xref: /openbmc/linux/fs/btrfs/extent_map.c (revision bdeeed09)
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 "extent_map.h"
8 #include "compression.h"
9 
10 
11 static struct kmem_cache *extent_map_cache;
12 
13 int __init extent_map_init(void)
14 {
15 	extent_map_cache = kmem_cache_create("btrfs_extent_map",
16 			sizeof(struct extent_map), 0,
17 			SLAB_MEM_SPREAD, NULL);
18 	if (!extent_map_cache)
19 		return -ENOMEM;
20 	return 0;
21 }
22 
23 void __cold extent_map_exit(void)
24 {
25 	kmem_cache_destroy(extent_map_cache);
26 }
27 
28 /**
29  * extent_map_tree_init - initialize extent map tree
30  * @tree:		tree to initialize
31  *
32  * Initialize the extent tree @tree.  Should be called for each new inode
33  * or other user of the extent_map interface.
34  */
35 void extent_map_tree_init(struct extent_map_tree *tree)
36 {
37 	tree->map = RB_ROOT;
38 	INIT_LIST_HEAD(&tree->modified_extents);
39 	rwlock_init(&tree->lock);
40 }
41 
42 /**
43  * alloc_extent_map - allocate new extent map structure
44  *
45  * Allocate a new extent_map structure.  The new structure is
46  * returned with a reference count of one and needs to be
47  * freed using free_extent_map()
48  */
49 struct extent_map *alloc_extent_map(void)
50 {
51 	struct extent_map *em;
52 	em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
53 	if (!em)
54 		return NULL;
55 	RB_CLEAR_NODE(&em->rb_node);
56 	em->flags = 0;
57 	em->compress_type = BTRFS_COMPRESS_NONE;
58 	em->generation = 0;
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  */
71 void free_extent_map(struct extent_map *em)
72 {
73 	if (!em)
74 		return;
75 	WARN_ON(refcount_read(&em->refs) == 0);
76 	if (refcount_dec_and_test(&em->refs)) {
77 		WARN_ON(extent_map_in_tree(em));
78 		WARN_ON(!list_empty(&em->list));
79 		if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
80 			kfree(em->map_lookup);
81 		kmem_cache_free(extent_map_cache, em);
82 	}
83 }
84 
85 /* simple helper to do math around the end of an extent, handling wrap */
86 static u64 range_end(u64 start, u64 len)
87 {
88 	if (start + len < start)
89 		return (u64)-1;
90 	return start + len;
91 }
92 
93 static int tree_insert(struct rb_root *root, struct extent_map *em)
94 {
95 	struct rb_node **p = &root->rb_node;
96 	struct rb_node *parent = NULL;
97 	struct extent_map *entry = NULL;
98 	struct rb_node *orig_parent = NULL;
99 	u64 end = range_end(em->start, em->len);
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 		else
110 			return -EEXIST;
111 	}
112 
113 	orig_parent = parent;
114 	while (parent && em->start >= extent_map_end(entry)) {
115 		parent = rb_next(parent);
116 		entry = rb_entry(parent, struct extent_map, rb_node);
117 	}
118 	if (parent)
119 		if (end > entry->start && em->start < extent_map_end(entry))
120 			return -EEXIST;
121 
122 	parent = orig_parent;
123 	entry = rb_entry(parent, struct extent_map, rb_node);
124 	while (parent && em->start < entry->start) {
125 		parent = rb_prev(parent);
126 		entry = rb_entry(parent, struct extent_map, rb_node);
127 	}
128 	if (parent)
129 		if (end > entry->start && em->start < extent_map_end(entry))
130 			return -EEXIST;
131 
132 	rb_link_node(&em->rb_node, orig_parent, p);
133 	rb_insert_color(&em->rb_node, root);
134 	return 0;
135 }
136 
137 /*
138  * search through the tree for an extent_map with a given offset.  If
139  * it can't be found, try to find some neighboring extents
140  */
141 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
142 				     struct rb_node **prev_ret,
143 				     struct rb_node **next_ret)
144 {
145 	struct rb_node *n = root->rb_node;
146 	struct rb_node *prev = NULL;
147 	struct rb_node *orig_prev = NULL;
148 	struct extent_map *entry;
149 	struct extent_map *prev_entry = NULL;
150 
151 	while (n) {
152 		entry = rb_entry(n, struct extent_map, rb_node);
153 		prev = n;
154 		prev_entry = entry;
155 
156 		if (offset < entry->start)
157 			n = n->rb_left;
158 		else if (offset >= extent_map_end(entry))
159 			n = n->rb_right;
160 		else
161 			return n;
162 	}
163 
164 	if (prev_ret) {
165 		orig_prev = prev;
166 		while (prev && offset >= extent_map_end(prev_entry)) {
167 			prev = rb_next(prev);
168 			prev_entry = rb_entry(prev, struct extent_map, rb_node);
169 		}
170 		*prev_ret = prev;
171 		prev = orig_prev;
172 	}
173 
174 	if (next_ret) {
175 		prev_entry = rb_entry(prev, struct extent_map, rb_node);
176 		while (prev && offset < prev_entry->start) {
177 			prev = rb_prev(prev);
178 			prev_entry = rb_entry(prev, struct extent_map, rb_node);
179 		}
180 		*next_ret = prev;
181 	}
182 	return NULL;
183 }
184 
185 /* check to see if two extent_map structs are adjacent and safe to merge */
186 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
187 {
188 	if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
189 		return 0;
190 
191 	/*
192 	 * don't merge compressed extents, we need to know their
193 	 * actual size
194 	 */
195 	if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
196 		return 0;
197 
198 	if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
199 	    test_bit(EXTENT_FLAG_LOGGING, &next->flags))
200 		return 0;
201 
202 	/*
203 	 * We don't want to merge stuff that hasn't been written to the log yet
204 	 * since it may not reflect exactly what is on disk, and that would be
205 	 * bad.
