xref: /openbmc/linux/fs/btrfs/ordered-data.c (revision 4f3db074)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
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.
7  *
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.
12  *
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.
17  */
18 
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"
28 
29 static struct kmem_cache *btrfs_ordered_extent_cache;
30 
31 static u64 entry_end(struct btrfs_ordered_extent *entry)
32 {
33 	if (entry->file_offset + entry->len < entry->file_offset)
34 		return (u64)-1;
35 	return entry->file_offset + entry->len;
36 }
37 
38 /* returns NULL if the insertion worked, or it returns the node it did find
39  * in the tree
40  */
41 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
42 				   struct rb_node *node)
43 {
44 	struct rb_node **p = &root->rb_node;
45 	struct rb_node *parent = NULL;
46 	struct btrfs_ordered_extent *entry;
47 
48 	while (*p) {
49 		parent = *p;
50 		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
51 
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;
58 	}
59 
60 	rb_link_node(node, parent, p);
61 	rb_insert_color(node, root);
62 	return NULL;
63 }
64 
65 static void ordered_data_tree_panic(struct inode *inode, int errno,
66 					       u64 offset)
67 {
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);
71 }
72 
73 /*
74  * look for a given offset in the tree, and if it can't be found return the
75  * first lesser offset
76  */
77 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
78 				     struct rb_node **prev_ret)
79 {
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;
85 
86 	while (n) {
87 		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
88 		prev = n;
89 		prev_entry = entry;
90 
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;
97 	}
98 	if (!prev_ret)
99 		return NULL;
100 
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;
109 
110 		prev = test;
111 	}
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;
122 	}
123 	*prev_ret = prev;
124 	return NULL;
125 }
126 
127 /*
128  * helper to check if a given offset is inside a given entry
129  */
130 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
131 {
132 	if (file_offset < entry->file_offset ||
133 	    entry->file_offset + entry->len <= file_offset)
134 		return 0;
135 	return 1;
136 }
137 
138 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
139 			  u64 len)
140 {
141 	if (file_offset + len <= entry->file_offset ||
142 	    entry->file_offset + entry->len <= file_offset)
143 		return 0;
144 	return 1;
145 }
146 
147 /*
148  * look find the first ordered struct that has this offset, otherwise
149  * the first one less than this offset
150  */
151 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
152 					  u64 file_offset)
153 {
154 	struct rb_root *root = &tree->tree;
155 	struct rb_node *prev = NULL;
156 	struct rb_node *ret;
157 	struct btrfs_ordered_extent *entry;
158 
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;
164 	}
165 	ret = __tree_search(root, file_offset, &prev);
166 	if (!ret)
167 		ret = prev;
168 	if (ret)
169 		tree->last = ret;
170 	return ret;
171 }
172 
173 /* allocate and add a new ordered_extent into the per-inode tree.
174  * file_offset is the logical offset in the file
175  *
176  * start is the disk block number of an extent already reserved in the
177  * extent allocation tree
178  *
179  * len is the length of the extent
180  *
181  * The tree is given a single reference on the ordered extent that was
182  * inserted.
