]>
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; | |
8cd8937a | 39 | const u32 *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 | ||
e562d860 | 65 | if (memcmp(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 | ||
42a55769 PNA |
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) | |
96518518 | 76 | { |
745f5450 | 77 | struct nft_hash *priv = nft_set_priv(set); |
20a69341 | 78 | const struct nft_hash_elem *he; |
bfd6e327 | 79 | struct nft_hash_cmp_arg arg = { |
42a55769 | 80 | .genmask = nft_genmask_cur(net), |
bfd6e327 PM |
81 | .set = set, |
82 | .key = key, | |
83 | }; | |
ce6eb0d7 | 84 | |
bfd6e327 | 85 | he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params); |
b2832dd6 PM |
86 | if (he != NULL) |
87 | *ext = &he->ext; | |
ce6eb0d7 | 88 | |
cfe4a9dd | 89 | return !!he; |
96518518 PM |
90 | } |
91 | ||
8cd8937a | 92 | static bool nft_hash_update(struct nft_set *set, const u32 *key, |
22fe54d5 PM |
93 | void *(*new)(struct nft_set *, |
94 | const struct nft_expr *, | |
a55e22e9 | 95 | struct nft_regs *regs), |
22fe54d5 | 96 | const struct nft_expr *expr, |
a55e22e9 | 97 | struct nft_regs *regs, |
22fe54d5 PM |
98 | const struct nft_set_ext **ext) |
99 | { | |
100 | struct nft_hash *priv = nft_set_priv(set); | |
dab45060 | 101 | struct nft_hash_elem *he, *prev; |
22fe54d5 PM |
102 | struct nft_hash_cmp_arg arg = { |
103 | .genmask = NFT_GENMASK_ANY, | |
104 | .set = set, | |
105 | .key = key, | |
106 | }; | |
107 | ||
108 | he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params); | |
109 | if (he != NULL) | |
110 | goto out; | |
111 | ||
a55e22e9 | 112 | he = new(set, expr, regs); |
22fe54d5 PM |
113 | if (he == NULL) |
114 | goto err1; | |
dab45060 LZ |
115 | |
116 | prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node, | |
117 | nft_hash_params); | |
118 | if (IS_ERR(prev)) | |
22fe54d5 | 119 | goto err2; |
dab45060 LZ |
120 | |
121 | /* Another cpu may race to insert the element with the same key */ | |
122 | if (prev) { | |
123 | nft_set_elem_destroy(set, he, true); | |
124 | he = prev; | |
125 | } | |
126 | ||
22fe54d5 PM |
127 | out: |
128 | *ext = &he->ext; | |
129 | return true; | |
130 | ||
131 | err2: | |
61f9e292 | 132 | nft_set_elem_destroy(set, he, true); |
22fe54d5 PM |
133 | err1: |
134 | return false; | |
135 | } | |
136 | ||
42a55769 | 137 | static int nft_hash_insert(const struct net *net, const struct nft_set *set, |
c016c7e4 PNA |
138 | const struct nft_set_elem *elem, |
139 | struct nft_set_ext **ext) | |
96518518 | 140 | { |
745f5450 | 141 | struct nft_hash *priv = nft_set_priv(set); |
fe2811eb | 142 | struct nft_hash_elem *he = elem->priv; |
bfd6e327 | 143 | struct nft_hash_cmp_arg arg = { |
42a55769 | 144 | .genmask = nft_genmask_next(net), |
bfd6e327 | 145 | .set = set, |
7d740264 | 146 | .key = elem->key.val.data, |
bfd6e327 | 147 | }; |
c016c7e4 PNA |
148 | struct nft_hash_elem *prev; |
149 | ||
150 | prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node, | |
151 | nft_hash_params); | |
152 | if (IS_ERR(prev)) | |
153 | return PTR_ERR(prev); | |
154 | if (prev) { | |
155 | *ext = &prev->ext; | |
156 | return -EEXIST; | |
157 | } | |
158 | return 0; | |
96518518 PM |
159 | } |
160 | ||
