Commit | Line | Data |
---|---|---|
f9a61eb4 JK |
1 | #include <linux/spinlock.h> |
2 | #include <linux/slab.h> | |
3 | #include <linux/list.h> | |
4 | #include <linux/list_bl.h> | |
5 | #include <linux/module.h> | |
6 | #include <linux/sched.h> | |
c2f3140f | 7 | #include <linux/workqueue.h> |
f9a61eb4 JK |
8 | #include <linux/mbcache2.h> |
9 | ||
10 | /* | |
11 | * Mbcache is a simple key-value store. Keys need not be unique, however | |
12 | * key-value pairs are expected to be unique (we use this fact in | |
13 | * mb2_cache_entry_delete_block()). | |
14 | * | |
15 | * Ext2 and ext4 use this cache for deduplication of extended attribute blocks. | |
16 | * They use hash of a block contents as a key and block number as a value. | |
17 | * That's why keys need not be unique (different xattr blocks may end up having | |
18 | * the same hash). However block number always uniquely identifies a cache | |
19 | * entry. | |
20 | * | |
21 | * We provide functions for creation and removal of entries, search by key, | |
22 | * and a special "delete entry with given key-value pair" operation. Fixed | |
23 | * size hash table is used for fast key lookups. | |
24 | */ | |
25 | ||
26 | struct mb2_cache { | |
27 | /* Hash table of entries */ | |
28 | struct hlist_bl_head *c_hash; | |
29 | /* log2 of hash table size */ | |
30 | int c_bucket_bits; | |
c2f3140f JK |
31 | /* Maximum entries in cache to avoid degrading hash too much */ |
32 | int c_max_entries; | |
f9a61eb4 JK |
33 | /* Protects c_lru_list, c_entry_count */ |
34 | spinlock_t c_lru_list_lock; | |
35 | struct list_head c_lru_list; | |
36 | /* Number of entries in cache */ | |
37 | unsigned long c_entry_count; | |
38 | struct shrinker c_shrink; | |
c2f3140f JK |
39 | /* Work for shrinking when the cache has too many entries */ |
40 | struct work_struct c_shrink_work; | |
f9a61eb4 JK |
41 | }; |
42 | ||
43 | static struct kmem_cache *mb2_entry_cache; | |
44 | ||
c2f3140f JK |
45 | static unsigned long mb2_cache_shrink(struct mb2_cache *cache, |
46 | unsigned int nr_to_scan); | |
47 | ||
48 | /* | |
49 | * Number of entries to reclaim synchronously when there are too many entries | |
50 | * in cache | |
51 | */ | |
52 | #define SYNC_SHRINK_BATCH 64 | |
53 | ||
f9a61eb4 JK |
54 | /* |
55 | * mb2_cache_entry_create - create entry in cache | |
56 | * @cache - cache where the entry should be created | |
57 | * @mask - gfp mask with which the entry should be allocated | |
58 | * @key - key of the entry | |
59 | * @block - block that contains data | |
60 | * | |
61 | * Creates entry in @cache with key @key and records that data is stored in | |
62 | * block @block. The function returns -EBUSY if entry with the same key | |
63 | * and for the same block already exists in cache. Otherwise 0 is returned. | |
64 | */ | |
65 | int mb2_cache_entry_create(struct mb2_cache *cache, gfp_t mask, u32 key, | |
66 | sector_t block) | |
67 | { | |
68 | struct mb2_cache_entry *entry, *dup; | |
69 | struct hlist_bl_node *dup_node; | |
70 | struct hlist_bl_head *head; | |
71 | ||
c2f3140f JK |
72 | /* Schedule background reclaim if there are too many entries */ |
73 | if (cache->c_entry_count >= cache->c_max_entries) | |
74 | schedule_work(&cache->c_shrink_work); | |
75 | /* Do some sync reclaim if background reclaim cannot keep up */ | |
76 | if (cache->c_entry_count >= 2*cache->c_max_entries) | |
77 | mb2_cache_shrink(cache, SYNC_SHRINK_BATCH); | |
78 | ||
f9a61eb4 JK |
79 | entry = kmem_cache_alloc(mb2_entry_cache, mask); |
80 | if (!entry) | |
81 | return -ENOMEM; | |
82 | ||
83 | INIT_LIST_HEAD(&entry->e_lru_list); | |
84 | /* One ref for hash, one ref returned */ | |
85 | atomic_set(&entry->e_refcnt, 1); | |
86 | entry->e_key = key; | |
87 | entry->e_block = block; | |
