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);
148 atomic_dec(&set->nelems);
157 nft_set_elem_destroy(set, he, true);
158 atomic_dec(&set->nelems);
163 static int nft_rhash_insert(const struct net *net, const struct nft_set *set,
164 const struct nft_set_elem *elem,
165 struct nft_set_ext **ext)
167 struct nft_rhash *priv = nft_set_priv(set);
168 struct nft_rhash_elem *he = elem->priv;
169 struct nft_rhash_cmp_arg arg = {
170 .genmask = nft_genmask_next(net),
172 .key = elem->key.val.data,
174 struct nft_rhash_elem *prev;
176 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
179 return PTR_ERR(prev);
187 static void nft_rhash_activate(const struct net *net, const struct nft_set *set,
188 const struct nft_set_elem *elem)
190 struct nft_rhash_elem *he = elem->priv;
192 nft_set_elem_change_active(net, set, &he->ext);
193 nft_set_elem_clear_busy(&he->ext);
196 static bool nft_rhash_flush(const struct net *net,
197 const struct nft_set *set, void *priv)
199 struct nft_rhash_elem *he = priv;
201 if (!nft_set_elem_mark_busy(&he->ext) ||
202 !nft_is_active(net, &he->ext)) {
203 nft_set_elem_change_active(net, set, &he->ext);
209 static void *nft_rhash_deactivate(const struct net *net,
210 const struct nft_set *set,
211 const struct nft_set_elem *elem)
213 struct nft_rhash *priv = nft_set_priv(set);
214 struct nft_rhash_elem *he;
215 struct nft_rhash_cmp_arg arg = {
216 .genmask = nft_genmask_next(net),
218 .key = elem->key.val.data,
222 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_rhash_params);
224 !nft_rhash_flush(net, set, he))
232 static void nft_rhash_remove(const struct net *net,
233 const struct nft_set *set,
234 const struct nft_set_elem *elem)
236 struct nft_rhash *priv = nft_set_priv(set);
237 struct nft_rhash_elem *he = elem->priv;
239 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
242 static void nft_rhash_walk(const struct nft_ctx *ctx, struct nft_set *set,
243 struct nft_set_iter *iter)
245 struct nft_rhash *priv = nft_set_priv(set);
246 struct nft_rhash_elem *he;
247 struct rhashtable_iter hti;
248 struct nft_set_elem elem;
251 err = rhashtable_walk_init(&priv->ht, &hti, GFP_ATOMIC);
256 rhashtable_walk_start(&hti);
258 while ((he = rhashtable_walk_next(&hti))) {
261 if (err != -EAGAIN) {
269 if (iter->count < iter->skip)
271 if (nft_set_elem_expired(&he->ext))
273 if (!nft_set_elem_active(&he->ext, iter->genmask))
278 iter->err = iter->fn(ctx, set, iter, &elem);
287 rhashtable_walk_stop(&hti);
288 rhashtable_walk_exit(&hti);
291 static void nft_rhash_gc(struct work_struct *work)
294 struct nft_rhash_elem *he;
295 struct nft_rhash *priv;
296 struct nft_set_gc_batch *gcb = NULL;
297 struct rhashtable_iter hti;
300 priv = container_of(work, struct nft_rhash, gc_work.work);
301 set = nft_set_container_of(priv);
303 err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
307 rhashtable_walk_start(&hti);
309 while ((he = rhashtable_walk_next(&hti))) {
311 if (PTR_ERR(he) != -EAGAIN)
316 if (nft_set_ext_exists(&he->ext, NFT_SET_EXT_EXPR)) {
317 struct nft_expr *expr = nft_set_ext_expr(&he->ext);
320 expr->ops->gc(read_pnet(&set->net), expr))
323 if (!nft_set_elem_expired(&he->ext))
326 if (nft_set_elem_mark_busy(&he->ext))
329 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
332 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
333 atomic_dec(&set->nelems);
334 nft_set_gc_batch_add(gcb, he);
337 rhashtable_walk_stop(&hti);
338 rhashtable_walk_exit(&hti);
340 nft_set_gc_batch_complete(gcb);
342 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
343 nft_set_gc_interval(set));
346 static u64 nft_rhash_privsize(const struct nlattr * const nla[],
347 const struct nft_set_desc *desc)
349 return sizeof(struct nft_rhash);
352 static void nft_rhash_gc_init(const struct nft_set *set)
354 struct nft_rhash *priv = nft_set_priv(set);
356 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
357 nft_set_gc_interval(set));
360 static int nft_rhash_init(const struct nft_set *set,
361 const struct nft_set_desc *desc,
362 const struct nlattr * const tb[])
364 struct nft_rhash *priv = nft_set_priv(set);
365 struct rhashtable_params params = nft_rhash_params;
368 params.nelem_hint = desc->size ?: NFT_RHASH_ELEMENT_HINT;
