drivers/staging/lustre/include/linux/libcfs/libcfs_hash.h

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * libcfs/include/libcfs/libcfs_hash.h
  37  *
  38  * Hashing routines
  39  *
  40  */
  41
  42 #ifndef __LIBCFS_HASH_H__
  43 #define __LIBCFS_HASH_H__
  44 /*
  45  * Knuth recommends primes in approximately golden ratio to the maximum
  46  * integer representable by a machine word for multiplicative hashing.
  47  * Chuck Lever verified the effectiveness of this technique:
  48  * http://www.citi.umich.edu/techreports/reports/citi-tr-00-1.pdf
  49  *
  50  * These primes are chosen to be bit-sparse, that is operations on
  51  * them can use shifts and additions instead of multiplications for
  52  * machines where multiplications are slow.
  53  */
  54 /* 2^31 + 2^29 - 2^25 + 2^22 - 2^19 - 2^16 + 1 */
  55 #define CFS_GOLDEN_RATIO_PRIME_32 0x9e370001UL
  56 /*  2^63 + 2^61 - 2^57 + 2^54 - 2^51 - 2^18 + 1 */
  57 #define CFS_GOLDEN_RATIO_PRIME_64 0x9e37fffffffc0001ULL
  58
  59 /*
  60  * Ideally we would use HAVE_HASH_LONG for this, but on linux we configure
  61  * the linux kernel and user space at the same time, so we need to differentiate
  62  * between them explicitely. If this is not needed on other architectures, then
  63  * we'll need to move the functions to archi specific headers.
  64  */
  65
  66 #include <linux/hash.h>
  67
  68 #define cfs_hash_long(val, bits)    hash_long(val, bits)
  69
  70 /** disable debug */
  71 #define CFS_HASH_DEBUG_NONE      0
  72 /** record hash depth and output to console when it's too deep,
  73  *  computing overhead is low but consume more memory */
  74 #define CFS_HASH_DEBUG_1            1
  75 /** expensive, check key validation */
  76 #define CFS_HASH_DEBUG_2            2
  77
  78 #define CFS_HASH_DEBUG_LEVEL    CFS_HASH_DEBUG_NONE
  79
  80 struct cfs_hash_ops;
  81 struct cfs_hash_lock_ops;
  82 struct cfs_hash_hlist_ops;
  83
  84 typedef union {
  85         rwlock_t                rw;             /**< rwlock */
  86         spinlock_t              spin;           /**< spinlock */
  87 } cfs_hash_lock_t;
  88
  89 /**
  90  * cfs_hash_bucket is a container of:
  91  * - lock, couter ...
  92  * - array of hash-head starting from hsb_head[0], hash-head can be one of
  93  *   . cfs_hash_head_t
  94  *   . cfs_hash_head_dep_t
  95  *   . cfs_hash_dhead_t
  96  *   . cfs_hash_dhead_dep_t
  97  *   which depends on requirement of user
  98  * - some extra bytes (caller can require it while creating hash)
  99  */
 100 typedef struct cfs_hash_bucket {
 101         cfs_hash_lock_t         hsb_lock;       /**< bucket lock */
 102         __u32                   hsb_count;      /**< current entries */
 103         __u32                   hsb_version;    /**< change version */
 104         unsigned int            hsb_index;      /**< index of bucket */
 105         int                     hsb_depmax;     /**< max depth on bucket */
 106         long                    hsb_head[0];    /**< hash-head array */
 107 } cfs_hash_bucket_t;
 108
 109 /**
 110  * cfs_hash bucket descriptor, it's normally in stack of caller
 111  */
 112 typedef struct cfs_hash_bd {
 113         cfs_hash_bucket_t         *bd_bucket;      /**< address of bucket */
 114         unsigned int            bd_offset;      /**< offset in bucket */
 115 } cfs_hash_bd_t;
 116
 117 #define CFS_HASH_NAME_LEN          16      /**< default name length */
 118 #define CFS_HASH_BIGNAME_LEN    64      /**< bigname for param tree */
 119
 120 #define CFS_HASH_BKT_BITS          3       /**< default bits of bucket */
 121 #define CFS_HASH_BITS_MAX          30      /**< max bits of bucket */
 122 #define CFS_HASH_BITS_MIN          CFS_HASH_BKT_BITS
 123
 124 /**
 125  * common hash attributes.
 126  */
 127 enum cfs_hash_tag {
 128         /**
 129          * don't need any lock, caller will protect operations with it's
 130          * own lock. With this flag:
 131          *  . CFS_HASH_NO_BKTLOCK, CFS_HASH_RW_BKTLOCK, CFS_HASH_SPIN_BKTLOCK
 132          *    will be ignored.
