Commit | Line | Data |
---|---|---|
96518518 | 1 | /* |
ce6eb0d7 | 2 | * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net> |
96518518 PM |
3 | * |
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. | |
7 | * | |
8 | * Development of this code funded by Astaro AG (http://www.astaro.com/) | |
9 | */ | |
10 | ||
11 | #include <linux/kernel.h> | |
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/list.h> | |
c50b960c | 15 | #include <linux/log2.h> |
96518518 PM |
16 | #include <linux/jhash.h> |
17 | #include <linux/netlink.h> | |
cfe4a9dd | 18 | #include <linux/rhashtable.h> |
96518518 PM |
19 | #include <linux/netfilter.h> |
20 | #include <linux/netfilter/nf_tables.h> | |
21 | #include <net/netfilter/nf_tables.h> | |
22 | ||
cfe4a9dd TG |
23 | /* We target a hash table size of 4, element hint is 75% of final size */ |
24 | #define NFT_HASH_ELEMENT_HINT 3 | |
96518518 PM |
25 | |
26 | struct nft_hash_elem { | |
cfe4a9dd | 27 | struct rhash_head node; |
ce6eb0d7 PM |
28 | struct nft_data key; |
29 | struct nft_data data[]; | |
96518518 PM |
30 | }; |
31 | ||
20a69341 PM |
32 | static bool nft_hash_lookup(const struct nft_set *set, |
33 | const struct nft_data *key, | |
34 | struct nft_data *data) | |
96518518 | 35 | { |
cfe4a9dd | 36 | const struct rhashtable *priv = nft_set_priv(set); |
20a69341 | 37 | const struct nft_hash_elem *he; |
ce6eb0d7 | 38 | |
cfe4a9dd TG |
39 | he = rhashtable_lookup(priv, key); |
40 | if (he && set->flags & NFT_SET_MAP) | |
41 | nft_data_copy(data, he->data); | |
ce6eb0d7 | 42 | |
cfe4a9dd | 43 | return !!he; |
96518518 PM |
44 | } |
45 | ||
20a69341 PM |
46 | static int nft_hash_insert(const struct nft_set *set, |
47 | const struct nft_set_elem *elem) | |
96518518 | 48 | { |
cfe4a9dd | 49 | struct rhashtable *priv = nft_set_priv(set); |
20a69341 | 50 | struct nft_hash_elem *he; |
cfe4a9dd | 51 | unsigned int size; |
96518518 | 52 | |
20a69341 | 53 | if (elem->flags != 0) |
96518518 | 54 | return -EINVAL; |
96518518 | 55 | |
20a69341 PM |
56 | size = sizeof(*he); |
57 | if (set->flags & NFT_SET_MAP) | |
58 | size += sizeof(he->data[0]); | |
59 | ||
60 | he = kzalloc(size, GFP_KERNEL); | |
61 | if (he == NULL) | |
96518518 PM |
62 | return -ENOMEM; |
63 | ||
20a69341 PM |
64 | nft_data_copy(&he->key, &elem->key); |
65 | if (set->flags & NFT_SET_MAP) | |
66 | nft_data_copy(he->data, &elem->data); | |
96518518 | 67 | |
6eba8224 | 68 | rhashtable_insert(priv, &he->node); |
ce6eb0d7 | 69 | |
96518518 | 70 | return 0; |
96518518 PM |
71 | } |
72 | ||
ce6eb0d7 PM |
73 | static void nft_hash_elem_destroy(const struct nft_set *set, |
74 | struct nft_hash_elem *he) | |
75 | { | |
76 | nft_data_uninit(&he->key, NFT_DATA_VALUE); | |
77 | if (set->flags & NFT_SET_MAP) | |
78 | nft_data_uninit(he->data, set->dtype); | |
79 | kfree(he); | |
80 | } | |
81 | ||
20a69341 PM |
82 | static void nft_hash_remove(const struct nft_set *set, |
83 | const struct nft_set_elem *elem) | |
96518518 | 84 | { |
cfe4a9dd TG |
85 | struct rhashtable *priv = nft_set_priv(set); |
86 | struct rhash_head *he, __rcu **pprev; | |
96518518 | 87 | |
ce6eb0d7 | 88 | pprev = elem->cookie; |
cfe4a9dd TG |
89 | he = rht_dereference((*pprev), priv); |
90 | ||
6eba8224 | 91 | rhashtable_remove_pprev(priv, he, pprev); |
ce6eb0d7 | 92 | |
