1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
5 * Development of this code funded by Astaro AG (http://www.astaro.com/)
8 #include <linux/kernel.h>
9 #include <linux/init.h>
10 #include <linux/module.h>
11 #include <linux/list.h>
12 #include <linux/log2.h>
13 #include <linux/jhash.h>
14 #include <linux/netlink.h>
15 #include <linux/workqueue.h>
16 #include <linux/rhashtable.h>
17 #include <linux/netfilter.h>
18 #include <linux/netfilter/nf_tables.h>
19 #include <net/netfilter/nf_tables_core.h>
21 /* We target a hash table size of 4, element hint is 75% of final size */
22 #define NFT_RHASH_ELEMENT_HINT 3
26 struct delayed_work gc_work;
29 struct nft_rhash_elem {
30 struct rhash_head node;
31 struct nft_set_ext ext;
34 struct nft_rhash_cmp_arg {
35 const struct nft_set *set;
40 static inline u32 nft_rhash_key(const void *data, u32 len, u32 seed)
42 const struct nft_rhash_cmp_arg *arg = data;
44 return jhash(arg->key, len, seed);
47 static inline u32 nft_rhash_obj(const void *data, u32 len, u32 seed)
49 const struct nft_rhash_elem *he = data;
51 return jhash(nft_set_ext_key(&he->ext), len, seed);
54 static inline int nft_rhash_cmp(struct rhashtable_compare_arg *arg,
57 const struct nft_rhash_cmp_arg *x = arg->key;
58 const struct nft_rhash_elem *he = ptr;
60 if (memcmp(nft_set_ext_key(&he->ext), x->key, x->set->klen))
62 if (nft_set_elem_expired(&he->ext))
64 if (!nft_set_elem_active(&he->ext, x->genmask))
69 static const struct rhashtable_params nft_rhash_params = {
70 .head_offset = offsetof(struct nft_rhash_elem, node),
71 .hashfn = nft_rhash_key,
72 .obj_hashfn = nft_rhash_obj,
73 .obj_cmpfn = nft_rhash_cmp,
74 .automatic_shrinking = true,
77 static bool nft_rhash_lookup(const struct net *net, const struct nft_set *set,
78 const u32 *key, const struct nft_set_ext **ext)
80 struct nft_rhash *priv = nft_set_priv(set);
81 const struct nft_rhash_elem *he;
82 struct nft_rhash_cmp_arg arg = {
83 .genmask = nft_genmask_cur(net),
88 he = rhashtable_lookup(&priv->ht, &arg, nft_rhash_params);
95 static void *nft_rhash_get(const struct net *net, const struct nft_set *set,
96 const struct nft_set_elem *elem, unsigned int flags)
98 struct nft_rhash *priv = nft_set_priv(set);
99 struct nft_rhash_elem *he;
100 struct nft_rhash_cmp_arg arg = {
101 .genmask = nft_genmask_cur(net),
103 .key = elem->key.val.data,
106 he = rhashtable_lookup(&priv->ht, &arg, nft_rhash_params);
110 return ERR_PTR(-ENOENT);
113 static bool nft_rhash_update(struct nft_set *set, const u32 *key,
114 void *(*new)(struct nft_set *,
115 const struct nft_expr *,
116 struct nft_regs *regs),
117 const struct nft_expr *expr,
118 struct nft_regs *regs,
119 const struct nft_set_ext **ext)
121 struct nft_rhash *priv = nft_set_priv(set);
122 struct nft_rhash_elem *he, *prev;
123 struct nft_rhash_cmp_arg arg = {
124 .genmask = NFT_GENMASK_ANY,
129 he = rhashtable_lookup(&priv->ht, &arg, nft_rhash_params);
133 he = new(set, expr, regs);
137 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
142 /* Another cpu may race to insert the element with the same key */
144 nft_set_elem_destroy(set, he, true);
153 nft_set_elem_destroy(set, he, true);
