2 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * Development of this code funded by Astaro AG (http://www.astaro.com/)
11 #include <linux/kernel.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/list.h>
15 #include <linux/log2.h>
16 #include <linux/jhash.h>
17 #include <linux/netlink.h>
18 #include <linux/workqueue.h>
19 #include <linux/rhashtable.h>
20 #include <linux/netfilter.h>
21 #include <linux/netfilter/nf_tables.h>
22 #include <net/netfilter/nf_tables.h>
24 /* We target a hash table size of 4, element hint is 75% of final size */
25 #define NFT_RHASH_ELEMENT_HINT 3
29 struct delayed_work gc_work;
32 struct nft_rhash_elem {
33 struct rhash_head node;
34 struct nft_set_ext ext;
37 struct nft_rhash_cmp_arg {
38 const struct nft_set *set;
43 static inline u32 nft_rhash_key(const void *data, u32 len, u32 seed)
45 const struct nft_rhash_cmp_arg *arg = data;
47 return jhash(arg->key, len, seed);
50 static inline u32 nft_rhash_obj(const void *data, u32 len, u32 seed)
52 const struct nft_rhash_elem *he = data;
54 return jhash(nft_set_ext_key(&he->ext), len, seed);
57 static inline int nft_rhash_cmp(struct rhashtable_compare_arg *arg,
60 const struct nft_rhash_cmp_arg *x = arg->key;
61 const struct nft_rhash_elem *he = ptr;
63 if (memcmp(nft_set_ext_key(&he->ext), x->key, x->set->klen))
65 if (nft_set_elem_expired(&he->ext))
67 if (!nft_set_elem_active(&he->ext, x->genmask))
72 static const struct rhashtable_params nft_rhash_params = {
73 .head_offset = offsetof(struct nft_rhash_elem, node),
74 .hashfn = nft_rhash_key,
75 .obj_hashfn = nft_rhash_obj,
76 .obj_cmpfn = nft_rhash_cmp,
77 .automatic_shrinking = true,
80 static bool nft_rhash_lookup(const struct net *net, const struct nft_set *set,
81 const u32 *key, const struct nft_set_ext **ext)
83 struct nft_rhash *priv = nft_set_priv(set);
84 const struct nft_rhash_elem *he;
85 struct nft_rhash_cmp_arg arg = {
86 .genmask = nft_genmask_cur(net),
91 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_rhash_params);
98 static bool nft_rhash_update(struct nft_set *set, const u32 *key,
99 void *(*new)(struct nft_set *,
100 const struct nft_expr *,
101 struct nft_regs *regs),
102 const struct nft_expr *expr,
103 struct nft_regs *regs,
104 const struct nft_set_ext **ext)
106 struct nft_rhash *priv = nft_set_priv(set);
107 struct nft_rhash_elem *he, *prev;
108 struct nft_rhash_cmp_arg arg = {
109 .genmask = NFT_GENMASK_ANY,
114 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_rhash_params);
118 he = new(set, expr, regs);
122 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
127 /* Another cpu may race to insert the element with the same key */
129 nft_set_elem_destroy(set, he, true);
138 nft_set_elem_destroy(set, he, true);
143 static int nft_rhash_insert(const struct net *net, const struct nft_set *set,
144 const struct nft_set_elem *elem,
145 struct nft_set_ext **ext)
147 struct nft_rhash *priv = nft_set_priv(set);
148 struct nft_rhash_elem *he = elem->priv;
149 struct nft_rhash_cmp_arg arg = {
150 .genmask = nft_genmask_next(net),
152 .key = elem->key.val.data,
154 struct nft_rhash_elem *prev;
156 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
159 return PTR_ERR(prev);
167 static void nft_rhash_activate(const struct net *net, const struct nft_set *set,
168 const struct nft_set_elem *elem)
170 struct nft_rhash_elem *he = elem->priv;
172 nft_set_elem_change_active(net, set, &he->ext);
173 nft_set_elem_clear_busy(&he->ext);
176 static bool nft_rhash_flush(const struct net *net,
177 const struct nft_set *set, void *priv)
179 struct nft_rhash_elem *he = priv;
181 if (!nft_set_elem_mark_busy(&he->ext) ||
182 !nft_is_active(net, &he->ext)) {
183 nft_set_elem_change_active(net, set, &he->ext);
