Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Resizable, Scalable, Concurrent Hash Table
4 *
5 * Copyright (c) 2015-2016 Herbert Xu <herbert@gondor.apana.org.au>
6 * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
7 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
8 *
9 * Code partially derived from nft_hash
10 * Rewritten with rehash code from br_multicast plus single list
11 * pointer as suggested by Josh Triplett
12 *
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License version 2 as
15 * published by the Free Software Foundation.
16 */
17
18#ifndef _LINUX_RHASHTABLE_H
19#define _LINUX_RHASHTABLE_H
20
21#include <linux/err.h>
22#include <linux/errno.h>
23#include <linux/jhash.h>
24#include <linux/list_nulls.h>
25#include <linux/workqueue.h>
26#include <linux/rculist.h>
27#include <linux/bit_spinlock.h>
28
29#include <linux/rhashtable-types.h>
30/*
31 * Objects in an rhashtable have an embedded struct rhash_head
32 * which is linked into as hash chain from the hash table - or one
33 * of two or more hash tables when the rhashtable is being resized.
34 * The end of the chain is marked with a special nulls marks which has
35 * the least significant bit set but otherwise stores the address of
36 * the hash bucket. This allows us to be sure we've found the end
37 * of the right list.
38 * The value stored in the hash bucket has BIT(0) used as a lock bit.
39 * This bit must be atomically set before any changes are made to
40 * the chain. To avoid dereferencing this pointer without clearing
41 * the bit first, we use an opaque 'struct rhash_lock_head *' for the
42 * pointer stored in the bucket. This struct needs to be defined so
43 * that rcu_dereference() works on it, but it has no content so a
44 * cast is needed for it to be useful. This ensures it isn't
45 * used by mistake with clearing the lock bit first.
46 */
47struct rhash_lock_head {};
48
49/* Maximum chain length before rehash
50 *
51 * The maximum (not average) chain length grows with the size of the hash
52 * table, at a rate of (log N)/(log log N).
53 *
54 * The value of 16 is selected so that even if the hash table grew to
55 * 2^32 you would not expect the maximum chain length to exceed it
56 * unless we are under attack (or extremely unlucky).
57 *
58 * As this limit is only to detect attacks, we don't need to set it to a
59 * lower value as you'd need the chain length to vastly exceed 16 to have
60 * any real effect on the system.
61 */
62#define RHT_ELASTICITY 16u
63
64/**
65 * struct bucket_table - Table of hash buckets
66 * @size: Number of hash buckets
67 * @nest: Number of bits of first-level nested table.
68 * @rehash: Current bucket being rehashed
69 * @hash_rnd: Random seed to fold into hash
70 * @walkers: List of active walkers
71 * @rcu: RCU structure for freeing the table
72 * @future_tbl: Table under construction during rehashing
73 * @ntbl: Nested table used when out of memory.
74 * @buckets: size * hash buckets
75 */
76struct bucket_table {
77 unsigned int size;
78 unsigned int nest;
79 u32 hash_rnd;
80 struct list_head walkers;
81 struct rcu_head rcu;
82
83 struct bucket_table __rcu *future_tbl;
84
85 struct lockdep_map dep_map;
86
87 struct rhash_lock_head __rcu *buckets[] ____cacheline_aligned_in_smp;
88};
89
90/*
91 * NULLS_MARKER() expects a hash value with the low
92 * bits mostly likely to be significant, and it discards
93 * the msb.
94 * We give it an address, in which the bottom bit is
95 * always 0, and the msb might be significant.
96 * So we shift the address down one bit to align with
97 * expectations and avoid losing a significant bit.
98 *
99 * We never store the NULLS_MARKER in the hash table
100 * itself as we need the lsb for locking.
101 * Instead we store a NULL
102 */
103#define RHT_NULLS_MARKER(ptr) \
104 ((void *)NULLS_MARKER(((unsigned long) (ptr)) >> 1))
105#define INIT_RHT_NULLS_HEAD(ptr) \
106 ((ptr) = NULL)
107
108static inline bool rht_is_a_nulls(const struct rhash_head *ptr)
109{
110 return ((unsigned long) ptr & 1);
111}
112
113static inline void *rht_obj(const struct rhashtable *ht,
114 const struct rhash_head *he)
115{
116 return (char *)he - ht->p.head_offset;
117}
118
119static inline unsigned int rht_bucket_index(const struct bucket_table *tbl,
120 unsigned int hash)
121{
122 return hash & (tbl->size - 1);
123}
124
125static __always_inline unsigned int rht_key_get_hash(struct rhashtable *ht,
126 const void *key, const struct rhashtable_params params,
127 unsigned int hash_rnd)
128{
129 unsigned int hash;
130
131 /* params must be equal to ht->p if it isn't constant. */
132 if (!__builtin_constant_p(params.key_len)) {
133 hash = ht->p.hashfn(key, ht->key_len, hash_rnd);
134 } else {
135 unsigned int key_len = params.key_len ? : ht->p.key_len;
136
137 if (params.hashfn)
138 hash = params.hashfn(key, key_len, hash_rnd);
139 else if (key_len & (sizeof(u32) - 1))
140 hash = jhash(key, key_len, hash_rnd);
141 else
142 hash = jhash2(key, key_len / sizeof(u32), hash_rnd);
143 }
144
145 return hash;
146}
147
148static __always_inline unsigned int rht_key_hashfn(
149 struct rhashtable *ht, const struct bucket_table *tbl,
150 const void *key, const struct rhashtable_params params)
151{
152 unsigned int hash = rht_key_get_hash(ht, key, params, tbl->hash_rnd);
153
154 return rht_bucket_index(tbl, hash);
155}
156
157static __always_inline unsigned int rht_head_hashfn(
158 struct rhashtable *ht, const struct bucket_table *tbl,
159 const struct rhash_head *he, const struct rhashtable_params params)
160{
161 const char *ptr = rht_obj(ht, he);
162
163 return likely(params.obj_hashfn) ?
