xref: /openbmc/linux/fs/btrfs/ordered-data.c (revision 76ce0265)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/slab.h>
7 #include <linux/blkdev.h>
8 #include <linux/writeback.h>
9 #include <linux/sched/mm.h>
10 #include "misc.h"
11 #include "ctree.h"
12 #include "transaction.h"
13 #include "btrfs_inode.h"
14 #include "extent_io.h"
15 #include "disk-io.h"
16 #include "compression.h"
17 #include "delalloc-space.h"
18 
19 static struct kmem_cache *btrfs_ordered_extent_cache;
20 
21 static u64 entry_end(struct btrfs_ordered_extent *entry)
22 {
23 	if (entry->file_offset + entry->num_bytes < entry->file_offset)
24 		return (u64)-1;
25 	return entry->file_offset + entry->num_bytes;
26 }
27 
28 /* returns NULL if the insertion worked, or it returns the node it did find
29  * in the tree
30  */
31 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
32 				   struct rb_node *node)
33 {
34 	struct rb_node **p = &root->rb_node;
35 	struct rb_node *parent = NULL;
36 	struct btrfs_ordered_extent *entry;
37 
38 	while (*p) {
39 		parent = *p;
40 		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
41 
42 		if (file_offset < entry->file_offset)
43 			p = &(*p)->rb_left;
44 		else if (file_offset >= entry_end(entry))
45 			p = &(*p)->rb_right;
46 		else
47 			return parent;
48 	}
49 
50 	rb_link_node(node, parent, p);
51 	rb_insert_color(node, root);
52 	return NULL;
53 }
54 
55 /*
56  * look for a given offset in the tree, and if it can't be found return the
57  * first lesser offset
58  */
59 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
60 				     struct rb_node **prev_ret)
61 {
62 	struct rb_node *n = root->rb_node;
63 	struct rb_node *prev = NULL;
64 	struct rb_node *test;
65 	struct btrfs_ordered_extent *entry;
66 	struct btrfs_ordered_extent *prev_entry = NULL;
67 
68 	while (n) {
69 		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
70 		prev = n;
71 		prev_entry = entry;
72 
73 		if (file_offset < entry->file_offset)
74 			n = n->rb_left;
75 		else if (file_offset >= entry_end(entry))
76 			n = n->rb_right;
77 		else
78 			return n;
79 	}
80 	if (!prev_ret)
81 		return NULL;
82 
83 	while (prev && file_offset >= entry_end(prev_entry)) {
84 		test = rb_next(prev);
85 		if (!test)
86 			break;
87 		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
88 				      rb_node);
89 		if (file_offset < entry_end(prev_entry))
90 			break;
91 
92 		prev = test;
93 	}
94 	if (prev)
95 		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
96 				      rb_node);
97 	while (prev && file_offset < entry_end(prev_entry)) {
98 		test = rb_prev(prev);
99 		if (!test)
100 			break;
101 		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
102 				      rb_node);
103 		prev = test;
104 	}
105 	*prev_ret = prev;
106 	return NULL;
107 }
108 
109 /*
110  * helper to check if a given offset is inside a given entry
111  */
112 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
113 {
114 	if (file_offset < entry->file_offset ||
115 	    entry->file_offset + entry->num_bytes <= file_offset)
116 		return 0;
117 	return 1;
118 }
119 
120 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
121 			  u64 len)
122 {
123 	if (file_offset + len <= entry->file_offset ||
124 	    entry->file_offset + entry->num_bytes <= file_offset)
125 		return 0;
126 	return 1;
127 }
128 
129 /*
130  * look find the first ordered struct that has this offset, otherwise
131  * the first one less than this offset
132  */
133 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
134 					  u64 file_offset)
135 {
136 	struct rb_root *root = &tree->tree;
137 	struct rb_node *prev = NULL;
138 	struct rb_node *ret;
139 	struct btrfs_ordered_extent *entry;
140 
141 	if (tree->last) {
142 		entry = rb_entry(tree->last, struct btrfs_ordered_extent,
143 				 rb_node);
144 		if (offset_in_entry(entry, file_offset))
145 			return tree->last;
146 	}
147 	ret = __tree_search(root, file_offset, &prev);
148 	if (!ret)
149 		ret = prev;
150 	if (ret)
151 		tree->last = ret;
152 	return ret;
153 }
154 
155 /* allocate and add a new ordered_extent into the per-inode tree.
