Linux 4.2.1
[linux/fpc-iii.git] / fs / btrfs / ordered-data.c
blob52170cf1757e3d25192a6cf6211a9d2edaa52751
1 /*
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/slab.h>
20 #include <linux/blkdev.h>
21 #include <linux/writeback.h>
22 #include <linux/pagevec.h>
23 #include "ctree.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "extent_io.h"
27 #include "disk-io.h"
29 static struct kmem_cache *btrfs_ordered_extent_cache;
31 static u64 entry_end(struct btrfs_ordered_extent *entry)
33 if (entry->file_offset + entry->len < entry->file_offset)
34 return (u64)-1;
35 return entry->file_offset + entry->len;
38 /* returns NULL if the insertion worked, or it returns the node it did find
39 * in the tree
41 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
42 struct rb_node *node)
44 struct rb_node **p = &root->rb_node;
45 struct rb_node *parent = NULL;
46 struct btrfs_ordered_extent *entry;
48 while (*p) {
49 parent = *p;
50 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
52 if (file_offset < entry->file_offset)
53 p = &(*p)->rb_left;
54 else if (file_offset >= entry_end(entry))
55 p = &(*p)->rb_right;
56 else
57 return parent;
60 rb_link_node(node, parent, p);
61 rb_insert_color(node, root);
62 return NULL;
65 static void ordered_data_tree_panic(struct inode *inode, int errno,
66 u64 offset)
68 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
69 btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
70 "%llu", offset);
74 * look for a given offset in the tree, and if it can't be found return the
75 * first lesser offset
77 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
78 struct rb_node **prev_ret)
80 struct rb_node *n = root->rb_node;
81 struct rb_node *prev = NULL;
82 struct rb_node *test;
83 struct btrfs_ordered_extent *entry;
84 struct btrfs_ordered_extent *prev_entry = NULL;
86 while (n) {
87 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
88 prev = n;
89 prev_entry = entry;
91 if (file_offset < entry->file_offset)
92 n = n->rb_left;
93 else if (file_offset >= entry_end(entry))
94 n = n->rb_right;
95 else
96 return n;
98 if (!prev_ret)
99 return NULL;
101 while (prev && file_offset >= entry_end(prev_entry)) {
102 test = rb_next(prev);
103 if (!test)
104 break;
105 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
106 rb_node);
107 if (file_offset < entry_end(prev_entry))
108 break;
110 prev = test;
112 if (prev)
113 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
114 rb_node);
115 while (prev && file_offset < entry_end(prev_entry)) {
116 test = rb_prev(prev);
117 if (!test)
118 break;
119 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
120 rb_node);
121 prev = test;
123 *prev_ret = prev;
124 return NULL;
128 * helper to check if a given offset is inside a given entry
130 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
132 if (file_offset < entry->file_offset ||
133 entry->file_offset + entry->len <= file_offset)
134 return 0;
135 return 1;
138 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
139 u64 len)
141 if (file_offset + len <= entry->file_offset ||
142 entry->file_offset + entry->len <= file_offset)
143 return 0;
144 return 1;
148 * look find the first ordered struct that has this offset, otherwise
149 * the first one less than this offset
151 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
152 u64 file_offset)
154 struct rb_root *root = &tree->tree;
155 struct rb_node *prev = NULL;
156 struct rb_node *ret;
157 struct btrfs_ordered_extent *entry;
159 if (tree->last) {
160 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
161 rb_node);
162 if (offset_in_entry(entry, file_offset))
163 return tree->last;
165 ret = __tree_search(root, file_offset, &prev);
166 if (!ret)
167 ret = prev;
168 if (ret)
169 tree->last = ret;
170 return ret;
173 /* allocate and add a new ordered_extent into the per-inode tree.
174 * file_offset is the logical offset in the file
176 * start is the disk block number of an extent already reserved in the
177 * extent allocation tree
179 * len is the length of the extent
181 * The tree is given a single reference on the ordered extent that was
182 * inserted.
