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_HASH_ELEMENT_HINT 3
29 struct delayed_work gc_work;
32 struct nft_hash_elem {
33 struct rhash_head node;
34 struct nft_set_ext ext;
37 struct nft_hash_cmp_arg {
38 const struct nft_set *set;
43 static const struct rhashtable_params nft_hash_params;
45 static inline u32 nft_hash_key(const void *data, u32 len, u32 seed)
47 const struct nft_hash_cmp_arg *arg = data;
49 return jhash(arg->key, len, seed);
52 static inline u32 nft_hash_obj(const void *data, u32 len, u32 seed)
54 const struct nft_hash_elem *he = data;
56 return jhash(nft_set_ext_key(&he->ext), len, seed);
59 static inline int nft_hash_cmp(struct rhashtable_compare_arg *arg,
62 const struct nft_hash_cmp_arg *x = arg->key;
63 const struct nft_hash_elem *he = ptr;
65 if (memcmp(nft_set_ext_key(&he->ext), x->key, x->set->klen))
67 if (nft_set_elem_expired(&he->ext))
69 if (!nft_set_elem_active(&he->ext, x->genmask))
74 static bool nft_hash_lookup(const struct net *net, const struct nft_set *set,
75 const u32 *key, const struct nft_set_ext **ext)
77 struct nft_hash *priv = nft_set_priv(set);
78 const struct nft_hash_elem *he;
79 struct nft_hash_cmp_arg arg = {
80 .genmask = nft_genmask_cur(net),
85 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params);
92 static bool nft_hash_update(struct nft_set *set, const u32 *key,
93 void *(*new)(struct nft_set *,
94 const struct nft_expr *,
95 struct nft_regs *regs),
96 const struct nft_expr *expr,
97 struct nft_regs *regs,
98 const struct nft_set_ext **ext)
100 struct nft_hash *priv = nft_set_priv(set);
101 struct nft_hash_elem *he, *prev;
102 struct nft_hash_cmp_arg arg = {
103 .genmask = NFT_GENMASK_ANY,
108 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params);
112 he = new(set, expr, regs);
116 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
121 /* Another cpu may race to insert the element with the same key */
123 nft_set_elem_destroy(set, he, true);
124 atomic_dec(&set->nelems);
133 nft_set_elem_destroy(set, he, true);
134 atomic_dec(&set->nelems);
139 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
140 const struct nft_set_elem *elem,
141 struct nft_set_ext **ext)
143 struct nft_hash *priv = nft_set_priv(set);
144 struct nft_hash_elem *he = elem->priv;
145 struct nft_hash_cmp_arg arg = {
146 .genmask = nft_genmask_next(net),
148 .key = elem->key.val.data,
150 struct nft_hash_elem *prev;
152 prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
155 return PTR_ERR(prev);
163 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
164 const struct nft_set_elem *elem)
166 struct nft_hash_elem *he = elem->priv;
168 nft_set_elem_change_active(net, set, &he->ext);
169 nft_set_elem_clear_busy(&he->ext);
172 static void *nft_hash_deactivate(const struct net *net,
173 const struct nft_set *set,
174 const struct nft_set_elem *elem)
176 struct nft_hash *priv = nft_set_priv(set);
177 struct nft_hash_elem *he;
178 struct nft_hash_cmp_arg arg = {
179 .genmask = nft_genmask_next(net),
181 .key = elem->key.val.data,
185 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params);
187 if (!nft_set_elem_mark_busy(&he->ext) ||
188 !nft_is_active(net, &he->ext))
189 nft_set_elem_change_active(net, set, &he->ext);
198 static void nft_hash_remove(const struct nft_set *set,
199 const struct nft_set_elem *elem)
201 struct nft_hash *priv = nft_set_priv(set);
202 struct nft_hash_elem *he = elem->priv;
204 rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
207 static void nft_hash_walk(const struct nft_ctx *ctx, const struct nft_set *set,
208 struct nft_set_iter *iter)
210 struct nft_hash *priv = nft_set_priv(set);
211 struct nft_hash_elem *he;
212 struct rhashtable_iter hti;
213 struct nft_set_elem elem;
216 err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
221 err = rhashtable_walk_start(&hti);
222 if (err && err != -EAGAIN) {
227 while ((he = rhashtable_walk_next(&hti))) {
230 if (err != -EAGAIN) {
