1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/err.h>
4 #include <linux/slab.h>
5 #include <linux/spinlock.h>
6 #include "ctree.h"
7 #include "volumes.h"
8 #include "extent_map.h"
9 #include "compression.h"
10 #include "btrfs_inode.h"
11 
12 
13 static struct kmem_cache *extent_map_cache;
14 
extent_map_init(void)15 int __init extent_map_init(void)
16 {
17 	extent_map_cache = kmem_cache_create("btrfs_extent_map",
18 			sizeof(struct extent_map), 0,
19 			SLAB_MEM_SPREAD, NULL);
20 	if (!extent_map_cache)
21 		return -ENOMEM;
22 	return 0;
23 }
24 
extent_map_exit(void)25 void __cold extent_map_exit(void)
26 {
27 	kmem_cache_destroy(extent_map_cache);
28 }
29 
30 /**
31  * extent_map_tree_init - initialize extent map tree
32  * @tree:		tree to initialize
33  *
34  * Initialize the extent tree @tree.  Should be called for each new inode
35  * or other user of the extent_map interface.
36  */
extent_map_tree_init(struct extent_map_tree * tree)37 void extent_map_tree_init(struct extent_map_tree *tree)
38 {
39 	tree->map = RB_ROOT_CACHED;
40 	INIT_LIST_HEAD(&tree->modified_extents);
41 	rwlock_init(&tree->lock);
42 }
43 
44 /**
45  * alloc_extent_map - allocate new extent map structure
46  *
47  * Allocate a new extent_map structure.  The new structure is
48  * returned with a reference count of one and needs to be
49  * freed using free_extent_map()
50  */
alloc_extent_map(void)51 struct extent_map *alloc_extent_map(void)
52 {
53 	struct extent_map *em;
54 	em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
55 	if (!em)
56 		return NULL;
57 	RB_CLEAR_NODE(&em->rb_node);
58 	em->compress_type = BTRFS_COMPRESS_NONE;
59 	refcount_set(&em->refs, 1);
60 	INIT_LIST_HEAD(&em->list);
61 	return em;
62 }
63 
64 /**
65  * free_extent_map - drop reference count of an extent_map
66  * @em:		extent map being released
67  *
68  * Drops the reference out on @em by one and free the structure
69  * if the reference count hits zero.
70  */
free_extent_map(struct extent_map * em)71 void free_extent_map(struct extent_map *em)
72 {
73 	if (!em)
74 		return;
75 	if (refcount_dec_and_test(&em->refs)) {
76 		WARN_ON(extent_map_in_tree(em));
77 		WARN_ON(!list_empty(&em->list));
78 		if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
79 			kfree(em->map_lookup);
80 		kmem_cache_free(extent_map_cache, em);
81 	}
82 }
83 
84 /* simple helper to do math around the end of an extent, handling wrap */
range_end(u64 start,u64 len)85 static u64 range_end(u64 start, u64 len)
86 {
87 	if (start + len < start)
88 		return (u64)-1;
89 	return start + len;
90 }
91 
tree_insert(struct rb_root_cached * root,struct extent_map * em)92 static int tree_insert(struct rb_root_cached *root, struct extent_map *em)
93 {
94 	struct rb_node **p = &root->rb_root.rb_node;
95 	struct rb_node *parent = NULL;
96 	struct extent_map *entry = NULL;
97 	struct rb_node *orig_parent = NULL;
98 	u64 end = range_end(em->start, em->len);
99 	bool leftmost = true;
100 
101 	while (*p) {
102 		parent = *p;
103 		entry = rb_entry(parent, struct extent_map, rb_node);
104 
105 		if (em->start < entry->start) {
106 			p = &(*p)->rb_left;
107 		} else if (em->start >= extent_map_end(entry)) {
108 			p = &(*p)->rb_right;
109 			leftmost = false;
110 		} else {
111 			return -EEXIST;
112 		}
113 	}
114 
115 	orig_parent = parent;
116 	while (parent && em->start >= extent_map_end(entry)) {
117 		parent = rb_next(parent);
118 		entry = rb_entry(parent, struct extent_map, rb_node);
119 	}
120 	if (parent)
121 		if (end > entry->start && em->start < extent_map_end(entry))
122 			return -EEXIST;
123 
124 	parent = orig_parent;
125 	entry = rb_entry(parent, struct extent_map, rb_node);
126 	while (parent && em->start < entry->start) {
127 		parent = rb_prev(parent);
128 		entry = rb_entry(parent, struct extent_map, rb_node);
129 	}
130 	if (parent)
131 		if (end > entry->start && em->start < extent_map_end(entry))
132 			return -EEXIST;
133 