184 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
185 u64 start, u64 len, u64 disk_len,
186 int type, int dio, int compress_type)
188 struct btrfs_root *root = BTRFS_I(inode)->root;
189 struct btrfs_ordered_inode_tree *tree;
190 struct rb_node *node;
191 struct btrfs_ordered_extent *entry;
193 tree = &BTRFS_I(inode)->ordered_tree;
194 entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
195 if (!entry)
196 return -ENOMEM;
198 entry->file_offset = file_offset;
199 entry->start = start;
200 entry->len = len;
201 entry->disk_len = disk_len;
202 entry->bytes_left = len;
203 entry->inode = igrab(inode);
204 entry->compress_type = compress_type;
205 entry->truncated_len = (u64)-1;
206 if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
207 set_bit(type, &entry->flags);
209 if (dio)
210 set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
212 /* one ref for the tree */
213 atomic_set(&entry->refs, 1);
214 init_waitqueue_head(&entry->wait);
215 INIT_LIST_HEAD(&entry->list);
216 INIT_LIST_HEAD(&entry->root_extent_list);
217 INIT_LIST_HEAD(&entry->work_list);
218 init_completion(&entry->completion);
219 INIT_LIST_HEAD(&entry->log_list);
220 INIT_LIST_HEAD(&entry->trans_list);
222 trace_btrfs_ordered_extent_add(inode, entry);
224 spin_lock_irq(&tree->lock);
225 node = tree_insert(&tree->tree, file_offset,
226 &entry->rb_node);
227 if (node)
228 ordered_data_tree_panic(inode, -EEXIST, file_offset);
229 spin_unlock_irq(&tree->lock);
231 spin_lock(&root->ordered_extent_lock);
232 list_add_tail(&entry->root_extent_list,
233 &root->ordered_extents);
234 root->nr_ordered_extents++;
235 if (root->nr_ordered_extents == 1) {
236 spin_lock(&root->fs_info->ordered_root_lock);
237 BUG_ON(!list_empty(&root->ordered_root));
238 list_add_tail(&root->ordered_root,
239 &root->fs_info->ordered_roots);
240 spin_unlock(&root->fs_info->ordered_root_lock);
242 spin_unlock(&root->ordered_extent_lock);
244 return 0;
247 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
248 u64 start, u64 len, u64 disk_len, int type)
250 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
251 disk_len, type, 0,
252 BTRFS_COMPRESS_NONE);
255 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
256 u64 start, u64 len, u64 disk_len, int type)
258 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
259 disk_len, type, 1,
260 BTRFS_COMPRESS_NONE);
263 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
264 u64 start, u64 len, u64 disk_len,
265 int type, int compress_type)
267 return __btrfs_add_ordered_extent(inode, file_offset, start, len,
268 disk_len, type, 0,
269 compress_type);
273 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
274 * when an ordered extent is finished. If the list covers more than one
275 * ordered extent, it is split across multiples.
277 void btrfs_add_ordered_sum(struct inode *inode,
278 struct btrfs_ordered_extent *entry,
279 struct btrfs_ordered_sum *sum)
281 struct btrfs_ordered_inode_tree *tree;
283 tree = &BTRFS_I(inode)->ordered_tree;
284 spin_lock_irq(&tree->lock);
285 list_add_tail(&sum->list, &entry->list);
286 spin_unlock_irq(&tree->lock);
290 * this is used to account for finished IO across a given range
291 * of the file. The IO may span ordered extents. If
292 * a given ordered_extent is completely done, 1 is returned, otherwise
293 * 0.
295 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
296 * to make sure this function only returns 1 once for a given ordered extent.
298 * file_offset is updated to one byte past the range that is recorded as
299 * complete. This allows you to walk forward in the file.
301 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
302 struct btrfs_ordered_extent **cached,
303 u64 *file_offset, u64 io_size, int uptodate)
305 struct btrfs_ordered_inode_tree *tree;
306 struct rb_node *node;
307 struct btrfs_ordered_extent *entry = NULL;
308 int ret;
309 unsigned long flags;
310 u64 dec_end;
311 u64 dec_start;
312 u64 to_dec;
314 tree = &BTRFS_I(inode)->ordered_tree;
315 spin_lock_irqsave(&tree->lock, flags);
316 node = tree_search(tree, *file_offset);
317 if (!node) {
318 ret = 1;
319 goto out;
322 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
323 if (!offset_in_entry(entry, *file_offset)) {
324 ret = 1;
325 goto out;
328 dec_start = max(*file_offset, entry->file_offset);
329 dec_end = min(*file_offset + io_size, entry->file_offset +
330 entry->len);
331 *file_offset = dec_end;
332 if (dec_start > dec_end) {
333 btrfs_crit(BTRFS_I(inode)->root->fs_info,
334 "bad ordering dec_start %llu end %llu", dec_start, dec_end);
336 to_dec = dec_end - dec_start;
337 if (to_dec > entry->bytes_left) {
338 btrfs_crit(BTRFS_I(inode)->root->fs_info,
339 "bad ordered accounting left %llu size %llu",
340 entry->bytes_left, to_dec);
342 entry->bytes_left -= to_dec;
343 if (!uptodate)
344 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
346 if (entry->bytes_left == 0) {
347 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
348 if (waitqueue_active(&entry->wait))
349 wake_up(&entry->wait);
350 } else {
351 ret = 1;
353 out:
354 if (!ret && cached && entry) {
355 *cached = entry;
356 atomic_inc(&entry->refs);
358 spin_unlock_irqrestore(&tree->lock, flags);
359 return ret == 0;
363 * this is used to account for finished IO across a given range
364 * of the file. The IO should not span ordered extents. If
365 * a given ordered_extent is completely done, 1 is returned, otherwise
366 * 0.
