1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * libcfs/include/libcfs/libcfs_hash.h
42 #ifndef __LIBCFS_HASH_H__
43 #define __LIBCFS_HASH_H__
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
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.
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
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.
66 #if (defined __linux__ && defined __KERNEL__)
67 #include <linux/hash.h>
69 /* Fast hashing routine for a long.
70 (C) 2002 William Lee Irwin III, IBM */
72 #if BITS_PER_LONG == 32
73 /* 2^31 + 2^29 - 2^25 + 2^22 - 2^19 - 2^16 + 1 */
74 #define CFS_GOLDEN_RATIO_PRIME CFS_GOLDEN_RATIO_PRIME_32
75 #elif BITS_PER_LONG == 64
76 /* 2^63 + 2^61 - 2^57 + 2^54 - 2^51 - 2^18 + 1 */
77 #define CFS_GOLDEN_RATIO_PRIME CFS_GOLDEN_RATIO_PRIME_64
79 #error Define CFS_GOLDEN_RATIO_PRIME for your wordsize.
82 static inline unsigned long hash_long(unsigned long val, unsigned int bits)
84 unsigned long hash = val;
86 #if BITS_PER_LONG == 64
87 /* Sigh, gcc can't optimise this alone like it does for 32 bits. */
88 unsigned long n = hash;
102 /* On some cpus multiply is faster, on others gcc will do shifts */
103 hash *= CFS_GOLDEN_RATIO_PRIME;
106 /* High bits are more random, so use them. */
107 return hash >> (BITS_PER_LONG - bits);
110 static inline unsigned long hash_ptr(void *ptr, unsigned int bits)
112 return hash_long((unsigned long)ptr, bits);
116 /* !(__linux__ && __KERNEL__) */
121 typedef struct cfs_hash_bucket {
122 struct hlist_head hsb_head; /* entries list */
123 atomic_t hsb_count; /* current entries */
124 rwlock_t hsb_rwlock; /* cfs_hash_bucket */
127 #define CFS_MAX_HASH_NAME 16
129 typedef struct cfs_hash {
130 int hs_cur_bits; /* current hash bits */
131 int hs_cur_mask; /* current hash mask */
132 int hs_min_bits; /* min hash bits */
133 int hs_max_bits; /* max hash bits */
134 int hs_min_theta; /* resize min threshold */
135 int hs_max_theta; /* resize max threshold */
136 int hs_flags; /* hash flags */
137 atomic_t hs_count; /* current entries */
138 atomic_t hs_rehash_count;/* resize count */
139 struct cfs_hash_bucket **hs_buckets; /* hash buckets */
140 struct cfs_hash_ops *hs_ops; /* hash operations */
141 rwlock_t hs_rwlock; /* cfs_hash */
142 char hs_name[CFS_MAX_HASH_NAME];
145 typedef struct cfs_hash_ops {
146 unsigned (*hs_hash)(cfs_hash_t *hs, void *key, unsigned mask);
147 void * (*hs_key)(struct hlist_node *hnode);
148 int (*hs_compare)(void *key, struct hlist_node *hnode);
149 void * (*hs_get)(struct hlist_node *hnode);
150 void * (*hs_put)(struct hlist_node *hnode);
151 void (*hs_exit)(struct hlist_node *hnode);
154 #define CFS_HASH_DEBUG 0x0001 /* Enable expensive debug checks */
155 #define CFS_HASH_REHASH 0x0002 /* Enable dynamic hash resizing */
157 #define CFS_HO(hs) (hs)->hs_ops
158 #define CFS_HOP(hs, op) (hs)->hs_ops->hs_ ## op
160 static inline unsigned
161 cfs_hash_id(cfs_hash_t *hs, void *key, unsigned mask)
165 LASSERT(CFS_HOP(hs, hash));
167 return CFS_HOP(hs, hash)(hs, key, mask);
171 cfs_hash_key(cfs_hash_t *hs, struct hlist_node *hnode)
177 if (CFS_HOP(hs, key))
178 return CFS_HOP(hs, key)(hnode);
183 /* Returns 1 on a match,
184 * XXX: This would be better if it returned, -1, 0, or 1 for
185 * <, =, > respectivly. It could then be used to implement
186 * a CFS_HASH_SORT feature flags which could keep each hash
187 * bucket in order. This would increase insertion times
188 * but could reduce lookup times for deep chains. Ideally,
189 * the rehash should keep chain depth short but if that
190 * ends up not being the case this would be a nice feature.