189 static void *nft_rhash_deactivate(const struct net *net,
190 const struct nft_set *set,
191 const struct nft_set_elem *elem)
193 struct nft_rhash *priv = nft_set_priv(set);
194 struct nft_rhash_elem *he;
195 struct nft_rhash_cmp_arg arg = {
196 .genmask = nft_genmask_next(net),
198 .key = elem->key.val.data,
202 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_rhash_params);
204 !nft_rhash_flush(net, set, he))
212 static void nft_rhash_remove(const struct net *net,
213 const struct nft_set *set,
214 const struct nft_set_elem *elem)
216 struct nft_rhash *priv = nft_set_priv(set);
217 struct nft_rhash_elem *he = elem->priv;
219 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
222 static void nft_rhash_walk(const struct nft_ctx *ctx, struct nft_set *set,
223 struct nft_set_iter *iter)
225 struct nft_rhash *priv = nft_set_priv(set);
226 struct nft_rhash_elem *he;
227 struct rhashtable_iter hti;
228 struct nft_set_elem elem;
231 err = rhashtable_walk_init(&priv->ht, &hti, GFP_ATOMIC);
236 err = rhashtable_walk_start(&hti);
237 if (err && err != -EAGAIN) {
242 while ((he = rhashtable_walk_next(&hti))) {
245 if (err != -EAGAIN) {
253 if (iter->count < iter->skip)
255 if (nft_set_elem_expired(&he->ext))
257 if (!nft_set_elem_active(&he->ext, iter->genmask))
262 iter->err = iter->fn(ctx, set, iter, &elem);
271 rhashtable_walk_stop(&hti);
272 rhashtable_walk_exit(&hti);
275 static void nft_rhash_gc(struct work_struct *work)
278 struct nft_rhash_elem *he;
279 struct nft_rhash *priv;
280 struct nft_set_gc_batch *gcb = NULL;
281 struct rhashtable_iter hti;
284 priv = container_of(work, struct nft_rhash, gc_work.work);
285 set = nft_set_container_of(priv);
287 err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
291 err = rhashtable_walk_start(&hti);
292 if (err && err != -EAGAIN)
295 while ((he = rhashtable_walk_next(&hti))) {
297 if (PTR_ERR(he) != -EAGAIN)
302 if (!nft_set_elem_expired(&he->ext))
304 if (nft_set_elem_mark_busy(&he->ext))
307 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
310 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
311 atomic_dec(&set->nelems);
312 nft_set_gc_batch_add(gcb, he);
315 rhashtable_walk_stop(&hti);
316 rhashtable_walk_exit(&hti);
318 nft_set_gc_batch_complete(gcb);
320 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
321 nft_set_gc_interval(set));
324 static unsigned int nft_rhash_privsize(const struct nlattr * const nla[],
325 const struct nft_set_desc *desc)
327 return sizeof(struct nft_rhash);
330 static int nft_rhash_init(const struct nft_set *set,
331 const struct nft_set_desc *desc,
332 const struct nlattr * const tb[])
334 struct nft_rhash *priv = nft_set_priv(set);
335 struct rhashtable_params params = nft_rhash_params;
338 params.nelem_hint = desc->size ?: NFT_RHASH_ELEMENT_HINT;
339 params.key_len = set->klen;
341 err = rhashtable_init(&priv->ht, ¶ms);
345 INIT_DEFERRABLE_WORK(&priv->gc_work, nft_rhash_gc);
346 if (set->flags & NFT_SET_TIMEOUT)
347 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
348 nft_set_gc_interval(set));
352 static void nft_rhash_elem_destroy(void *ptr, void *arg)
354 nft_set_elem_destroy(arg, ptr, true);
357 static void nft_rhash_destroy(const struct nft_set *set)
359 struct nft_rhash *priv = nft_set_priv(set);
361 cancel_delayed_work_sync(&priv->gc_work);
363 rhashtable_free_and_destroy(&priv->ht, nft_rhash_elem_destroy,
367 /* Number of buckets is stored in u32, so cap our result to 1U<<31 */
368 #define NFT_MAX_BUCKETS (1U << 31)
370 static u32 nft_hash_buckets(u32 size)
372 u64 val = div_u64((u64)size * 4, 3);
374 if (val >= NFT_MAX_BUCKETS)
375 return NFT_MAX_BUCKETS;
377 return roundup_pow_of_two(val);
380 static bool nft_rhash_estimate(const struct nft_set_desc *desc, u32 features,
