linux/fs/mbcache.c
<<
>>
Prefs
   1/*
   2 * linux/fs/mbcache.c
   3 * (C) 2001-2002 Andreas Gruenbacher, <a.gruenbacher@computer.org>
   4 */
   5
   6/*
   7 * Filesystem Meta Information Block Cache (mbcache)
   8 *
   9 * The mbcache caches blocks of block devices that need to be located
  10 * by their device/block number, as well as by other criteria (such
  11 * as the block's contents).
  12 *
  13 * There can only be one cache entry in a cache per device and block number.
  14 * Additional indexes need not be unique in this sense. The number of
  15 * additional indexes (=other criteria) can be hardwired at compile time
  16 * or specified at cache create time.
  17 *
  18 * Each cache entry is of fixed size. An entry may be `valid' or `invalid'
  19 * in the cache. A valid entry is in the main hash tables of the cache,
  20 * and may also be in the lru list. An invalid entry is not in any hashes
  21 * or lists.
  22 *
  23 * A valid cache entry is only in the lru list if no handles refer to it.
  24 * Invalid cache entries will be freed when the last handle to the cache
  25 * entry is released. Entries that cannot be freed immediately are put
  26 * back on the lru list.
  27 */
  28
  29#include <linux/kernel.h>
  30#include <linux/module.h>
  31
  32#include <linux/hash.h>
  33#include <linux/fs.h>
  34#include <linux/mm.h>
  35#include <linux/slab.h>
  36#include <linux/sched.h>
  37#include <linux/init.h>
  38#include <linux/mbcache.h>
  39
  40
  41#ifdef MB_CACHE_DEBUG
  42# define mb_debug(f...) do { \
  43                printk(KERN_DEBUG f); \
  44                printk("\n"); \
  45        } while (0)
  46#define mb_assert(c) do { if (!(c)) \
  47                printk(KERN_ERR "assertion " #c " failed\n"); \
  48        } while(0)
  49#else
  50# define mb_debug(f...) do { } while(0)
  51# define mb_assert(c) do { } while(0)
  52#endif
  53#define mb_error(f...) do { \
  54                printk(KERN_ERR f); \
  55                printk("\n"); \
  56        } while(0)
  57
  58#define MB_CACHE_WRITER ((unsigned short)~0U >> 1)
  59
  60static DECLARE_WAIT_QUEUE_HEAD(mb_cache_queue);
  61                
  62MODULE_AUTHOR("Andreas Gruenbacher <a.gruenbacher@computer.org>");
  63MODULE_DESCRIPTION("Meta block cache (for extended attributes)");
  64MODULE_LICENSE("GPL");
  65
  66EXPORT_SYMBOL(mb_cache_create);
  67EXPORT_SYMBOL(mb_cache_shrink);
  68EXPORT_SYMBOL(mb_cache_destroy);
  69EXPORT_SYMBOL(mb_cache_entry_alloc);
  70EXPORT_SYMBOL(mb_cache_entry_insert);
  71EXPORT_SYMBOL(mb_cache_entry_release);
  72EXPORT_SYMBOL(mb_cache_entry_free);
  73EXPORT_SYMBOL(mb_cache_entry_get);
  74#if !defined(MB_CACHE_INDEXES_COUNT) || (MB_CACHE_INDEXES_COUNT > 0)
  75EXPORT_SYMBOL(mb_cache_entry_find_first);
  76EXPORT_SYMBOL(mb_cache_entry_find_next);
  77#endif
  78
  79/*
  80 * Global data: list of all mbcache's, lru list, and a spinlock for
  81 * accessing cache data structures on SMP machines. The lru list is
  82 * global across all mbcaches.
  83 */
  84
  85static LIST_HEAD(mb_cache_list);
  86static LIST_HEAD(mb_cache_lru_list);
  87static DEFINE_SPINLOCK(mb_cache_spinlock);
  88
  89/*
  90 * What the mbcache registers as to get shrunk dynamically.
  91 */
  92
  93static int mb_cache_shrink_fn(struct shrinker *shrink,
  94                              struct shrink_control *sc);
  95
  96static struct shrinker mb_cache_shrinker = {
  97        .shrink = mb_cache_shrink_fn,
  98        .seeks = DEFAULT_SEEKS,
  99};
 100
 101static inline int
 102__mb_cache_entry_is_hashed(struct mb_cache_entry *ce)
 103{
 104        return !list_empty(&ce->e_block_list);
 105}
 106
 107
 108static void
 109__mb_cache_entry_unhash(struct mb_cache_entry *ce)
 110{
 111        if (__mb_cache_entry_is_hashed(ce)) {
 112                list_del_init(&ce->e_block_list);
 113                list_del(&ce->e_index.o_list);
 114        }
 115}
 116
 117
 118static void
 119__mb_cache_entry_forget(struct mb_cache_entry *ce, gfp_t gfp_mask)
 120{
 121        struct mb_cache *cache = ce->e_cache;
 122
 123        mb_assert(!(ce->e_used || ce->e_queued));
 124        kmem_cache_free(cache->c_entry_cache, ce);