183  */
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)
187 {
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;
192 
193 	tree = &BTRFS_I(inode)->ordered_tree;
194 	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
195 	if (!entry)
196 		return -ENOMEM;
197 
198 	entry->file_offset = file_offset;
199 	entry->start = start;
200 	entry->len = len;
201 	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) &&
202 	    !(type == BTRFS_ORDERED_NOCOW))
203 		entry->csum_bytes_left = disk_len;
204 	entry->disk_len = disk_len;
205 	entry->bytes_left = len;
206 	entry->inode = igrab(inode);
207 	entry->compress_type = compress_type;
208 	entry->truncated_len = (u64)-1;
209 	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
210 		set_bit(type, &entry->flags);
211 
212 	if (dio)
213 		set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
214 
215 	/* one ref for the tree */
216 	atomic_set(&entry->refs, 1);
217 	init_waitqueue_head(&entry->wait);
218 	INIT_LIST_HEAD(&entry->list);
219 	INIT_LIST_HEAD(&entry->root_extent_list);
220 	INIT_LIST_HEAD(&entry->work_list);
221 	init_completion(&entry->completion);
222 	INIT_LIST_HEAD(&entry->log_list);
223 	INIT_LIST_HEAD(&entry->trans_list);
224 
225 	trace_btrfs_ordered_extent_add(inode, entry);
226 
227 	spin_lock_irq(&tree->lock);
228 	node = tree_insert(&tree->tree, file_offset,
229 			   &entry->rb_node);
230 	if (node)
231 		ordered_data_tree_panic(inode, -EEXIST, file_offset);
232 	spin_unlock_irq(&tree->lock);
233 
234 	spin_lock(&root->ordered_extent_lock);
235 	list_add_tail(&entry->root_extent_list,
236 		      &root->ordered_extents);
237 	root->nr_ordered_extents++;
238 	if (root->nr_ordered_extents == 1) {
239 		spin_lock(&root->fs_info->ordered_root_lock);
240 		BUG_ON(!list_empty(&root->ordered_root));
241 		list_add_tail(&root->ordered_root,
242 			      &root->fs_info->ordered_roots);
243 		spin_unlock(&root->fs_info->ordered_root_lock);
244 	}
245 	spin_unlock(&root->ordered_extent_lock);
246 
247 	return 0;
248 }
249 
250 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
251 			     u64 start, u64 len, u64 disk_len, int type)
252 {
253 	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
254 					  disk_len, type, 0,
255 					  BTRFS_COMPRESS_NONE);
256 }
257 
258 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
259 				 u64 start, u64 len, u64 disk_len, int type)
260 {
261 	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
262 					  disk_len, type, 1,
263 					  BTRFS_COMPRESS_NONE);
264 }
265 
266 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
267 				      u64 start, u64 len, u64 disk_len,
268 				      int type, int compress_type)
269 {
270 	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
271 					  disk_len, type, 0,
272 					  compress_type);
273 }
274 
275 /*
276  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
277  * when an ordered extent is finished.  If the list covers more than one
278  * ordered extent, it is split across multiples.
279  */
280 void btrfs_add_ordered_sum(struct inode *inode,
281 			   struct btrfs_ordered_extent *entry,
282 			   struct btrfs_ordered_sum *sum)
283 {
284 	struct btrfs_ordered_inode_tree *tree;
285 
286 	tree = &BTRFS_I(inode)->ordered_tree;
287 	spin_lock_irq(&tree->lock);
288 	list_add_tail(&sum->list, &entry->list);
289 	WARN_ON(entry->csum_bytes_left < sum->len);
290 	entry->csum_bytes_left -= sum->len;
291 	if (entry->csum_bytes_left == 0)
292 		wake_up(&entry->wait);
293 	spin_unlock_irq(&tree->lock);
294 }
295 
296 /*
297  * this is used to account for finished IO across a given range
298  * of the file.  The IO may span ordered extents.  If
299  * a given ordered_extent is completely done, 1 is returned, otherwise
300  * 0.
301  *
302  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
303  * to make sure this function only returns 1 once for a given ordered extent.
304  *
305  * file_offset is updated to one byte past the range that is recorded as
306  * complete.  This allows you to walk forward in the file.