369 params.key_len = set->klen;
371 err = rhashtable_init(&priv->ht, ¶ms);
375 INIT_DEFERRABLE_WORK(&priv->gc_work, nft_rhash_gc);
376 if (set->flags & NFT_SET_TIMEOUT)
377 nft_rhash_gc_init(set);
382 static void nft_rhash_elem_destroy(void *ptr, void *arg)
384 nft_set_elem_destroy(arg, ptr, true);
387 static void nft_rhash_destroy(const struct nft_set *set)
389 struct nft_rhash *priv = nft_set_priv(set);
391 cancel_delayed_work_sync(&priv->gc_work);
393 rhashtable_free_and_destroy(&priv->ht, nft_rhash_elem_destroy,
397 /* Number of buckets is stored in u32, so cap our result to 1U<<31 */
398 #define NFT_MAX_BUCKETS (1U << 31)
400 static u32 nft_hash_buckets(u32 size)
402 u64 val = div_u64((u64)size * 4, 3);
404 if (val >= NFT_MAX_BUCKETS)
405 return NFT_MAX_BUCKETS;
407 return roundup_pow_of_two(val);
410 static bool nft_rhash_estimate(const struct nft_set_desc *desc, u32 features,
411 struct nft_set_estimate *est)
414 est->lookup = NFT_SET_CLASS_O_1;
415 est->space = NFT_SET_CLASS_O_N;
423 struct hlist_head table[];
426 struct nft_hash_elem {
427 struct hlist_node node;
428 struct nft_set_ext ext;
431 static bool nft_hash_lookup(const struct net *net, 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 hash = jhash(key, set->klen, priv->seed);
440 hash = reciprocal_scale(hash, priv->buckets);
441 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
442 if (!memcmp(nft_set_ext_key(&he->ext), key, set->klen) &&
443 nft_set_elem_active(&he->ext, genmask)) {
451 static void *nft_hash_get(const struct net *net, const struct nft_set *set,
452 const struct nft_set_elem *elem, unsigned int flags)
454 struct nft_hash *priv = nft_set_priv(set);
455 u8 genmask = nft_genmask_cur(net);
456 struct nft_hash_elem *he;
459 hash = jhash(elem->key.val.data, set->klen, priv->seed);
460 hash = reciprocal_scale(hash, priv->buckets);
461 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
462 if (!memcmp(nft_set_ext_key(&he->ext), elem->key.val.data, set->klen) &&
463 nft_set_elem_active(&he->ext, genmask))
466 return ERR_PTR(-ENOENT);
469 /* nft_hash_select_ops() makes sure key size can be either 2 or 4 bytes . */
470 static inline u32 nft_hash_key(const u32 *key, u32 klen)
478 static bool nft_hash_lookup_fast(const struct net *net,
479 const struct nft_set *set,
480 const u32 *key, const struct nft_set_ext **ext)
482 struct nft_hash *priv = nft_set_priv(set);
483 u8 genmask = nft_genmask_cur(net);
484 const struct nft_hash_elem *he;
487 k1 = nft_hash_key(key, set->klen);
488 hash = jhash_1word(k1, priv->seed);
489 hash = reciprocal_scale(hash, priv->buckets);
490 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
491 k2 = nft_hash_key(nft_set_ext_key(&he->ext)->data, set->klen);
493 nft_set_elem_active(&he->ext, genmask)) {
501 static u32 nft_jhash(const struct nft_set *set, const struct nft_hash *priv,
502 const struct nft_set_ext *ext)
504 const struct nft_data *key = nft_set_ext_key(ext);
507 if (set->klen == 4) {
509 hash = jhash_1word(k1, priv->seed);
511 hash = jhash(key, set->klen, priv->seed);
513 hash = reciprocal_scale(hash, priv->buckets);
518 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
519 const struct nft_set_elem *elem,
520 struct nft_set_ext **ext)
522 struct nft_hash_elem *this = elem->priv, *he;
523 struct nft_hash *priv = nft_set_priv(set);
524 u8 genmask = nft_genmask_next(net);
527 hash = nft_jhash(set, priv, &this->ext);
528 hlist_for_each_entry(he, &priv->table[hash], node) {
529 if (!memcmp(nft_set_ext_key(&this->ext),
530 nft_set_ext_key(&he->ext), set->klen) &&
531 nft_set_elem_active(&he->ext, genmask)) {
536 hlist_add_head_rcu(&this->node, &priv->table[hash]);
540 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
541 const struct nft_set_elem *elem)
543 struct nft_hash_elem *he = elem->priv;
545 nft_set_elem_change_active(net, set, &he->ext);
548 static bool nft_hash_flush(const struct net *net,
549 const struct nft_set *set, void *priv)
551 struct nft_hash_elem *he = priv;
553 nft_set_elem_change_active(net, set, &he->ext);
557 static void *nft_hash_deactivate(const struct net *net,
558 const struct nft_set *set,
559 const struct nft_set_elem *elem)
561 struct nft_hash *priv = nft_set_priv(set);
562 struct nft_hash_elem *this = elem->priv, *he;
