Loading...
1/*
2 * Resizable, Scalable, Concurrent Hash Table
3 *
4 * Copyright (c) 2015-2016 Herbert Xu <herbert@gondor.apana.org.au>
5 * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
6 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
7 *
8 * Code partially derived from nft_hash
9 * Rewritten with rehash code from br_multicast plus single list
10 * pointer as suggested by Josh Triplett
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 */
16
17#ifndef _LINUX_RHASHTABLE_H
18#define _LINUX_RHASHTABLE_H
19
20#include <linux/atomic.h>
21#include <linux/compiler.h>
22#include <linux/err.h>
23#include <linux/errno.h>
24#include <linux/jhash.h>
25#include <linux/list_nulls.h>
26#include <linux/workqueue.h>
27#include <linux/mutex.h>
28#include <linux/rcupdate.h>
29
30/*
31 * The end of the chain is marked with a special nulls marks which has
32 * the following format:
33 *
34 * +-------+-----------------------------------------------------+-+
35 * | Base | Hash |1|
36 * +-------+-----------------------------------------------------+-+
37 *
38 * Base (4 bits) : Reserved to distinguish between multiple tables.
39 * Specified via &struct rhashtable_params.nulls_base.
40 * Hash (27 bits): Full hash (unmasked) of first element added to bucket
41 * 1 (1 bit) : Nulls marker (always set)
42 *
43 * The remaining bits of the next pointer remain unused for now.
44 */
45#define RHT_BASE_BITS 4
46#define RHT_HASH_BITS 27
47#define RHT_BASE_SHIFT RHT_HASH_BITS
48
49/* Base bits plus 1 bit for nulls marker */
50#define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1)
51
52struct rhash_head {
53 struct rhash_head __rcu *next;
54};
55
56struct rhlist_head {
57 struct rhash_head rhead;
58 struct rhlist_head __rcu *next;
59};
60
61/**
62 * struct bucket_table - Table of hash buckets
63 * @size: Number of hash buckets
64 * @rehash: Current bucket being rehashed
65 * @hash_rnd: Random seed to fold into hash
66 * @locks_mask: Mask to apply before accessing locks[]
67 * @locks: Array of spinlocks protecting individual buckets
68 * @walkers: List of active walkers
69 * @rcu: RCU structure for freeing the table
70 * @future_tbl: Table under construction during rehashing
71 * @buckets: size * hash buckets
72 */
73struct bucket_table {
74 unsigned int size;
75 unsigned int rehash;
76 u32 hash_rnd;
77 unsigned int locks_mask;
78 spinlock_t *locks;
79 struct list_head walkers;
80 struct rcu_head rcu;
81
82 struct bucket_table __rcu *future_tbl;
83
84 struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp;
85};
86
87/**
88 * struct rhashtable_compare_arg - Key for the function rhashtable_compare
89 * @ht: Hash table
90 * @key: Key to compare against
91 */
92struct rhashtable_compare_arg {
93 struct rhashtable *ht;
94 const void *key;
95};
96
97typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed);
98typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed);
99typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg,
100 const void *obj);
101
102struct rhashtable;
103
104/**
105 * struct rhashtable_params - Hash table construction parameters
106 * @nelem_hint: Hint on number of elements, should be 75% of desired size
107 * @key_len: Length of key
108 * @key_offset: Offset of key in struct to be hashed
109 * @head_offset: Offset of rhash_head in struct to be hashed
110 * @insecure_max_entries: Maximum number of entries (may be exceeded)
111 * @max_size: Maximum size while expanding
112 * @min_size: Minimum size while shrinking
113 * @nulls_base: Base value to generate nulls marker
114 * @insecure_elasticity: Set to true to disable chain length checks
115 * @automatic_shrinking: Enable automatic shrinking of tables
116 * @locks_mul: Number of bucket locks to allocate per cpu (default: 128)
117 * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
118 * @obj_hashfn: Function to hash object
119 * @obj_cmpfn: Function to compare key with object
120 */
121struct rhashtable_params {
122 size_t nelem_hint;
123 size_t key_len;
124 size_t key_offset;
125 size_t head_offset;
126 unsigned int insecure_max_entries;
127 unsigned int max_size;
128 unsigned int min_size;
129 u32 nulls_base;
130 bool insecure_elasticity;
131 bool automatic_shrinking;
132 size_t locks_mul;
133 rht_hashfn_t hashfn;
134 rht_obj_hashfn_t obj_hashfn;
135 rht_obj_cmpfn_t obj_cmpfn;
136};
137
138/**
139 * struct rhashtable - Hash table handle
140 * @tbl: Bucket table
141 * @nelems: Number of elements in table
142 * @key_len: Key length for hashfn
143 * @elasticity: Maximum chain length before rehash
144 * @p: Configuration parameters
145 * @rhlist: True if this is an rhltable
146 * @run_work: Deferred worker to expand/shrink asynchronously
147 * @mutex: Mutex to protect current/future table swapping
148 * @lock: Spin lock to protect walker list
149 */
150struct rhashtable {
151 struct bucket_table __rcu *tbl;
152 atomic_t nelems;
153 unsigned int key_len;
154 unsigned int elasticity;
155 struct rhashtable_params p;
156 bool rhlist;
157 struct work_struct run_work;
158 struct mutex mutex;
159 spinlock_t lock;
160};
161
162/**
163 * struct rhltable - Hash table with duplicate objects in a list
164 * @ht: Underlying rhtable
165 */
166struct rhltable {
167 struct rhashtable ht;
168};
169
170/**
171 * struct rhashtable_walker - Hash table walker
172 * @list: List entry on list of walkers
173 * @tbl: The table that we were walking over
174 */
175struct rhashtable_walker {
176 struct list_head list;
177 struct bucket_table *tbl;
178};
179
180/**
181 * struct rhashtable_iter - Hash table iterator
182 * @ht: Table to iterate through
183 * @p: Current pointer
184 * @list: Current hash list pointer
185 * @walker: Associated rhashtable walker
186 * @slot: Current slot
187 * @skip: Number of entries to skip in slot
188 */
189struct rhashtable_iter {
190 struct rhashtable *ht;
191 struct rhash_head *p;
192 struct rhlist_head *list;
193 struct rhashtable_walker walker;
194 unsigned int slot;
195 unsigned int skip;
196};
197
198static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash)
199{
200 return NULLS_MARKER(ht->p.nulls_base + hash);
201}
202
203#define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \
204 ((ptr) = (typeof(ptr)) rht_marker(ht, hash))
205
206static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
207{
208 return ((unsigned long) ptr & 1);
209}
210
211static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr)
212{
213 return ((unsigned long) ptr) >> 1;
214}
215
216static inline void *rht_obj(const struct rhashtable *ht,
217 const struct rhash_head *he)
218{
219 return (char *)he - ht->p.head_offset;
220}
221
222static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
223 unsigned int hash)
224{
225 return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1);
226}
227
228static inline unsigned int rht_key_hashfn(
229 struct rhashtable *ht, const struct bucket_table *tbl,
230 const void *key, const struct rhashtable_params params)
231{
232 unsigned int hash;
233
234 /* params must be equal to ht->p if it isn't constant. */
235 if (!__builtin_constant_p(params.key_len))
236 hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd);
237 else if (params.key_len) {
238 unsigned int key_len = params.key_len;
239
240 if (params.hashfn)
241 hash = params.hashfn(key, key_len, tbl->hash_rnd);
242 else if (key_len & (sizeof(u32) - 1))
243 hash = jhash(key, key_len, tbl->hash_rnd);
244 else
245 hash = jhash2(key, key_len / sizeof(u32),
246 tbl->hash_rnd);
247 } else {
248 unsigned int key_len = ht->p.key_len;
249
250 if (params.hashfn)
251 hash = params.hashfn(key, key_len, tbl->hash_rnd);
252 else
253 hash = jhash(key, key_len, tbl->hash_rnd);
254 }
255
256 return rht_bucket_index(tbl, hash);
257}
258
259static inline unsigned int rht_head_hashfn(
260 struct rhashtable *ht, const struct bucket_table *tbl,
261 const struct rhash_head *he, const struct rhashtable_params params)
262{
263 const char *ptr = rht_obj(ht, he);
264
265 return likely(params.obj_hashfn) ?
266 rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
267 ht->p.key_len,
268 tbl->hash_rnd)) :
269 rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
270}
271
272/**
273 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
274 * @ht: hash table
275 * @tbl: current table
276 */
277static inline bool rht_grow_above_75(const struct rhashtable *ht,
278 const struct bucket_table *tbl)
279{
280 /* Expand table when exceeding 75% load */
281 return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
282 (!ht->p.max_size || tbl->size < ht->p.max_size);
283}
284
285/**
286 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
287 * @ht: hash table
288 * @tbl: current table
289 */
290static inline bool rht_shrink_below_30(const struct rhashtable *ht,
291 const struct bucket_table *tbl)
292{
293 /* Shrink table beneath 30% load */
294 return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
295 tbl->size > ht->p.min_size;
296}
297
298/**
299 * rht_grow_above_100 - returns true if nelems > table-size
300 * @ht: hash table
301 * @tbl: current table
302 */
303static inline bool rht_grow_above_100(const struct rhashtable *ht,
304 const struct bucket_table *tbl)
305{
306 return atomic_read(&ht->nelems) > tbl->size &&
307 (!ht->p.max_size || tbl->size < ht->p.max_size);
308}
309
310/**
311 * rht_grow_above_max - returns true if table is above maximum
312 * @ht: hash table
313 * @tbl: current table
314 */
315static inline bool rht_grow_above_max(const struct rhashtable *ht,
316 const struct bucket_table *tbl)
317{
318 return ht->p.insecure_max_entries &&
319 atomic_read(&ht->nelems) >= ht->p.insecure_max_entries;
320}
321
322/* The bucket lock is selected based on the hash and protects mutations
323 * on a group of hash buckets.
324 *
325 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
326 * a single lock always covers both buckets which may both contains
327 * entries which link to the same bucket of the old table during resizing.
328 * This allows to simplify the locking as locking the bucket in both
329 * tables during resize always guarantee protection.
330 *
331 * IMPORTANT: When holding the bucket lock of both the old and new table
332 * during expansions and shrinking, the old bucket lock must always be
333 * acquired first.