164 rht_bucket_index(tbl, params.obj_hashfn(ptr, params.key_len ?:
165 ht->p.key_len,
166 tbl->hash_rnd)) :
167 rht_key_hashfn(ht, tbl, ptr + params.key_offset, params);
168}
169
170/**
171 * rht_grow_above_75 - returns true if nelems > 0.75 * table-size
172 * @ht: hash table
173 * @tbl: current table
174 */
175static inline bool rht_grow_above_75(const struct rhashtable *ht,
176 const struct bucket_table *tbl)
177{
178 /* Expand table when exceeding 75% load */
179 return atomic_read(&ht->nelems) > (tbl->size / 4 * 3) &&
180 (!ht->p.max_size || tbl->size < ht->p.max_size);
181}
182
183/**
184 * rht_shrink_below_30 - returns true if nelems < 0.3 * table-size
185 * @ht: hash table
186 * @tbl: current table
187 */
188static inline bool rht_shrink_below_30(const struct rhashtable *ht,
189 const struct bucket_table *tbl)
190{
191 /* Shrink table beneath 30% load */
192 return atomic_read(&ht->nelems) < (tbl->size * 3 / 10) &&
193 tbl->size > ht->p.min_size;
194}
195
196/**
197 * rht_grow_above_100 - returns true if nelems > table-size
198 * @ht: hash table
199 * @tbl: current table
200 */
201static inline bool rht_grow_above_100(const struct rhashtable *ht,
202 const struct bucket_table *tbl)
203{
204 return atomic_read(&ht->nelems) > tbl->size &&
205 (!ht->p.max_size || tbl->size < ht->p.max_size);
206}
207
208/**
209 * rht_grow_above_max - returns true if table is above maximum
210 * @ht: hash table
211 * @tbl: current table
212 */
213static inline bool rht_grow_above_max(const struct rhashtable *ht,
214 const struct bucket_table *tbl)
215{
216 return atomic_read(&ht->nelems) >= ht->max_elems;
217}
218
219#ifdef CONFIG_PROVE_LOCKING
220int lockdep_rht_mutex_is_held(struct rhashtable *ht);
221int lockdep_rht_bucket_is_held(const struct bucket_table *tbl, u32 hash);
222#else
223static inline int lockdep_rht_mutex_is_held(struct rhashtable *ht)
224{
225 return 1;
226}
227
228static inline int lockdep_rht_bucket_is_held(const struct bucket_table *tbl,
229 u32 hash)
230{
231 return 1;
232}
233#endif /* CONFIG_PROVE_LOCKING */
234
235void *rhashtable_insert_slow(struct rhashtable *ht, const void *key,
236 struct rhash_head *obj);
237
238void rhashtable_walk_enter(struct rhashtable *ht,
239 struct rhashtable_iter *iter);
240void rhashtable_walk_exit(struct rhashtable_iter *iter);
241int rhashtable_walk_start_check(struct rhashtable_iter *iter) __acquires_shared(RCU);
242
243static inline void rhashtable_walk_start(struct rhashtable_iter *iter)
244 __acquires_shared(RCU)
245{
246 (void)rhashtable_walk_start_check(iter);
247}
248
249void *rhashtable_walk_next(struct rhashtable_iter *iter);
250void *rhashtable_walk_peek(struct rhashtable_iter *iter);
251void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases_shared(RCU);
252
253void rhashtable_free_and_destroy(struct rhashtable *ht,
254 void (*free_fn)(void *ptr, void *arg),
255 void *arg);
256void rhashtable_destroy(struct rhashtable *ht);
257
258struct rhash_lock_head __rcu **rht_bucket_nested(
259 const struct bucket_table *tbl, unsigned int hash);
260struct rhash_lock_head __rcu **__rht_bucket_nested(
261 const struct bucket_table *tbl, unsigned int hash);
262struct rhash_lock_head __rcu **rht_bucket_nested_insert(
263 struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash);
264
265#define rht_dereference(p, ht) \
266 rcu_dereference_protected(p, lockdep_rht_mutex_is_held(ht))
267
268#define rht_dereference_rcu(p, ht) \
269 rcu_dereference_all_check(p, lockdep_rht_mutex_is_held(ht))
270
271#define rht_dereference_bucket(p, tbl, hash) \
272 rcu_dereference_protected(p, lockdep_rht_bucket_is_held(tbl, hash))
273
274#define rht_dereference_bucket_rcu(p, tbl, hash) \
275 rcu_dereference_all_check(p, lockdep_rht_bucket_is_held(tbl, hash))
276
277#define rht_entry(tpos, pos, member) \
278 ({ tpos = container_of(pos, typeof(*tpos), member); 1; })
279
280static inline struct rhash_lock_head __rcu *const *rht_bucket(
281 const struct bucket_table *tbl, unsigned int hash)
282{
283 return unlikely(tbl->nest) ? rht_bucket_nested(tbl, hash) :
284 &tbl->buckets[hash];
285}
286
287static inline struct rhash_lock_head __rcu **rht_bucket_var(
288 struct bucket_table *tbl, unsigned int hash)
289{
290 return unlikely(tbl->nest) ? __rht_bucket_nested(tbl, hash) :
291 &tbl->buckets[hash];
292}
293
294static inline struct rhash_lock_head __rcu **rht_bucket_insert(
295 struct rhashtable *ht, struct bucket_table *tbl, unsigned int hash)
296{
297 return unlikely(tbl->nest) ? rht_bucket_nested_insert(ht, tbl, hash) :
298 &tbl->buckets[hash];
299}
300
301/*
302 * We lock a bucket by setting BIT(0) in the pointer - this is always
303 * zero in real pointers. The NULLS mark is never stored in the bucket,
304 * rather we store NULL if the bucket is empty.
305 * bit_spin_locks do not handle contention well, but the whole point
306 * of the hashtable design is to achieve minimum per-bucket contention.
307 * A nested hash table might not have a bucket pointer. In that case
308 * we cannot get a lock. For remove and replace the bucket cannot be
309 * interesting and doesn't need locking.
310 * For insert we allocate the bucket if this is the last bucket_table,
311 * and then take the lock.
312 * Sometimes we unlock a bucket by writing a new pointer there. In that
313 * case we don't need to unlock, but we do need to reset state such as
314 * local_bh. For that we have rht_assign_unlock(). As rcu_assign_pointer()
315 * provides the same release semantics that bit_spin_unlock() provides,
316 * this is safe.
317 * When we write to a bucket without unlocking, we use rht_assign_locked().