ce6eb0d7 | 93 | synchronize_rcu(); |
20a69341 PM |
94 | kfree(he); |
95 | } | |
96518518 | 96 | |
8d24c0b4 TG |
97 | struct nft_compare_arg { |
98 | const struct nft_set *set; | |
99 | struct nft_set_elem *elem; | |
100 | }; | |
101 | ||
102 | static bool nft_hash_compare(void *ptr, void *arg) | |
103 | { | |
104 | struct nft_hash_elem *he = ptr; | |
105 | struct nft_compare_arg *x = arg; | |
106 | ||
107 | if (!nft_data_cmp(&he->key, &x->elem->key, x->set->klen)) { | |
108 | x->elem->cookie = &he->node; | |
109 | x->elem->flags = 0; | |
110 | if (x->set->flags & NFT_SET_MAP) | |
111 | nft_data_copy(&x->elem->data, he->data); | |
112 | ||
113 | return true; | |
114 | } | |
115 | ||
116 | return false; | |
117 | } | |
118 | ||
20a69341 PM |
119 | static int nft_hash_get(const struct nft_set *set, struct nft_set_elem *elem) |
120 | { | |
cfe4a9dd | 121 | const struct rhashtable *priv = nft_set_priv(set); |
8d24c0b4 TG |
122 | struct nft_compare_arg arg = { |
123 | .set = set, | |
124 | .elem = elem, | |
125 | }; | |
126 | ||
127 | if (rhashtable_lookup_compare(priv, &elem->key, | |
128 | &nft_hash_compare, &arg)) | |
20a69341 | 129 | return 0; |
8d24c0b4 | 130 | |
20a69341 | 131 | return -ENOENT; |
96518518 PM |
132 | } |
133 | ||
20a69341 PM |
134 | static void nft_hash_walk(const struct nft_ctx *ctx, const struct nft_set *set, |
135 | struct nft_set_iter *iter) | |
96518518 | 136 | { |
cfe4a9dd TG |
137 | const struct rhashtable *priv = nft_set_priv(set); |
138 | const struct bucket_table *tbl; | |
20a69341 PM |
139 | const struct nft_hash_elem *he; |
140 | struct nft_set_elem elem; | |
96518518 PM |
141 | unsigned int i; |
142 | ||
cfe4a9dd | 143 | tbl = rht_dereference_rcu(priv->tbl, priv); |
ce6eb0d7 | 144 | for (i = 0; i < tbl->size; i++) { |
cfe4a9dd | 145 | rht_for_each_entry_rcu(he, tbl->buckets[i], node) { |
20a69341 PM |
146 | if (iter->count < iter->skip) |
147 | goto cont; | |
148 | ||
149 | memcpy(&elem.key, &he->key, sizeof(elem.key)); | |
150 | if (set->flags & NFT_SET_MAP) | |
151 | memcpy(&elem.data, he->data, sizeof(elem.data)); | |
152 | elem.flags = 0; | |
153 | ||
154 | iter->err = iter->fn(ctx, set, iter, &elem); | |
155 | if (iter->err < 0) | |
156 | return; | |
157 | cont: | |
158 | iter->count++; | |
96518518 PM |
159 | } |
160 | } | |
96518518 PM |
161 | } |
162 | ||
20a69341 PM |
163 | static unsigned int nft_hash_privsize(const struct nlattr * const nla[]) |
164 | { | |
cfe4a9dd TG |
165 | return sizeof(struct rhashtable); |
166 | } | |
167 | ||
1f501d62 | 168 | #ifdef CONFIG_PROVE_LOCKING |
7b4ce235 | 169 | static int lockdep_nfnl_lock_is_held(void *parent) |
cfe4a9dd TG |
170 | { |
171 | return lockdep_nfnl_is_held(NFNL_SUBSYS_NFTABLES); | |
20a69341 | 172 | } |
1f501d62 | 173 | #endif |
96518518 | 174 | |
20a69341 | 175 | static int nft_hash_init(const struct nft_set *set, |
c50b960c | 176 | const struct nft_set_desc *desc, |
96518518 PM |
177 | const struct nlattr * const tb[]) |
178 | { | |
cfe4a9dd TG |
179 | struct rhashtable *priv = nft_set_priv(set); |
180 | struct rhashtable_params params = { | |
181 | .nelem_hint = desc->size ? : NFT_HASH_ELEMENT_HINT, | |
182 | .head_offset = offsetof(struct nft_hash_elem, node), | |
183 | .key_offset = offsetof(struct nft_hash_elem, key), | |