 125        atomic_dec(&cache->c_entry_count);
 126}
 127
 128
 129static void
 130__mb_cache_entry_release_unlock(struct mb_cache_entry *ce)
 131        __releases(mb_cache_spinlock)
 132{
 133        /* Wake up all processes queuing for this cache entry. */
 134        if (ce->e_queued)
 135                wake_up_all(&mb_cache_queue);
 136        if (ce->e_used >= MB_CACHE_WRITER)
 137                ce->e_used -= MB_CACHE_WRITER;
 138        ce->e_used--;
 139        if (!(ce->e_used || ce->e_queued)) {
 140                if (!__mb_cache_entry_is_hashed(ce))
 141                        goto forget;
 142                mb_assert(list_empty(&ce->e_lru_list));
 143                list_add_tail(&ce->e_lru_list, &mb_cache_lru_list);
 144        }
 145        spin_unlock(&mb_cache_spinlock);
 146        return;
 147forget:
 148        spin_unlock(&mb_cache_spinlock);
 149        __mb_cache_entry_forget(ce, GFP_KERNEL);
 150}
 151
 152
 153/*
 154 * mb_cache_shrink_fn()  memory pressure callback
 155 *
 156 * This function is called by the kernel memory management when memory
 157 * gets low.
 158 *
 159 * @shrink: (ignored)
 160 * @sc: shrink_control passed from reclaim
 161 *
 162 * Returns the number of objects which are present in the cache.
 163 */
 164static int
 165mb_cache_shrink_fn(struct shrinker *shrink, struct shrink_control *sc)
 166{
 167        LIST_HEAD(free_list);
 168        struct mb_cache *cache;
 169        struct mb_cache_entry *entry, *tmp;
 170        int count = 0;
 171        int nr_to_scan = sc->nr_to_scan;
 172        gfp_t gfp_mask = sc->gfp_mask;
 173
 174        mb_debug("trying to free %d entries", nr_to_scan);
 175        spin_lock(&mb_cache_spinlock);
 176        while (nr_to_scan-- && !list_empty(&mb_cache_lru_list)) {
 177                struct mb_cache_entry *ce =
 178                        list_entry(mb_cache_lru_list.next,
 179                                   struct mb_cache_entry, e_lru_list);
 180                list_move_tail(&ce->e_lru_list, &free_list);
 181                __mb_cache_entry_unhash(ce);
 182        }
 183        list_for_each_entry(cache, &mb_cache_list, c_cache_list) {
 184                mb_debug("cache %s (%d)", cache->c_name,
 185                          atomic_read(&cache->c_entry_count));
 186                count += atomic_read(&cache->c_entry_count);
 187        }
 188        spin_unlock(&mb_cache_spinlock);
 189        list_for_each_entry_safe(entry, tmp, &free_list, e_lru_list) {
 190                __mb_cache_entry_forget(entry, gfp_mask);
 191        }
 192        return (count / 100) * sysctl_vfs_cache_pressure;
 193}
 194
 195
 196/*
 197 * mb_cache_create()  create a new cache
 198 *
 199 * All entries in one cache are equal size. Cache entries may be from
 200 * multiple devices. If this is the first mbcache created, registers
 201 * the cache with kernel memory management. Returns NULL if no more
 202 * memory was available.
 203 *
 204 * @name: name of the cache (informal)
 205 * @bucket_bits: log2(number of hash buckets)
 206 */
 207struct mb_cache *
 208mb_cache_create(const char *name, int bucket_bits)
 209{
 210        int n, bucket_count = 1 << bucket_bits;
 211        struct mb_cache *cache = NULL;
 212
 213        cache = kmalloc(sizeof(struct mb_cache), GFP_KERNEL);
 214        if (!cache)
 215                return NULL;
 216        cache->c_name = name;
 217        atomic_set(&cache->c_entry_count, 0);
 218        cache->c_bucket_bits = bucket_bits;
 219        cache->c_block_hash = kmalloc(bucket_count * sizeof(struct list_head),
 220                                      GFP_KERNEL);
 221        if (!cache->c_block_hash)
 222                goto fail;
 223        for (n=0; n<bucket_count; n++)
 224                INIT_LIST_HEAD(&cache->c_block_hash[n]);
 225        cache->c_index_hash = kmalloc(bucket_count * sizeof(struct list_head),
 226                                      GFP_KERNEL);
 227        if (!cache->c_index_hash)
 228                goto fail;
 229        for (n=0; n<bucket_count; n++)
 230                INIT_LIST_HEAD(&cache->c_index_hash[n]);
