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