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 #include "qgroup.h"
19 #include "subpage.h"
20 
21 static struct kmem_cache *btrfs_ordered_extent_cache;
22 
entry_end(struct btrfs_ordered_extent * entry)23 static u64 entry_end(struct btrfs_ordered_extent *entry)
24 {
25 	if (entry->file_offset + entry->num_bytes < entry->file_offset)
26 		return (u64)-1;
27 	return entry->file_offset + entry->num_bytes;
28 }
29 
30 /* returns NULL if the insertion worked, or it returns the node it did find
31  * in the tree
32  */
tree_insert(struct rb_root * root,u64 file_offset,struct rb_node * node)33 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
34 				   struct rb_node *node)
35 {
36 	struct rb_node **p = &root->rb_node;
37 	struct rb_node *parent = NULL;
38 	struct btrfs_ordered_extent *entry;
39 
40 	while (*p) {
41 		parent = *p;
42 		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
43 
44 		if (file_offset < entry->file_offset)
45 			p = &(*p)->rb_left;
46 		else if (file_offset >= entry_end(entry))
47 			p = &(*p)->rb_right;
48 		else
49 			return parent;
50 	}
51 
52 	rb_link_node(node, parent, p);
53 	rb_insert_color(node, root);
54 	return NULL;
55 }
56 
57 /*
58  * look for a given offset in the tree, and if it can't be found return the
59  * first lesser offset
60  */
__tree_search(struct rb_root * root,u64 file_offset,struct rb_node ** prev_ret)61 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
62 				     struct rb_node **prev_ret)
63 {
64 	struct rb_node *n = root->rb_node;
65 	struct rb_node *prev = NULL;
66 	struct rb_node *test;
67 	struct btrfs_ordered_extent *entry;
68 	struct btrfs_ordered_extent *prev_entry = NULL;
69 
70 	while (n) {
71 		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
72 		prev = n;
73 		prev_entry = entry;
74 
75 		if (file_offset < entry->file_offset)
76 			n = n->rb_left;
77 		else if (file_offset >= entry_end(entry))
78 			n = n->rb_right;
79 		else
80 			return n;
81 	}
82 	if (!prev_ret)
83 		return NULL;
84 
85 	while (prev && file_offset >= entry_end(prev_entry)) {
86 		test = rb_next(prev);
87 		if (!test)
88 			break;
89 		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
90 				      rb_node);
91 		if (file_offset < entry_end(prev_entry))
92 			break;
93 
94 		prev = test;
95 	}
96 	if (prev)
97 		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
98 				      rb_node);
99 	while (prev && file_offset < entry_end(prev_entry)) {
100 		test = rb_prev(prev);
101 		if (!test)
102 			break;
103 		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
104 				      rb_node);
105 		prev = test;
106 	}
107 	*prev_ret = prev;
108 	return NULL;
109 }
110 
range_overlaps(struct btrfs_ordered_extent * entry,u64 file_offset,u64 len)111 static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
112 			  u64 len)
113 {
114 	if (file_offset + len <= entry->file_offset ||
115 	    entry->file_offset + entry->num_bytes <= file_offset)
116 		return 0;
117 	return 1;
118 }
119 
120 /*
121  * look find the first ordered struct that has this offset, otherwise
122  * the first one less than this offset
123  */
tree_search(struct btrfs_ordered_inode_tree * tree,u64 file_offset)124 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
125 					  u64 file_offset)
126 {
127 	struct rb_root *root = &tree->tree;
128 	struct rb_node *prev = NULL;
129 	struct rb_node *ret;
130 	struct btrfs_ordered_extent *entry;
131 
132 	if (tree->last) {
133 		entry = rb_entry(tree->last, struct btrfs_ordered_extent,
134 				 rb_node);
135 		if (in_range(file_offset, entry->file_offset, entry->num_bytes))
136 			return tree->last;
137 	}
138 	ret = __tree_search(root, file_offset, &prev);
139 	if (!ret)
140 		ret = prev;
141 	if (ret)
142 		tree->last = ret;
143 	return ret;
144 }
145 
146 /*
147  * Allocate and add a new ordered_extent into the per-inode tree.
148  *
149  * The tree is given a single reference on the ordered extent that was
150  * inserted.
151  */
__btrfs_add_ordered_extent(struct btrfs_inode * inode,u64 file_offset,u64 disk_bytenr,u64 num_bytes,u64 disk_num_bytes,int type,int dio,int compress_type)152 static int __btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
153 				      u64 disk_bytenr, u64 num_bytes,
154 				      u64 disk_num_bytes, int type, int dio,
155 				      int compress_type)
156 {
157 	struct btrfs_root *root = inode->root;
158 	struct btrfs_fs_info *fs_info = root->fs_info;
159 	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
160 	struct rb_node *node;
161 	struct btrfs_ordered_extent *entry;
162 	int ret;
163 
164 	if (type == BTRFS_ORDERED_NOCOW || type == BTRFS_ORDERED_PREALLOC) {
165 		/* For nocow write, we can release the qgroup rsv right now */
166 		ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes);
167 		if (ret < 0)
168 			return ret;
169 		ret = 0;
170 	} else {
171 		/*
172 		 * The ordered extent has reserved qgroup space, release now
173 		 * and pass the reserved number for qgroup_record to free.