156  *
157  * The tree is given a single reference on the ordered extent that was
158  * inserted.
159  */
160 static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
161 				      u64 disk_bytenr, u64 num_bytes,
162 				      u64 disk_num_bytes, int type, int dio,
163 				      int compress_type)
164 {
165 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
166 	struct btrfs_root *root = BTRFS_I(inode)->root;
167 	struct btrfs_ordered_inode_tree *tree;
168 	struct rb_node *node;
169 	struct btrfs_ordered_extent *entry;
170 
171 	tree = &BTRFS_I(inode)->ordered_tree;
172 	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
173 	if (!entry)
174 		return -ENOMEM;
175 
176 	entry->file_offset = file_offset;
177 	entry->disk_bytenr = disk_bytenr;
178 	entry->num_bytes = num_bytes;
179 	entry->disk_num_bytes = disk_num_bytes;
180 	entry->bytes_left = num_bytes;
181 	entry->inode = igrab(inode);
182 	entry->compress_type = compress_type;
183 	entry->truncated_len = (u64)-1;
184 	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
185 		set_bit(type, &entry->flags);
186 
187 	if (dio) {
188 		percpu_counter_add_batch(&fs_info->dio_bytes, num_bytes,
189 					 fs_info->delalloc_batch);
190 		set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
191 	}
192 
193 	/* one ref for the tree */
194 	refcount_set(&entry->refs, 1);
195 	init_waitqueue_head(&entry->wait);
196 	INIT_LIST_HEAD(&entry->list);
197 	INIT_LIST_HEAD(&entry->root_extent_list);
198 	INIT_LIST_HEAD(&entry->work_list);
199 	init_completion(&entry->completion);
200 	INIT_LIST_HEAD(&entry->log_list);
201 	INIT_LIST_HEAD(&entry->trans_list);
202 
203 	trace_btrfs_ordered_extent_add(inode, entry);
204 
205 	spin_lock_irq(&tree->lock);
206 	node = tree_insert(&tree->tree, file_offset,
207 			   &entry->rb_node);
208 	if (node)
209 		btrfs_panic(fs_info, -EEXIST,
210 				"inconsistency in ordered tree at offset %llu",
211 				file_offset);
212 	spin_unlock_irq(&tree->lock);
213 
214 	spin_lock(&root->ordered_extent_lock);
215 	list_add_tail(&entry->root_extent_list,
216 		      &root->ordered_extents);
217 	root->nr_ordered_extents++;
218 	if (root->nr_ordered_extents == 1) {
219 		spin_lock(&fs_info->ordered_root_lock);
220 		BUG_ON(!list_empty(&root->ordered_root));
221 		list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
222 		spin_unlock(&fs_info->ordered_root_lock);
223 	}
224 	spin_unlock(&root->ordered_extent_lock);
225 
226 	/*
227 	 * We don't need the count_max_extents here, we can assume that all of
228 	 * that work has been done at higher layers, so this is truly the
229 	 * smallest the extent is going to get.
230 	 */
231 	spin_lock(&BTRFS_I(inode)->lock);
232 	btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
233 	spin_unlock(&BTRFS_I(inode)->lock);
234 
235 	return 0;
236 }
237 
238 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
239 			     u64 disk_bytenr, u64 num_bytes, u64 disk_num_bytes,
240 			     int type)
241 {
242 	return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
243 					  num_bytes, disk_num_bytes, type, 0,
244 					  BTRFS_COMPRESS_NONE);
245 }
246 
247 int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
248 				 u64 disk_bytenr, u64 num_bytes,
249 				 u64 disk_num_bytes, int type)
250 {
251 	return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
252 					  num_bytes, disk_num_bytes, type, 1,
253 					  BTRFS_COMPRESS_NONE);
254 }
255 
256 int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
257 				      u64 disk_bytenr, u64 num_bytes,
258 				      u64 disk_num_bytes, int type,
259 				      int compress_type)
260 {
261 	return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
262 					  num_bytes, disk_num_bytes, type, 0,
263 					  compress_type);
264 }
265 
266 /*
267  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
268  * when an ordered extent is finished.  If the list covers more than one
269  * ordered extent, it is split across multiples.