 231        cache->c_entry_cache = kmem_cache_create(name,
 232                sizeof(struct mb_cache_entry), 0,
 233                SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD, NULL);
 234        if (!cache->c_entry_cache)
 235                goto fail2;
 236
 237        /*
 238         * Set an upper limit on the number of cache entries so that the hash
 239         * chains won't grow too long.
 240         */
 241        cache->c_max_entries = bucket_count << 4;
 242
 243        spin_lock(&mb_cache_spinlock);
 244        list_add(&cache->c_cache_list, &mb_cache_list);
 245        spin_unlock(&mb_cache_spinlock);
 246        return cache;
 247
 248fail2:
 249        kfree(cache->c_index_hash);
 250
 251fail:
 252        kfree(cache->c_block_hash);
 253        kfree(cache);
 254        return NULL;
 255}
 256
 257
 258/*
 259 * mb_cache_shrink()
 260 *
 261 * Removes all cache entries of a device from the cache. All cache entries
 262 * currently in use cannot be freed, and thus remain in the cache. All others
 263 * are freed.
 264 *
 265 * @bdev: which device's cache entries to shrink
 266 */
 267void
 268mb_cache_shrink(struct block_device *bdev)
 269{
 270        LIST_HEAD(free_list);
 271        struct list_head *l, *ltmp;
 272
 273        spin_lock(&mb_cache_spinlock);
 274        list_for_each_safe(l, ltmp, &mb_cache_lru_list) {
 275                struct mb_cache_entry *ce =
 276                        list_entry(l, struct mb_cache_entry, e_lru_list);
 277                if (ce->e_bdev == bdev) {
 278                        list_move_tail(&ce->e_lru_list, &free_list);
 279                        __mb_cache_entry_unhash(ce);
 280                }
 281        }
 282        spin_unlock(&mb_cache_spinlock);
 283        list_for_each_safe(l, ltmp, &free_list) {
 284                __mb_cache_entry_forget(list_entry(l, struct mb_cache_entry,
 285                                                   e_lru_list), GFP_KERNEL);
 286        }
 287}
 288
 289
 290/*
 291 * mb_cache_destroy()
 292 *
 293 * Shrinks the cache to its minimum possible size (hopefully 0 entries),
 294 * and then destroys it. If this was the last mbcache, un-registers the
 295 * mbcache from kernel memory management.
 296 */
 297void
 298mb_cache_destroy(struct mb_cache *cache)
 299{
 300        LIST_HEAD(free_list);
 301        struct list_head *l, *ltmp;
 302
 303        spin_lock(&mb_cache_spinlock);
 304        list_for_each_safe(l, ltmp, &mb_cache_lru_list) {
 305                struct mb_cache_entry *ce =
 306                        list_entry(l, struct mb_cache_entry, e_lru_list);
 307                if (ce->e_cache == cache) {
 308                        list_move_tail(&ce->e_lru_list, &free_list);
 309                        __mb_cache_entry_unhash(ce);
 310                }
 311        }
 312        list_del(&cache->c_cache_list);
 313        spin_unlock(&mb_cache_spinlock);
 314
 315        list_for_each_safe(l, ltmp, &free_list) {
 316                __mb_cache_entry_forget(list_entry(l, struct mb_cache_entry,
 317                                                   e_lru_list), GFP_KERNEL);
 318        }
 319
 320        if (atomic_read(&cache->c_entry_count) > 0) {
 321                mb_error("cache %s: %d orphaned entries",
 322                          cache->c_name,
 323                          atomic_read(&cache->c_entry_count));
 324        }
 325
 326        kmem_cache_destroy(cache->c_entry_cache);
 327
 328        kfree(cache->c_index_hash);
 329        kfree(cache->c_block_hash);
 330        kfree(cache);
 331}
 332
 333/*
 334 * mb_cache_entry_alloc()
 335 *
 336 * Allocates a new cache entry. The new entry will not be valid initially,
 337 * and thus cannot be looked up yet. It should be filled with data, and
 338 * then inserted into the cache using mb_cache_entry_insert(). Returns NULL
 339 * if no more memory was available.
 340 */
 341struct mb_cache_entry *
 342mb_cache_entry_alloc(struct mb_cache *cache, gfp_t gfp_flags)
 343{
 344        struct mb_cache_entry *ce = NULL;
 345
 346        if (atomic_read(&cache->c_entry_count) >= cache->c_max_entries) {