206 	 */
207 	if (!list_empty(&prev->list) || !list_empty(&next->list))
208 		return 0;
209 
210 	if (extent_map_end(prev) == next->start &&
211 	    prev->flags == next->flags &&
212 	    prev->bdev == next->bdev &&
213 	    ((next->block_start == EXTENT_MAP_HOLE &&
214 	      prev->block_start == EXTENT_MAP_HOLE) ||
215 	     (next->block_start == EXTENT_MAP_INLINE &&
216 	      prev->block_start == EXTENT_MAP_INLINE) ||
217 	     (next->block_start == EXTENT_MAP_DELALLOC &&
218 	      prev->block_start == EXTENT_MAP_DELALLOC) ||
219 	     (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
220 	      next->block_start == extent_map_block_end(prev)))) {
221 		return 1;
222 	}
223 	return 0;
224 }
225 
226 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
227 {
228 	struct extent_map *merge = NULL;
229 	struct rb_node *rb;
230 
231 	if (em->start != 0) {
232 		rb = rb_prev(&em->rb_node);
233 		if (rb)
234 			merge = rb_entry(rb, struct extent_map, rb_node);
235 		if (rb && mergable_maps(merge, em)) {
236 			em->start = merge->start;
237 			em->orig_start = merge->orig_start;
238 			em->len += merge->len;
239 			em->block_len += merge->block_len;
240 			em->block_start = merge->block_start;
241 			em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
242 			em->mod_start = merge->mod_start;
243 			em->generation = max(em->generation, merge->generation);
244 
245 			rb_erase(&merge->rb_node, &tree->map);
246 			RB_CLEAR_NODE(&merge->rb_node);
247 			free_extent_map(merge);
248 		}
249 	}
250 
251 	rb = rb_next(&em->rb_node);
252 	if (rb)
253 		merge = rb_entry(rb, struct extent_map, rb_node);
254 	if (rb && mergable_maps(em, merge)) {
255 		em->len += merge->len;
256 		em->block_len += merge->block_len;
257 		rb_erase(&merge->rb_node, &tree->map);
258 		RB_CLEAR_NODE(&merge->rb_node);
259 		em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
260 		em->generation = max(em->generation, merge->generation);
261 		free_extent_map(merge);
262 	}
263 }
264 
265 /**
266  * unpin_extent_cache - unpin an extent from the cache
267  * @tree:	tree to unpin the extent in
268  * @start:	logical offset in the file
269  * @len:	length of the extent
270  * @gen:	generation that this extent has been modified in
271  *
272  * Called after an extent has been written to disk properly.  Set the generation
273  * to the generation that actually added the file item to the inode so we know
274  * we need to sync this extent when we call fsync().