318 */
319
320static inline unsigned long rht_lock(struct bucket_table *tbl,
321 struct rhash_lock_head __rcu **bkt)
322 __acquires(__bitlock(0, bkt))
323{
324 unsigned long flags;
325
326 local_irq_save(flags);
327 bit_spin_lock(0, (unsigned long *)bkt);
328 lock_map_acquire(&tbl->dep_map);
329 return flags;
330}
331
332static inline unsigned long rht_lock_nested(struct bucket_table *tbl,
333 struct rhash_lock_head __rcu **bucket,
334 unsigned int subclass)
335 __acquires(__bitlock(0, bucket))
336{
337 unsigned long flags;
338
339 local_irq_save(flags);
340 bit_spin_lock(0, (unsigned long *)bucket);
341 lock_acquire_exclusive(&tbl->dep_map, subclass, 0, NULL, _THIS_IP_);
342 return flags;
343}
344
345static inline void rht_unlock(struct bucket_table *tbl,
346 struct rhash_lock_head __rcu **bkt,
347 unsigned long flags)
348 __releases(__bitlock(0, bkt))
349{
350 lock_map_release(&tbl->dep_map);
351 bit_spin_unlock(0, (unsigned long *)bkt);
352 local_irq_restore(flags);
353}
354
355enum rht_lookup_freq {
356 RHT_LOOKUP_NORMAL,
357 RHT_LOOKUP_LIKELY,
358};
359
360static __always_inline struct rhash_head *__rht_ptr(
361 struct rhash_lock_head *p, struct rhash_lock_head __rcu *const *bkt,
362 const enum rht_lookup_freq freq)
363{
364 unsigned long p_val = (unsigned long)p & ~BIT(0);
365
366 BUILD_BUG_ON(!__builtin_constant_p(freq));
367
368 if (freq == RHT_LOOKUP_LIKELY)
369 return (struct rhash_head *)
370 (likely(p_val) ? p_val : (unsigned long)RHT_NULLS_MARKER(bkt));
371 else
372 return (struct rhash_head *)
373 (p_val ?: (unsigned long)RHT_NULLS_MARKER(bkt));
374}
375
376/*
377 * Where 'bkt' is a bucket and might be locked:
378 * rht_ptr_rcu() dereferences that pointer and clears the lock bit.
379 * rht_ptr() dereferences in a context where the bucket is locked.
380 * rht_ptr_exclusive() dereferences in a context where exclusive
381 * access is guaranteed, such as when destroying the table.
382 */
383static __always_inline struct rhash_head *__rht_ptr_rcu(
384 struct rhash_lock_head __rcu *const *bkt,
385 const enum rht_lookup_freq freq)
386{
387 return __rht_ptr(rcu_dereference_all(*bkt), bkt, freq);
388}
389
390static inline struct rhash_head *rht_ptr_rcu(
391 struct rhash_lock_head __rcu *const *bkt)
392{
393 return __rht_ptr_rcu(bkt, RHT_LOOKUP_NORMAL);
394}
395
396static inline struct rhash_head *rht_ptr(
397 struct rhash_lock_head __rcu *const *bkt,
398 struct bucket_table *tbl,
399 unsigned int hash)
400{
401 return __rht_ptr(rht_dereference_bucket(*bkt, tbl, hash), bkt,
402 RHT_LOOKUP_NORMAL);
403}
404
405static inline struct rhash_head *rht_ptr_exclusive(
406 struct rhash_lock_head __rcu *const *bkt)
407{
408 return __rht_ptr(rcu_dereference_protected(*bkt, 1), bkt,
409 RHT_LOOKUP_NORMAL);
410}
411
412static inline void rht_assign_locked(struct rhash_lock_head __rcu **bkt,
413 struct rhash_head *obj)
414{
415 if (rht_is_a_nulls(obj))
416 obj = NULL;
417 rcu_assign_pointer(*bkt, (void *)((unsigned long)obj | BIT(0)));
418}
419
420static inline void rht_assign_unlock(struct bucket_table *tbl,
421 struct rhash_lock_head __rcu **bkt,
422 struct rhash_head *obj,
423 unsigned long flags)
424 __releases(__bitlock(0, bkt))
425{
426 if (rht_is_a_nulls(obj))
427 obj = NULL;
428 lock_map_release(&tbl->dep_map);
429 rcu_assign_pointer(*bkt, (void *)obj);
430 preempt_enable();
431 __release(__bitlock(0, bkt));
432 local_irq_restore(flags);
433}
434
435/**
436 * rht_for_each_from - iterate over hash chain from given head
437 * @pos: the &struct rhash_head to use as a loop cursor.
438 * @head: the &struct rhash_head to start from
439 * @tbl: the &struct bucket_table
440 * @hash: the hash value / bucket index
441 */
442#define rht_for_each_from(pos, head, tbl, hash) \
443 for (pos = head; \
444 !rht_is_a_nulls(pos); \
445 pos = rht_dereference_bucket((pos)->next, tbl, hash))
446
447/**
448 * rht_for_each - iterate over hash chain
449 * @pos: the &struct rhash_head to use as a loop cursor.
450 * @tbl: the &struct bucket_table
451 * @hash: the hash value / bucket index
452 */
453#define rht_for_each(pos, tbl, hash) \
454 rht_for_each_from(pos, rht_ptr(rht_bucket(tbl, hash), tbl, hash), \
455 tbl, hash)
456
457/**
458 * rht_for_each_entry_from - iterate over hash chain from given head
459 * @tpos: the type * to use as a loop cursor.
460 * @pos: the &struct rhash_head to use as a loop cursor.
461 * @head: the &struct rhash_head to start from
462 * @tbl: the &struct bucket_table
463 * @hash: the hash value / bucket index
464 * @member: name of the &struct rhash_head within the hashable struct.
465 */
466#define rht_for_each_entry_from(tpos, pos, head, tbl, hash, member) \
467 for (pos = head; \
468 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
469 pos = rht_dereference_bucket((pos)->next, tbl, hash))
470
471/**
472 * rht_for_each_entry - iterate over hash chain of given type
473 * @tpos: the type * to use as a loop cursor.
474 * @pos: the &struct rhash_head to use as a loop cursor.
475 * @tbl: the &struct bucket_table
476 * @hash: the hash value / bucket index
477 * @member: name of the &struct rhash_head within the hashable struct.
