xref: /linux/fs/btrfs/extent_map.c (revision 00a6d7b6762c27d441e9ac8faff36384bc0fc180)
1 #include <linux/err.h>
2 #include <linux/slab.h>
3 #include <linux/spinlock.h>
4 #include <linux/hardirq.h>
5 #include "ctree.h"
6 #include "extent_map.h"
7 
8 
9 static struct kmem_cache *extent_map_cache;
10 
11 int __init extent_map_init(void)
12 {
13 	extent_map_cache = kmem_cache_create("btrfs_extent_map",
14 			sizeof(struct extent_map), 0,
15 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
16 	if (!extent_map_cache)
17 		return -ENOMEM;
18 	return 0;
19 }
20 
21 void extent_map_exit(void)
22 {
23 	if (extent_map_cache)
24 		kmem_cache_destroy(extent_map_cache);
25 }
26 
27 /**
28  * extent_map_tree_init - initialize extent map tree
29  * @tree:		tree to initialize
30  *
31  * Initialize the extent tree @tree.  Should be called for each new inode
32  * or other user of the extent_map interface.
33  */
34 void extent_map_tree_init(struct extent_map_tree *tree)
35 {
36 	tree->map = RB_ROOT;
37 	INIT_LIST_HEAD(&tree->modified_extents);
38 	rwlock_init(&tree->lock);
39 }
40 
41 /**
42  * alloc_extent_map - allocate new extent map structure
43  *
44  * Allocate a new extent_map structure.  The new structure is
45  * returned with a reference count of one and needs to be
46  * freed using free_extent_map()
47  */
48 struct extent_map *alloc_extent_map(void)
49 {
50 	struct extent_map *em;
51 	em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
52 	if (!em)
53 		return NULL;
54 	RB_CLEAR_NODE(&em->rb_node);
55 	em->flags = 0;
56 	em->compress_type = BTRFS_COMPRESS_NONE;
57 	em->generation = 0;
58 	atomic_set(&em->refs, 1);
59 	INIT_LIST_HEAD(&em->list);
60 	return em;
61 }
62 
63 /**
64  * free_extent_map - drop reference count of an extent_map
65  * @em:		extent map beeing releasead
66  *
67  * Drops the reference out on @em by one and free the structure
68  * if the reference count hits zero.
69  */
70 void free_extent_map(struct extent_map *em)
71 {
72 	if (!em)
73 		return;
74 	WARN_ON(atomic_read(&em->refs) == 0);
75 	if (atomic_dec_and_test(&em->refs)) {
76 		WARN_ON(extent_map_in_tree(em));
77 		WARN_ON(!list_empty(&em->list));
78 		kmem_cache_free(extent_map_cache, em);
79 	}
80 }
81 
82 /* simple helper to do math around the end of an extent, handling wrap */
83 static u64 range_end(u64 start, u64 len)
84 {
85 	if (start + len < start)
86 		return (u64)-1;
87 	return start + len;
88 }
89 
90 static int tree_insert(struct rb_root *root, struct extent_map *em)
91 {
92 	struct rb_node **p = &root->rb_node;
93 	struct rb_node *parent = NULL;
94 	struct extent_map *entry = NULL;
95 	struct rb_node *orig_parent = NULL;
96 	u64 end = range_end(em->start, em->len);
97 
98 	while (*p) {
99 		parent = *p;
100 		entry = rb_entry(parent, struct extent_map, rb_node);
101 
102 		if (em->start < entry->start)
103 			p = &(*p)->rb_left;
104 		else if (em->start >= extent_map_end(entry))
105 			p = &(*p)->rb_right;
106 		else
107 			return -EEXIST;
108 	}
109 
110 	orig_parent = parent;
111 	while (parent && em->start >= extent_map_end(entry)) {
112 		parent = rb_next(parent);
113 		entry = rb_entry(parent, struct extent_map, rb_node);
114 	}
115 	if (parent)
116 		if (end > entry->start && em->start < extent_map_end(entry))
117 			return -EEXIST;
118 
119 	parent = orig_parent;
120 	entry = rb_entry(parent, struct extent_map, rb_node);
121 	while (parent && em->start < entry->start) {
122 		parent = rb_prev(parent);
123 		entry = rb_entry(parent, struct extent_map, rb_node);
124 	}
125 	if (parent)
126 		if (end > entry->start && em->start < extent_map_end(entry))
127 			return -EEXIST;
128 
129 	rb_link_node(&em->rb_node, orig_parent, p);
130 	rb_insert_color(&em->rb_node, root);
131 	return 0;
132 }
133 
134 /*
135  * search through the tree for an extent_map with a given offset.  If
136  * it can't be found, try to find some neighboring extents
137  */
138 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
139 				     struct rb_node **prev_ret,
140 				     struct rb_node **next_ret)
141 {
142 	struct rb_node *n = root->rb_node;
143 	struct rb_node *prev = NULL;