88 | head = &cache->c_hash[hash_32(key, cache->c_bucket_bits)]; | |
89 | entry->e_hash_list_head = head; | |
90 | hlist_bl_lock(head); | |
91 | hlist_bl_for_each_entry(dup, dup_node, head, e_hash_list) { | |
92 | if (dup->e_key == key && dup->e_block == block) { | |
93 | hlist_bl_unlock(head); | |
94 | kmem_cache_free(mb2_entry_cache, entry); | |
95 | return -EBUSY; | |
96 | } | |
97 | } | |
98 | hlist_bl_add_head(&entry->e_hash_list, head); | |
99 | hlist_bl_unlock(head); | |
100 | ||
101 | spin_lock(&cache->c_lru_list_lock); | |
102 | list_add_tail(&entry->e_lru_list, &cache->c_lru_list); | |
103 | /* Grab ref for LRU list */ | |
104 | atomic_inc(&entry->e_refcnt); | |
105 | cache->c_entry_count++; | |
106 | spin_unlock(&cache->c_lru_list_lock); | |
107 | ||
108 | return 0; | |
109 | } | |
110 | EXPORT_SYMBOL(mb2_cache_entry_create); | |
111 | ||
112 | void __mb2_cache_entry_free(struct mb2_cache_entry *entry) | |
113 | { | |
114 | kmem_cache_free(mb2_entry_cache, entry); | |
115 | } | |
116 | EXPORT_SYMBOL(__mb2_cache_entry_free); | |
117 | ||
118 | static struct mb2_cache_entry *__entry_find(struct mb2_cache *cache, | |
119 | struct mb2_cache_entry *entry, | |
120 | u32 key) | |
121 | { | |
122 | struct mb2_cache_entry *old_entry = entry; | |
123 | struct hlist_bl_node *node; | |
124 | struct hlist_bl_head *head; | |
125 | ||
126 | if (entry) | |
127 | head = entry->e_hash_list_head; | |
128 | else | |
129 | head = &cache->c_hash[hash_32(key, cache->c_bucket_bits)]; | |
130 | hlist_bl_lock(head); | |
131 | if (entry && !hlist_bl_unhashed(&entry->e_hash_list)) | |
132 | node = entry->e_hash_list.next; | |
133 | else | |
134 | node = hlist_bl_first(head); | |
135 | while (node) { | |
136 | entry = hlist_bl_entry(node, struct mb2_cache_entry, | |
137 | e_hash_list); | |
138 | if (entry->e_key == key) { | |
139 | atomic_inc(&entry->e_refcnt); | |
140 | goto out; | |
141 | } | |
142 | node = node->next; | |
143 | } | |
144 | entry = NULL; | |
145 | out: | |
146 | hlist_bl_unlock(head); | |
147 | if (old_entry) | |
148 | mb2_cache_entry_put(cache, old_entry); | |
149 | ||
150 | return entry; | |
151 | } | |
152 | ||
153 | /* | |
154 | * mb2_cache_entry_find_first - find the first entry in cache with given key | |
155 | * @cache: cache where we should search | |
156 | * @key: key to look for | |
157 | * | |
158 | * Search in @cache for entry with key @key. Grabs reference to the first | |
159 | * entry found and returns the entry. | |
160 | */ | |
161 | struct mb2_cache_entry *mb2_cache_entry_find_first(struct mb2_cache *cache, | |
162 | u32 key) | |
163 | { | |
164 | return __entry_find(cache, NULL, key); | |
165 | } | |
166 | EXPORT_SYMBOL(mb2_cache_entry_find_first); | |
167 | ||
168 | /* | |
169 | * mb2_cache_entry_find_next - find next entry in cache with the same | |
170 | * @cache: cache where we should search | |
171 | * @entry: entry to start search from | |
172 | * | |
173 | * Finds next entry in the hash chain which has the same key as @entry. | |
174 | * If @entry is unhashed (which can happen when deletion of entry races | |
175 | * with the search), finds the first entry in the hash chain. The function | |
176 | * drops reference to @entry and returns with a reference to the found entry. | |
177 | */ | |
178 | struct mb2_cache_entry *mb2_cache_entry_find_next(struct mb2_cache *cache, | |
179 | struct mb2_cache_entry *entry) | |
180 | { | |
181 | return __entry_find(cache, entry, entry->e_key); | |
182 | } | |
183 | EXPORT_SYMBOL(mb2_cache_entry_find_next); | |
184 | ||
185 | /* mb2_cache_entry_delete_block - remove information about block from cache | |
186 | * @cache - cache we work with | |
187 | * @key - key of the entry to remove | |
188 | * @block - block containing data for @key | |
189 | * | |
