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 void *nft_rhash_get(const struct net *net, const struct nft_set *set,
99 const struct nft_set_elem *elem, unsigned int flags)
101 struct nft_rhash *priv = nft_set_priv(set);
102 struct nft_rhash_elem *he;
103 struct nft_rhash_cmp_arg arg = {
104 .genmask = nft_genmask_cur(net),
106 .key = elem->key.val.data,
109 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_rhash_params);
113 return ERR_PTR(-ENOENT);
116 static bool nft_rhash_update(struct nft_set *set, const u32 *key,
117 void *(*new)(struct nft_set *,
118 const struct nft_expr *,
119 struct nft_regs *regs),
120 const struct nft_expr *expr,
121 struct nft_regs *regs,
122 const struct nft_set_ext **ext)
124 struct nft_rhash *priv = nft_set_priv(set);
125 struct nft_rhash_elem *he, *prev;
126 struct nft_rhash_cmp_arg arg = {
127 .genmask = NFT_GENMASK_ANY,
132 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_rhash_params);
136 he = new(set, expr, regs);
140 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
145 /* Another cpu may race to insert the element with the same key */
147 nft_set_elem_destroy(set, he, true);
156 nft_set_elem_destroy(set, he, true);
161 static int nft_rhash_insert(const struct net *net, const struct nft_set *set,
162 const struct nft_set_elem *elem,
163 struct nft_set_ext **ext)
165 struct nft_rhash *priv = nft_set_priv(set);
166 struct nft_rhash_elem *he = elem->priv;
167 struct nft_rhash_cmp_arg arg = {
168 .genmask = nft_genmask_next(net),
170 .key = elem->key.val.data,
172 struct nft_rhash_elem *prev;
174 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
177 return PTR_ERR(prev);
185 static void nft_rhash_activate(const struct net *net, const struct nft_set *set,
186 const struct nft_set_elem *elem)
188 struct nft_rhash_elem *he = elem->priv;
190 nft_set_elem_change_active(net, set, &he->ext);
191 nft_set_elem_clear_busy(&he->ext);
194 static bool nft_rhash_flush(const struct net *net,
195 const struct nft_set *set, void *priv)
197 struct nft_rhash_elem *he = priv;
199 if (!nft_set_elem_mark_busy(&he->ext) ||
200 !nft_is_active(net, &he->ext)) {
201 nft_set_elem_change_active(net, set, &he->ext);
207 static void *nft_rhash_deactivate(const struct net *net,
208 const struct nft_set *set,
209 const struct nft_set_elem *elem)
211 struct nft_rhash *priv = nft_set_priv(set);
212 struct nft_rhash_elem *he;
213 struct nft_rhash_cmp_arg arg = {
214 .genmask = nft_genmask_next(net),
216 .key = elem->key.val.data,
220 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_rhash_params);
222 !nft_rhash_flush(net, set, he))
230 static void nft_rhash_remove(const struct net *net,
231 const struct nft_set *set,
232 const struct nft_set_elem *elem)
234 struct nft_rhash *priv = nft_set_priv(set);
235 struct nft_rhash_elem *he = elem->priv;
237 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
240 static void nft_rhash_walk(const struct nft_ctx *ctx, struct nft_set *set,
241 struct nft_set_iter *iter)
243 struct nft_rhash *priv = nft_set_priv(set);
244 struct nft_rhash_elem *he;
245 struct rhashtable_iter hti;
246 struct nft_set_elem elem;
249 err = rhashtable_walk_init(&priv->ht, &hti, GFP_ATOMIC);
254 rhashtable_walk_start(&hti);
256 while ((he = rhashtable_walk_next(&hti))) {
259 if (err != -EAGAIN) {
267 if (iter->count < iter->skip)
269 if (nft_set_elem_expired(&he->ext))