42a55769 | 161 | static void nft_hash_activate(const struct net *net, const struct nft_set *set, |
cc02e457 | 162 | const struct nft_set_elem *elem) |
96518518 | 163 | { |
cc02e457 | 164 | struct nft_hash_elem *he = elem->priv; |
ce6eb0d7 | 165 | |
42a55769 | 166 | nft_set_elem_change_active(net, set, &he->ext); |
9d098292 | 167 | nft_set_elem_clear_busy(&he->ext); |
20a69341 | 168 | } |
96518518 | 169 | |
1ba1c414 PNA |
170 | static bool nft_hash_flush(const struct net *net, |
171 | const struct nft_set *set, void *priv) | |
37df5301 PNA |
172 | { |
173 | struct nft_hash_elem *he = priv; | |
174 | ||
175 | if (!nft_set_elem_mark_busy(&he->ext) || | |
176 | !nft_is_active(net, &he->ext)) { | |
177 | nft_set_elem_change_active(net, set, &he->ext); | |
178 | return true; | |
179 | } | |
180 | return false; | |
181 | } | |
182 | ||
42a55769 PNA |
183 | static void *nft_hash_deactivate(const struct net *net, |
184 | const struct nft_set *set, | |
cc02e457 | 185 | const struct nft_set_elem *elem) |
20a69341 | 186 | { |
745f5450 | 187 | struct nft_hash *priv = nft_set_priv(set); |
fa377321 | 188 | struct nft_hash_elem *he; |
bfd6e327 | 189 | struct nft_hash_cmp_arg arg = { |
8eee54be | 190 | .genmask = nft_genmask_next(net), |
bfd6e327 | 191 | .set = set, |
7d740264 | 192 | .key = elem->key.val.data, |
bfd6e327 | 193 | }; |
fa377321 | 194 | |
9d098292 | 195 | rcu_read_lock(); |
bfd6e327 | 196 | he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params); |
37df5301 | 197 | if (he != NULL && |
1ba1c414 | 198 | !nft_hash_flush(net, set, he)) |
37df5301 PNA |
199 | he = NULL; |
200 | ||
9d098292 | 201 | rcu_read_unlock(); |
8d24c0b4 | 202 | |
cc02e457 PM |
203 | return he; |
204 | } | |
8d24c0b4 | 205 | |
5cb82a38 PNA |
206 | static void nft_hash_remove(const struct net *net, |
207 | const struct nft_set *set, | |
cc02e457 PM |
208 | const struct nft_set_elem *elem) |
209 | { | |
210 | struct nft_hash *priv = nft_set_priv(set); | |
211 | struct nft_hash_elem *he = elem->priv; | |
212 | ||
213 | rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params); | |
96518518 PM |
214 | } |
215 | ||
de70185d | 216 | static void nft_hash_walk(const struct nft_ctx *ctx, struct nft_set *set, |
20a69341 | 217 | struct nft_set_iter *iter) |
96518518 | 218 | { |
745f5450 | 219 | struct nft_hash *priv = nft_set_priv(set); |
fe2811eb | 220 | struct nft_hash_elem *he; |
9a776628 | 221 | struct rhashtable_iter hti; |
20a69341 | 222 | struct nft_set_elem elem; |
9a776628 | 223 | int err; |
96518518 | 224 | |
8f6fd83c | 225 | err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL); |
9a776628 HX |
226 | iter->err = err; |
227 | if (err) | |
228 | return; | |
88d6ed15 | 229 | |
9a776628 HX |
230 | err = rhashtable_walk_start(&hti); |
231 | if (err && err != -EAGAIN) { | |
232 | iter->err = err; | |
233 | goto out; | |
234 | } | |
235 | ||
236 | while ((he = rhashtable_walk_next(&hti))) { | |
237 | if (IS_ERR(he)) { | |
238 | err = PTR_ERR(he); | |
239 | if (err != -EAGAIN) { | |
240 | iter->err = err; | |
241 | goto out; | |
242 | } | |
d8bdff59 HX |
243 | |
244 | continue; | |
9a776628 HX |
245 | } |
246 | ||
247 | if (iter->count < iter->skip) | |
248 | goto cont; | |
9d098292 PM |
249 | if (nft_set_elem_expired(&he->ext)) |
