_ordered_sum, list);
287 list_del(&sum->list);
296 * remove an ordered extent from the tree. No references are dropped
297 * but, anyone waiting on this extent is woken up.
299 int btrfs_remove_ordered_extent(struct inode *inode,
300 struct btrfs_ordered_extent *entry)
302 struct btrfs_ordered_inode_tree *tree;
303 struct rb_node *node;
305 tree = &BTRFS_I(inode)->ordered_tree;
306 mutex_lock(&tree->mutex);
307 node = &entry->rb_node;
308 rb_erase(node, &tree->tree);
310 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
312 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
313 list_del_init(&entry->root_extent_list);
314 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
316 mutex_unlock(&tree->mutex);
317 wake_up(&entry->wait);
322 * wait for all the ordered extents in a root. This is done when balancing
323 * space between drives.
325 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
327 struct list_head splice;
328 struct list_head *cur;
329 struct btrfs_ordered_extent *ordered;
332 INIT_LIST_HEAD(&splice);
334 spin_lock(&root->fs_info->ordered_extent_lock);
335 list_splice_init(&root->fs_info->ordered_extents, &splice);
336 while (!list_empty(&splice)) {
338 ordered = list_entry(cur, struct btrfs_ordered_extent,
341 !test_bit(BTRFS_ORDERED_NOCOW, &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_NONE);
393 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
398 * Used to wait on ordered extents across a large range of bytes.
400 void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
405 struct btrfs_ordered_extent *ordered;
407 if (start + len < start) {
408 orig_end = INT_LIMIT(loff_t);
410 orig_end = start + len - 1;
411 if (orig_end > INT_LIMIT(loff_t))
412 orig_end = INT_LIMIT(loff_t);
416 /* start IO across the range first to instantiate any delalloc
419 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
421 btrfs_wait_on_page_writeback_range(inode->i_mapping,
422 start >> PAGE_CACHE_SHIFT,
423 orig_end >> PAGE_CACHE_SHIFT);
427 ordered = btrfs_lookup_first_ordered_extent(inode, end);
431 if (ordered->file_offset > orig_end) {
432 btrfs_put_ordered_extent(ordered);
435 if (ordered->file_offset + ordered->len < start) {
436 btrfs_put_ordered_extent(ordered);
439 btrfs_start_ordered_extent(inode, ordered, 1);
440 end = ordered->file_offset;
441 btrfs_put_ordered_extent(ordered);
442 if (end == 0 || end == start)
446 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
447 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
448 printk("inode %lu still ordered or delalloc after wait "
449 "%llu %llu\n", inode->i_ino,
450 (unsigned long long)start,
451 (unsigned long long)orig_end);
457 * find an ordered extent corresponding to file_offset. return NULL if
458 * nothing is found, otherwise take a reference on the extent and return it
460 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
463 struct btrfs_ordered_inode_tree *tree;
464 struct rb_node *node;
465 struct btrfs_ordered_extent *entry = NULL;
467 tree = &BTRFS_I(inode)->ordered_tree;
468 mutex_lock(&tree->mutex);
469 node = tree_search(tree, file_offset);
473 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
474 if (!offset_in_entry(entry, file_offset))
477 atomic_inc(&entry->refs);
479 mutex_unlock(&tree->mutex);
484 * lookup and return any extent before 'file_offset'. NULL is returned
487 struct btrfs_ordered_extent *
488 btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset)
490 struct btrfs_ordered_inode_tree *tree;
491 struct rb_node *node;
492 struct btrfs_ordered_extent *entry = NULL;
494 tree = &BTRFS_I(inode)->ordered_tree;
495 mutex_lock(&tree->mutex);
496 node = tree_search(tree, file_offset);
500 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
501 atomic_inc(&entry->refs);
503 mutex_unlock(&tree->mutex);
508 * After an extent is done, call this to conditionally update the on disk
509 * i_size. i_size is updated to cover any fully written part of the file.
511 int btrfs_ordered_update_i_size(struct inode *inode,
512 struct btrfs_ordered_extent *ordered)
514 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
515 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
519 struct rb_node *node;
520 struct btrfs_ordered_extent *test;
522 mutex_lock(&tree->mutex);
523 disk_i_size = BTRFS_I(inode)->disk_i_size;
526 * if the disk i_size is already at the inode->i_size, or
527 * this ordered extent is inside the disk i_size, we're done
529 if (disk_i_size >= inode->i_size ||
530 ordered->file_offset + ordered->len <= disk_i_size) {
535 * we can't update the disk_isize if there are delalloc bytes
536 * between disk_i_size and this ordered extent
538 if (test_range_bit(io_tree, disk_i_size,
539 ordered->file_offset + ordered->len - 1,
540 EXTENT_DELALLOC, 0)) {
544 * walk backward from this ordered extent to disk_i_size.