174 		 */
175 		ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes);
176 		if (ret < 0)
177 			return ret;
178 	}
179 	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
180 	if (!entry)
181 		return -ENOMEM;
182 
183 	entry->file_offset = file_offset;
184 	entry->disk_bytenr = disk_bytenr;
185 	entry->num_bytes = num_bytes;
186 	entry->disk_num_bytes = disk_num_bytes;
187 	entry->bytes_left = num_bytes;
188 	entry->inode = igrab(&inode->vfs_inode);
189 	entry->compress_type = compress_type;
190 	entry->truncated_len = (u64)-1;
191 	entry->qgroup_rsv = ret;
192 	entry->physical = (u64)-1;
193 
194 	ASSERT(type == BTRFS_ORDERED_REGULAR ||
195 	       type == BTRFS_ORDERED_NOCOW ||
196 	       type == BTRFS_ORDERED_PREALLOC ||
197 	       type == BTRFS_ORDERED_COMPRESSED);
198 	set_bit(type, &entry->flags);
199 
200 	percpu_counter_add_batch(&fs_info->ordered_bytes, num_bytes,
201 				 fs_info->delalloc_batch);
202 
203 	if (dio)
204 		set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
205 
206 	/* one ref for the tree */
207 	refcount_set(&entry->refs, 1);
208 	init_waitqueue_head(&entry->wait);
209 	INIT_LIST_HEAD(&entry->list);
210 	INIT_LIST_HEAD(&entry->log_list);
211 	INIT_LIST_HEAD(&entry->root_extent_list);
212 	INIT_LIST_HEAD(&entry->work_list);
213 	init_completion(&entry->completion);
214 
215 	trace_btrfs_ordered_extent_add(inode, entry);
216 
217 	spin_lock_irq(&tree->lock);
218 	node = tree_insert(&tree->tree, file_offset,
219 			   &entry->rb_node);
220 	if (node)
221 		btrfs_panic(fs_info, -EEXIST,
222 				"inconsistency in ordered tree at offset %llu",
223 				file_offset);
224 	spin_unlock_irq(&tree->lock);
225 
226 	spin_lock(&root->ordered_extent_lock);
227 	list_add_tail(&entry->root_extent_list,
228 		      &root->ordered_extents);
229 	root->nr_ordered_extents++;
230 	if (root->nr_ordered_extents == 1) {
231 		spin_lock(&fs_info->ordered_root_lock);
232 		BUG_ON(!list_empty(&root->ordered_root));
233 		list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
234 		spin_unlock(&fs_info->ordered_root_lock);
235 	}
236 	spin_unlock(&root->ordered_extent_lock);
237 
238 	/*
239 	 * We don't need the count_max_extents here, we can assume that all of
240 	 * that work has been done at higher layers, so this is truly the
241 	 * smallest the extent is going to get.
242 	 */
243 	spin_lock(&inode->lock);
244 	btrfs_mod_outstanding_extents(inode, 1);
245 	spin_unlock(&inode->lock);
246 
247 	return 0;
248 }
249 
btrfs_add_ordered_extent(struct btrfs_inode * inode,u64 file_offset,u64 disk_bytenr,u64 num_bytes,u64 disk_num_bytes,int type)250 int btrfs_add_ordered_extent(struct btrfs_inode *inode, u64 file_offset,
251 			     u64 disk_bytenr, u64 num_bytes, u64 disk_num_bytes,
252 			     int type)
253 {
254 	ASSERT(type == BTRFS_ORDERED_REGULAR ||
255 	       type == BTRFS_ORDERED_NOCOW ||
256 	       type == BTRFS_ORDERED_PREALLOC);
257 	return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
258 					  num_bytes, disk_num_bytes, type, 0,
259 					  BTRFS_COMPRESS_NONE);
260 }
261 
btrfs_add_ordered_extent_dio(struct btrfs_inode * inode,u64 file_offset,u64 disk_bytenr,u64 num_bytes,u64 disk_num_bytes,int type)262 int btrfs_add_ordered_extent_dio(struct btrfs_inode *inode, u64 file_offset,
263 				 u64 disk_bytenr, u64 num_bytes,
264 				 u64 disk_num_bytes, int type)
265 {
266 	ASSERT(type == BTRFS_ORDERED_REGULAR ||
267 	       type == BTRFS_ORDERED_NOCOW ||
268 	       type == BTRFS_ORDERED_PREALLOC);
269 	return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
270 					  num_bytes, disk_num_bytes, type, 1,
271 					  BTRFS_COMPRESS_NONE);
272 }
273 
btrfs_add_ordered_extent_compress(struct btrfs_inode * inode,u64 file_offset,u64 disk_bytenr,u64 num_bytes,u64 disk_num_bytes,int compress_type)274 int btrfs_add_ordered_extent_compress(struct btrfs_inode *inode, u64 file_offset,
275 				      u64 disk_bytenr, u64 num_bytes,
276 				      u64 disk_num_bytes, int compress_type)
277 {
278 	ASSERT(compress_type != BTRFS_COMPRESS_NONE);
279 	return __btrfs_add_ordered_extent(inode, file_offset, disk_bytenr,
280 					  num_bytes, disk_num_bytes,
281 					  BTRFS_ORDERED_COMPRESSED, 0,
282 					  compress_type);
283 }
284 
285 /*
286  * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
287  * when an ordered extent is finished.  If the list covers more than one
288  * ordered extent, it is split across multiples.