193 cfs_hash_compare(cfs_hash_t *hs, void *key, struct hlist_node *hnode)
199 if (CFS_HOP(hs, compare))
200 return CFS_HOP(hs, compare)(key, hnode);
206 cfs_hash_get(cfs_hash_t *hs, struct hlist_node *hnode)
212 if (CFS_HOP(hs, get))
213 return CFS_HOP(hs, get)(hnode);
219 cfs_hash_put(cfs_hash_t *hs, struct hlist_node *hnode)
225 if (CFS_HOP(hs, put))
226 return CFS_HOP(hs, put)(hnode);
232 cfs_hash_exit(cfs_hash_t *hs, struct hlist_node *hnode)
238 if (CFS_HOP(hs, exit))
239 return CFS_HOP(hs, exit)(hnode);
242 /* Validate hnode references the correct key */
244 __cfs_hash_key_validate(cfs_hash_t *hs, void *key,
245 struct hlist_node *hnode)
247 if (unlikely(hs->hs_flags & CFS_HASH_DEBUG))
248 LASSERT(cfs_hash_compare(hs, key, hnode) > 0);
251 /* Validate hnode is in the correct bucket */
253 __cfs_hash_bucket_validate(cfs_hash_t *hs, cfs_hash_bucket_t *hsb,
254 struct hlist_node *hnode)
258 if (unlikely(hs->hs_flags & CFS_HASH_DEBUG)) {
259 i = cfs_hash_id(hs, cfs_hash_key(hs, hnode), hs->hs_cur_mask);
260 LASSERT(hs->hs_buckets[i] == hsb);
264 static inline struct hlist_node *
265 __cfs_hash_bucket_lookup(cfs_hash_t *hs,
266 cfs_hash_bucket_t *hsb, void *key)
268 struct hlist_node *hnode;
270 hlist_for_each(hnode, &hsb->hsb_head)
271 if (cfs_hash_compare(hs, key, hnode) > 0)
278 __cfs_hash_bucket_add(cfs_hash_t *hs,
279 cfs_hash_bucket_t *hsb,
280 struct hlist_node *hnode)
282 hlist_add_head(hnode, &(hsb->hsb_head));
283 atomic_inc(&hsb->hsb_count);
284 atomic_inc(&hs->hs_count);
286 return cfs_hash_get(hs, hnode);
290 __cfs_hash_bucket_del(cfs_hash_t *hs,
291 cfs_hash_bucket_t *hsb,
292 struct hlist_node *hnode)
294 hlist_del_init(hnode);
295 LASSERT(atomic_read(&hsb->hsb_count) > 0);
296 atomic_dec(&hsb->hsb_count);
297 LASSERT(atomic_read(&hs->hs_count) > 0);
298 atomic_dec(&hs->hs_count);
300 return cfs_hash_put(hs, hnode);
303 /* Hash init/cleanup functions */
304 cfs_hash_t *cfs_hash_create(char *name, unsigned int cur_bits,
305 unsigned int max_bits,
306 cfs_hash_ops_t *ops, int flags);
307 void cfs_hash_destroy(cfs_hash_t *hs);
309 /* Hash addition functions */
310 void cfs_hash_add(cfs_hash_t *hs, void *key,
311 struct hlist_node *hnode);
312 int cfs_hash_add_unique(cfs_hash_t *hs, void *key,
313 struct hlist_node *hnode);
314 void *cfs_hash_findadd_unique(cfs_hash_t *hs, void *key,
315 struct hlist_node *hnode);
317 /* Hash deletion functions */
318 void *cfs_hash_del(cfs_hash_t *hs, void *key, struct hlist_node *hnode);
319 void *cfs_hash_del_key(cfs_hash_t *hs, void *key);
321 /* Hash lookup/for_each functions */
322 void *cfs_hash_lookup(cfs_hash_t *hs, void *key);
323 typedef void (*cfs_hash_for_each_cb_t)(void *obj, void *data);
324 void cfs_hash_for_each(cfs_hash_t *hs, cfs_hash_for_each_cb_t, void *data);
325 void cfs_hash_for_each_safe(cfs_hash_t *hs, cfs_hash_for_each_cb_t, void *data);
326 void cfs_hash_for_each_empty(cfs_hash_t *hs, cfs_hash_for_each_cb_t, void *data);
327 void cfs_hash_for_each_key(cfs_hash_t *hs, void *key,
328 cfs_hash_for_each_cb_t, void *data);
331 * Rehash - Theta is calculated to be the average chained
332 * hash depth assuming a perfectly uniform hash funcion.
334 int cfs_hash_rehash(cfs_hash_t *hs, int bits);
335 void cfs_hash_rehash_key(cfs_hash_t *hs, void *old_key,
336 void *new_key, struct hlist_node *hnode);
339 #define CFS_HASH_THETA_BITS 10
341 /* Return integer component of theta */
342 static inline int __cfs_hash_theta_int(int theta)
344 return (theta >> CFS_HASH_THETA_BITS);
347 /* Return a fractional value between 0 and 999 */
348 static inline int __cfs_hash_theta_frac(int theta)
350 return ((theta * 1000) >> CFS_HASH_THETA_BITS) -
351 (__cfs_hash_theta_int(theta) * 1000);
354 static inline int __cfs_hash_theta(cfs_hash_t *hs)
356 return (atomic_read(&hs->hs_count) <<
357 CFS_HASH_THETA_BITS) >> hs->hs_cur_bits;
360 static inline void __cfs_hash_set_theta(cfs_hash_t *hs, int min, int max)
363 hs->hs_min_theta = min;
364 hs->hs_max_theta = max;
367 /* Generic debug formatting routines mainly for proc handler */
368 int cfs_hash_debug_header(char *str, int size);
369 int cfs_hash_debug_str(cfs_hash_t *hs, char *str, int size);
372 * Generic djb2 hash algorithm for character arrays.
374 static inline unsigned
375 cfs_hash_djb2_hash(void *key, size_t size, unsigned mask)
377 unsigned i, hash = 5381;
379 LASSERT(key != NULL);
381 for (i = 0; i < size; i++)
382 hash = hash * 33 + ((char *)key)[i];
384 return (hash & mask);
388 * Generic u32 hash algorithm.
390 static inline unsigned
391 cfs_hash_u32_hash(__u32 key, unsigned mask)
393 return ((key * CFS_GOLDEN_RATIO_PRIME_32) & mask);
397 * Generic u64 hash algorithm.
399 static inline unsigned
400 cfs_hash_u64_hash(__u64 key, unsigned mask)
402 return ((unsigned)(key * CFS_GOLDEN_RATIO_PRIME_64) & mask);
405 #define cfs_hash_for_each_bucket(hs, hsb, pos) \
407 pos <= hs->hs_cur_mask && \
408 ({ hsb = hs->hs_buckets[i]; 1; }); \
411 /* !__LIBCFS__HASH_H__ */