Commit | Line | Data |
---|---|---|
09c434b8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
f9a61eb4 JK |
2 | #include <linux/spinlock.h> |
3 | #include <linux/slab.h> | |
4 | #include <linux/list.h> | |
5 | #include <linux/list_bl.h> | |
6 | #include <linux/module.h> | |
7 | #include <linux/sched.h> | |
c2f3140f | 8 | #include <linux/workqueue.h> |
7a2508e1 | 9 | #include <linux/mbcache.h> |
f9a61eb4 JK |
10 | |
11 | /* | |
12 | * Mbcache is a simple key-value store. Keys need not be unique, however | |
13 | * key-value pairs are expected to be unique (we use this fact in | |
3dc96bba | 14 | * mb_cache_entry_delete_or_get()). |
f9a61eb4 JK |
15 | * |
16 | * Ext2 and ext4 use this cache for deduplication of extended attribute blocks. | |
dec214d0 TE |
17 | * Ext4 also uses it for deduplication of xattr values stored in inodes. |
18 | * They use hash of data as a key and provide a value that may represent a | |
19 | * block or inode number. That's why keys need not be unique (hash of different | |
20 | * data may be the same). However user provided value always uniquely | |
21 | * identifies a cache entry. | |
f9a61eb4 JK |
22 | * |
23 | * We provide functions for creation and removal of entries, search by key, | |
24 | * and a special "delete entry with given key-value pair" operation. Fixed | |
25 | * size hash table is used for fast key lookups. | |
26 | */ | |
27 | ||
7a2508e1 | 28 | struct mb_cache { |
f9a61eb4 JK |
29 | /* Hash table of entries */ |
30 | struct hlist_bl_head *c_hash; | |
31 | /* log2 of hash table size */ | |
32 | int c_bucket_bits; | |
c2f3140f | 33 | /* Maximum entries in cache to avoid degrading hash too much */ |
132d4e2d | 34 | unsigned long c_max_entries; |
f0c8b462 JK |
35 | /* Protects c_list, c_entry_count */ |
36 | spinlock_t c_list_lock; | |
37 | struct list_head c_list; | |
f9a61eb4 JK |
38 | /* Number of entries in cache */ |
39 | unsigned long c_entry_count; | |
40 | struct shrinker c_shrink; | |
c2f3140f JK |
41 | /* Work for shrinking when the cache has too many entries */ |
42 | struct work_struct c_shrink_work; | |
f9a61eb4 JK |
43 | }; |
44 | ||
7a2508e1 | 45 | static struct kmem_cache *mb_entry_cache; |
f9a61eb4 | 46 | |
7a2508e1 | 47 | static unsigned long mb_cache_shrink(struct mb_cache *cache, |
132d4e2d | 48 | unsigned long nr_to_scan); |
c2f3140f | 49 | |
dc8d5e56 AG |
50 | static inline struct hlist_bl_head *mb_cache_entry_head(struct mb_cache *cache, |
51 | u32 key) | |
f0c8b462 | 52 | { |
dc8d5e56 | 53 | return &cache->c_hash[hash_32(key, cache->c_bucket_bits)]; |
f0c8b462 JK |
54 | } |
55 | ||
c2f3140f JK |
56 | /* |
57 | * Number of entries to reclaim synchronously when there are too many entries | |
58 | * in cache | |
59 | */ | |
60 | #define SYNC_SHRINK_BATCH 64 | |
61 | ||
f9a61eb4 | 62 | /* |
7a2508e1 | 63 | * mb_cache_entry_create - create entry in cache |
f9a61eb4 JK |
64 | * @cache - cache where the entry should be created |
65 | * @mask - gfp mask with which the entry should be allocated | |
66 | * @key - key of the entry | |
c07dfcb4 TE |
67 | * @value - value of the entry |
