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