]>
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> |
7a2508e1 | 8 | #include <linux/mbcache.h> |
f9a61eb4 JK |
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 | |
7a2508e1 | 13 | * mb_cache_entry_delete_block()). |
f9a61eb4 JK |
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 | ||
7a2508e1 | 26 | struct mb_cache { |
f9a61eb4 JK |
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; | |
f0c8b462 JK |
33 | /* Protects c_list, c_entry_count */ |
34 | spinlock_t c_list_lock; | |
35 | struct list_head c_list; | |
f9a61eb4 JK |
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 | ||
7a2508e1 | 43 | static struct kmem_cache *mb_entry_cache; |
f9a61eb4 | 44 | |
7a2508e1 JK |
45 | static unsigned long mb_cache_shrink(struct mb_cache *cache, |
46 | unsigned int nr_to_scan); | |
c2f3140f | 47 | |
dc8d5e56 AG |
48 | static inline struct hlist_bl_head *mb_cache_entry_head(struct mb_cache *cache, |
49 | u32 key) | |
f0c8b462 | 50 | { |
dc8d5e56 | 51 | return &cache->c_hash[hash_32(key, cache->c_bucket_bits)]; |
f0c8b462 JK |
52 | } |
53 | ||
c2f3140f JK |
54 | /* |
55 | * Number of entries to reclaim synchronously when there are too many entries | |
56 | * in cache | |
57 | */ | |
58 | #define SYNC_SHRINK_BATCH 64 | |
59 | ||
f9a61eb4 | 60 | /* |
7a2508e1 | 61 | * mb_cache_entry_create - create entry in cache |
f9a61eb4 JK |
62 | * @cache - cache where the entry should be created |
63 | * @mask - gfp mask with which the entry should be allocated | |
64 | * @key - key of the entry | |
65 | * @block - block that contains data | |
66 | * | |
67 | * Creates entry in @cache with key @key and records that data is stored in | |
68 | * block @block. The function returns -EBUSY if entry with the same key | |
69 | * and for the same block already exists in cache. Otherwise 0 is returned. | |
70 | */ | |
7a2508e1 JK |
71 | int mb_cache_entry_create(struct mb_cache *cache, gfp_t mask, u32 key, |
72 | sector_t block) | |
f9a61eb4 | 73 | { |
7a2508e1 | 74 | struct mb_cache_entry *entry, *dup; |
f9a61eb4 JK |
75 | struct hlist_bl_node *dup_node; |
76 | struct hlist_bl_head *head; | |
77 | ||
c2f3140f JK |
78 | /* Schedule background reclaim if there are too many entries */ |
79 | if (cache->c_entry_count >= cache->c_max_entries) | |
80 | schedule_work(&cache->c_shrink_work); | |
81 | /* Do some sync reclaim if background reclaim cannot keep up */ | |
82 | if (cache->c_entry_count >= 2*cache->c_max_entries) | |
7a2508e1 | 83 | mb_cache_shrink(cache, SYNC_SHRINK_BATCH); |
c2f3140f | 84 | |
7a2508e1 | 85 | entry = kmem_cache_alloc(mb_entry_cache, mask); |
f9a61eb4 JK |
86 | if (!entry) |
87 | return -ENOMEM; | |
88 | ||
f0c8b462 | 89 | INIT_LIST_HEAD(&entry->e_list); |
f9a61eb4 JK |
90 | /* One ref for hash, one ref returned */ |
91 | atomic_set(&entry->e_refcnt, 1); | |
92 | entry->e_key = key; | |
93 | entry->e_block = block; | |
dc8d5e56 | 94 | head = mb_cache_entry_head(cache, key); |
f9a61eb4 JK |
95 | hlist_bl_lock(head); |
96 | hlist_bl_for_each_entry(dup, dup_node, head, e_hash_list) { | |
