xref: /openbmc/linux/fs/btrfs/extent_map.c (revision 8fdff1dc)
1 #include <linux/err.h>
2 #include <linux/slab.h>
3 #include <linux/module.h>
4 #include <linux/spinlock.h>
5 #include <linux/hardirq.h>
6 #include "ctree.h"
7 #include "extent_map.h"
8 
9 
10 static struct kmem_cache *extent_map_cache;
11 
12 int __init extent_map_init(void)
13 {
14 	extent_map_cache = kmem_cache_create("btrfs_extent_map",
15 			sizeof(struct extent_map), 0,
16 			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
17 	if (!extent_map_cache)
18 		return -ENOMEM;
19 	return 0;
20 }
21 
22 void extent_map_exit(void)
23 {
24 	if (extent_map_cache)
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 	em->in_tree = 0;
56 	em->flags = 0;
57 	em->compress_type = BTRFS_COMPRESS_NONE;
58 	em->generation = 0;
59 	atomic_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 beeing releasead
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(atomic_read(&em->refs) == 0);
76 	if (atomic_dec_and_test(&em->refs)) {
77 		WARN_ON(em->in_tree);
78 		WARN_ON(!list_empty(&em->list));
79 		kmem_cache_free(extent_map_cache, em);
80 	}
81 }
82 
83 static struct rb_node *tree_insert(struct rb_root *root, u64 offset,
84 				   struct rb_node *node)
85 {
86 	struct rb_node **p = &root->rb_node;
87 	struct rb_node *parent = NULL;
88 	struct extent_map *entry;
89 
90 	while (*p) {
91 		parent = *p;
92 		entry = rb_entry(parent, struct extent_map, rb_node);
93 
94 		WARN_ON(!entry->in_tree);
95 
96 		if (offset < entry->start)
97 			p = &(*p)->rb_left;
98 		else if (offset >= extent_map_end(entry))
99 			p = &(*p)->rb_right;
100 		else
101 			return parent;
102 	}
103 
104 	entry = rb_entry(node, struct extent_map, rb_node);
105 	entry->in_tree = 1;
106 	rb_link_node(node, parent, p);
107 	rb_insert_color(node, root);
108 	return NULL;
109 }
110 
111 /*
112  * search through the tree for an extent_map with a given offset.  If
113  * it can't be found, try to find some neighboring extents
114  */
115 static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
116 				     struct rb_node **prev_ret,
117 				     struct rb_node **next_ret)
118 {
119 	struct rb_node *n = root->rb_node;
120 	struct rb_node *prev = NULL;
121 	struct rb_node *orig_prev = NULL;
122 	struct extent_map *entry;
123 	struct extent_map *prev_entry = NULL;
124 
125 	while (n) {
126 		entry = rb_entry(n, struct extent_map, rb_node);
127 		prev = n;
128 		prev_entry = entry;
129 
130 		WARN_ON(!entry->in_tree);
131 
132 		if (offset < entry->start)
133 			n = n->rb_left;
134 		else if (offset >= extent_map_end(entry))
135 			n = n->rb_right;
136 		else
137 			return n;
138 	}
139 
140 	if (prev_ret) {
141 		orig_prev = prev;
142 		while (prev && offset >= extent_map_end(prev_entry)) {
143 			prev = rb_next(prev);
144 			prev_entry = rb_entry(prev, struct extent_map, rb_node);
145 		}
146 		*prev_ret = prev;
147 		prev = orig_prev;
148 	}
149 
150 	if (next_ret) {
151 		prev_entry = rb_entry(prev, struct extent_map, rb_node);
152 		while (prev && offset < prev_entry->start) {
153 			prev = rb_prev(prev);
154 			prev_entry = rb_entry(prev, struct extent_map, rb_node);
155 		}
156 		*next_ret = prev;
157 	}
158 	return NULL;
159 }
160 
161 /* check to see if two extent_map structs are adjacent and safe to merge */
162 static int mergable_maps(struct extent_map *prev, struct extent_map *next)
163 {
164 	if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
165 		return 0;
166 
167 	/*
168 	 * don't merge compressed extents, we need to know their
169 	 * actual size
170 	 */
171 	if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
172 		return 0;
173 
174 	if (extent_map_end(prev) == next->start &&
175 	    prev->flags == next->flags &&