334 */
335static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
336 unsigned int hash)
337{
338 return &tbl->locks[hash & tbl->locks_mask];
339}
340
341#ifdef CONFIG_PROVE_LOCKING
342int lockdep_rht_mutex_is_held(struct rhashtable *ht);
343int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
344#else
345static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
346{
347 return 1;
348}
349
350static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
351 u32 hash)
352{
353 return 1;
354}
355#endif /* CONFIG_PROVE_LOCKING */
356
357int rhashtable_init(struct rhashtable *ht,
358 const struct rhashtable_params *params);
359int rhltable_init(struct rhltable *hlt,
360 const struct rhashtable_params *params);
361
362void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
363 struct rhash_head *obj);
364
365void rhashtable_walk_enter(struct rhashtable *ht,
366 struct rhashtable_iter *iter);
367void rhashtable_walk_exit(struct rhashtable_iter *iter);
368int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU);
369void *rhashtable_walk_next(struct rhashtable_iter *iter);
370void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
371
372void rhashtable_free_and_destroy(struct rhashtable *ht,
373 void (*free_fn)(void *ptr, void *arg),
374 void *arg);
375void rhashtable_destroy(struct rhashtable *ht);
376
377#define rht_dereference(p, ht) \
378 rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
379
380#define rht_dereference_rcu(p, ht) \
381 rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
382
383#define rht_dereference_bucket(p, tbl, hash) \
384 rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
385
386#define rht_dereference_bucket_rcu(p, tbl, hash) \
387 rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
388
389#define rht_entry(tpos, pos, member) \
390 ({ tpos = container_of(pos, typeof(*tpos), member); 1; })
391
392/**
393 * rht_for_each_continue - continue iterating over hash chain
394 * @pos: the &struct rhash_head to use as a loop cursor.
395 * @head: the previous &struct rhash_head to continue from
396 * @tbl: the &struct bucket_table
397 * @hash: the hash value / bucket index
398 */
399#define rht_for_each_continue(pos, head, tbl, hash) \
400 for (pos = rht_dereference_bucket(head, tbl, hash); \
401 !rht_is_a_nulls(pos); \
402 pos = rht_dereference_bucket((pos)->next, tbl, hash))
403
404/**
405 * rht_for_each - iterate over hash chain
406 * @pos: the &struct rhash_head to use as a loop cursor.
407 * @tbl: the &struct bucket_table
408 * @hash: the hash value / bucket index
409 */
410#define rht_for_each(pos, tbl, hash) \
411 rht_for_each_continue(pos, (tbl)->buckets[hash], tbl, hash)
412
413/**
414 * rht_for_each_entry_continue - continue iterating over hash chain
415 * @tpos: the type * to use as a loop cursor.
416 * @pos: the &struct rhash_head to use as a loop cursor.
417 * @head: the previous &struct rhash_head to continue from
418 * @tbl: the &struct bucket_table
419 * @hash: the hash value / bucket index
420 * @member: name of the &struct rhash_head within the hashable struct.
421 */
422#define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \
423 for (pos = rht_dereference_bucket(head, tbl, hash); \
424 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
425 pos = rht_dereference_bucket((pos)->next, tbl, hash))
426
427/**
428 * rht_for_each_entry - iterate over hash chain of given type
429 * @tpos: the type * to use as a loop cursor.
430 * @pos: the &struct rhash_head to use as a loop cursor.
431 * @tbl: the &struct bucket_table
432 * @hash: the hash value / bucket index
433 * @member: name of the &struct rhash_head within the hashable struct.
434 */
435#define rht_for_each_entry(tpos, pos, tbl, hash, member) \
436 rht_for_each_entry_continue(tpos, pos, (tbl)->buckets[hash], \
437 tbl, hash, member)
438
439/**
440 * rht_for_each_entry_safe - safely iterate over hash chain of given type
441 * @tpos: the type * to use as a loop cursor.
442 * @pos: the &struct rhash_head to use as a loop cursor.
443 * @next: the &struct rhash_head to use as next in loop cursor.
444 * @tbl: the &struct bucket_table
445 * @hash: the hash value / bucket index
446 * @member: name of the &struct rhash_head within the hashable struct.
447 *
448 * This hash chain list-traversal primitive allows for the looped code to
449 * remove the loop cursor from the list.
450 */
451#define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \
452 for (pos = rht_dereference_bucket((tbl)->buckets[hash], tbl, hash), \
453 next = !rht_is_a_nulls(pos) ? \
454 rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
455 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
456 pos = next, \
457 next = !rht_is_a_nulls(pos) ? \
458 rht_dereference_bucket(pos->next, tbl, hash) : NULL)
459
460/**
461 * rht_for_each_rcu_continue - continue iterating over rcu hash chain
462 * @pos: the &struct rhash_head to use as a loop cursor.
463 * @head: the previous &struct rhash_head to continue from
464 * @tbl: the &struct bucket_table
465 * @hash: the hash value / bucket index
466 *
467 * This hash chain list-traversal primitive may safely run concurrently with
468 * the _rcu mutation primitives such as rhashtable_insert() as long as the
469 * traversal is guarded by rcu_read_lock().
470 */
471#define rht_for_each_rcu_continue(pos, head, tbl, hash) \
472 for (({barrier(); }), \
473 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
474 !rht_is_a_nulls(pos); \
475 pos = rcu_dereference_raw(pos->next))
476
477/**
478 * rht_for_each_rcu - iterate over rcu hash chain
479 * @pos: the &struct rhash_head to use as a loop cursor.
480 * @tbl: the &struct bucket_table
481 * @hash: the hash value / bucket index
482 *
483 * This hash chain list-traversal primitive may safely run concurrently with
484 * the _rcu mutation primitives such as rhashtable_insert() as long as the
485 * traversal is guarded by rcu_read_lock().
486 */
487#define rht_for_each_rcu(pos, tbl, hash) \
488 rht_for_each_rcu_continue(pos, (tbl)->buckets[hash], tbl, hash)
489
490/**
491 * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
492 * @tpos: the type * to use as a loop cursor.
493 * @pos: the &struct rhash_head to use as a loop cursor.
494 * @head: the previous &struct rhash_head to continue from
495 * @tbl: the &struct bucket_table
496 * @hash: the hash value / bucket index
497 * @member: name of the &struct rhash_head within the hashable struct.
498 *
499 * This hash chain list-traversal primitive may safely run concurrently with
500 * the _rcu mutation primitives such as rhashtable_insert() as long as the
501 * traversal is guarded by rcu_read_lock().
502 */
503#define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
504 for (({barrier(); }), \
505 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
506 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
507 pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
508
509/**
510 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
511 * @tpos: the type * to use as a loop cursor.
512 * @pos: the &struct rhash_head to use as a loop cursor.
513 * @tbl: the &struct bucket_table
514 * @hash: the hash value / bucket index
515 * @member: name of the &struct rhash_head within the hashable struct.
516 *
517 * This hash chain list-traversal primitive may safely run concurrently with
518 * the _rcu mutation primitives such as rhashtable_insert() as long as the
519 * traversal is guarded by rcu_read_lock().
520 */
521#define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \
522 rht_for_each_entry_rcu_continue(tpos, pos, (tbl)->buckets[hash],\
523 tbl, hash, member)
524
525/**
526 * rhl_for_each_rcu - iterate over rcu hash table list
527 * @pos: the &struct rlist_head to use as a loop cursor.
528 * @list: the head of the list
529 *
530 * This hash chain list-traversal primitive should be used on the
531 * list returned by rhltable_lookup.
532 */
533#define rhl_for_each_rcu(pos, list) \
534 for (pos = list; pos; pos = rcu_dereference_raw(pos->next))
535
536/**
537 * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type
538 * @tpos: the type * to use as a loop cursor.
539 * @pos: the &struct rlist_head to use as a loop cursor.
540 * @list: the head of the list
541 * @member: name of the &struct rlist_head within the hashable struct.
542 *
543 * This hash chain list-traversal primitive should be used on the
544 * list returned by rhltable_lookup.
545 */
546#define rhl_for_each_entry_rcu(tpos, pos, list, member) \
547 for (pos = list; pos && rht_entry(tpos, pos, member); \
548 pos = rcu_dereference_raw(pos->next))
549
550static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
551 const void *obj)
552{
553 struct rhashtable *ht = arg->ht;
554 const char *ptr = obj;
555
556 return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
557}
558
559/* Internal function, do not use. */
560static inline struct rhash_head *__rhashtable_lookup(
561 struct rhashtable *ht, const void *key,
562 const struct rhashtable_params params)
563{
564 struct rhashtable_compare_arg arg = {
565 .ht = ht,
566 .key = key,
567 };
568 const struct bucket_table *tbl;
569 struct rhash_head *he;
570 unsigned int hash;
571
572 tbl = rht_dereference_rcu(ht->tbl, ht);
573restart:
574 hash = rht_key_hashfn(ht, tbl, key, params);
575 rht_for_each_rcu(he, tbl, hash) {
576 if (params.obj_cmpfn ?
577 params.obj_cmpfn(&arg, rht_obj(ht, he)) :
578 rhashtable_compare(&arg, rht_obj(ht, he)))
579 continue;
580 return he;
581 }
582
583 /* Ensure we see any new tables. */
584 smp_rmb();
585
586 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
587 if (unlikely(tbl))
588 goto restart;
589
590 return NULL;
591}
592
593/**
594 * rhashtable_lookup - search hash table
595 * @ht: hash table
596 * @key: the pointer to the key
597 * @params: hash table parameters
598 *
599 * Computes the hash value for the key and traverses the bucket chain looking
600 * for a entry with an identical key. The first matching entry is returned.
601 *
602 * This must only be called under the RCU read lock.
603 *
604 * Returns the first entry on which the compare function returned true.
605 */
606static inline void *rhashtable_lookup(
607 struct rhashtable *ht, const void *key,
608 const struct rhashtable_params params)
609{
610 struct rhash_head *he = __rhashtable_lookup(ht, key, params);
611
612 return he ? rht_obj(ht, he) : NULL;
613}
614
615/**
616 * rhashtable_lookup_fast - search hash table, without RCU read lock
617 * @ht: hash table
618 * @key: the pointer to the key
619 * @params: hash table parameters
620 *
621 * Computes the hash value for the key and traverses the bucket chain looking
622 * for a entry with an identical key. The first matching entry is returned.
623 *
624 * Only use this function when you have other mechanisms guaranteeing
625 * that the object won't go away after the RCU read lock is released.
626 *
627 * Returns the first entry on which the compare function returned true.
628 */
629static inline void *rhashtable_lookup_fast(
630 struct rhashtable *ht, const void *key,
631 const struct rhashtable_params params)
632{
633 void *obj;
634
635 rcu_read_lock();
636 obj = rhashtable_lookup(ht, key, params);
637 rcu_read_unlock();
638
639 return obj;
640}
641
642/**
643 * rhltable_lookup - search hash list table
644 * @hlt: hash table
645 * @key: the pointer to the key
646 * @params: hash table parameters
647 *
648 * Computes the hash value for the key and traverses the bucket chain looking
649 * for a entry with an identical key. All matching entries are returned
650 * in a list.