289  */
btrfs_add_ordered_sum(struct btrfs_ordered_extent * entry,struct btrfs_ordered_sum * sum)290 void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
291 			   struct btrfs_ordered_sum *sum)
292 {
293 	struct btrfs_ordered_inode_tree *tree;
294 
295 	tree = &BTRFS_I(entry->inode)->ordered_tree;
296 	spin_lock_irq(&tree->lock);
297 	list_add_tail(&sum->list, &entry->list);
298 	spin_unlock_irq(&tree->lock);
299 }
300 
301 /*
302  * Mark all ordered extents io inside the specified range finished.
303  *
304  * @page:	 The invovled page for the opeartion.
305  *		 For uncompressed buffered IO, the page status also needs to be
306  *		 updated to indicate whether the pending ordered io is finished.
307  *		 Can be NULL for direct IO and compressed write.
308  *		 For these cases, callers are ensured they won't execute the
309  *		 endio function twice.
310  * @finish_func: The function to be executed when all the IO of an ordered
311  *		 extent are finished.
312  *
313  * This function is called for endio, thus the range must have ordered
314  * extent(s) coveri it.
315  */
btrfs_mark_ordered_io_finished(struct btrfs_inode * inode,struct page * page,u64 file_offset,u64 num_bytes,btrfs_func_t finish_func,bool uptodate)316 void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
317 				struct page *page, u64 file_offset,
318 				u64 num_bytes, btrfs_func_t finish_func,
319 				bool uptodate)
320 {
321 	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
322 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
323 	struct btrfs_workqueue *wq;
324 	struct rb_node *node;
325 	struct btrfs_ordered_extent *entry = NULL;
326 	unsigned long flags;
327 	u64 cur = file_offset;
328 
329 	if (btrfs_is_free_space_inode(inode))
330 		wq = fs_info->endio_freespace_worker;
331 	else
332 		wq = fs_info->endio_write_workers;
333 
334 	if (page)
335 		ASSERT(page->mapping && page_offset(page) <= file_offset &&
336 		       file_offset + num_bytes <= page_offset(page) + PAGE_SIZE);
337 
338 	spin_lock_irqsave(&tree->lock, flags);
339 	while (cur < file_offset + num_bytes) {
340 		u64 entry_end;
341 		u64 end;
342 		u32 len;
343 
344 		node = tree_search(tree, cur);
345 		/* No ordered extents at all */
346 		if (!node)
347 			break;
348 
349 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
350 		entry_end = entry->file_offset + entry->num_bytes;
351 		/*
352 		 * |<-- OE --->|  |
353 		 *		  cur
354 		 * Go to next OE.
355 		 */
356 		if (cur >= entry_end) {
357 			node = rb_next(node);
358 			/* No more ordered extents, exit */
359 			if (!node)
360 				break;
361 			entry = rb_entry(node, struct btrfs_ordered_extent,
362 					 rb_node);
363 
364 			/* Go to next ordered extent and continue */
365 			cur = entry->file_offset;
366 			continue;
367 		}
368 		/*
369 		 * |	|<--- OE --->|
370 		 * cur
371 		 * Go to the start of OE.
372 		 */
373 		if (cur < entry->file_offset) {
374 			cur = entry->file_offset;
375 			continue;
376 		}
377 
378 		/*
379 		 * Now we are definitely inside one ordered extent.
380 		 *
381 		 * |<--- OE --->|
382 		 *	|
383 		 *	cur
384 		 */
385 		end = min(entry->file_offset + entry->num_bytes,
386 			  file_offset + num_bytes) - 1;
387 		ASSERT(end + 1 - cur < U32_MAX);
388 		len = end + 1 - cur;
389 
390 		if (page) {
391 			/*
392 			 * Ordered (Private2) bit indicates whether we still
393 			 * have pending io unfinished for the ordered extent.
394 			 *
395 			 * If there's no such bit, we need to skip to next range.
396 			 */
397 			if (!btrfs_page_test_ordered(fs_info, page, cur, len)) {
398 				cur += len;
399 				continue;
400 			}
401 			btrfs_page_clear_ordered(fs_info, page, cur, len);
402 		}
403 
404 		/* Now we're fine to update the accounting */
405 		if (unlikely(len > entry->bytes_left)) {
406 			WARN_ON(1);
407 			btrfs_crit(fs_info,
408 "bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%u left=%llu",
409 				   inode->root->root_key.objectid,
410 				   btrfs_ino(inode),
411 				   entry->file_offset,
412 				   entry->num_bytes,
413 				   len, entry->bytes_left);
414 			entry->bytes_left = 0;
415 		} else {
416 			entry->bytes_left -= len;
417 		}
418 
419 		if (!uptodate)
420 			set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
421 
422 		/*
423 		 * All the IO of the ordered extent is finished, we need to queue
424 		 * the finish_func to be executed.