 133          *  . Some functions will be disabled with this flag, i.e:
 134          *    cfs_hash_for_each_empty, cfs_hash_rehash
 135          */
 136         CFS_HASH_NO_LOCK        = 1 << 0,
 137         /** no bucket lock, use one spinlock to protect the whole hash */
 138         CFS_HASH_NO_BKTLOCK     = 1 << 1,
 139         /** rwlock to protect bucket */
 140         CFS_HASH_RW_BKTLOCK     = 1 << 2,
 141         /** spinlcok to protect bucket */
 142         CFS_HASH_SPIN_BKTLOCK   = 1 << 3,
 143         /** always add new item to tail */
 144         CFS_HASH_ADD_TAIL       = 1 << 4,
 145         /** hash-table doesn't have refcount on item */
 146         CFS_HASH_NO_ITEMREF     = 1 << 5,
 147         /** big name for param-tree */
 148         CFS_HASH_BIGNAME        = 1 << 6,
 149         /** track global count */
 150         CFS_HASH_COUNTER        = 1 << 7,
 151         /** rehash item by new key */
 152         CFS_HASH_REHASH_KEY     = 1 << 8,
 153         /** Enable dynamic hash resizing */
 154         CFS_HASH_REHASH  = 1 << 9,
 155         /** can shrink hash-size */
 156         CFS_HASH_SHRINK  = 1 << 10,
 157         /** assert hash is empty on exit */
 158         CFS_HASH_ASSERT_EMPTY   = 1 << 11,
 159         /** record hlist depth */
 160         CFS_HASH_DEPTH    = 1 << 12,
 161         /**
 162          * rehash is always scheduled in a different thread, so current
 163          * change on hash table is non-blocking
 164          */
 165         CFS_HASH_NBLK_CHANGE    = 1 << 13,
 166         /** NB, we typed hs_flags as  __u16, please change it
 167          * if you need to extend >=16 flags */
 168 };
 169
 170 /** most used attributes */
 171 #define CFS_HASH_DEFAULT       (CFS_HASH_RW_BKTLOCK | \
 172                                 CFS_HASH_COUNTER | CFS_HASH_REHASH)
 173
 174 /**
 175  * cfs_hash is a hash-table implementation for general purpose, it can support:
 176  *    . two refcount modes
 177  *      hash-table with & without refcount
 178  *    . four lock modes
 179  *      nolock, one-spinlock, rw-bucket-lock, spin-bucket-lock
 180  *    . general operations
 181  *      lookup, add(add_tail or add_head), delete
 182  *    . rehash
 183  *      grows or shrink
 184  *    . iteration
 185  *      locked iteration and unlocked iteration
 186  *    . bigname
 187  *      support long name hash
 188  *    . debug
 189  *      trace max searching depth
 190  *
 191  * Rehash:
 192  * When the htable grows or shrinks, a separate task (cfs_hash_rehash_worker)
 193  * is spawned to handle the rehash in the background, it's possible that other
 194  * processes can concurrently perform additions, deletions, and lookups
 195  * without being blocked on rehash completion, because rehash will release
 196  * the global wrlock for each bucket.
 197  *
 198  * rehash and iteration can't run at the same time because it's too tricky
 199  * to keep both of them safe and correct.
 200  * As they are relatively rare operations, so:
 201  *   . if iteration is in progress while we try to launch rehash, then
 202  *     it just giveup, iterator will launch rehash at the end.
 203  *   . if rehash is in progress while we try to iterate the hash table,
 204  *     then we just wait (shouldn't be very long time), anyway, nobody
 205  *     should expect iteration of whole hash-table to be non-blocking.
 206  *
 207  * During rehashing, a (key,object) pair may be in one of two buckets,
 208  * depending on whether the worker task has yet to transfer the object
 209  * to its new location in the table. Lookups and deletions need to search both
 210  * locations; additions must take care to only insert into the new bucket.
