mbcache.c 11.1 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
Jan Kara's avatar
Jan Kara committed
13
 * mb_cache_entry_delete_block()).
Jan Kara's avatar
Jan Kara committed
14 15 16 17 18 19 20 21 22 23 24 25
 *
 * Ext2 and ext4 use this cache for deduplication of extended attribute blocks.
 * They use hash of a block contents as a key and block number as a value.
 * That's why keys need not be unique (different xattr blocks may end up having
 * the same hash). However block number always uniquely identifies a cache
 * entry.
 *
 * 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
26
struct mb_cache {
Jan Kara's avatar
Jan Kara committed
27 28 29 30
	/* 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
31 32
	/* Maximum entries in cache to avoid degrading hash too much */
	int			c_max_entries;
33 34 35
	/* Protects c_list, c_entry_count */
	spinlock_t		c_list_lock;
	struct list_head	c_list;
Jan Kara's avatar
Jan Kara committed
36 37 38
	/* Number of entries in cache */
	unsigned long		c_entry_count;
	struct shrinker		c_shrink;
Jan Kara's avatar
Jan Kara committed
39 40
	/* Work for shrinking when the cache has too many entries */
	struct work_struct	c_shrink_work;
Jan Kara's avatar
Jan Kara committed
41 42
};

Jan Kara's avatar
Jan Kara committed
43
static struct kmem_cache *mb_entry_cache;
Jan Kara's avatar
Jan Kara committed
44

Jan Kara's avatar
Jan Kara committed
45 46
static unsigned long mb_cache_shrink(struct mb_cache *cache,
				     unsigned int nr_to_scan);
Jan Kara's avatar
Jan Kara committed
47

48 49
static inline struct hlist_bl_head *mb_cache_entry_head(struct mb_cache *cache,
							u32 key)
50
{
51
	return &cache->c_hash[hash_32(key, cache->c_bucket_bits)];
52 53
}

Jan Kara's avatar
Jan Kara committed
54 55 56 57 58 59
/*
 * 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
60
/*
Jan Kara's avatar
Jan Kara committed
61
 * mb_cache_entry_create - create entry in cache
Jan Kara's avatar
Jan Kara committed
62 63 64 65 66 67 68 69 70
 * @cache - cache where the entry should be created
 * @mask - gfp mask with which the entry should be allocated
 * @key - key of the entry
 * @block - block that contains data
 *
 * Creates entry in @cache with key @key and records that data is stored in
 * block @block. The function returns -EBUSY if entry with the same key
 * and for the same block already exists in cache. Otherwise 0 is returned.
 */
Jan Kara's avatar
Jan Kara committed
71 72
int mb_cache_entry_create(struct mb_cache *cache, gfp_t mask, u32 key,
			  sector_t block)
Jan Kara's avatar
Jan Kara committed
73
{
Jan Kara's avatar
Jan Kara committed
74
	struct mb_cache_entry *entry, *dup;
Jan Kara's avatar
Jan Kara committed
75 76 77
	struct hlist_bl_node *dup_node;
	struct hlist_bl_head *head;

Jan Kara's avatar
Jan Kara committed
78 79 80 81 82
	/* 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
83
		mb_cache_shrink(cache, SYNC_SHRINK_BATCH);
Jan Kara's avatar
Jan Kara committed
84

Jan Kara's avatar
Jan Kara committed
85
	entry = kmem_cache_alloc(mb_entry_cache, mask);
Jan Kara's avatar
Jan Kara committed
86 87 88
	if (!entry)
		return -ENOMEM;

89
	INIT_LIST_HEAD(&entry->e_list);
Jan Kara's avatar
Jan Kara committed
90 91 92 93
	/* One ref for hash, one ref returned */
	atomic_set(&entry->e_refcnt, 1);
	entry->e_key = key;
	entry->e_block = block;
94
	head = mb_cache_entry_head(cache, key);
Jan Kara's avatar
Jan Kara committed
95 96 97 98
	hlist_bl_lock(head);
	hlist_bl_for_each_entry(dup, dup_node, head, e_hash_list) {
		if (dup->e_key == key && dup->e_block == block) {
			hlist_bl_unlock(head);
Jan Kara's avatar
Jan Kara committed
99
			kmem_cache_free(mb_entry_cache, entry);
Jan Kara's avatar
Jan Kara committed
100 101 102 103 104 105
			return -EBUSY;
		}
	}
	hlist_bl_add_head(&entry->e_hash_list, head);
	hlist_bl_unlock(head);