425 		 */
426 		if (entry->bytes_left == 0) {
427 			set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
428 			cond_wake_up(&entry->wait);
429 			refcount_inc(&entry->refs);
430 			spin_unlock_irqrestore(&tree->lock, flags);
431 			btrfs_init_work(&entry->work, finish_func, NULL, NULL);
432 			btrfs_queue_work(wq, &entry->work);
433 			spin_lock_irqsave(&tree->lock, flags);
434 		}
435 		cur += len;
436 	}
437 	spin_unlock_irqrestore(&tree->lock, flags);
438 }
439 
440 /*
441  * Finish IO for one ordered extent across a given range.  The range can only
442  * contain one ordered extent.
443  *
444  * @cached:	 The cached ordered extent. If not NULL, we can skip the tree
445  *               search and use the ordered extent directly.
446  * 		 Will be also used to store the finished ordered extent.
447  * @file_offset: File offset for the finished IO
448  * @io_size:	 Length of the finish IO range
449  *
450  * Return true if the ordered extent is finished in the range, and update
451  * @cached.
452  * Return false otherwise.
453  *
454  * NOTE: The range can NOT cross multiple ordered extents.
455  * Thus caller should ensure the range doesn't cross ordered extents.
456  */
btrfs_dec_test_ordered_pending(struct btrfs_inode * inode,struct btrfs_ordered_extent ** cached,u64 file_offset,u64 io_size)457 bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
458 				    struct btrfs_ordered_extent **cached,
459 				    u64 file_offset, u64 io_size)
460 {
461 	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
462 	struct rb_node *node;
463 	struct btrfs_ordered_extent *entry = NULL;
464 	unsigned long flags;
465 	bool finished = false;
466 
467 	spin_lock_irqsave(&tree->lock, flags);
468 	if (cached && *cached) {
469 		entry = *cached;
470 		goto have_entry;
471 	}
472 
473 	node = tree_search(tree, file_offset);
474 	if (!node)
475 		goto out;
476 
477 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
478 have_entry:
479 	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
480 		goto out;
481 
482 	if (io_size > entry->bytes_left)
483 		btrfs_crit(inode->root->fs_info,
484 			   "bad ordered accounting left %llu size %llu",
485 		       entry->bytes_left, io_size);
486 
487 	entry->bytes_left -= io_size;
488 
489 	if (entry->bytes_left == 0) {
490 		/*
491 		 * Ensure only one caller can set the flag and finished_ret
492 		 * accordingly
493 		 */
494 		finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
495 		/* test_and_set_bit implies a barrier */
496 		cond_wake_up_nomb(&entry->wait);
497 	}
498 out:
499 	if (finished && cached && entry) {
500 		*cached = entry;
501 		refcount_inc(&entry->refs);
502 	}
503 	spin_unlock_irqrestore(&tree->lock, flags);
504 	return finished;
505 }
506 
507 /*
508  * used to drop a reference on an ordered extent.  This will free
509  * the extent if the last reference is dropped
510  */
btrfs_put_ordered_extent(struct btrfs_ordered_extent * entry)511 void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
512 {
513 	struct list_head *cur;
514 	struct btrfs_ordered_sum *sum;
515 
516 	trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
517 
518 	if (refcount_dec_and_test(&entry->refs)) {
519 		ASSERT(list_empty(&entry->root_extent_list));
520 		ASSERT(list_empty(&entry->log_list));
521 		ASSERT(RB_EMPTY_NODE(&entry->rb_node));
522 		if (entry->inode)
523 			btrfs_add_delayed_iput(entry->inode);
524 		while (!list_empty(&entry->list)) {
525 			cur = entry->list.next;
526 			sum = list_entry(cur, struct btrfs_ordered_sum, list);
527 			list_del(&sum->list);
528 			kvfree(sum);
529 		}
530 		kmem_cache_free(btrfs_ordered_extent_cache, entry);
531 	}
532 }
533 
534 /*
535  * remove an ordered extent from the tree.  No references are dropped
536  * and waiters are woken up.
537  */
btrfs_remove_ordered_extent(struct btrfs_inode * btrfs_inode,struct btrfs_ordered_extent * entry)538 void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
539 				 struct btrfs_ordered_extent *entry)
540 {
541 	struct btrfs_ordered_inode_tree *tree;
542 	struct btrfs_root *root = btrfs_inode->root;
543 	struct btrfs_fs_info *fs_info = root->fs_info;
544 	struct rb_node *node;
545 	bool pending;
546 
547 	/* This is paired with btrfs_add_ordered_extent. */
548 	spin_lock(&btrfs_inode->lock);
549 	btrfs_mod_outstanding_extents(btrfs_inode, -1);
550 	spin_unlock(&btrfs_inode->lock);
551 	if (root != fs_info->tree_root)
552 		btrfs_delalloc_release_metadata(btrfs_inode, entry->num_bytes,
553 						false);
554 
555 	percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
556 				 fs_info->delalloc_batch);
557 
558 	tree = &btrfs_inode->ordered_tree;
559 	spin_lock_irq(&tree->lock);
560 	node = &entry->rb_node;
561 	rb_erase(node, &tree->tree);
562 	RB_CLEAR_NODE(node);
563 	if (tree->last == node)
564 		tree->last = NULL;
565 	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
566 	pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
567 	spin_unlock_irq(&tree->lock);
568 
569 	/*
570 	 * The current running transaction is waiting on us, we need to let it
571 	 * know that we're complete and wake it up.
