1 /*
2 * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
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>
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>
23
24 /* We target a hash table size of 4, element hint is 75% of final size */
25 #define NFT_HASH_ELEMENT_HINT 3
26
27 struct nft_hash {
28 struct rhashtable ht;
29 struct delayed_work gc_work;
30 };
31
32 struct nft_hash_elem {
33 struct rhash_head node;
34 struct nft_set_ext ext;
35 };
36
37 struct nft_hash_cmp_arg {
38 const struct nft_set *set;
39 const u32 *key;
40 u8 genmask;
41 };
42
43 static const struct rhashtable_params nft_hash_params;
44
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
56 return jhash(nft_set_ext_key(&he->ext), len, seed);
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
65 if (memcmp(nft_set_ext_key(&he->ext), x->key, x->set->klen))
66 return 1;
67 if (nft_set_elem_expired(&he->ext))
68 return 1;
69 if (!nft_set_elem_active(&he->ext, x->genmask))
70 return 1;
71 return 0;
72 }
73
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)
76 {
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),
81 .set = set,
82 .key = key,
83 };
84
85 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params);
86 if (he != NULL)
87 *ext = &he->ext;
88
89 return !!he;
90 }
91
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)
99 {
100 struct nft_hash *priv = nft_set_priv(set);
101 struct nft_hash_elem *he;
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
112 he = new(set, expr, regs);
113 if (he == NULL)
114 goto err1;
115 if (rhashtable_lookup_insert_key(&priv->ht, &arg, &he->node,
116 nft_hash_params))
117 goto err2;
118 out:
119 *ext = &he->ext;
120 return true;
121
122 err2:
123 nft_set_elem_destroy(set, he);
124 err1:
125 return false;
126 }
127
128 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
129 const struct nft_set_elem *elem)
130 {
131 struct nft_hash *priv = nft_set_priv(set);
132 struct nft_hash_elem *he = elem->priv;
133 struct nft_hash_cmp_arg arg = {
134 .genmask = nft_genmask_next(net),
135 .set = set,
136 .key = elem->key.val.data,
137 };
138
139 return rhashtable_lookup_insert_key(&priv->ht, &arg, &he->node,
140 nft_hash_params);
141 }
142
143 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
144 const struct nft_set_elem *elem)
145 {
146 struct nft_hash_elem *he = elem->priv;
147
148 nft_set_elem_change_active(net, set, &he->ext);
149 nft_set_elem_clear_busy(&he->ext);
150 }
151
152 static void *nft_hash_deactivate(const struct net *net,
153 const struct nft_set *set,
154 const struct nft_set_elem *elem)
155 {
156 struct nft_hash *priv = nft_set_priv(set);
157 struct nft_hash_elem *he;
158 struct nft_hash_cmp_arg arg = {
159 .genmask = nft_genmask_next(net),
160 .set = set,
161 .key = elem->key.val.data,
162 };
163
164 rcu_read_lock();
165 he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params);
166 if (he != NULL) {
167 if (!nft_set_elem_mark_busy(&he->ext) ||
168 !nft_is_active(net, &he->ext))
169 nft_set_elem_change_active(net, set, &he->ext);
170 else
171 he = NULL;
172 }
173 rcu_read_unlock();
174
175 return he;
176 }
177
178 static void nft_hash_remove(const struct nft_set *set,
179 const struct nft_set_elem *elem)
180 {
181 struct nft_hash *priv = nft_set_priv(set);
182 struct nft_hash_elem *he = elem->priv;
183
184 rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
185 }
186
187 static void nft_hash_walk(const struct nft_ctx *ctx, const struct nft_set *set,
188 struct nft_set_iter *iter)
189 {
190 struct nft_hash *priv = nft_set_priv(set);
191 struct nft_hash_elem *he;
192 struct rhashtable_iter hti;
193 struct nft_set_elem elem;
194 int err;
195
196 err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
197 iter->err = err;
198 if (err)
199 return;
200
201 err = rhashtable_walk_start(&hti);
202 if (err && err != -EAGAIN) {
203 iter->err = err;
204 goto out;
205 }
206
207 while ((he = rhashtable_walk_next(&hti))) {
208 if (IS_ERR(he)) {
209 err = PTR_ERR(he);
210 if (err != -EAGAIN) {
211 iter->err = err;
212 goto out;
213 }
214
215 continue;
216 }
217
218 if (iter->count < iter->skip)
219 goto cont;
220 if (nft_set_elem_expired(&he->ext))
221 goto cont;
222 if (!nft_set_elem_active(&he->ext, iter->genmask))
223 goto cont;
224
225 elem.priv = he;
226
227 iter->err = iter->fn(ctx, set, iter, &elem);
228 if (iter->err < 0)
229 goto out;
230
231 cont:
232 iter->count++;
233 }
234
235 out:
236 rhashtable_walk_stop(&hti);