 211  */
 212
 213 typedef struct cfs_hash {
 214         /** serialize with rehash, or serialize all operations if
 215          * the hash-table has CFS_HASH_NO_BKTLOCK */
 216         cfs_hash_lock_t      hs_lock;
 217         /** hash operations */
 218         struct cfs_hash_ops     *hs_ops;
 219         /** hash lock operations */
 220         struct cfs_hash_lock_ops   *hs_lops;
 221         /** hash list operations */
 222         struct cfs_hash_hlist_ops  *hs_hops;
 223         /** hash buckets-table */
 224         cfs_hash_bucket_t        **hs_buckets;
 225         /** total number of items on this hash-table */
 226         atomic_t                hs_count;
 227         /** hash flags, see cfs_hash_tag for detail */
 228         __u16                  hs_flags;
 229         /** # of extra-bytes for bucket, for user saving extended attributes */
 230         __u16                  hs_extra_bytes;
 231         /** wants to iterate */
 232         __u8                    hs_iterating;
 233         /** hash-table is dying */
 234         __u8                    hs_exiting;
 235         /** current hash bits */
 236         __u8                    hs_cur_bits;
 237         /** min hash bits */
 238         __u8                    hs_min_bits;
 239         /** max hash bits */
 240         __u8                    hs_max_bits;
 241         /** bits for rehash */
 242         __u8                    hs_rehash_bits;
 243         /** bits for each bucket */
 244         __u8                    hs_bkt_bits;
 245         /** resize min threshold */
 246         __u16                  hs_min_theta;
 247         /** resize max threshold */
 248         __u16                  hs_max_theta;
 249         /** resize count */
 250         __u32                  hs_rehash_count;
 251         /** # of iterators (caller of cfs_hash_for_each_*) */
 252         __u32                  hs_iterators;
 253         /** rehash workitem */
 254         cfs_workitem_t        hs_rehash_wi;
 255         /** refcount on this hash table */
 256         atomic_t                hs_refcount;
 257         /** rehash buckets-table */
 258         cfs_hash_bucket_t        **hs_rehash_buckets;
 259 #if CFS_HASH_DEBUG_LEVEL >= CFS_HASH_DEBUG_1
 260         /** serialize debug members */
 261         spinlock_t                      hs_dep_lock;
 262         /** max depth */
 263         unsigned int            hs_dep_max;
 264         /** id of the deepest bucket */
 265         unsigned int            hs_dep_bkt;
 266         /** offset in the deepest bucket */
 267         unsigned int            hs_dep_off;
 268         /** bits when we found the max depth */
 269         unsigned int            hs_dep_bits;
 270         /** workitem to output max depth */
 271         cfs_workitem_t        hs_dep_wi;
 272 #endif
 273         /** name of htable */
 274         char                    hs_name[0];
 275 } cfs_hash_t;
 276
 277 typedef struct cfs_hash_lock_ops {
 278         /** lock the hash table */
 279         void    (*hs_lock)(cfs_hash_lock_t *lock, int exclusive);
 280         /** unlock the hash table */
 281         void    (*hs_unlock)(cfs_hash_lock_t *lock, int exclusive);
 282         /** lock the hash bucket */
 283         void    (*hs_bkt_lock)(cfs_hash_lock_t *lock, int exclusive);
 284         /** unlock the hash bucket */
 285         void    (*hs_bkt_unlock)(cfs_hash_lock_t *lock, int exclusive);
 286 } cfs_hash_lock_ops_t;
 287
 288 typedef struct cfs_hash_hlist_ops {
 289         /** return hlist_head of hash-head of @bd */
 290         struct hlist_head *(*hop_hhead)(cfs_hash_t *hs, cfs_hash_bd_t *bd);
 291         /** return hash-head size */
 292         int (*hop_hhead_size)(cfs_hash_t *hs);
 293         /** add @hnode to hash-head of @bd */
 294         int (*hop_hnode_add)(cfs_hash_t *hs,
 295                              cfs_hash_bd_t *bd, struct hlist_node *hnode);
 296         /** remove @hnode from hash-head of @bd */
 297         int (*hop_hnode_del)(cfs_hash_t *hs,