271 if (!nft_set_elem_active(&he->ext, iter->genmask))
276 iter->err = iter->fn(ctx, set, iter, &elem);
285 rhashtable_walk_stop(&hti);
286 rhashtable_walk_exit(&hti);
289 static void nft_rhash_gc(struct work_struct *work)
292 struct nft_rhash_elem *he;
293 struct nft_rhash *priv;
294 struct nft_set_gc_batch *gcb = NULL;
295 struct rhashtable_iter hti;
298 priv = container_of(work, struct nft_rhash, gc_work.work);
299 set = nft_set_container_of(priv);
301 err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
305 rhashtable_walk_start(&hti);
307 while ((he = rhashtable_walk_next(&hti))) {
309 if (PTR_ERR(he) != -EAGAIN)
314 if (!nft_set_elem_expired(&he->ext))
316 if (nft_set_elem_mark_busy(&he->ext))
319 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
322 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
323 atomic_dec(&set->nelems);
324 nft_set_gc_batch_add(gcb, he);
327 rhashtable_walk_stop(&hti);
328 rhashtable_walk_exit(&hti);
330 nft_set_gc_batch_complete(gcb);
332 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
333 nft_set_gc_interval(set));
336 static unsigned int nft_rhash_privsize(const struct nlattr * const nla[],
337 const struct nft_set_desc *desc)
339 return sizeof(struct nft_rhash);
342 static int nft_rhash_init(const struct nft_set *set,
343 const struct nft_set_desc *desc,
344 const struct nlattr * const tb[])
346 struct nft_rhash *priv = nft_set_priv(set);
347 struct rhashtable_params params = nft_rhash_params;
350 params.nelem_hint = desc->size ?: NFT_RHASH_ELEMENT_HINT;
351 params.key_len = set->klen;
353 err = rhashtable_init(&priv->ht, ¶ms);
357 INIT_DEFERRABLE_WORK(&priv->gc_work, nft_rhash_gc);
358 if (set->flags & NFT_SET_TIMEOUT)
359 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
360 nft_set_gc_interval(set));
364 static void nft_rhash_elem_destroy(void *ptr, void *arg)
366 nft_set_elem_destroy(arg, ptr, true);
369 static void nft_rhash_destroy(const struct nft_set *set)
371 struct nft_rhash *priv = nft_set_priv(set);
373 cancel_delayed_work_sync(&priv->gc_work);
374 rhashtable_free_and_destroy(&priv->ht, nft_rhash_elem_destroy,
378 static u32 nft_hash_buckets(u32 size)
380 return roundup_pow_of_two(size * 4 / 3);
383 static bool nft_rhash_estimate(const struct nft_set_desc *desc, u32 features,
384 struct nft_set_estimate *est)
387 est->lookup = NFT_SET_CLASS_O_1;
388 est->space = NFT_SET_CLASS_O_N;
396 struct hlist_head table[];
399 struct nft_hash_elem {
400 struct hlist_node node;
401 struct nft_set_ext ext;
404 static bool nft_hash_lookup(const struct net *net, const struct nft_set *set,
405 const u32 *key, const struct nft_set_ext **ext)
407 struct nft_hash *priv = nft_set_priv(set);
408 u8 genmask = nft_genmask_cur(net);
409 const struct nft_hash_elem *he;
412 hash = jhash(key, set->klen, priv->seed);
413 hash = reciprocal_scale(hash, priv->buckets);
414 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
415 if (!memcmp(nft_set_ext_key(&he->ext), key, set->klen) &&
416 nft_set_elem_active(&he->ext, genmask)) {
424 static void *nft_hash_get(const struct net *net, const struct nft_set *set,
425 const struct nft_set_elem *elem, unsigned int flags)
427 struct nft_hash *priv = nft_set_priv(set);
428 u8 genmask = nft_genmask_cur(net);
429 struct nft_hash_elem *he;
432 hash = jhash(elem->key.val.data, set->klen, priv->seed);
433 hash = reciprocal_scale(hash, priv->buckets);
434 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