651 *
652 * This must only be called under the RCU read lock.
653 *
654 * Returns the list of entries that match the given key.
655 */
656static inline struct rhlist_head *rhltable_lookup(
657 struct rhltable *hlt, const void *key,
658 const struct rhashtable_params params)
659{
660 struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params);
661
662 return he ? container_of(he, struct rhlist_head, rhead) : NULL;
663}
664
665/* Internal function, please use rhashtable_insert_fast() instead. This
666 * function returns the existing element already in hashes in there is a clash,
667 * otherwise it returns an error via ERR_PTR().
668 */
669static inline void *__rhashtable_insert_fast(
670 struct rhashtable *ht, const void *key, struct rhash_head *obj,
671 const struct rhashtable_params params, bool rhlist)
672{
673 struct rhashtable_compare_arg arg = {
674 .ht = ht,
675 .key = key,
676 };
677 struct rhash_head __rcu **pprev;
678 struct bucket_table *tbl;
679 struct rhash_head *head;
680 spinlock_t *lock;
681 unsigned int hash;
682 int elasticity;
683 void *data;
684
685 rcu_read_lock();
686
687 tbl = rht_dereference_rcu(ht->tbl, ht);
688 hash = rht_head_hashfn(ht, tbl, obj, params);
689 lock = rht_bucket_lock(tbl, hash);
690 spin_lock_bh(lock);
691
692 if (unlikely(rht_dereference_bucket(tbl->future_tbl, tbl, hash))) {
693slow_path:
694 spin_unlock_bh(lock);
695 rcu_read_unlock();
696 return rhashtable_insert_slow(ht, key, obj);
697 }
698
699 elasticity = ht->elasticity;
700 pprev = &tbl->buckets[hash];
701 rht_for_each(head, tbl, hash) {
702 struct rhlist_head *plist;
703 struct rhlist_head *list;
704
705 elasticity--;
706 if (!key ||
707 (params.obj_cmpfn ?
708 params.obj_cmpfn(&arg, rht_obj(ht, head)) :
709 rhashtable_compare(&arg, rht_obj(ht, head))))
710 continue;
711
712 data = rht_obj(ht, head);
713
714 if (!rhlist)
715 goto out;
716
717
718 list = container_of(obj, struct rhlist_head, rhead);
719 plist = container_of(head, struct rhlist_head, rhead);
720
721 RCU_INIT_POINTER(list->next, plist);
722 head = rht_dereference_bucket(head->next, tbl, hash);
723 RCU_INIT_POINTER(list->rhead.next, head);
724 rcu_assign_pointer(*pprev, obj);
725
726 goto good;
727 }
728
729 if (elasticity <= 0)
730 goto slow_path;
731
732 data = ERR_PTR(-E2BIG);
733 if (unlikely(rht_grow_above_max(ht, tbl)))
734 goto out;
735
736 if (unlikely(rht_grow_above_100(ht, tbl)))
737 goto slow_path;
738
739 head = rht_dereference_bucket(tbl->buckets[hash], tbl, hash);
740
741 RCU_INIT_POINTER(obj->next, head);
742 if (rhlist) {
743 struct rhlist_head *list;
744
745 list = container_of(obj, struct rhlist_head, rhead);
746 RCU_INIT_POINTER(list->next, NULL);
747 }
748
749 rcu_assign_pointer(tbl->buckets[hash], obj);
750
751 atomic_inc(&ht->nelems);
752 if (rht_grow_above_75(ht, tbl))
753 schedule_work(&ht->run_work);
754
755good:
756 data = NULL;
757
758out:
759 spin_unlock_bh(lock);
760 rcu_read_unlock();
761
762 return data;
763}
764
765/**
766 * rhashtable_insert_fast - insert object into hash table
767 * @ht: hash table
768 * @obj: pointer to hash head inside object
769 * @params: hash table parameters
770 *
771 * Will take a per bucket spinlock to protect against mutual mutations
772 * on the same bucket. Multiple insertions may occur in parallel unless
773 * they map to the same bucket lock.
774 *
775 * It is safe to call this function from atomic context.
776 *
777 * Will trigger an automatic deferred table resizing if the size grows
778 * beyond the watermark indicated by grow_decision() which can be passed
779 * to rhashtable_init().
780 */
781static inline int rhashtable_insert_fast(
782 struct rhashtable *ht, struct rhash_head *obj,
783 const struct rhashtable_params params)
784{
785 void *ret;
786
787 ret = __rhashtable_insert_fast(ht, NULL, obj, params, false);
788 if (IS_ERR(ret))
789 return PTR_ERR(ret);
790
791 return ret == NULL ? 0 : -EEXIST;
792}
793
794/**
795 * rhltable_insert_key - insert object into hash list table
796 * @hlt: hash list table
797 * @key: the pointer to the key
798 * @list: pointer to hash list head inside object
799 * @params: hash table parameters
800 *
801 * Will take a per bucket spinlock to protect against mutual mutations
802 * on the same bucket. Multiple insertions may occur in parallel unless
803 * they map to the same bucket lock.
804 *
805 * It is safe to call this function from atomic context.
806 *
807 * Will trigger an automatic deferred table resizing if the size grows
808 * beyond the watermark indicated by grow_decision() which can be passed
809 * to rhashtable_init().
810 */
811static inline int rhltable_insert_key(
812 struct rhltable *hlt, const void *key, struct rhlist_head *list,
813 const struct rhashtable_params params)
814{
815 return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead,
816 params, true));
817}
818
819/**
820 * rhltable_insert - insert object into hash list table
821 * @hlt: hash list table
822 * @list: pointer to hash list head inside object
823 * @params: hash table parameters
824 *
825 * Will take a per bucket spinlock to protect against mutual mutations
826 * on the same bucket. Multiple insertions may occur in parallel unless
827 * they map to the same bucket lock.
828 *
829 * It is safe to call this function from atomic context.
830 *
831 * Will trigger an automatic deferred table resizing if the size grows
832 * beyond the watermark indicated by grow_decision() which can be passed
833 * to rhashtable_init().
834 */
835static inline int rhltable_insert(
836 struct rhltable *hlt, struct rhlist_head *list,
837 const struct rhashtable_params params)
838{
839 const char *key = rht_obj(&hlt->ht, &list->rhead);
840
841 key += params.key_offset;
842
843 return rhltable_insert_key(hlt, key, list, params);
844}
845
846/**
847 * rhashtable_lookup_insert_fast - lookup and insert object into hash table
848 * @ht: hash table
849 * @obj: pointer to hash head inside object
850 * @params: hash table parameters
851 *
852 * Locks down the bucket chain in both the old and new table if a resize
853 * is in progress to ensure that writers can't remove from the old table
854 * and can't insert to the new table during the atomic operation of search
855 * and insertion. Searches for duplicates in both the old and new table if
856 * a resize is in progress.
857 *
858 * This lookup function may only be used for fixed key hash table (key_len
859 * parameter set). It will BUG() if used inappropriately.
860 *
861 * It is safe to call this function from atomic context.
862 *
863 * Will trigger an automatic deferred table resizing if the size grows
864 * beyond the watermark indicated by grow_decision() which can be passed
865 * to rhashtable_init().
866 */
867static inline int rhashtable_lookup_insert_fast(
868 struct rhashtable *ht, struct rhash_head *obj,
869 const struct rhashtable_params params)
870{
871 const char *key = rht_obj(ht, obj);
872 void *ret;
873
874 BUG_ON(ht->p.obj_hashfn);
875
876 ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
877 false);
878 if (IS_ERR(ret))
879 return PTR_ERR(ret);
880
881 return ret == NULL ? 0 : -EEXIST;
882}
883
884/**
885 * rhashtable_lookup_insert_key - search and insert object to hash table
886 * with explicit key
887 * @ht: hash table
888 * @key: key
889 * @obj: pointer to hash head inside object
890 * @params: hash table parameters
891 *
892 * Locks down the bucket chain in both the old and new table if a resize
893 * is in progress to ensure that writers can't remove from the old table
894 * and can't insert to the new table during the atomic operation of search
895 * and insertion. Searches for duplicates in both the old and new table if
896 * a resize is in progress.
897 *
898 * Lookups may occur in parallel with hashtable mutations and resizing.
899 *
900 * Will trigger an automatic deferred table resizing if the size grows
901 * beyond the watermark indicated by grow_decision() which can be passed
902 * to rhashtable_init().
903 *
904 * Returns zero on success.
905 */
906static inline int rhashtable_lookup_insert_key(
907 struct rhashtable *ht, const void *key, struct rhash_head *obj,
908 const struct rhashtable_params params)
909{
910 void *ret;
911
912 BUG_ON(!ht->p.obj_hashfn || !key);
913
914 ret = __rhashtable_insert_fast(ht, key, obj, params, false);
915 if (IS_ERR(ret))
916 return PTR_ERR(ret);
917
918 return ret == NULL ? 0 : -EEXIST;
919}
920
921/**
922 * rhashtable_lookup_get_insert_key - lookup and insert object into hash table
923 * @ht: hash table
924 * @obj: pointer to hash head inside object
925 * @params: hash table parameters
926 * @data: pointer to element data already in hashes
927 *
928 * Just like rhashtable_lookup_insert_key(), but this function returns the
929 * object if it exists, NULL if it does not and the insertion was successful,
930 * and an ERR_PTR otherwise.