134 	rb_link_node(&em->rb_node, orig_parent, p);
135 	rb_insert_color_cached(&em->rb_node, root, leftmost);
136 	return 0;
137 }
138 
139 /*
140  * search through the tree for an extent_map with a given offset.  If
141  * it can't be found, try to find some neighboring extents
142  */
__tree_search(struct rb_root * root,u64 offset,struct rb_node ** prev_or_next_ret)143 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
144 				     struct rb_node **prev_or_next_ret)
145 {
146 	struct rb_node *n = root->rb_node;
147 	struct rb_node *prev = NULL;
148 	struct rb_node *orig_prev = NULL;
149 	struct extent_map *entry;
150 	struct extent_map *prev_entry = NULL;
151 
152 	ASSERT(prev_or_next_ret);
153 
154 	while (n) {
155 		entry = rb_entry(n, struct extent_map, rb_node);
156 		prev = n;
157 		prev_entry = entry;
158 
159 		if (offset < entry->start)
160 			n = n->rb_left;
161 		else if (offset >= extent_map_end(entry))
162 			n = n->rb_right;
163 		else
164 			return n;
165 	}
166 
167 	orig_prev = prev;
168 	while (prev && offset >= extent_map_end(prev_entry)) {
169 		prev = rb_next(prev);
170 		prev_entry = rb_entry(prev, struct extent_map, rb_node);
171 	}
172 
173 	/*
174 	 * Previous extent map found, return as in this case the caller does not
175 	 * care about the next one.
176 	 */
177 	if (prev) {
178 		*prev_or_next_ret = prev;
179 		return NULL;
180 	}
181 
182 	prev = orig_prev;
183 	prev_entry = rb_entry(prev, struct extent_map, rb_node);
184 	while (prev && offset < prev_entry->start) {
185 		prev = rb_prev(prev);
186 		prev_entry = rb_entry(prev, struct extent_map, rb_node);
187 	}
188 	*prev_or_next_ret = prev;
189 
190 	return NULL;
191 }
192 
193 /* check to see if two extent_map structs are adjacent and safe to merge */
mergable_maps(struct extent_map * prev,struct extent_map * next)194 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
195 {
196 	if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
197 		return 0;
198 
199 	/*
200 	 * don't merge compressed extents, we need to know their
201 	 * actual size
202 	 */
203 	if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
204 		return 0;
205 
206 	if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
207 	    test_bit(EXTENT_FLAG_LOGGING, &next->flags))
208 		return 0;
209 
210 	/*
211 	 * We don't want to merge stuff that hasn't been written to the log yet
212 	 * since it may not reflect exactly what is on disk, and that would be
213 	 * bad.
214 	 */
215 	if (!list_empty(&prev->list) || !list_empty(&next->list))
216 		return 0;
217 
218 	ASSERT(next->block_start != EXTENT_MAP_DELALLOC &&
219 	       prev->block_start != EXTENT_MAP_DELALLOC);
220 
221 	if (prev->map_lookup || next->map_lookup)
222 		ASSERT(test_bit(EXTENT_FLAG_FS_MAPPING, &prev->flags) &&
223 		       test_bit(EXTENT_FLAG_FS_MAPPING, &next->flags));
224 
225 	if (extent_map_end(prev) == next->start &&
226 	    prev->flags == next->flags &&
227 	    prev->map_lookup == next->map_lookup &&
228 	    ((next->block_start == EXTENT_MAP_HOLE &&
229 	      prev->block_start == EXTENT_MAP_HOLE) ||
230 	     (next->block_start == EXTENT_MAP_INLINE &&
231 	      prev->block_start == EXTENT_MAP_INLINE) ||
232 	     (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
233 	      next->block_start == extent_map_block_end(prev)))) {
234 		return 1;
235 	}
236 	return 0;
237 }
238 
try_merge_map(struct extent_map_tree * tree,struct extent_map * em)239 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
240 {
241 	struct extent_map *merge = NULL;
242 	struct rb_node *rb;
243 
244 	/*
245 	 * We can't modify an extent map that is in the tree and that is being
246 	 * used by another task, as it can cause that other task to see it in
247 	 * inconsistent state during the merging. We always have 1 reference for
248 	 * the tree and 1 for this task (which is unpinning the extent map or
249 	 * clearing the logging flag), so anything > 2 means it's being used by
250 	 * other tasks too.