368 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
369 * to make sure this function only returns 1 once for a given ordered extent.
371 int btrfs_dec_test_ordered_pending(struct inode *inode,
372 struct btrfs_ordered_extent **cached,
373 u64 file_offset, u64 io_size, int uptodate)
375 struct btrfs_ordered_inode_tree *tree;
376 struct rb_node *node;
377 struct btrfs_ordered_extent *entry = NULL;
378 unsigned long flags;
379 int ret;
381 tree = &BTRFS_I(inode)->ordered_tree;
382 spin_lock_irqsave(&tree->lock, flags);
383 if (cached && *cached) {
384 entry = *cached;
385 goto have_entry;
388 node = tree_search(tree, file_offset);
389 if (!node) {
390 ret = 1;
391 goto out;
394 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
395 have_entry:
396 if (!offset_in_entry(entry, file_offset)) {
397 ret = 1;
398 goto out;
401 if (io_size > entry->bytes_left) {
402 btrfs_crit(BTRFS_I(inode)->root->fs_info,
403 "bad ordered accounting left %llu size %llu",
404 entry->bytes_left, io_size);
406 entry->bytes_left -= io_size;
407 if (!uptodate)
408 set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
410 if (entry->bytes_left == 0) {
411 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
412 if (waitqueue_active(&entry->wait))
413 wake_up(&entry->wait);
414 } else {
415 ret = 1;
417 out:
418 if (!ret && cached && entry) {
419 *cached = entry;
420 atomic_inc(&entry->refs);
422 spin_unlock_irqrestore(&tree->lock, flags);
423 return ret == 0;
426 /* Needs to either be called under a log transaction or the log_mutex */
427 void btrfs_get_logged_extents(struct inode *inode,
428 struct list_head *logged_list,
429 const loff_t start,
430 const loff_t end)
432 struct btrfs_ordered_inode_tree *tree;
433 struct btrfs_ordered_extent *ordered;
434 struct rb_node *n;
435 struct rb_node *prev;
437 tree = &BTRFS_I(inode)->ordered_tree;
438 spin_lock_irq(&tree->lock);
439 n = __tree_search(&tree->tree, end, &prev);
440 if (!n)
441 n = prev;
442 for (; n; n = rb_prev(n)) {
443 ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
444 if (ordered->file_offset > end)
445 continue;
446 if (entry_end(ordered) <= start)
447 break;
448 if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
449 continue;
450 list_add(&ordered->log_list, logged_list);
451 atomic_inc(&ordered->refs);
453 spin_unlock_irq(&tree->lock);
456 void btrfs_put_logged_extents(struct list_head *logged_list)
458 struct btrfs_ordered_extent *ordered;
460 while (!list_empty(logged_list)) {
461 ordered = list_first_entry(logged_list,
462 struct btrfs_ordered_extent,
463 log_list);
464 list_del_init(&ordered->log_list);
465 btrfs_put_ordered_extent(ordered);
469 void btrfs_submit_logged_extents(struct list_head *logged_list,
470 struct btrfs_root *log)
472 int index = log->log_transid % 2;
474 spin_lock_irq(&log->log_extents_lock[index]);
475 list_splice_tail(logged_list, &log->logged_list[index]);
476 spin_unlock_irq(&log->log_extents_lock[index]);
479 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
480 struct btrfs_root *log, u64 transid)
482 struct btrfs_ordered_extent *ordered;
483 int index = transid % 2;
485 spin_lock_irq(&log->log_extents_lock[index]);
486 while (!list_empty(&log->logged_list[index])) {
487 ordered = list_first_entry(&log->logged_list[index],
488 struct btrfs_ordered_extent,
489 log_list);
490 list_del_init(&ordered->log_list);
491 spin_unlock_irq(&log->log_extents_lock[index]);
493 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
494 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
495 struct inode *inode = ordered->inode;
496 u64 start = ordered->file_offset;
497 u64 end = ordered->file_offset + ordered->len - 1;
499 WARN_ON(!inode);
500 filemap_fdatawrite_range(inode->i_mapping, start, end);
502 wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
503 &ordered->flags));
506 * If our ordered extent completed it means it updated the
507 * fs/subvol and csum trees already, so no need to make the
508 * current transaction's commit wait for it, as we end up
509 * holding memory unnecessarily and delaying the inode's iput
510 * until the transaction commit (we schedule an iput for the
511 * inode when the ordered extent's refcount drops to 0), which
512 * prevents it from being evictable until the transaction
513 * commits.