106 107
	spin_lock(&cache->c_list_lock);
	list_add_tail(&entry->e_list, &cache->c_list);
Jan Kara's avatar
Jan Kara committed
108 109 110
	/* Grab ref for LRU list */
	atomic_inc(&entry->e_refcnt);
	cache->c_entry_count++;
111
	spin_unlock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
112 113 114

	return 0;
}
Jan Kara's avatar
Jan Kara committed
115
EXPORT_SYMBOL(mb_cache_entry_create);
Jan Kara's avatar
Jan Kara committed
116

Jan Kara's avatar
Jan Kara committed
117
void __mb_cache_entry_free(struct mb_cache_entry *entry)
Jan Kara's avatar
Jan Kara committed
118
{
Jan Kara's avatar
Jan Kara committed
119
	kmem_cache_free(mb_entry_cache, entry);
Jan Kara's avatar
Jan Kara committed
120
}
Jan Kara's avatar
Jan Kara committed
121
EXPORT_SYMBOL(__mb_cache_entry_free);
Jan Kara's avatar
Jan Kara committed
122

Jan Kara's avatar
Jan Kara committed
123 124 125
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
126
{
Jan Kara's avatar
Jan Kara committed
127
	struct mb_cache_entry *old_entry = entry;
Jan Kara's avatar
Jan Kara committed
128 129 130
	struct hlist_bl_node *node;
	struct hlist_bl_head *head;

131
	head = mb_cache_entry_head(cache, key);
Jan Kara's avatar
Jan Kara committed
132 133 134 135 136 137
	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
138
		entry = hlist_bl_entry(node, struct mb_cache_entry,
Jan Kara's avatar
Jan Kara committed
139 140 141 142 143 144 145 146 147 148 149
				       e_hash_list);
		if (entry->e_key == key) {
			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
150
		mb_cache_entry_put(cache, old_entry);
Jan Kara's avatar
Jan Kara committed
151 152 153 154 155

	return entry;
}

/*
Jan Kara's avatar
Jan Kara committed
156
 * mb_cache_entry_find_first - find the first entry in cache with given key
Jan Kara's avatar
Jan Kara committed
157 158 159 160 161 162
 * @cache: cache where we should search
 * @key: key to look for
 *
 * Search in @cache for entry with key @key. Grabs reference to the first
 * entry found and returns the entry.
 */
Jan Kara's avatar
Jan Kara committed
163 164
struct mb_cache_entry *mb_cache_entry_find_first(struct mb_cache *cache,
						 u32 key)
Jan Kara's avatar
Jan Kara committed
165 166 167
{
	return __entry_find(cache, NULL, key);
}
Jan Kara's avatar
Jan Kara committed
168
EXPORT_SYMBOL(mb_cache_entry_find_first);
Jan Kara's avatar
Jan Kara committed
169 170

/*
Jan Kara's avatar
Jan Kara committed
171
 * mb_cache_entry_find_next - find next entry in cache with the same
Jan Kara's avatar
Jan Kara committed
172 173 174 175 176 177 178 179
 * @cache: cache where we should search
 * @entry: entry to start search from
 *
 * Finds next 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 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
180 181
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
182 183 184
{
	return __entry_find(cache, entry, entry->e_key);
}
Jan Kara's avatar
Jan Kara committed
185
EXPORT_SYMBOL(mb_cache_entry_find_next);
Jan Kara's avatar
Jan Kara committed
186

Jan Kara's avatar
Jan Kara committed
187
/* mb_cache_entry_delete_block - remove information about block from cache
Jan Kara's avatar
Jan Kara committed
188 189 190 191 192 193
 * @cache - cache we work with
 * @key - key of the entry to remove
 * @block - block containing data for @key
 *
 * Remove entry from cache @cache with key @key with data stored in @block.
 */
Jan Kara's avatar
Jan Kara committed
194 195
void mb_cache_entry_delete_block(struct mb_cache *cache, u32 key,
				 sector_t block)
Jan Kara's avatar
Jan Kara committed
196 197 198
{
	struct hlist_bl_node *node;
	struct hlist_bl_head *head;
Jan Kara's avatar
Jan Kara committed
199
	struct mb_cache_entry *entry;
Jan Kara's avatar
Jan Kara committed
200