 298                              cfs_hash_bd_t *bd, struct hlist_node *hnode);
 299 } cfs_hash_hlist_ops_t;
 300
 301 typedef struct cfs_hash_ops {
 302         /** return hashed value from @key */
 303         unsigned (*hs_hash)(cfs_hash_t *hs, const void *key, unsigned mask);
 304         /** return key address of @hnode */
 305         void *   (*hs_key)(struct hlist_node *hnode);
 306         /** copy key from @hnode to @key */
 307         void     (*hs_keycpy)(struct hlist_node *hnode, void *key);
 308         /**
 309          *  compare @key with key of @hnode
 310          *  returns 1 on a match
 311          */
 312         int      (*hs_keycmp)(const void *key, struct hlist_node *hnode);
 313         /** return object address of @hnode, i.e: container_of(...hnode) */
 314         void *   (*hs_object)(struct hlist_node *hnode);
 315         /** get refcount of item, always called with holding bucket-lock */
 316         void     (*hs_get)(cfs_hash_t *hs, struct hlist_node *hnode);
 317         /** release refcount of item */
 318         void     (*hs_put)(cfs_hash_t *hs, struct hlist_node *hnode);
 319         /** release refcount of item, always called with holding bucket-lock */
 320         void     (*hs_put_locked)(cfs_hash_t *hs, struct hlist_node *hnode);
 321         /** it's called before removing of @hnode */
 322         void     (*hs_exit)(cfs_hash_t *hs, struct hlist_node *hnode);
 323 } cfs_hash_ops_t;
 324
 325 /** total number of buckets in @hs */
 326 #define CFS_HASH_NBKT(hs)       \
 327         (1U << ((hs)->hs_cur_bits - (hs)->hs_bkt_bits))
 328
 329 /** total number of buckets in @hs while rehashing */
 330 #define CFS_HASH_RH_NBKT(hs)    \
 331         (1U << ((hs)->hs_rehash_bits - (hs)->hs_bkt_bits))
 332
 333 /** number of hlist for in bucket */
 334 #define CFS_HASH_BKT_NHLIST(hs) (1U << (hs)->hs_bkt_bits)
 335
 336 /** total number of hlist in @hs */
 337 #define CFS_HASH_NHLIST(hs)     (1U << (hs)->hs_cur_bits)
 338
 339 /** total number of hlist in @hs while rehashing */
 340 #define CFS_HASH_RH_NHLIST(hs)  (1U << (hs)->hs_rehash_bits)
 341
 342 static inline int
 343 cfs_hash_with_no_lock(cfs_hash_t *hs)
 344 {
 345         /* caller will serialize all operations for this hash-table */
 346         return (hs->hs_flags & CFS_HASH_NO_LOCK) != 0;
 347 }
 348
 349 static inline int
 350 cfs_hash_with_no_bktlock(cfs_hash_t *hs)
 351 {
 352         /* no bucket lock, one single lock to protect the hash-table */
 353         return (hs->hs_flags & CFS_HASH_NO_BKTLOCK) != 0;
 354 }
 355
 356 static inline int
 357 cfs_hash_with_rw_bktlock(cfs_hash_t *hs)
 358 {
 359         /* rwlock to protect hash bucket */
 360         return (hs->hs_flags & CFS_HASH_RW_BKTLOCK) != 0;
 361 }
 362
 363 static inline int
 364 cfs_hash_with_spin_bktlock(cfs_hash_t *hs)
 365 {
 366         /* spinlock to protect hash bucket */
 367         return (hs->hs_flags & CFS_HASH_SPIN_BKTLOCK) != 0;
 368 }
 369
 370 static inline int
 371 cfs_hash_with_add_tail(cfs_hash_t *hs)
 372 {
 373         return (hs->hs_flags & CFS_HASH_ADD_TAIL) != 0;
 374 }
 375
 376 static inline int
 377 cfs_hash_with_no_itemref(cfs_hash_t *hs)
 378 {
 379         /* hash-table doesn't keep refcount on item,
 380          * item can't be removed from hash unless it's
 381          * ZERO refcount */
 382         return (hs->hs_flags & CFS_HASH_NO_ITEMREF) != 0;
 383 }
 384
 385 static inline int
 386 cfs_hash_with_bigname(cfs_hash_t *hs)
 387 {
 388         return (hs->hs_flags & CFS_HASH_BIGNAME) != 0;
 389 }
 390
 391 static inline int
 392 cfs_hash_with_counter(cfs_hash_t *hs)
 393 {
 394         return (hs->hs_flags & CFS_HASH_COUNTER) != 0;
 395 }
 396
 397 static inline int
 398 cfs_hash_with_rehash(cfs_hash_t *hs)
 399 {
 400         return (hs->hs_flags & CFS_HASH_REHASH) != 0;
 401 }
 402
 403 static inline int
 404 cfs_hash_with_rehash_key(cfs_hash_t *hs)
 405 {
 406         return (hs->hs_flags & CFS_HASH_REHASH_KEY) != 0;
 407 }