515 if (test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags))
516 btrfs_put_ordered_extent(ordered);
517 else
518 list_add_tail(&ordered->trans_list, &trans->ordered);
520 spin_lock_irq(&log->log_extents_lock[index]);
522 spin_unlock_irq(&log->log_extents_lock[index]);
525 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
527 struct btrfs_ordered_extent *ordered;
528 int index = transid % 2;
530 spin_lock_irq(&log->log_extents_lock[index]);
531 while (!list_empty(&log->logged_list[index])) {
532 ordered = list_first_entry(&log->logged_list[index],
533 struct btrfs_ordered_extent,
534 log_list);
535 list_del_init(&ordered->log_list);
536 spin_unlock_irq(&log->log_extents_lock[index]);
537 btrfs_put_ordered_extent(ordered);
538 spin_lock_irq(&log->log_extents_lock[index]);
540 spin_unlock_irq(&log->log_extents_lock[index]);
544 * used to drop a reference on an ordered extent. This will free
545 * the extent if the last reference is dropped
547 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
549 struct list_head *cur;
550 struct btrfs_ordered_sum *sum;
552 trace_btrfs_ordered_extent_put(entry->inode, entry);
554 if (atomic_dec_and_test(&entry->refs)) {
555 ASSERT(list_empty(&entry->log_list));
556 ASSERT(list_empty(&entry->trans_list));
557 ASSERT(list_empty(&entry->root_extent_list));
558 ASSERT(RB_EMPTY_NODE(&entry->rb_node));
559 if (entry->inode)
560 btrfs_add_delayed_iput(entry->inode);
561 while (!list_empty(&entry->list)) {
562 cur = entry->list.next;
563 sum = list_entry(cur, struct btrfs_ordered_sum, list);
564 list_del(&sum->list);
565 kfree(sum);
567 kmem_cache_free(btrfs_ordered_extent_cache, entry);
572 * remove an ordered extent from the tree. No references are dropped
573 * and waiters are woken up.
575 void btrfs_remove_ordered_extent(struct inode *inode,
576 struct btrfs_ordered_extent *entry)
578 struct btrfs_ordered_inode_tree *tree;
579 struct btrfs_root *root = BTRFS_I(inode)->root;
580 struct rb_node *node;
582 tree = &BTRFS_I(inode)->ordered_tree;
583 spin_lock_irq(&tree->lock);
584 node = &entry->rb_node;
585 rb_erase(node, &tree->tree);
586 RB_CLEAR_NODE(node);
587 if (tree->last == node)
588 tree->last = NULL;
589 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
590 spin_unlock_irq(&tree->lock);
592 spin_lock(&root->ordered_extent_lock);
593 list_del_init(&entry->root_extent_list);
594 root->nr_ordered_extents--;
596 trace_btrfs_ordered_extent_remove(inode, entry);
598 if (!root->nr_ordered_extents) {
599 spin_lock(&root->fs_info->ordered_root_lock);
600 BUG_ON(list_empty(&root->ordered_root));
601 list_del_init(&root->ordered_root);
602 spin_unlock(&root->fs_info->ordered_root_lock);
604 spin_unlock(&root->ordered_extent_lock);
605 wake_up(&entry->wait);
608 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
610 struct btrfs_ordered_extent *ordered;
612 ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
613 btrfs_start_ordered_extent(ordered->inode, ordered, 1);
614 complete(&ordered->completion);
618 * wait for all the ordered extents in a root. This is done when balancing
619 * space between drives.