270  */
271 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
272 			   struct btrfs_ordered_sum *sum)
273 {
274 	struct btrfs_ordered_inode_tree *tree;
275 
276 	tree = &BTRFS_I(entry->inode)->ordered_tree;
277 	spin_lock_irq(&tree->lock);
278 	list_add_tail(&sum->list, &entry->list);
279 	spin_unlock_irq(&tree->lock);
280 }
281 
282 /*
283  * this is used to account for finished IO across a given range
284  * of the file.  The IO may span ordered extents.  If
285  * a given ordered_extent is completely done, 1 is returned, otherwise
286  * 0.
287  *
288  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
289  * to make sure this function only returns 1 once for a given ordered extent.
290  *
291  * file_offset is updated to one byte past the range that is recorded as
292  * complete.  This allows you to walk forward in the file.
293  */
294 int btrfs_dec_test_first_ordered_pending(struct inode *inode,
295 				   struct btrfs_ordered_extent **cached,
296 				   u64 *file_offset, u64 io_size, int uptodate)
297 {
298 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
299 	struct btrfs_ordered_inode_tree *tree;
300 	struct rb_node *node;
301 	struct btrfs_ordered_extent *entry = NULL;
302 	int ret;
303 	unsigned long flags;
304 	u64 dec_end;
305 	u64 dec_start;
306 	u64 to_dec;
307 
308 	tree = &BTRFS_I(inode)->ordered_tree;
309 	spin_lock_irqsave(&tree->lock, flags);
310 	node = tree_search(tree, *file_offset);
311 	if (!node) {
312 		ret = 1;
313 		goto out;
314 	}
315 
316 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
317 	if (!offset_in_entry(entry, *file_offset)) {
318 		ret = 1;
319 		goto out;
320 	}
321 
322 	dec_start = max(*file_offset, entry->file_offset);
323 	dec_end = min(*file_offset + io_size,
324 		      entry->file_offset + entry->num_bytes);
325 	*file_offset = dec_end;
326 	if (dec_start > dec_end) {
327 		btrfs_crit(fs_info, "bad ordering dec_start %llu end %llu",
328 			   dec_start, dec_end);
329 	}
330 	to_dec = dec_end - dec_start;
331 	if (to_dec > entry->bytes_left) {
332 		btrfs_crit(fs_info,
333 			   "bad ordered accounting left %llu size %llu",
334 			   entry->bytes_left, to_dec);
335 	}
336 	entry->bytes_left -= to_dec;
337 	if (!uptodate)
338 		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
339 
340 	if (entry->bytes_left == 0) {
341 		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
342 		/* test_and_set_bit implies a barrier */
343 		cond_wake_up_nomb(&entry->wait);
344 	} else {
345 		ret = 1;
346 	}
347 out:
348 	if (!ret && cached && entry) {
349 		*cached = entry;
350 		refcount_inc(&entry->refs);
351 	}
352 	spin_unlock_irqrestore(&tree->lock, flags);
353 	return ret == 0;
354 }
355 
356 /*
357  * this is used to account for finished IO across a given range
358  * of the file.  The IO should not span ordered extents.  If
359  * a given ordered_extent is completely done, 1 is returned, otherwise
360  * 0.
361  *
362  * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
363  * to make sure this function only returns 1 once for a given ordered extent.