144 	struct rb_node *orig_prev = NULL;
145 	struct extent_map *entry;
146 	struct extent_map *prev_entry = NULL;
147 
148 	while (n) {
149 		entry = rb_entry(n, struct extent_map, rb_node);
150 		prev = n;
151 		prev_entry = entry;
152 
153 		if (offset < entry->start)
154 			n = n->rb_left;
155 		else if (offset >= extent_map_end(entry))
156 			n = n->rb_right;
157 		else
158 			return n;
159 	}
160 
161 	if (prev_ret) {
162 		orig_prev = prev;
163 		while (prev && offset >= extent_map_end(prev_entry)) {
164 			prev = rb_next(prev);
165 			prev_entry = rb_entry(prev, struct extent_map, rb_node);
166 		}
167 		*prev_ret = prev;
168 		prev = orig_prev;
169 	}
170 
171 	if (next_ret) {
172 		prev_entry = rb_entry(prev, struct extent_map, rb_node);
173 		while (prev && offset < prev_entry->start) {
174 			prev = rb_prev(prev);
175 			prev_entry = rb_entry(prev, struct extent_map, rb_node);
176 		}
177 		*next_ret = prev;
178 	}
179 	return NULL;
180 }
181 
182 /* check to see if two extent_map structs are adjacent and safe to merge */
183 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
184 {
185 	if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
186 		return 0;
187 
188 	/*
189 	 * don't merge compressed extents, we need to know their
190 	 * actual size
191 	 */
192 	if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
193 		return 0;
194 
195 	if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
196 	    test_bit(EXTENT_FLAG_LOGGING, &next->flags))
197 		return 0;
198 
199 	/*
200 	 * We don't want to merge stuff that hasn't been written to the log yet
201 	 * since it may not reflect exactly what is on disk, and that would be
202 	 * bad.
203 	 */
204 	if (!list_empty(&prev->list) || !list_empty(&next->list))
205 		return 0;
206 
207 	if (extent_map_end(prev) == next->start &&
208 	    prev->flags == next->flags &&
209 	    prev->bdev == next->bdev &&
210 	    ((next->block_start == EXTENT_MAP_HOLE &&
211 	      prev->block_start == EXTENT_MAP_HOLE) ||
212 	     (next->block_start == EXTENT_MAP_INLINE &&
213 	      prev->block_start == EXTENT_MAP_INLINE) ||
214 	     (next->block_start == EXTENT_MAP_DELALLOC &&
215 	      prev->block_start == EXTENT_MAP_DELALLOC) ||
216 	     (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
217 	      next->block_start == extent_map_block_end(prev)))) {
218 		return 1;
219 	}
220 	return 0;
221 }
222 
223 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
224 {
225 	struct extent_map *merge = NULL;
226 	struct rb_node *rb;
227 
228 	if (em->start != 0) {
229 		rb = rb_prev(&em->rb_node);
230 		if (rb)
231 			merge = rb_entry(rb, struct extent_map, rb_node);
232 		if (rb && mergable_maps(merge, em)) {
233 			em->start = merge->start;
234 			em->orig_start = merge->orig_start;
235 			em->len += merge->len;
236 			em->block_len += merge->block_len;
237 			em->block_start = merge->block_start;
238 			em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
239 			em->mod_start = merge->mod_start;
240 			em->generation = max(em->generation, merge->generation);
241 
242 			rb_erase(&merge->rb_node, &tree->map);
243 			RB_CLEAR_NODE(&merge->rb_node);
244 			free_extent_map(merge);
245 		}
246 	}
247 
248 	rb = rb_next(&em->rb_node);
249 	if (rb)
250 		merge = rb_entry(rb, struct extent_map, rb_node);
251 	if (rb && mergable_maps(em, merge)) {
252 		em->len += merge->len;
253 		em->block_len += merge->block_len;
254 		rb_erase(&merge->rb_node, &tree->map);
255 		RB_CLEAR_NODE(&merge->rb_node);
256 		em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
257 		em->generation = max(em->generation, merge->generation);
258 		free_extent_map(merge);
259 	}
260 }
261 
262 /**
263  * unpin_extent_cache - unpin an extent from the cache
264  * @tree:	tree to unpin the extent in
265  * @start:	logical offset in the file
266  * @len:	length of the extent
267  * @gen:	generation that this extent has been modified in
268  *
269  * Called after an extent has been written to disk properly.  Set the generation
270  * to the generation that actually added the file item to the inode so we know
271  * we need to sync this extent when we call fsync().