307  */
308 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
309 				   struct btrfs_ordered_extent **cached,
310 				   u64 *file_offset, u64 io_size, int uptodate)
311 {
312 	struct btrfs_ordered_inode_tree *tree;
313 	struct rb_node *node;
314 	struct btrfs_ordered_extent *entry = NULL;
315 	int ret;
316 	unsigned long flags;
317 	u64 dec_end;
318 	u64 dec_start;
319 	u64 to_dec;
320 
321 	tree = &BTRFS_I(inode)->ordered_tree;
322 	spin_lock_irqsave(&tree->lock, flags);
323 	node = tree_search(tree, *file_offset);
324 	if (!node) {
325 		ret = 1;
326 		goto out;
327 	}
328 
329 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
330 	if (!offset_in_entry(entry, *file_offset)) {
331 		ret = 1;
332 		goto out;
333 	}
334 
335 	dec_start = max(*file_offset, entry->file_offset);
336 	dec_end = min(*file_offset + io_size, entry->file_offset +
337 		      entry->len);
338 	*file_offset = dec_end;
339 	if (dec_start > dec_end) {
340 		btrfs_crit(BTRFS_I(inode)->root->fs_info,
341 			"bad ordering dec_start %llu end %llu", dec_start, dec_end);
342 	}
343 	to_dec = dec_end - dec_start;
344 	if (to_dec > entry->bytes_left) {
345 		btrfs_crit(BTRFS_I(inode)->root->fs_info,
346 			"bad ordered accounting left %llu size %llu",
347 			entry->bytes_left, to_dec);
348 	}
349 	entry->bytes_left -= to_dec;
350 	if (!uptodate)
351 		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
352 
353 	if (entry->bytes_left == 0) {
354 		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
355 		if (waitqueue_active(&entry->wait))
356 			wake_up(&entry->wait);
357 	} else {
358 		ret = 1;
359 	}
360 out:
361 	if (!ret && cached && entry) {
362 		*cached = entry;
363 		atomic_inc(&entry->refs);
364 	}
365 	spin_unlock_irqrestore(&tree->lock, flags);
366 	return ret == 0;
367 }
368 
369 /*
370  * this is used to account for finished IO across a given range
371  * of the file.  The IO should not span ordered extents.  If
372  * a given ordered_extent is completely done, 1 is returned, otherwise
373  * 0.
374  *
375  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
376  * to make sure this function only returns 1 once for a given ordered extent.
377  */
378 int btrfs_dec_test_ordered_pending(struct inode *inode,
379 				   struct btrfs_ordered_extent **cached,
380 				   u64 file_offset, u64 io_size, int uptodate)
381 {
382 	struct btrfs_ordered_inode_tree *tree;
383 	struct rb_node *node;
384 	struct btrfs_ordered_extent *entry = NULL;
385 	unsigned long flags;
386 	int ret;
387 
388 	tree = &BTRFS_I(inode)->ordered_tree;
389 	spin_lock_irqsave(&tree->lock, flags);
390 	if (cached && *cached) {
391 		entry = *cached;
392 		goto have_entry;
393 	}
394 
395 	node = tree_search(tree, file_offset);
396 	if (!node) {
397 		ret = 1;
398 		goto out;
399 	}
400 
401 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
402 have_entry:
403 	if (!offset_in_entry(entry, file_offset)) {
404 		ret = 1;
405 		goto out;
406 	}
407 
408 	if (io_size > entry->bytes_left) {
409 		btrfs_crit(BTRFS_I(inode)->root->fs_info,
410 			   "bad ordered accounting left %llu size %llu",
411 		       entry->bytes_left, io_size);
412 	}
413 	entry->bytes_left -= io_size;
414 	if (!uptodate)
415 		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
416 
417 	if (entry->bytes_left == 0) {
418 		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
419 		if (waitqueue_active(&entry->wait))
420 			wake_up(&entry->wait);
421 	} else {
422 		ret = 1;
423 	}
424 out:
425 	if (!ret && cached && entry) {
426 		*cached = entry;
427 		atomic_inc(&entry->refs);
428 	}
429 	spin_unlock_irqrestore(&tree->lock, flags);
430 	return ret == 0;
431 }
432 
433 /* Needs to either be called under a log transaction or the log_mutex */
434 void btrfs_get_logged_extents(struct inode *inode,
435 			      struct list_head *logged_list,
436 			      const loff_t start,
437 			      const loff_t end)
438 {
439 	struct btrfs_ordered_inode_tree *tree;
440 	struct btrfs_ordered_extent *ordered;