250 | goto cont; | |
8588ac09 | 251 | if (!nft_set_elem_active(&he->ext, iter->genmask)) |
cc02e457 | 252 | goto cont; |
20a69341 | 253 | |
fe2811eb | 254 | elem.priv = he; |
9a776628 HX |
255 | |
256 | iter->err = iter->fn(ctx, set, iter, &elem); | |
257 | if (iter->err < 0) | |
258 | goto out; | |
20a69341 | 259 | |
20a69341 | 260 | cont: |
9a776628 | 261 | iter->count++; |
96518518 | 262 | } |
9a776628 HX |
263 | |
264 | out: | |
265 | rhashtable_walk_stop(&hti); | |
266 | rhashtable_walk_exit(&hti); | |
96518518 PM |
267 | } |
268 | ||
9d098292 PM |
269 | static void nft_hash_gc(struct work_struct *work) |
270 | { | |
3dd0673a | 271 | struct nft_set *set; |
9d098292 PM |
272 | struct nft_hash_elem *he; |
273 | struct nft_hash *priv; | |
274 | struct nft_set_gc_batch *gcb = NULL; | |
275 | struct rhashtable_iter hti; | |
276 | int err; | |
277 | ||
278 | priv = container_of(work, struct nft_hash, gc_work.work); | |
279 | set = nft_set_container_of(priv); | |
280 | ||
8f6fd83c | 281 | err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL); |
9d098292 PM |
282 | if (err) |
283 | goto schedule; | |
284 | ||
285 | err = rhashtable_walk_start(&hti); | |
286 | if (err && err != -EAGAIN) | |
287 | goto out; | |
288 | ||
289 | while ((he = rhashtable_walk_next(&hti))) { | |
290 | if (IS_ERR(he)) { | |
291 | if (PTR_ERR(he) != -EAGAIN) | |
292 | goto out; | |
293 | continue; | |
294 | } | |
295 | ||
296 | if (!nft_set_elem_expired(&he->ext)) | |
297 | continue; | |
298 | if (nft_set_elem_mark_busy(&he->ext)) | |
299 | continue; | |
300 | ||
301 | gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC); | |
302 | if (gcb == NULL) | |
303 | goto out; | |
304 | rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params); | |
3dd0673a | 305 | atomic_dec(&set->nelems); |
9d098292 PM |
306 | nft_set_gc_batch_add(gcb, he); |
307 | } | |
308 | out: | |
309 | rhashtable_walk_stop(&hti); | |
310 | rhashtable_walk_exit(&hti); | |
311 | ||
312 | nft_set_gc_batch_complete(gcb); | |
313 | schedule: | |
314 | queue_delayed_work(system_power_efficient_wq, &priv->gc_work, | |
315 | nft_set_gc_interval(set)); | |
316 | } | |
317 | ||
20a69341 PM |
318 | static unsigned int nft_hash_privsize(const struct nlattr * const nla[]) |
319 | { | |
745f5450 | 320 | return sizeof(struct nft_hash); |
cfe4a9dd TG |
321 | } |
322 | ||
fa377321 | 323 | static const struct rhashtable_params nft_hash_params = { |
45d84751 | 324 | .head_offset = offsetof(struct nft_hash_elem, node), |
bfd6e327 PM |
325 | .hashfn = nft_hash_key, |
326 | .obj_hashfn = nft_hash_obj, | |
327 | .obj_cmpfn = nft_hash_cmp, | |
45d84751 | 328 | .automatic_shrinking = true, |
fa377321 HX |
329 | }; |
330 | ||
20a69341 | 331 | static int nft_hash_init(const struct nft_set *set, |
c50b960c | 332 | const struct nft_set_desc *desc, |
96518518 PM |
333 | const struct nlattr * const tb[]) |
334 | { | |
745f5450 | 335 | struct nft_hash *priv = nft_set_priv(set); |
fa377321 | 336 | struct rhashtable_params params = nft_hash_params; |
9d098292 | 337 | int err; |
fa377321 HX |
338 | |
339 | params.nelem_hint = desc->size ?: NFT_HASH_ELEMENT_HINT; | |
45d84751 | 340 | params.key_len = set->klen; |
96518518 | 341 | |
9d098292 PM |
342 | err = rhashtable_init(&priv->ht, ¶ms); |