184 | .key_len = set->klen, | |
185 | .hashfn = jhash, | |
186 | .grow_decision = rht_grow_above_75, | |
187 | .shrink_decision = rht_shrink_below_30, | |
1f501d62 | 188 | #ifdef CONFIG_PROVE_LOCKING |
cfe4a9dd | 189 | .mutex_is_held = lockdep_nfnl_lock_is_held, |
1f501d62 | 190 | #endif |
cfe4a9dd | 191 | }; |
96518518 | 192 | |
cfe4a9dd | 193 | return rhashtable_init(priv, ¶ms); |
96518518 PM |
194 | } |
195 | ||
20a69341 | 196 | static void nft_hash_destroy(const struct nft_set *set) |
96518518 | 197 | { |
cfe4a9dd | 198 | const struct rhashtable *priv = nft_set_priv(set); |
39f39016 | 199 | const struct bucket_table *tbl = priv->tbl; |
ce6eb0d7 | 200 | struct nft_hash_elem *he, *next; |
96518518 PM |
201 | unsigned int i; |
202 | ||
39f39016 PNA |
203 | for (i = 0; i < tbl->size; i++) { |
204 | for (he = rht_entry(tbl->buckets[i], struct nft_hash_elem, node); | |
205 | he != NULL; he = next) { | |
206 | next = rht_entry(he->node.next, struct nft_hash_elem, node); | |
ce6eb0d7 | 207 | nft_hash_elem_destroy(set, he); |
39f39016 PNA |
208 | } |
209 | } | |
cfe4a9dd | 210 | rhashtable_destroy(priv); |
96518518 PM |
211 | } |
212 | ||
c50b960c PM |
213 | static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features, |
214 | struct nft_set_estimate *est) | |
215 | { | |
216 | unsigned int esize; | |
217 | ||
218 | esize = sizeof(struct nft_hash_elem); | |
219 | if (features & NFT_SET_MAP) | |
220 | esize += FIELD_SIZEOF(struct nft_hash_elem, data[0]); | |
221 | ||
222 | if (desc->size) { | |
cfe4a9dd TG |
223 | est->size = sizeof(struct rhashtable) + |
224 | roundup_pow_of_two(desc->size * 4 / 3) * | |
c50b960c PM |
225 | sizeof(struct nft_hash_elem *) + |
226 | desc->size * esize; | |
227 | } else { | |
228 | /* Resizing happens when the load drops below 30% or goes | |
229 | * above 75%. The average of 52.5% load (approximated by 50%) | |
230 | * is used for the size estimation of the hash buckets, | |
231 | * meaning we calculate two buckets per element. | |
232 | */ | |
233 | est->size = esize + 2 * sizeof(struct nft_hash_elem *); | |
234 | } | |
235 | ||
236 | est->class = NFT_SET_CLASS_O_1; | |
237 | ||
238 | return true; | |
239 | } | |
240 | ||
20a69341 PM |
241 | static struct nft_set_ops nft_hash_ops __read_mostly = { |
242 | .privsize = nft_hash_privsize, | |
c50b960c | 243 | .estimate = nft_hash_estimate, |
96518518 PM |
244 | .init = nft_hash_init, |
245 | .destroy = nft_hash_destroy, | |
20a69341 PM |
246 | .get = nft_hash_get, |
247 | .insert = nft_hash_insert, | |
248 | .remove = nft_hash_remove, | |
249 | .lookup = nft_hash_lookup, | |
250 | .walk = nft_hash_walk, | |
251 | .features = NFT_SET_MAP, | |
252 | .owner = THIS_MODULE, | |
96518518 PM |
253 | }; |
254 | ||
255 | static int __init nft_hash_module_init(void) | |
256 | { | |
20a69341 | 257 | return nft_register_set(&nft_hash_ops); |
96518518 PM |
258 | } |
259 | ||
260 | static void __exit nft_hash_module_exit(void) | |
261 | { | |
20a69341 | 262 | nft_unregister_set(&nft_hash_ops); |
96518518 PM |
263 | } |
264 | ||
265 | module_init(nft_hash_module_init); | |
266 | module_exit(nft_hash_module_exit); | |
267 | ||
268 | MODULE_LICENSE("GPL"); | |
269 | MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>"); | |
20a69341 | 270 | MODULE_ALIAS_NFT_SET(); |