251 	 */
252 	if (refcount_read(&em->refs) > 2)
253 		return;
254 
255 	if (em->start != 0) {
256 		rb = rb_prev(&em->rb_node);
257 		if (rb)
258 			merge = rb_entry(rb, struct extent_map, rb_node);
259 		if (rb && mergable_maps(merge, em)) {
260 			em->start = merge->start;
261 			em->orig_start = merge->orig_start;
262 			em->len += merge->len;
263 			em->block_len += merge->block_len;
264 			em->block_start = merge->block_start;
265 			em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
266 			em->mod_start = merge->mod_start;
267 			em->generation = max(em->generation, merge->generation);
268 			set_bit(EXTENT_FLAG_MERGED, &em->flags);
269 
270 			rb_erase_cached(&merge->rb_node, &tree->map);
271 			RB_CLEAR_NODE(&merge->rb_node);
272 			free_extent_map(merge);
273 		}
274 	}
275 
276 	rb = rb_next(&em->rb_node);
277 	if (rb)
278 		merge = rb_entry(rb, struct extent_map, rb_node);
279 	if (rb && mergable_maps(em, merge)) {
280 		em->len += merge->len;
281 		em->block_len += merge->block_len;
282 		rb_erase_cached(&merge->rb_node, &tree->map);
283 		RB_CLEAR_NODE(&merge->rb_node);
284 		em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
285 		em->generation = max(em->generation, merge->generation);
286 		set_bit(EXTENT_FLAG_MERGED, &em->flags);
287 		free_extent_map(merge);
288 	}
289 }
290 
291 /**
292  * unpin_extent_cache - unpin an extent from the cache
293  * @tree:	tree to unpin the extent in
294  * @start:	logical offset in the file
295  * @len:	length of the extent
296  * @gen:	generation that this extent has been modified in
297  *
298  * Called after an extent has been written to disk properly.  Set the generation
299  * to the generation that actually added the file item to the inode so we know
300  * we need to sync this extent when we call fsync().
301  */
unpin_extent_cache(struct extent_map_tree * tree,u64 start,u64 len,u64 gen)302 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
303 		       u64 gen)
304 {
305 	int ret = 0;
306 	struct extent_map *em;
307 	bool prealloc = false;
308 
309 	write_lock(&tree->lock);
310 	em = lookup_extent_mapping(tree, start, len);
311 
312 	WARN_ON(!em || em->start != start);
313 
314 	if (!em)
315 		goto out;
316 
317 	em->generation = gen;
318 	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
319 	em->mod_start = em->start;
320 	em->mod_len = em->len;
321 
322 	if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
323 		prealloc = true;
324 		clear_bit(EXTENT_FLAG_FILLING, &em->flags);
325 	}
326 
327 	try_merge_map(tree, em);
328 
329 	if (prealloc) {
330 		em->mod_start = em->start;
331 		em->mod_len = em->len;
332 	}
333 
334 	free_extent_map(em);
335 out:
336 	write_unlock(&tree->lock);
337 	return ret;
338 
339 }
340 
clear_em_logging(struct extent_map_tree * tree,struct extent_map * em)341 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
342 {
343 	lockdep_assert_held_write(&tree->lock);
344 
345 	clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
346 	if (extent_map_in_tree(em))
347 		try_merge_map(tree, em);
348 }
349 
setup_extent_mapping(struct extent_map_tree * tree,struct extent_map * em,int modified)350 static inline void setup_extent_mapping(struct extent_map_tree *tree,
351 					struct extent_map *em,
352 					int modified)
353 {
354 	refcount_inc(&em->refs);
355 	em->mod_start = em->start;
356 	em->mod_len = em->len;
357 
358 	if (modified)
359 		list_move(&em->list, &tree->modified_extents);
360 	else
361 		try_merge_map(tree, em);
362 }
363 
extent_map_device_set_bits(struct extent_map * em,unsigned bits)364 static void extent_map_device_set_bits(struct extent_map *em, unsigned bits)
365 {
366 	struct map_lookup *map = em->map_lookup;
367 	u64 stripe_size = em->orig_block_len;
368 	int i;
369 
370 	for (i = 0; i < map->num_stripes; i++) {