931 */
932static inline void *rhashtable_lookup_get_insert_key(
933 struct rhashtable *ht, const void *key, struct rhash_head *obj,
934 const struct rhashtable_params params)
935{
936 BUG_ON(!ht->p.obj_hashfn || !key);
937
938 return __rhashtable_insert_fast(ht, key, obj, params, false);
939}
940
941/* Internal function, please use rhashtable_remove_fast() instead */
942static inline int __rhashtable_remove_fast_one(
943 struct rhashtable *ht, struct bucket_table *tbl,
944 struct rhash_head *obj, const struct rhashtable_params params,
945 bool rhlist)
946{
947 struct rhash_head __rcu **pprev;
948 struct rhash_head *he;
949 spinlock_t * lock;
950 unsigned int hash;
951 int err = -ENOENT;
952
953 hash = rht_head_hashfn(ht, tbl, obj, params);
954 lock = rht_bucket_lock(tbl, hash);
955
956 spin_lock_bh(lock);
957
958 pprev = &tbl->buckets[hash];
959 rht_for_each(he, tbl, hash) {
960 struct rhlist_head *list;
961
962 list = container_of(he, struct rhlist_head, rhead);
963
964 if (he != obj) {
965 struct rhlist_head __rcu **lpprev;
966
967 pprev = &he->next;
968
969 if (!rhlist)
970 continue;
971
972 do {
973 lpprev = &list->next;
974 list = rht_dereference_bucket(list->next,
975 tbl, hash);
976 } while (list && obj != &list->rhead);
977
978 if (!list)
979 continue;
980
981 list = rht_dereference_bucket(list->next, tbl, hash);
982 RCU_INIT_POINTER(*lpprev, list);
983 err = 0;
984 break;
985 }
986
987 obj = rht_dereference_bucket(obj->next, tbl, hash);
988 err = 1;
989
990 if (rhlist) {
991 list = rht_dereference_bucket(list->next, tbl, hash);
992 if (list) {
993 RCU_INIT_POINTER(list->rhead.next, obj);
994 obj = &list->rhead;
995 err = 0;
996 }
997 }
998
999 rcu_assign_pointer(*pprev, obj);
1000 break;
1001 }
1002
1003 spin_unlock_bh(lock);
1004
1005 if (err > 0) {
1006 atomic_dec(&ht->nelems);
1007 if (unlikely(ht->p.automatic_shrinking &&
1008 rht_shrink_below_30(ht, tbl)))
1009 schedule_work(&ht->run_work);
1010 err = 0;
1011 }
1012
1013 return err;
1014}
1015
1016/* Internal function, please use rhashtable_remove_fast() instead */
1017static inline int __rhashtable_remove_fast(
1018 struct rhashtable *ht, struct rhash_head *obj,
1019 const struct rhashtable_params params, bool rhlist)
1020{
1021 struct bucket_table *tbl;
1022 int err;
1023
1024 rcu_read_lock();
1025
1026 tbl = rht_dereference_rcu(ht->tbl, ht);
1027
1028 /* Because we have already taken (and released) the bucket
1029 * lock in old_tbl, if we find that future_tbl is not yet
1030 * visible then that guarantees the entry to still be in
1031 * the old tbl if it exists.
1032 */
1033 while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params,
1034 rhlist)) &&
1035 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1036 ;
1037
1038 rcu_read_unlock();
1039
1040 return err;
1041}
1042
1043/**
1044 * rhashtable_remove_fast - remove object from hash table
1045 * @ht: hash table
1046 * @obj: pointer to hash head inside object
1047 * @params: hash table parameters
1048 *
1049 * Since the hash chain is single linked, the removal operation needs to
1050 * walk the bucket chain upon removal. The removal operation is thus
1051 * considerable slow if the hash table is not correctly sized.
1052 *
1053 * Will automatically shrink the table via rhashtable_expand() if the
1054 * shrink_decision function specified at rhashtable_init() returns true.
1055 *
1056 * Returns zero on success, -ENOENT if the entry could not be found.
1057 */
1058static inline int rhashtable_remove_fast(
1059 struct rhashtable *ht, struct rhash_head *obj,
1060 const struct rhashtable_params params)
1061{
1062 return __rhashtable_remove_fast(ht, obj, params, false);
1063}
1064
1065/**
1066 * rhltable_remove - remove object from hash list table
1067 * @hlt: hash list table
1068 * @list: pointer to hash list head inside object
1069 * @params: hash table parameters
1070 *
1071 * Since the hash chain is single linked, the removal operation needs to
1072 * walk the bucket chain upon removal. The removal operation is thus
1073 * considerable slow if the hash table is not correctly sized.
1074 *
1075 * Will automatically shrink the table via rhashtable_expand() if the
1076 * shrink_decision function specified at rhashtable_init() returns true.
1077 *
1078 * Returns zero on success, -ENOENT if the entry could not be found.
1079 */
1080static inline int rhltable_remove(
1081 struct rhltable *hlt, struct rhlist_head *list,
1082 const struct rhashtable_params params)
1083{
1084 return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true);
1085}
1086
1087/* Internal function, please use rhashtable_replace_fast() instead */
1088static inline int __rhashtable_replace_fast(
1089 struct rhashtable *ht, struct bucket_table *tbl,
1090 struct rhash_head *obj_old, struct rhash_head *obj_new,
1091 const struct rhashtable_params params)
1092{
1093 struct rhash_head __rcu **pprev;
1094 struct rhash_head *he;
1095 spinlock_t *lock;
1096 unsigned int hash;
1097 int err = -ENOENT;
1098
1099 /* Minimally, the old and new objects must have same hash
1100 * (which should mean identifiers are the same).
1101 */
1102 hash = rht_head_hashfn(ht, tbl, obj_old, params);
1103 if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
1104 return -EINVAL;
1105
1106 lock = rht_bucket_lock(tbl, hash);
1107
1108 spin_lock_bh(lock);
1109
1110 pprev = &tbl->buckets[hash];
1111 rht_for_each(he, tbl, hash) {
1112 if (he != obj_old) {
1113 pprev = &he->next;
1114 continue;
1115 }
1116
1117 rcu_assign_pointer(obj_new->next, obj_old->next);
1118 rcu_assign_pointer(*pprev, obj_new);
1119 err = 0;
1120 break;
1121 }
1122
1123 spin_unlock_bh(lock);
1124
1125 return err;
1126}
1127
1128/**
1129 * rhashtable_replace_fast - replace an object in hash table
1130 * @ht: hash table
1131 * @obj_old: pointer to hash head inside object being replaced
1132 * @obj_new: pointer to hash head inside object which is new
1133 * @params: hash table parameters
1134 *
1135 * Replacing an object doesn't affect the number of elements in the hash table
1136 * or bucket, so we don't need to worry about shrinking or expanding the
1137 * table here.
1138 *
1139 * Returns zero on success, -ENOENT if the entry could not be found,
1140 * -EINVAL if hash is not the same for the old and new objects.
1141 */
1142static inline int rhashtable_replace_fast(
1143 struct rhashtable *ht, struct rhash_head *obj_old,
1144 struct rhash_head *obj_new,
1145 const struct rhashtable_params params)
1146{
1147 struct bucket_table *tbl;
1148 int err;
1149
1150 rcu_read_lock();
1151
1152 tbl = rht_dereference_rcu(ht->tbl, ht);
1153
1154 /* Because we have already taken (and released) the bucket
1155 * lock in old_tbl, if we find that future_tbl is not yet
1156 * visible then that guarantees the entry to still be in
1157 * the old tbl if it exists.
1158 */
1159 while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
1160 obj_new, params)) &&
1161 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1162 ;
1163
1164 rcu_read_unlock();
1165
1166 return err;
1167}
1168
1169/* Obsolete function, do not use in new code. */
1170static inline int rhashtable_walk_init(struct rhashtable *ht,
1171 struct rhashtable_iter *iter, gfp_t gfp)
1172{
1173 rhashtable_walk_enter(ht, iter);
1174 return 0;
1175}
1176
1177/**
1178 * rhltable_walk_enter - Initialise an iterator
1179 * @hlt: Table to walk over
1180 * @iter: Hash table Iterator
1181 *
1182 * This function prepares a hash table walk.
1183 *
1184 * Note that if you restart a walk after rhashtable_walk_stop you
1185 * may see the same object twice. Also, you may miss objects if
1186 * there are removals in between rhashtable_walk_stop and the next
1187 * call to rhashtable_walk_start.
1188 *
1189 * For a completely stable walk you should construct your own data
1190 * structure outside the hash table.
1191 *
1192 * This function may sleep so you must not call it from interrupt
1193 * context or with spin locks held.
1194 *
1195 * You must call rhashtable_walk_exit after this function returns.
1196 */
1197static inline void rhltable_walk_enter(struct rhltable *hlt,
1198 struct rhashtable_iter *iter)
1199{
1200 return rhashtable_walk_enter(&hlt->ht, iter);
1201}
1202
1203/**
1204 * rhltable_free_and_destroy - free elements and destroy hash list table
1205 * @hlt: the hash list table to destroy
1206 * @free_fn: callback to release resources of element
1207 * @arg: pointer passed to free_fn
1208 *
1209 * See documentation for rhashtable_free_and_destroy.
1210 */
1211static inline void rhltable_free_and_destroy(struct rhltable *hlt,
1212 void (*free_fn)(void *ptr,
1213 void *arg),
1214 void *arg)
1215{
1216 return rhashtable_free_and_destroy(&hlt->ht, free_fn, arg);
1217}
1218
1219static inline void rhltable_destroy(struct rhltable *hlt)
1220{
1221 return rhltable_free_and_destroy(hlt, NULL, NULL);
1222}
1223
1224#endif /* _LINUX_RHASHTABLE_H */
1/*
2 * Resizable, Scalable, Concurrent Hash Table
3 *
4 * Copyright (c) 2015-2016 Herbert Xu <herbert@gondor.apana.org.au>
5 * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
6 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
7 *
8 * Code partially derived from nft_hash
9 * Rewritten with rehash code from br_multicast plus single list
10 * pointer as suggested by Josh Triplett
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 */
16
17#ifndef _LINUX_RHASHTABLE_H
18#define _LINUX_RHASHTABLE_H
19
20#include <linux/atomic.h>
21#include <linux/compiler.h>
22#include <linux/err.h>
23#include <linux/errno.h>
24#include <linux/jhash.h>
25#include <linux/list_nulls.h>
26#include <linux/workqueue.h>
27#include <linux/mutex.h>
28#include <linux/rculist.h>
29
30/*
31 * The end of the chain is marked with a special nulls marks which has
32 * the following format:
33 *
34 * +-------+-----------------------------------------------------+-+
35 * | Base | Hash |1|
36 * +-------+-----------------------------------------------------+-+
37 *
38 * Base (4 bits) : Reserved to distinguish between multiple tables.
39 * Specified via &struct rhashtable_params.nulls_base.
40 * Hash (27 bits): Full hash (unmasked) of first element added to bucket
41 * 1 (1 bit) : Nulls marker (always set)
42 *
43 * The remaining bits of the next pointer remain unused for now.
44 */
45#define RHT_BASE_BITS 4
46#define RHT_HASH_BITS 27
47#define RHT_BASE_SHIFT RHT_HASH_BITS
48
49/* Base bits plus 1 bit for nulls marker */
50#define RHT_HASH_RESERVED_SPACE (RHT_BASE_BITS + 1)
51
52/* Maximum chain length before rehash
53 *
54 * The maximum (not average) chain length grows with the size of the hash
55 * table, at a rate of (log N)/(log log N).
56 *
57 * The value of 16 is selected so that even if the hash table grew to
58 * 2^32 you would not expect the maximum chain length to exceed it
59 * unless we are under attack (or extremely unlucky).