381 struct nft_set_estimate *est)
384 est->lookup = NFT_SET_CLASS_O_1;
385 est->space = NFT_SET_CLASS_O_N;
393 struct hlist_head table[];
396 struct nft_hash_elem {
397 struct hlist_node node;
398 struct nft_set_ext ext;
401 static bool nft_hash_lookup(const struct net *net, const struct nft_set *set,
402 const u32 *key, const struct nft_set_ext **ext)
404 struct nft_hash *priv = nft_set_priv(set);
405 u8 genmask = nft_genmask_cur(net);
406 const struct nft_hash_elem *he;
409 hash = jhash(key, set->klen, priv->seed);
410 hash = reciprocal_scale(hash, priv->buckets);
411 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
412 if (!memcmp(nft_set_ext_key(&he->ext), key, set->klen) &&
413 nft_set_elem_active(&he->ext, genmask)) {
421 /* nft_hash_select_ops() makes sure key size can be either 2 or 4 bytes . */
422 static inline u32 nft_hash_key(const u32 *key, u32 klen)
430 static bool nft_hash_lookup_fast(const struct net *net,
431 const struct nft_set *set,
432 const u32 *key, const struct nft_set_ext **ext)
434 struct nft_hash *priv = nft_set_priv(set);
435 u8 genmask = nft_genmask_cur(net);
436 const struct nft_hash_elem *he;
439 k1 = nft_hash_key(key, set->klen);
440 hash = jhash_1word(k1, priv->seed);
441 hash = reciprocal_scale(hash, priv->buckets);
442 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
443 k2 = nft_hash_key(nft_set_ext_key(&he->ext)->data, set->klen);
445 nft_set_elem_active(&he->ext, genmask)) {
453 static u32 nft_jhash(const struct nft_set *set, const struct nft_hash *priv,
454 const struct nft_set_ext *ext)
456 const struct nft_data *key = nft_set_ext_key(ext);
459 if (set->klen == 4) {
461 hash = jhash_1word(k1, priv->seed);
463 hash = jhash(key, set->klen, priv->seed);
465 hash = reciprocal_scale(hash, priv->buckets);
470 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
471 const struct nft_set_elem *elem,
472 struct nft_set_ext **ext)
474 struct nft_hash_elem *this = elem->priv, *he;
475 struct nft_hash *priv = nft_set_priv(set);
476 u8 genmask = nft_genmask_next(net);
479 hash = nft_jhash(set, priv, &this->ext);
480 hlist_for_each_entry(he, &priv->table[hash], node) {
481 if (!memcmp(nft_set_ext_key(&this->ext),
482 nft_set_ext_key(&he->ext), set->klen) &&
483 nft_set_elem_active(&he->ext, genmask)) {
488 hlist_add_head_rcu(&this->node, &priv->table[hash]);
492 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
493 const struct nft_set_elem *elem)
495 struct nft_hash_elem *he = elem->priv;
497 nft_set_elem_change_active(net, set, &he->ext);
500 static bool nft_hash_flush(const struct net *net,
501 const struct nft_set *set, void *priv)
503 struct nft_hash_elem *he = priv;
505 nft_set_elem_change_active(net, set, &he->ext);
509 static void *nft_hash_deactivate(const struct net *net,
510 const struct nft_set *set,
511 const struct nft_set_elem *elem)
513 struct nft_hash *priv = nft_set_priv(set);
514 struct nft_hash_elem *this = elem->priv, *he;
515 u8 genmask = nft_genmask_next(net);
518 hash = nft_jhash(set, priv, &this->ext);
519 hlist_for_each_entry(he, &priv->table[hash], node) {
520 if (!memcmp(nft_set_ext_key(&this->ext), &elem->key.val,
522 nft_set_elem_active(&he->ext, genmask)) {
523 nft_set_elem_change_active(net, set, &he->ext);
530 static void nft_hash_remove(const struct net *net,
531 const struct nft_set *set,
532 const struct nft_set_elem *elem)
534 struct nft_hash_elem *he = elem->priv;
536 hlist_del_rcu(&he->node);
539 static void nft_hash_walk(const struct nft_ctx *ctx, struct nft_set *set,
540 struct nft_set_iter *iter)
542 struct nft_hash *priv = nft_set_priv(set);
543 struct nft_hash_elem *he;
544 struct nft_set_elem elem;
547 for (i = 0; i < priv->buckets; i++) {
548 hlist_for_each_entry_rcu(he, &priv->table[i], node) {
