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> | |
9d098292 | 18 | #include <linux/workqueue.h> |
cfe4a9dd | 19 | #include <linux/rhashtable.h> |
96518518 PM |
20 | #include <linux/netfilter.h> |
21 | #include <linux/netfilter/nf_tables.h> | |
22 | #include <net/netfilter/nf_tables.h> | |
23 | ||
cfe4a9dd TG |
24 | /* We target a hash table size of 4, element hint is 75% of final size */ |
25 | #define NFT_HASH_ELEMENT_HINT 3 | |
96518518 | 26 | |
745f5450 PM |
27 | struct nft_hash { |
28 | struct rhashtable ht; | |
9d098292 | 29 | struct delayed_work gc_work; |
745f5450 PM |
30 | }; |
31 | ||
96518518 | 32 | struct nft_hash_elem { |
cfe4a9dd | 33 | struct rhash_head node; |
fe2811eb | 34 | struct nft_set_ext ext; |
96518518 PM |
35 | }; |
36 | ||
bfd6e327 PM |
37 | struct nft_hash_cmp_arg { |
38 | const struct nft_set *set; | |
39 | const struct nft_data *key; | |
cc02e457 | 40 | u8 genmask; |
bfd6e327 PM |
41 | }; |
42 | ||
fa377321 HX |
43 | static const struct rhashtable_params nft_hash_params; |
44 | ||
bfd6e327 PM |
45 | static inline u32 nft_hash_key(const void *data, u32 len, u32 seed) |
46 | { | |
47 | const struct nft_hash_cmp_arg *arg = data; | |
48 | ||
49 | return jhash(arg->key, len, seed); | |
50 | } | |
51 | ||
52 | static inline u32 nft_hash_obj(const void *data, u32 len, u32 seed) | |
53 | { | |
54 | const struct nft_hash_elem *he = data; | |
55 | ||
fe2811eb | 56 | return jhash(nft_set_ext_key(&he->ext), len, seed); |
bfd6e327 PM |
57 | } |
58 | ||
59 | static inline int nft_hash_cmp(struct rhashtable_compare_arg *arg, | |
60 | const void *ptr) | |
61 | { | |
62 | const struct nft_hash_cmp_arg *x = arg->key; | |
63 | const struct nft_hash_elem *he = ptr; | |
64 | ||
fe2811eb | 65 | if (nft_data_cmp(nft_set_ext_key(&he->ext), x->key, x->set->klen)) |
bfd6e327 | 66 | return 1; |
9d098292 PM |
67 | if (nft_set_elem_expired(&he->ext)) |
68 | return 1; | |
cc02e457 PM |
69 | if (!nft_set_elem_active(&he->ext, x->genmask)) |
70 | return 1; | |
bfd6e327 PM |
71 | return 0; |
72 | } | |
73 | ||
20a69341 PM |
74 | static bool nft_hash_lookup(const struct nft_set *set, |
75 | const struct nft_data *key, | |
b2832dd6 | 76 | const struct nft_set_ext **ext) |
96518518 | 77 | { |
745f5450 | 78 | struct nft_hash *priv = nft_set_priv(set); |
20a69341 | 79 | const struct nft_hash_elem *he; |
bfd6e327 | 80 | struct nft_hash_cmp_arg arg = { |
cc02e457 | 81 | .genmask = nft_genmask_cur(read_pnet(&set->pnet)), |
bfd6e327 PM |
82 | .set = set, |
83 | .key = key, | |
84 | }; | |
ce6eb0d7 | 85 | |
bfd6e327 | 86 | he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params); |
b2832dd6 PM |
87 | if (he != NULL) |
88 | *ext = &he->ext; | |
ce6eb0d7 | 89 | |
cfe4a9dd | 90 | return !!he; |
96518518 PM |
91 | } |
92 | ||
20a69341 PM |
93 | static int nft_hash_insert(const struct nft_set *set, |
94 | const struct nft_set_elem *elem) | |
96518518 | 95 | { |
745f5450 | 96 | struct nft_hash *priv = nft_set_priv(set); |
fe2811eb | 97 | struct nft_hash_elem *he = elem->priv; |
bfd6e327 | 98 | struct nft_hash_cmp_arg arg = { |
cc02e457 | 99 | .genmask = nft_genmask_next(read_pnet(&set->pnet)), |