158 static int nft_rhash_insert(const struct net *net, const struct nft_set *set,
159 const struct nft_set_elem *elem,
160 struct nft_set_ext **ext)
162 struct nft_rhash *priv = nft_set_priv(set);
163 struct nft_rhash_elem *he = elem->priv;
164 struct nft_rhash_cmp_arg arg = {
165 .genmask = nft_genmask_next(net),
167 .key = elem->key.val.data,
169 struct nft_rhash_elem *prev;
171 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
174 return PTR_ERR(prev);
182 static void nft_rhash_activate(const struct net *net, const struct nft_set *set,
183 const struct nft_set_elem *elem)
185 struct nft_rhash_elem *he = elem->priv;
187 nft_set_elem_change_active(net, set, &he->ext);
188 nft_set_elem_clear_busy(&he->ext);
191 static bool nft_rhash_flush(const struct net *net,
192 const struct nft_set *set, void *priv)
194 struct nft_rhash_elem *he = priv;
196 if (!nft_set_elem_mark_busy(&he->ext) ||
197 !nft_is_active(net, &he->ext)) {
198 nft_set_elem_change_active(net, set, &he->ext);
204 static void *nft_rhash_deactivate(const struct net *net,
205 const struct nft_set *set,
206 const struct nft_set_elem *elem)
208 struct nft_rhash *priv = nft_set_priv(set);
209 struct nft_rhash_elem *he;
210 struct nft_rhash_cmp_arg arg = {
211 .genmask = nft_genmask_next(net),
213 .key = elem->key.val.data,
217 he = rhashtable_lookup(&priv->ht, &arg, nft_rhash_params);
219 !nft_rhash_flush(net, set, he))
227 static void nft_rhash_remove(const struct net *net,
228 const struct nft_set *set,
229 const struct nft_set_elem *elem)
231 struct nft_rhash *priv = nft_set_priv(set);
232 struct nft_rhash_elem *he = elem->priv;
234 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
237 static bool nft_rhash_delete(const struct nft_set *set,
240 struct nft_rhash *priv = nft_set_priv(set);
241 struct nft_rhash_cmp_arg arg = {
242 .genmask = NFT_GENMASK_ANY,
246 struct nft_rhash_elem *he;
248 he = rhashtable_lookup(&priv->ht, &arg, nft_rhash_params);
252 return rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params) == 0;
255 static void nft_rhash_walk(const struct nft_ctx *ctx, struct nft_set *set,
256 struct nft_set_iter *iter)
258 struct nft_rhash *priv = nft_set_priv(set);
259 struct nft_rhash_elem *he;
260 struct rhashtable_iter hti;
261 struct nft_set_elem elem;
263 rhashtable_walk_enter(&priv->ht, &hti);
264 rhashtable_walk_start(&hti);
266 while ((he = rhashtable_walk_next(&hti))) {
268 if (PTR_ERR(he) != -EAGAIN) {
269 iter->err = PTR_ERR(he);
276 if (iter->count < iter->skip)
278 if (nft_set_elem_expired(&he->ext))
280 if (!nft_set_elem_active(&he->ext, iter->genmask))
285 iter->err = iter->fn(ctx, set, iter, &elem);
292 rhashtable_walk_stop(&hti);
293 rhashtable_walk_exit(&hti);
296 static bool nft_rhash_expr_needs_gc_run(const struct nft_set *set,
297 struct nft_set_ext *ext)
299 struct nft_set_elem_expr *elem_expr = nft_set_ext_expr(ext);
300 struct nft_expr *expr;
303 nft_setelem_expr_foreach(expr, elem_expr, size) {
305 expr->ops->gc(read_pnet(&set->net), expr))
312 static void nft_rhash_gc(struct work_struct *work)
315 struct nft_rhash_elem *he;
316 struct nft_rhash *priv;
317 struct nft_set_gc_batch *gcb = NULL;
318 struct rhashtable_iter hti;