97 | if (dup->e_key == key && dup->e_block == block) { | |
98 | hlist_bl_unlock(head); | |
7a2508e1 | 99 | kmem_cache_free(mb_entry_cache, entry); |
f9a61eb4 JK |
100 | return -EBUSY; |
101 | } | |
102 | } | |
103 | hlist_bl_add_head(&entry->e_hash_list, head); | |
104 | hlist_bl_unlock(head); | |
105 | ||
f0c8b462 JK |
106 | spin_lock(&cache->c_list_lock); |
107 | list_add_tail(&entry->e_list, &cache->c_list); | |
f9a61eb4 JK |
108 | /* Grab ref for LRU list */ |
109 | atomic_inc(&entry->e_refcnt); | |
110 | cache->c_entry_count++; | |
f0c8b462 | 111 | spin_unlock(&cache->c_list_lock); |
f9a61eb4 JK |
112 | |
113 | return 0; | |
114 | } | |
7a2508e1 | 115 | EXPORT_SYMBOL(mb_cache_entry_create); |
f9a61eb4 | 116 | |
7a2508e1 | 117 | void __mb_cache_entry_free(struct mb_cache_entry *entry) |
f9a61eb4 | 118 | { |
7a2508e1 | 119 | kmem_cache_free(mb_entry_cache, entry); |
f9a61eb4 | 120 | } |
7a2508e1 | 121 | EXPORT_SYMBOL(__mb_cache_entry_free); |
f9a61eb4 | 122 | |
7a2508e1 JK |
123 | static struct mb_cache_entry *__entry_find(struct mb_cache *cache, |
124 | struct mb_cache_entry *entry, | |
125 | u32 key) | |
f9a61eb4 | 126 | { |
7a2508e1 | 127 | struct mb_cache_entry *old_entry = entry; |
f9a61eb4 JK |
128 | struct hlist_bl_node *node; |
129 | struct hlist_bl_head *head; | |
130 | ||
dc8d5e56 | 131 | head = mb_cache_entry_head(cache, key); |
f9a61eb4 JK |
132 | hlist_bl_lock(head); |
133 | if (entry && !hlist_bl_unhashed(&entry->e_hash_list)) | |
134 | node = entry->e_hash_list.next; | |
135 | else | |
136 | node = hlist_bl_first(head); | |
137 | while (node) { | |
7a2508e1 | 138 | entry = hlist_bl_entry(node, struct mb_cache_entry, |
f9a61eb4 JK |
139 | e_hash_list); |
140 | if (entry->e_key == key) { | |
141 | atomic_inc(&entry->e_refcnt); | |
142 | goto out; | |
143 | } | |
144 | node = node->next; | |
145 | } | |
146 | entry = NULL; | |
147 | out: | |
148 | hlist_bl_unlock(head); | |
149 | if (old_entry) | |
7a2508e1 | 150 | mb_cache_entry_put(cache, old_entry); |
f9a61eb4 JK |
151 | |
152 | return entry; | |
153 | } | |
154 | ||
155 | /* | |
7a2508e1 | 156 | * mb_cache_entry_find_first - find the first entry in cache with given key |
f9a61eb4 JK |
157 | * @cache: cache where we should search |
158 | * @key: key to look for | |
159 | * | |
160 | * Search in @cache for entry with key @key. Grabs reference to the first | |
161 | * entry found and returns the entry. | |
162 | */ | |
7a2508e1 JK |
163 | struct mb_cache_entry *mb_cache_entry_find_first(struct mb_cache *cache, |
164 | u32 key) | |
f9a61eb4 JK |
165 | { |
166 | return __entry_find(cache, NULL, key); | |
167 | } | |
7a2508e1 | 168 | EXPORT_SYMBOL(mb_cache_entry_find_first); |
f9a61eb4 JK |
169 | |
170 | /* | |
7a2508e1 | 171 | * mb_cache_entry_find_next - find next entry in cache with the same |
f9a61eb4 JK |
172 | * @cache: cache where we should search |
173 | * @entry: entry to start search from | |
174 | * | |
175 | * Finds next entry in the hash chain which has the same key as @entry. | |
176 | * If @entry is unhashed (which can happen when deletion of entry races | |