572 	 */
573 	if (pending) {
574 		struct btrfs_transaction *trans;
575 
576 		/*
577 		 * The checks for trans are just a formality, it should be set,
578 		 * but if it isn't we don't want to deref/assert under the spin
579 		 * lock, so be nice and check if trans is set, but ASSERT() so
580 		 * if it isn't set a developer will notice.
581 		 */
582 		spin_lock(&fs_info->trans_lock);
583 		trans = fs_info->running_transaction;
584 		if (trans)
585 			refcount_inc(&trans->use_count);
586 		spin_unlock(&fs_info->trans_lock);
587 
588 		ASSERT(trans);
589 		if (trans) {
590 			if (atomic_dec_and_test(&trans->pending_ordered))
591 				wake_up(&trans->pending_wait);
592 			btrfs_put_transaction(trans);
593 		}
594 	}
595 
596 	spin_lock(&root->ordered_extent_lock);
597 	list_del_init(&entry->root_extent_list);
598 	root->nr_ordered_extents--;
599 
600 	trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
601 
602 	if (!root->nr_ordered_extents) {
603 		spin_lock(&fs_info->ordered_root_lock);
604 		BUG_ON(list_empty(&root->ordered_root));
605 		list_del_init(&root->ordered_root);
606 		spin_unlock(&fs_info->ordered_root_lock);
607 	}
608 	spin_unlock(&root->ordered_extent_lock);
609 	wake_up(&entry->wait);
610 }
611 
btrfs_run_ordered_extent_work(struct btrfs_work * work)612 static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
613 {
614 	struct btrfs_ordered_extent *ordered;
615 
616 	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
617 	btrfs_start_ordered_extent(ordered, 1);
618 	complete(&ordered->completion);
619 }
620 
621 /*
622  * wait for all the ordered extents in a root.  This is done when balancing
623  * space between drives.
624  */
btrfs_wait_ordered_extents(struct btrfs_root * root,u64 nr,const u64 range_start,const u64 range_len)625 u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
626 			       const u64 range_start, const u64 range_len)
627 {
628 	struct btrfs_fs_info *fs_info = root->fs_info;
629 	LIST_HEAD(splice);
630 	LIST_HEAD(skipped);
631 	LIST_HEAD(works);
632 	struct btrfs_ordered_extent *ordered, *next;
633 	u64 count = 0;
634 	const u64 range_end = range_start + range_len;
635 
636 	mutex_lock(&root->ordered_extent_mutex);
637 	spin_lock(&root->ordered_extent_lock);
638 	list_splice_init(&root->ordered_extents, &splice);
639 	while (!list_empty(&splice) && nr) {
640 		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
641 					   root_extent_list);
642 
643 		if (range_end <= ordered->disk_bytenr ||
644 		    ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
645 			list_move_tail(&ordered->root_extent_list, &skipped);
646 			cond_resched_lock(&root->ordered_extent_lock);
647 			continue;
648 		}
649 
650 		list_move_tail(&ordered->root_extent_list,
651 			       &root->ordered_extents);
652 		refcount_inc(&ordered->refs);
653 		spin_unlock(&root->ordered_extent_lock);
654 
655 		btrfs_init_work(&ordered->flush_work,
656 				btrfs_run_ordered_extent_work, NULL, NULL);
657 		list_add_tail(&ordered->work_list, &works);
658 		btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
659 
660 		cond_resched();
661 		spin_lock(&root->ordered_extent_lock);
662 		if (nr != U64_MAX)
663 			nr--;
664 		count++;
665 	}
666 	list_splice_tail(&skipped, &root->ordered_extents);
667 	list_splice_tail(&splice, &root->ordered_extents);
668 	spin_unlock(&root->ordered_extent_lock);
669 
670 	list_for_each_entry_safe(ordered, next, &works, work_list) {
671 		list_del_init(&ordered->work_list);
672 		wait_for_completion(&ordered->completion);
673 		btrfs_put_ordered_extent(ordered);
674 		cond_resched();
675 	}
676 	mutex_unlock(&root->ordered_extent_mutex);
677 
678 	return count;
679 }
680 
btrfs_wait_ordered_roots(struct btrfs_fs_info * fs_info,u64 nr,const u64 range_start,const u64 range_len)681 void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
682 			     const u64 range_start, const u64 range_len)
683 {
684 	struct btrfs_root *root;
685 	struct list_head splice;
686 	u64 done;
687 
688 	INIT_LIST_HEAD(&splice);
689 
690 	mutex_lock(&fs_info->ordered_operations_mutex);
691 	spin_lock(&fs_info->ordered_root_lock);
692 	list_splice_init(&fs_info->ordered_roots, &splice);
693 	while (!list_empty(&splice) && nr) {
694 		root = list_first_entry(&splice, struct btrfs_root,
695 					ordered_root);
696 		root = btrfs_grab_root(root);
697 		BUG_ON(!root);
698 		list_move_tail(&root->ordered_root,
699 			       &fs_info->ordered_roots);
700 		spin_unlock(&fs_info->ordered_root_lock);
701 
702 		done = btrfs_wait_ordered_extents(root, nr,
703 						  range_start, range_len);
704 		btrfs_put_root(root);
705 
706 		spin_lock(&fs_info->ordered_root_lock);
707 		if (nr != U64_MAX) {
708 			nr -= done;
709 		}
710 	}
711 	list_splice_tail(&splice, &fs_info->ordered_roots);
712 	spin_unlock(&fs_info->ordered_root_lock);
713 	mutex_unlock(&fs_info->ordered_operations_mutex);
714 }
715 
716 /*
717  * Used to start IO or wait for a given ordered extent to finish.