364  */
365 int btrfs_dec_test_ordered_pending(struct inode *inode,
366 				   struct btrfs_ordered_extent **cached,
367 				   u64 file_offset, u64 io_size, int uptodate)
368 {
369 	struct btrfs_ordered_inode_tree *tree;
370 	struct rb_node *node;
371 	struct btrfs_ordered_extent *entry = NULL;
372 	unsigned long flags;
373 	int ret;
374 
375 	tree = &BTRFS_I(inode)->ordered_tree;
376 	spin_lock_irqsave(&tree->lock, flags);
377 	if (cached && *cached) {
378 		entry = *cached;
379 		goto have_entry;
380 	}
381 
382 	node = tree_search(tree, file_offset);
383 	if (!node) {
384 		ret = 1;
385 		goto out;
386 	}
387 
388 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
389 have_entry:
390 	if (!offset_in_entry(entry, file_offset)) {
391 		ret = 1;
392 		goto out;
393 	}
394 
395 	if (io_size > entry->bytes_left) {
396 		btrfs_crit(BTRFS_I(inode)->root->fs_info,
397 			   "bad ordered accounting left %llu size %llu",
398 		       entry->bytes_left, io_size);
399 	}
400 	entry->bytes_left -= io_size;
401 	if (!uptodate)
402 		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
403 
404 	if (entry->bytes_left == 0) {
405 		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
406 		/* test_and_set_bit implies a barrier */
407 		cond_wake_up_nomb(&entry->wait);
408 	} else {
409 		ret = 1;
410 	}
411 out:
412 	if (!ret && cached && entry) {
413 		*cached = entry;
414 		refcount_inc(&entry->refs);
415 	}
416 	spin_unlock_irqrestore(&tree->lock, flags);
417 	return ret == 0;
418 }
419 
420 /*
421  * used to drop a reference on an ordered extent.  This will free
422  * the extent if the last reference is dropped
423  */
424 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
425 {
426 	struct list_head *cur;
427 	struct btrfs_ordered_sum *sum;
428 
429 	trace_btrfs_ordered_extent_put(entry->inode, entry);
430 
431 	if (refcount_dec_and_test(&entry->refs)) {
432 		ASSERT(list_empty(&entry->log_list));
433 		ASSERT(list_empty(&entry->trans_list));
434 		ASSERT(list_empty(&entry->root_extent_list));
435 		ASSERT(RB_EMPTY_NODE(&entry->rb_node));
436 		if (entry->inode)
437 			btrfs_add_delayed_iput(entry->inode);
438 		while (!list_empty(&entry->list)) {
439 			cur = entry->list.next;
440 			sum = list_entry(cur, struct btrfs_ordered_sum, list);
441 			list_del(&sum->list);
442 			kvfree(sum);
443 		}
444 		kmem_cache_free(btrfs_ordered_extent_cache, entry);
445 	}
446 }
447 
448 /*
449  * remove an ordered extent from the tree.  No references are dropped
450  * and waiters are woken up.
451  */
452 void btrfs_remove_ordered_extent(struct inode *inode,
453 				 struct btrfs_ordered_extent *entry)
454 {
455 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
456 	struct btrfs_ordered_inode_tree *tree;
457 	struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
458 	struct btrfs_root *root = btrfs_inode->root;
459 	struct rb_node *node;
460 
461 	/* This is paired with btrfs_add_ordered_extent. */
462 	spin_lock(&btrfs_inode->lock);
463 	btrfs_mod_outstanding_extents(btrfs_inode, -1);
464 	spin_unlock(&btrfs_inode->lock);
465 	if (root != fs_info->tree_root)
466 		btrfs_delalloc_release_metadata(btrfs_inode, entry->num_bytes,
467 						false);
468 
469 	if (test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
470 		percpu_counter_add_batch(&fs_info->dio_bytes, -entry->num_bytes,
471 					 fs_info->delalloc_batch);
472 
473 	tree = &btrfs_inode->ordered_tree;
474 	spin_lock_irq(&tree->lock);
475 	node = &entry->rb_node;
476 	rb_erase(node, &tree->tree);
477 	RB_CLEAR_NODE(node);
478 	if (tree->last == node)
479 		tree->last = NULL;
480 	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
481 	spin_unlock_irq(&tree->lock);
482 
483 	spin_lock(&root->ordered_extent_lock);
484 	list_del_init(&entry->root_extent_list);
485 	root->nr_ordered_extents--;
486 
487 	trace_btrfs_ordered_extent_remove(inode, entry);
488 
489 	if (!root->nr_ordered_extents) {
490 		spin_lock(&fs_info->ordered_root_lock);
491 		BUG_ON(list_empty(&root->ordered_root));
492 		list_del_init(&root->ordered_root);
493 		spin_unlock(&fs_info->ordered_root_lock);
494 	}
495 	spin_unlock(&root->ordered_extent_lock);
496 	wake_up(&entry->wait);
497 }
498 
499 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
500 {
501 	struct btrfs_ordered_extent *ordered;
502 
503 	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
504 	btrfs_start_ordered_extent(ordered->inode, ordered, 1);
505 	complete(&ordered->completion);
506 }
507 
508 /*
509  * wait for all the ordered extents in a root.  This is done when balancing
510  * space between drives.