177 | * with the search), finds the first entry in the hash chain. The function | |
178 | * drops reference to @entry and returns with a reference to the found entry. | |
179 | */ | |
7a2508e1 JK |
180 | struct mb_cache_entry *mb_cache_entry_find_next(struct mb_cache *cache, |
181 | struct mb_cache_entry *entry) | |
f9a61eb4 JK |
182 | { |
183 | return __entry_find(cache, entry, entry->e_key); | |
184 | } | |
7a2508e1 | 185 | EXPORT_SYMBOL(mb_cache_entry_find_next); |
f9a61eb4 | 186 | |
7a2508e1 | 187 | /* mb_cache_entry_delete_block - remove information about block from cache |
f9a61eb4 JK |
188 | * @cache - cache we work with |
189 | * @key - key of the entry to remove | |
190 | * @block - block containing data for @key | |
191 | * | |
192 | * Remove entry from cache @cache with key @key with data stored in @block. | |
193 | */ | |
7a2508e1 JK |
194 | void mb_cache_entry_delete_block(struct mb_cache *cache, u32 key, |
195 | sector_t block) | |
f9a61eb4 JK |
196 | { |
197 | struct hlist_bl_node *node; | |
198 | struct hlist_bl_head *head; | |
7a2508e1 | 199 | struct mb_cache_entry *entry; |
f9a61eb4 | 200 | |
dc8d5e56 | 201 | head = mb_cache_entry_head(cache, key); |
f9a61eb4 JK |
202 | hlist_bl_lock(head); |
203 | hlist_bl_for_each_entry(entry, node, head, e_hash_list) { | |
204 | if (entry->e_key == key && entry->e_block == block) { | |
205 | /* We keep hash list reference to keep entry alive */ | |
206 | hlist_bl_del_init(&entry->e_hash_list); | |
207 | hlist_bl_unlock(head); | |
f0c8b462 JK |
208 | spin_lock(&cache->c_list_lock); |
209 | if (!list_empty(&entry->e_list)) { | |
210 | list_del_init(&entry->e_list); | |
f9a61eb4 JK |
211 | cache->c_entry_count--; |
212 | atomic_dec(&entry->e_refcnt); | |
213 | } | |
f0c8b462 | 214 | spin_unlock(&cache->c_list_lock); |
7a2508e1 | 215 | mb_cache_entry_put(cache, entry); |
f9a61eb4 JK |
216 | return; |
217 | } | |
218 | } | |
219 | hlist_bl_unlock(head); | |
220 | } | |
7a2508e1 | 221 | EXPORT_SYMBOL(mb_cache_entry_delete_block); |
f9a61eb4 | 222 | |
7a2508e1 | 223 | /* mb_cache_entry_touch - cache entry got used |
f9a61eb4 JK |
224 | * @cache - cache the entry belongs to |
225 | * @entry - entry that got used | |
226 | * | |
f0c8b462 | 227 | * Marks entry as used to give hit higher chances of surviving in cache. |
f9a61eb4 | 228 | */ |
7a2508e1 JK |
229 | void mb_cache_entry_touch(struct mb_cache *cache, |
230 | struct mb_cache_entry *entry) | |
f9a61eb4 | 231 | { |
dc8d5e56 | 232 | entry->e_referenced = 1; |
f9a61eb4 | 233 | } |
7a2508e1 | 234 | EXPORT_SYMBOL(mb_cache_entry_touch); |
f9a61eb4 | 235 | |
7a2508e1 JK |
236 | static unsigned long mb_cache_count(struct shrinker *shrink, |
237 | struct shrink_control *sc) | |
f9a61eb4 | 238 | { |
7a2508e1 JK |
239 | struct mb_cache *cache = container_of(shrink, struct mb_cache, |
240 | c_shrink); | |
f9a61eb4 JK |
241 | |
242 | return cache->c_entry_count; | |
243 | } | |
244 | ||
245 | /* Shrink number of entries in cache */ | |
7a2508e1 JK |
246 | static unsigned long mb_cache_shrink(struct mb_cache *cache, |