 408
 409 static inline int
 410 cfs_hash_with_shrink(cfs_hash_t *hs)
 411 {
 412         return (hs->hs_flags & CFS_HASH_SHRINK) != 0;
 413 }
 414
 415 static inline int
 416 cfs_hash_with_assert_empty(cfs_hash_t *hs)
 417 {
 418         return (hs->hs_flags & CFS_HASH_ASSERT_EMPTY) != 0;
 419 }
 420
 421 static inline int
 422 cfs_hash_with_depth(cfs_hash_t *hs)
 423 {
 424         return (hs->hs_flags & CFS_HASH_DEPTH) != 0;
 425 }
 426
 427 static inline int
 428 cfs_hash_with_nblk_change(cfs_hash_t *hs)
 429 {
 430         return (hs->hs_flags & CFS_HASH_NBLK_CHANGE) != 0;
 431 }
 432
 433 static inline int
 434 cfs_hash_is_exiting(cfs_hash_t *hs)
 435 {       /* cfs_hash_destroy is called */
 436         return hs->hs_exiting;
 437 }
 438
 439 static inline int
 440 cfs_hash_is_rehashing(cfs_hash_t *hs)
 441 {       /* rehash is launched */
 442         return hs->hs_rehash_bits != 0;
 443 }
 444
 445 static inline int
 446 cfs_hash_is_iterating(cfs_hash_t *hs)
 447 {       /* someone is calling cfs_hash_for_each_* */
 448         return hs->hs_iterating || hs->hs_iterators != 0;
 449 }
 450
 451 static inline int
 452 cfs_hash_bkt_size(cfs_hash_t *hs)
 453 {
 454         return offsetof(cfs_hash_bucket_t, hsb_head[0]) +
 455                hs->hs_hops->hop_hhead_size(hs) * CFS_HASH_BKT_NHLIST(hs) +
 456                hs->hs_extra_bytes;
 457 }
 458
 459 #define CFS_HOP(hs, op)    (hs)->hs_ops->hs_ ## op
 460
 461 static inline unsigned
 462 cfs_hash_id(cfs_hash_t *hs, const void *key, unsigned mask)
 463 {
 464         return CFS_HOP(hs, hash)(hs, key, mask);
 465 }
 466
 467 static inline void *
 468 cfs_hash_key(cfs_hash_t *hs, struct hlist_node *hnode)
 469 {
 470         return CFS_HOP(hs, key)(hnode);
 471 }
 472
 473 static inline void
 474 cfs_hash_keycpy(cfs_hash_t *hs, struct hlist_node *hnode, void *key)
 475 {
 476         if (CFS_HOP(hs, keycpy) != NULL)
 477                 CFS_HOP(hs, keycpy)(hnode, key);
 478 }
 479
 480 /**
 481  * Returns 1 on a match,
 482  */
 483 static inline int
 484 cfs_hash_keycmp(cfs_hash_t *hs, const void *key, struct hlist_node *hnode)
 485 {
 486         return CFS_HOP(hs, keycmp)(key, hnode);
 487 }
 488
 489 static inline void *
 490 cfs_hash_object(cfs_hash_t *hs, struct hlist_node *hnode)
 491 {
 492         return CFS_HOP(hs, object)(hnode);
 493 }
 494
 495 static inline void
 496 cfs_hash_get(cfs_hash_t *hs, struct hlist_node *hnode)
 497 {
 498         return CFS_HOP(hs, get)(hs, hnode);
 499 }
 500
 501 static inline void
 502 cfs_hash_put_locked(cfs_hash_t *hs, struct hlist_node *hnode)
 503 {
 504         LASSERT(CFS_HOP(hs, put_locked) != NULL);
 505
 506         return CFS_HOP(hs, put_locked)(hs, hnode);
 507 }
 508
 509 static inline void
 510 cfs_hash_put(cfs_hash_t *hs, struct hlist_node *hnode)
 511 {
 512         LASSERT(CFS_HOP(hs, put) != NULL);
 513
 514         return CFS_HOP(hs, put)(hs, hnode);
 515 }
 516
 517 static inline void
 518 cfs_hash_exit(cfs_hash_t *hs, struct hlist_node *hnode)
 519 {
 520         if (CFS_HOP(hs, exit))
 521                 CFS_HOP(hs, exit)(hs, hnode);
 522 }
 523
 524 static inline void cfs_hash_lock(cfs_hash_t *hs, int excl)
 525 {
 526         hs->hs_lops->hs_lock(&hs->hs_lock, excl);
 527 }
 528
 529 static inline void cfs_hash_unlock(cfs_hash_t *hs, int excl)
 530 {
 531         hs->hs_lops->hs_unlock(&hs->hs_lock, excl);
 532 }
 533
 534 static inline int cfs_hash_dec_and_lock(cfs_hash_t *hs,
 535                                         atomic_t *condition)
 536 {
 537         LASSERT(cfs_hash_with_no_bktlock(hs));
 538         return atomic_dec_and_lock(condition, &hs->hs_lock.spin);
 539 }
 540
 541 static inline void cfs_hash_bd_lock(cfs_hash_t *hs,
 542                                     cfs_hash_bd_t *bd, int excl)
 543 {
 544         hs->hs_lops->hs_bkt_lock(&bd->bd_bucket->hsb_lock, excl);
 545 }
 546
 547 static inline void cfs_hash_bd_unlock(cfs_hash_t *hs,
 548                                       cfs_hash_bd_t *bd, int excl)
