fs/mbcache.c

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