68 | * @reusable - is the entry reusable by others? | |
f9a61eb4 | 69 | * |
c07dfcb4 TE |
70 | * Creates entry in @cache with key @key and value @value. The function returns |
71 | * -EBUSY if entry with the same key and value already exists in cache. | |
72 | * Otherwise 0 is returned. | |
f9a61eb4 | 73 | */ |
7a2508e1 | 74 | int mb_cache_entry_create(struct mb_cache *cache, gfp_t mask, u32 key, |
c07dfcb4 | 75 | u64 value, bool reusable) |
f9a61eb4 | 76 | { |
7a2508e1 | 77 | struct mb_cache_entry *entry, *dup; |
f9a61eb4 JK |
78 | struct hlist_bl_node *dup_node; |
79 | struct hlist_bl_head *head; | |
80 | ||
c2f3140f JK |
81 | /* Schedule background reclaim if there are too many entries */ |
82 | if (cache->c_entry_count >= cache->c_max_entries) | |
83 | schedule_work(&cache->c_shrink_work); | |
84 | /* Do some sync reclaim if background reclaim cannot keep up */ | |
85 | if (cache->c_entry_count >= 2*cache->c_max_entries) | |
7a2508e1 | 86 | mb_cache_shrink(cache, SYNC_SHRINK_BATCH); |
c2f3140f | 87 | |
7a2508e1 | 88 | entry = kmem_cache_alloc(mb_entry_cache, mask); |
f9a61eb4 JK |
89 | if (!entry) |
90 | return -ENOMEM; | |
91 | ||
f0c8b462 | 92 | INIT_LIST_HEAD(&entry->e_list); |
307af6c8 | 93 | /* Initial hash reference */ |
f9a61eb4 JK |
94 | atomic_set(&entry->e_refcnt, 1); |
95 | entry->e_key = key; | |
c07dfcb4 | 96 | entry->e_value = value; |
6048c64b | 97 | entry->e_reusable = reusable; |
3876bbe2 | 98 | entry->e_referenced = 0; |
dc8d5e56 | 99 | head = mb_cache_entry_head(cache, key); |
f9a61eb4 JK |
100 | hlist_bl_lock(head); |
101 | hlist_bl_for_each_entry(dup, dup_node, head, e_hash_list) { | |
c07dfcb4 | 102 | if (dup->e_key == key && dup->e_value == value) { |
f9a61eb4 | 103 | hlist_bl_unlock(head); |
7a2508e1 | 104 | kmem_cache_free(mb_entry_cache, entry); |
f9a61eb4 JK |
105 | return -EBUSY; |
106 | } | |
107 | } | |
108 | hlist_bl_add_head(&entry->e_hash_list, head); | |
307af6c8 JK |
109 | /* |
110 | * Add entry to LRU list before it can be found by | |
111 | * mb_cache_entry_delete() to avoid races | |
112 | */ | |
f0c8b462 JK |
113 | spin_lock(&cache->c_list_lock); |
114 | list_add_tail(&entry->e_list, &cache->c_list); | |
f9a61eb4 | 115 | cache->c_entry_count++; |
f0c8b462 | 116 | spin_unlock(&cache->c_list_lock); |
307af6c8 | 117 | hlist_bl_unlock(head); |
f9a61eb4 JK |
118 | |
119 | return 0; | |
120 | } | |
7a2508e1 | 121 | EXPORT_SYMBOL(mb_cache_entry_create); |
f9a61eb4 | 122 | |
307af6c8 | 123 | void __mb_cache_entry_free(struct mb_cache *cache, struct mb_cache_entry *entry) |
f9a61eb4 | 124 | { |
307af6c8 JK |
125 | struct hlist_bl_head *head; |
126 | ||
127 | head = mb_cache_entry_head(cache, entry->e_key); | |
128 | hlist_bl_lock(head); | |
129 | hlist_bl_del(&entry->e_hash_list); | |
130 | hlist_bl_unlock(head); | |
7a2508e1 | 131 | kmem_cache_free(mb_entry_cache, entry); |
f9a61eb4 | 132 | } |
7a2508e1 | 133 | EXPORT_SYMBOL(__mb_cache_entry_free); |
f9a61eb4 | 134 | |
3dc96bba JK |
135 | /* |
136 | * mb_cache_entry_wait_unused - wait to be the last user of the entry | |