 549 {
 550         hs->hs_lops->hs_bkt_unlock(&bd->bd_bucket->hsb_lock, excl);
 551 }
 552
 553 /**
 554  * operations on cfs_hash bucket (bd: bucket descriptor),
 555  * they are normally for hash-table without rehash
 556  */
 557 void cfs_hash_bd_get(cfs_hash_t *hs, const void *key, cfs_hash_bd_t *bd);
 558
 559 static inline void cfs_hash_bd_get_and_lock(cfs_hash_t *hs, const void *key,
 560                                             cfs_hash_bd_t *bd, int excl)
 561 {
 562         cfs_hash_bd_get(hs, key, bd);
 563         cfs_hash_bd_lock(hs, bd, excl);
 564 }
 565
 566 static inline unsigned cfs_hash_bd_index_get(cfs_hash_t *hs, cfs_hash_bd_t *bd)
 567 {
 568         return bd->bd_offset | (bd->bd_bucket->hsb_index << hs->hs_bkt_bits);
 569 }
 570
 571 static inline void cfs_hash_bd_index_set(cfs_hash_t *hs,
 572                                          unsigned index, cfs_hash_bd_t *bd)
 573 {
 574         bd->bd_bucket = hs->hs_buckets[index >> hs->hs_bkt_bits];
 575         bd->bd_offset = index & (CFS_HASH_BKT_NHLIST(hs) - 1U);
 576 }
 577
 578 static inline void *
 579 cfs_hash_bd_extra_get(cfs_hash_t *hs, cfs_hash_bd_t *bd)
 580 {
 581         return (void *)bd->bd_bucket +
 582                cfs_hash_bkt_size(hs) - hs->hs_extra_bytes;
 583 }
 584
 585 static inline __u32
 586 cfs_hash_bd_version_get(cfs_hash_bd_t *bd)
 587 {
 588         /* need hold cfs_hash_bd_lock */
 589         return bd->bd_bucket->hsb_version;
 590 }
 591
 592 static inline __u32
 593 cfs_hash_bd_count_get(cfs_hash_bd_t *bd)
 594 {
 595         /* need hold cfs_hash_bd_lock */
 596         return bd->bd_bucket->hsb_count;
 597 }
 598
 599 static inline int
 600 cfs_hash_bd_depmax_get(cfs_hash_bd_t *bd)
 601 {
 602         return bd->bd_bucket->hsb_depmax;
 603 }
 604
 605 static inline int
 606 cfs_hash_bd_compare(cfs_hash_bd_t *bd1, cfs_hash_bd_t *bd2)
 607 {
 608         if (bd1->bd_bucket->hsb_index != bd2->bd_bucket->hsb_index)
 609                 return bd1->bd_bucket->hsb_index - bd2->bd_bucket->hsb_index;
 610
 611         if (bd1->bd_offset != bd2->bd_offset)
 612                 return bd1->bd_offset - bd2->bd_offset;
 613
 614         return 0;
 615 }
 616
 617 void cfs_hash_bd_add_locked(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 618                             struct hlist_node *hnode);
 619 void cfs_hash_bd_del_locked(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 620                             struct hlist_node *hnode);
 621 void cfs_hash_bd_move_locked(cfs_hash_t *hs, cfs_hash_bd_t *bd_old,
 622                              cfs_hash_bd_t *bd_new, struct hlist_node *hnode);
 623
 624 static inline int cfs_hash_bd_dec_and_lock(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 625                                            atomic_t *condition)
 626 {
 627         LASSERT(cfs_hash_with_spin_bktlock(hs));
 628         return atomic_dec_and_lock(condition,
 629                                        &bd->bd_bucket->hsb_lock.spin);
 630 }
 631
 632 static inline struct hlist_head *cfs_hash_bd_hhead(cfs_hash_t *hs,
 633                                                   cfs_hash_bd_t *bd)
 634 {
 635         return hs->hs_hops->hop_hhead(hs, bd);
 636 }
 637
 638 struct hlist_node *cfs_hash_bd_lookup_locked(cfs_hash_t *hs,
 639                                             cfs_hash_bd_t *bd, const void *key);
 640 struct hlist_node *cfs_hash_bd_peek_locked(cfs_hash_t *hs,
 641                                           cfs_hash_bd_t *bd, const void *key);
 642 struct hlist_node *cfs_hash_bd_findadd_locked(cfs_hash_t *hs,
 643                                              cfs_hash_bd_t *bd, const void *key,
 644                                              struct hlist_node *hnode,
 645                                              int insist_add);
 646 struct hlist_node *cfs_hash_bd_finddel_locked(cfs_hash_t *hs,
 647                                              cfs_hash_bd_t *bd, const void *key,
 648                                              struct hlist_node *hnode);