272  */
273 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
274 		       u64 gen)
275 {
276 	int ret = 0;
277 	struct extent_map *em;
278 	bool prealloc = false;
279 
280 	write_lock(&tree->lock);
281 	em = lookup_extent_mapping(tree, start, len);
282 
283 	WARN_ON(!em || em->start != start);
284 
285 	if (!em)
286 		goto out;
287 
288 	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
289 		list_move(&em->list, &tree->modified_extents);
290 	em->generation = gen;
291 	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
292 	em->mod_start = em->start;
293 	em->mod_len = em->len;
294 
295 	if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
296 		prealloc = true;
297 		clear_bit(EXTENT_FLAG_FILLING, &em->flags);
298 	}
299 
300 	try_merge_map(tree, em);
301 
302 	if (prealloc) {
303 		em->mod_start = em->start;
304 		em->mod_len = em->len;
305 	}
306 
307 	free_extent_map(em);
308 out:
309 	write_unlock(&tree->lock);
310 	return ret;
311 
312 }
313 
314 void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
315 {
316 	clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
317 	if (extent_map_in_tree(em))
318 		try_merge_map(tree, em);
319 }
320 
321 static inline void setup_extent_mapping(struct extent_map_tree *tree,
322 					struct extent_map *em,
323 					int modified)
324 {
325 	atomic_inc(&em->refs);
326 	em->mod_start = em->start;
327 	em->mod_len = em->len;
328 
329 	if (modified)
330 		list_move(&em->list, &tree->modified_extents);
331 	else
332 		try_merge_map(tree, em);
333 }
334 
335 /**
336  * add_extent_mapping - add new extent map to the extent tree
337  * @tree:	tree to insert new map in
338  * @em:		map to insert
339  *
340  * Insert @em into @tree or perform a simple forward/backward merge with
341  * existing mappings.  The extent_map struct passed in will be inserted
342  * into the tree directly, with an additional reference taken, or a
343  * reference dropped if the merge attempt was successful.
344  */
345 int add_extent_mapping(struct extent_map_tree *tree,
346 		       struct extent_map *em, int modified)
347 {
348 	int ret = 0;
349 
350 	ret = tree_insert(&tree->map, em);
351 	if (ret)
352 		goto out;
353 
354 	setup_extent_mapping(tree, em, modified);
355 out:
356 	return ret;
357 }
358 
359 static struct extent_map *
360 __lookup_extent_mapping(struct extent_map_tree *tree,
361 			u64 start, u64 len, int strict)
362 {
363 	struct extent_map *em;
364 	struct rb_node *rb_node;
365 	struct rb_node *prev = NULL;
366 	struct rb_node *next = NULL;
367 	u64 end = range_end(start, len);
368 
369 	rb_node = __tree_search(&tree->map, start, &prev, &next);
370 	if (!rb_node) {
371 		if (prev)
372 			rb_node = prev;
373 		else if (next)
374 			rb_node = next;
375 		else
376 			return NULL;
377 	}
378 
379 	em = rb_entry(rb_node, struct extent_map, rb_node);
380 
381 	if (strict && !(end > em->start && start < extent_map_end(em)))
382 		return NULL;
383 
384 	atomic_inc(&em->refs);
385 	return em;
386 }
387 
388 /**
389  * lookup_extent_mapping - lookup extent_map
390  * @tree:	tree to lookup in
391  * @start:	byte offset to start the search
392  * @len:	length of the lookup range
393  *
394  * Find and return the first extent_map struct in @tree that intersects the
395  * [start, len] range.  There may be additional objects in the tree that
396  * intersect, so check the object returned carefully to make sure that no
397  * additional lookups are needed.
398  */
399 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
400 					 u64 start, u64 len)
401 {
402 	return __lookup_extent_mapping(tree, start, len, 1);
403 }
404 
405 /**
406  * search_extent_mapping - find a nearby extent map
407  * @tree:	tree to lookup in
408  * @start:	byte offset to start the search
409  * @len:	length of the lookup range
410  *
411  * Find and return the first extent_map struct in @tree that intersects the
412  * [start, len] range.
413  *
414  * If one can't be found, any nearby extent may be returned
415  */
416 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
417 					 u64 start, u64 len)
418 {
419 	return __lookup_extent_mapping(tree, start, len, 0);
420 }
421 
422 /**
423  * remove_extent_mapping - removes an extent_map from the extent tree
424  * @tree:	extent tree to remove from
425  * @em:		extent map beeing removed
426  *
427  * Removes @em from @tree.  No reference counts are dropped, and no checks
428  * are done to see if the range is in use
429  */
430 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
431 {
432 	int ret = 0;
433 
434 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
435 	rb_erase(&em->rb_node, &tree->map);
436 	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
437 		list_del_init(&em->list);
438 	RB_CLEAR_NODE(&em->rb_node);
439 	return ret;
440 }
441 
442 void replace_extent_mapping(struct extent_map_tree *tree,
443 			    struct extent_map *cur,
444 			    struct extent_map *new,
445 			    int modified)
446 {
447 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
448 	ASSERT(extent_map_in_tree(cur));
449 	if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
450 		list_del_init(&cur->list);
451 	rb_replace_node(&cur->rb_node, &new->rb_node, &tree->map);
452 	RB_CLEAR_NODE(&cur->rb_node);
453 
454 	setup_extent_mapping(tree, new, modified);
455 }
456