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(&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(&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(&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(&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;
248 rhashtable_walk_enter(&priv->ht, &hti);
249 rhashtable_walk_start(&hti);
251 while ((he = rhashtable_walk_next(&hti))) {
253 if (PTR_ERR(he) != -EAGAIN) {
254 iter->err = PTR_ERR(he);
261 if (iter->count < iter->skip)
263 if (nft_set_elem_expired(&he->ext))
265 if (!nft_set_elem_active(&he->ext, iter->genmask))
270 iter->err = iter->fn(ctx, set, iter, &elem);
277 rhashtable_walk_stop(&hti);
278 rhashtable_walk_exit(&hti);
281 static void nft_rhash_gc(struct work_struct *work)
284 struct nft_rhash_elem *he;
285 struct nft_rhash *priv;
286 struct nft_set_gc_batch *gcb = NULL;
287 struct rhashtable_iter hti;
289 priv = container_of(work, struct nft_rhash, gc_work.work);
290 set = nft_set_container_of(priv);
292 rhashtable_walk_enter(&priv->ht, &hti);
293 rhashtable_walk_start(&hti);
295 while ((he = rhashtable_walk_next(&hti))) {
297 if (PTR_ERR(he) != -EAGAIN)
302 if (nft_set_ext_exists(&he->ext, NFT_SET_EXT_EXPR)) {
303 struct nft_expr *expr = nft_set_ext_expr(&he->ext);
306 expr->ops->gc(read_pnet(&set->net), expr))
309 if (!nft_set_elem_expired(&he->ext))
312 if (nft_set_elem_mark_busy(&he->ext))
315 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
318 rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
319 atomic_dec(&set->nelems);
320 nft_set_gc_batch_add(gcb, he);
322 rhashtable_walk_stop(&hti);
323 rhashtable_walk_exit(&hti);
325 nft_set_gc_batch_complete(gcb);
326 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
327 nft_set_gc_interval(set));
330 static u64 nft_rhash_privsize(const struct nlattr * const nla[],
331 const struct nft_set_desc *desc)
333 return sizeof(struct nft_rhash);
336 static void nft_rhash_gc_init(const struct nft_set *set)
338 struct nft_rhash *priv = nft_set_priv(set);
340 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
341 nft_set_gc_interval(set));
344 static int nft_rhash_init(const struct nft_set *set,
345 const struct nft_set_desc *desc,
346 const struct nlattr * const tb[])
348 struct nft_rhash *priv = nft_set_priv(set);
349 struct rhashtable_params params = nft_rhash_params;
352 params.nelem_hint = desc->size ?: NFT_RHASH_ELEMENT_HINT;
353 params.key_len = set->klen;
355 err = rhashtable_init(&priv->ht, ¶ms);
359 INIT_DEFERRABLE_WORK(&priv->gc_work, nft_rhash_gc);
360 if (set->flags & NFT_SET_TIMEOUT)
361 nft_rhash_gc_init(set);
366 static void nft_rhash_elem_destroy(void *ptr, void *arg)
368 nft_set_elem_destroy(arg, ptr, true);
371 static void nft_rhash_destroy(const struct nft_set *set)
373 struct nft_rhash *priv = nft_set_priv(set);
375 cancel_delayed_work_sync(&priv->gc_work);
377 rhashtable_free_and_destroy(&priv->ht, nft_rhash_elem_destroy,
381 static u32 nft_hash_buckets(u32 size)
383 return roundup_pow_of_two(size * 4 / 3);
386 static bool nft_rhash_estimate(const struct nft_set_desc *desc, u32 features,
387 struct nft_set_estimate *est)
390 est->lookup = NFT_SET_CLASS_O_1;
391 est->space = NFT_SET_CLASS_O_N;
399 struct hlist_head table[];
402 struct nft_hash_elem {
403 struct hlist_node node;
404 struct nft_set_ext ext;
407 static bool nft_hash_lookup(const struct net *net, const struct nft_set *set,
408 const u32 *key, const struct nft_set_ext **ext)
410 struct nft_hash *priv = nft_set_priv(set);
411 u8 genmask = nft_genmask_cur(net);