 649
 650 /**
 651  * operations on cfs_hash bucket (bd: bucket descriptor),
 652  * they are safe for hash-table with rehash
 653  */
 654 void cfs_hash_dual_bd_get(cfs_hash_t *hs, const void *key, cfs_hash_bd_t *bds);
 655 void cfs_hash_dual_bd_lock(cfs_hash_t *hs, cfs_hash_bd_t *bds, int excl);
 656 void cfs_hash_dual_bd_unlock(cfs_hash_t *hs, cfs_hash_bd_t *bds, int excl);
 657
 658 static inline void cfs_hash_dual_bd_get_and_lock(cfs_hash_t *hs, const void *key,
 659                                                  cfs_hash_bd_t *bds, int excl)
 660 {
 661         cfs_hash_dual_bd_get(hs, key, bds);
 662         cfs_hash_dual_bd_lock(hs, bds, excl);
 663 }
 664
 665 struct hlist_node *cfs_hash_dual_bd_lookup_locked(cfs_hash_t *hs,
 666                                                  cfs_hash_bd_t *bds,
 667                                                  const void *key);
 668 struct hlist_node *cfs_hash_dual_bd_findadd_locked(cfs_hash_t *hs,
 669                                                   cfs_hash_bd_t *bds,
 670                                                   const void *key,
 671                                                   struct hlist_node *hnode,
 672                                                   int insist_add);
 673 struct hlist_node *cfs_hash_dual_bd_finddel_locked(cfs_hash_t *hs,
 674                                                   cfs_hash_bd_t *bds,
 675                                                   const void *key,
 676                                                   struct hlist_node *hnode);
 677
 678 /* Hash init/cleanup functions */
 679 cfs_hash_t *cfs_hash_create(char *name, unsigned cur_bits, unsigned max_bits,
 680                             unsigned bkt_bits, unsigned extra_bytes,
 681                             unsigned min_theta, unsigned max_theta,
 682                             cfs_hash_ops_t *ops, unsigned flags);
 683
 684 cfs_hash_t *cfs_hash_getref(cfs_hash_t *hs);
 685 void cfs_hash_putref(cfs_hash_t *hs);
 686
 687 /* Hash addition functions */
 688 void cfs_hash_add(cfs_hash_t *hs, const void *key,
 689                   struct hlist_node *hnode);
 690 int cfs_hash_add_unique(cfs_hash_t *hs, const void *key,
 691                         struct hlist_node *hnode);
 692 void *cfs_hash_findadd_unique(cfs_hash_t *hs, const void *key,
 693                               struct hlist_node *hnode);
 694
 695 /* Hash deletion functions */
 696 void *cfs_hash_del(cfs_hash_t *hs, const void *key, struct hlist_node *hnode);
 697 void *cfs_hash_del_key(cfs_hash_t *hs, const void *key);
 698
 699 /* Hash lookup/for_each functions */
 700 #define CFS_HASH_LOOP_HOG       1024
 701
 702 typedef int (*cfs_hash_for_each_cb_t)(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 703                                       struct hlist_node *node, void *data);
 704 void *cfs_hash_lookup(cfs_hash_t *hs, const void *key);
 705 void cfs_hash_for_each(cfs_hash_t *hs, cfs_hash_for_each_cb_t, void *data);
 706 void cfs_hash_for_each_safe(cfs_hash_t *hs, cfs_hash_for_each_cb_t, void *data);
 707 int  cfs_hash_for_each_nolock(cfs_hash_t *hs,
 708                               cfs_hash_for_each_cb_t, void *data);
 709 int  cfs_hash_for_each_empty(cfs_hash_t *hs,
 710                              cfs_hash_for_each_cb_t, void *data);
 711 void cfs_hash_for_each_key(cfs_hash_t *hs, const void *key,
 712                            cfs_hash_for_each_cb_t, void *data);
 713 typedef int (*cfs_hash_cond_opt_cb_t)(void *obj, void *data);
 714 void cfs_hash_cond_del(cfs_hash_t *hs, cfs_hash_cond_opt_cb_t, void *data);
 715
 716 void cfs_hash_hlist_for_each(cfs_hash_t *hs, unsigned hindex,
 717                              cfs_hash_for_each_cb_t, void *data);
 718 int  cfs_hash_is_empty(cfs_hash_t *hs);
 719 __u64 cfs_hash_size_get(cfs_hash_t *hs);
 720
 721 /*
 722  * Rehash - Theta is calculated to be the average chained
 723  * hash depth assuming a perfectly uniform hash function.