60 *
61 * As this limit is only to detect attacks, we don't need to set it to a
62 * lower value as you'd need the chain length to vastly exceed 16 to have
63 * any real effect on the system.
64 */
65#define RHT_ELASTICITY 16u
66
67struct rhash_head {
68 struct rhash_head __rcu *next;
69};
70
71struct rhlist_head {
72 struct rhash_head rhead;
73 struct rhlist_head __rcu *next;
74};
75
76/**
77 * struct bucket_table - Table of hash buckets
78 * @size: Number of hash buckets
79 * @nest: Number of bits of first-level nested table.
80 * @rehash: Current bucket being rehashed
81 * @hash_rnd: Random seed to fold into hash
82 * @locks_mask: Mask to apply before accessing locks[]
83 * @locks: Array of spinlocks protecting individual buckets
84 * @walkers: List of active walkers
85 * @rcu: RCU structure for freeing the table
86 * @future_tbl: Table under construction during rehashing
87 * @ntbl: Nested table used when out of memory.
88 * @buckets: size * hash buckets
89 */
90struct bucket_table {
91 unsigned int size;
92 unsigned int nest;
93 unsigned int rehash;
94 u32 hash_rnd;
95 unsigned int locks_mask;
96 spinlock_t *locks;
97 struct list_head walkers;
98 struct rcu_head rcu;
99
100 struct bucket_table __rcu *future_tbl;
101
102 struct rhash_head __rcu *buckets[] ____cacheline_aligned_in_smp;
103};
104
105/**
106 * struct rhashtable_compare_arg - Key for the function rhashtable_compare
107 * @ht: Hash table
108 * @key: Key to compare against
109 */
110struct rhashtable_compare_arg {
111 struct rhashtable *ht;
112 const void *key;
113};
114
115typedef u32 (*rht_hashfn_t)(const void *data, u32 len, u32 seed);
116typedef u32 (*rht_obj_hashfn_t)(const void *data, u32 len, u32 seed);
117typedef int (*rht_obj_cmpfn_t)(struct rhashtable_compare_arg *arg,
118 const void *obj);
119
120struct rhashtable;
121
122/**
123 * struct rhashtable_params - Hash table construction parameters
124 * @nelem_hint: Hint on number of elements, should be 75% of desired size
125 * @key_len: Length of key
126 * @key_offset: Offset of key in struct to be hashed
127 * @head_offset: Offset of rhash_head in struct to be hashed
128 * @max_size: Maximum size while expanding
129 * @min_size: Minimum size while shrinking
130 * @locks_mul: Number of bucket locks to allocate per cpu (default: 32)
131 * @automatic_shrinking: Enable automatic shrinking of tables
132 * @nulls_base: Base value to generate nulls marker
133 * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
134 * @obj_hashfn: Function to hash object
135 * @obj_cmpfn: Function to compare key with object
136 */
137struct rhashtable_params {
138 u16 nelem_hint;
139 u16 key_len;
140 u16 key_offset;
141 u16 head_offset;
142 unsigned int max_size;
143 u16 min_size;
144 bool automatic_shrinking;
145 u8 locks_mul;
146 u32 nulls_base;
147 rht_hashfn_t hashfn;
148 rht_obj_hashfn_t obj_hashfn;
149 rht_obj_cmpfn_t obj_cmpfn;
150};
151
152/**
153 * struct rhashtable - Hash table handle
154 * @tbl: Bucket table
155 * @key_len: Key length for hashfn
156 * @max_elems: Maximum number of elements in table
157 * @p: Configuration parameters
158 * @rhlist: True if this is an rhltable
159 * @run_work: Deferred worker to expand/shrink asynchronously
160 * @mutex: Mutex to protect current/future table swapping
161 * @lock: Spin lock to protect walker list
162 * @nelems: Number of elements in table
163 */
164struct rhashtable {
165 struct bucket_table __rcu *tbl;
166 unsigned int key_len;
167 unsigned int max_elems;
168 struct rhashtable_params p;
169 bool rhlist;
170 struct work_struct run_work;
171 struct mutex mutex;
172 spinlock_t lock;
173 atomic_t nelems;
174};
175
176/**
177 * struct rhltable - Hash table with duplicate objects in a list
178 * @ht: Underlying rhtable
179 */
180struct rhltable {
181 struct rhashtable ht;
182};
183
184/**
185 * struct rhashtable_walker - Hash table walker
186 * @list: List entry on list of walkers
187 * @tbl: The table that we were walking over
188 */
189struct rhashtable_walker {
190 struct list_head list;
191 struct bucket_table *tbl;
192};
193
194/**
195 * struct rhashtable_iter - Hash table iterator
196 * @ht: Table to iterate through
197 * @p: Current pointer
198 * @list: Current hash list pointer
199 * @walker: Associated rhashtable walker
200 * @slot: Current slot
201 * @skip: Number of entries to skip in slot
202 */
203struct rhashtable_iter {
204 struct rhashtable *ht;
205 struct rhash_head *p;
206 struct rhlist_head *list;
207 struct rhashtable_walker walker;
208 unsigned int slot;
209 unsigned int skip;
210 bool end_of_table;
211};
212
213static inline unsigned long rht_marker(const struct rhashtable *ht, u32 hash)
214{
215 return NULLS_MARKER(ht->p.nulls_base + hash);
216}
217
218#define INIT_RHT_NULLS_HEAD(ptr, ht, hash) \
219 ((ptr) = (typeof(ptr)) rht_marker(ht, hash))
220
221static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
222{
223 return ((unsigned long) ptr & 1);
224}
225
226static inline unsigned long rht_get_nulls_value(const struct rhash_head *ptr)
227{
228 return ((unsigned long) ptr) >> 1;
229}
230
231static inline void *rht_obj(const struct rhashtable *ht,
232 const struct rhash_head *he)
233{
234 return (char *)he - ht->p.head_offset;
235}
236
237static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
238 unsigned int hash)
239{
240 return (hash >> RHT_HASH_RESERVED_SPACE) & (tbl->size - 1);
241}
242
243static inline unsigned int rht_key_get_hash(struct rhashtable *ht,
244 const void *key, const struct rhashtable_params params,
245 unsigned int hash_rnd)
246{
247 unsigned int hash;
248
249 /* params must be equal to ht->p if it isn't constant. */
250 if (!__builtin_constant_p(params.key_len))
251 hash = ht->p.hashfn(key, ht->key_len, hash_rnd);
252 else if (params.key_len) {
253 unsigned int key_len = params.key_len;
254
255 if (params.hashfn)
256 hash = params.hashfn(key, key_len, hash_rnd);
257 else if (key_len & (sizeof(u32) - 1))
258 hash = jhash(key, key_len, hash_rnd);
259 else
260 hash = jhash2(key, key_len / sizeof(u32), hash_rnd);
261 } else {
262 unsigned int key_len = ht->p.key_len;
263
264 if (params.hashfn)
265 hash = params.hashfn(key, key_len, hash_rnd);
266 else
267 hash = jhash(key, key_len, hash_rnd);
268 }
269
270 return hash;
271}
272
273static inline unsigned int rht_key_hashfn(
274 struct rhashtable *ht, const struct bucket_table *tbl,
275 const void *key, const struct rhashtable_params params)
276{
277 unsigned int hash = rht_key_get_hash(ht, key, params, tbl->hash_rnd);
278
279 return rht_bucket_index(tbl, hash);
280}
281
282static inline unsigned int rht_head_hashfn(
283 struct rhashtable *ht, const struct bucket_table *tbl,
284 const struct rhash_head *he, const struct rhashtable_params params)
285{
286 const char *ptr = rht_obj(ht, he);
287
288 return likely(params.obj_hashfn) ?
289 rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
290 ht->p.key_len,
291 tbl->hash_rnd)) :
292 rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
293}
294
295/**
296 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
297 * @ht: hash table
298 * @tbl: current table
299 */
300static inline bool rht_grow_above_75(const struct rhashtable *ht,
301 const struct bucket_table *tbl)
302{
303 /* Expand table when exceeding 75% load */
304 return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
305 (!ht->p.max_size || tbl->size < ht->p.max_size);
306}
307
308/**
309 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
310 * @ht: hash table
311 * @tbl: current table
312 */
313static inline bool rht_shrink_below_30(const struct rhashtable *ht,
314 const struct bucket_table *tbl)
315{
316 /* Shrink table beneath 30% load */
317 return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
318 tbl->size > ht->p.min_size;
319}
320
321/**
322 * rht_grow_above_100 - returns true if nelems > table-size
323 * @ht: hash table
324 * @tbl: current table
325 */
326static inline bool rht_grow_above_100(const struct rhashtable *ht,
327 const struct bucket_table *tbl)
328{
329 return atomic_read(&ht->nelems) > tbl->size &&
330 (!ht->p.max_size || tbl->size < ht->p.max_size);
331}
332
333/**
334 * rht_grow_above_max - returns true if table is above maximum
335 * @ht: hash table
336 * @tbl: current table
337 */
338static inline bool rht_grow_above_max(const struct rhashtable *ht,
339 const struct bucket_table *tbl)
340{
341 return atomic_read(&ht->nelems) >= ht->max_elems;
342}
343
344/* The bucket lock is selected based on the hash and protects mutations
345 * on a group of hash buckets.
346 *
347 * A maximum of tbl->size/2 bucket locks is allocated. This ensures that
348 * a single lock always covers both buckets which may both contains
349 * entries which link to the same bucket of the old table during resizing.
350 * This allows to simplify the locking as locking the bucket in both
351 * tables during resize always guarantee protection.
352 *
353 * IMPORTANT: When holding the bucket lock of both the old and new table
354 * during expansions and shrinking, the old bucket lock must always be
355 * acquired first.