 347                spin_lock(&mb_cache_spinlock);
 348                if (!list_empty(&mb_cache_lru_list)) {
 349                        ce = list_entry(mb_cache_lru_list.next,
 350                                        struct mb_cache_entry, e_lru_list);
 351                        list_del_init(&ce->e_lru_list);
 352                        __mb_cache_entry_unhash(ce);
 353                }
 354                spin_unlock(&mb_cache_spinlock);
 355        }
 356        if (!ce) {
 357                ce = kmem_cache_alloc(cache->c_entry_cache, gfp_flags);
 358                if (!ce)
 359                        return NULL;
 360                atomic_inc(&cache->c_entry_count);
 361                INIT_LIST_HEAD(&ce->e_lru_list);
 362                INIT_LIST_HEAD(&ce->e_block_list);
 363                ce->e_cache = cache;
 364                ce->e_queued = 0;
 365        }
 366        ce->e_used = 1 + MB_CACHE_WRITER;
 367        return ce;
 368}
 369
 370
 371/*
 372 * mb_cache_entry_insert()
 373 *
 374 * Inserts an entry that was allocated using mb_cache_entry_alloc() into
 375 * the cache. After this, the cache entry can be looked up, but is not yet
 376 * in the lru list as the caller still holds a handle to it. Returns 0 on
 377 * success, or -EBUSY if a cache entry for that device + inode exists
 378 * already (this may happen after a failed lookup, but when another process
 379 * has inserted the same cache entry in the meantime).
 380 *
 381 * @bdev: device the cache entry belongs to
 382 * @block: block number
 383 * @key: lookup key
 384 */
 385int
 386mb_cache_entry_insert(struct mb_cache_entry *ce, struct block_device *bdev,
 387                      sector_t block, unsigned int key)
 388{
 389        struct mb_cache *cache = ce->e_cache;
 390        unsigned int bucket;
 391        struct list_head *l;
 392        int error = -EBUSY;
 393
 394        bucket = hash_long((unsigned long)bdev + (block & 0xffffffff), 
 395                           cache->c_bucket_bits);
 396        spin_lock(&mb_cache_spinlock);
 397        list_for_each_prev(l, &cache->c_block_hash[bucket]) {
 398                struct mb_cache_entry *ce =
 399                        list_entry(l, struct mb_cache_entry, e_block_list);
 400                if (ce->e_bdev == bdev && ce->e_block == block)
 401                        goto out;
 402        }
 403        __mb_cache_entry_unhash(ce);
 404        ce->e_bdev = bdev;
 405        ce->e_block = block;
 406        list_add(&ce->e_block_list, &cache->c_block_hash[bucket]);
 407        ce->e_index.o_key = key;
 408        bucket = hash_long(key, cache->c_bucket_bits);
 409        list_add(&ce->e_index.o_list, &cache->c_index_hash[bucket]);
 410        error = 0;
 411out:
 412        spin_unlock(&mb_cache_spinlock);
 413        return error;
 414}
 415
 416
 417/*
 418 * mb_cache_entry_release()
 419 *
 420 * Release a handle to a cache entry. When the last handle to a cache entry
 421 * is released it is either freed (if it is invalid) or otherwise inserted
 422 * in to the lru list.
 423 */
 424void
 425mb_cache_entry_release(struct mb_cache_entry *ce)
 426{
 427        spin_lock(&mb_cache_spinlock);
 428        __mb_cache_entry_release_unlock(ce);
 429}
 430
 431
 432/*
 433 * mb_cache_entry_free()
 434 *
 435 * This is equivalent to the sequence mb_cache_entry_takeout() --
 436 * mb_cache_entry_release().
 437 */
 438void
 439mb_cache_entry_free(struct mb_cache_entry *ce)
 440{
 441        spin_lock(&mb_cache_spinlock);
 442        mb_assert(list_empty(&ce->e_lru_list));
 443        __mb_cache_entry_unhash(ce);
 444        __mb_cache_entry_release_unlock(ce);
 445}
 446
 447
 448/*
 449 * mb_cache_entry_get()
 45380 *
Get if a cache ent byev: devi/ck: block num. (T ot use  onlntry onache entry
 * ie the cacperev: devi * ak: bl.)(). Returns N> * if sucho a cache entry
de exitry. T   retedame cache entrs : bledary fexclusivvi  succ (ies&qud usl the
writeries&quse().
 */
 356struct mb_cache_entry *
 357cache, struct block_device *bdev,
 358                   sector_t block)
{
        unsigned int bucket;
        struct list_head *l;
 362     a>(struct mb_cache_entry *ce;