176 	    prev->bdev == next->bdev &&
177 	    ((next->block_start == EXTENT_MAP_HOLE &&
178 	      prev->block_start == EXTENT_MAP_HOLE) ||
179 	     (next->block_start == EXTENT_MAP_INLINE &&
180 	      prev->block_start == EXTENT_MAP_INLINE) ||
181 	     (next->block_start == EXTENT_MAP_DELALLOC &&
182 	      prev->block_start == EXTENT_MAP_DELALLOC) ||
183 	     (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
184 	      next->block_start == extent_map_block_end(prev)))) {
185 		return 1;
186 	}
187 	return 0;
188 }
189 
190 static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
191 {
192 	struct extent_map *merge = NULL;
193 	struct rb_node *rb;
194 
195 	if (em->start != 0) {
196 		rb = rb_prev(&em->rb_node);
197 		if (rb)
198 			merge = rb_entry(rb, struct extent_map, rb_node);
199 		if (rb && mergable_maps(merge, em)) {
200 			em->start = merge->start;
201 			em->orig_start = merge->orig_start;
202 			em->len += merge->len;
203 			em->block_len += merge->block_len;
204 			em->block_start = merge->block_start;
205 			merge->in_tree = 0;
206 			em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
207 			em->mod_start = merge->mod_start;
208 			em->generation = max(em->generation, merge->generation);
209 			list_move(&em->list, &tree->modified_extents);
210 
211 			list_del_init(&merge->list);
212 			rb_erase(&merge->rb_node, &tree->map);
213 			free_extent_map(merge);
214 		}
215 	}
216 
217 	rb = rb_next(&em->rb_node);
218 	if (rb)
219 		merge = rb_entry(rb, struct extent_map, rb_node);
220 	if (rb && mergable_maps(em, merge)) {
221 		em->len += merge->len;
222 		em->block_len += merge->len;
223 		rb_erase(&merge->rb_node, &tree->map);
224 		merge->in_tree = 0;
225 		em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
226 		em->generation = max(em->generation, merge->generation);
227 		list_del_init(&merge->list);
228 		free_extent_map(merge);
229 	}
230 }
231 
232 /**
233  * unpin_extent_cache - unpin an extent from the cache
234  * @tree:	tree to unpin the extent in
235  * @start:	logical offset in the file
236  * @len:	length of the extent
237  * @gen:	generation that this extent has been modified in
238  *
239  * Called after an extent has been written to disk properly.  Set the generation
240  * to the generation that actually added the file item to the inode so we know
241  * we need to sync this extent when we call fsync().
242  */
243 int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
244 		       u64 gen)
245 {
246 	int ret = 0;
247 	struct extent_map *em;
248 	bool prealloc = false;
249 
250 	write_lock(&tree->lock);
251 	em = lookup_extent_mapping(tree, start, len);
252 
253 	WARN_ON(!em || em->start != start);
254 
255 	if (!em)
256 		goto out;
257 
258 	list_move(&em->list, &tree->modified_extents);
259 	em->generation = gen;
260 	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
261 	em->mod_start = em->start;
262 	em->mod_len = em->len;
263 
264 	if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
265 		prealloc = true;
266 		clear_bit(EXTENT_FLAG_FILLING, &em->flags);
267 	}
268 
269 	try_merge_map(tree, em);
270 
271 	if (prealloc) {
272 		em->mod_start = em->start;
273 		em->mod_len = em->len;
274 	}
275 
276 	free_extent_map(em);
277 out:
278 	write_unlock(&tree->lock);
279 	return ret;
280 
281 }
282 
283 /**
284  * add_extent_mapping - add new extent map to the extent tree
285  * @tree:	tree to insert new map in
286  * @em:		map to insert
287  *
288  * Insert @em into @tree or perform a simple forward/backward merge with
289  * existing mappings.  The extent_map struct passed in will be inserted
290  * into the tree directly, with an additional reference taken, or a
291  * reference dropped if the merge attempt was successful.