320 priv = container_of(work, struct nft_rhash, gc_work.work);
321 set = nft_set_container_of(priv);
323 rhashtable_walk_enter(&priv->ht, &hti);
324 rhashtable_walk_start(&hti);
326 while ((he = rhashtable_walk_next(&hti))) {
328 if (PTR_ERR(he) != -EAGAIN)
333 if (nft_set_ext_exists(&he->ext, NFT_SET_EXT_EXPRESSIONS) &&
334 nft_rhash_expr_needs_gc_run(set, &he->ext))
337 if (!nft_set_elem_expired(&he->ext))
340 if (nft_set_elem_mark_busy(&he->ext))
343 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
346 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
347 atomic_dec(&set->nelems);
348 nft_set_gc_batch_add(gcb, he);
350 rhashtable_walk_stop(&hti);
351 rhashtable_walk_exit(&hti);
353 nft_set_gc_batch_complete(gcb);
354 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
355 nft_set_gc_interval(set));
358 static u64 nft_rhash_privsize(const struct nlattr * const nla[],
359 const struct nft_set_desc *desc)
361 return sizeof(struct nft_rhash);
364 static void nft_rhash_gc_init(const struct nft_set *set)
366 struct nft_rhash *priv = nft_set_priv(set);
368 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
369 nft_set_gc_interval(set));
372 static int nft_rhash_init(const struct nft_set *set,
373 const struct nft_set_desc *desc,
374 const struct nlattr * const tb[])
376 struct nft_rhash *priv = nft_set_priv(set);
377 struct rhashtable_params params = nft_rhash_params;
380 params.nelem_hint = desc->size ?: NFT_RHASH_ELEMENT_HINT;
381 params.key_len = set->klen;
383 err = rhashtable_init(&priv->ht, ¶ms);
387 INIT_DEFERRABLE_WORK(&priv->gc_work, nft_rhash_gc);
388 if (set->flags & NFT_SET_TIMEOUT)
389 nft_rhash_gc_init(set);
394 static void nft_rhash_elem_destroy(void *ptr, void *arg)
396 nft_set_elem_destroy(arg, ptr, true);
399 static void nft_rhash_destroy(const struct nft_set *set)
401 struct nft_rhash *priv = nft_set_priv(set);
403 cancel_delayed_work_sync(&priv->gc_work);
405 rhashtable_free_and_destroy(&priv->ht, nft_rhash_elem_destroy,
409 static u32 nft_hash_buckets(u32 size)
411 return roundup_pow_of_two(size * 4 / 3);
414 static bool nft_rhash_estimate(const struct nft_set_desc *desc, u32 features,
415 struct nft_set_estimate *est)
418 est->lookup = NFT_SET_CLASS_O_1;
419 est->space = NFT_SET_CLASS_O_N;
427 struct hlist_head table[];
430 struct nft_hash_elem {
431 struct hlist_node node;
432 struct nft_set_ext ext;
435 static bool nft_hash_lookup(const struct net *net, const struct nft_set *set,
436 const u32 *key, const struct nft_set_ext **ext)
438 struct nft_hash *priv = nft_set_priv(set);
439 u8 genmask = nft_genmask_cur(net);
440 const struct nft_hash_elem *he;
443 hash = jhash(key, set->klen, priv->seed);
444 hash = reciprocal_scale(hash, priv->buckets);
445 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
446 if (!memcmp(nft_set_ext_key(&he->ext), key, set->klen) &&
447 nft_set_elem_active(&he->ext, genmask)) {
455 static void *nft_hash_get(const struct net *net, const struct nft_set *set,
456 const struct nft_set_elem *elem, unsigned int flags)
458 struct nft_hash *priv = nft_set_priv(set);
459 u8 genmask = nft_genmask_cur(net);
460 struct nft_hash_elem *he;
463 hash = jhash(elem->key.val.data, set->klen, priv->seed);