        bucket = hash_long((unsigned long)bdev + (block & 0xfffffff/a>
 365     /a>                   cache->c_bucket_bits);
 366        spin_lock(&mb_cache_spinlock);
        list_for_eert(l, &cache->c_block_hash[bucket]) {
             > *ce = list_entry(l, struct mb_cache_entry, e_block_list);
                if (ce->e_bdev == bdev && ce->e_block == block]) {
                     == 
                        if (!list_empty(&ce->e_lru_list
                       >        list_del_init(&ce->e_lru_list);

     /a>             while if (ce->e_used) > 0) {
                       >        ce->e_queued<++a>);
                             gt;e_qutry" class="sref">mb_cac>e_qu">l, &                                             gt;                             gt;spin_unlock(&mb_cache_spinlock);
                             gt;                             gt;spin_lock(&mb_cache_spinlock);
                             gt;ce->e_queued<--a>);
                     /a>}
                     gt;e_qutry" class="sref">mb_cac>e_qu">l, &     /a>             f (ce->e_used = 1 + MB_CACHE_WRITER;

 387                        if (!__mb_cache_entisock_h_used" class="sref">__mb_cache_entisock_h_uock(ce                                __mb_cache_entry_release_unlock(ce);
                                return NULL;
                     /a>}
                        goto              /a>}
     /a>}
 394        ce = NULL;