435 if (!memcmp(nft_set_ext_key(&he->ext), elem->key.val.data, set->klen) &&
436 nft_set_elem_active(&he->ext, genmask))
439 return ERR_PTR(-ENOENT);
442 /* nft_hash_select_ops() makes sure key size can be either 2 or 4 bytes . */
443 static inline u32 nft_hash_key(const u32 *key, u32 klen)
451 static bool nft_hash_lookup_fast(const struct net *net,
452 const struct nft_set *set,
453 const u32 *key, const struct nft_set_ext **ext)
455 struct nft_hash *priv = nft_set_priv(set);
456 u8 genmask = nft_genmask_cur(net);
457 const struct nft_hash_elem *he;
460 k1 = nft_hash_key(key, set->klen);
461 hash = jhash_1word(k1, priv->seed);
462 hash = reciprocal_scale(hash, priv->buckets);
463 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
464 k2 = nft_hash_key(nft_set_ext_key(&he->ext)->data, set->klen);
466 nft_set_elem_active(&he->ext, genmask)) {
474 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
475 const struct nft_set_elem *elem,
476 struct nft_set_ext **ext)
478 struct nft_hash_elem *this = elem->priv, *he;
479 struct nft_hash *priv = nft_set_priv(set);
480 u8 genmask = nft_genmask_next(net);
483 hash = jhash(nft_set_ext_key(&this->ext), set->klen, priv->seed);
484 hash = reciprocal_scale(hash, priv->buckets);
485 hlist_for_each_entry(he, &priv->table[hash], node) {
486 if (!memcmp(nft_set_ext_key(&this->ext),
487 nft_set_ext_key(&he->ext), set->klen) &&
488 nft_set_elem_active(&he->ext, genmask)) {
493 hlist_add_head_rcu(&this->node, &priv->table[hash]);
497 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
498 const struct nft_set_elem *elem)
500 struct nft_hash_elem *he = elem->priv;
502 nft_set_elem_change_active(net, set, &he->ext);
505 static bool nft_hash_flush(const struct net *net,
506 const struct nft_set *set, void *priv)
508 struct nft_hash_elem *he = priv;
510 nft_set_elem_change_active(net, set, &he->ext);
514 static void *nft_hash_deactivate(const struct net *net,
515 const struct nft_set *set,
516 const struct nft_set_elem *elem)
518 struct nft_hash *priv = nft_set_priv(set);
519 struct nft_hash_elem *this = elem->priv, *he;
520 u8 genmask = nft_genmask_next(net);
523 hash = jhash(nft_set_ext_key(&this->ext), set->klen, priv->seed);
524 hash = reciprocal_scale(hash, priv->buckets);
525 hlist_for_each_entry(he, &priv->table[hash], node) {
526 if (!memcmp(nft_set_ext_key(&this->ext), &elem->key.val,
528 nft_set_elem_active(&he->ext, genmask)) {
529 nft_set_elem_change_active(net, set, &he->ext);
536 static void nft_hash_remove(const struct net *net,
537 const struct nft_set *set,
538 const struct nft_set_elem *elem)
540 struct nft_hash_elem *he = elem->priv;
542 hlist_del_rcu(&he->node);
545 static void nft_hash_walk(const struct nft_ctx *ctx, struct nft_set *set,
546 struct nft_set_iter *iter)
548 struct nft_hash *priv = nft_set_priv(set);
549 struct nft_hash_elem *he;
550 struct nft_set_elem elem;
553 for (i = 0; i < priv->buckets; i++) {
554 hlist_for_each_entry_rcu(he, &priv->table[i], node) {
555 if (iter->count < iter->skip)
557 if (!nft_set_elem_active(&he->ext, iter->genmask))
562 iter->err = iter->fn(ctx, set, iter, &elem);
571 static unsigned int nft_hash_privsize(const struct nlattr * const nla[],
572 const struct nft_set_desc *desc)
574 return sizeof(struct nft_hash) +
575 nft_hash_buckets(desc->size) * sizeof(struct hlist_head);
578 static int nft_hash_init(const struct nft_set *set,