718  *
719  * If wait is one, this effectively waits on page writeback for all the pages
720  * in the extent, and it waits on the io completion code to insert
721  * metadata into the btree corresponding to the extent
722  */
btrfs_start_ordered_extent(struct btrfs_ordered_extent * entry,int wait)723 void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry, int wait)
724 {
725 	u64 start = entry->file_offset;
726 	u64 end = start + entry->num_bytes - 1;
727 	struct btrfs_inode *inode = BTRFS_I(entry->inode);
728 
729 	trace_btrfs_ordered_extent_start(inode, entry);
730 
731 	/*
732 	 * pages in the range can be dirty, clean or writeback.  We
733 	 * start IO on any dirty ones so the wait doesn't stall waiting
734 	 * for the flusher thread to find them
735 	 */
736 	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
737 		filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
738 	if (wait) {
739 		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
740 						 &entry->flags));
741 	}
742 }
743 
744 /*
745  * Used to wait on ordered extents across a large range of bytes.
746  */
btrfs_wait_ordered_range(struct inode * inode,u64 start,u64 len)747 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
748 {
749 	int ret = 0;
750 	int ret_wb = 0;
751 	u64 end;
752 	u64 orig_end;
753 	struct btrfs_ordered_extent *ordered;
754 
755 	if (start + len < start) {
756 		orig_end = INT_LIMIT(loff_t);
757 	} else {
758 		orig_end = start + len - 1;
759 		if (orig_end > INT_LIMIT(loff_t))
760 			orig_end = INT_LIMIT(loff_t);
761 	}
762 
763 	/* start IO across the range first to instantiate any delalloc
764 	 * extents
765 	 */
766 	ret = btrfs_fdatawrite_range(inode, start, orig_end);
767 	if (ret)
768 		return ret;
769 
770 	/*
771 	 * If we have a writeback error don't return immediately. Wait first
772 	 * for any ordered extents that haven't completed yet. This is to make
773 	 * sure no one can dirty the same page ranges and call writepages()
774 	 * before the ordered extents complete - to avoid failures (-EEXIST)
775 	 * when adding the new ordered extents to the ordered tree.
776 	 */
777 	ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
778 
779 	end = orig_end;
780 	while (1) {
781 		ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
782 		if (!ordered)
783 			break;
784 		if (ordered->file_offset > orig_end) {
785 			btrfs_put_ordered_extent(ordered);
786 			break;
787 		}
788 		if (ordered->file_offset + ordered->num_bytes <= start) {
789 			btrfs_put_ordered_extent(ordered);
790 			break;
791 		}
792 		btrfs_start_ordered_extent(ordered, 1);
793 		end = ordered->file_offset;
794 		/*
795 		 * If the ordered extent had an error save the error but don't
796 		 * exit without waiting first for all other ordered extents in
797 		 * the range to complete.
798 		 */
799 		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
800 			ret = -EIO;
801 		btrfs_put_ordered_extent(ordered);
802 		if (end == 0 || end == start)
803 			break;
804 		end--;
805 	}
806 	return ret_wb ? ret_wb : ret;
807 }
808 
809 /*
810  * find an ordered extent corresponding to file_offset.  return NULL if
811  * nothing is found, otherwise take a reference on the extent and return it
812  */
btrfs_lookup_ordered_extent(struct btrfs_inode * inode,u64 file_offset)813 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
814 							 u64 file_offset)
815 {
816 	struct btrfs_ordered_inode_tree *tree;
817 	struct rb_node *node;
818 	struct btrfs_ordered_extent *entry = NULL;
819 	unsigned long flags;
820 
821 	tree = &inode->ordered_tree;
822 	spin_lock_irqsave(&tree->lock, flags);
823 	node = tree_search(tree, file_offset);
824 	if (!node)
825 		goto out;
826 
827 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
828 	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
829 		entry = NULL;
830 	if (entry)
831 		refcount_inc(&entry->refs);
832 out:
833 	spin_unlock_irqrestore(&tree->lock, flags);
834 	return entry;
835 }
836 
837 /* Since the DIO code tries to lock a wide area we need to look for any ordered
838  * extents that exist in the range, rather than just the start of the range.
839  */
btrfs_lookup_ordered_range(struct btrfs_inode * inode,u64 file_offset,u64 len)840 struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
841 		struct btrfs_inode *inode, u64 file_offset, u64 len)
842 {
843 	struct btrfs_ordered_inode_tree *tree;
844 	struct rb_node *node;
845 	struct btrfs_ordered_extent *entry = NULL;
846 
847 	tree = &inode->ordered_tree;
848 	spin_lock_irq(&tree->lock);
849 	node = tree_search(tree, file_offset);
850 	if (!node) {
851 		node = tree_search(tree, file_offset + len);
852 		if (!node)
853 			goto out;
854 	}
855 
856 	while (1) {
857 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
858 		if (range_overlaps(entry, file_offset, len))
859 			break;
860 
861 		if (entry->file_offset >= file_offset + len) {
862 			entry = NULL;
863 			break;
864 		}
865 		entry = NULL;
866 		node = rb_next(node);
867 		if (!node)
868 			break;
869 	}
870 out:
871 	if (entry)
872 		refcount_inc(&entry->refs);
873 	spin_unlock_irq(&tree->lock);
874 	return entry;
875 }
876 
877 /*
878  * Adds all ordered extents to the given list. The list ends up sorted by the
879  * file_offset of the ordered extents.
