mbcache.c 11.9 KB
Newer Older
Jan Kara's avatar
Jan Kara committed
1 2 3 4 5 6
#include <linux/spinlock.h>
#include <linux/slab.h>
#include <linux/list.h>
#include <linux/list_bl.h>
#include <linux/module.h>
#include <linux/sched.h>
Jan Kara's avatar
Jan Kara committed
7
#include <linux/workqueue.h>
Jan Kara's avatar
Jan Kara committed
8
#include <linux/mbcache.h>
Jan Kara's avatar
Jan Kara committed
9 10 11 12

/*
 * Mbcache is a simple key-value store. Keys need not be unique, however
 * key-value pairs are expected to be unique (we use this fact in
13
 * mb_cache_entry_delete()).
Jan Kara's avatar
Jan Kara committed
14 15
 *
 * Ext2 and ext4 use this cache for deduplication of extended attribute blocks.
16 17 18 19 20
 * Ext4 also uses it for deduplication of xattr values stored in inodes.
 * They use hash of data as a key and provide a value that may represent a
 * block or inode number. That's why keys need not be unique (hash of different
 * data may be the same). However user provided value always uniquely
 * identifies a cache entry.
Jan Kara's avatar
Jan Kara committed
21 22 23 24 25 26
 *
 * We provide functions for creation and removal of entries, search by key,
 * and a special "delete entry with given key-value pair" operation. Fixed
 * size hash table is used for fast key lookups.
 */

Jan Kara's avatar
Jan Kara committed
27
struct mb_cache {
Jan Kara's avatar
Jan Kara committed
28 29 30 31
	/* Hash table of entries */
	struct hlist_bl_head	*c_hash;
	/* log2 of hash table size */
	int			c_bucket_bits;
Jan Kara's avatar
Jan Kara committed
32
	/* Maximum entries in cache to avoid degrading hash too much */
33
	unsigned long		c_max_entries;
34 35 36
	/* Protects c_list, c_entry_count */
	spinlock_t		c_list_lock;
	struct list_head	c_list;
Jan Kara's avatar
Jan Kara committed
37 38 39
	/* Number of entries in cache */
	unsigned long		c_entry_count;
	struct shrinker		c_shrink;
Jan Kara's avatar
Jan Kara committed
40 41
	/* Work for shrinking when the cache has too many entries */
	struct work_struct	c_shrink_work;
Jan Kara's avatar
Jan Kara committed
42 43
};

Jan Kara's avatar
Jan Kara committed
44
static struct kmem_cache *mb_entry_cache;
Jan Kara's avatar
Jan Kara committed
45

Jan Kara's avatar
Jan Kara committed
46
static unsigned long mb_cache_shrink(struct mb_cache *cache,
47
				     unsigned long nr_to_scan);
Jan Kara's avatar
Jan Kara committed
48

49 50
static inline struct hlist_bl_head *mb_cache_entry_head(struct mb_cache *cache,
							u32 key)
51
{
52
	return &cache->c_hash[hash_32(key, cache->c_bucket_bits)];
53 54
}

Jan Kara's avatar
Jan Kara committed
55 56 57 58 59 60
/*
 * Number of entries to reclaim synchronously when there are too many entries
 * in cache
 */
#define SYNC_SHRINK_BATCH 64

Jan Kara's avatar
Jan Kara committed
61
/*
Jan Kara's avatar
Jan Kara committed
62
 * mb_cache_entry_create - create entry in cache
Jan Kara's avatar
Jan Kara committed
63 64 65
 * @cache - cache where the entry should be created
 * @mask - gfp mask with which the entry should be allocated
 * @key - key of the entry
66 67
 * @value - value of the entry
 * @reusable - is the entry reusable by others?
Jan Kara's avatar
Jan Kara committed
68
 *
69 70 71
 * Creates entry in @cache with key @key and value @value. The function returns
 * -EBUSY if entry with the same key and value already exists in cache.
 * Otherwise 0 is returned.
Jan Kara's avatar
Jan Kara committed
72
 */
Jan Kara's avatar
Jan Kara committed
73
int mb_cache_entry_create(struct mb_cache *cache, gfp_t mask, u32 key,
74
			  u64 value, bool reusable)
Jan Kara's avatar
Jan Kara committed
75
{
Jan Kara's avatar
Jan Kara committed
76
	struct mb_cache_entry *entry, *dup;
Jan Kara's avatar
Jan Kara committed
77 78 79
	struct hlist_bl_node *dup_node;
	struct hlist_bl_head *head;