237 rhashtable_walk_exit(&hti);
238 }
239
240 static void nft_hash_gc(struct work_struct *work)
241 {
242 struct nft_set *set;
243 struct nft_hash_elem *he;
244 struct nft_hash *priv;
245 struct nft_set_gc_batch *gcb = NULL;
246 struct rhashtable_iter hti;
247 int err;
248
249 priv = container_of(work, struct nft_hash, gc_work.work);
250 set = nft_set_container_of(priv);
251
252 err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
253 if (err)
254 goto schedule;
255
256 err = rhashtable_walk_start(&hti);
257 if (err && err != -EAGAIN)
258 goto out;
259
260 while ((he = rhashtable_walk_next(&hti))) {
261 if (IS_ERR(he)) {
262 if (PTR_ERR(he) != -EAGAIN)
263 goto out;
264 continue;
265 }
266
267 if (!nft_set_elem_expired(&he->ext))
268 continue;
269 if (nft_set_elem_mark_busy(&he->ext))
270 continue;
271
272 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
273 if (gcb == NULL)
274 goto out;
275 rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
276 atomic_dec(&set->nelems);
277 nft_set_gc_batch_add(gcb, he);
278 }
279 out:
280 rhashtable_walk_stop(&hti);
281 rhashtable_walk_exit(&hti);
282
283 nft_set_gc_batch_complete(gcb);
284 schedule:
285 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
286 nft_set_gc_interval(set));
287 }
288
289 static unsigned int nft_hash_privsize(const struct nlattr * const nla[])
290 {
291 return sizeof(struct nft_hash);
292 }
293
294 static const struct rhashtable_params nft_hash_params = {
295 .head_offset = offsetof(struct nft_hash_elem, node),
296 .hashfn = nft_hash_key,
297 .obj_hashfn = nft_hash_obj,
298 .obj_cmpfn = nft_hash_cmp,
299 .automatic_shrinking = true,
300 };
301
302 static int nft_hash_init(const struct nft_set *set,
303 const struct nft_set_desc *desc,
304 const struct nlattr * const tb[])
305 {
306 struct nft_hash *priv = nft_set_priv(set);
307 struct rhashtable_params params = nft_hash_params;
308 int err;
309
310 params.nelem_hint = desc->size ?: NFT_HASH_ELEMENT_HINT;
311 params.key_len = set->klen;
312
313 err = rhashtable_init(&priv->ht, ¶ms);
314 if (err < 0)
315 return err;
316
317 INIT_DEFERRABLE_WORK(&priv->gc_work, nft_hash_gc);
318 if (set->flags & NFT_SET_TIMEOUT)
319 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
320 nft_set_gc_interval(set));
321 return 0;
322 }
323
324 static void nft_hash_elem_destroy(void *ptr, void *arg)
325 {
326 nft_set_elem_destroy((const struct nft_set *)arg, ptr);
327 }
328
329 static void nft_hash_destroy(const struct nft_set *set)
330 {
331 struct nft_hash *priv = nft_set_priv(set);
332
333 cancel_delayed_work_sync(&priv->gc_work);
334 rhashtable_free_and_destroy(&priv->ht, nft_hash_elem_destroy,
335 (void *)set);
336 }
337
338 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
339 struct nft_set_estimate *est)
340 {
341 unsigned int esize;
342
343 esize = sizeof(struct nft_hash_elem);
344 if (desc->size) {
345 est->size = sizeof(struct nft_hash) +
346 roundup_pow_of_two(desc->size * 4 / 3) *
347 sizeof(struct nft_hash_elem *) +
348 desc->size * esize;
349 } else {
350 /* Resizing happens when the load drops below 30% or goes
351 * above 75%. The average of 52.5% load (approximated by 50%)
352 * is used for the size estimation of the hash buckets,
353 * meaning we calculate two buckets per element.
354 */
355 est->size = esize + 2 * sizeof(struct nft_hash_elem *);
356 }
357
358 est->class = NFT_SET_CLASS_O_1;
359
360 return true;
361 }
362
363 static struct nft_set_ops nft_hash_ops __read_mostly = {
364 .privsize = nft_hash_privsize,
365 .elemsize = offsetof(struct nft_hash_elem, ext),
366 .estimate = nft_hash_estimate,
367 .init = nft_hash_init,
368 .destroy = nft_hash_destroy,
369 .insert = nft_hash_insert,
370 .activate = nft_hash_activate,
371 .deactivate = nft_hash_deactivate,
372 .remove = nft_hash_remove,
373 .lookup = nft_hash_lookup,
374 .update = nft_hash_update,
375 .walk = nft_hash_walk,
376 .features = NFT_SET_MAP | NFT_SET_TIMEOUT,
377 .owner = THIS_MODULE,
378 };
379
380 static int __init nft_hash_module_init(void)
381 {
382 return nft_register_set(&nft_hash_ops);
383 }
384
385 static void __exit nft_hash_module_exit(void)
386 {
387 nft_unregister_set(&nft_hash_ops);
388 }
389
390 module_init(nft_hash_module_init);
391 module_exit(nft_hash_module_exit);
392
393 MODULE_LICENSE("GPL");
394 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
395 MODULE_ALIAS_NFT_SET();
396
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.