371 		struct btrfs_io_stripe *stripe = &map->stripes[i];
372 		struct btrfs_device *device = stripe->dev;
373 
374 		set_extent_bits_nowait(&device->alloc_state, stripe->physical,
375 				 stripe->physical + stripe_size - 1, bits);
376 	}
377 }
378 
extent_map_device_clear_bits(struct extent_map * em,unsigned bits)379 static void extent_map_device_clear_bits(struct extent_map *em, unsigned bits)
380 {
381 	struct map_lookup *map = em->map_lookup;
382 	u64 stripe_size = em->orig_block_len;
383 	int i;
384 
385 	for (i = 0; i < map->num_stripes; i++) {
386 		struct btrfs_io_stripe *stripe = &map->stripes[i];
387 		struct btrfs_device *device = stripe->dev;
388 
389 		__clear_extent_bit(&device->alloc_state, stripe->physical,
390 				   stripe->physical + stripe_size - 1, bits,
391 				   NULL, GFP_NOWAIT, NULL);
392 	}
393 }
394 
395 /**
396  * Add new extent map to the extent tree
397  *
398  * @tree:	tree to insert new map in
399  * @em:		map to insert
400  * @modified:	indicate whether the given @em should be added to the
401  *	        modified list, which indicates the extent needs to be logged
402  *
403  * Insert @em into @tree or perform a simple forward/backward merge with
404  * existing mappings.  The extent_map struct passed in will be inserted
405  * into the tree directly, with an additional reference taken, or a
406  * reference dropped if the merge attempt was successful.
407  */
add_extent_mapping(struct extent_map_tree * tree,struct extent_map * em,int modified)408 int add_extent_mapping(struct extent_map_tree *tree,
409 		       struct extent_map *em, int modified)
410 {
411 	int ret = 0;
412 
413 	lockdep_assert_held_write(&tree->lock);
414 
415 	ret = tree_insert(&tree->map, em);
416 	if (ret)
417 		goto out;
418 
419 	setup_extent_mapping(tree, em, modified);
420 	if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags)) {
421 		extent_map_device_set_bits(em, CHUNK_ALLOCATED);
422 		extent_map_device_clear_bits(em, CHUNK_TRIMMED);
423 	}
424 out:
425 	return ret;
426 }
427 
428 static struct extent_map *
__lookup_extent_mapping(struct extent_map_tree * tree,u64 start,u64 len,int strict)429 __lookup_extent_mapping(struct extent_map_tree *tree,
430 			u64 start, u64 len, int strict)
431 {
432 	struct extent_map *em;
433 	struct rb_node *rb_node;
434 	struct rb_node *prev_or_next = NULL;
435 	u64 end = range_end(start, len);
436 
437 	rb_node = __tree_search(&tree->map.rb_root, start, &prev_or_next);
438 	if (!rb_node) {
439 		if (prev_or_next)
440 			rb_node = prev_or_next;
441 		else
442 			return NULL;
443 	}
444 
445 	em = rb_entry(rb_node, struct extent_map, rb_node);
446 
447 	if (strict && !(end > em->start && start < extent_map_end(em)))
448 		return NULL;
449 
450 	refcount_inc(&em->refs);
451 	return em;
452 }
453 
454 /**
455  * lookup_extent_mapping - lookup extent_map
456  * @tree:	tree to lookup in
457  * @start:	byte offset to start the search
458  * @len:	length of the lookup range
459  *
460  * Find and return the first extent_map struct in @tree that intersects the
461  * [start, len] range.  There may be additional objects in the tree that
462  * intersect, so check the object returned carefully to make sure that no
463  * additional lookups are needed.
464  */
lookup_extent_mapping(struct extent_map_tree * tree,u64 start,u64 len)465 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
466 					 u64 start, u64 len)
467 {
468 	return __lookup_extent_mapping(tree, start, len, 1);
469 }
470 
471 /**
472  * search_extent_mapping - find a nearby extent map
473  * @tree:	tree to lookup in
474  * @start:	byte offset to start the search
475  * @len:	length of the lookup range
476  *
477  * Find and return the first extent_map struct in @tree that intersects the
478  * [start, len] range.