Jan Kara's avatar
Jan Kara committed
80 81 82 83 84
	/* Schedule background reclaim if there are too many entries */
	if (cache->c_entry_count >= cache->c_max_entries)
		schedule_work(&cache->c_shrink_work);
	/* Do some sync reclaim if background reclaim cannot keep up */
	if (cache->c_entry_count >= 2*cache->c_max_entries)
Jan Kara's avatar
Jan Kara committed
85
		mb_cache_shrink(cache, SYNC_SHRINK_BATCH);
Jan Kara's avatar
Jan Kara committed
86

Jan Kara's avatar
Jan Kara committed
87
	entry = kmem_cache_alloc(mb_entry_cache, mask);
Jan Kara's avatar
Jan Kara committed
88 89 90
	if (!entry)
		return -ENOMEM;

91
	INIT_LIST_HEAD(&entry->e_list);
Jan Kara's avatar
Jan Kara committed
92 93 94
	/* One ref for hash, one ref returned */
	atomic_set(&entry->e_refcnt, 1);
	entry->e_key = key;
95
	entry->e_value = value;
96
	entry->e_reusable = reusable;
97
	entry->e_referenced = 0;
98
	head = mb_cache_entry_head(cache, key);
Jan Kara's avatar
Jan Kara committed
99 100
	hlist_bl_lock(head);
	hlist_bl_for_each_entry(dup, dup_node, head, e_hash_list) {
101
		if (dup->e_key == key && dup->e_value == value) {
Jan Kara's avatar
Jan Kara committed
102
			hlist_bl_unlock(head);
Jan Kara's avatar
Jan Kara committed
103
			kmem_cache_free(mb_entry_cache, entry);
Jan Kara's avatar
Jan Kara committed
104 105 106 107 108 109
			return -EBUSY;
		}
	}
	hlist_bl_add_head(&entry->e_hash_list, head);
	hlist_bl_unlock(head);

110 111
	spin_lock(&cache->c_list_lock);
	list_add_tail(&entry->e_list, &cache->c_list);
Jan Kara's avatar
Jan Kara committed
112 113 114
	/* Grab ref for LRU list */
	atomic_inc(&entry->e_refcnt);
	cache->c_entry_count++;
115
	spin_unlock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
116 117 118

	return 0;
}
Jan Kara's avatar
Jan Kara committed
119
EXPORT_SYMBOL(mb_cache_entry_create);
Jan Kara's avatar
Jan Kara committed
120

Jan Kara's avatar
Jan Kara committed
121
void __mb_cache_entry_free(struct mb_cache_entry *entry)
Jan Kara's avatar
Jan Kara committed
122
{
Jan Kara's avatar
Jan Kara committed
123
	kmem_cache_free(mb_entry_cache, entry);
Jan Kara's avatar
Jan Kara committed
124
}
Jan Kara's avatar
Jan Kara committed
125
EXPORT_SYMBOL(__mb_cache_entry_free);
Jan Kara's avatar
Jan Kara committed
126

Jan Kara's avatar
Jan Kara committed
127 128 129
static struct mb_cache_entry *__entry_find(struct mb_cache *cache,
					   struct mb_cache_entry *entry,
					   u32 key)
Jan Kara's avatar
Jan Kara committed
130
{
Jan Kara's avatar
Jan Kara committed
131
	struct mb_cache_entry *old_entry = entry;
Jan Kara's avatar
Jan Kara committed
132 133 134
	struct hlist_bl_node *node;
	struct hlist_bl_head *head;