563 u8 genmask = nft_genmask_next(net);
566 hash = nft_jhash(set, priv, &this->ext);
567 hlist_for_each_entry(he, &priv->table[hash], node) {
568 if (!memcmp(nft_set_ext_key(&he->ext), &elem->key.val,
570 nft_set_elem_active(&he->ext, genmask)) {
571 nft_set_elem_change_active(net, set, &he->ext);
578 static void nft_hash_remove(const struct net *net,
579 const struct nft_set *set,
580 const struct nft_set_elem *elem)
582 struct nft_hash_elem *he = elem->priv;
584 hlist_del_rcu(&he->node);
587 static void nft_hash_walk(const struct nft_ctx *ctx, struct nft_set *set,
588 struct nft_set_iter *iter)
590 struct nft_hash *priv = nft_set_priv(set);
591 struct nft_hash_elem *he;
592 struct nft_set_elem elem;
595 for (i = 0; i < priv->buckets; i++) {
596 hlist_for_each_entry_rcu(he, &priv->table[i], node) {
597 if (iter->count < iter->skip)
599 if (!nft_set_elem_active(&he->ext, iter->genmask))
604 iter->err = iter->fn(ctx, set, iter, &elem);
613 static u64 nft_hash_privsize(const struct nlattr * const nla[],
614 const struct nft_set_desc *desc)
616 return sizeof(struct nft_hash) +
617 nft_hash_buckets(desc->size) * sizeof(struct hlist_head);
620 static int nft_hash_init(const struct nft_set *set,
621 const struct nft_set_desc *desc,
622 const struct nlattr * const tb[])
624 struct nft_hash *priv = nft_set_priv(set);
626 priv->buckets = nft_hash_buckets(desc->size);
627 get_random_bytes(&priv->seed, sizeof(priv->seed));
632 static void nft_hash_destroy(const struct nft_set *set)
634 struct nft_hash *priv = nft_set_priv(set);
635 struct nft_hash_elem *he;
636 struct hlist_node *next;
639 for (i = 0; i < priv->buckets; i++) {
640 hlist_for_each_entry_safe(he, next, &priv->table[i], node) {
641 hlist_del_rcu(&he->node);
642 nft_set_elem_destroy(set, he, true);
647 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
648 struct nft_set_estimate *est)
656 est->size = sizeof(struct nft_hash) +
657 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
658 desc->size * sizeof(struct nft_hash_elem);
659 est->lookup = NFT_SET_CLASS_O_1;
660 est->space = NFT_SET_CLASS_O_N;
665 static bool nft_hash_fast_estimate(const struct nft_set_desc *desc, u32 features,
666 struct nft_set_estimate *est)
674 est->size = sizeof(struct nft_hash) +
675 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
676 desc->size * sizeof(struct nft_hash_elem);
677 est->lookup = NFT_SET_CLASS_O_1;
678 est->space = NFT_SET_CLASS_O_N;
683 struct nft_set_type nft_set_rhash_type __read_mostly = {
684 .owner = THIS_MODULE,
685 .features = NFT_SET_MAP | NFT_SET_OBJECT |
686 NFT_SET_TIMEOUT | NFT_SET_EVAL,
688 .privsize = nft_rhash_privsize,
689 .elemsize = offsetof(struct nft_rhash_elem, ext),
690 .estimate = nft_rhash_estimate,
691 .init = nft_rhash_init,
692 .gc_init = nft_rhash_gc_init,
693 .destroy = nft_rhash_destroy,
694 .insert = nft_rhash_insert,
695 .activate = nft_rhash_activate,
696 .deactivate = nft_rhash_deactivate,
697 .flush = nft_rhash_flush,
698 .remove = nft_rhash_remove,
699 .lookup = nft_rhash_lookup,
700 .update = nft_rhash_update,
701 .walk = nft_rhash_walk,
702 .get = nft_rhash_get,
706 struct nft_set_type nft_set_hash_type __read_mostly = {
707 .owner = THIS_MODULE,
708 .features = NFT_SET_MAP | NFT_SET_OBJECT,
710 .privsize = nft_hash_privsize,
711 .elemsize = offsetof(struct nft_hash_elem, ext),
712 .estimate = nft_hash_estimate,
713 .init = nft_hash_init,
714 .destroy = nft_hash_destroy,
715 .insert = nft_hash_insert,
716 .activate = nft_hash_activate,
717 .deactivate = nft_hash_deactivate,
718 .flush = nft_hash_flush,
719 .remove = nft_hash_remove,
720 .lookup = nft_hash_lookup,
721 .walk = nft_hash_walk,
726 struct nft_set_type nft_set_hash_fast_type __read_mostly = {
727 .owner = THIS_MODULE,
728 .features = NFT_SET_MAP | NFT_SET_OBJECT,
730 .privsize = nft_hash_privsize,
731 .elemsize = offsetof(struct nft_hash_elem, ext),
732 .estimate = nft_hash_fast_estimate,
733 .init = nft_hash_init,
734 .destroy = nft_hash_destroy,
735 .insert = nft_hash_insert,
736 .activate = nft_hash_activate,
737 .deactivate = nft_hash_deactivate,
738 .flush = nft_hash_flush,
739 .remove = nft_hash_remove,
740 .lookup = nft_hash_lookup_fast,
741 .walk = nft_hash_walk,