479  *
480  * If one can't be found, any nearby extent may be returned
481  */
search_extent_mapping(struct extent_map_tree * tree,u64 start,u64 len)482 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
483 					 u64 start, u64 len)
484 {
485 	return __lookup_extent_mapping(tree, start, len, 0);
486 }
487 
488 /**
489  * remove_extent_mapping - removes an extent_map from the extent tree
490  * @tree:	extent tree to remove from
491  * @em:		extent map being removed
492  *
493  * Removes @em from @tree.  No reference counts are dropped, and no checks
494  * are done to see if the range is in use
495  */
remove_extent_mapping(struct extent_map_tree * tree,struct extent_map * em)496 void remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
497 {
498 	lockdep_assert_held_write(&tree->lock);
499 
500 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
501 	rb_erase_cached(&em->rb_node, &tree->map);
502 	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
503 		list_del_init(&em->list);
504 	if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
505 		extent_map_device_clear_bits(em, CHUNK_ALLOCATED);
506 	RB_CLEAR_NODE(&em->rb_node);
507 }
508 
replace_extent_mapping(struct extent_map_tree * tree,struct extent_map * cur,struct extent_map * new,int modified)509 void replace_extent_mapping(struct extent_map_tree *tree,
510 			    struct extent_map *cur,
511 			    struct extent_map *new,
512 			    int modified)
513 {
514 	lockdep_assert_held_write(&tree->lock);
515 
516 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
517 	ASSERT(extent_map_in_tree(cur));
518 	if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
519 		list_del_init(&cur->list);
520 	rb_replace_node_cached(&cur->rb_node, &new->rb_node, &tree->map);
521 	RB_CLEAR_NODE(&cur->rb_node);
522 
523 	setup_extent_mapping(tree, new, modified);
524 }
525 
next_extent_map(struct extent_map * em)526 static struct extent_map *next_extent_map(struct extent_map *em)
527 {
528 	struct rb_node *next;
529 
530 	next = rb_next(&em->rb_node);
531 	if (!next)
532 		return NULL;
533 	return container_of(next, struct extent_map, rb_node);
534 }
535 
prev_extent_map(struct extent_map * em)536 static struct extent_map *prev_extent_map(struct extent_map *em)
537 {
538 	struct rb_node *prev;
539 
540 	prev = rb_prev(&em->rb_node);
541 	if (!prev)
542 		return NULL;
543 	return container_of(prev, struct extent_map, rb_node);
544 }
545 
546 /*
547  * Helper for btrfs_get_extent.  Given an existing extent in the tree,
548  * the existing extent is the nearest extent to map_start,
549  * and an extent that you want to insert, deal with overlap and insert
550  * the best fitted new extent into the tree.
551  */
merge_extent_mapping(struct extent_map_tree * em_tree,struct extent_map * existing,struct extent_map * em,u64 map_start)552 static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
553 					 struct extent_map *existing,
554 					 struct extent_map *em,
555 					 u64 map_start)
556 {
557 	struct extent_map *prev;
558 	struct extent_map *next;
559 	u64 start;
560 	u64 end;
561 	u64 start_diff;
562 
563 	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
564 
565 	if (existing->start > map_start) {
566 		next = existing;
567 		prev = prev_extent_map(next);
568 	} else {
569 		prev = existing;
570 		next = next_extent_map(prev);
571 	}
572 
573 	start = prev ? extent_map_end(prev) : em->start;
574 	start = max_t(u64, start, em->start);
575 	end = next ? next->start : extent_map_end(em);
576 	end = min_t(u64, end, extent_map_end(em));
577 	start_diff = start - em->start;
578 	em->start = start;
579 	em->len = end - start;
580 	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
581 	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
582 		em->block_start += start_diff;
583 		em->block_len = em->len;
584 	}
585 	return add_extent_mapping(em_tree, em, 0);
586 }
587 
588 /**
589  * Add extent mapping into em_tree
590  *
591  * @fs_info:  the filesystem
592  * @em_tree:  extent tree into which we want to insert the extent mapping
593  * @em_in:    extent we are inserting
594  * @start:    start of the logical range btrfs_get_extent() is requesting
595  * @len:      length of the logical range btrfs_get_extent() is requesting
596  *
597  * Note that @em_in's range may be different from [start, start+len),
598  * but they must be overlapped.