135
	head = mb_cache_entry_head(cache, key);
Jan Kara's avatar
Jan Kara committed
136 137 138 139 140 141
	hlist_bl_lock(head);
	if (entry && !hlist_bl_unhashed(&entry->e_hash_list))
		node = entry->e_hash_list.next;
	else
		node = hlist_bl_first(head);
	while (node) {
Jan Kara's avatar
Jan Kara committed
142
		entry = hlist_bl_entry(node, struct mb_cache_entry,
Jan Kara's avatar
Jan Kara committed
143
				       e_hash_list);
144
		if (entry->e_key == key && entry->e_reusable) {
Jan Kara's avatar
Jan Kara committed
145 146 147 148 149 150 151 152 153
			atomic_inc(&entry->e_refcnt);
			goto out;
		}
		node = node->next;
	}
	entry = NULL;
out:
	hlist_bl_unlock(head);
	if (old_entry)
Jan Kara's avatar
Jan Kara committed
154
		mb_cache_entry_put(cache, old_entry);
Jan Kara's avatar
Jan Kara committed
155 156 157 158 159

	return entry;
}

/*
160
 * mb_cache_entry_find_first - find the first reusable entry with the given key
Jan Kara's avatar
Jan Kara committed
161 162 163
 * @cache: cache where we should search
 * @key: key to look for
 *
164 165
 * Search in @cache for a reusable entry with key @key. Grabs reference to the
 * first reusable entry found and returns the entry.
Jan Kara's avatar
Jan Kara committed
166
 */
Jan Kara's avatar
Jan Kara committed
167 168
struct mb_cache_entry *mb_cache_entry_find_first(struct mb_cache *cache,
						 u32 key)
Jan Kara's avatar
Jan Kara committed
169 170 171
{
	return __entry_find(cache, NULL, key);
}
Jan Kara's avatar
Jan Kara committed
172
EXPORT_SYMBOL(mb_cache_entry_find_first);
Jan Kara's avatar
Jan Kara committed
173 174

/*
175
 * mb_cache_entry_find_next - find next reusable entry with the same key
Jan Kara's avatar
Jan Kara committed
176 177 178
 * @cache: cache where we should search
 * @entry: entry to start search from
 *
179 180 181 182
 * Finds next reusable entry in the hash chain which has the same key as @entry.
 * If @entry is unhashed (which can happen when deletion of entry races with the
 * search), finds the first reusable entry in the hash chain. The function drops
 * reference to @entry and returns with a reference to the found entry.
Jan Kara's avatar
Jan Kara committed
183
 */
Jan Kara's avatar
Jan Kara committed
184 185
struct mb_cache_entry *mb_cache_entry_find_next(struct mb_cache *cache,
						struct mb_cache_entry *entry)
Jan Kara's avatar
Jan Kara committed
186 187 188
{
	return __entry_find(cache, entry, entry->e_key);
}
Jan Kara's avatar
Jan Kara committed
189
EXPORT_SYMBOL(mb_cache_entry_find_next);
Jan Kara's avatar
Jan Kara committed
190

191
/*
192
 * mb_cache_entry_get - get a cache entry by value (and key)
193
 * @cache - cache we work with
194 195
 * @key - key
 * @value - value
196 197
 */
struct mb_cache_entry *mb_cache_entry_get(struct mb_cache *cache, u32 key,
198
					  u64 value)
199 200 201 202 203 204 205 206
{
	struct hlist_bl_node *node;
	struct hlist_bl_head *head;
	struct mb_cache_entry *entry;

	head = mb_cache_entry_head(cache, key);
	hlist_bl_lock(head);
	hlist_bl_for_each_entry(entry, node, head, e_hash_list) {
207
		if (entry->e_key == key && entry->e_value == value) {
208 209 210 211 212 213 214 215 216 217 218
			atomic_inc(&entry->e_refcnt);
			goto out;
		}
	}
	entry = NULL;
out:
	hlist_bl_unlock(head);
	return entry;
}
EXPORT_SYMBOL(mb_cache_entry_get);