bfd6e327 PM |
100 | .set = set, |
101 | .key = &elem->key, | |
102 | }; | |
ce6eb0d7 | 103 | |
fe2811eb PM |
104 | return rhashtable_lookup_insert_key(&priv->ht, &arg, &he->node, |
105 | nft_hash_params); | |
96518518 PM |
106 | } |
107 | ||
cc02e457 PM |
108 | static void nft_hash_activate(const struct nft_set *set, |
109 | const struct nft_set_elem *elem) | |
96518518 | 110 | { |
cc02e457 | 111 | struct nft_hash_elem *he = elem->priv; |
ce6eb0d7 | 112 | |
cc02e457 | 113 | nft_set_elem_change_active(set, &he->ext); |
9d098292 | 114 | nft_set_elem_clear_busy(&he->ext); |
20a69341 | 115 | } |
96518518 | 116 | |
cc02e457 PM |
117 | static void *nft_hash_deactivate(const struct nft_set *set, |
118 | const struct nft_set_elem *elem) | |
20a69341 | 119 | { |
745f5450 | 120 | struct nft_hash *priv = nft_set_priv(set); |
fa377321 | 121 | struct nft_hash_elem *he; |
bfd6e327 | 122 | struct nft_hash_cmp_arg arg = { |
cc02e457 | 123 | .genmask = nft_genmask_next(read_pnet(&set->pnet)), |
bfd6e327 PM |
124 | .set = set, |
125 | .key = &elem->key, | |
126 | }; | |
fa377321 | 127 | |
9d098292 | 128 | rcu_read_lock(); |
bfd6e327 | 129 | he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params); |
9d098292 PM |
130 | if (he != NULL) { |
131 | if (!nft_set_elem_mark_busy(&he->ext)) | |
132 | nft_set_elem_change_active(set, &he->ext); | |
133 | else | |
134 | he = NULL; | |
135 | } | |
136 | rcu_read_unlock(); | |
8d24c0b4 | 137 | |
cc02e457 PM |
138 | return he; |
139 | } | |
8d24c0b4 | 140 | |
cc02e457 PM |
141 | static void nft_hash_remove(const struct nft_set *set, |
142 | const struct nft_set_elem *elem) | |
143 | { | |
144 | struct nft_hash *priv = nft_set_priv(set); | |
145 | struct nft_hash_elem *he = elem->priv; | |
146 | ||
147 | rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params); | |
96518518 PM |
148 | } |
149 | ||
20a69341 PM |
150 | static void nft_hash_walk(const struct nft_ctx *ctx, const struct nft_set *set, |
151 | struct nft_set_iter *iter) | |
96518518 | 152 | { |
745f5450 | 153 | struct nft_hash *priv = nft_set_priv(set); |
fe2811eb | 154 | struct nft_hash_elem *he; |
9a776628 | 155 | struct rhashtable_iter hti; |
20a69341 | 156 | struct nft_set_elem elem; |
cc02e457 | 157 | u8 genmask = nft_genmask_cur(read_pnet(&set->pnet)); |
9a776628 | 158 | int err; |
96518518 | 159 | |
745f5450 | 160 | err = rhashtable_walk_init(&priv->ht, &hti); |
9a776628 HX |
161 | iter->err = err; |
162 | if (err) | |
163 | return; | |
88d6ed15 | 164 | |
9a776628 HX |
165 | err = rhashtable_walk_start(&hti); |
166 | if (err && err != -EAGAIN) { | |
167 | iter->err = err; | |
168 | goto out; | |
169 | } | |
170 | ||
171 | while ((he = rhashtable_walk_next(&hti))) { | |
172 | if (IS_ERR(he)) { | |
173 | err = PTR_ERR(he); | |
174 | if (err != -EAGAIN) { | |
175 | iter->err = err; | |
176 | goto out; | |
177 | } | |
d8bdff59 HX |
178 | |
179 | continue; | |
9a776628 HX |
180 | } |
181 | ||
182 | if (iter->count < iter->skip) | |
183 | goto cont; | |
9d098292 PM |
184 | if (nft_set_elem_expired(&he->ext)) |
185 | goto cont; | |
cc02e457 PM |
186 | if (!nft_set_elem_active(&he->ext, genmask)) |