464 hash = reciprocal_scale(hash, priv->buckets);
465 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
466 if (!memcmp(nft_set_ext_key(&he->ext), elem->key.val.data, set->klen) &&
467 nft_set_elem_active(&he->ext, genmask))
470 return ERR_PTR(-ENOENT);
473 static bool nft_hash_lookup_fast(const struct net *net,
474 const struct nft_set *set,
475 const u32 *key, const struct nft_set_ext **ext)
477 struct nft_hash *priv = nft_set_priv(set);
478 u8 genmask = nft_genmask_cur(net);
479 const struct nft_hash_elem *he;
483 hash = jhash_1word(k1, priv->seed);
484 hash = reciprocal_scale(hash, priv->buckets);
485 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
486 k2 = *(u32 *)nft_set_ext_key(&he->ext)->data;
488 nft_set_elem_active(&he->ext, genmask)) {
496 static u32 nft_jhash(const struct nft_set *set, const struct nft_hash *priv,
497 const struct nft_set_ext *ext)
499 const struct nft_data *key = nft_set_ext_key(ext);
502 if (set->klen == 4) {
504 hash = jhash_1word(k1, priv->seed);
506 hash = jhash(key, set->klen, priv->seed);
508 hash = reciprocal_scale(hash, priv->buckets);
513 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
514 const struct nft_set_elem *elem,
515 struct nft_set_ext **ext)
517 struct nft_hash_elem *this = elem->priv, *he;
518 struct nft_hash *priv = nft_set_priv(set);
519 u8 genmask = nft_genmask_next(net);
522 hash = nft_jhash(set, priv, &this->ext);
523 hlist_for_each_entry(he, &priv->table[hash], node) {
524 if (!memcmp(nft_set_ext_key(&this->ext),
525 nft_set_ext_key(&he->ext), set->klen) &&
526 nft_set_elem_active(&he->ext, genmask)) {
531 hlist_add_head_rcu(&this->node, &priv->table[hash]);
535 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
536 const struct nft_set_elem *elem)
538 struct nft_hash_elem *he = elem->priv;
540 nft_set_elem_change_active(net, set, &he->ext);
543 static bool nft_hash_flush(const struct net *net,
544 const struct nft_set *set, void *priv)
546 struct nft_hash_elem *he = priv;
548 nft_set_elem_change_active(net, set, &he->ext);
552 static void *nft_hash_deactivate(const struct net *net,
553 const struct nft_set *set,
554 const struct nft_set_elem *elem)
556 struct nft_hash *priv = nft_set_priv(set);
557 struct nft_hash_elem *this = elem->priv, *he;
558 u8 genmask = nft_genmask_next(net);
561 hash = nft_jhash(set, priv, &this->ext);
562 hlist_for_each_entry(he, &priv->table[hash], node) {
563 if (!memcmp(nft_set_ext_key(&he->ext), &elem->key.val,
565 nft_set_elem_active(&he->ext, genmask)) {
566 nft_set_elem_change_active(net, set, &he->ext);
573 static void nft_hash_remove(const struct net *net,
574 const struct nft_set *set,
575 const struct nft_set_elem *elem)
577 struct nft_hash_elem *he = elem->priv;
579 hlist_del_rcu(&he->node);
582 static void nft_hash_walk(const struct nft_ctx *ctx, struct nft_set *set,
583 struct nft_set_iter *iter)
585 struct nft_hash *priv = nft_set_priv(set);
586 struct nft_hash_elem *he;
587 struct nft_set_elem elem;
590 for (i = 0; i < priv->buckets; i++) {
591 hlist_for_each_entry_rcu(he, &priv->table[i], node) {
592 if (iter->count < iter->skip)
594 if (!nft_set_elem_active(&he->ext, iter->genmask))
599 iter->err = iter->fn(ctx, set, iter, &elem);