219
/* mb_cache_entry_delete - remove a cache entry
Jan Kara's avatar
Jan Kara committed
220
 * @cache - cache we work with
221 222
 * @key - key
 * @value - value
Jan Kara's avatar
Jan Kara committed
223
 *
224
 * Remove entry from cache @cache with key @key and value @value.
Jan Kara's avatar
Jan Kara committed
225
 */
226
void mb_cache_entry_delete(struct mb_cache *cache, u32 key, u64 value)
Jan Kara's avatar
Jan Kara committed
227 228 229
{
	struct hlist_bl_node *node;
	struct hlist_bl_head *head;
Jan Kara's avatar
Jan Kara committed
230
	struct mb_cache_entry *entry;
Jan Kara's avatar
Jan Kara committed
231

232
	head = mb_cache_entry_head(cache, key);
Jan Kara's avatar
Jan Kara committed
233 234
	hlist_bl_lock(head);
	hlist_bl_for_each_entry(entry, node, head, e_hash_list) {
235
		if (entry->e_key == key && entry->e_value == value) {
Jan Kara's avatar
Jan Kara committed
236 237 238
			/* We keep hash list reference to keep entry alive */
			hlist_bl_del_init(&entry->e_hash_list);
			hlist_bl_unlock(head);
239 240 241
			spin_lock(&cache->c_list_lock);
			if (!list_empty(&entry->e_list)) {
				list_del_init(&entry->e_list);
Jan Kara's avatar
Jan Kara committed
242 243 244
				cache->c_entry_count--;
				atomic_dec(&entry->e_refcnt);
			}
245
			spin_unlock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
246
			mb_cache_entry_put(cache, entry);
Jan Kara's avatar
Jan Kara committed
247 248 249 250 251
			return;
		}
	}
	hlist_bl_unlock(head);
}
252
EXPORT_SYMBOL(mb_cache_entry_delete);
Jan Kara's avatar
Jan Kara committed
253

Jan Kara's avatar
Jan Kara committed
254
/* mb_cache_entry_touch - cache entry got used
Jan Kara's avatar
Jan Kara committed
255 256 257
 * @cache - cache the entry belongs to
 * @entry - entry that got used
 *
258
 * Marks entry as used to give hit higher chances of surviving in cache.
Jan Kara's avatar
Jan Kara committed
259
 */
Jan Kara's avatar
Jan Kara committed
260 261
void mb_cache_entry_touch(struct mb_cache *cache,
			  struct mb_cache_entry *entry)
Jan Kara's avatar
Jan Kara committed
262
{
263
	entry->e_referenced = 1;
Jan Kara's avatar
Jan Kara committed
264
}
Jan Kara's avatar
Jan Kara committed
265
EXPORT_SYMBOL(mb_cache_entry_touch);
Jan Kara's avatar
Jan Kara committed
266

Jan Kara's avatar
Jan Kara committed
267 268
static unsigned long mb_cache_count(struct shrinker *shrink,
				    struct shrink_control *sc)
Jan Kara's avatar
Jan Kara committed
269
{
Jan Kara's avatar
Jan Kara committed
270 271
	struct mb_cache *cache = container_of(shrink, struct mb_cache,
					      c_shrink);
Jan Kara's avatar
Jan Kara committed
272 273 274 275 276

	return cache->c_entry_count;
}

/* Shrink number of entries in cache */
Jan Kara's avatar
Jan Kara committed
277
static unsigned long mb_cache_shrink(struct mb_cache *cache,
278
				     unsigned long nr_to_scan)
Jan Kara's avatar
Jan Kara committed
279
{
Jan Kara's avatar
Jan Kara committed
280
	struct mb_cache_entry *entry;
Jan Kara's avatar
Jan Kara committed
281
	struct hlist_bl_head *head;
282
	unsigned long shrunk = 0;
Jan Kara's avatar
Jan Kara committed
283