 397        spin_unlock(&mb_cache_spinlock);
 398        return ce;
}

 401<#   f (!MB_CACINDEXES_COUNTock" class="sref">MB_CACINDEXES_COUNTock|| if (MB_CACINDEXES_COUNTock" class="sref">MB_CACINDEXES_COUNTock) >/a>

 403(struct mb_cache_entry
 404<   __mb_cache_entfeluree(struct list_head *l, struct list_head *     /a>           >, struct block_device *bdev, unsigned int key)
{
 407     while if (l< !a> =                 struct mb_cache_entry *ce =
                      = list_entry(l, struct mb_cache_entry, e_index.o_list);
                if (ce->e_bdev == bdev && ce->e_index.o_key = key<; 0) {
                        
                        if (!list_empty(&ce->e_lru_list
                              = list_del_init(&ce->e_lru_list);
 415
                     /a>
 417 */
                     f (ce->e_used<++a>);
                     while if (ce->e_used) &a> = MB_CACHE_WRITER<; 0) {
                             gt;ce->e_queued<++a>);
                             gt;e_qutry" class="sref">mb_cac>e_qu">l, &                                             gt;                       >        spin_unlock(&mb_cache_spinlock);
                              =      /a>                        spin_lock(&mb_cache_spinlock);
                       >        ce->e_queued<--a>);
                     /a>}
                     f (e_qutry" class="sref">mb_cac>e_qu">l, &
                        if (!__mb_cache_entisock_h_used" class="sref">__mb_cache_entisock_h_uock(ce                             gt;__mb_cache_entry_release_unlock(ce);
                             gt;spin_lock(&mb_cache_spinlock);
                       >        return                      /a>}
     /a>                return ce;
             /a>}
                l< a> = l->     /a>}
        return NULL;
 440}


/*
>__mb_cache_entfelutfers_get()

Feluito tfers_ame cache enton aogivenev: deviwith if ertainkup  i0 on
e_in. Additional matbcastry can fouluiwith on
 448>__mb_cache_entfelutnexi().(). Returns N> * if matbchat wfoulutry.  on
 449   retedame cache entrs : bledary fshareda  succ (ies&qumultiple isadercies&quse().
 45380 *
@fs/mb: ie the cacto search on
 * @bdeie tv: device the cache entshouldtry belngs to
 * @kece tup  i0ece t>e_indeo
 */
(struct mb_cache_entry
mb_cache_entfelutfers_ree(struct cache, struct block_device *bdev,
                       >, unsigned int key)
{
        unsigned int bucket = hash_long(key, cache->c_bucket_bits);
        struct list_head *l;
        struct mb_cache_entry *ce;

        spin_lock(&mb_cache_spinlock);
        l< a> = cache->c_index_hash[bucket. 365     > *ce< a> = __mb_cache_entfelureel, &cache->c_index_hash[bucket, bdevkey;
 366        spin_unlock(&mb_cache_spinlock);
        return ce;
}


/*
>__mb_cache_entfelutnexi()>/*
/*
Feluito tnexiame cache enton aogivenev: deviwith if ertainkup  i0 a0 on
additional >e_in. ). Returns N> * if matbchcouldtrywfoulutry.  ch_pioucess
he entrs atomatically is relea/a>t toat>>__mb_cache_entfelutnexi()try ess
b thelligneikviceis:ess

he ent=>>__mb_cache_entfelutfers_ge;ess
while ihe_en) {ess
>>>>>..e().
>>>>>he ent=>>__mb_cache_entfelutnexi(he_en,>..ee;ess
}ess
 *
@ch_p:ry.  ch_piouc matbc> *
 * @bdeie tv: device the cache entshouldtry belngs to
 * @kece tup  i0ece t>e_indeo
 */
(struct mb_cache_entry
mb_cache_entry *                      >, struct block_device *bdev, unsigned int key)
{
     a>(struct cache< a> =  394        unsigned int bucket = hash_long(key, cache->c_bucket_bits);
     a>(struct list_head *l;
     a>(struct mb_cache_entry *ce;

        spin_lock(&mb_cache_spinlock);
        l< a> = e_index.o_list.        ce< a> = __mb_cache_entfelureel, &cache->c_index_hash[bucket, bdevkey;
        __mb_cache_entry_release_unlock(        return ce;
 403}

MB_CACINDEXES_COUNT/>|| if">MB_CACINDEXES_COUNT/>) > "> */

 407{
                retu0/a>;
}

vint {
     > *}





y. original LXR software byece ta> LXR s="cul_i>key fxr.s="ux.no k>c_ly ho_led byea> mb_came6abfooler"> fxr.s+coprepare_to_wa_init" ce< a> =