599  *
600  * Insert @em_in into @em_tree. In case there is an overlapping range, handle
601  * the -EEXIST by either:
602  * a) Returning the existing extent in @em_in if @start is within the
603  *    existing em.
604  * b) Merge the existing extent with @em_in passed in.
605  *
606  * Return 0 on success, otherwise -EEXIST.
607  *
608  */
btrfs_add_extent_mapping(struct btrfs_fs_info * fs_info,struct extent_map_tree * em_tree,struct extent_map ** em_in,u64 start,u64 len)609 int btrfs_add_extent_mapping(struct btrfs_fs_info *fs_info,
610 			     struct extent_map_tree *em_tree,
611 			     struct extent_map **em_in, u64 start, u64 len)
612 {
613 	int ret;
614 	struct extent_map *em = *em_in;
615 
616 	ret = add_extent_mapping(em_tree, em, 0);
617 	/* it is possible that someone inserted the extent into the tree
618 	 * while we had the lock dropped.  It is also possible that
619 	 * an overlapping map exists in the tree
620 	 */
621 	if (ret == -EEXIST) {
622 		struct extent_map *existing;
623 
624 		ret = 0;
625 
626 		existing = search_extent_mapping(em_tree, start, len);
627 
628 		trace_btrfs_handle_em_exist(fs_info, existing, em, start, len);
629 
630 		/*
631 		 * existing will always be non-NULL, since there must be
632 		 * extent causing the -EEXIST.
633 		 */
634 		if (start >= existing->start &&
635 		    start < extent_map_end(existing)) {
636 			free_extent_map(em);
637 			*em_in = existing;
638 			ret = 0;
639 		} else {
640 			u64 orig_start = em->start;
641 			u64 orig_len = em->len;
642 
643 			/*
644 			 * The existing extent map is the one nearest to
645 			 * the [start, start + len) range which overlaps
646 			 */
647 			ret = merge_extent_mapping(em_tree, existing,
648 						   em, start);
649 			if (ret) {
650 				free_extent_map(em);
651 				*em_in = NULL;
652 				WARN_ONCE(ret,
653 "unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
654 					  ret, existing->start, existing->len,
655 					  orig_start, orig_len);
656 			}
657 			free_extent_map(existing);
658 		}
659 	}
660 
661 	ASSERT(ret == 0 || ret == -EEXIST);
662 	return ret;
663 }
664 
665 /*
666  * Drop all extent maps from a tree in the fastest possible way, rescheduling
667  * if needed. This avoids searching the tree, from the root down to the first
668  * extent map, before each deletion.
669  */
drop_all_extent_maps_fast(struct extent_map_tree * tree)670 static void drop_all_extent_maps_fast(struct extent_map_tree *tree)
671 {
672 	write_lock(&tree->lock);
673 	while (!RB_EMPTY_ROOT(&tree->map.rb_root)) {
674 		struct extent_map *em;
675 		struct rb_node *node;
676 
677 		node = rb_first_cached(&tree->map);
678 		em = rb_entry(node, struct extent_map, rb_node);
679 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
680 		clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
681 		remove_extent_mapping(tree, em);
682 		free_extent_map(em);
683 		cond_resched_rwlock_write(&tree->lock);
684 	}
685 	write_unlock(&tree->lock);
686 }
687 
688 /*
689  * Drop all extent maps in a given range.
690  *
691  * @inode:       The target inode.
692  * @start:       Start offset of the range.
693  * @end:         End offset of the range (inclusive value).
694  * @skip_pinned: Indicate if pinned extent maps should be ignored or not.
695  *
696  * This drops all the extent maps that intersect the given range [@start, @end].
697  * Extent maps that partially overlap the range and extend behind or beyond it,
698  * are split.
699  * The caller should have locked an appropriate file range in the inode's io
700  * tree before calling this function.