292  */
293 int add_extent_mapping(struct extent_map_tree *tree,
294 		       struct extent_map *em)
295 {
296 	int ret = 0;
297 	struct rb_node *rb;
298 	struct extent_map *exist;
299 
300 	exist = lookup_extent_mapping(tree, em->start, em->len);
301 	if (exist) {
302 		free_extent_map(exist);
303 		ret = -EEXIST;
304 		goto out;
305 	}
306 	rb = tree_insert(&tree->map, em->start, &em->rb_node);
307 	if (rb) {
308 		ret = -EEXIST;
309 		goto out;
310 	}
311 	atomic_inc(&em->refs);
312 
313 	em->mod_start = em->start;
314 	em->mod_len = em->len;
315 
316 	try_merge_map(tree, em);
317 out:
318 	return ret;
319 }
320 
321 /* simple helper to do math around the end of an extent, handling wrap */
322 static u64 range_end(u64 start, u64 len)
323 {
324 	if (start + len < start)
325 		return (u64)-1;
326 	return start + len;
327 }
328 
329 struct extent_map *__lookup_extent_mapping(struct extent_map_tree *tree,
330 					   u64 start, u64 len, int strict)
331 {
332 	struct extent_map *em;
333 	struct rb_node *rb_node;
334 	struct rb_node *prev = NULL;
335 	struct rb_node *next = NULL;
336 	u64 end = range_end(start, len);
337 
338 	rb_node = __tree_search(&tree->map, start, &prev, &next);
339 	if (!rb_node) {
340 		if (prev)
341 			rb_node = prev;
342 		else if (next)
343 			rb_node = next;
344 		else
345 			return NULL;
346 	}
347 
348 	em = rb_entry(rb_node, struct extent_map, rb_node);
349 
350 	if (strict && !(end > em->start && start < extent_map_end(em)))
351 		return NULL;
352 
353 	atomic_inc(&em->refs);
354 	return em;
355 }
356 
357 /**
358  * lookup_extent_mapping - lookup extent_map
359  * @tree:	tree to lookup in
360  * @start:	byte offset to start the search
361  * @len:	length of the lookup range
362  *
363  * Find and return the first extent_map struct in @tree that intersects the
364  * [start, len] range.  There may be additional objects in the tree that
365  * intersect, so check the object returned carefully to make sure that no
366  * additional lookups are needed.
367  */
368 struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
369 					 u64 start, u64 len)
370 {
371 	return __lookup_extent_mapping(tree, start, len, 1);
372 }
373 
374 /**
375  * search_extent_mapping - find a nearby extent map
376  * @tree:	tree to lookup in
377  * @start:	byte offset to start the search
378  * @len:	length of the lookup range
379  *
380  * Find and return the first extent_map struct in @tree that intersects the
381  * [start, len] range.
382  *
383  * If one can't be found, any nearby extent may be returned
384  */
385 struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
386 					 u64 start, u64 len)
387 {
388 	return __lookup_extent_mapping(tree, start, len, 0);
389 }
390 
391 /**
392  * remove_extent_mapping - removes an extent_map from the extent tree
393  * @tree:	extent tree to remove from
394  * @em:		extent map beeing removed
395  *
396  * Removes @em from @tree.  No reference counts are dropped, and no checks
397  * are done to see if the range is in use
398  */
399 int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
400 {
401 	int ret = 0;
402 
403 	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
404 	rb_erase(&em->rb_node, &tree->map);
405 	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
406 		list_del_init(&em->list);
407 	em->in_tree = 0;
408 	return ret;
409 }
410