511  */
512 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
513 			       const u64 range_start, const u64 range_len)
514 {
515 	struct btrfs_fs_info *fs_info = root->fs_info;
516 	LIST_HEAD(splice);
517 	LIST_HEAD(skipped);
518 	LIST_HEAD(works);
519 	struct btrfs_ordered_extent *ordered, *next;
520 	u64 count = 0;
521 	const u64 range_end = range_start + range_len;
522 
523 	mutex_lock(&root->ordered_extent_mutex);
524 	spin_lock(&root->ordered_extent_lock);
525 	list_splice_init(&root->ordered_extents, &splice);
526 	while (!list_empty(&splice) && nr) {
527 		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
528 					   root_extent_list);
529 
530 		if (range_end <= ordered->disk_bytenr ||
531 		    ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
532 			list_move_tail(&ordered->root_extent_list, &skipped);
533 			cond_resched_lock(&root->ordered_extent_lock);
534 			continue;
535 		}
536 
537 		list_move_tail(&ordered->root_extent_list,
538 			       &root->ordered_extents);
539 		refcount_inc(&ordered->refs);
540 		spin_unlock(&root->ordered_extent_lock);
541 
542 		btrfs_init_work(&ordered->flush_work,
543 				btrfs_run_ordered_extent_work, NULL, NULL);
544 		list_add_tail(&ordered->work_list, &works);
545 		btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
546 
547 		cond_resched();
548 		spin_lock(&root->ordered_extent_lock);
549 		if (nr != U64_MAX)
550 			nr--;
551 		count++;
552 	}
553 	list_splice_tail(&skipped, &root->ordered_extents);
554 	list_splice_tail(&splice, &root->ordered_extents);
555 	spin_unlock(&root->ordered_extent_lock);
556 
557 	list_for_each_entry_safe(ordered, next, &works, work_list) {
558 		list_del_init(&ordered->work_list);
559 		wait_for_completion(&ordered->completion);
560 		btrfs_put_ordered_extent(ordered);
561 		cond_resched();
562 	}
563 	mutex_unlock(&root->ordered_extent_mutex);
564 
565 	return count;
566 }
567 
568 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
569 			     const u64 range_start, const u64 range_len)
570 {
571 	struct btrfs_root *root;
572 	struct list_head splice;
573 	u64 done;
574 
575 	INIT_LIST_HEAD(&splice);
576 
577 	mutex_lock(&fs_info->ordered_operations_mutex);
578 	spin_lock(&fs_info->ordered_root_lock);
579 	list_splice_init(&fs_info->ordered_roots, &splice);
580 	while (!list_empty(&splice) && nr) {
581 		root = list_first_entry(&splice, struct btrfs_root,
582 					ordered_root);
583 		root = btrfs_grab_fs_root(root);
584 		BUG_ON(!root);
585 		list_move_tail(&root->ordered_root,
586 			       &fs_info->ordered_roots);
587 		spin_unlock(&fs_info->ordered_root_lock);
588 
589 		done = btrfs_wait_ordered_extents(root, nr,
590 						  range_start, range_len);
591 		btrfs_put_fs_root(root);
592 
593 		spin_lock(&fs_info->ordered_root_lock);
594 		if (nr != U64_MAX) {
595 			nr -= done;
596 		}
597 	}
598 	list_splice_tail(&splice, &fs_info->ordered_roots);
599 	spin_unlock(&fs_info->ordered_root_lock);
600 	mutex_unlock(&fs_info->ordered_operations_mutex);
601 }
602 
603 /*
604  * Used to start IO or wait for a given ordered extent to finish.
605  *
606  * If wait is one, this effectively waits on page writeback for all the pages
607  * in the extent, and it waits on the io completion code to insert
608  * metadata into the btree corresponding to the extent
609  */
610 void btrfs_start_ordered_extent(struct inode *inode,
611 				       struct btrfs_ordered_extent *entry,
612 				       int wait)
613 {
614 	u64 start = entry->file_offset;
615 	u64 end = start + entry->num_bytes - 1;
616 
617 	trace_btrfs_ordered_extent_start(inode, entry);
618 
619 	/*
620 	 * pages in the range can be dirty, clean or writeback.  We
621 	 * start IO on any dirty ones so the wait doesn't stall waiting
622 	 * for the flusher thread to find them
623 	 */
624 	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
625 		filemap_fdatawrite_range(inode->i_mapping, start, end);
626 	if (wait) {
627 		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
628 						 &entry->flags));
629 	}
630 }
631 
632 /*
633  * Used to wait on ordered extents across a large range of bytes.