 724  */
 725 void cfs_hash_rehash_cancel_locked(cfs_hash_t *hs);
 726 void cfs_hash_rehash_cancel(cfs_hash_t *hs);
 727 int  cfs_hash_rehash(cfs_hash_t *hs, int do_rehash);
 728 void cfs_hash_rehash_key(cfs_hash_t *hs, const void *old_key,
 729                          void *new_key, struct hlist_node *hnode);
 730
 731 #if CFS_HASH_DEBUG_LEVEL > CFS_HASH_DEBUG_1
 732 /* Validate hnode references the correct key */
 733 static inline void
 734 cfs_hash_key_validate(cfs_hash_t *hs, const void *key,
 735                       struct hlist_node *hnode)
 736 {
 737         LASSERT(cfs_hash_keycmp(hs, key, hnode));
 738 }
 739
 740 /* Validate hnode is in the correct bucket */
 741 static inline void
 742 cfs_hash_bucket_validate(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 743                          struct hlist_node *hnode)
 744 {
 745         cfs_hash_bd_t   bds[2];
 746
 747         cfs_hash_dual_bd_get(hs, cfs_hash_key(hs, hnode), bds);
 748         LASSERT(bds[0].bd_bucket == bd->bd_bucket ||
 749                 bds[1].bd_bucket == bd->bd_bucket);
 750 }
 751
 752 #else /* CFS_HASH_DEBUG_LEVEL > CFS_HASH_DEBUG_1 */
 753
 754 static inline void
 755 cfs_hash_key_validate(cfs_hash_t *hs, const void *key,
 756                       struct hlist_node *hnode) {}
 757
 758 static inline void
 759 cfs_hash_bucket_validate(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 760                          struct hlist_node *hnode) {}
 761
 762 #endif /* CFS_HASH_DEBUG_LEVEL */
 763
 764 #define CFS_HASH_THETA_BITS  10
 765 #define CFS_HASH_MIN_THETA  (1U << (CFS_HASH_THETA_BITS - 1))
 766 #define CFS_HASH_MAX_THETA  (1U << (CFS_HASH_THETA_BITS + 1))
 767
 768 /* Return integer component of theta */
 769 static inline int __cfs_hash_theta_int(int theta)
 770 {
 771         return (theta >> CFS_HASH_THETA_BITS);
 772 }
 773
 774 /* Return a fractional value between 0 and 999 */
 775 static inline int __cfs_hash_theta_frac(int theta)
 776 {
 777         return ((theta * 1000) >> CFS_HASH_THETA_BITS) -
 778                (__cfs_hash_theta_int(theta) * 1000);
 779 }
 780
 781 static inline int __cfs_hash_theta(cfs_hash_t *hs)
 782 {
 783         return (atomic_read(&hs->hs_count) <<
 784                 CFS_HASH_THETA_BITS) >> hs->hs_cur_bits;
 785 }
 786
 787 static inline void __cfs_hash_set_theta(cfs_hash_t *hs, int min, int max)
 788 {
 789         LASSERT(min < max);
 790         hs->hs_min_theta = (__u16)min;
 791         hs->hs_max_theta = (__u16)max;
 792 }
 793
 794 /* Generic debug formatting routines mainly for proc handler */
 795 struct seq_file;
 796 int cfs_hash_debug_header(struct seq_file *m);
 797 int cfs_hash_debug_str(cfs_hash_t *hs, struct seq_file *m);
 798
 799 /*
 800  * Generic djb2 hash algorithm for character arrays.
 801  */
 802 static inline unsigned
 803 cfs_hash_djb2_hash(const void *key, size_t size, unsigned mask)
 804 {
 805         unsigned i, hash = 5381;
 806
 807         LASSERT(key != NULL);
 808
 809         for (i = 0; i < size; i++)
 810                 hash = hash * 33 + ((char *)key)[i];
 811
 812         return (hash & mask);
 813 }
 814
 815 /*
 816  * Generic u32 hash algorithm.
 817  */
 818 static inline unsigned
 819 cfs_hash_u32_hash(const __u32 key, unsigned mask)
 820 {
 821         return ((key * CFS_GOLDEN_RATIO_PRIME_32) & mask);
 822 }
 823
 824 /*
 825  * Generic u64 hash algorithm.
 826  */
 827 static inline unsigned
 828 cfs_hash_u64_hash(const __u64 key, unsigned mask)
 829 {
 830         return ((unsigned)(key * CFS_GOLDEN_RATIO_PRIME_64) & mask);
 831 }
 832
 833 /** iterate over all buckets in @bds (array of cfs_hash_bd_t) */
 834 #define cfs_hash_for_each_bd(bds, n, i) \
 835         for (i = 0; i < n && (bds)[i].bd_bucket != NULL; i++)
 836
 837 /** iterate over all buckets of @hs */
 838 #define cfs_hash_for_each_bucket(hs, bd, pos)              \
 839         for (pos = 0;                                      \
 840              pos < CFS_HASH_NBKT(hs) &&                  \
 841              ((bd)->bd_bucket = (hs)->hs_buckets[pos]) != NULL; pos++)
 842
 843 /** iterate over all hlist of bucket @bd */
 844 #define cfs_hash_bd_for_each_hlist(hs, bd, hlist)              \
 845         for ((bd)->bd_offset = 0;                              \
 846              (bd)->bd_offset < CFS_HASH_BKT_NHLIST(hs) &&       \
 847              (hlist = cfs_hash_bd_hhead(hs, bd)) != NULL;       \
 848              (bd)->bd_offset++)
 849
 850 /* !__LIBCFS__HASH_H__ */
 851 #endif