441 	struct rb_node *n;
442 	struct rb_node *prev;
443 
444 	tree = &BTRFS_I(inode)->ordered_tree;
445 	spin_lock_irq(&tree->lock);
446 	n = __tree_search(&tree->tree, end, &prev);
447 	if (!n)
448 		n = prev;
449 	for (; n; n = rb_prev(n)) {
450 		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
451 		if (ordered->file_offset > end)
452 			continue;
453 		if (entry_end(ordered) <= start)
454 			break;
455 		if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
456 			continue;
457 		list_add(&ordered->log_list, logged_list);
458 		atomic_inc(&ordered->refs);
459 	}
460 	spin_unlock_irq(&tree->lock);
461 }
462 
463 void btrfs_put_logged_extents(struct list_head *logged_list)
464 {
465 	struct btrfs_ordered_extent *ordered;
466 
467 	while (!list_empty(logged_list)) {
468 		ordered = list_first_entry(logged_list,
469 					   struct btrfs_ordered_extent,
470 					   log_list);
471 		list_del_init(&ordered->log_list);
472 		btrfs_put_ordered_extent(ordered);
473 	}
474 }
475 
476 void btrfs_submit_logged_extents(struct list_head *logged_list,
477 				 struct btrfs_root *log)
478 {
479 	int index = log->log_transid % 2;
480 
481 	spin_lock_irq(&log->log_extents_lock[index]);
482 	list_splice_tail(logged_list, &log->logged_list[index]);
483 	spin_unlock_irq(&log->log_extents_lock[index]);
484 }
485 
486 void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans,
487 			       struct btrfs_root *log, u64 transid)
488 {
489 	struct btrfs_ordered_extent *ordered;
490 	int index = transid % 2;
491 
492 	spin_lock_irq(&log->log_extents_lock[index]);
493 	while (!list_empty(&log->logged_list[index])) {
494 		ordered = list_first_entry(&log->logged_list[index],
495 					   struct btrfs_ordered_extent,
496 					   log_list);
497 		list_del_init(&ordered->log_list);
498 		spin_unlock_irq(&log->log_extents_lock[index]);
499 
500 		if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
501 		    !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
502 			struct inode *inode = ordered->inode;
503 			u64 start = ordered->file_offset;
504 			u64 end = ordered->file_offset + ordered->len - 1;
505 
506 			WARN_ON(!inode);
507 			filemap_fdatawrite_range(inode->i_mapping, start, end);
508 		}
509 		wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
510 						   &ordered->flags));
511 
512 		list_add_tail(&ordered->trans_list, &trans->ordered);
513 		spin_lock_irq(&log->log_extents_lock[index]);
514 	}
515 	spin_unlock_irq(&log->log_extents_lock[index]);
516 }
517 
518 void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
519 {
520 	struct btrfs_ordered_extent *ordered;
521 	int index = transid % 2;
522 
523 	spin_lock_irq(&log->log_extents_lock[index]);
524 	while (!list_empty(&log->logged_list[index])) {
525 		ordered = list_first_entry(&log->logged_list[index],
526 					   struct btrfs_ordered_extent,
527 					   log_list);
528 		list_del_init(&ordered->log_list);
529 		spin_unlock_irq(&log->log_extents_lock[index]);
530 		btrfs_put_ordered_extent(ordered);
531 		spin_lock_irq(&log->log_extents_lock[index]);
532 	}
533 	spin_unlock_irq(&log->log_extents_lock[index]);
534 }
535 
536 /*
537  * used to drop a reference on an ordered extent.  This will free
538  * the extent if the last reference is dropped
539  */
540 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
541 {
542 	struct list_head *cur;
543 	struct btrfs_ordered_sum *sum;
544 
545 	trace_btrfs_ordered_extent_put(entry->inode, entry);
546 
547 	if (atomic_dec_and_test(&entry->refs)) {
548 		if (entry->inode)
549 			btrfs_add_delayed_iput(entry->inode);
550 		while (!list_empty(&entry->list)) {
551 			cur = entry->list.next;
552 			sum = list_entry(cur, struct btrfs_ordered_sum, list);
553 			list_del(&sum->list);
554 			kfree(sum);
555 		}
556 		kmem_cache_free(btrfs_ordered_extent_cache, entry);
557 	}
558 }
559 
560 /*
561  * remove an ordered extent from the tree.  No references are dropped
562  * and waiters are woken up.