608 static u64 nft_hash_privsize(const struct nlattr * const nla[],
609 const struct nft_set_desc *desc)
611 return sizeof(struct nft_hash) +
612 nft_hash_buckets(desc->size) * sizeof(struct hlist_head);
615 static int nft_hash_init(const struct nft_set *set,
616 const struct nft_set_desc *desc,
617 const struct nlattr * const tb[])
619 struct nft_hash *priv = nft_set_priv(set);
621 priv->buckets = nft_hash_buckets(desc->size);
622 get_random_bytes(&priv->seed, sizeof(priv->seed));
627 static void nft_hash_destroy(const struct nft_set *set)
629 struct nft_hash *priv = nft_set_priv(set);
630 struct nft_hash_elem *he;
631 struct hlist_node *next;
634 for (i = 0; i < priv->buckets; i++) {
635 hlist_for_each_entry_safe(he, next, &priv->table[i], node) {
636 hlist_del_rcu(&he->node);
637 nft_set_elem_destroy(set, he, true);
642 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
643 struct nft_set_estimate *est)
651 est->size = sizeof(struct nft_hash) +
652 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
653 desc->size * sizeof(struct nft_hash_elem);
654 est->lookup = NFT_SET_CLASS_O_1;
655 est->space = NFT_SET_CLASS_O_N;
660 static bool nft_hash_fast_estimate(const struct nft_set_desc *desc, u32 features,
661 struct nft_set_estimate *est)
669 est->size = sizeof(struct nft_hash) +
670 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
671 desc->size * sizeof(struct nft_hash_elem);
672 est->lookup = NFT_SET_CLASS_O_1;
673 est->space = NFT_SET_CLASS_O_N;
678 const struct nft_set_type nft_set_rhash_type = {
679 .features = NFT_SET_MAP | NFT_SET_OBJECT |
680 NFT_SET_TIMEOUT | NFT_SET_EVAL,
682 .privsize = nft_rhash_privsize,
683 .elemsize = offsetof(struct nft_rhash_elem, ext),
684 .estimate = nft_rhash_estimate,
685 .init = nft_rhash_init,
686 .gc_init = nft_rhash_gc_init,
687 .destroy = nft_rhash_destroy,
688 .insert = nft_rhash_insert,
689 .activate = nft_rhash_activate,
690 .deactivate = nft_rhash_deactivate,
691 .flush = nft_rhash_flush,
692 .remove = nft_rhash_remove,
693 .lookup = nft_rhash_lookup,
694 .update = nft_rhash_update,
695 .delete = nft_rhash_delete,
696 .walk = nft_rhash_walk,
697 .get = nft_rhash_get,
701 const struct nft_set_type nft_set_hash_type = {
702 .features = NFT_SET_MAP | NFT_SET_OBJECT,
704 .privsize = nft_hash_privsize,
705 .elemsize = offsetof(struct nft_hash_elem, ext),
706 .estimate = nft_hash_estimate,
707 .init = nft_hash_init,
708 .destroy = nft_hash_destroy,
709 .insert = nft_hash_insert,
710 .activate = nft_hash_activate,
711 .deactivate = nft_hash_deactivate,
712 .flush = nft_hash_flush,
713 .remove = nft_hash_remove,
714 .lookup = nft_hash_lookup,
715 .walk = nft_hash_walk,
720 const struct nft_set_type nft_set_hash_fast_type = {
721 .features = NFT_SET_MAP | NFT_SET_OBJECT,
723 .privsize = nft_hash_privsize,
724 .elemsize = offsetof(struct nft_hash_elem, ext),
725 .estimate = nft_hash_fast_estimate,
726 .init = nft_hash_init,
727 .destroy = nft_hash_destroy,
728 .insert = nft_hash_insert,
729 .activate = nft_hash_activate,
730 .deactivate = nft_hash_deactivate,
731 .flush = nft_hash_flush,
732 .remove = nft_hash_remove,
733 .lookup = nft_hash_lookup_fast,
734 .walk = nft_hash_walk,