579 const struct nft_set_desc *desc,
580 const struct nlattr * const tb[])
582 struct nft_hash *priv = nft_set_priv(set);
584 priv->buckets = nft_hash_buckets(desc->size);
585 get_random_bytes(&priv->seed, sizeof(priv->seed));
590 static void nft_hash_destroy(const struct nft_set *set)
592 struct nft_hash *priv = nft_set_priv(set);
593 struct nft_hash_elem *he;
594 struct hlist_node *next;
597 for (i = 0; i < priv->buckets; i++) {
598 hlist_for_each_entry_safe(he, next, &priv->table[i], node) {
599 hlist_del_rcu(&he->node);
600 nft_set_elem_destroy(set, he, true);
605 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
606 struct nft_set_estimate *est)
608 est->size = sizeof(struct nft_hash) +
609 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
610 desc->size * sizeof(struct nft_hash_elem);
611 est->lookup = NFT_SET_CLASS_O_1;
612 est->space = NFT_SET_CLASS_O_N;
617 static struct nft_set_type nft_hash_type;
618 static struct nft_set_ops nft_rhash_ops __read_mostly = {
619 .type = &nft_hash_type,
620 .privsize = nft_rhash_privsize,
621 .elemsize = offsetof(struct nft_rhash_elem, ext),
622 .estimate = nft_rhash_estimate,
623 .init = nft_rhash_init,
624 .destroy = nft_rhash_destroy,
625 .insert = nft_rhash_insert,
626 .activate = nft_rhash_activate,
627 .deactivate = nft_rhash_deactivate,
628 .flush = nft_rhash_flush,
629 .remove = nft_rhash_remove,
630 .lookup = nft_rhash_lookup,
631 .update = nft_rhash_update,
632 .walk = nft_rhash_walk,
633 .get = nft_rhash_get,
634 .features = NFT_SET_MAP | NFT_SET_OBJECT | NFT_SET_TIMEOUT,
637 static struct nft_set_ops nft_hash_ops __read_mostly = {
638 .type = &nft_hash_type,
639 .privsize = nft_hash_privsize,
640 .elemsize = offsetof(struct nft_hash_elem, ext),
641 .estimate = nft_hash_estimate,
642 .init = nft_hash_init,
643 .destroy = nft_hash_destroy,
644 .insert = nft_hash_insert,
645 .activate = nft_hash_activate,
646 .deactivate = nft_hash_deactivate,
647 .flush = nft_hash_flush,
648 .remove = nft_hash_remove,
649 .lookup = nft_hash_lookup,
650 .walk = nft_hash_walk,
652 .features = NFT_SET_MAP | NFT_SET_OBJECT,
655 static struct nft_set_ops nft_hash_fast_ops __read_mostly = {
656 .type = &nft_hash_type,
657 .privsize = nft_hash_privsize,
658 .elemsize = offsetof(struct nft_hash_elem, ext),
659 .estimate = nft_hash_estimate,
660 .init = nft_hash_init,
661 .destroy = nft_hash_destroy,
662 .insert = nft_hash_insert,
663 .activate = nft_hash_activate,
664 .deactivate = nft_hash_deactivate,
665 .flush = nft_hash_flush,
666 .remove = nft_hash_remove,
667 .lookup = nft_hash_lookup_fast,
668 .walk = nft_hash_walk,
670 .features = NFT_SET_MAP | NFT_SET_OBJECT,
673 static const struct nft_set_ops *
674 nft_hash_select_ops(const struct nft_ctx *ctx, const struct nft_set_desc *desc,
678 switch (desc->klen) {
680 return &nft_hash_fast_ops;
682 return &nft_hash_ops;
686 return &nft_rhash_ops;
689 static struct nft_set_type nft_hash_type __read_mostly = {
690 .select_ops = nft_hash_select_ops,
691 .owner = THIS_MODULE,
694 static int __init nft_hash_module_init(void)
696 return nft_register_set(&nft_hash_type);
699 static void __exit nft_hash_module_exit(void)
701 nft_unregister_set(&nft_hash_type);
704 module_init(nft_hash_module_init);
705 module_exit(nft_hash_module_exit);
707 MODULE_LICENSE("GPL");
708 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
709 MODULE_ALIAS_NFT_SET();