247 | unsigned int nr_to_scan) | |
f9a61eb4 | 248 | { |
7a2508e1 | 249 | struct mb_cache_entry *entry; |
f9a61eb4 JK |
250 | struct hlist_bl_head *head; |
251 | unsigned int shrunk = 0; | |
252 | ||
f0c8b462 JK |
253 | spin_lock(&cache->c_list_lock); |
254 | while (nr_to_scan-- && !list_empty(&cache->c_list)) { | |
255 | entry = list_first_entry(&cache->c_list, | |
7a2508e1 | 256 | struct mb_cache_entry, e_list); |
dc8d5e56 AG |
257 | if (entry->e_referenced) { |
258 | entry->e_referenced = 0; | |
f0c8b462 JK |
259 | list_move_tail(&cache->c_list, &entry->e_list); |
260 | continue; | |
261 | } | |
262 | list_del_init(&entry->e_list); | |
f9a61eb4 JK |
263 | cache->c_entry_count--; |
264 | /* | |
265 | * We keep LRU list reference so that entry doesn't go away | |
266 | * from under us. | |
267 | */ | |
f0c8b462 | 268 | spin_unlock(&cache->c_list_lock); |
dc8d5e56 | 269 | head = mb_cache_entry_head(cache, entry->e_key); |
f9a61eb4 JK |
270 | hlist_bl_lock(head); |
271 | if (!hlist_bl_unhashed(&entry->e_hash_list)) { | |
272 | hlist_bl_del_init(&entry->e_hash_list); | |
273 | atomic_dec(&entry->e_refcnt); | |
274 | } | |
275 | hlist_bl_unlock(head); | |
7a2508e1 | 276 | if (mb_cache_entry_put(cache, entry)) |
f9a61eb4 JK |
277 | shrunk++; |
278 | cond_resched(); | |
f0c8b462 | 279 | spin_lock(&cache->c_list_lock); |
f9a61eb4 | 280 | } |
f0c8b462 | 281 | spin_unlock(&cache->c_list_lock); |
f9a61eb4 JK |
282 | |
283 | return shrunk; | |
284 | } | |
285 | ||
7a2508e1 JK |
286 | static unsigned long mb_cache_scan(struct shrinker *shrink, |
287 | struct shrink_control *sc) | |
c2f3140f JK |
288 | { |
289 | int nr_to_scan = sc->nr_to_scan; | |
7a2508e1 | 290 | struct mb_cache *cache = container_of(shrink, struct mb_cache, |
c2f3140f | 291 | c_shrink); |
7a2508e1 | 292 | return mb_cache_shrink(cache, nr_to_scan); |
c2f3140f JK |
293 | } |
294 | ||
295 | /* We shrink 1/X of the cache when we have too many entries in it */ | |
296 | #define SHRINK_DIVISOR 16 | |
297 | ||
7a2508e1 | 298 | static void mb_cache_shrink_worker(struct work_struct *work) |
c2f3140f | 299 | { |
7a2508e1 JK |
300 | struct mb_cache *cache = container_of(work, struct mb_cache, |
301 | c_shrink_work); | |
302 | mb_cache_shrink(cache, cache->c_max_entries / SHRINK_DIVISOR); | |
c2f3140f JK |
303 | } |
304 | ||
f9a61eb4 | 305 | /* |
7a2508e1 | 306 | * mb_cache_create - create cache |
f9a61eb4 JK |
307 | * @bucket_bits: log2 of the hash table size |
308 | * | |
309 | * Create cache for keys with 2^bucket_bits hash entries. | |
310 | */ | |
7a2508e1 | 311 | struct mb_cache *mb_cache_create(int bucket_bits) |
f9a61eb4 | 312 | { |
7a2508e1 | 313 | struct mb_cache *cache; |
f9a61eb4 JK |
314 | int bucket_count = 1 << bucket_bits; |
315 | int i; | |
316 | ||
317 | if (!try_module_get(THIS_MODULE)) | |
318 | return NULL; | |
319 | ||
7a2508e1 | 320 | cache = kzalloc(sizeof(struct mb_cache), GFP_KERNEL); |
f9a61eb4 JK |
321 | if (!cache) |
322 | goto err_out; | |
323 | cache->c_bucket_bits = bucket_bits; | |