137 | * | |
138 | * @entry - entry to work on | |
139 | * | |
140 | * Wait to be the last user of the entry. | |
141 | */ | |
142 | void mb_cache_entry_wait_unused(struct mb_cache_entry *entry) | |
143 | { | |
307af6c8 | 144 | wait_var_event(&entry->e_refcnt, atomic_read(&entry->e_refcnt) <= 2); |
3dc96bba JK |
145 | } |
146 | EXPORT_SYMBOL(mb_cache_entry_wait_unused); | |
147 | ||
7a2508e1 JK |
148 | static struct mb_cache_entry *__entry_find(struct mb_cache *cache, |
149 | struct mb_cache_entry *entry, | |
150 | u32 key) | |
f9a61eb4 | 151 | { |
7a2508e1 | 152 | struct mb_cache_entry *old_entry = entry; |
f9a61eb4 JK |
153 | struct hlist_bl_node *node; |
154 | struct hlist_bl_head *head; | |
155 | ||
dc8d5e56 | 156 | head = mb_cache_entry_head(cache, key); |
f9a61eb4 JK |
157 | hlist_bl_lock(head); |
158 | if (entry && !hlist_bl_unhashed(&entry->e_hash_list)) | |
159 | node = entry->e_hash_list.next; | |
160 | else | |
161 | node = hlist_bl_first(head); | |
162 | while (node) { | |
7a2508e1 | 163 | entry = hlist_bl_entry(node, struct mb_cache_entry, |
f9a61eb4 | 164 | e_hash_list); |
307af6c8 JK |
165 | if (entry->e_key == key && entry->e_reusable && |
166 | atomic_inc_not_zero(&entry->e_refcnt)) | |
f9a61eb4 | 167 | goto out; |
f9a61eb4 JK |
168 | node = node->next; |
169 | } | |
170 | entry = NULL; | |
171 | out: | |
172 | hlist_bl_unlock(head); | |
173 | if (old_entry) | |
7a2508e1 | 174 | mb_cache_entry_put(cache, old_entry); |
f9a61eb4 JK |
175 | |
176 | return entry; | |
177 | } | |
178 | ||
179 | /* | |
b649668c | 180 | * mb_cache_entry_find_first - find the first reusable entry with the given key |
f9a61eb4 JK |
181 | * @cache: cache where we should search |
182 | * @key: key to look for | |
183 | * | |
b649668c EB |
184 | * Search in @cache for a reusable entry with key @key. Grabs reference to the |
185 | * first reusable entry found and returns the entry. | |
f9a61eb4 | 186 | */ |
7a2508e1 JK |
187 | struct mb_cache_entry *mb_cache_entry_find_first(struct mb_cache *cache, |
188 | u32 key) | |
f9a61eb4 JK |
189 | { |
190 | return __entry_find(cache, NULL, key); | |
191 | } | |
7a2508e1 | 192 | EXPORT_SYMBOL(mb_cache_entry_find_first); |
f9a61eb4 JK |
193 | |
194 | /* | |
b649668c | 195 | * mb_cache_entry_find_next - find next reusable entry with the same key |
f9a61eb4 JK |
196 | * @cache: cache where we should search |
197 | * @entry: entry to start search from | |
198 | * | |
b649668c EB |
199 | * Finds next reusable entry in the hash chain which has the same key as @entry. |
200 | * If @entry is unhashed (which can happen when deletion of entry races with the | |
201 | * search), finds the first reusable entry in the hash chain. The function drops | |
202 | * reference to @entry and returns with a reference to the found entry. | |
f9a61eb4 | 203 | */ |
7a2508e1 JK |
204 | struct mb_cache_entry *mb_cache_entry_find_next(struct mb_cache *cache, |
205 | struct mb_cache_entry *entry) | |
f9a61eb4 JK |
206 | { |
207 | return __entry_find(cache, entry, entry->e_key); | |
208 | } | |
7a2508e1 | 209 | EXPORT_SYMBOL(mb_cache_entry_find_next); |