563  */
564 void btrfs_remove_ordered_extent(struct inode *inode,
565 				 struct btrfs_ordered_extent *entry)
566 {
567 	struct btrfs_ordered_inode_tree *tree;
568 	struct btrfs_root *root = BTRFS_I(inode)->root;
569 	struct rb_node *node;
570 
571 	tree = &BTRFS_I(inode)->ordered_tree;
572 	spin_lock_irq(&tree->lock);
573 	node = &entry->rb_node;
574 	rb_erase(node, &tree->tree);
575 	if (tree->last == node)
576 		tree->last = NULL;
577 	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
578 	spin_unlock_irq(&tree->lock);
579 
580 	spin_lock(&root->ordered_extent_lock);
581 	list_del_init(&entry->root_extent_list);
582 	root->nr_ordered_extents--;
583 
584 	trace_btrfs_ordered_extent_remove(inode, entry);
585 
586 	if (!root->nr_ordered_extents) {
587 		spin_lock(&root->fs_info->ordered_root_lock);
588 		BUG_ON(list_empty(&root->ordered_root));
589 		list_del_init(&root->ordered_root);
590 		spin_unlock(&root->fs_info->ordered_root_lock);
591 	}
592 	spin_unlock(&root->ordered_extent_lock);
593 	wake_up(&entry->wait);
594 }
595 
596 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
597 {
598 	struct btrfs_ordered_extent *ordered;
599 
600 	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
601 	btrfs_start_ordered_extent(ordered->inode, ordered, 1);
602 	complete(&ordered->completion);
603 }
604 
605 /*
606  * wait for all the ordered extents in a root.  This is done when balancing
607  * space between drives.
608  */
609 int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
610 {
611 	struct list_head splice, works;
612 	struct btrfs_ordered_extent *ordered, *next;
613 	int count = 0;
614 
615 	INIT_LIST_HEAD(&splice);
616 	INIT_LIST_HEAD(&works);
617 
618 	mutex_lock(&root->ordered_extent_mutex);
619 	spin_lock(&root->ordered_extent_lock);
620 	list_splice_init(&root->ordered_extents, &splice);
621 	while (!list_empty(&splice) && nr) {
622 		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
623 					   root_extent_list);
624 		list_move_tail(&ordered->root_extent_list,
625 			       &root->ordered_extents);
626 		atomic_inc(&ordered->refs);
627 		spin_unlock(&root->ordered_extent_lock);
628 
629 		btrfs_init_work(&ordered->flush_work,
630 				btrfs_flush_delalloc_helper,
631 				btrfs_run_ordered_extent_work, NULL, NULL);
632 		list_add_tail(&ordered->work_list, &works);
633 		btrfs_queue_work(root->fs_info->flush_workers,
634 				 &ordered->flush_work);
635 
636 		cond_resched();
637 		spin_lock(&root->ordered_extent_lock);
638 		if (nr != -1)
639 			nr--;
640 		count++;
641 	}
642 	list_splice_tail(&splice, &root->ordered_extents);
643 	spin_unlock(&root->ordered_extent_lock);
644 
645 	list_for_each_entry_safe(ordered, next, &works, work_list) {
646 		list_del_init(&ordered->work_list);
647 		wait_for_completion(&ordered->completion);
648 		btrfs_put_ordered_extent(ordered);
649 		cond_resched();
650 	}
651 	mutex_unlock(&root->ordered_extent_mutex);
652 
653 	return count;
654 }
655 
656 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
657 {
658 	struct btrfs_root *root;
659 	struct list_head splice;
660 	int done;
661 
662 	INIT_LIST_HEAD(&splice);
663 
664 	mutex_lock(&fs_info->ordered_operations_mutex);
665 	spin_lock(&fs_info->ordered_root_lock);
666 	list_splice_init(&fs_info->ordered_roots, &splice);
667 	while (!list_empty(&splice) && nr) {
668 		root = list_first_entry(&splice, struct btrfs_root,
669 					ordered_root);
670 		root = btrfs_grab_fs_root(root);
671 		BUG_ON(!root);
672 		list_move_tail(&root->ordered_root,
673 			       &fs_info->ordered_roots);
674 		spin_unlock(&fs_info->ordered_root_lock);
675 
676 		done = btrfs_wait_ordered_extents(root, nr);
677 		btrfs_put_fs_root(root);
678 
679 		spin_lock(&fs_info->ordered_root_lock);
680 		if (nr != -1) {
681 			nr -= done;
682 			WARN_ON(nr < 0);
683 		}
684 	}
685 	list_splice_tail(&splice, &fs_info->ordered_roots);
686 	spin_unlock(&fs_info->ordered_root_lock);
687 	mutex_unlock(&fs_info->ordered_operations_mutex);
688 }
689 
690 /*
691  * Used to start IO or wait for a given ordered extent to finish.