187 | goto cont; | |
20a69341 | 188 | |
fe2811eb | 189 | elem.priv = he; |
9a776628 HX |
190 | |
191 | iter->err = iter->fn(ctx, set, iter, &elem); | |
192 | if (iter->err < 0) | |
193 | goto out; | |
20a69341 | 194 | |
20a69341 | 195 | cont: |
9a776628 | 196 | iter->count++; |
96518518 | 197 | } |
9a776628 HX |
198 | |
199 | out: | |
200 | rhashtable_walk_stop(&hti); | |
201 | rhashtable_walk_exit(&hti); | |
96518518 PM |
202 | } |
203 | ||
9d098292 PM |
204 | static void nft_hash_gc(struct work_struct *work) |
205 | { | |
3dd0673a | 206 | struct nft_set *set; |
9d098292 PM |
207 | struct nft_hash_elem *he; |
208 | struct nft_hash *priv; | |
209 | struct nft_set_gc_batch *gcb = NULL; | |
210 | struct rhashtable_iter hti; | |
211 | int err; | |
212 | ||
213 | priv = container_of(work, struct nft_hash, gc_work.work); | |
214 | set = nft_set_container_of(priv); | |
215 | ||
216 | err = rhashtable_walk_init(&priv->ht, &hti); | |
217 | if (err) | |
218 | goto schedule; | |
219 | ||
220 | err = rhashtable_walk_start(&hti); | |
221 | if (err && err != -EAGAIN) | |
222 | goto out; | |
223 | ||
224 | while ((he = rhashtable_walk_next(&hti))) { | |
225 | if (IS_ERR(he)) { | |
226 | if (PTR_ERR(he) != -EAGAIN) | |
227 | goto out; | |
228 | continue; | |
229 | } | |
230 | ||
231 | if (!nft_set_elem_expired(&he->ext)) | |
232 | continue; | |
233 | if (nft_set_elem_mark_busy(&he->ext)) | |
234 | continue; | |
235 | ||
236 | gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC); | |
237 | if (gcb == NULL) | |
238 | goto out; | |
239 | rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params); | |
3dd0673a | 240 | atomic_dec(&set->nelems); |
9d098292 PM |
241 | nft_set_gc_batch_add(gcb, he); |
242 | } | |
243 | out: | |
244 | rhashtable_walk_stop(&hti); | |
245 | rhashtable_walk_exit(&hti); | |
246 | ||
247 | nft_set_gc_batch_complete(gcb); | |
248 | schedule: | |
249 | queue_delayed_work(system_power_efficient_wq, &priv->gc_work, | |
250 | nft_set_gc_interval(set)); | |
251 | } | |
252 | ||
20a69341 PM |
253 | static unsigned int nft_hash_privsize(const struct nlattr * const nla[]) |
254 | { | |
745f5450 | 255 | return sizeof(struct nft_hash); |
cfe4a9dd TG |
256 | } |
257 | ||
fa377321 | 258 | static const struct rhashtable_params nft_hash_params = { |
45d84751 | 259 | .head_offset = offsetof(struct nft_hash_elem, node), |
bfd6e327 PM |
260 | .hashfn = nft_hash_key, |
261 | .obj_hashfn = nft_hash_obj, | |
262 | .obj_cmpfn = nft_hash_cmp, | |
45d84751 | 263 | .automatic_shrinking = true, |
fa377321 HX |
264 | }; |
265 | ||
20a69341 | 266 | static int nft_hash_init(const struct nft_set *set, |
c50b960c | 267 | const struct nft_set_desc *desc, |
96518518 PM |
268 | const struct nlattr * const tb[]) |
269 | { | |
745f5450 | 270 | struct nft_hash *priv = nft_set_priv(set); |
fa377321 | 271 | struct rhashtable_params params = nft_hash_params; |
9d098292 | 272 | int err; |
fa377321 HX |
273 | |
274 | params.nelem_hint = desc->size ?: NFT_HASH_ELEMENT_HINT; | |
45d84751 | 275 | params.key_len = set->klen; |
96518518 | 276 | |
9d098292 PM |
277 | err = rhashtable_init(&priv->ht, ¶ms); |