478 */
479#define rht_for_each_entry(tpos, pos, tbl, hash, member) \
480 rht_for_each_entry_from(tpos, pos, \
481 rht_ptr(rht_bucket(tbl, hash), tbl, hash), \
482 tbl, hash, member)
483
484/**
485 * rht_for_each_entry_safe - safely iterate over hash chain of given type
486 * @tpos: the type * to use as a loop cursor.
487 * @pos: the &struct rhash_head to use as a loop cursor.
488 * @next: the &struct rhash_head to use as next in loop cursor.
489 * @tbl: the &struct bucket_table
490 * @hash: the hash value / bucket index
491 * @member: name of the &struct rhash_head within the hashable struct.
492 *
493 * This hash chain list-traversal primitive allows for the looped code to
494 * remove the loop cursor from the list.
495 */
496#define rht_for_each_entry_safe(tpos, pos, next, tbl, hash, member) \
497 for (pos = rht_ptr(rht_bucket(tbl, hash), tbl, hash), \
498 next = !rht_is_a_nulls(pos) ? \
499 rht_dereference_bucket(pos->next, tbl, hash) : NULL; \
500 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
501 pos = next, \
502 next = !rht_is_a_nulls(pos) ? \
503 rht_dereference_bucket(pos->next, tbl, hash) : NULL)
504
505/**
506 * rht_for_each_rcu_from - iterate over rcu hash chain from given head
507 * @pos: the &struct rhash_head to use as a loop cursor.
508 * @head: the &struct rhash_head to start from
509 * @tbl: the &struct bucket_table
510 * @hash: the hash value / bucket index
511 *
512 * This hash chain list-traversal primitive may safely run concurrently with
513 * the _rcu mutation primitives such as rhashtable_insert() as long as the
514 * traversal is guarded by rcu_read_lock().
515 */
516#define rht_for_each_rcu_from(pos, head, tbl, hash) \
517 for (({barrier(); }), \
518 pos = head; \
519 !rht_is_a_nulls(pos); \
520 pos = rcu_dereference_all(pos->next))
521
522/**
523 * rht_for_each_rcu - iterate over rcu hash chain
524 * @pos: the &struct rhash_head to use as a loop cursor.
525 * @tbl: the &struct bucket_table
526 * @hash: the hash value / bucket index
527 *
528 * This hash chain list-traversal primitive may safely run concurrently with
529 * the _rcu mutation primitives such as rhashtable_insert() as long as the
530 * traversal is guarded by rcu_read_lock().
531 */
532#define rht_for_each_rcu(pos, tbl, hash) \
533 for (({barrier(); }), \
534 pos = rht_ptr_rcu(rht_bucket(tbl, hash)); \
535 !rht_is_a_nulls(pos); \
536 pos = rcu_dereference_all(pos->next))
537
538/**
539 * rht_for_each_entry_rcu_from - iterated over rcu hash chain from given head
540 * @tpos: the type * to use as a loop cursor.
541 * @pos: the &struct rhash_head to use as a loop cursor.
542 * @head: the &struct rhash_head to start from
543 * @tbl: the &struct bucket_table
544 * @hash: the hash value / bucket index
545 * @member: name of the &struct rhash_head within the hashable struct.
546 *
547 * This hash chain list-traversal primitive may safely run concurrently with
548 * the _rcu mutation primitives such as rhashtable_insert() as long as the
549 * traversal is guarded by rcu_read_lock().
550 */
551#define rht_for_each_entry_rcu_from(tpos, pos, head, tbl, hash, member) \
552 for (({barrier(); }), \
553 pos = head; \
554 (!rht_is_a_nulls(pos)) && rht_entry(tpos, pos, member); \
555 pos = rht_dereference_bucket_rcu(pos->next, tbl, hash))
556
557/**
558 * rht_for_each_entry_rcu - iterate over rcu hash chain of given type
559 * @tpos: the type * to use as a loop cursor.
560 * @pos: the &struct rhash_head to use as a loop cursor.
561 * @tbl: the &struct bucket_table
562 * @hash: the hash value / bucket index
563 * @member: name of the &struct rhash_head within the hashable struct.
564 *
565 * This hash chain list-traversal primitive may safely run concurrently with
566 * the _rcu mutation primitives such as rhashtable_insert() as long as the
567 * traversal is guarded by rcu_read_lock().
568 */
569#define rht_for_each_entry_rcu(tpos, pos, tbl, hash, member) \
570 rht_for_each_entry_rcu_from(tpos, pos, \
571 rht_ptr_rcu(rht_bucket(tbl, hash)), \
572 tbl, hash, member)
573
574/**
575 * rhl_for_each_rcu - iterate over rcu hash table list
576 * @pos: the &struct rlist_head to use as a loop cursor.
577 * @list: the head of the list
578 *
579 * This hash chain list-traversal primitive should be used on the
580 * list returned by rhltable_lookup.
581 */
582#define rhl_for_each_rcu(pos, list) \
583 for (pos = list; pos; pos = rcu_dereference_all(pos->next))
584
585/**
586 * rhl_for_each_entry_rcu - iterate over rcu hash table list of given type
587 * @tpos: the type * to use as a loop cursor.
588 * @pos: the &struct rlist_head to use as a loop cursor.
589 * @list: the head of the list
590 * @member: name of the &struct rlist_head within the hashable struct.
591 *
592 * This hash chain list-traversal primitive should be used on the
593 * list returned by rhltable_lookup.
594 */
595#define rhl_for_each_entry_rcu(tpos, pos, list, member) \
596 for (pos = list; pos && rht_entry(tpos, pos, member); \
597 pos = rcu_dereference_all(pos->next))
598
599static inline int rhashtable_compare(struct rhashtable_compare_arg *arg,
600 const void *obj)
601{
602 struct rhashtable *ht = arg->ht;
603 const char *ptr = obj;
604
605 return memcmp(ptr + ht->p.key_offset, arg->key, ht->p.key_len);
606}
607
608/* Internal function, do not use. */
609static __always_inline struct rhash_head *__rhashtable_lookup(
610 struct rhashtable *ht, const void *key,
611 const struct rhashtable_params params,
612 const enum rht_lookup_freq freq)
613 __must_hold_shared(RCU)
614{
615 struct rhashtable_compare_arg arg = {
616 .ht = ht,
617 .key = key,
618 };
619 struct rhash_lock_head __rcu *const *bkt;
620 struct bucket_table *tbl;
621 struct rhash_head *he;
622 unsigned int hash;
623
624 BUILD_BUG_ON(!__builtin_constant_p(freq));
625 tbl = rht_dereference_rcu(ht->tbl, ht);
626restart:
627 hash = rht_key_hashfn(ht, tbl, key, params);
628 bkt = rht_bucket(tbl, hash);
629 do {
630 rht_for_each_rcu_from(he, __rht_ptr_rcu(bkt, freq), tbl, hash) {
631 if (params.obj_cmpfn ?