284 285 286
	spin_lock(&cache->c_list_lock);
	while (nr_to_scan-- && !list_empty(&cache->c_list)) {
		entry = list_first_entry(&cache->c_list,
Jan Kara's avatar
Jan Kara committed
287
					 struct mb_cache_entry, e_list);
288 289
		if (entry->e_referenced) {
			entry->e_referenced = 0;
290
			list_move_tail(&entry->e_list, &cache->c_list);
291 292 293
			continue;
		}
		list_del_init(&entry->e_list);
Jan Kara's avatar
Jan Kara committed
294 295 296 297 298
		cache->c_entry_count--;
		/*
		 * We keep LRU list reference so that entry doesn't go away
		 * from under us.
		 */
299
		spin_unlock(&cache->c_list_lock);
300
		head = mb_cache_entry_head(cache, entry->e_key);
Jan Kara's avatar
Jan Kara committed
301 302 303 304 305 306
		hlist_bl_lock(head);
		if (!hlist_bl_unhashed(&entry->e_hash_list)) {
			hlist_bl_del_init(&entry->e_hash_list);
			atomic_dec(&entry->e_refcnt);
		}
		hlist_bl_unlock(head);
Jan Kara's avatar
Jan Kara committed
307
		if (mb_cache_entry_put(cache, entry))
Jan Kara's avatar
Jan Kara committed
308 309
			shrunk++;
		cond_resched();
310
		spin_lock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
311
	}
312
	spin_unlock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
313 314 315 316

	return shrunk;
}

Jan Kara's avatar
Jan Kara committed
317 318
static unsigned long mb_cache_scan(struct shrinker *shrink,
				   struct shrink_control *sc)
Jan Kara's avatar
Jan Kara committed
319
{
Jan Kara's avatar
Jan Kara committed
320
	struct mb_cache *cache = container_of(shrink, struct mb_cache,
Jan Kara's avatar
Jan Kara committed
321
					      c_shrink);
322
	return mb_cache_shrink(cache, sc->nr_to_scan);
Jan Kara's avatar
Jan Kara committed
323 324 325 326 327
}

/* We shrink 1/X of the cache when we have too many entries in it */
#define SHRINK_DIVISOR 16

Jan Kara's avatar
Jan Kara committed
328
static void mb_cache_shrink_worker(struct work_struct *work)
Jan Kara's avatar
Jan Kara committed
329
{
Jan Kara's avatar
Jan Kara committed
330 331 332
	struct mb_cache *cache = container_of(work, struct mb_cache,
					      c_shrink_work);
	mb_cache_shrink(cache, cache->c_max_entries / SHRINK_DIVISOR);
Jan Kara's avatar
Jan Kara committed
333 334
}

Jan Kara's avatar
Jan Kara committed
335
/*
Jan Kara's avatar
Jan Kara committed
336
 * mb_cache_create - create cache
Jan Kara's avatar
Jan Kara committed
337 338 339 340
 * @bucket_bits: log2 of the hash table size
 *
 * Create cache for keys with 2^bucket_bits hash entries.
 */
Jan Kara's avatar
Jan Kara committed
341
struct mb_cache *mb_cache_create(int bucket_bits)
Jan Kara's avatar
Jan Kara committed
342
{
Jan Kara's avatar
Jan Kara committed
343
	struct mb_cache *cache;
344 345
	unsigned long bucket_count = 1UL << bucket_bits;
	unsigned long i;
Jan Kara's avatar
Jan Kara committed
346

Jan Kara's avatar
Jan Kara committed
347
	cache = kzalloc(sizeof(struct mb_cache), GFP_KERNEL);
Jan Kara's avatar
Jan Kara committed
348 349 350
	if (!cache)
		goto err_out;
	cache->c_bucket_bits = bucket_bits;
Jan Kara's avatar
Jan Kara committed
351
	cache->c_max_entries = bucket_count << 4;
352 353
	INIT_LIST_HEAD(&cache->c_list);
	spin_lock_init(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
354 355 356 357 358 359 360 361 362
	cache->c_hash = kmalloc(bucket_count * sizeof(struct hlist_bl_head),
				GFP_KERNEL);
	if (!cache->c_hash) {
		kfree(cache);
		goto err_out;
	}
	for (i = 0; i < bucket_count; i++)
		INIT_HLIST_BL_HEAD(&cache->c_hash[i]);