880  */
btrfs_get_ordered_extents_for_logging(struct btrfs_inode * inode,struct list_head * list)881 void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
882 					   struct list_head *list)
883 {
884 	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
885 	struct rb_node *n;
886 
887 	ASSERT(inode_is_locked(&inode->vfs_inode));
888 
889 	spin_lock_irq(&tree->lock);
890 	for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
891 		struct btrfs_ordered_extent *ordered;
892 
893 		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
894 
895 		if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
896 			continue;
897 
898 		ASSERT(list_empty(&ordered->log_list));
899 		list_add_tail(&ordered->log_list, list);
900 		refcount_inc(&ordered->refs);
901 	}
902 	spin_unlock_irq(&tree->lock);
903 }
904 
905 /*
906  * lookup and return any extent before 'file_offset'.  NULL is returned
907  * if none is found
908  */
909 struct btrfs_ordered_extent *
btrfs_lookup_first_ordered_extent(struct btrfs_inode * inode,u64 file_offset)910 btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
911 {
912 	struct btrfs_ordered_inode_tree *tree;
913 	struct rb_node *node;
914 	struct btrfs_ordered_extent *entry = NULL;
915 
916 	tree = &inode->ordered_tree;
917 	spin_lock_irq(&tree->lock);
918 	node = tree_search(tree, file_offset);
919 	if (!node)
920 		goto out;
921 
922 	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
923 	refcount_inc(&entry->refs);
924 out:
925 	spin_unlock_irq(&tree->lock);
926 	return entry;
927 }
928 
929 /*
930  * Lookup the first ordered extent that overlaps the range
931  * [@file_offset, @file_offset + @len).
932  *
933  * The difference between this and btrfs_lookup_first_ordered_extent() is
934  * that this one won't return any ordered extent that does not overlap the range.
935  * And the difference against btrfs_lookup_ordered_extent() is, this function
936  * ensures the first ordered extent gets returned.
937  */
btrfs_lookup_first_ordered_range(struct btrfs_inode * inode,u64 file_offset,u64 len)938 struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
939 			struct btrfs_inode *inode, u64 file_offset, u64 len)
940 {
941 	struct btrfs_ordered_inode_tree *tree = &inode->ordered_tree;
942 	struct rb_node *node;
943 	struct rb_node *cur;
944 	struct rb_node *prev;
945 	struct rb_node *next;
946 	struct btrfs_ordered_extent *entry = NULL;
947 
948 	spin_lock_irq(&tree->lock);
949 	node = tree->tree.rb_node;
950 	/*
951 	 * Here we don't want to use tree_search() which will use tree->last
952 	 * and screw up the search order.
953 	 * And __tree_search() can't return the adjacent ordered extents
954 	 * either, thus here we do our own search.
955 	 */
956 	while (node) {
957 		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
958 
959 		if (file_offset < entry->file_offset) {
960 			node = node->rb_left;
961 		} else if (file_offset >= entry_end(entry)) {
962 			node = node->rb_right;
963 		} else {
964 			/*
965 			 * Direct hit, got an ordered extent that starts at
966 			 * @file_offset
967 			 */
968 			goto out;
969 		}
970 	}
971 	if (!entry) {
972 		/* Empty tree */
973 		goto out;
974 	}
975 
976 	cur = &entry->rb_node;
977 	/* We got an entry around @file_offset, check adjacent entries */
978 	if (entry->file_offset < file_offset) {
979 		prev = cur;
980 		next = rb_next(cur);
981 	} else {
982 		prev = rb_prev(cur);
983 		next = cur;
984 	}
985 	if (prev) {
986 		entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
987 		if (range_overlaps(entry, file_offset, len))
988 			goto out;
989 	}
990 	if (next) {
991 		entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
992 		if (range_overlaps(entry, file_offset, len))
993 			goto out;
994 	}
995 	/* No ordered extent in the range */
996 	entry = NULL;
997 out:
998 	if (entry)
999 		refcount_inc(&entry->refs);
1000 	spin_unlock_irq(&tree->lock);
1001 	return entry;
1002 }
1003 
1004 /*
1005  * btrfs_flush_ordered_range - Lock the passed range and ensures all pending
1006  * ordered extents in it are run to completion.
1007  *
1008  * @inode:        Inode whose ordered tree is to be searched
1009  * @start:        Beginning of range to flush
1010  * @end:          Last byte of range to lock
1011  * @cached_state: If passed, will return the extent state responsible for the
1012  * locked range. It's the caller's responsibility to free the cached state.
1013  *
1014  * This function always returns with the given range locked, ensuring after it's
1015  * called no order extent can be pending.