c2f3140f | 324 | cache->c_max_entries = bucket_count << 4; |
f0c8b462 JK |
325 | INIT_LIST_HEAD(&cache->c_list); |
326 | spin_lock_init(&cache->c_list_lock); | |
f9a61eb4 JK |
327 | cache->c_hash = kmalloc(bucket_count * sizeof(struct hlist_bl_head), |
328 | GFP_KERNEL); | |
329 | if (!cache->c_hash) { | |
330 | kfree(cache); | |
331 | goto err_out; | |
332 | } | |
333 | for (i = 0; i < bucket_count; i++) | |
334 | INIT_HLIST_BL_HEAD(&cache->c_hash[i]); | |
335 | ||
7a2508e1 JK |
336 | cache->c_shrink.count_objects = mb_cache_count; |
337 | cache->c_shrink.scan_objects = mb_cache_scan; | |
f9a61eb4 JK |
338 | cache->c_shrink.seeks = DEFAULT_SEEKS; |
339 | register_shrinker(&cache->c_shrink); | |
340 | ||
7a2508e1 | 341 | INIT_WORK(&cache->c_shrink_work, mb_cache_shrink_worker); |
c2f3140f | 342 | |
f9a61eb4 JK |
343 | return cache; |
344 | ||
345 | err_out: | |
346 | module_put(THIS_MODULE); | |
347 | return NULL; | |
348 | } | |
7a2508e1 | 349 | EXPORT_SYMBOL(mb_cache_create); |
f9a61eb4 JK |
350 | |
351 | /* | |
7a2508e1 | 352 | * mb_cache_destroy - destroy cache |
f9a61eb4 JK |
353 | * @cache: the cache to destroy |
354 | * | |
355 | * Free all entries in cache and cache itself. Caller must make sure nobody | |
356 | * (except shrinker) can reach @cache when calling this. | |
357 | */ | |
7a2508e1 | 358 | void mb_cache_destroy(struct mb_cache *cache) |
f9a61eb4 | 359 | { |
7a2508e1 | 360 | struct mb_cache_entry *entry, *next; |
f9a61eb4 JK |
361 | |
362 | unregister_shrinker(&cache->c_shrink); | |
363 | ||
364 | /* | |
365 | * We don't bother with any locking. Cache must not be used at this | |
366 | * point. | |
367 | */ | |
f0c8b462 | 368 | list_for_each_entry_safe(entry, next, &cache->c_list, e_list) { |
f9a61eb4 JK |
369 | if (!hlist_bl_unhashed(&entry->e_hash_list)) { |
370 | hlist_bl_del_init(&entry->e_hash_list); | |
371 | atomic_dec(&entry->e_refcnt); | |
372 | } else | |
373 | WARN_ON(1); | |
f0c8b462 | 374 | list_del(&entry->e_list); |
f9a61eb4 | 375 | WARN_ON(atomic_read(&entry->e_refcnt) != 1); |
7a2508e1 | 376 | mb_cache_entry_put(cache, entry); |
f9a61eb4 JK |
377 | } |
378 | kfree(cache->c_hash); | |
379 | kfree(cache); | |
380 | module_put(THIS_MODULE); | |
381 | } | |
7a2508e1 | 382 | EXPORT_SYMBOL(mb_cache_destroy); |
f9a61eb4 | 383 | |
7a2508e1 | 384 | static int __init mbcache_init(void) |
f9a61eb4 | 385 | { |
7a2508e1 JK |
386 | mb_entry_cache = kmem_cache_create("mbcache", |
387 | sizeof(struct mb_cache_entry), 0, | |
f9a61eb4 | 388 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL); |
7a2508e1 | 389 | BUG_ON(!mb_entry_cache); |
f9a61eb4 JK |
390 | return 0; |
391 | } | |
392 | ||
7a2508e1 | 393 | static void __exit mbcache_exit(void) |
f9a61eb4 | 394 | { |
7a2508e1 | 395 | kmem_cache_destroy(mb_entry_cache); |
f9a61eb4 JK |
396 | } |
397 | ||
7a2508e1 JK |
398 | module_init(mbcache_init) |
399 | module_exit(mbcache_exit) | |
f9a61eb4 JK |
400 | |
401 | MODULE_AUTHOR("Jan Kara <jack@suse.cz>"); | |
402 | MODULE_DESCRIPTION("Meta block cache (for extended attributes)"); | |
403 | MODULE_LICENSE("GPL"); |