692  *
693  * If wait is one, this effectively waits on page writeback for all the pages
694  * in the extent, and it waits on the io completion code to insert
695  * metadata into the btree corresponding to the extent
696  */
697 void btrfs_start_ordered_extent(struct inode *inode,
698 				       struct btrfs_ordered_extent *entry,
699 				       int wait)
700 {
701 	u64 start = entry->file_offset;
702 	u64 end = start + entry->len - 1;
703 
704 	trace_btrfs_ordered_extent_start(inode, entry);
705 
706 	/*
707 	 * pages in the range can be dirty, clean or writeback.  We
708 	 * start IO on any dirty ones so the wait doesn't stall waiting
709 	 * for the flusher thread to find them
710 	 */
711 	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
712 		filemap_fdatawrite_range(inode->i_mapping, start, end);
713 	if (wait) {
714 		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
715 						 &entry->flags));
716 	}
717 }
718 
719 /*
720  * Used to wait on ordered extents across a large range of bytes.
721  */
722 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
723 {
724 	int ret = 0;
725 	u64 end;
726 	u64 orig_end;
727 	struct btrfs_ordered_extent *ordered;
728 
729 	if (start + len < start) {
730 		orig_end = INT_LIMIT(loff_t);
731 	} else {
732 		orig_end = start + len - 1;
733 		if (orig_end > INT_LIMIT(loff_t))
734 			orig_end = INT_LIMIT(loff_t);
735 	}
736 
737 	/* start IO across the range first to instantiate any delalloc
738 	 * extents
739 	 */
740 	ret = btrfs_fdatawrite_range(inode, start, orig_end);
741 	if (ret)
742 		return ret;
743 
744 	ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
745 	if (ret)
746 		return ret;
747 
748 	end = orig_end;
749 	while (1) {
750 		ordered = btrfs_lookup_first_ordered_extent(inode, end);
751 		if (!ordered)
752 			break;
753 		if (ordered->file_offset > orig_end) {
754 			btrfs_put_ordered_extent(ordered);
755 			break;
756 		}
757 		if (ordered->file_offset + ordered->len <= start) {
758 			btrfs_put_ordered_extent(ordered);
759 			break;
760 		}
761 		btrfs_start_ordered_extent(inode, ordered, 1);
762 		end = ordered->file_offset;
763 		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
764 			ret = -EIO;
765 		btrfs_put_ordered_extent(ordered);
766 		if (ret || end == 0 || end == start)
767 			break;
768 		end--;
769 	}
770 	return ret;
771 }
772 
773 /*
774  * find an ordered extent corresponding to file_offset.  return NULL if
775  * nothing is found, otherwise take a reference on the extent and return it
776  */
777 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
778 							 u64 file_offset)
779 {
780 	struct btrfs_ordered_inode_tree *tree;
781 	struct rb_node *node;
782 	struct btrfs_ordered_extent *entry = NULL;
783 
784 	tree = &BTRFS_I(inode)->ordered_tree;
785 	spin_lock_irq(&tree->lock);
786 	node = tree_search(tree, file_offset);
787 	if (!node)
788 		goto out;
789 
790 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
791 	if (!offset_in_entry(entry, file_offset))
792 		entry = NULL;
793 	if (entry)
794 		atomic_inc(&entry->refs);
795 out:
796 	spin_unlock_irq(&tree->lock);
797 	return entry;
798 }
799 
800 /* Since the DIO code tries to lock a wide area we need to look for any ordered
801  * extents that exist in the range, rather than just the start of the range.