549 if (iter->count < iter->skip)
551 if (!nft_set_elem_active(&he->ext, iter->genmask))
556 iter->err = iter->fn(ctx, set, iter, &elem);
565 static unsigned int nft_hash_privsize(const struct nlattr * const nla[],
566 const struct nft_set_desc *desc)
568 return sizeof(struct nft_hash) +
569 nft_hash_buckets(desc->size) * sizeof(struct hlist_head);
572 static int nft_hash_init(const struct nft_set *set,
573 const struct nft_set_desc *desc,
574 const struct nlattr * const tb[])
576 struct nft_hash *priv = nft_set_priv(set);
578 priv->buckets = nft_hash_buckets(desc->size);
579 get_random_bytes(&priv->seed, sizeof(priv->seed));
584 static void nft_hash_destroy(const struct nft_set *set)
586 struct nft_hash *priv = nft_set_priv(set);
587 struct nft_hash_elem *he;
588 struct hlist_node *next;
591 for (i = 0; i < priv->buckets; i++) {
592 hlist_for_each_entry_safe(he, next, &priv->table[i], node) {
593 hlist_del_rcu(&he->node);
594 nft_set_elem_destroy(set, he, true);
599 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
600 struct nft_set_estimate *est)
602 est->size = sizeof(struct nft_hash) +
603 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
604 desc->size * sizeof(struct nft_hash_elem);
605 est->lookup = NFT_SET_CLASS_O_1;
606 est->space = NFT_SET_CLASS_O_N;
611 static struct nft_set_type nft_hash_type;
612 static struct nft_set_ops nft_rhash_ops __read_mostly = {
613 .type = &nft_hash_type,
614 .privsize = nft_rhash_privsize,
615 .elemsize = offsetof(struct nft_rhash_elem, ext),
616 .estimate = nft_rhash_estimate,
617 .init = nft_rhash_init,
618 .destroy = nft_rhash_destroy,
619 .insert = nft_rhash_insert,
620 .activate = nft_rhash_activate,
621 .deactivate = nft_rhash_deactivate,
622 .flush = nft_rhash_flush,
623 .remove = nft_rhash_remove,
624 .lookup = nft_rhash_lookup,
625 .update = nft_rhash_update,
626 .walk = nft_rhash_walk,
627 .features = NFT_SET_MAP | NFT_SET_OBJECT | NFT_SET_TIMEOUT,
630 static struct nft_set_ops nft_hash_ops __read_mostly = {
631 .type = &nft_hash_type,
632 .privsize = nft_hash_privsize,
633 .elemsize = offsetof(struct nft_hash_elem, ext),
634 .estimate = nft_hash_estimate,
635 .init = nft_hash_init,
636 .destroy = nft_hash_destroy,
637 .insert = nft_hash_insert,
638 .activate = nft_hash_activate,
639 .deactivate = nft_hash_deactivate,
640 .flush = nft_hash_flush,
641 .remove = nft_hash_remove,
642 .lookup = nft_hash_lookup,
643 .walk = nft_hash_walk,
644 .features = NFT_SET_MAP | NFT_SET_OBJECT,
647 static struct nft_set_ops nft_hash_fast_ops __read_mostly = {
648 .type = &nft_hash_type,
649 .privsize = nft_hash_privsize,
650 .elemsize = offsetof(struct nft_hash_elem, ext),
651 .estimate = nft_hash_estimate,
652 .init = nft_hash_init,
653 .destroy = nft_hash_destroy,
654 .insert = nft_hash_insert,
655 .activate = nft_hash_activate,
656 .deactivate = nft_hash_deactivate,
657 .flush = nft_hash_flush,
658 .remove = nft_hash_remove,
659 .lookup = nft_hash_lookup_fast,
660 .walk = nft_hash_walk,
661 .features = NFT_SET_MAP | NFT_SET_OBJECT,
664 static const struct nft_set_ops *
665 nft_hash_select_ops(const struct nft_ctx *ctx, const struct nft_set_desc *desc,
669 switch (desc->klen) {
671 return &nft_hash_fast_ops;
673 return &nft_hash_ops;
677 return &nft_rhash_ops;
680 static struct nft_set_type nft_hash_type __read_mostly = {
681 .select_ops = nft_hash_select_ops,
682 .owner = THIS_MODULE,
685 static int __init nft_hash_module_init(void)
687 return nft_register_set(&nft_hash_type);
690 static void __exit nft_hash_module_exit(void)
692 nft_unregister_set(&nft_hash_type);
695 module_init(nft_hash_module_init);
696 module_exit(nft_hash_module_exit);
698 MODULE_LICENSE("GPL");
699 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
700 MODULE_ALIAS_NFT_SET();