545 * if we find an ordered extent then we can't update disk i_size
548 node = &ordered->rb_node;
550 node = rb_prev(node);
553 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
554 if (test->file_offset + test->len <= disk_i_size)
556 if (test->file_offset >= inode->i_size)
558 if (test->file_offset >= disk_i_size)
561 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
564 * at this point, we know we can safely update i_size to at least
565 * the offset from this ordered extent. But, we need to
566 * walk forward and see if ios from higher up in the file have
569 node = rb_next(&ordered->rb_node);
573 * do we have an area where IO might have finished
574 * between our ordered extent and the next one.
576 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
577 if (test->file_offset > entry_end(ordered)) {
578 i_size_test = test->file_offset;
581 i_size_test = i_size_read(inode);
585 * i_size_test is the end of a region after this ordered
586 * extent where there are no ordered extents. As long as there
587 * are no delalloc bytes in this area, it is safe to update
588 * disk_i_size to the end of the region.
590 if (i_size_test > entry_end(ordered) &&
591 !test_range_bit(io_tree, entry_end(ordered), i_size_test - 1,
592 EXTENT_DELALLOC, 0)) {
593 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
595 BTRFS_I(inode)->disk_i_size = new_i_size;
597 mutex_unlock(&tree->mutex);
602 * search the ordered extents for one corresponding to 'offset' and
603 * try to find a checksum. This is used because we allow pages to
604 * be reclaimed before their checksum is actually put into the btree
606 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u32 *sum)
608 struct btrfs_ordered_sum *ordered_sum;
609 struct btrfs_sector_sum *sector_sums;
610 struct btrfs_ordered_extent *ordered;
611 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
612 struct list_head *cur;
613 unsigned long num_sectors;
615 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
618 ordered = btrfs_lookup_ordered_extent(inode, offset);
622 mutex_lock(&tree->mutex);
623 list_for_each_prev(cur, &ordered->list) {
624 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
625 if (offset >= ordered_sum->file_offset) {
626 num_sectors = ordered_sum->len / sectorsize;
627 sector_sums = ordered_sum->sums;
628 for (i = 0; i < num_sectors; i++) {
629 if (sector_sums[i].offset == offset) {
630 *sum = sector_sums[i].sum;
638 mutex_unlock(&tree->mutex);
639 btrfs_put_ordered_extent(ordered);
645 * taken from mm/filemap.c because it isn't exported
647 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
648 * @mapping: address space structure to write
649 * @start: offset in bytes where the range starts
650 * @end: offset in bytes where the range ends (inclusive)
651 * @sync_mode: enable synchronous operation
653 * Start writeback against all of a mapping's dirty pages that lie
654 * within the byte offsets <start, end> inclusive.
656 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
657 * opposed to a regular memory cleansing writeback. The difference between
658 * these two operations is that if a dirty page/buffer is encountered, it must
659 * be waited upon, and not just skipped over.
661 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
662 loff_t end, int sync_mode)
664 struct writeback_control wbc = {
665 .sync_mode = sync_mode,
666 .nr_to_write = mapping->nrpages * 2,
667 .range_start = start,
671 return btrfs_writepages(mapping, &wbc);
675 * taken from mm/filemap.c because it isn't exported
677 * wait_on_page_writeback_range - wait for writeback to complete
678 * @mapping: target address_space
679 * @start: beginning page index
680 * @end: ending page index
682 * Wait for writeback to complete against pages indexed by start->end
685 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
686 pgoff_t start, pgoff_t end)
696 pagevec_init(&pvec, 0);
698 while ((index <= end) &&
699 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
700 PAGECACHE_TAG_WRITEBACK,
701 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
704 for (i = 0; i < nr_pages; i++) {
705 struct page *page = pvec.pages[i];
707 /* until radix tree lookup accepts end_index */
708 if (page->index > end)
711 wait_on_page_writeback(page);
715 pagevec_release(&pvec);
719 /* Check for outstanding write errors */
720 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
722 if (test_and_clear_bit(AS_EIO, &mapping->flags))