356 */
357static inline spinlock_t *rht_bucket_lock(const struct bucket_table *tbl,
358 unsigned int hash)
359{
360 return &tbl->locks[hash & tbl->locks_mask];
361}
362
363#ifdef CONFIG_PROVE_LOCKING
364int lockdep_rht_mutex_is_held(struct rhashtable *ht);
365int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
366#else
367static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
368{
369 return 1;
370}
371
372static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
373 u32 hash)
374{
375 return 1;
376}
377#endif /* CONFIG_PROVE_LOCKING */
378
379int rhashtable_init(struct rhashtable *ht,
380 const struct rhashtable_params *params);
381int rhltable_init(struct rhltable *hlt,
382 const struct rhashtable_params *params);
383
384void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
385 struct rhash_head *obj);
386
387void rhashtable_walk_enter(struct rhashtable *ht,
388 struct rhashtable_iter *iter);
389void rhashtable_walk_exit(struct rhashtable_iter *iter);
390int rhashtable_walk_start_check(struct rhashtable_iter *iter) __acquires(RCU);
391
392static inline void rhashtable_walk_start(struct rhashtable_iter *iter)
393{
394 (void)rhashtable_walk_start_check(iter);
395}
396
397void *rhashtable_walk_next(struct rhashtable_iter *iter);
398void *rhashtable_walk_peek(struct rhashtable_iter *iter);
399void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
400
401void rhashtable_free_and_destroy(struct rhashtable *ht,
402 void (*free_fn)(void *ptr, void *arg),
403 void *arg);
404void rhashtable_destroy(struct rhashtable *ht);
405
406struct rhash_head __rcu **rht_bucket_nested(const struct bucket_table *tbl,
407 unsigned int hash);
408struct rhash_head __rcu **rht_bucket_nested_insert(struct rhashtable *ht,
409 struct bucket_table *tbl,
410 unsigned int hash);
411
412#define rht_dereference(p, ht) \
413 rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
414
415#define rht_dereference_rcu(p, ht) \
416 rcu_dereference_check(p, lockdep_rht_mutex_is_held(ht))
417
418#define rht_dereference_bucket(p, tbl, hash) \
419 rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
420
421#define rht_dereference_bucket_rcu(p, tbl, hash) \
422 rcu_dereference_check(p, lockdep_rht_bucket_is_held(tbl, hash))
423
424#define rht_entry(tpos, pos, member) \
425 ({ tpos = container_of(pos, typeof(*tpos), member); 1; })
426
427static inline struct rhash_head __rcu *const *rht_bucket(
428 const struct bucket_table *tbl, unsigned int hash)
429{
430 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
431 &tbl->buckets[hash];
432}
433
434static inline struct rhash_head __rcu **rht_bucket_var(
435 struct bucket_table *tbl, unsigned int hash)
436{
437 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
438 &tbl->buckets[hash];
439}
440
441static inline struct rhash_head __rcu **rht_bucket_insert(
442 struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash)
443{
444 return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) :
445 &tbl->buckets[hash];
446}
447
448/**
449 * rht_for_each_continue - continue iterating over hash chain
450 * @pos: the &struct rhash_head to use as a loop cursor.
451 * @head: the previous &struct rhash_head to continue from
452 * @tbl: the &struct bucket_table
453 * @hash: the hash value / bucket index
454 */
455#define rht_for_each_continue(pos, head, tbl, hash) \
456 for (pos = rht_dereference_bucket(head, tbl, hash); \
457 !rht_is_a_nulls(pos); \
458 pos = rht_dereference_bucket((pos)->next, tbl, hash))
459
460/**
461 * rht_for_each - iterate over hash chain
462 * @pos: the &struct rhash_head to use as a loop cursor.
463 * @tbl: the &struct bucket_table
464 * @hash: the hash value / bucket index
465 */
466#define rht_for_each(pos, tbl, hash) \
467 rht_for_each_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
468
469/**
470 * rht_for_each_entry_continue - continue iterating over hash chain
471 * @tpos: the type * to use as a loop cursor.
472 * @pos: the &struct rhash_head to use as a loop cursor.
473 * @head: the previous &struct rhash_head to continue from
474 * @tbl: the &struct bucket_table
475 * @hash: the hash value / bucket index
476 * @member: name of the &struct rhash_head within the hashable struct.
477 */
478#define rht_for_each_entry_continue(tpos, pos, head, tbl, hash, member) \
479 for (pos = rht_dereference_bucket(head, tbl, hash); \
480 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
481 pos = rht_dereference_bucket((pos)->next, tbl, hash))
482
483/**
484 * rht_for_each_entry - iterate over hash chain of given type
485 * @tpos: the type * to use as a loop cursor.
486 * @pos: the &struct rhash_head to use as a loop cursor.
487 * @tbl: the &struct bucket_table
488 * @hash: the hash value / bucket index
489 * @member: name of the &struct rhash_head within the hashable struct.
490 */
491#define rht_for_each_entry(tpos, pos, tbl, hash, member) \
492 rht_for_each_entry_continue(tpos, pos, *rht_bucket(tbl, hash), \
493 tbl, hash, member)
494
495/**
496 * rht_for_each_entry_safe - safely iterate over hash chain of given type
497 * @tpos: the type * to use as a loop cursor.
498 * @pos: the &struct rhash_head to use as a loop cursor.
499 * @next: the &struct rhash_head to use as next in loop cursor.
500 * @tbl: the &struct bucket_table
501 * @hash: the hash value / bucket index
502 * @member: name of the &struct rhash_head within the hashable struct.
503 *
504 * This hash chain list-traversal primitive allows for the looped code to
505 * remove the loop cursor from the list.
506 */
507#define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \
508 for (pos = rht_dereference_bucket(*rht_bucket(tbl, hash), tbl, hash), \
509 next = !rht_is_a_nulls(pos) ? \
510 rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
511 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
512 pos = next, \
513 next = !rht_is_a_nulls(pos) ? \
514 rht_dereference_bucket(pos->next, tbl, hash) : NULL)
515
516/**
517 * rht_for_each_rcu_continue - continue iterating over rcu hash chain
518 * @pos: the &struct rhash_head to use as a loop cursor.
519 * @head: the previous &struct rhash_head to continue from
520 * @tbl: the &struct bucket_table
521 * @hash: the hash value / bucket index
522 *
523 * This hash chain list-traversal primitive may safely run concurrently with
524 * the _rcu mutation primitives such as rhashtable_insert() as long as the
525 * traversal is guarded by rcu_read_lock().
526 */
527#define rht_for_each_rcu_continue(pos, head, tbl, hash) \
528 for (({barrier(); }), \
529 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
530 !rht_is_a_nulls(pos); \
531 pos = rcu_dereference_raw(pos->next))
532
533/**
534 * rht_for_each_rcu - iterate over rcu hash chain
535 * @pos: the &struct rhash_head to use as a loop cursor.
536 * @tbl: the &struct bucket_table
537 * @hash: the hash value / bucket index
538 *
539 * This hash chain list-traversal primitive may safely run concurrently with
540 * the _rcu mutation primitives such as rhashtable_insert() as long as the
541 * traversal is guarded by rcu_read_lock().
542 */
543#define rht_for_each_rcu(pos, tbl, hash) \
544 rht_for_each_rcu_continue(pos, *rht_bucket(tbl, hash), tbl, hash)
545
546/**
547 * rht_for_each_entry_rcu_continue - continue iterating over rcu hash chain
548 * @tpos: the type * to use as a loop cursor.
549 * @pos: the &struct rhash_head to use as a loop cursor.
550 * @head: the previous &struct rhash_head to continue from
551 * @tbl: the &struct bucket_table
552 * @hash: the hash value / bucket index
553 * @member: name of the &struct rhash_head within the hashable struct.
554 *
555 * This hash chain list-traversal primitive may safely run concurrently with
556 * the _rcu mutation primitives such as rhashtable_insert() as long as the
557 * traversal is guarded by rcu_read_lock().
558 */
559#define rht_for_each_entry_rcu_continue(tpos, pos, head, tbl, hash, member) \
560 for (({barrier(); }), \
561 pos = rht_dereference_bucket_rcu(head, tbl, hash); \
562 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
563 pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
564
565/**
566 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
567 * @tpos: the type * to use as a loop cursor.
568 * @pos: the &struct rhash_head to use as a loop cursor.
569 * @tbl: the &struct bucket_table
570 * @hash: the hash value / bucket index
571 * @member: name of the &struct rhash_head within the hashable struct.
572 *
573 * This hash chain list-traversal primitive may safely run concurrently with
574 * the _rcu mutation primitives such as rhashtable_insert() as long as the
575 * traversal is guarded by rcu_read_lock().
576 */
577#define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \
578 rht_for_each_entry_rcu_continue(tpos, pos, *rht_bucket(tbl, hash), \
579 tbl, hash, member)
580
581/**
582 * rhl_for_each_rcu - iterate over rcu hash table list
583 * @pos: the &struct rlist_head to use as a loop cursor.
584 * @list: the head of the list
585 *
586 * This hash chain list-traversal primitive should be used on the
587 * list returned by rhltable_lookup.
588 */
589#define rhl_for_each_rcu(pos, list) \
590 for (pos = list; pos; pos = rcu_dereference_raw(pos->next))
591
592/**
593 * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type
594 * @tpos: the type * to use as a loop cursor.
595 * @pos: the &struct rlist_head to use as a loop cursor.
596 * @list: the head of the list
597 * @member: name of the &struct rlist_head within the hashable struct.
598 *
599 * This hash chain list-traversal primitive should be used on the
600 * list returned by rhltable_lookup.
601 */
602#define rhl_for_each_entry_rcu(tpos, pos, list, member) \
603 for (pos = list; pos && rht_entry(tpos, pos, member); \
604 pos = rcu_dereference_raw(pos->next))
605
606static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
607 const void *obj)
608{
609 struct rhashtable *ht = arg->ht;
610 const char *ptr = obj;
611
612 return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
613}
614
615/* Internal function, do not use. */
616static inline struct rhash_head *__rhashtable_lookup(
617 struct rhashtable *ht, const void *key,
618 const struct rhashtable_params params)
619{
620 struct rhashtable_compare_arg arg = {
621 .ht = ht,
622 .key = key,
623 };
624 struct bucket_table *tbl;
625 struct rhash_head *he;
626 unsigned int hash;
627
628 tbl = rht_dereference_rcu(ht->tbl, ht);
629restart:
630 hash = rht_key_hashfn(ht, tbl, key, params);
631 rht_for_each_rcu(he, tbl, hash) {
632 if (params.obj_cmpfn ?
633 params.obj_cmpfn(&arg, rht_obj(ht, he)) :
634 rhashtable_compare(&arg, rht_obj(ht, he)))
635 continue;
636 return he;
637 }
638
639 /* Ensure we see any new tables. */
640 smp_rmb();
641
642 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
643 if (unlikely(tbl))
644 goto restart;
645
646 return NULL;
647}
648
649/**
650 * rhashtable_lookup - search hash table
651 * @ht: hash table
652 * @key: the pointer to the key
653 * @params: hash table parameters
654 *
655 * Computes the hash value for the key and traverses the bucket chain looking
656 * for a entry with an identical key. The first matching entry is returned.
657 *
658 * This must only be called under the RCU read lock.
659 *
660 * Returns the first entry on which the compare function returned true.
661 */
662static inline void *rhashtable_lookup(
663 struct rhashtable *ht, const void *key,
664 const struct rhashtable_params params)
665{
666 struct rhash_head *he = __rhashtable_lookup(ht, key, params);
667
668 return he ? rht_obj(ht, he) : NULL;
669}
670
671/**
672 * rhashtable_lookup_fast - search hash table, without RCU read lock
673 * @ht: hash table
674 * @key: the pointer to the key
675 * @params: hash table parameters
676 *
677 * Computes the hash value for the key and traverses the bucket chain looking
678 * for a entry with an identical key. The first matching entry is returned.