f9a61eb4 | 210 | |
6048c64b | 211 | /* |
c07dfcb4 | 212 | * mb_cache_entry_get - get a cache entry by value (and key) |
6048c64b | 213 | * @cache - cache we work with |
c07dfcb4 TE |
214 | * @key - key |
215 | * @value - value | |
6048c64b AG |
216 | */ |
217 | struct mb_cache_entry *mb_cache_entry_get(struct mb_cache *cache, u32 key, | |
c07dfcb4 | 218 | u64 value) |
6048c64b AG |
219 | { |
220 | struct hlist_bl_node *node; | |
221 | struct hlist_bl_head *head; | |
222 | struct mb_cache_entry *entry; | |
223 | ||
224 | head = mb_cache_entry_head(cache, key); | |
225 | hlist_bl_lock(head); | |
226 | hlist_bl_for_each_entry(entry, node, head, e_hash_list) { | |
307af6c8 JK |
227 | if (entry->e_key == key && entry->e_value == value && |
228 | atomic_inc_not_zero(&entry->e_refcnt)) | |
6048c64b | 229 | goto out; |
6048c64b AG |
230 | } |
231 | entry = NULL; | |
232 | out: | |
233 | hlist_bl_unlock(head); | |
234 | return entry; | |
235 | } | |
236 | EXPORT_SYMBOL(mb_cache_entry_get); | |
237 | ||
3dc96bba JK |
238 | /* mb_cache_entry_delete_or_get - remove a cache entry if it has no users |
239 | * @cache - cache we work with | |
240 | * @key - key | |
241 | * @value - value | |
242 | * | |
243 | * Remove entry from cache @cache with key @key and value @value. The removal | |
244 | * happens only if the entry is unused. The function returns NULL in case the | |
245 | * entry was successfully removed or there's no entry in cache. Otherwise the | |
246 | * function grabs reference of the entry that we failed to delete because it | |
247 | * still has users and return it. | |
248 | */ | |
249 | struct mb_cache_entry *mb_cache_entry_delete_or_get(struct mb_cache *cache, | |
250 | u32 key, u64 value) | |
251 | { | |
3dc96bba JK |
252 | struct mb_cache_entry *entry; |
253 | ||
307af6c8 JK |
254 | entry = mb_cache_entry_get(cache, key, value); |
255 | if (!entry) | |
256 | return NULL; | |
257 | ||
258 | /* | |
259 | * Drop the ref we got from mb_cache_entry_get() and the initial hash | |
260 | * ref if we are the last user | |
261 | */ | |
262 | if (atomic_cmpxchg(&entry->e_refcnt, 2, 0) != 2) | |
263 | return entry; | |
3dc96bba | 264 | |
307af6c8 JK |
265 | spin_lock(&cache->c_list_lock); |
266 | if (!list_empty(&entry->e_list)) | |
267 | list_del_init(&entry->e_list); | |
268 | cache->c_entry_count--; | |
269 | spin_unlock(&cache->c_list_lock); | |
270 | __mb_cache_entry_free(cache, entry); | |
3dc96bba JK |
271 | return NULL; |
272 | } | |
273 | EXPORT_SYMBOL(mb_cache_entry_delete_or_get); | |
274 | ||
7a2508e1 | 275 | /* mb_cache_entry_touch - cache entry got used |
f9a61eb4 JK |
276 | * @cache - cache the entry belongs to |
277 | * @entry - entry that got used | |
278 | * | |
f0c8b462 | 279 | * Marks entry as used to give hit higher chances of surviving in cache. |
f9a61eb4 | 280 | */ |
7a2508e1 JK |
281 | void mb_cache_entry_touch(struct mb_cache *cache, |
282 | struct mb_cache_entry *entry) | |
f9a61eb4 | 283 | { |
dc8d5e56 | 284 | entry->e_referenced = 1; |
f9a61eb4 | 285 | } |