412 const struct nft_hash_elem *he;
415 hash = jhash(key, set->klen, priv->seed);
416 hash = reciprocal_scale(hash, priv->buckets);
417 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
418 if (!memcmp(nft_set_ext_key(&he->ext), key, set->klen) &&
419 nft_set_elem_active(&he->ext, genmask)) {
427 static void *nft_hash_get(const struct net *net, const struct nft_set *set,
428 const struct nft_set_elem *elem, unsigned int flags)
430 struct nft_hash *priv = nft_set_priv(set);
431 u8 genmask = nft_genmask_cur(net);
432 struct nft_hash_elem *he;
435 hash = jhash(elem->key.val.data, set->klen, priv->seed);
436 hash = reciprocal_scale(hash, priv->buckets);
437 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
438 if (!memcmp(nft_set_ext_key(&he->ext), elem->key.val.data, set->klen) &&
439 nft_set_elem_active(&he->ext, genmask))
442 return ERR_PTR(-ENOENT);
445 /* nft_hash_select_ops() makes sure key size can be either 2 or 4 bytes . */
446 static inline u32 nft_hash_key(const u32 *key, u32 klen)
454 static bool nft_hash_lookup_fast(const struct net *net,
455 const struct nft_set *set,
456 const u32 *key, const struct nft_set_ext **ext)
458 struct nft_hash *priv = nft_set_priv(set);
459 u8 genmask = nft_genmask_cur(net);
460 const struct nft_hash_elem *he;
463 k1 = nft_hash_key(key, set->klen);
464 hash = jhash_1word(k1, priv->seed);
465 hash = reciprocal_scale(hash, priv->buckets);
466 hlist_for_each_entry_rcu(he, &priv->table[hash], node) {
467 k2 = nft_hash_key(nft_set_ext_key(&he->ext)->data, set->klen);
469 nft_set_elem_active(&he->ext, genmask)) {
477 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
478 const struct nft_set_elem *elem,
479 struct nft_set_ext **ext)
481 struct nft_hash_elem *this = elem->priv, *he;
482 struct nft_hash *priv = nft_set_priv(set);
483 u8 genmask = nft_genmask_next(net);
486 hash = jhash(nft_set_ext_key(&this->ext), set->klen, priv->seed);
487 hash = reciprocal_scale(hash, priv->buckets);
488 hlist_for_each_entry(he, &priv->table[hash], node) {
489 if (!memcmp(nft_set_ext_key(&this->ext),
490 nft_set_ext_key(&he->ext), set->klen) &&
491 nft_set_elem_active(&he->ext, genmask)) {
496 hlist_add_head_rcu(&this->node, &priv->table[hash]);
500 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
501 const struct nft_set_elem *elem)
503 struct nft_hash_elem *he = elem->priv;
505 nft_set_elem_change_active(net, set, &he->ext);
508 static bool nft_hash_flush(const struct net *net,
509 const struct nft_set *set, void *priv)
511 struct nft_hash_elem *he = priv;
513 nft_set_elem_change_active(net, set, &he->ext);
517 static void *nft_hash_deactivate(const struct net *net,
518 const struct nft_set *set,
519 const struct nft_set_elem *elem)
521 struct nft_hash *priv = nft_set_priv(set);
522 struct nft_hash_elem *this = elem->priv, *he;
523 u8 genmask = nft_genmask_next(net);
526 hash = jhash(nft_set_ext_key(&this->ext), set->klen, priv->seed);
527 hash = reciprocal_scale(hash, priv->buckets);
528 hlist_for_each_entry(he, &priv->table[hash], node) {
529 if (!memcmp(nft_set_ext_key(&this->ext), &elem->key.val,
531 nft_set_elem_active(&he->ext, genmask)) {
532 nft_set_elem_change_active(net, set, &he->ext);
539 static void nft_hash_remove(const struct net *net,
540 const struct nft_set *set,
541 const struct nft_set_elem *elem)
543 struct nft_hash_elem *he = elem->priv;
545 hlist_del_rcu(&he->node);
548 static void nft_hash_walk(const struct nft_ctx *ctx, struct nft_set *set,
549 struct nft_set_iter *iter)
551 struct nft_hash *priv = nft_set_priv(set);