701  */
btrfs_drop_extent_map_range(struct btrfs_inode * inode,u64 start,u64 end,bool skip_pinned)702 void btrfs_drop_extent_map_range(struct btrfs_inode *inode, u64 start, u64 end,
703 				 bool skip_pinned)
704 {
705 	struct extent_map *split;
706 	struct extent_map *split2;
707 	struct extent_map *em;
708 	struct extent_map_tree *em_tree = &inode->extent_tree;
709 	u64 len = end - start + 1;
710 
711 	WARN_ON(end < start);
712 	if (end == (u64)-1) {
713 		if (start == 0 && !skip_pinned) {
714 			drop_all_extent_maps_fast(em_tree);
715 			return;
716 		}
717 		len = (u64)-1;
718 	} else {
719 		/* Make end offset exclusive for use in the loop below. */
720 		end++;
721 	}
722 
723 	/*
724 	 * It's ok if we fail to allocate the extent maps, see the comment near
725 	 * the bottom of the loop below. We only need two spare extent maps in
726 	 * the worst case, where the first extent map that intersects our range
727 	 * starts before the range and the last extent map that intersects our
728 	 * range ends after our range (and they might be the same extent map),
729 	 * because we need to split those two extent maps at the boundaries.
730 	 */
731 	split = alloc_extent_map();
732 	split2 = alloc_extent_map();
733 
734 	write_lock(&em_tree->lock);
735 	em = lookup_extent_mapping(em_tree, start, len);
736 
737 	while (em) {
738 		/* extent_map_end() returns exclusive value (last byte + 1). */
739 		const u64 em_end = extent_map_end(em);
740 		struct extent_map *next_em = NULL;
741 		u64 gen;
742 		unsigned long flags;
743 		bool modified;
744 		bool compressed;
745 
746 		if (em_end < end) {
747 			next_em = next_extent_map(em);
748 			if (next_em) {
749 				if (next_em->start < end)
750 					refcount_inc(&next_em->refs);
751 				else
752 					next_em = NULL;
753 			}
754 		}
755 
756 		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
757 			start = em_end;
758 			if (end != (u64)-1)
759 				len = start + len - em_end;
760 			goto next;
761 		}
762 
763 		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
764 		clear_bit(EXTENT_FLAG_LOGGING, &flags);
765 		modified = !list_empty(&em->list);
766 
767 		/*
768 		 * The extent map does not cross our target range, so no need to
769 		 * split it, we can remove it directly.
770 		 */
771 		if (em->start >= start && em_end <= end)
772 			goto remove_em;
773 
774 		flags = em->flags;
775 		gen = em->generation;
776 		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
777 
778 		if (em->start < start) {
779 			if (!split) {
780 				split = split2;
781 				split2 = NULL;
782 				if (!split)
783 					goto remove_em;
784 			}
785 			split->start = em->start;
786 			split->len = start - em->start;
787 
788 			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
789 				split->orig_start = em->orig_start;
790 				split->block_start = em->block_start;
791 
792 				if (compressed)
793 					split->block_len = em->block_len;
794 				else
795 					split->block_len = split->len;
796 				split->orig_block_len = max(split->block_len,
797 						em->orig_block_len);
798 				split->ram_bytes = em->ram_bytes;
799 			} else {
800 				split->orig_start = split->start;
801 				split->block_len = 0;
802 				split->block_start = em->block_start;
803 				split->orig_block_len = 0;
804 				split->ram_bytes = split->len;
805 			}
806 
807 			split->generation = gen;
808 			split->flags = flags;
809 			split->compress_type = em->compress_type;
810 			replace_extent_mapping(em_tree, em, split, modified);
811 			free_extent_map(split);
812 			split = split2;
813 			split2 = NULL;
814 		}
815 		if (em_end > end) {
816 			if (!split) {
817 				split = split2;
818 				split2 = NULL;
819 				if (!split)
820 					goto remove_em;
821 			}
822 			split->start = start + len;
823 			split->len = em_end - (start + len);
824 			split->block_start = em->block_start;
825 			split->flags = flags;
826 			split->compress_type = em->compress_type;
827 			split->generation = gen;
828 
829 			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
830 				split->orig_block_len = max(em->block_len,
831 						    em->orig_block_len);
832 
833 				split->ram_bytes = em->ram_bytes;
834 				if (compressed) {
835 					split->block_len = em->block_len;