679 *
680 * Only use this function when you have other mechanisms guaranteeing
681 * that the object won't go away after the RCU read lock is released.
682 *
683 * Returns the first entry on which the compare function returned true.
684 */
685static inline void *rhashtable_lookup_fast(
686 struct rhashtable *ht, const void *key,
687 const struct rhashtable_params params)
688{
689 void *obj;
690
691 rcu_read_lock();
692 obj = rhashtable_lookup(ht, key, params);
693 rcu_read_unlock();
694
695 return obj;
696}
697
698/**
699 * rhltable_lookup - search hash list table
700 * @hlt: hash table
701 * @key: the pointer to the key
702 * @params: hash table parameters
703 *
704 * Computes the hash value for the key and traverses the bucket chain looking
705 * for a entry with an identical key. All matching entries are returned
706 * in a list.
707 *
708 * This must only be called under the RCU read lock.
709 *
710 * Returns the list of entries that match the given key.
711 */
712static inline struct rhlist_head *rhltable_lookup(
713 struct rhltable *hlt, const void *key,
714 const struct rhashtable_params params)
715{
716 struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params);
717
718 return he ? container_of(he, struct rhlist_head, rhead) : NULL;
719}
720
721/* Internal function, please use rhashtable_insert_fast() instead. This
722 * function returns the existing element already in hashes in there is a clash,
723 * otherwise it returns an error via ERR_PTR().
724 */
725static inline void *__rhashtable_insert_fast(
726 struct rhashtable *ht, const void *key, struct rhash_head *obj,
727 const struct rhashtable_params params, bool rhlist)
728{
729 struct rhashtable_compare_arg arg = {
730 .ht = ht,
731 .key = key,
732 };
733 struct rhash_head __rcu **pprev;
734 struct bucket_table *tbl;
735 struct rhash_head *head;
736 spinlock_t *lock;
737 unsigned int hash;
738 int elasticity;
739 void *data;
740
741 rcu_read_lock();
742
743 tbl = rht_dereference_rcu(ht->tbl, ht);
744 hash = rht_head_hashfn(ht, tbl, obj, params);
745 lock = rht_bucket_lock(tbl, hash);
746 spin_lock_bh(lock);
747
748 if (unlikely(rht_dereference_bucket(tbl->future_tbl, tbl, hash))) {
749slow_path:
750 spin_unlock_bh(lock);
751 rcu_read_unlock();
752 return rhashtable_insert_slow(ht, key, obj);
753 }
754
755 elasticity = RHT_ELASTICITY;
756 pprev = rht_bucket_insert(ht, tbl, hash);
757 data = ERR_PTR(-ENOMEM);
758 if (!pprev)
759 goto out;
760
761 rht_for_each_continue(head, *pprev, tbl, hash) {
762 struct rhlist_head *plist;
763 struct rhlist_head *list;
764
765 elasticity--;
766 if (!key ||
767 (params.obj_cmpfn ?
768 params.obj_cmpfn(&arg, rht_obj(ht, head)) :
769 rhashtable_compare(&arg, rht_obj(ht, head)))) {
770 pprev = &head->next;
771 continue;
772 }
773
774 data = rht_obj(ht, head);
775
776 if (!rhlist)
777 goto out;
778
779
780 list = container_of(obj, struct rhlist_head, rhead);
781 plist = container_of(head, struct rhlist_head, rhead);
782
783 RCU_INIT_POINTER(list->next, plist);
784 head = rht_dereference_bucket(head->next, tbl, hash);
785 RCU_INIT_POINTER(list->rhead.next, head);
786 rcu_assign_pointer(*pprev, obj);
787
788 goto good;
789 }
790
791 if (elasticity <= 0)
792 goto slow_path;
793
794 data = ERR_PTR(-E2BIG);
795 if (unlikely(rht_grow_above_max(ht, tbl)))
796 goto out;
797
798 if (unlikely(rht_grow_above_100(ht, tbl)))
799 goto slow_path;
800
801 head = rht_dereference_bucket(*pprev, tbl, hash);
802
803 RCU_INIT_POINTER(obj->next, head);
804 if (rhlist) {
805 struct rhlist_head *list;
806
807 list = container_of(obj, struct rhlist_head, rhead);
808 RCU_INIT_POINTER(list->next, NULL);
809 }
810
811 rcu_assign_pointer(*pprev, obj);
812
813 atomic_inc(&ht->nelems);
814 if (rht_grow_above_75(ht, tbl))
815 schedule_work(&ht->run_work);
816
817good:
818 data = NULL;
819
820out:
821 spin_unlock_bh(lock);
822 rcu_read_unlock();
823
824 return data;
825}
826
827/**
828 * rhashtable_insert_fast - insert object into hash table
829 * @ht: hash table
830 * @obj: pointer to hash head inside object
831 * @params: hash table parameters
832 *
833 * Will take a per bucket spinlock to protect against mutual mutations
834 * on the same bucket. Multiple insertions may occur in parallel unless
835 * they map to the same bucket lock.
836 *
837 * It is safe to call this function from atomic context.
838 *
839 * Will trigger an automatic deferred table resizing if the size grows
840 * beyond the watermark indicated by grow_decision() which can be passed
841 * to rhashtable_init().
842 */
843static inline int rhashtable_insert_fast(
844 struct rhashtable *ht, struct rhash_head *obj,
845 const struct rhashtable_params params)
846{
847 void *ret;
848
849 ret = __rhashtable_insert_fast(ht, NULL, obj, params, false);
850 if (IS_ERR(ret))
851 return PTR_ERR(ret);
852
853 return ret == NULL ? 0 : -EEXIST;
854}
855
856/**
857 * rhltable_insert_key - insert object into hash list table
858 * @hlt: hash list table
859 * @key: the pointer to the key
860 * @list: pointer to hash list head inside object
861 * @params: hash table parameters
862 *
863 * Will take a per bucket spinlock to protect against mutual mutations
864 * on the same bucket. Multiple insertions may occur in parallel unless
865 * they map to the same bucket lock.
866 *
867 * It is safe to call this function from atomic context.
868 *
869 * Will trigger an automatic deferred table resizing if the size grows
870 * beyond the watermark indicated by grow_decision() which can be passed
871 * to rhashtable_init().
872 */
873static inline int rhltable_insert_key(
874 struct rhltable *hlt, const void *key, struct rhlist_head *list,
875 const struct rhashtable_params params)
876{
877 return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead,
878 params, true));
879}
880
881/**
882 * rhltable_insert - insert object into hash list table
883 * @hlt: hash list table
884 * @list: pointer to hash list head inside object
885 * @params: hash table parameters
886 *
887 * Will take a per bucket spinlock to protect against mutual mutations
888 * on the same bucket. Multiple insertions may occur in parallel unless
889 * they map to the same bucket lock.
890 *
891 * It is safe to call this function from atomic context.
892 *
893 * Will trigger an automatic deferred table resizing if the size grows
894 * beyond the watermark indicated by grow_decision() which can be passed
895 * to rhashtable_init().
896 */
897static inline int rhltable_insert(
898 struct rhltable *hlt, struct rhlist_head *list,
899 const struct rhashtable_params params)
900{
901 const char *key = rht_obj(&hlt->ht, &list->rhead);
902
903 key += params.key_offset;
904
905 return rhltable_insert_key(hlt, key, list, params);
906}
907
908/**
909 * rhashtable_lookup_insert_fast - lookup and insert object into hash table
910 * @ht: hash table
911 * @obj: pointer to hash head inside object
912 * @params: hash table parameters
913 *
914 * Locks down the bucket chain in both the old and new table if a resize
915 * is in progress to ensure that writers can't remove from the old table
916 * and can't insert to the new table during the atomic operation of search
917 * and insertion. Searches for duplicates in both the old and new table if
918 * a resize is in progress.
919 *
920 * This lookup function may only be used for fixed key hash table (key_len
921 * parameter set). It will BUG() if used inappropriately.
922 *
923 * It is safe to call this function from atomic context.
924 *
925 * Will trigger an automatic deferred table resizing if the size grows
926 * beyond the watermark indicated by grow_decision() which can be passed
927 * to rhashtable_init().
928 */
929static inline int rhashtable_lookup_insert_fast(
930 struct rhashtable *ht, struct rhash_head *obj,
931 const struct rhashtable_params params)
932{
933 const char *key = rht_obj(ht, obj);
934 void *ret;
935
936 BUG_ON(ht->p.obj_hashfn);
937
938 ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
939 false);
940 if (IS_ERR(ret))
941 return PTR_ERR(ret);
942
943 return ret == NULL ? 0 : -EEXIST;
944}
945
946/**
947 * rhashtable_lookup_get_insert_fast - lookup and insert object into hash table
948 * @ht: hash table
949 * @obj: pointer to hash head inside object
950 * @params: hash table parameters
951 *
952 * Just like rhashtable_lookup_insert_fast(), but this function returns the
953 * object if it exists, NULL if it did not and the insertion was successful,
954 * and an ERR_PTR otherwise.
955 */
956static inline void *rhashtable_lookup_get_insert_fast(
957 struct rhashtable *ht, struct rhash_head *obj,
958 const struct rhashtable_params params)
959{
960 const char *key = rht_obj(ht, obj);
961
962 BUG_ON(ht->p.obj_hashfn);
963
964 return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
965 false);
966}
967
968/**
969 * rhashtable_lookup_insert_key - search and insert object to hash table
970 * with explicit key
971 * @ht: hash table
972 * @key: key
973 * @obj: pointer to hash head inside object
974 * @params: hash table parameters
975 *
976 * Locks down the bucket chain in both the old and new table if a resize
977 * is in progress to ensure that writers can't remove from the old table
978 * and can't insert to the new table during the atomic operation of search
979 * and insertion. Searches for duplicates in both the old and new table if
980 * a resize is in progress.
981 *
982 * Lookups may occur in parallel with hashtable mutations and resizing.
983 *
984 * Will trigger an automatic deferred table resizing if the size grows
985 * beyond the watermark indicated by grow_decision() which can be passed
986 * to rhashtable_init().
987 *
988 * Returns zero on success.