1016  */
btrfs_lock_and_flush_ordered_range(struct btrfs_inode * inode,u64 start,u64 end,struct extent_state ** cached_state)1017 void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1018 					u64 end,
1019 					struct extent_state **cached_state)
1020 {
1021 	struct btrfs_ordered_extent *ordered;
1022 	struct extent_state *cache = NULL;
1023 	struct extent_state **cachedp = &cache;
1024 
1025 	if (cached_state)
1026 		cachedp = cached_state;
1027 
1028 	while (1) {
1029 		lock_extent_bits(&inode->io_tree, start, end, cachedp);
1030 		ordered = btrfs_lookup_ordered_range(inode, start,
1031 						     end - start + 1);
1032 		if (!ordered) {
1033 			/*
1034 			 * If no external cached_state has been passed then
1035 			 * decrement the extra ref taken for cachedp since we
1036 			 * aren't exposing it outside of this function
1037 			 */
1038 			if (!cached_state)
1039 				refcount_dec(&cache->refs);
1040 			break;
1041 		}
1042 		unlock_extent_cached(&inode->io_tree, start, end, cachedp);
1043 		btrfs_start_ordered_extent(ordered, 1);
1044 		btrfs_put_ordered_extent(ordered);
1045 	}
1046 }
1047 
clone_ordered_extent(struct btrfs_ordered_extent * ordered,u64 pos,u64 len)1048 static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
1049 				u64 len)
1050 {
1051 	struct inode *inode = ordered->inode;
1052 	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1053 	u64 file_offset = ordered->file_offset + pos;
1054 	u64 disk_bytenr = ordered->disk_bytenr + pos;
1055 	u64 num_bytes = len;
1056 	u64 disk_num_bytes = len;
1057 	int type;
1058 	unsigned long flags_masked = ordered->flags & ~(1 << BTRFS_ORDERED_DIRECT);
1059 	int compress_type = ordered->compress_type;
1060 	unsigned long weight;
1061 	int ret;
1062 
1063 	weight = hweight_long(flags_masked);
1064 	WARN_ON_ONCE(weight > 1);
1065 	if (!weight)
1066 		type = 0;
1067 	else
1068 		type = __ffs(flags_masked);
1069 
1070 	/*
1071 	 * The splitting extent is already counted and will be added again
1072 	 * in btrfs_add_ordered_extent_*(). Subtract num_bytes to avoid
1073 	 * double counting.
1074 	 */
1075 	percpu_counter_add_batch(&fs_info->ordered_bytes, -num_bytes,
1076 				 fs_info->delalloc_batch);
1077 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered->flags)) {
1078 		WARN_ON_ONCE(1);
1079 		ret = btrfs_add_ordered_extent_compress(BTRFS_I(inode),
1080 				file_offset, disk_bytenr, num_bytes,
1081 				disk_num_bytes, compress_type);
1082 	} else if (test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
1083 		ret = btrfs_add_ordered_extent_dio(BTRFS_I(inode), file_offset,
1084 				disk_bytenr, num_bytes, disk_num_bytes, type);
1085 	} else {
1086 		ret = btrfs_add_ordered_extent(BTRFS_I(inode), file_offset,
1087 				disk_bytenr, num_bytes, disk_num_bytes, type);
1088 	}
1089 
1090 	return ret;
1091 }
1092 
btrfs_split_ordered_extent(struct btrfs_ordered_extent * ordered,u64 pre,u64 post)1093 int btrfs_split_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pre,
1094 				u64 post)
1095 {
1096 	struct inode *inode = ordered->inode;
1097 	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1098 	struct rb_node *node;
1099 	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1100 	int ret = 0;
1101 
1102 	spin_lock_irq(&tree->lock);
1103 	/* Remove from tree once */
1104 	node = &ordered->rb_node;
1105 	rb_erase(node, &tree->tree);
1106 	RB_CLEAR_NODE(node);
1107 	if (tree->last == node)
1108 		tree->last = NULL;
1109 
1110 	ordered->file_offset += pre;
1111 	ordered->disk_bytenr += pre;
1112 	ordered->num_bytes -= (pre + post);
1113 	ordered->disk_num_bytes -= (pre + post);
1114 	ordered->bytes_left -= (pre + post);
1115 
1116 	/* Re-insert the node */
1117 	node = tree_insert(&tree->tree, ordered->file_offset, &ordered->rb_node);
1118 	if (node)
1119 		btrfs_panic(fs_info, -EEXIST,
1120 			"zoned: inconsistency in ordered tree at offset %llu",
1121 			    ordered->file_offset);
1122 
1123 	spin_unlock_irq(&tree->lock);
1124 
1125 	if (pre)
1126 		ret = clone_ordered_extent(ordered, 0, pre);
1127 	if (ret == 0 && post)
1128 		ret = clone_ordered_extent(ordered, pre + ordered->disk_num_bytes,
1129 					   post);
1130 
1131 	return ret;
1132 }
1133 
ordered_data_init(void)1134 int __init ordered_data_init(void)
1135 {
1136 	btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1137 				     sizeof(struct btrfs_ordered_extent), 0,
1138 				     SLAB_MEM_SPREAD,
1139 				     NULL);
1140 	if (!btrfs_ordered_extent_cache)
1141 		return -ENOMEM;
1142 
1143 	return 0;
1144 }
1145 
ordered_data_exit(void)1146 void __cold ordered_data_exit(void)
1147 {
1148 	kmem_cache_destroy(btrfs_ordered_extent_cache);
1149 }
1150