634  */
635 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
636 {
637 	int ret = 0;
638 	int ret_wb = 0;
639 	u64 end;
640 	u64 orig_end;
641 	struct btrfs_ordered_extent *ordered;
642 
643 	if (start + len < start) {
644 		orig_end = INT_LIMIT(loff_t);
645 	} else {
646 		orig_end = start + len - 1;
647 		if (orig_end > INT_LIMIT(loff_t))
648 			orig_end = INT_LIMIT(loff_t);
649 	}
650 
651 	/* start IO across the range first to instantiate any delalloc
652 	 * extents
653 	 */
654 	ret = btrfs_fdatawrite_range(inode, start, orig_end);
655 	if (ret)
656 		return ret;
657 
658 	/*
659 	 * If we have a writeback error don't return immediately. Wait first
660 	 * for any ordered extents that haven't completed yet. This is to make
661 	 * sure no one can dirty the same page ranges and call writepages()
662 	 * before the ordered extents complete - to avoid failures (-EEXIST)
663 	 * when adding the new ordered extents to the ordered tree.
664 	 */
665 	ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
666 
667 	end = orig_end;
668 	while (1) {
669 		ordered = btrfs_lookup_first_ordered_extent(inode, end);
670 		if (!ordered)
671 			break;
672 		if (ordered->file_offset > orig_end) {
673 			btrfs_put_ordered_extent(ordered);
674 			break;
675 		}
676 		if (ordered->file_offset + ordered->num_bytes <= start) {
677 			btrfs_put_ordered_extent(ordered);
678 			break;
679 		}
680 		btrfs_start_ordered_extent(inode, ordered, 1);
681 		end = ordered->file_offset;
682 		/*
683 		 * If the ordered extent had an error save the error but don't
684 		 * exit without waiting first for all other ordered extents in
685 		 * the range to complete.
686 		 */
687 		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
688 			ret = -EIO;
689 		btrfs_put_ordered_extent(ordered);
690 		if (end == 0 || end == start)
691 			break;
692 		end--;
693 	}
694 	return ret_wb ? ret_wb : ret;
695 }
696 
697 /*
698  * find an ordered extent corresponding to file_offset.  return NULL if
699  * nothing is found, otherwise take a reference on the extent and return it
700  */
701 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
702 							 u64 file_offset)
703 {
704 	struct btrfs_ordered_inode_tree *tree;
705 	struct rb_node *node;
706 	struct btrfs_ordered_extent *entry = NULL;
707 
708 	tree = &BTRFS_I(inode)->ordered_tree;
709 	spin_lock_irq(&tree->lock);
710 	node = tree_search(tree, file_offset);
711 	if (!node)
712 		goto out;
713 
714 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
715 	if (!offset_in_entry(entry, file_offset))
716 		entry = NULL;
717 	if (entry)
718 		refcount_inc(&entry->refs);
719 out:
720 	spin_unlock_irq(&tree->lock);
721 	return entry;
722 }
723 
724 /* Since the DIO code tries to lock a wide area we need to look for any ordered
725  * extents that exist in the range, rather than just the start of the range.
726  */
727 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
728 		struct btrfs_inode *inode, u64 file_offset, u64 len)
729 {
730 	struct btrfs_ordered_inode_tree *tree;
731 	struct rb_node *node;
732 	struct btrfs_ordered_extent *entry = NULL;
733 
734 	tree = &inode->ordered_tree;
735 	spin_lock_irq(&tree->lock);
736 	node = tree_search(tree, file_offset);
737 	if (!node) {
738 		node = tree_search(tree, file_offset + len);
739 		if (!node)
740 			goto out;
741 	}
742 
743 	while (1) {
744 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
745 		if (range_overlaps(entry, file_offset, len))
746 			break;
747 
748 		if (entry->file_offset >= file_offset + len) {
749 			entry = NULL;
750 			break;
751 		}
752 		entry = NULL;
753 		node = rb_next(node);
754 		if (!node)
755 			break;
756 	}
757 out:
758 	if (entry)
759 		refcount_inc(&entry->refs);
760 	spin_unlock_irq(&tree->lock);
761 	return entry;
762 }
763 
764 /*
765  * lookup and return any extent before 'file_offset'.  NULL is returned
766  * if none is found
767  */
768 struct btrfs_ordered_extent *
769 btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
770 {
771 	struct btrfs_ordered_inode_tree *tree;
772 	struct rb_node *node;
773 	struct btrfs_ordered_extent *entry = NULL;
774 
775 	tree = &BTRFS_I(inode)->ordered_tree;
776 	spin_lock_irq(&tree->lock);
777 	node = tree_search(tree, file_offset);
778 	if (!node)
779 		goto out;
780 
781 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
782 	refcount_inc(&entry->refs);
783 out:
784 	spin_unlock_irq(&tree->lock);
785 	return entry;
786 }
787 
788 /*
789  * After an extent is done, call this to conditionally update the on disk
790  * i_size.  i_size is updated to cover any fully written part of the file.