989 */
990static inline int rhashtable_lookup_insert_key(
991 struct rhashtable *ht, const void *key, struct rhash_head *obj,
992 const struct rhashtable_params params)
993{
994 void *ret;
995
996 BUG_ON(!ht->p.obj_hashfn || !key);
997
998 ret = __rhashtable_insert_fast(ht, key, obj, params, false);
999 if (IS_ERR(ret))
1000 return PTR_ERR(ret);
1001
1002 return ret == NULL ? 0 : -EEXIST;
1003}
1004
1005/**
1006 * rhashtable_lookup_get_insert_key - lookup and insert object into hash table
1007 * @ht: hash table
1008 * @obj: pointer to hash head inside object
1009 * @params: hash table parameters
1010 * @data: pointer to element data already in hashes
1011 *
1012 * Just like rhashtable_lookup_insert_key(), but this function returns the
1013 * object if it exists, NULL if it does not and the insertion was successful,
1014 * and an ERR_PTR otherwise.
1015 */
1016static inline void *rhashtable_lookup_get_insert_key(
1017 struct rhashtable *ht, const void *key, struct rhash_head *obj,
1018 const struct rhashtable_params params)
1019{
1020 BUG_ON(!ht->p.obj_hashfn || !key);
1021
1022 return __rhashtable_insert_fast(ht, key, obj, params, false);
1023}
1024
1025/* Internal function, please use rhashtable_remove_fast() instead */
1026static inline int __rhashtable_remove_fast_one(
1027 struct rhashtable *ht, struct bucket_table *tbl,
1028 struct rhash_head *obj, const struct rhashtable_params params,
1029 bool rhlist)
1030{
1031 struct rhash_head __rcu **pprev;
1032 struct rhash_head *he;
1033 spinlock_t * lock;
1034 unsigned int hash;
1035 int err = -ENOENT;
1036
1037 hash = rht_head_hashfn(ht, tbl, obj, params);
1038 lock = rht_bucket_lock(tbl, hash);
1039
1040 spin_lock_bh(lock);
1041
1042 pprev = rht_bucket_var(tbl, hash);
1043 rht_for_each_continue(he, *pprev, tbl, hash) {
1044 struct rhlist_head *list;
1045
1046 list = container_of(he, struct rhlist_head, rhead);
1047
1048 if (he != obj) {
1049 struct rhlist_head __rcu **lpprev;
1050
1051 pprev = &he->next;
1052
1053 if (!rhlist)
1054 continue;
1055
1056 do {
1057 lpprev = &list->next;
1058 list = rht_dereference_bucket(list->next,
1059 tbl, hash);
1060 } while (list && obj != &list->rhead);
1061
1062 if (!list)
1063 continue;
1064
1065 list = rht_dereference_bucket(list->next, tbl, hash);
1066 RCU_INIT_POINTER(*lpprev, list);
1067 err = 0;
1068 break;
1069 }
1070
1071 obj = rht_dereference_bucket(obj->next, tbl, hash);
1072 err = 1;
1073
1074 if (rhlist) {
1075 list = rht_dereference_bucket(list->next, tbl, hash);
1076 if (list) {
1077 RCU_INIT_POINTER(list->rhead.next, obj);
1078 obj = &list->rhead;
1079 err = 0;
1080 }
1081 }
1082
1083 rcu_assign_pointer(*pprev, obj);
1084 break;
1085 }
1086
1087 spin_unlock_bh(lock);
1088
1089 if (err > 0) {
1090 atomic_dec(&ht->nelems);
1091 if (unlikely(ht->p.automatic_shrinking &&
1092 rht_shrink_below_30(ht, tbl)))
1093 schedule_work(&ht->run_work);
1094 err = 0;
1095 }
1096
1097 return err;
1098}
1099
1100/* Internal function, please use rhashtable_remove_fast() instead */
1101static inline int __rhashtable_remove_fast(
1102 struct rhashtable *ht, struct rhash_head *obj,
1103 const struct rhashtable_params params, bool rhlist)
1104{
1105 struct bucket_table *tbl;
1106 int err;
1107
1108 rcu_read_lock();
1109
1110 tbl = rht_dereference_rcu(ht->tbl, ht);
1111
1112 /* Because we have already taken (and released) the bucket
1113 * lock in old_tbl, if we find that future_tbl is not yet
1114 * visible then that guarantees the entry to still be in
1115 * the old tbl if it exists.
1116 */
1117 while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params,
1118 rhlist)) &&
1119 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1120 ;
1121
1122 rcu_read_unlock();
1123
1124 return err;
1125}
1126
1127/**
1128 * rhashtable_remove_fast - remove object from hash table
1129 * @ht: hash table
1130 * @obj: pointer to hash head inside object
1131 * @params: hash table parameters
1132 *
1133 * Since the hash chain is single linked, the removal operation needs to
1134 * walk the bucket chain upon removal. The removal operation is thus
1135 * considerable slow if the hash table is not correctly sized.
1136 *
1137 * Will automatically shrink the table via rhashtable_expand() if the
1138 * shrink_decision function specified at rhashtable_init() returns true.
1139 *
1140 * Returns zero on success, -ENOENT if the entry could not be found.
1141 */
1142static inline int rhashtable_remove_fast(
1143 struct rhashtable *ht, struct rhash_head *obj,
1144 const struct rhashtable_params params)
1145{
1146 return __rhashtable_remove_fast(ht, obj, params, false);
1147}
1148
1149/**
1150 * rhltable_remove - remove object from hash list table
1151 * @hlt: hash list table
1152 * @list: pointer to hash list head inside object
1153 * @params: hash table parameters
1154 *
1155 * Since the hash chain is single linked, the removal operation needs to
1156 * walk the bucket chain upon removal. The removal operation is thus
1157 * considerable slow if the hash table is not correctly sized.
1158 *
1159 * Will automatically shrink the table via rhashtable_expand() if the
1160 * shrink_decision function specified at rhashtable_init() returns true.
1161 *
1162 * Returns zero on success, -ENOENT if the entry could not be found.
1163 */
1164static inline int rhltable_remove(
1165 struct rhltable *hlt, struct rhlist_head *list,
1166 const struct rhashtable_params params)
1167{
1168 return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true);
1169}
1170
1171/* Internal function, please use rhashtable_replace_fast() instead */
1172static inline int __rhashtable_replace_fast(
1173 struct rhashtable *ht, struct bucket_table *tbl,
1174 struct rhash_head *obj_old, struct rhash_head *obj_new,
1175 const struct rhashtable_params params)
1176{
1177 struct rhash_head __rcu **pprev;
1178 struct rhash_head *he;
1179 spinlock_t *lock;
1180 unsigned int hash;
1181 int err = -ENOENT;
1182
1183 /* Minimally, the old and new objects must have same hash
1184 * (which should mean identifiers are the same).
1185 */
1186 hash = rht_head_hashfn(ht, tbl, obj_old, params);
1187 if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
1188 return -EINVAL;
1189
1190 lock = rht_bucket_lock(tbl, hash);
1191
1192 spin_lock_bh(lock);
1193
1194 pprev = rht_bucket_var(tbl, hash);
1195 rht_for_each_continue(he, *pprev, tbl, hash) {
1196 if (he != obj_old) {
1197 pprev = &he->next;
1198 continue;
1199 }
1200
1201 rcu_assign_pointer(obj_new->next, obj_old->next);
1202 rcu_assign_pointer(*pprev, obj_new);
1203 err = 0;
1204 break;
1205 }
1206
1207 spin_unlock_bh(lock);
1208
1209 return err;
1210}
1211
1212/**
1213 * rhashtable_replace_fast - replace an object in hash table
1214 * @ht: hash table
1215 * @obj_old: pointer to hash head inside object being replaced
1216 * @obj_new: pointer to hash head inside object which is new
1217 * @params: hash table parameters
1218 *
1219 * Replacing an object doesn't affect the number of elements in the hash table
1220 * or bucket, so we don't need to worry about shrinking or expanding the
1221 * table here.
1222 *
1223 * Returns zero on success, -ENOENT if the entry could not be found,
1224 * -EINVAL if hash is not the same for the old and new objects.
1225 */
1226static inline int rhashtable_replace_fast(
1227 struct rhashtable *ht, struct rhash_head *obj_old,
1228 struct rhash_head *obj_new,
1229 const struct rhashtable_params params)
1230{
1231 struct bucket_table *tbl;
1232 int err;
1233
1234 rcu_read_lock();
1235
1236 tbl = rht_dereference_rcu(ht->tbl, ht);
1237
1238 /* Because we have already taken (and released) the bucket
1239 * lock in old_tbl, if we find that future_tbl is not yet
1240 * visible then that guarantees the entry to still be in
1241 * the old tbl if it exists.
1242 */
1243 while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
1244 obj_new, params)) &&
1245 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1246 ;
1247
1248 rcu_read_unlock();
1249
1250 return err;
1251}
1252
1253/* Obsolete function, do not use in new code. */
1254static inline int rhashtable_walk_init(struct rhashtable *ht,
1255 struct rhashtable_iter *iter, gfp_t gfp)
1256{
1257 rhashtable_walk_enter(ht, iter);
1258 return 0;
1259}
1260
1261/**
1262 * rhltable_walk_enter - Initialise an iterator
1263 * @hlt: Table to walk over
1264 * @iter: Hash table Iterator
1265 *
1266 * This function prepares a hash table walk.
1267 *
1268 * Note that if you restart a walk after rhashtable_walk_stop you
1269 * may see the same object twice. Also, you may miss objects if
1270 * there are removals in between rhashtable_walk_stop and the next
1271 * call to rhashtable_walk_start.
1272 *
1273 * For a completely stable walk you should construct your own data
1274 * structure outside the hash table.
1275 *
1276 * This function may sleep so you must not call it from interrupt
1277 * context or with spin locks held.
1278 *
1279 * You must call rhashtable_walk_exit after this function returns.
1280 */
1281static inline void rhltable_walk_enter(struct rhltable *hlt,
1282 struct rhashtable_iter *iter)
1283{
1284 return rhashtable_walk_enter(&hlt->ht, iter);
1285}
1286
1287/**
1288 * rhltable_free_and_destroy - free elements and destroy hash list table
1289 * @hlt: the hash list table to destroy
1290 * @free_fn: callback to release resources of element
1291 * @arg: pointer passed to free_fn
1292 *
1293 * See documentation for rhashtable_free_and_destroy.
1294 */
1295static inline void rhltable_free_and_destroy(struct rhltable *hlt,
1296 void (*free_fn)(void *ptr,
1297 void *arg),
1298 void *arg)
1299{
1300 return rhashtable_free_and_destroy(&hlt->ht, free_fn, arg);
1301}
1302
1303static inline void rhltable_destroy(struct rhltable *hlt)
1304{
1305 return rhltable_free_and_destroy(hlt, NULL, NULL);
1306}
1307
1308#endif /* _LINUX_RHASHTABLE_H */