836 					split->orig_start = em->orig_start;
837 				} else {
838 					const u64 diff = start + len - em->start;
839 
840 					split->block_len = split->len;
841 					split->block_start += diff;
842 					split->orig_start = em->orig_start;
843 				}
844 			} else {
845 				split->ram_bytes = split->len;
846 				split->orig_start = split->start;
847 				split->block_len = 0;
848 				split->orig_block_len = 0;
849 			}
850 
851 			if (extent_map_in_tree(em)) {
852 				replace_extent_mapping(em_tree, em, split,
853 						       modified);
854 			} else {
855 				int ret;
856 
857 				ret = add_extent_mapping(em_tree, split,
858 							 modified);
859 				/* Logic error, shouldn't happen. */
860 				ASSERT(ret == 0);
861 				if (WARN_ON(ret != 0) && modified)
862 					btrfs_set_inode_full_sync(inode);
863 			}
864 			free_extent_map(split);
865 			split = NULL;
866 		}
867 remove_em:
868 		if (extent_map_in_tree(em)) {
869 			/*
870 			 * If the extent map is still in the tree it means that
871 			 * either of the following is true:
872 			 *
873 			 * 1) It fits entirely in our range (doesn't end beyond
874 			 *    it or starts before it);
875 			 *
876 			 * 2) It starts before our range and/or ends after our
877 			 *    range, and we were not able to allocate the extent
878 			 *    maps for split operations, @split and @split2.
879 			 *
880 			 * If we are at case 2) then we just remove the entire
881 			 * extent map - this is fine since if anyone needs it to
882 			 * access the subranges outside our range, will just
883 			 * load it again from the subvolume tree's file extent
884 			 * item. However if the extent map was in the list of
885 			 * modified extents, then we must mark the inode for a
886 			 * full fsync, otherwise a fast fsync will miss this
887 			 * extent if it's new and needs to be logged.
888 			 */
889 			if ((em->start < start || em_end > end) && modified) {
890 				ASSERT(!split);
891 				btrfs_set_inode_full_sync(inode);
892 			}
893 			remove_extent_mapping(em_tree, em);
894 		}
895 
896 		/*
897 		 * Once for the tree reference (we replaced or removed the
898 		 * extent map from the tree).
899 		 */
900 		free_extent_map(em);
901 next:
902 		/* Once for us (for our lookup reference). */
903 		free_extent_map(em);
904 
905 		em = next_em;
906 	}
907 
908 	write_unlock(&em_tree->lock);
909 
910 	free_extent_map(split);
911 	free_extent_map(split2);
912 }
913 
914 /*
915  * Replace a range in the inode's extent map tree with a new extent map.
916  *
917  * @inode:      The target inode.
918  * @new_em:     The new extent map to add to the inode's extent map tree.
919  * @modified:   Indicate if the new extent map should be added to the list of
920  *              modified extents (for fast fsync tracking).
921  *
922  * Drops all the extent maps in the inode's extent map tree that intersect the
923  * range of the new extent map and adds the new extent map to the tree.
924  * The caller should have locked an appropriate file range in the inode's io
925  * tree before calling this function.
926  */
btrfs_replace_extent_map_range(struct btrfs_inode * inode,struct extent_map * new_em,bool modified)927 int btrfs_replace_extent_map_range(struct btrfs_inode *inode,
928 				   struct extent_map *new_em,
929 				   bool modified)
930 {
931 	const u64 end = new_em->start + new_em->len - 1;
932 	struct extent_map_tree *tree = &inode->extent_tree;
933 	int ret;
934 
935 	ASSERT(!extent_map_in_tree(new_em));
936 
937 	/*
938 	 * The caller has locked an appropriate file range in the inode's io
939 	 * tree, but getting -EEXIST when adding the new extent map can still
940 	 * happen in case there are extents that partially cover the range, and
941 	 * this is due to two tasks operating on different parts of the extent.
942 	 * See commit 18e83ac75bfe67 ("Btrfs: fix unexpected EEXIST from
943 	 * btrfs_get_extent") for an example and details.
944 	 */
945 	do {
946 		btrfs_drop_extent_map_range(inode, new_em->start, end, false);
947 		write_lock(&tree->lock);
948 		ret = add_extent_mapping(tree, new_em, modified);
949 		write_unlock(&tree->lock);
950 	} while (ret == -EEXIST);
951 
952 	return ret;
953 }
954