Jan Kara's avatar
Jan Kara committed
363 364
	cache->c_shrink.count_objects = mb_cache_count;
	cache->c_shrink.scan_objects = mb_cache_scan;
Jan Kara's avatar
Jan Kara committed
365
	cache->c_shrink.seeks = DEFAULT_SEEKS;
366 367 368 369 370
	if (register_shrinker(&cache->c_shrink)) {
		kfree(cache->c_hash);
		kfree(cache);
		goto err_out;
	}
Jan Kara's avatar
Jan Kara committed
371

Jan Kara's avatar
Jan Kara committed
372
	INIT_WORK(&cache->c_shrink_work, mb_cache_shrink_worker);
Jan Kara's avatar
Jan Kara committed
373

Jan Kara's avatar
Jan Kara committed
374 375 376 377 378
	return cache;

err_out:
	return NULL;
}
Jan Kara's avatar
Jan Kara committed
379
EXPORT_SYMBOL(mb_cache_create);
Jan Kara's avatar
Jan Kara committed
380 381

/*
Jan Kara's avatar
Jan Kara committed
382
 * mb_cache_destroy - destroy cache
Jan Kara's avatar
Jan Kara committed
383 384 385 386 387
 * @cache: the cache to destroy
 *
 * Free all entries in cache and cache itself. Caller must make sure nobody
 * (except shrinker) can reach @cache when calling this.
 */
Jan Kara's avatar
Jan Kara committed
388
void mb_cache_destroy(struct mb_cache *cache)
Jan Kara's avatar
Jan Kara committed
389
{
Jan Kara's avatar
Jan Kara committed
390
	struct mb_cache_entry *entry, *next;
Jan Kara's avatar
Jan Kara committed
391 392 393 394 395 396 397

	unregister_shrinker(&cache->c_shrink);

	/*
	 * We don't bother with any locking. Cache must not be used at this
	 * point.
	 */
398
	list_for_each_entry_safe(entry, next, &cache->c_list, e_list) {
Jan Kara's avatar
Jan Kara committed
399 400 401 402 403
		if (!hlist_bl_unhashed(&entry->e_hash_list)) {
			hlist_bl_del_init(&entry->e_hash_list);
			atomic_dec(&entry->e_refcnt);
		} else
			WARN_ON(1);
404
		list_del(&entry->e_list);
Jan Kara's avatar
Jan Kara committed
405
		WARN_ON(atomic_read(&entry->e_refcnt) != 1);
Jan Kara's avatar
Jan Kara committed
406
		mb_cache_entry_put(cache, entry);
Jan Kara's avatar
Jan Kara committed
407 408 409 410
	}
	kfree(cache->c_hash);
	kfree(cache);
}
Jan Kara's avatar
Jan Kara committed
411
EXPORT_SYMBOL(mb_cache_destroy);
Jan Kara's avatar
Jan Kara committed
412

Jan Kara's avatar
Jan Kara committed
413
static int __init mbcache_init(void)
Jan Kara's avatar
Jan Kara committed
414
{
Jan Kara's avatar
Jan Kara committed
415 416
	mb_entry_cache = kmem_cache_create("mbcache",
				sizeof(struct mb_cache_entry), 0,
Jan Kara's avatar
Jan Kara committed
417
				SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL);
418 419
	if (!mb_entry_cache)
		return -ENOMEM;
Jan Kara's avatar
Jan Kara committed
420 421 422
	return 0;
}

Jan Kara's avatar
Jan Kara committed
423
static void __exit mbcache_exit(void)
Jan Kara's avatar
Jan Kara committed
424
{
Jan Kara's avatar
Jan Kara committed
425
	kmem_cache_destroy(mb_entry_cache);
Jan Kara's avatar
Jan Kara committed
426 427
}

Jan Kara's avatar
Jan Kara committed
428 429
module_init(mbcache_init)
module_exit(mbcache_exit)
Jan Kara's avatar
Jan Kara committed
430 431 432 433

MODULE_AUTHOR("Jan Kara <jack@suse.cz>");
MODULE_DESCRIPTION("Meta block cache (for extended attributes)");
MODULE_LICENSE("GPL");