621 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
623 struct list_head splice, works;
624 struct btrfs_ordered_extent *ordered, *next;
625 int count = 0;
627 INIT_LIST_HEAD(&splice);
628 INIT_LIST_HEAD(&works);
630 mutex_lock(&root->ordered_extent_mutex);
631 spin_lock(&root->ordered_extent_lock);
632 list_splice_init(&root->ordered_extents, &splice);
633 while (!list_empty(&splice) && nr) {
634 ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
635 root_extent_list);
636 list_move_tail(&ordered->root_extent_list,
637 &root->ordered_extents);
638 atomic_inc(&ordered->refs);
639 spin_unlock(&root->ordered_extent_lock);
641 btrfs_init_work(&ordered->flush_work,
642 btrfs_flush_delalloc_helper,
643 btrfs_run_ordered_extent_work, NULL, NULL);
644 list_add_tail(&ordered->work_list, &works);
645 btrfs_queue_work(root->fs_info->flush_workers,
646 &ordered->flush_work);
648 cond_resched();
649 spin_lock(&root->ordered_extent_lock);
650 if (nr != -1)
651 nr--;
652 count++;
654 list_splice_tail(&splice, &root->ordered_extents);
655 spin_unlock(&root->ordered_extent_lock);
657 list_for_each_entry_safe(ordered, next, &works, work_list) {
658 list_del_init(&ordered->work_list);
659 wait_for_completion(&ordered->completion);
660 btrfs_put_ordered_extent(ordered);
661 cond_resched();
663 mutex_unlock(&root->ordered_extent_mutex);
665 return count;
668 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
670 struct btrfs_root *root;
671 struct list_head splice;
672 int done;
674 INIT_LIST_HEAD(&splice);
676 mutex_lock(&fs_info->ordered_operations_mutex);
677 spin_lock(&fs_info->ordered_root_lock);
678 list_splice_init(&fs_info->ordered_roots, &splice);
679 while (!list_empty(&splice) && nr) {
680 root = list_first_entry(&splice, struct btrfs_root,
681 ordered_root);
682 root = btrfs_grab_fs_root(root);
683 BUG_ON(!root);
684 list_move_tail(&root->ordered_root,
685 &fs_info->ordered_roots);
686 spin_unlock(&fs_info->ordered_root_lock);
688 done = btrfs_wait_ordered_extents(root, nr);
689 btrfs_put_fs_root(root);
691 spin_lock(&fs_info->ordered_root_lock);
692 if (nr != -1) {
693 nr -= done;
694 WARN_ON(nr < 0);
697 list_splice_tail(&splice, &fs_info->ordered_roots);
698 spin_unlock(&fs_info->ordered_root_lock);
699 mutex_unlock(&fs_info->ordered_operations_mutex);
703 * Used to start IO or wait for a given ordered extent to finish.
705 * If wait is one, this effectively waits on page writeback for all the pages
706 * in the extent, and it waits on the io completion code to insert
707 * metadata into the btree corresponding to the extent
709 void btrfs_start_ordered_extent(struct inode *inode,
710 struct btrfs_ordered_extent *entry,
711 int wait)
713 u64 start = entry->file_offset;
714 u64 end = start + entry->len - 1;
716 trace_btrfs_ordered_extent_start(inode, entry);
719 * pages in the range can be dirty, clean or writeback. We
720 * start IO on any dirty ones so the wait doesn't stall waiting
721 * for the flusher thread to find them
723 if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
724 filemap_fdatawrite_range(inode->i_mapping, start, end);
725 if (wait) {
726 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
727 &entry->flags));
732 * Used to wait on ordered extents across a large range of bytes.
734 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
736 int ret = 0;
737 int ret_wb = 0;
738 u64 end;
739 u64 orig_end;
740 struct btrfs_ordered_extent *ordered;
742 if (start + len < start) {
743 orig_end = INT_LIMIT(loff_t);
744 } else {
745 orig_end = start + len - 1;
746 if (orig_end > INT_LIMIT(loff_t))
747 orig_end = INT_LIMIT(loff_t);
750 /* start IO across the range first to instantiate any delalloc
751 * extents
753 ret = btrfs_fdatawrite_range(inode, start, orig_end);
754 if (ret)
755 return ret;
758 * If we have a writeback error don't return immediately. Wait first
759 * for any ordered extents that haven't completed yet. This is to make
760 * sure no one can dirty the same page ranges and call writepages()
761 * before the ordered extents complete - to avoid failures (-EEXIST)
762 * when adding the new ordered extents to the ordered tree.