802  */
803 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
804 							u64 file_offset,
805 							u64 len)
806 {
807 	struct btrfs_ordered_inode_tree *tree;
808 	struct rb_node *node;
809 	struct btrfs_ordered_extent *entry = NULL;
810 
811 	tree = &BTRFS_I(inode)->ordered_tree;
812 	spin_lock_irq(&tree->lock);
813 	node = tree_search(tree, file_offset);
814 	if (!node) {
815 		node = tree_search(tree, file_offset + len);
816 		if (!node)
817 			goto out;
818 	}
819 
820 	while (1) {
821 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
822 		if (range_overlaps(entry, file_offset, len))
823 			break;
824 
825 		if (entry->file_offset >= file_offset + len) {
826 			entry = NULL;
827 			break;
828 		}
829 		entry = NULL;
830 		node = rb_next(node);
831 		if (!node)
832 			break;
833 	}
834 out:
835 	if (entry)
836 		atomic_inc(&entry->refs);
837 	spin_unlock_irq(&tree->lock);
838 	return entry;
839 }
840 
841 /*
842  * lookup and return any extent before 'file_offset'.  NULL is returned
843  * if none is found
844  */
845 struct btrfs_ordered_extent *
846 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
847 {
848 	struct btrfs_ordered_inode_tree *tree;
849 	struct rb_node *node;
850 	struct btrfs_ordered_extent *entry = NULL;
851 
852 	tree = &BTRFS_I(inode)->ordered_tree;
853 	spin_lock_irq(&tree->lock);
854 	node = tree_search(tree, file_offset);
855 	if (!node)
856 		goto out;
857 
858 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
859 	atomic_inc(&entry->refs);
860 out:
861 	spin_unlock_irq(&tree->lock);
862 	return entry;
863 }
864 
865 /*
866  * After an extent is done, call this to conditionally update the on disk
867  * i_size.  i_size is updated to cover any fully written part of the file.
868  */
869 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
870 				struct btrfs_ordered_extent *ordered)
871 {
872 	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
873 	u64 disk_i_size;
874 	u64 new_i_size;
875 	u64 i_size = i_size_read(inode);
876 	struct rb_node *node;
877 	struct rb_node *prev = NULL;
878 	struct btrfs_ordered_extent *test;
879 	int ret = 1;
880 
881 	spin_lock_irq(&tree->lock);
882 	if (ordered) {
883 		offset = entry_end(ordered);
884 		if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
885 			offset = min(offset,
886 				     ordered->file_offset +
887 				     ordered->truncated_len);
888 	} else {
889 		offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
890 	}
891 	disk_i_size = BTRFS_I(inode)->disk_i_size;
892 
893 	/* truncate file */
894 	if (disk_i_size > i_size) {
895 		BTRFS_I(inode)->disk_i_size = i_size;
896 		ret = 0;
897 		goto out;
898 	}
899 
900 	/*
901 	 * if the disk i_size is already at the inode->i_size, or
902 	 * this ordered extent is inside the disk i_size, we're done
903 	 */
904 	if (disk_i_size == i_size)
905 		goto out;
906 
907 	/*
908 	 * We still need to update disk_i_size if outstanding_isize is greater
909 	 * than disk_i_size.
910 	 */
911 	if (offset <= disk_i_size &&
912 	    (!ordered || ordered->outstanding_isize <= disk_i_size))
913 		goto out;
914 
915 	/*
916 	 * walk backward from this ordered extent to disk_i_size.