632 params.obj_cmpfn(&arg, rht_obj(ht, he)) :
633 rhashtable_compare(&arg, rht_obj(ht, he)))
634 continue;
635 return he;
636 }
637 /* An object might have been moved to a different hash chain,
638 * while we walk along it - better check and retry.
639 */
640 } while (he != RHT_NULLS_MARKER(bkt));
641
642 /* Ensure we see any new tables. */
643 smp_rmb();
644
645 tbl = rht_dereference_rcu(tbl->future_tbl, ht);
646 if (unlikely(tbl))
647 goto restart;
648
649 return NULL;
650}
651
652/**
653 * rhashtable_lookup - search hash table
654 * @ht: hash table
655 * @key: the pointer to the key
656 * @params: hash table parameters
657 *
658 * Computes the hash value for the key and traverses the bucket chain looking
659 * for an entry with an identical key. The first matching entry is returned.
660 *
661 * This must only be called under the RCU read lock.
662 *
663 * Returns the first entry on which the compare function returned true.
664 */
665static __always_inline void *rhashtable_lookup(
666 struct rhashtable *ht, const void *key,
667 const struct rhashtable_params params)
668 __must_hold_shared(RCU)
669{
670 struct rhash_head *he = __rhashtable_lookup(ht, key, params,
671 RHT_LOOKUP_NORMAL);
672
673 return he ? rht_obj(ht, he) : NULL;
674}
675
676static __always_inline void *rhashtable_lookup_likely(
677 struct rhashtable *ht, const void *key,
678 const struct rhashtable_params params)
679 __must_hold_shared(RCU)
680{
681 struct rhash_head *he = __rhashtable_lookup(ht, key, params,
682 RHT_LOOKUP_LIKELY);
683
684 return likely(he) ? rht_obj(ht, he) : NULL;
685}
686
687/**
688 * rhashtable_lookup_fast - search hash table, without RCU read lock
689 * @ht: hash table
690 * @key: the pointer to the key
691 * @params: hash table parameters
692 *
693 * Computes the hash value for the key and traverses the bucket chain looking
694 * for an entry with an identical key. The first matching entry is returned.
695 *
696 * Only use this function when you have other mechanisms guaranteeing
697 * that the object won't go away after the RCU read lock is released.
698 *
699 * Returns the first entry on which the compare function returned true.
700 */
701static __always_inline void *rhashtable_lookup_fast(
702 struct rhashtable *ht, const void *key,
703 const struct rhashtable_params params)
704{
705 void *obj;
706
707 rcu_read_lock();
708 obj = rhashtable_lookup(ht, key, params);
709 rcu_read_unlock();
710
711 return obj;
712}
713
714/**
715 * rhltable_lookup - search hash list table
716 * @hlt: hash table
717 * @key: the pointer to the key
718 * @params: hash table parameters
719 *
720 * Computes the hash value for the key and traverses the bucket chain looking
721 * for an entry with an identical key. All matching entries are returned
722 * in a list.
723 *
724 * This must only be called under the RCU read lock.
725 *
726 * Returns the list of entries that match the given key.
727 */
728static __always_inline struct rhlist_head *rhltable_lookup(
729 struct rhltable *hlt, const void *key,
730 const struct rhashtable_params params)
731 __must_hold_shared(RCU)
732{
733 struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params,
734 RHT_LOOKUP_NORMAL);
735
736 return he ? container_of(he, struct rhlist_head, rhead) : NULL;
737}
738
739static __always_inline struct rhlist_head *rhltable_lookup_likely(
740 struct rhltable *hlt, const void *key,
741 const struct rhashtable_params params)
742 __must_hold_shared(RCU)
743{
744 struct rhash_head *he = __rhashtable_lookup(&hlt->ht, key, params,
745 RHT_LOOKUP_LIKELY);
746
747 return likely(he) ? container_of(he, struct rhlist_head, rhead) : NULL;
748}
749
750/* Internal function, please use rhashtable_insert_fast() instead. This
751 * function returns the existing element already in hashes if there is a clash,
752 * otherwise it returns an error via ERR_PTR().
753 */
754static __always_inline void *__rhashtable_insert_fast(
755 struct rhashtable *ht, const void *key, struct rhash_head *obj,
756 const struct rhashtable_params params, bool rhlist)
757{
758 struct rhashtable_compare_arg arg = {
759 .ht = ht,
760 .key = key,
761 };
762 struct rhash_lock_head __rcu **bkt;
763 struct rhash_head __rcu **pprev;
764 struct bucket_table *tbl;
765 struct rhash_head *head;
766 unsigned long flags;
767 unsigned int hash;
768 int elasticity;
769 void *data;
770
771 rcu_read_lock();
772
773 tbl = rht_dereference_rcu(ht->tbl, ht);
774 hash = rht_head_hashfn(ht, tbl, obj, params);
775 elasticity = RHT_ELASTICITY;
776 bkt = rht_bucket_insert(ht, tbl, hash);
777 data = ERR_PTR(-ENOMEM);
778 if (!bkt)
779 goto out;
780 pprev = NULL;
781 flags = rht_lock(tbl, bkt);
782
783 if (unlikely(rcu_access_pointer(tbl->future_tbl))) {
784slow_path:
785 rht_unlock(tbl, bkt, flags);
786 rcu_read_unlock();
787 return rhashtable_insert_slow(ht, key, obj);
788 }
789
790 rht_for_each_from(head, rht_ptr(bkt, tbl, hash), tbl, hash) {
791 struct rhlist_head *plist;
792 struct rhlist_head *list;
793
794 elasticity--;
795 if (!key ||
796 (params.obj_cmpfn ?