764 ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
766 end = orig_end;
767 while (1) {
768 ordered = btrfs_lookup_first_ordered_extent(inode, end);
769 if (!ordered)
770 break;
771 if (ordered->file_offset > orig_end) {
772 btrfs_put_ordered_extent(ordered);
773 break;
775 if (ordered->file_offset + ordered->len <= start) {
776 btrfs_put_ordered_extent(ordered);
777 break;
779 btrfs_start_ordered_extent(inode, ordered, 1);
780 end = ordered->file_offset;
781 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
782 ret = -EIO;
783 btrfs_put_ordered_extent(ordered);
784 if (ret || end == 0 || end == start)
785 break;
786 end--;
788 return ret_wb ? ret_wb : ret;
792 * find an ordered extent corresponding to file_offset. return NULL if
793 * nothing is found, otherwise take a reference on the extent and return it
795 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
796 u64 file_offset)
798 struct btrfs_ordered_inode_tree *tree;
799 struct rb_node *node;
800 struct btrfs_ordered_extent *entry = NULL;
802 tree = &BTRFS_I(inode)->ordered_tree;
803 spin_lock_irq(&tree->lock);
804 node = tree_search(tree, file_offset);
805 if (!node)
806 goto out;
808 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
809 if (!offset_in_entry(entry, file_offset))
810 entry = NULL;
811 if (entry)
812 atomic_inc(&entry->refs);
813 out:
814 spin_unlock_irq(&tree->lock);
815 return entry;
818 /* Since the DIO code tries to lock a wide area we need to look for any ordered
819 * extents that exist in the range, rather than just the start of the range.
821 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
822 u64 file_offset,
823 u64 len)
825 struct btrfs_ordered_inode_tree *tree;
826 struct rb_node *node;
827 struct btrfs_ordered_extent *entry = NULL;
829 tree = &BTRFS_I(inode)->ordered_tree;
830 spin_lock_irq(&tree->lock);
831 node = tree_search(tree, file_offset);
832 if (!node) {
833 node = tree_search(tree, file_offset + len);
834 if (!node)
835 goto out;
838 while (1) {
839 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
840 if (range_overlaps(entry, file_offset, len))
841 break;
843 if (entry->file_offset >= file_offset + len) {
844 entry = NULL;
845 break;
847 entry = NULL;
848 node = rb_next(node);
849 if (!node)
850 break;
852 out:
853 if (entry)
854 atomic_inc(&entry->refs);
855 spin_unlock_irq(&tree->lock);
856 return entry;
859 bool btrfs_have_ordered_extents_in_range(struct inode *inode,
860 u64 file_offset,
861 u64 len)
863 struct btrfs_ordered_extent *oe;
865 oe = btrfs_lookup_ordered_range(inode, file_offset, len);
866 if (oe) {
867 btrfs_put_ordered_extent(oe);
868 return true;
870 return false;
874 * lookup and return any extent before 'file_offset'. NULL is returned
875 * if none is found
877 struct btrfs_ordered_extent *
878 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
880 struct btrfs_ordered_inode_tree *tree;
881 struct rb_node *node;
882 struct btrfs_ordered_extent *entry = NULL;
884 tree = &BTRFS_I(inode)->ordered_tree;
885 spin_lock_irq(&tree->lock);
886 node = tree_search(tree, file_offset);
887 if (!node)
888 goto out;
890 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
891 atomic_inc(&entry->refs);
892 out:
893 spin_unlock_irq(&tree->lock);
894 return entry;
898 * After an extent is done, call this to conditionally update the on disk
899 * i_size. i_size is updated to cover any fully written part of the file.
901 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
902 struct btrfs_ordered_extent *ordered)
904 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
905 u64 disk_i_size;
906 u64 new_i_size;
907 u64 i_size = i_size_read(inode);
908 struct rb_node *node;
909 struct rb_node *prev = NULL;
910 struct btrfs_ordered_extent *test;
911 int ret = 1;
913 spin_lock_irq(&tree->lock);
914 if (ordered) {
915 offset = entry_end(ordered);
916 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
917 offset = min(offset,
918 ordered->file_offset +
919 ordered->truncated_len);
920 } else {
921 offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
923 disk_i_size = BTRFS_I(inode)->disk_i_size;
925 /* truncate file */
926 if (disk_i_size > i_size) {
927 BTRFS_I(inode)->disk_i_size = i_size;
928 ret = 0;
929 goto out;
933 * if the disk i_size is already at the inode->i_size, or
934 * this ordered extent is inside the disk i_size, we're done
936 if (disk_i_size == i_size)
937 goto out;
940 * We still need to update disk_i_size if outstanding_isize is greater
941 * than disk_i_size.