791  */
792 int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
793 				struct btrfs_ordered_extent *ordered)
794 {
795 	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
796 	u64 disk_i_size;
797 	u64 new_i_size;
798 	u64 i_size = i_size_read(inode);
799 	struct rb_node *node;
800 	struct rb_node *prev = NULL;
801 	struct btrfs_ordered_extent *test;
802 	int ret = 1;
803 	u64 orig_offset = offset;
804 
805 	spin_lock_irq(&tree->lock);
806 	if (ordered) {
807 		offset = entry_end(ordered);
808 		if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
809 			offset = min(offset,
810 				     ordered->file_offset +
811 				     ordered->truncated_len);
812 	} else {
813 		offset = ALIGN(offset, btrfs_inode_sectorsize(inode));
814 	}
815 	disk_i_size = BTRFS_I(inode)->disk_i_size;
816 
817 	/*
818 	 * truncate file.
819 	 * If ordered is not NULL, then this is called from endio and
820 	 * disk_i_size will be updated by either truncate itself or any
821 	 * in-flight IOs which are inside the disk_i_size.
822 	 *
823 	 * Because btrfs_setsize() may set i_size with disk_i_size if truncate
824 	 * fails somehow, we need to make sure we have a precise disk_i_size by
825 	 * updating it as usual.
826 	 *
827 	 */
828 	if (!ordered && disk_i_size > i_size) {
829 		BTRFS_I(inode)->disk_i_size = orig_offset;
830 		ret = 0;
831 		goto out;
832 	}
833 
834 	/*
835 	 * if the disk i_size is already at the inode->i_size, or
836 	 * this ordered extent is inside the disk i_size, we're done
837 	 */
838 	if (disk_i_size == i_size)
839 		goto out;
840 
841 	/*
842 	 * We still need to update disk_i_size if outstanding_isize is greater
843 	 * than disk_i_size.
844 	 */
845 	if (offset <= disk_i_size &&
846 	    (!ordered || ordered->outstanding_isize <= disk_i_size))
847 		goto out;
848 
849 	/*
850 	 * walk backward from this ordered extent to disk_i_size.
851 	 * if we find an ordered extent then we can't update disk i_size
852 	 * yet
853 	 */
854 	if (ordered) {
855 		node = rb_prev(&ordered->rb_node);
856 	} else {
857 		prev = tree_search(tree, offset);
858 		/*
859 		 * we insert file extents without involving ordered struct,
860 		 * so there should be no ordered struct cover this offset
861 		 */
862 		if (prev) {
863 			test = rb_entry(prev, struct btrfs_ordered_extent,
864 					rb_node);
865 			BUG_ON(offset_in_entry(test, offset));
866 		}
867 		node = prev;
868 	}
869 	for (; node; node = rb_prev(node)) {
870 		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
871 
872 		/* We treat this entry as if it doesn't exist */
873 		if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
874 			continue;
875 
876 		if (entry_end(test) <= disk_i_size)
877 			break;
878 		if (test->file_offset >= i_size)
879 			break;
880 
881 		/*
882 		 * We don't update disk_i_size now, so record this undealt
883 		 * i_size. Or we will not know the real i_size.
884 		 */
885 		if (test->outstanding_isize < offset)
886 			test->outstanding_isize = offset;
887 		if (ordered &&
888 		    ordered->outstanding_isize > test->outstanding_isize)
889 			test->outstanding_isize = ordered->outstanding_isize;
890 		goto out;
891 	}
892 	new_i_size = min_t(u64, offset, i_size);
893 
894 	/*
895 	 * Some ordered extents may completed before the current one, and
896 	 * we hold the real i_size in ->outstanding_isize.