275  */
276 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
277 		       u64 gen)
278 {
279 	int ret = 0;
280 	struct extent_map *em;
281 	bool prealloc = false;
282 
283 	write_lock(&tree->lock);
284 	em = lookup_extent_mapping(tree, start, len);
285 
286 	WARN_ON(!em || em->start != start);
287 
288 	if (!em)
289 		goto out;
290 
291 	em->generation = gen;
292 	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
293 	em->mod_start = em->start;
294 	em->mod_len = em->len;
295 
296 	if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
297 		prealloc = true;
298 		clear_bit(EXTENT_FLAG_FILLING, &em->flags);
299 	}
300 
301 	try_merge_map(tree, em);
302 
303 	if (prealloc) {
304 		em->mod_start = em->start;
305 		em->mod_len = em->len;
306 	}
307 
308 	free_extent_map(em);
309 out:
310 	write_unlock(&tree->lock);
311 	return ret;
312 
313 }
314 
315 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
316 {
317 	clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
318 	if (extent_map_in_tree(em))
319 		try_merge_map(tree, em);
320 }
321 
322 static inline void setup_extent_mapping(struct extent_map_tree *tree,
323 					struct extent_map *em,
324 					int modified)
325 {
326 	refcount_inc(&em->refs);
327 	em->mod_start = em->start;
328 	em->mod_len = em->len;
329 
330 	if (modified)
331 		list_move(&em->list, &tree->modified_extents);
332 	else
333 		try_merge_map(tree, em);
334 }
335 
336 /**
337  * add_extent_mapping - add new extent map to the extent tree
338  * @tree:	tree to insert new map in
339  * @em:		map to insert
340  *
341  * Insert @em into @tree or perform a simple forward/backward merge with
342  * existing mappings.  The extent_map struct passed in will be inserted
343  * into the tree directly, with an additional reference taken, or a
344  * reference dropped if the merge attempt was successful.
345  */
346 int add_extent_mapping(struct extent_map_tree *tree,
347 		       struct extent_map *em, int modified)
348 {
349 	int ret = 0;
350 
351 	ret = tree_insert(&tree->map, em);
352 	if (ret)
353 		goto out;
354 
355 	setup_extent_mapping(tree, em, modified);
356 out:
357 	return ret;
358 }
359 
360 static struct extent_map *
361 __lookup_extent_mapping(struct extent_map_tree *tree,
362 			u64 start, u64 len, int strict)
363 {
364 	struct extent_map *em;
365 	struct rb_node *rb_node;
366 	struct rb_node *prev = NULL;
367 	struct rb_node *next = NULL;
368 	u64 end = range_end(start, len);
369 
370 	rb_node = __tree_search(&tree->map, start, &prev, &next);
371 	if (!rb_node) {
372 		if (prev)
373 			rb_node = prev;
374 		else if (next)
375 			rb_node = next;
376 		else
377 			return NULL;
378 	}
379 
380 	em = rb_entry(rb_node, struct extent_map, rb_node);
381 
382 	if (strict && !(end > em->start && start < extent_map_end(em)))
383 		return NULL;
384 
385 	refcount_inc(&em->refs);
386 	return em;
387 }
388 
389 /**
390  * lookup_extent_mapping - lookup extent_map
391  * @tree:	tree to lookup in
392  * @start:	byte offset to start the search
393  * @len:	length of the lookup range
394  *
395  * Find and return the first extent_map struct in @tree that intersects the
396  * [start, len] range.  There may be additional objects in the tree that
397  * intersect, so check the object returned carefully to make sure that no
398  * additional lookups are needed.
399  */
400 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
401 					 u64 start, u64 len)
402 {
403 	return __lookup_extent_mapping(tree, start, len, 1);
404 }
405 
406 /**
407  * search_extent_mapping - find a nearby extent map
408  * @tree:	tree to lookup in
409  * @start:	byte offset to start the search
410  * @len:	length of the lookup range
411  *
412  * Find and return the first extent_map struct in @tree that intersects the
413  * [start, len] range.
414  *
415  * If one can't be found, any nearby extent may be returned
416  */
417 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
418 					 u64 start, u64 len)
419 {
420 	return __lookup_extent_mapping(tree, start, len, 0);
421 }
422 
423 /**
424  * remove_extent_mapping - removes an extent_map from the extent tree
425  * @tree:	extent tree to remove from
426  * @em:		extent map being removed
427  *
428  * Removes @em from @tree.  No reference counts are dropped, and no checks
429  * are done to see if the range is in use
430  */
431 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
432 {
433 	int ret = 0;
434 
435 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
436 	rb_erase(&em->rb_node, &tree->map);
437 	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
438 		list_del_init(&em->list);
439 	RB_CLEAR_NODE(&em->rb_node);
440 	return ret;
441 }
442 
443 void replace_extent_mapping(struct extent_map_tree *tree,
444 			    struct extent_map *cur,
445 			    struct extent_map *new,
446 			    int modified)
447 {
448 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
449 	ASSERT(extent_map_in_tree(cur));
450 	if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
451 		list_del_init(&cur->list);
452 	rb_replace_node(&cur->rb_node, &new->rb_node, &tree->map);
453 	RB_CLEAR_NODE(&cur->rb_node);
454 
455 	setup_extent_mapping(tree, new, modified);
456 }
457 
458 static struct extent_map *next_extent_map(struct extent_map *em)
459 {
460 	struct rb_node *next;
461 
462 	next = rb_next(&em->rb_node);
463 	if (!next)
464 		return NULL;
465 	return container_of(next, struct extent_map, rb_node);
466 }
467 
468 static struct extent_map *prev_extent_map(struct extent_map *em)
469 {
470 	struct rb_node *prev;
471 
472 	prev = rb_prev(&em->rb_node);
473 	if (!prev)
474 		return NULL;
475 	return container_of(prev, struct extent_map, rb_node);
476 }
477 
478 /* helper for btfs_get_extent.  Given an existing extent in the tree,
479  * the existing extent is the nearest extent to map_start,
480  * and an extent that you want to insert, deal with overlap and insert
481  * the best fitted new extent into the tree.