943 if (offset <= disk_i_size &&
944 (!ordered || ordered->outstanding_isize <= disk_i_size))
945 goto out;
948 * walk backward from this ordered extent to disk_i_size.
949 * if we find an ordered extent then we can't update disk i_size
950 * yet
952 if (ordered) {
953 node = rb_prev(&ordered->rb_node);
954 } else {
955 prev = tree_search(tree, offset);
957 * we insert file extents without involving ordered struct,
958 * so there should be no ordered struct cover this offset
960 if (prev) {
961 test = rb_entry(prev, struct btrfs_ordered_extent,
962 rb_node);
963 BUG_ON(offset_in_entry(test, offset));
965 node = prev;
967 for (; node; node = rb_prev(node)) {
968 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
970 /* We treat this entry as if it doesnt exist */
971 if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
972 continue;
973 if (test->file_offset + test->len <= disk_i_size)
974 break;
975 if (test->file_offset >= i_size)
976 break;
977 if (entry_end(test) > disk_i_size) {
979 * we don't update disk_i_size now, so record this
980 * undealt i_size. Or we will not know the real
981 * i_size.
983 if (test->outstanding_isize < offset)
984 test->outstanding_isize = offset;
985 if (ordered &&
986 ordered->outstanding_isize >
987 test->outstanding_isize)
988 test->outstanding_isize =
989 ordered->outstanding_isize;
990 goto out;
993 new_i_size = min_t(u64, offset, i_size);
996 * Some ordered extents may completed before the current one, and
997 * we hold the real i_size in ->outstanding_isize.
999 if (ordered && ordered->outstanding_isize > new_i_size)
1000 new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1001 BTRFS_I(inode)->disk_i_size = new_i_size;
1002 ret = 0;
1003 out:
1005 * We need to do this because we can't remove ordered extents until
1006 * after the i_disk_size has been updated and then the inode has been
1007 * updated to reflect the change, so we need to tell anybody who finds
1008 * this ordered extent that we've already done all the real work, we
1009 * just haven't completed all the other work.
1011 if (ordered)
1012 set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1013 spin_unlock_irq(&tree->lock);
1014 return ret;
1018 * search the ordered extents for one corresponding to 'offset' and
1019 * try to find a checksum. This is used because we allow pages to
1020 * be reclaimed before their checksum is actually put into the btree
1022 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1023 u32 *sum, int len)
1025 struct btrfs_ordered_sum *ordered_sum;
1026 struct btrfs_ordered_extent *ordered;
1027 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1028 unsigned long num_sectors;
1029 unsigned long i;
1030 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1031 int index = 0;
1033 ordered = btrfs_lookup_ordered_extent(inode, offset);
1034 if (!ordered)
1035 return 0;
1037 spin_lock_irq(&tree->lock);
1038 list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1039 if (disk_bytenr >= ordered_sum->bytenr &&
1040 disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1041 i = (disk_bytenr - ordered_sum->bytenr) >>
1042 inode->i_sb->s_blocksize_bits;
1043 num_sectors = ordered_sum->len >>
1044 inode->i_sb->s_blocksize_bits;
1045 num_sectors = min_t(int, len - index, num_sectors - i);
1046 memcpy(sum + index, ordered_sum->sums + i,
1047 num_sectors);
1049 index += (int)num_sectors;
1050 if (index == len)
1051 goto out;
1052 disk_bytenr += num_sectors * sectorsize;
1055 out:
1056 spin_unlock_irq(&tree->lock);
1057 btrfs_put_ordered_extent(ordered);
1058 return index;
1061 int __init ordered_data_init(void)
1063 btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1064 sizeof(struct btrfs_ordered_extent), 0,
1065 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1066 NULL);
1067 if (!btrfs_ordered_extent_cache)
1068 return -ENOMEM;
1070 return 0;
1073 void ordered_data_exit(void)
1075 if (btrfs_ordered_extent_cache)
1076 kmem_cache_destroy(btrfs_ordered_extent_cache);