190 | * Remove entry from cache @cache with key @key with data stored in @block. | |
191 | */ | |
192 | void mb2_cache_entry_delete_block(struct mb2_cache *cache, u32 key, | |
193 | sector_t block) | |
194 | { | |
195 | struct hlist_bl_node *node; | |
196 | struct hlist_bl_head *head; | |
197 | struct mb2_cache_entry *entry; | |
198 | ||
199 | head = &cache->c_hash[hash_32(key, cache->c_bucket_bits)]; | |
200 | hlist_bl_lock(head); | |
201 | hlist_bl_for_each_entry(entry, node, head, e_hash_list) { | |
202 | if (entry->e_key == key && entry->e_block == block) { | |
203 | /* We keep hash list reference to keep entry alive */ | |
204 | hlist_bl_del_init(&entry->e_hash_list); | |
205 | hlist_bl_unlock(head); | |
206 | spin_lock(&cache->c_lru_list_lock); | |
207 | if (!list_empty(&entry->e_lru_list)) { | |
208 | list_del_init(&entry->e_lru_list); | |
209 | cache->c_entry_count--; | |
210 | atomic_dec(&entry->e_refcnt); | |
211 | } | |
212 | spin_unlock(&cache->c_lru_list_lock); | |
213 | mb2_cache_entry_put(cache, entry); | |
214 | return; | |
215 | } | |
216 | } | |
217 | hlist_bl_unlock(head); | |
218 | } | |
219 | EXPORT_SYMBOL(mb2_cache_entry_delete_block); | |
220 | ||
221 | /* mb2_cache_entry_touch - cache entry got used | |
222 | * @cache - cache the entry belongs to | |
223 | * @entry - entry that got used | |
224 | * | |
225 | * Move entry in lru list to reflect the fact that it was used. | |
226 | */ | |
227 | void mb2_cache_entry_touch(struct mb2_cache *cache, | |
228 | struct mb2_cache_entry *entry) | |
229 | { | |
230 | spin_lock(&cache->c_lru_list_lock); | |
231 | if (!list_empty(&entry->e_lru_list)) | |
232 | list_move_tail(&cache->c_lru_list, &entry->e_lru_list); | |
233 | spin_unlock(&cache->c_lru_list_lock); | |
234 | } | |
235 | EXPORT_SYMBOL(mb2_cache_entry_touch); | |
236 | ||
237 | static unsigned long mb2_cache_count(struct shrinker *shrink, | |
238 | struct shrink_control *sc) | |
239 | { | |
240 | struct mb2_cache *cache = container_of(shrink, struct mb2_cache, | |
241 | c_shrink); | |
242 | ||
243 | return cache->c_entry_count; | |
244 | } | |
245 | ||
246 | /* Shrink number of entries in cache */ | |
c2f3140f JK |
247 | static unsigned long mb2_cache_shrink(struct mb2_cache *cache, |
248 | unsigned int nr_to_scan) | |
f9a61eb4 | 249 | { |
f9a61eb4 JK |
250 | struct mb2_cache_entry *entry; |
251 | struct hlist_bl_head *head; | |
252 | unsigned int shrunk = 0; | |
253 | ||
254 | spin_lock(&cache->c_lru_list_lock); | |
255 | while (nr_to_scan-- && !list_empty(&cache->c_lru_list)) { | |
256 | entry = list_first_entry(&cache->c_lru_list, | |
257 | struct mb2_cache_entry, e_lru_list); | |
258 | list_del_init(&entry->e_lru_list); | |
259 | cache->c_entry_count--; | |
260 | /* | |
261 | * We keep LRU list reference so that entry doesn't go away | |
262 | * from under us. | |
263 | */ | |
264 | spin_unlock(&cache->c_lru_list_lock); | |
265 | head = entry->e_hash_list_head; | |
266 | hlist_bl_lock(head); | |
267 | if (!hlist_bl_unhashed(&entry->e_hash_list)) { | |
268 | hlist_bl_del_init(&entry->e_hash_list); | |
269 | atomic_dec(&entry->e_refcnt); | |
270 | } | |
271 | hlist_bl_unlock(head); | |
272 | if (mb2_cache_entry_put(cache, entry)) | |
273 | shrunk++; | |
274 | cond_resched(); | |
275 | spin_lock(&cache->c_lru_list_lock); | |
276 | } | |
277 | spin_unlock(&cache->c_lru_list_lock); | |
278 | ||
279 | return shrunk; | |
280 | } | |
281 | ||
c2f3140f JK |
282 | static unsigned long mb2_cache_scan(struct shrinker *shrink, |
283 | struct shrink_control *sc) | |
284 | { | |
285 | int nr_to_scan = sc->nr_to_scan; | |
286 | struct mb2_cache *cache = container_of(shrink, struct mb2_cache, | |
287 | c_shrink); | |
288 | return mb2_cache_shrink(cache, nr_to_scan); | |
289 | } | |
290 | ||
291 | /* We shrink 1/X of the cache when we have too many entries in it */ | |