797 params.obj_cmpfn(&arg, rht_obj(ht, head)) :
798 rhashtable_compare(&arg, rht_obj(ht, head)))) {
799 pprev = &head->next;
800 continue;
801 }
802
803 data = rht_obj(ht, head);
804
805 if (!rhlist)
806 goto out_unlock;
807
808
809 list = container_of(obj, struct rhlist_head, rhead);
810 plist = container_of(head, struct rhlist_head, rhead);
811
812 RCU_INIT_POINTER(list->next, plist);
813 head = rht_dereference_bucket(head->next, tbl, hash);
814 RCU_INIT_POINTER(list->rhead.next, head);
815 if (pprev) {
816 rcu_assign_pointer(*pprev, obj);
817 rht_unlock(tbl, bkt, flags);
818 } else
819 rht_assign_unlock(tbl, bkt, obj, flags);
820 data = NULL;
821 goto out;
822 }
823
824 if (elasticity <= 0)
825 goto slow_path;
826
827 data = ERR_PTR(-E2BIG);
828 if (unlikely(rht_grow_above_max(ht, tbl)))
829 goto out_unlock;
830
831 if (unlikely(rht_grow_above_100(ht, tbl)))
832 goto slow_path;
833
834 /* Inserting at head of list makes unlocking free. */
835 head = rht_ptr(bkt, tbl, hash);
836
837 RCU_INIT_POINTER(obj->next, head);
838 if (rhlist) {
839 struct rhlist_head *list;
840
841 list = container_of(obj, struct rhlist_head, rhead);
842 RCU_INIT_POINTER(list->next, NULL);
843 }
844
845 atomic_inc(&ht->nelems);
846 rht_assign_unlock(tbl, bkt, obj, flags);
847
848 if (rht_grow_above_75(ht, tbl))
849 schedule_work(&ht->run_work);
850
851 data = NULL;
852out:
853 rcu_read_unlock();
854
855 return data;
856
857out_unlock:
858 rht_unlock(tbl, bkt, flags);
859 goto out;
860}
861
862/**
863 * rhashtable_insert_fast - insert object into hash table
864 * @ht: hash table
865 * @obj: pointer to hash head inside object
866 * @params: hash table parameters
867 *
868 * Will take the per bucket bitlock to protect against mutual mutations
869 * on the same bucket. Multiple insertions may occur in parallel unless
870 * they map to the same bucket.
871 *
872 * It is safe to call this function from atomic context.
873 *
874 * Will trigger an automatic deferred table resizing if residency in the
875 * table grows beyond 70%.
876 */
877static __always_inline int rhashtable_insert_fast(
878 struct rhashtable *ht, struct rhash_head *obj,
879 const struct rhashtable_params params)
880{
881 void *ret;
882
883 ret = __rhashtable_insert_fast(ht, NULL, obj, params, false);
884 if (IS_ERR(ret))
885 return PTR_ERR(ret);
886
887 return ret == NULL ? 0 : -EEXIST;
888}
889
890/**
891 * rhltable_insert_key - insert object into hash list table
892 * @hlt: hash list table
893 * @key: the pointer to the key
894 * @list: pointer to hash list head inside object
895 * @params: hash table parameters
896 *
897 * Will take the per bucket bitlock to protect against mutual mutations
898 * on the same bucket. Multiple insertions may occur in parallel unless
899 * they map to the same bucket.
900 *
901 * It is safe to call this function from atomic context.
902 *
903 * Will trigger an automatic deferred table resizing if residency in the
904 * table grows beyond 70%.
905 */
906static __always_inline int rhltable_insert_key(
907 struct rhltable *hlt, const void *key, struct rhlist_head *list,
908 const struct rhashtable_params params)
909{
910 return PTR_ERR(__rhashtable_insert_fast(&hlt->ht, key, &list->rhead,
911 params, true));
912}
913
914/**
915 * rhltable_insert - insert object into hash list table
916 * @hlt: hash list table
917 * @list: pointer to hash list head inside object
918 * @params: hash table parameters
919 *
920 * Will take the per bucket bitlock to protect against mutual mutations
921 * on the same bucket. Multiple insertions may occur in parallel unless
922 * they map to the same bucket.
923 *
924 * It is safe to call this function from atomic context.
925 *
926 * Will trigger an automatic deferred table resizing if residency in the
927 * table grows beyond 70%.
928 */
929static __always_inline int rhltable_insert(
930 struct rhltable *hlt, struct rhlist_head *list,
931 const struct rhashtable_params params)
932{
933 const char *key = rht_obj(&hlt->ht, &list->rhead);
934
935 key += params.key_offset;
936
937 return rhltable_insert_key(hlt, key, list, params);
938}
939
940/**
941 * rhashtable_lookup_insert_fast - lookup and insert object into hash table
942 * @ht: hash table
943 * @obj: pointer to hash head inside object
944 * @params: hash table parameters
945 *
946 * This lookup function may only be used for fixed key hash table (key_len
947 * parameter set). It will BUG() if used inappropriately.
948 *
949 * It is safe to call this function from atomic context.
950 *
951 * Will trigger an automatic deferred table resizing if residency in the
952 * table grows beyond 70%.
953 */
954static __always_inline int rhashtable_lookup_insert_fast(
955 struct rhashtable *ht, struct rhash_head *obj,
956 const struct rhashtable_params params)
957{
958 const char *key = rht_obj(ht, obj);
959 void *ret;
960
961 BUG_ON(ht->p.obj_hashfn);
962
963 ret = __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
964 false);
965 if (IS_ERR(ret))
966 return PTR_ERR(ret);
967
968 return ret == NULL ? 0 : -EEXIST;
969}
970
971/**
972 * rhashtable_lookup_get_insert_fast - lookup and insert object into hash table
973 * @ht: hash table
974 * @obj: pointer to hash head inside object
975 * @params: hash table parameters
976 *
977 * Just like rhashtable_lookup_insert_fast(), but this function returns the
978 * object if it exists, NULL if it did not and the insertion was successful,
979 * and an ERR_PTR otherwise.