201
	head = mb_cache_entry_head(cache, key);
Jan Kara's avatar
Jan Kara committed
202 203 204 205 206 207
	hlist_bl_lock(head);
	hlist_bl_for_each_entry(entry, node, head, e_hash_list) {
		if (entry->e_key == key && entry->e_block == block) {
			/* We keep hash list reference to keep entry alive */
			hlist_bl_del_init(&entry->e_hash_list);
			hlist_bl_unlock(head);
208 209 210
			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
211 212 213
				cache->c_entry_count--;
				atomic_dec(&entry->e_refcnt);
			}
214
			spin_unlock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
215
			mb_cache_entry_put(cache, entry);
Jan Kara's avatar
Jan Kara committed
216 217 218 219 220
			return;
		}
	}
	hlist_bl_unlock(head);
}
Jan Kara's avatar
Jan Kara committed
221
EXPORT_SYMBOL(mb_cache_entry_delete_block);
Jan Kara's avatar
Jan Kara committed
222

Jan Kara's avatar
Jan Kara committed
223
/* mb_cache_entry_touch - cache entry got used
Jan Kara's avatar
Jan Kara committed
224 225 226
 * @cache - cache the entry belongs to
 * @entry - entry that got used
 *
227
 * Marks entry as used to give hit higher chances of surviving in cache.
Jan Kara's avatar
Jan Kara committed
228
 */
Jan Kara's avatar
Jan Kara committed
229 230
void mb_cache_entry_touch(struct mb_cache *cache,
			  struct mb_cache_entry *entry)
Jan Kara's avatar
Jan Kara committed
231
{
232
	entry->e_referenced = 1;
Jan Kara's avatar
Jan Kara committed
233
}
Jan Kara's avatar
Jan Kara committed
234
EXPORT_SYMBOL(mb_cache_entry_touch);
Jan Kara's avatar
Jan Kara committed
235

Jan Kara's avatar
Jan Kara committed
236 237
static unsigned long mb_cache_count(struct shrinker *shrink,
				    struct shrink_control *sc)
Jan Kara's avatar
Jan Kara committed
238
{
Jan Kara's avatar
Jan Kara committed
239 240
	struct mb_cache *cache = container_of(shrink, struct mb_cache,
					      c_shrink);
Jan Kara's avatar
Jan Kara committed
241 242 243 244 245

	return cache->c_entry_count;
}

/* Shrink number of entries in cache */
Jan Kara's avatar
Jan Kara committed
246 247
static unsigned long mb_cache_shrink(struct mb_cache *cache,
				     unsigned int nr_to_scan)
Jan Kara's avatar
Jan Kara committed
248
{
Jan Kara's avatar
Jan Kara committed
249
	struct mb_cache_entry *entry;
Jan Kara's avatar
Jan Kara committed
250 251 252
	struct hlist_bl_head *head;
	unsigned int shrunk = 0;

253 254 255
	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
256
					 struct mb_cache_entry, e_list);
257 258
		if (entry->e_referenced) {
			entry->e_referenced = 0;
259 260 261 262
			list_move_tail(&cache->c_list, &entry->e_list);
			continue;
		}
		list_del_init(&entry->e_list);
Jan Kara's avatar
Jan Kara committed
263 264 265 266 267
		cache->c_entry_count--;
		/*
		 * We keep LRU list reference so that entry doesn't go away
		 * from under us.
		 */
268
		spin_unlock(&cache->c_list_lock);
269
		head = mb_cache_entry_head(cache, entry->e_key);
Jan Kara's avatar
Jan Kara committed
270 271 272 273 274 275
		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
276
		if (mb_cache_entry_put(cache, entry))
Jan Kara's avatar
Jan Kara committed
277 278
			shrunk++;
		cond_resched();
279
		spin_lock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
280
	}
281
	spin_unlock(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
282 283 284 285

	return shrunk;
}

Jan Kara's avatar
Jan Kara committed
286 287
static unsigned long mb_cache_scan(struct shrinker *shrink,
				   struct shrink_control *sc)
Jan Kara's avatar
Jan Kara committed
288 289
{
	int nr_to_scan = sc->nr_to_scan;
Jan Kara's avatar
Jan Kara committed
290
	struct mb_cache *cache = container_of(shrink, struct mb_cache,
Jan Kara's avatar
Jan Kara committed
291
					      c_shrink);
Jan Kara's avatar
Jan Kara committed
292
	return mb_cache_shrink(cache, nr_to_scan);
Jan Kara's avatar
Jan Kara committed
293 294 295 296 297
}

/* 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
298
static void mb_cache_shrink_worker(struct work_struct *work)
Jan Kara's avatar
Jan Kara committed
299
{
Jan Kara's avatar
Jan Kara committed
300 301 302
	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
303 304
}