292 | #define SHRINK_DIVISOR 16 | |
293 | ||
294 | static void mb2_cache_shrink_worker(struct work_struct *work) | |
295 | { | |
296 | struct mb2_cache *cache = container_of(work, struct mb2_cache, | |
297 | c_shrink_work); | |
298 | mb2_cache_shrink(cache, cache->c_max_entries / SHRINK_DIVISOR); | |
299 | } | |
300 | ||
f9a61eb4 JK |
301 | /* |
302 | * mb2_cache_create - create cache | |
303 | * @bucket_bits: log2 of the hash table size | |
304 | * | |
305 | * Create cache for keys with 2^bucket_bits hash entries. | |
306 | */ | |
307 | struct mb2_cache *mb2_cache_create(int bucket_bits) | |
308 | { | |
309 | struct mb2_cache *cache; | |
310 | int bucket_count = 1 << bucket_bits; | |
311 | int i; | |
312 | ||
313 | if (!try_module_get(THIS_MODULE)) | |
314 | return NULL; | |
315 | ||
316 | cache = kzalloc(sizeof(struct mb2_cache), GFP_KERNEL); | |
317 | if (!cache) | |
318 | goto err_out; | |
319 | cache->c_bucket_bits = bucket_bits; | |
c2f3140f | 320 | cache->c_max_entries = bucket_count << 4; |
f9a61eb4 JK |
321 | INIT_LIST_HEAD(&cache->c_lru_list); |
322 | spin_lock_init(&cache->c_lru_list_lock); | |
323 | cache->c_hash = kmalloc(bucket_count * sizeof(struct hlist_bl_head), | |
324 | GFP_KERNEL); | |
325 | if (!cache->c_hash) { | |
326 | kfree(cache); | |
327 | goto err_out; | |
328 | } | |
329 | for (i = 0; i < bucket_count; i++) | |
330 | INIT_HLIST_BL_HEAD(&cache->c_hash[i]); | |
331 | ||
332 | cache->c_shrink.count_objects = mb2_cache_count; | |
333 | cache->c_shrink.scan_objects = mb2_cache_scan; | |
334 | cache->c_shrink.seeks = DEFAULT_SEEKS; | |
335 | register_shrinker(&cache->c_shrink); | |
336 | ||
c2f3140f JK |
337 | INIT_WORK(&cache->c_shrink_work, mb2_cache_shrink_worker); |
338 | ||
f9a61eb4 JK |
339 | return cache; |
340 | ||
341 | err_out: | |
342 | module_put(THIS_MODULE); | |
343 | return NULL; | |
344 | } | |
345 | EXPORT_SYMBOL(mb2_cache_create); | |
346 | ||
347 | /* | |
348 | * mb2_cache_destroy - destroy cache | |
349 | * @cache: the cache to destroy | |
350 | * | |
351 | * Free all entries in cache and cache itself. Caller must make sure nobody | |
352 | * (except shrinker) can reach @cache when calling this. | |
353 | */ | |
354 | void mb2_cache_destroy(struct mb2_cache *cache) | |
355 | { | |
356 | struct mb2_cache_entry *entry, *next; | |
357 | ||
358 | unregister_shrinker(&cache->c_shrink); | |
359 | ||
360 | /* | |
361 | * We don't bother with any locking. Cache must not be used at this | |
362 | * point. | |
363 | */ | |
364 | list_for_each_entry_safe(entry, next, &cache->c_lru_list, e_lru_list) { | |
365 | if (!hlist_bl_unhashed(&entry->e_hash_list)) { | |
366 | hlist_bl_del_init(&entry->e_hash_list); | |
367 | atomic_dec(&entry->e_refcnt); | |
368 | } else | |
369 | WARN_ON(1); | |
370 | list_del(&entry->e_lru_list); | |
371 | WARN_ON(atomic_read(&entry->e_refcnt) != 1); | |
372 | mb2_cache_entry_put(cache, entry); | |
373 | } | |
374 | kfree(cache->c_hash); | |
375 | kfree(cache); | |
376 | module_put(THIS_MODULE); | |
377 | } | |
378 | EXPORT_SYMBOL(mb2_cache_destroy); | |
379 | ||
380 | static int __init mb2cache_init(void) | |
381 | { | |
382 | mb2_entry_cache = kmem_cache_create("mbcache", | |
383 | sizeof(struct mb2_cache_entry), 0, | |
384 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL); | |
385 | BUG_ON(!mb2_entry_cache); | |
386 | return 0; | |
387 | } | |
388 | ||
389 | static void __exit mb2cache_exit(void) | |
390 | { | |
391 | kmem_cache_destroy(mb2_entry_cache); | |
392 | } | |
393 | ||
394 | module_init(mb2cache_init) | |
395 | module_exit(mb2cache_exit) | |
396 | ||
397 | MODULE_AUTHOR("Jan Kara <jack@suse.cz>"); | |
398 | MODULE_DESCRIPTION("Meta block cache (for extended attributes)"); | |
399 | MODULE_LICENSE("GPL"); |