7a2508e1 | 286 | EXPORT_SYMBOL(mb_cache_entry_touch); |
f9a61eb4 | 287 | |
7a2508e1 JK |
288 | static unsigned long mb_cache_count(struct shrinker *shrink, |
289 | struct shrink_control *sc) | |
f9a61eb4 | 290 | { |
7a2508e1 JK |
291 | struct mb_cache *cache = container_of(shrink, struct mb_cache, |
292 | c_shrink); | |
f9a61eb4 JK |
293 | |
294 | return cache->c_entry_count; | |
295 | } | |
296 | ||
297 | /* Shrink number of entries in cache */ | |
7a2508e1 | 298 | static unsigned long mb_cache_shrink(struct mb_cache *cache, |
132d4e2d | 299 | unsigned long nr_to_scan) |
f9a61eb4 | 300 | { |
7a2508e1 | 301 | struct mb_cache_entry *entry; |
132d4e2d | 302 | unsigned long shrunk = 0; |
f9a61eb4 | 303 | |
f0c8b462 JK |
304 | spin_lock(&cache->c_list_lock); |
305 | while (nr_to_scan-- && !list_empty(&cache->c_list)) { | |
306 | entry = list_first_entry(&cache->c_list, | |
7a2508e1 | 307 | struct mb_cache_entry, e_list); |
307af6c8 JK |
308 | /* Drop initial hash reference if there is no user */ |
309 | if (entry->e_referenced || | |
310 | atomic_cmpxchg(&entry->e_refcnt, 1, 0) != 1) { | |
dc8d5e56 | 311 | entry->e_referenced = 0; |
918b7306 | 312 | list_move_tail(&entry->e_list, &cache->c_list); |
f0c8b462 JK |
313 | continue; |
314 | } | |
315 | list_del_init(&entry->e_list); | |
f9a61eb4 | 316 | cache->c_entry_count--; |
f0c8b462 | 317 | spin_unlock(&cache->c_list_lock); |
307af6c8 JK |
318 | __mb_cache_entry_free(cache, entry); |
319 | shrunk++; | |
f9a61eb4 | 320 | cond_resched(); |
f0c8b462 | 321 | spin_lock(&cache->c_list_lock); |
f9a61eb4 | 322 | } |
f0c8b462 | 323 | spin_unlock(&cache->c_list_lock); |
f9a61eb4 JK |
324 | |
325 | return shrunk; | |
326 | } | |
327 | ||
7a2508e1 JK |
328 | static unsigned long mb_cache_scan(struct shrinker *shrink, |
329 | struct shrink_control *sc) | |
c2f3140f | 330 | { |
7a2508e1 | 331 | struct mb_cache *cache = container_of(shrink, struct mb_cache, |
c2f3140f | 332 | c_shrink); |
132d4e2d | 333 | return mb_cache_shrink(cache, sc->nr_to_scan); |
c2f3140f JK |
334 | } |
335 | ||
336 | /* We shrink 1/X of the cache when we have too many entries in it */ | |
337 | #define SHRINK_DIVISOR 16 | |
338 | ||
7a2508e1 | 339 | static void mb_cache_shrink_worker(struct work_struct *work) |
c2f3140f | 340 | { |
7a2508e1 JK |
341 | struct mb_cache *cache = container_of(work, struct mb_cache, |
342 | c_shrink_work); | |
343 | mb_cache_shrink(cache, cache->c_max_entries / SHRINK_DIVISOR); | |
c2f3140f JK |
344 | } |
345 | ||
f9a61eb4 | 346 | /* |
7a2508e1 | 347 | * mb_cache_create - create cache |
f9a61eb4 JK |
348 | * @bucket_bits: log2 of the hash table size |
349 | * | |
350 | * Create cache for keys with 2^bucket_bits hash entries. | |
351 | */ | |
7a2508e1 | 352 | struct mb_cache *mb_cache_create(int bucket_bits) |
f9a61eb4 | 353 | { |
7a2508e1 | 354 | struct mb_cache *cache; |
132d4e2d EB |
355 | unsigned long bucket_count = 1UL << bucket_bits; |
356 | unsigned long i; | |
f9a61eb4 | 357 | |
7a2508e1 | 358 | cache = kzalloc(sizeof(struct mb_cache), GFP_KERNEL); |
f9a61eb4 JK |
359 | if (!cache) |
360 | goto err_out; | |