552 struct nft_hash_elem *he;
553 struct nft_set_elem elem;
556 for (i = 0; i < priv->buckets; i++) {
557 hlist_for_each_entry_rcu(he, &priv->table[i], node) {
558 if (iter->count < iter->skip)
560 if (!nft_set_elem_active(&he->ext, iter->genmask))
565 iter->err = iter->fn(ctx, set, iter, &elem);
574 static u64 nft_hash_privsize(const struct nlattr * const nla[],
575 const struct nft_set_desc *desc)
577 return sizeof(struct nft_hash) +
578 nft_hash_buckets(desc->size) * sizeof(struct hlist_head);
581 static int nft_hash_init(const struct nft_set *set,
582 const struct nft_set_desc *desc,
583 const struct nlattr * const tb[])
585 struct nft_hash *priv = nft_set_priv(set);
587 priv->buckets = nft_hash_buckets(desc->size);
588 get_random_bytes(&priv->seed, sizeof(priv->seed));
593 static void nft_hash_destroy(const struct nft_set *set)
595 struct nft_hash *priv = nft_set_priv(set);
596 struct nft_hash_elem *he;
597 struct hlist_node *next;
600 for (i = 0; i < priv->buckets; i++) {
601 hlist_for_each_entry_safe(he, next, &priv->table[i], node) {
602 hlist_del_rcu(&he->node);
603 nft_set_elem_destroy(set, he, true);
608 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
609 struct nft_set_estimate *est)
617 est->size = sizeof(struct nft_hash) +
618 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
619 desc->size * sizeof(struct nft_hash_elem);
620 est->lookup = NFT_SET_CLASS_O_1;
621 est->space = NFT_SET_CLASS_O_N;
626 static bool nft_hash_fast_estimate(const struct nft_set_desc *desc, u32 features,
627 struct nft_set_estimate *est)
635 est->size = sizeof(struct nft_hash) +
636 nft_hash_buckets(desc->size) * sizeof(struct hlist_head) +
637 desc->size * sizeof(struct nft_hash_elem);
638 est->lookup = NFT_SET_CLASS_O_1;
639 est->space = NFT_SET_CLASS_O_N;
644 struct nft_set_type nft_set_rhash_type __read_mostly = {
645 .owner = THIS_MODULE,
646 .features = NFT_SET_MAP | NFT_SET_OBJECT |
647 NFT_SET_TIMEOUT | NFT_SET_EVAL,
649 .privsize = nft_rhash_privsize,
650 .elemsize = offsetof(struct nft_rhash_elem, ext),
651 .estimate = nft_rhash_estimate,
652 .init = nft_rhash_init,
653 .gc_init = nft_rhash_gc_init,
654 .destroy = nft_rhash_destroy,
655 .insert = nft_rhash_insert,
656 .activate = nft_rhash_activate,
657 .deactivate = nft_rhash_deactivate,
658 .flush = nft_rhash_flush,
659 .remove = nft_rhash_remove,
660 .lookup = nft_rhash_lookup,
661 .update = nft_rhash_update,
662 .walk = nft_rhash_walk,
663 .get = nft_rhash_get,
667 struct nft_set_type nft_set_hash_type __read_mostly = {
668 .owner = THIS_MODULE,
669 .features = NFT_SET_MAP | NFT_SET_OBJECT,
671 .privsize = nft_hash_privsize,
672 .elemsize = offsetof(struct nft_hash_elem, ext),
673 .estimate = nft_hash_estimate,
674 .init = nft_hash_init,
675 .destroy = nft_hash_destroy,
676 .insert = nft_hash_insert,
677 .activate = nft_hash_activate,
678 .deactivate = nft_hash_deactivate,
679 .flush = nft_hash_flush,
680 .remove = nft_hash_remove,
681 .lookup = nft_hash_lookup,
682 .walk = nft_hash_walk,
687 struct nft_set_type nft_set_hash_fast_type __read_mostly = {
688 .owner = THIS_MODULE,
689 .features = NFT_SET_MAP | NFT_SET_OBJECT,
691 .privsize = nft_hash_privsize,
692 .elemsize = offsetof(struct nft_hash_elem, ext),
693 .estimate = nft_hash_fast_estimate,
694 .init = nft_hash_init,
695 .destroy = nft_hash_destroy,
696 .insert = nft_hash_insert,
697 .activate = nft_hash_activate,
698 .deactivate = nft_hash_deactivate,
699 .flush = nft_hash_flush,
700 .remove = nft_hash_remove,
701 .lookup = nft_hash_lookup_fast,
702 .walk = nft_hash_walk,