917 	 * if we find an ordered extent then we can't update disk i_size
918 	 * yet
919 	 */
920 	if (ordered) {
921 		node = rb_prev(&ordered->rb_node);
922 	} else {
923 		prev = tree_search(tree, offset);
924 		/*
925 		 * we insert file extents without involving ordered struct,
926 		 * so there should be no ordered struct cover this offset
927 		 */
928 		if (prev) {
929 			test = rb_entry(prev, struct btrfs_ordered_extent,
930 					rb_node);
931 			BUG_ON(offset_in_entry(test, offset));
932 		}
933 		node = prev;
934 	}
935 	for (; node; node = rb_prev(node)) {
936 		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
937 
938 		/* We treat this entry as if it doesnt exist */
939 		if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
940 			continue;
941 		if (test->file_offset + test->len <= disk_i_size)
942 			break;
943 		if (test->file_offset >= i_size)
944 			break;
945 		if (entry_end(test) > disk_i_size) {
946 			/*
947 			 * we don't update disk_i_size now, so record this
948 			 * undealt i_size. Or we will not know the real
949 			 * i_size.
950 			 */
951 			if (test->outstanding_isize < offset)
952 				test->outstanding_isize = offset;
953 			if (ordered &&
954 			    ordered->outstanding_isize >
955 			    test->outstanding_isize)
956 				test->outstanding_isize =
957 						ordered->outstanding_isize;
958 			goto out;
959 		}
960 	}
961 	new_i_size = min_t(u64, offset, i_size);
962 
963 	/*
964 	 * Some ordered extents may completed before the current one, and
965 	 * we hold the real i_size in ->outstanding_isize.
966 	 */
967 	if (ordered && ordered->outstanding_isize > new_i_size)
968 		new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
969 	BTRFS_I(inode)->disk_i_size = new_i_size;
970 	ret = 0;
971 out:
972 	/*
973 	 * We need to do this because we can't remove ordered extents until
974 	 * after the i_disk_size has been updated and then the inode has been
975 	 * updated to reflect the change, so we need to tell anybody who finds
976 	 * this ordered extent that we've already done all the real work, we
977 	 * just haven't completed all the other work.
978 	 */
979 	if (ordered)
980 		set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
981 	spin_unlock_irq(&tree->lock);
982 	return ret;
983 }
984 
985 /*
986  * search the ordered extents for one corresponding to 'offset' and
987  * try to find a checksum.  This is used because we allow pages to
988  * be reclaimed before their checksum is actually put into the btree
989  */
990 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
991 			   u32 *sum, int len)
992 {
993 	struct btrfs_ordered_sum *ordered_sum;
994 	struct btrfs_ordered_extent *ordered;
995 	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
996 	unsigned long num_sectors;
997 	unsigned long i;
998 	u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
999 	int index = 0;
1000 
1001 	ordered = btrfs_lookup_ordered_extent(inode, offset);
1002 	if (!ordered)
1003 		return 0;
1004 
1005 	spin_lock_irq(&tree->lock);
1006 	list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1007 		if (disk_bytenr >= ordered_sum->bytenr &&
1008 		    disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1009 			i = (disk_bytenr - ordered_sum->bytenr) >>
1010 			    inode->i_sb->s_blocksize_bits;
1011 			num_sectors = ordered_sum->len >>
1012 				      inode->i_sb->s_blocksize_bits;
1013 			num_sectors = min_t(int, len - index, num_sectors - i);
1014 			memcpy(sum + index, ordered_sum->sums + i,
1015 			       num_sectors);
1016 
1017 			index += (int)num_sectors;
1018 			if (index == len)
1019 				goto out;
1020 			disk_bytenr += num_sectors * sectorsize;
1021 		}
1022 	}
1023 out:
1024 	spin_unlock_irq(&tree->lock);
1025 	btrfs_put_ordered_extent(ordered);
1026 	return index;
1027 }
1028 
1029 int __init ordered_data_init(void)
1030 {
1031 	btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1032 				     sizeof(struct btrfs_ordered_extent), 0,
1033 				     SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1034 				     NULL);
1035 	if (!btrfs_ordered_extent_cache)
1036 		return -ENOMEM;
1037 
1038 	return 0;
1039 }
1040 
1041 void ordered_data_exit(void)
1042 {
1043 	if (btrfs_ordered_extent_cache)
1044 		kmem_cache_destroy(btrfs_ordered_extent_cache);
1045 }
1046