980 */
981static __always_inline void *rhashtable_lookup_get_insert_fast(
982 struct rhashtable *ht, struct rhash_head *obj,
983 const struct rhashtable_params params)
984{
985 const char *key = rht_obj(ht, obj);
986
987 BUG_ON(ht->p.obj_hashfn);
988
989 return __rhashtable_insert_fast(ht, key + ht->p.key_offset, obj, params,
990 false);
991}
992
993/**
994 * rhashtable_lookup_insert_key - search and insert object to hash table
995 * with explicit key
996 * @ht: hash table
997 * @key: key
998 * @obj: pointer to hash head inside object
999 * @params: hash table parameters
1000 *
1001 * Lookups may occur in parallel with hashtable mutations and resizing.
1002 *
1003 * Will trigger an automatic deferred table resizing if residency in the
1004 * table grows beyond 70%.
1005 *
1006 * Returns zero on success.
1007 */
1008static __always_inline int rhashtable_lookup_insert_key(
1009 struct rhashtable *ht, const void *key, struct rhash_head *obj,
1010 const struct rhashtable_params params)
1011{
1012 void *ret;
1013
1014 BUG_ON(!ht->p.obj_hashfn || !key);
1015
1016 ret = __rhashtable_insert_fast(ht, key, obj, params, false);
1017 if (IS_ERR(ret))
1018 return PTR_ERR(ret);
1019
1020 return ret == NULL ? 0 : -EEXIST;
1021}
1022
1023/**
1024 * rhashtable_lookup_get_insert_key - lookup and insert object into hash table
1025 * @ht: hash table
1026 * @key: key
1027 * @obj: pointer to hash head inside object
1028 * @params: hash table parameters
1029 *
1030 * Just like rhashtable_lookup_insert_key(), but this function returns the
1031 * object if it exists, NULL if it does not and the insertion was successful,
1032 * and an ERR_PTR otherwise.
1033 */
1034static __always_inline void *rhashtable_lookup_get_insert_key(
1035 struct rhashtable *ht, const void *key, struct rhash_head *obj,
1036 const struct rhashtable_params params)
1037{
1038 BUG_ON(!ht->p.obj_hashfn || !key);
1039
1040 return __rhashtable_insert_fast(ht, key, obj, params, false);
1041}
1042
1043/* Internal function, please use rhashtable_remove_fast() instead */
1044static __always_inline int __rhashtable_remove_fast_one(
1045 struct rhashtable *ht, struct bucket_table *tbl,
1046 struct rhash_head *obj, const struct rhashtable_params params,
1047 bool rhlist)
1048{
1049 struct rhash_lock_head __rcu **bkt;
1050 struct rhash_head __rcu **pprev;
1051 struct rhash_head *he;
1052 unsigned long flags;
1053 unsigned int hash;
1054 int err = -ENOENT;
1055
1056 hash = rht_head_hashfn(ht, tbl, obj, params);
1057 bkt = rht_bucket_var(tbl, hash);
1058 if (!bkt)
1059 return -ENOENT;
1060 pprev = NULL;
1061 flags = rht_lock(tbl, bkt);
1062
1063 rht_for_each_from(he, rht_ptr(bkt, tbl, hash), tbl, hash) {
1064 struct rhlist_head *list;
1065
1066 list = container_of(he, struct rhlist_head, rhead);
1067
1068 if (he != obj) {
1069 struct rhlist_head __rcu **lpprev;
1070
1071 pprev = &he->next;
1072
1073 if (!rhlist)
1074 continue;
1075
1076 do {
1077 lpprev = &list->next;
1078 list = rht_dereference_bucket(list->next,
1079 tbl, hash);
1080 } while (list && obj != &list->rhead);
1081
1082 if (!list)
1083 continue;
1084
1085 list = rht_dereference_bucket(list->next, tbl, hash);
1086 RCU_INIT_POINTER(*lpprev, list);
1087 err = 0;
1088 break;
1089 }
1090
1091 obj = rht_dereference_bucket(obj->next, tbl, hash);
1092 err = 1;
1093
1094 if (rhlist) {
1095 list = rht_dereference_bucket(list->next, tbl, hash);
1096 if (list) {
1097 RCU_INIT_POINTER(list->rhead.next, obj);
1098 obj = &list->rhead;
1099 err = 0;
1100 }
1101 }
1102
1103 if (pprev) {
1104 rcu_assign_pointer(*pprev, obj);
1105 rht_unlock(tbl, bkt, flags);
1106 } else {
1107 rht_assign_unlock(tbl, bkt, obj, flags);
1108 }
1109 goto unlocked;
1110 }
1111
1112 rht_unlock(tbl, bkt, flags);
1113unlocked:
1114 if (err > 0) {
1115 atomic_dec(&ht->nelems);
1116 if (unlikely(ht->p.automatic_shrinking &&
1117 rht_shrink_below_30(ht, tbl)))
1118 schedule_work(&ht->run_work);
1119 err = 0;
1120 }
1121
1122 return err;
1123}
1124
1125/* Internal function, please use rhashtable_remove_fast() instead */
1126static __always_inline int __rhashtable_remove_fast(
1127 struct rhashtable *ht, struct rhash_head *obj,
1128 const struct rhashtable_params params, bool rhlist)
1129{
1130 struct bucket_table *tbl;
1131 int err;
1132
1133 rcu_read_lock();
1134
1135 tbl = rht_dereference_rcu(ht->tbl, ht);
1136
1137 /* Because we have already taken (and released) the bucket
1138 * lock in old_tbl, if we find that future_tbl is not yet
1139 * visible then that guarantees the entry to still be in
1140 * the old tbl if it exists.
1141 */
1142 while ((err = __rhashtable_remove_fast_one(ht, tbl, obj, params,
1143 rhlist)) &&
1144 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1145 ;
1146
1147 rcu_read_unlock();
1148
1149 return err;
1150}
1151
1152/**
1153 * rhashtable_remove_fast - remove object from hash table
1154 * @ht: hash table
1155 * @obj: pointer to hash head inside object
1156 * @params: hash table parameters
1157 *
1158 * Since the hash chain is single linked, the removal operation needs to
1159 * walk the bucket chain upon removal. The removal operation is thus
1160 * considerable slow if the hash table is not correctly sized.
1161 *
1162 * Will automatically shrink the table if permitted when residency drops
1163 * below 30%.