Jan Kara's avatar
Jan Kara committed
305
/*
Jan Kara's avatar
Jan Kara committed
306
 * mb_cache_create - create cache
Jan Kara's avatar
Jan Kara committed
307 308 309 310
 * @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
311
struct mb_cache *mb_cache_create(int bucket_bits)
Jan Kara's avatar
Jan Kara committed
312
{
Jan Kara's avatar
Jan Kara committed
313
	struct mb_cache *cache;
Jan Kara's avatar
Jan Kara committed
314 315 316 317 318 319
	int bucket_count = 1 << bucket_bits;
	int i;

	if (!try_module_get(THIS_MODULE))
		return NULL;

Jan Kara's avatar
Jan Kara committed
320
	cache = kzalloc(sizeof(struct mb_cache), GFP_KERNEL);
Jan Kara's avatar
Jan Kara committed
321 322 323
	if (!cache)
		goto err_out;
	cache->c_bucket_bits = bucket_bits;
Jan Kara's avatar
Jan Kara committed
324
	cache->c_max_entries = bucket_count << 4;
325 326
	INIT_LIST_HEAD(&cache->c_list);
	spin_lock_init(&cache->c_list_lock);
Jan Kara's avatar
Jan Kara committed
327 328 329 330 331 332 333 334 335
	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
336 337
	cache->c_shrink.count_objects = mb_cache_count;
	cache->c_shrink.scan_objects = mb_cache_scan;
Jan Kara's avatar
Jan Kara committed
338 339 340
	cache->c_shrink.seeks = DEFAULT_SEEKS;
	register_shrinker(&cache->c_shrink);

Jan Kara's avatar
Jan Kara committed
341
	INIT_WORK(&cache->c_shrink_work, mb_cache_shrink_worker);
Jan Kara's avatar
Jan Kara committed
342

Jan Kara's avatar
Jan Kara committed
343 344 345 346 347 348
	return cache;

err_out:
	module_put(THIS_MODULE);
	return NULL;
}
Jan Kara's avatar
Jan Kara committed
349
EXPORT_SYMBOL(mb_cache_create);
Jan Kara's avatar
Jan Kara committed
350 351

/*
Jan Kara's avatar
Jan Kara committed
352
 * mb_cache_destroy - destroy cache
Jan Kara's avatar
Jan Kara committed
353 354 355 356 357
 * @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
358
void mb_cache_destroy(struct mb_cache *cache)
Jan Kara's avatar
Jan Kara committed
359
{
Jan Kara's avatar
Jan Kara committed
360
	struct mb_cache_entry *entry, *next;
Jan Kara's avatar
Jan Kara committed
361 362 363 364 365 366 367

	unregister_shrinker(&cache->c_shrink);

	/*
	 * We don't bother with any locking. Cache must not be used at this
	 * point.
	 */
368
	list_for_each_entry_safe(entry, next, &cache->c_list, e_list) {
Jan Kara's avatar
Jan Kara committed
369 370 371 372 373
		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);
374
		list_del(&entry->e_list);
Jan Kara's avatar
Jan Kara committed
375
		WARN_ON(atomic_read(&entry->e_refcnt) != 1);
Jan Kara's avatar
Jan Kara committed
376
		mb_cache_entry_put(cache, entry);
Jan Kara's avatar
Jan Kara committed
377 378 379 380 381
	}
	kfree(cache->c_hash);
	kfree(cache);
	module_put(THIS_MODULE);
}
Jan Kara's avatar
Jan Kara committed
382
EXPORT_SYMBOL(mb_cache_destroy);
Jan Kara's avatar
Jan Kara committed
383

Jan Kara's avatar
Jan Kara committed
384
static int __init mbcache_init(void)
Jan Kara's avatar
Jan Kara committed
385
{
Jan Kara's avatar
Jan Kara committed
386 387
	mb_entry_cache = kmem_cache_create("mbcache",
				sizeof(struct mb_cache_entry), 0,
Jan Kara's avatar
Jan Kara committed
388
				SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL);
Jan Kara's avatar
Jan Kara committed
389
	BUG_ON(!mb_entry_cache);
Jan Kara's avatar
Jan Kara committed
390 391 392
	return 0;
}

Jan Kara's avatar
Jan Kara committed
393
static void __exit mbcache_exit(void)
Jan Kara's avatar
Jan Kara committed
394
{
Jan Kara's avatar
Jan Kara committed
395
	kmem_cache_destroy(mb_entry_cache);
Jan Kara's avatar
Jan Kara committed
396 397
}

Jan Kara's avatar
Jan Kara committed
398 399
module_init(mbcache_init)
module_exit(mbcache_exit)
Jan Kara's avatar
Jan Kara committed
400 401 402 403

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