482  */
483 static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
484 					 struct extent_map *existing,
485 					 struct extent_map *em,
486 					 u64 map_start)
487 {
488 	struct extent_map *prev;
489 	struct extent_map *next;
490 	u64 start;
491 	u64 end;
492 	u64 start_diff;
493 
494 	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
495 
496 	if (existing->start > map_start) {
497 		next = existing;
498 		prev = prev_extent_map(next);
499 	} else {
500 		prev = existing;
501 		next = next_extent_map(prev);
502 	}
503 
504 	start = prev ? extent_map_end(prev) : em->start;
505 	start = max_t(u64, start, em->start);
506 	end = next ? next->start : extent_map_end(em);
507 	end = min_t(u64, end, extent_map_end(em));
508 	start_diff = start - em->start;
509 	em->start = start;
510 	em->len = end - start;
511 	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
512 	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
513 		em->block_start += start_diff;
514 		em->block_len = em->len;
515 	}
516 	return add_extent_mapping(em_tree, em, 0);
517 }
518 
519 /**
520  * btrfs_add_extent_mapping - add extent mapping into em_tree
521  * @em_tree - the extent tree into which we want to insert the extent mapping
522  * @em_in   - extent we are inserting
523  * @start   - start of the logical range btrfs_get_extent() is requesting
524  * @len     - length of the logical range btrfs_get_extent() is requesting
525  *
526  * Note that @em_in's range may be different from [start, start+len),
527  * but they must be overlapped.
528  *
529  * Insert @em_in into @em_tree. In case there is an overlapping range, handle
530  * the -EEXIST by either:
531  * a) Returning the existing extent in @em_in if @start is within the
532  *    existing em.
533  * b) Merge the existing extent with @em_in passed in.
534  *
535  * Return 0 on success, otherwise -EEXIST.
536  *
537  */
538 int btrfs_add_extent_mapping(struct extent_map_tree *em_tree,
539 			     struct extent_map **em_in, u64 start, u64 len)
540 {
541 	int ret;
542 	struct extent_map *em = *em_in;
543 
544 	ret = add_extent_mapping(em_tree, em, 0);
545 	/* it is possible that someone inserted the extent into the tree
546 	 * while we had the lock dropped.  It is also possible that
547 	 * an overlapping map exists in the tree
548 	 */
549 	if (ret == -EEXIST) {
550 		struct extent_map *existing;
551 
552 		ret = 0;
553 
554 		existing = search_extent_mapping(em_tree, start, len);
555 
556 		trace_btrfs_handle_em_exist(existing, em, start, len);
557 
558 		/*
559 		 * existing will always be non-NULL, since there must be
560 		 * extent causing the -EEXIST.
561 		 */
562 		if (start >= existing->start &&
563 		    start < extent_map_end(existing)) {
564 			free_extent_map(em);
565 			*em_in = existing;
566 			ret = 0;
567 		} else {
568 			u64 orig_start = em->start;
569 			u64 orig_len = em->len;
570 
571 			/*
572 			 * The existing extent map is the one nearest to
573 			 * the [start, start + len) range which overlaps
574 			 */
575 			ret = merge_extent_mapping(em_tree, existing,
576 						   em, start);
577 			if (ret) {
578 				free_extent_map(em);
579 				*em_in = NULL;
580 				WARN_ONCE(ret,
581 "unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
582 					  ret, existing->start, existing->len,
583 					  orig_start, orig_len);
584 			}
585 			free_extent_map(existing);
586 		}
587 	}
588 
589 	ASSERT(ret == 0 || ret == -EEXIST);
590 	return ret;
591 }
592