361 | cache->c_bucket_bits = bucket_bits; | |
c2f3140f | 362 | cache->c_max_entries = bucket_count << 4; |
f0c8b462 JK |
363 | INIT_LIST_HEAD(&cache->c_list); |
364 | spin_lock_init(&cache->c_list_lock); | |
6da2ec56 KC |
365 | cache->c_hash = kmalloc_array(bucket_count, |
366 | sizeof(struct hlist_bl_head), | |
367 | GFP_KERNEL); | |
f9a61eb4 JK |
368 | if (!cache->c_hash) { |
369 | kfree(cache); | |
370 | goto err_out; | |
371 | } | |
372 | for (i = 0; i < bucket_count; i++) | |
373 | INIT_HLIST_BL_HEAD(&cache->c_hash[i]); | |
374 | ||
7a2508e1 JK |
375 | cache->c_shrink.count_objects = mb_cache_count; |
376 | cache->c_shrink.scan_objects = mb_cache_scan; | |
f9a61eb4 | 377 | cache->c_shrink.seeks = DEFAULT_SEEKS; |
e33c267a | 378 | if (register_shrinker(&cache->c_shrink, "mbcache-shrinker")) { |
8913f343 CY |
379 | kfree(cache->c_hash); |
380 | kfree(cache); | |
381 | goto err_out; | |
382 | } | |
f9a61eb4 | 383 | |
7a2508e1 | 384 | INIT_WORK(&cache->c_shrink_work, mb_cache_shrink_worker); |
c2f3140f | 385 | |
f9a61eb4 JK |
386 | return cache; |
387 | ||
388 | err_out: | |
f9a61eb4 JK |
389 | return NULL; |
390 | } | |
7a2508e1 | 391 | EXPORT_SYMBOL(mb_cache_create); |
f9a61eb4 JK |
392 | |
393 | /* | |
7a2508e1 | 394 | * mb_cache_destroy - destroy cache |
f9a61eb4 JK |
395 | * @cache: the cache to destroy |
396 | * | |
397 | * Free all entries in cache and cache itself. Caller must make sure nobody | |
398 | * (except shrinker) can reach @cache when calling this. | |
399 | */ | |
7a2508e1 | 400 | void mb_cache_destroy(struct mb_cache *cache) |
f9a61eb4 | 401 | { |
7a2508e1 | 402 | struct mb_cache_entry *entry, *next; |
f9a61eb4 JK |
403 | |
404 | unregister_shrinker(&cache->c_shrink); | |
405 | ||
406 | /* | |
407 | * We don't bother with any locking. Cache must not be used at this | |
408 | * point. | |
409 | */ | |
f0c8b462 | 410 | list_for_each_entry_safe(entry, next, &cache->c_list, e_list) { |
f0c8b462 | 411 | list_del(&entry->e_list); |
f9a61eb4 | 412 | WARN_ON(atomic_read(&entry->e_refcnt) != 1); |
7a2508e1 | 413 | mb_cache_entry_put(cache, entry); |
f9a61eb4 JK |
414 | } |
415 | kfree(cache->c_hash); | |
416 | kfree(cache); | |
f9a61eb4 | 417 | } |
7a2508e1 | 418 | EXPORT_SYMBOL(mb_cache_destroy); |
f9a61eb4 | 419 | |
7a2508e1 | 420 | static int __init mbcache_init(void) |
f9a61eb4 | 421 | { |
7a2508e1 JK |
422 | mb_entry_cache = kmem_cache_create("mbcache", |
423 | sizeof(struct mb_cache_entry), 0, | |
f9a61eb4 | 424 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL); |
21d0f4fa EB |
425 | if (!mb_entry_cache) |
426 | return -ENOMEM; | |
f9a61eb4 JK |
427 | return 0; |
428 | } | |
429 | ||
7a2508e1 | 430 | static void __exit mbcache_exit(void) |
f9a61eb4 | 431 | { |
7a2508e1 | 432 | kmem_cache_destroy(mb_entry_cache); |
f9a61eb4 JK |
433 | } |
434 | ||
7a2508e1 JK |
435 | module_init(mbcache_init) |
436 | module_exit(mbcache_exit) | |
f9a61eb4 JK |
437 | |
438 | MODULE_AUTHOR("Jan Kara <jack@suse.cz>"); | |
439 | MODULE_DESCRIPTION("Meta block cache (for extended attributes)"); | |
440 | MODULE_LICENSE("GPL"); |