238 if (iter->count < iter->skip)
240 if (nft_set_elem_expired(&he->ext))
242 if (!nft_set_elem_active(&he->ext, iter->genmask))
247 iter->err = iter->fn(ctx, set, iter, &elem);
256 rhashtable_walk_stop(&hti);
257 rhashtable_walk_exit(&hti);
260 static void nft_hash_gc(struct work_struct *work)
263 struct nft_hash_elem *he;
264 struct nft_hash *priv;
265 struct nft_set_gc_batch *gcb = NULL;
266 struct rhashtable_iter hti;
269 priv = container_of(work, struct nft_hash, gc_work.work);
270 set = nft_set_container_of(priv);
272 err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
276 err = rhashtable_walk_start(&hti);
277 if (err && err != -EAGAIN)
280 while ((he = rhashtable_walk_next(&hti))) {
282 if (PTR_ERR(he) != -EAGAIN)
287 if (!nft_set_elem_expired(&he->ext))
289 if (nft_set_elem_mark_busy(&he->ext))
292 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
295 rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
296 atomic_dec(&set->nelems);
297 nft_set_gc_batch_add(gcb, he);
300 rhashtable_walk_stop(&hti);
301 rhashtable_walk_exit(&hti);
303 nft_set_gc_batch_complete(gcb);
305 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
306 nft_set_gc_interval(set));
309 static unsigned int nft_hash_privsize(const struct nlattr * const nla[])
311 return sizeof(struct nft_hash);
314 static const struct rhashtable_params nft_hash_params = {
315 .head_offset = offsetof(struct nft_hash_elem, node),
316 .hashfn = nft_hash_key,
317 .obj_hashfn = nft_hash_obj,
318 .obj_cmpfn = nft_hash_cmp,
319 .automatic_shrinking = true,
322 static int nft_hash_init(const struct nft_set *set,
323 const struct nft_set_desc *desc,
324 const struct nlattr * const tb[])
326 struct nft_hash *priv = nft_set_priv(set);
327 struct rhashtable_params params = nft_hash_params;
330 params.nelem_hint = desc->size ?: NFT_HASH_ELEMENT_HINT;
331 params.key_len = set->klen;
333 err = rhashtable_init(&priv->ht, ¶ms);
337 INIT_DEFERRABLE_WORK(&priv->gc_work, nft_hash_gc);
338 if (set->flags & NFT_SET_TIMEOUT)
339 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
340 nft_set_gc_interval(set));
344 static void nft_hash_elem_destroy(void *ptr, void *arg)
346 nft_set_elem_destroy((const struct nft_set *)arg, ptr, true);
349 static void nft_hash_destroy(const struct nft_set *set)
351 struct nft_hash *priv = nft_set_priv(set);
353 cancel_delayed_work_sync(&priv->gc_work);
354 rhashtable_free_and_destroy(&priv->ht, nft_hash_elem_destroy,
358 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
359 struct nft_set_estimate *est)
363 esize = sizeof(struct nft_hash_elem);
365 est->size = sizeof(struct nft_hash) +
366 roundup_pow_of_two(desc->size * 4 / 3) *
367 sizeof(struct nft_hash_elem *) +
370 /* Resizing happens when the load drops below 30% or goes
371 * above 75%. The average of 52.5% load (approximated by 50%)
372 * is used for the size estimation of the hash buckets,
373 * meaning we calculate two buckets per element.
375 est->size = esize + 2 * sizeof(struct nft_hash_elem *);
378 est->class = NFT_SET_CLASS_O_1;
383 static struct nft_set_ops nft_hash_ops __read_mostly = {
384 .privsize = nft_hash_privsize,
385 .elemsize = offsetof(struct nft_hash_elem, ext),
386 .estimate = nft_hash_estimate,
387 .init = nft_hash_init,
388 .destroy = nft_hash_destroy,
389 .insert = nft_hash_insert,
390 .activate = nft_hash_activate,
391 .deactivate = nft_hash_deactivate,
392 .remove = nft_hash_remove,
393 .lookup = nft_hash_lookup,
394 .update = nft_hash_update,
395 .walk = nft_hash_walk,
396 .features = NFT_SET_MAP | NFT_SET_TIMEOUT,
397 .owner = THIS_MODULE,
400 static int __init nft_hash_module_init(void)
402 return nft_register_set(&nft_hash_ops);
405 static void __exit nft_hash_module_exit(void)
407 nft_unregister_set(&nft_hash_ops);
410 module_init(nft_hash_module_init);
411 module_exit(nft_hash_module_exit);
413 MODULE_LICENSE("GPL");
414 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
415 MODULE_ALIAS_NFT_SET();