1164 *
1165 * Returns zero on success, -ENOENT if the entry could not be found.
1166 */
1167static __always_inline int rhashtable_remove_fast(
1168 struct rhashtable *ht, struct rhash_head *obj,
1169 const struct rhashtable_params params)
1170{
1171 return __rhashtable_remove_fast(ht, obj, params, false);
1172}
1173
1174/**
1175 * rhltable_remove - remove object from hash list table
1176 * @hlt: hash list table
1177 * @list: pointer to hash list head inside object
1178 * @params: hash table parameters
1179 *
1180 * Since the hash chain is single linked, the removal operation needs to
1181 * walk the bucket chain upon removal. The removal operation is thus
1182 * considerably slower if the hash table is not correctly sized.
1183 *
1184 * Will automatically shrink the table if permitted when residency drops
1185 * below 30%
1186 *
1187 * Returns zero on success, -ENOENT if the entry could not be found.
1188 */
1189static __always_inline int rhltable_remove(
1190 struct rhltable *hlt, struct rhlist_head *list,
1191 const struct rhashtable_params params)
1192{
1193 return __rhashtable_remove_fast(&hlt->ht, &list->rhead, params, true);
1194}
1195
1196/* Internal function, please use rhashtable_replace_fast() instead */
1197static __always_inline int __rhashtable_replace_fast(
1198 struct rhashtable *ht, struct bucket_table *tbl,
1199 struct rhash_head *obj_old, struct rhash_head *obj_new,
1200 const struct rhashtable_params params)
1201{
1202 struct rhash_lock_head __rcu **bkt;
1203 struct rhash_head __rcu **pprev;
1204 struct rhash_head *he;
1205 unsigned long flags;
1206 unsigned int hash;
1207 int err = -ENOENT;
1208
1209 /* Minimally, the old and new objects must have same hash
1210 * (which should mean identifiers are the same).
1211 */
1212 hash = rht_head_hashfn(ht, tbl, obj_old, params);
1213 if (hash != rht_head_hashfn(ht, tbl, obj_new, params))
1214 return -EINVAL;
1215
1216 bkt = rht_bucket_var(tbl, hash);
1217 if (!bkt)
1218 return -ENOENT;
1219
1220 pprev = NULL;
1221 flags = rht_lock(tbl, bkt);
1222
1223 rht_for_each_from(he, rht_ptr(bkt, tbl, hash), tbl, hash) {
1224 if (he != obj_old) {
1225 pprev = &he->next;
1226 continue;
1227 }
1228
1229 rcu_assign_pointer(obj_new->next, obj_old->next);
1230 if (pprev) {
1231 rcu_assign_pointer(*pprev, obj_new);
1232 rht_unlock(tbl, bkt, flags);
1233 } else {
1234 rht_assign_unlock(tbl, bkt, obj_new, flags);
1235 }
1236 err = 0;
1237 goto unlocked;
1238 }
1239
1240 rht_unlock(tbl, bkt, flags);
1241
1242unlocked:
1243 return err;
1244}
1245
1246/**
1247 * rhashtable_replace_fast - replace an object in hash table
1248 * @ht: hash table
1249 * @obj_old: pointer to hash head inside object being replaced
1250 * @obj_new: pointer to hash head inside object which is new
1251 * @params: hash table parameters
1252 *
1253 * Replacing an object doesn't affect the number of elements in the hash table
1254 * or bucket, so we don't need to worry about shrinking or expanding the
1255 * table here.
1256 *
1257 * Returns zero on success, -ENOENT if the entry could not be found,
1258 * -EINVAL if hash is not the same for the old and new objects.
1259 */
1260static __always_inline int rhashtable_replace_fast(
1261 struct rhashtable *ht, struct rhash_head *obj_old,
1262 struct rhash_head *obj_new,
1263 const struct rhashtable_params params)
1264{
1265 struct bucket_table *tbl;
1266 int err;
1267
1268 rcu_read_lock();
1269
1270 tbl = rht_dereference_rcu(ht->tbl, ht);
1271
1272 /* Because we have already taken (and released) the bucket
1273 * lock in old_tbl, if we find that future_tbl is not yet
1274 * visible then that guarantees the entry to still be in
1275 * the old tbl if it exists.
1276 */
1277 while ((err = __rhashtable_replace_fast(ht, tbl, obj_old,
1278 obj_new, params)) &&
1279 (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
1280 ;
1281
1282 rcu_read_unlock();
1283
1284 return err;
1285}
1286
1287/**
1288 * rhltable_walk_enter - Initialise an iterator
1289 * @hlt: Table to walk over
1290 * @iter: Hash table Iterator
1291 *
1292 * This function prepares a hash table walk.
1293 *
1294 * Note that if you restart a walk after rhashtable_walk_stop you
1295 * may see the same object twice. Also, you may miss objects if
1296 * there are removals in between rhashtable_walk_stop and the next
1297 * call to rhashtable_walk_start.
1298 *
1299 * For a completely stable walk you should construct your own data
1300 * structure outside the hash table.
1301 *
1302 * This function may be called from any process context, including
1303 * non-preemptable context, but cannot be called from softirq or
1304 * hardirq context.
1305 *
1306 * You must call rhashtable_walk_exit after this function returns.
1307 */
1308static inline void rhltable_walk_enter(struct rhltable *hlt,
1309 struct rhashtable_iter *iter)
1310{
1311 rhashtable_walk_enter(&hlt->ht, iter);
1312}
1313
1314/**
1315 * rhltable_free_and_destroy - free elements and destroy hash list table
1316 * @hlt: the hash list table to destroy
1317 * @free_fn: callback to release resources of element
1318 * @arg: pointer passed to free_fn
1319 *
1320 * See documentation for rhashtable_free_and_destroy.
1321 */
1322static inline void rhltable_free_and_destroy(struct rhltable *hlt,
1323 void (*free_fn)(void *ptr,
1324 void *arg),
1325 void *arg)
1326{
1327 rhashtable_free_and_destroy(&hlt->ht, free_fn, arg);
1328}
1329
1330static inline void rhltable_destroy(struct rhltable *hlt)
1331{
1332 rhltable_free_and_destroy(hlt, NULL, NULL);
1333}
1334
1335#endif /* _LINUX_RHASHTABLE_H */