897 	 */
898 	if (ordered && ordered->outstanding_isize > new_i_size)
899 		new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
900 	BTRFS_I(inode)->disk_i_size = new_i_size;
901 	ret = 0;
902 out:
903 	/*
904 	 * We need to do this because we can't remove ordered extents until
905 	 * after the i_disk_size has been updated and then the inode has been
906 	 * updated to reflect the change, so we need to tell anybody who finds
907 	 * this ordered extent that we've already done all the real work, we
908 	 * just haven't completed all the other work.
909 	 */
910 	if (ordered)
911 		set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
912 	spin_unlock_irq(&tree->lock);
913 	return ret;
914 }
915 
916 /*
917  * search the ordered extents for one corresponding to 'offset' and
918  * try to find a checksum.  This is used because we allow pages to
919  * be reclaimed before their checksum is actually put into the btree
920  */
921 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
922 			   u8 *sum, int len)
923 {
924 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
925 	struct btrfs_ordered_sum *ordered_sum;
926 	struct btrfs_ordered_extent *ordered;
927 	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
928 	unsigned long num_sectors;
929 	unsigned long i;
930 	u32 sectorsize = btrfs_inode_sectorsize(inode);
931 	const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
932 	int index = 0;
933 
934 	ordered = btrfs_lookup_ordered_extent(inode, offset);
935 	if (!ordered)
936 		return 0;
937 
938 	spin_lock_irq(&tree->lock);
939 	list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
940 		if (disk_bytenr >= ordered_sum->bytenr &&
941 		    disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
942 			i = (disk_bytenr - ordered_sum->bytenr) >>
943 			    inode->i_sb->s_blocksize_bits;
944 			num_sectors = ordered_sum->len >>
945 				      inode->i_sb->s_blocksize_bits;
946 			num_sectors = min_t(int, len - index, num_sectors - i);
947 			memcpy(sum + index, ordered_sum->sums + i * csum_size,
948 			       num_sectors * csum_size);
949 
950 			index += (int)num_sectors * csum_size;
951 			if (index == len)
952 				goto out;
953 			disk_bytenr += num_sectors * sectorsize;
954 		}
955 	}
956 out:
957 	spin_unlock_irq(&tree->lock);
958 	btrfs_put_ordered_extent(ordered);
959 	return index;
960 }
961 
962 /*
963  * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
964  * ordered extents in it are run to completion.
965  *
966  * @tree:         IO tree used for locking out other users of the range
967  * @inode:        Inode whose ordered tree is to be searched
968  * @start:        Beginning of range to flush
969  * @end:          Last byte of range to lock
970  * @cached_state: If passed, will return the extent state responsible for the
971  * locked range. It's the caller's responsibility to free the cached state.
972  *
973  * This function always returns with the given range locked, ensuring after it's
974  * called no order extent can be pending.
975  */
976 void btrfs_lock_and_flush_ordered_range(struct extent_io_tree *tree,
977 					struct btrfs_inode *inode, u64 start,
978 					u64 end,
979 					struct extent_state **cached_state)
980 {
981 	struct btrfs_ordered_extent *ordered;
982 	struct extent_state *cache = NULL;
983 	struct extent_state **cachedp = &cache;
984 
985 	if (cached_state)
986 		cachedp = cached_state;
987 
988 	while (1) {
989 		lock_extent_bits(tree, start, end, cachedp);
990 		ordered = btrfs_lookup_ordered_range(inode, start,
991 						     end - start + 1);
992 		if (!ordered) {
993 			/*
994 			 * If no external cached_state has been passed then
995 			 * decrement the extra ref taken for cachedp since we
996 			 * aren't exposing it outside of this function
997 			 */
998 			if (!cached_state)
999 				refcount_dec(&cache->refs);
1000 			break;
1001 		}
1002 		unlock_extent_cached(tree, start, end, cachedp);
1003 		btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1);
1004 		btrfs_put_ordered_extent(ordered);
1005 	}
1006 }
1007 
1008 int __init ordered_data_init(void)
1009 {
1010 	btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1011 				     sizeof(struct btrfs_ordered_extent), 0,
1012 				     SLAB_MEM_SPREAD,
1013 				     NULL);
1014 	if (!btrfs_ordered_extent_cache)
1015 		return -ENOMEM;
1016 
1017 	return 0;
1018 }
1019 
1020 void __cold ordered_data_exit(void)
1021 {
1022 	kmem_cache_destroy(btrfs_ordered_extent_cache);
1023 }
1024