343 | if (err < 0) | |
344 | return err; | |
345 | ||
346 | INIT_DEFERRABLE_WORK(&priv->gc_work, nft_hash_gc); | |
347 | if (set->flags & NFT_SET_TIMEOUT) | |
348 | queue_delayed_work(system_power_efficient_wq, &priv->gc_work, | |
349 | nft_set_gc_interval(set)); | |
350 | return 0; | |
96518518 PM |
351 | } |
352 | ||
61edafbb | 353 | static void nft_hash_elem_destroy(void *ptr, void *arg) |
96518518 | 354 | { |
61f9e292 | 355 | nft_set_elem_destroy((const struct nft_set *)arg, ptr, true); |
6b6f302c | 356 | } |
97defe1e | 357 | |
6b6f302c TG |
358 | static void nft_hash_destroy(const struct nft_set *set) |
359 | { | |
745f5450 PM |
360 | struct nft_hash *priv = nft_set_priv(set); |
361 | ||
9d098292 | 362 | cancel_delayed_work_sync(&priv->gc_work); |
61edafbb PM |
363 | rhashtable_free_and_destroy(&priv->ht, nft_hash_elem_destroy, |
364 | (void *)set); | |
96518518 PM |
365 | } |
366 | ||
c50b960c PM |
367 | static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features, |
368 | struct nft_set_estimate *est) | |
369 | { | |
370 | unsigned int esize; | |
371 | ||
372 | esize = sizeof(struct nft_hash_elem); | |
c50b960c | 373 | if (desc->size) { |
745f5450 | 374 | est->size = sizeof(struct nft_hash) + |
cfe4a9dd | 375 | roundup_pow_of_two(desc->size * 4 / 3) * |
c50b960c PM |
376 | sizeof(struct nft_hash_elem *) + |
377 | desc->size * esize; | |
378 | } else { | |
379 | /* Resizing happens when the load drops below 30% or goes | |
380 | * above 75%. The average of 52.5% load (approximated by 50%) | |
381 | * is used for the size estimation of the hash buckets, | |
382 | * meaning we calculate two buckets per element. | |
383 | */ | |
384 | est->size = esize + 2 * sizeof(struct nft_hash_elem *); | |
385 | } | |
386 | ||
387 | est->class = NFT_SET_CLASS_O_1; | |
388 | ||
389 | return true; | |
390 | } | |
391 | ||
20a69341 PM |
392 | static struct nft_set_ops nft_hash_ops __read_mostly = { |
393 | .privsize = nft_hash_privsize, | |
fe2811eb | 394 | .elemsize = offsetof(struct nft_hash_elem, ext), |
c50b960c | 395 | .estimate = nft_hash_estimate, |
96518518 PM |
396 | .init = nft_hash_init, |
397 | .destroy = nft_hash_destroy, | |
20a69341 | 398 | .insert = nft_hash_insert, |
cc02e457 PM |
399 | .activate = nft_hash_activate, |
400 | .deactivate = nft_hash_deactivate, | |
1ba1c414 | 401 | .flush = nft_hash_flush, |
20a69341 PM |
402 | .remove = nft_hash_remove, |
403 | .lookup = nft_hash_lookup, | |
22fe54d5 | 404 | .update = nft_hash_update, |
20a69341 | 405 | .walk = nft_hash_walk, |
9d098292 | 406 | .features = NFT_SET_MAP | NFT_SET_TIMEOUT, |
20a69341 | 407 | .owner = THIS_MODULE, |
96518518 PM |
408 | }; |
409 | ||
410 | static int __init nft_hash_module_init(void) | |
411 | { | |
20a69341 | 412 | return nft_register_set(&nft_hash_ops); |
96518518 PM |
413 | } |
414 | ||
415 | static void __exit nft_hash_module_exit(void) | |
416 | { | |
20a69341 | 417 | nft_unregister_set(&nft_hash_ops); |
96518518 PM |
418 | } |
419 | ||
420 | module_init(nft_hash_module_init); | |
421 | module_exit(nft_hash_module_exit); | |
422 | ||
423 | MODULE_LICENSE("GPL"); | |
424 | MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>"); | |
20a69341 | 425 | MODULE_ALIAS_NFT_SET(); |