278 | if (err < 0) | |
279 | return err; | |
280 | ||
281 | INIT_DEFERRABLE_WORK(&priv->gc_work, nft_hash_gc); | |
282 | if (set->flags & NFT_SET_TIMEOUT) | |
283 | queue_delayed_work(system_power_efficient_wq, &priv->gc_work, | |
284 | nft_set_gc_interval(set)); | |
285 | return 0; | |
96518518 PM |
286 | } |
287 | ||
61edafbb | 288 | static void nft_hash_elem_destroy(void *ptr, void *arg) |
96518518 | 289 | { |
61edafbb | 290 | nft_set_elem_destroy((const struct nft_set *)arg, ptr); |
6b6f302c | 291 | } |
97defe1e | 292 | |
6b6f302c TG |
293 | static void nft_hash_destroy(const struct nft_set *set) |
294 | { | |
745f5450 PM |
295 | struct nft_hash *priv = nft_set_priv(set); |
296 | ||
9d098292 | 297 | cancel_delayed_work_sync(&priv->gc_work); |
61edafbb PM |
298 | rhashtable_free_and_destroy(&priv->ht, nft_hash_elem_destroy, |
299 | (void *)set); | |
96518518 PM |
300 | } |
301 | ||
c50b960c PM |
302 | static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features, |
303 | struct nft_set_estimate *est) | |
304 | { | |
305 | unsigned int esize; | |
306 | ||
307 | esize = sizeof(struct nft_hash_elem); | |
c50b960c | 308 | if (desc->size) { |
745f5450 | 309 | est->size = sizeof(struct nft_hash) + |
cfe4a9dd | 310 | roundup_pow_of_two(desc->size * 4 / 3) * |
c50b960c PM |
311 | sizeof(struct nft_hash_elem *) + |
312 | desc->size * esize; | |
313 | } else { | |
314 | /* Resizing happens when the load drops below 30% or goes | |
315 | * above 75%. The average of 52.5% load (approximated by 50%) | |
316 | * is used for the size estimation of the hash buckets, | |
317 | * meaning we calculate two buckets per element. | |
318 | */ | |
319 | est->size = esize + 2 * sizeof(struct nft_hash_elem *); | |
320 | } | |
321 | ||
322 | est->class = NFT_SET_CLASS_O_1; | |
323 | ||
324 | return true; | |
325 | } | |
326 | ||
20a69341 PM |
327 | static struct nft_set_ops nft_hash_ops __read_mostly = { |
328 | .privsize = nft_hash_privsize, | |
fe2811eb | 329 | .elemsize = offsetof(struct nft_hash_elem, ext), |
c50b960c | 330 | .estimate = nft_hash_estimate, |
96518518 PM |
331 | .init = nft_hash_init, |
332 | .destroy = nft_hash_destroy, | |
20a69341 | 333 | .insert = nft_hash_insert, |
cc02e457 PM |
334 | .activate = nft_hash_activate, |
335 | .deactivate = nft_hash_deactivate, | |
20a69341 PM |
336 | .remove = nft_hash_remove, |
337 | .lookup = nft_hash_lookup, | |
338 | .walk = nft_hash_walk, | |
9d098292 | 339 | .features = NFT_SET_MAP | NFT_SET_TIMEOUT, |
20a69341 | 340 | .owner = THIS_MODULE, |
96518518 PM |
341 | }; |
342 | ||
343 | static int __init nft_hash_module_init(void) | |
344 | { | |
20a69341 | 345 | return nft_register_set(&nft_hash_ops); |
96518518 PM |
346 | } |
347 | ||
348 | static void __exit nft_hash_module_exit(void) | |
349 | { | |
20a69341 | 350 | nft_unregister_set(&nft_hash_ops); |
96518518 PM |
351 | } |
352 | ||
353 | module_init(nft_hash_module_init); | |
354 | module_exit(nft_hash_module_exit); | |
355 | ||
356 | MODULE_LICENSE("GPL"); | |
357 | MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>"); | |
20a69341 | 358 | MODULE_ALIAS_NFT_SET(); |