net/netfilter/nft_set_hash.c

   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, *prev;
 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
 116         prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
 117                                                 nft_hash_params);
 118         if (IS_ERR(prev))
 119                 goto err2;
 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
 127 out:
 128         *ext = &he->ext;
 129         return true;
 130
 131 err2:
 132         nft_set_elem_destroy(set, he, true);
 133 err1:
 134         return false;
 135 }
 136
 137 static int nft_hash_insert(const struct net *net, const struct nft_set *set,
 138                            const struct nft_set_elem *elem,
 139                            struct nft_set_ext **ext)
 140 {
 141         struct nft_hash *priv = nft_set_priv(set);
 142         struct nft_hash_elem *he = elem->priv;
 143         struct nft_hash_cmp_arg arg = {
 144                 .genmask = nft_genmask_next(net),
 145                 .set     = set,
 146                 .key     = elem->key.val.data,
 147         };
 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;
 159 }
 160
 161 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
 162                               const struct nft_set_elem *elem)
 163 {
 164         struct nft_hash_elem *he = elem->priv;
 165
 166         nft_set_elem_change_active(net, set, &he->ext);
 167         nft_set_elem_clear_busy(&he->ext);
 168 }
 169
 170 static bool nft_hash_deactivate_one(const struct net *net,
 171                                     const struct nft_set *set, void *priv)
 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
 183 static void *nft_hash_deactivate(const struct net *net,
 184                                  const struct nft_set *set,
 185                                  const struct nft_set_elem *elem)
 186 {
 187         struct nft_hash *priv = nft_set_priv(set);
 188         struct nft_hash_elem *he;
 189         struct nft_hash_cmp_arg arg = {
 190                 .genmask = nft_genmask_next(net),
 191                 .set     = set,
 192                 .key     = elem->key.val.data,
 193         };
 194
 195         rcu_read_lock();
 196         he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params);
 197         if (he != NULL &&
 198             !nft_hash_deactivate_one(net, set, he))
 199                 he = NULL;
 200
 201         rcu_read_unlock();
 202
 203         return he;
 204 }
 205
 206 static void nft_hash_remove(const struct nft_set *set,
 207                             const struct nft_set_elem *elem)
 208 {
 209         struct nft_hash *priv = nft_set_priv(set);
 210         struct nft_hash_elem *he = elem->priv;
 211
 212         rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
 213 }
 214
 215 static void nft_hash_walk(const struct nft_ctx *ctx, struct nft_set *set,
 216                           struct nft_set_iter *iter)
 217 {
 218         struct nft_hash *priv = nft_set_priv(set);
 219         struct nft_hash_elem *he;
 220         struct rhashtable_iter hti;
 221         struct nft_set_elem elem;
 222         int err;
 223
 224         err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
 225         iter->err = err;
 226         if (err)
 227                 return;
 228
 229         err = rhashtable_walk_start(&hti);
 230         if (err && err != -EAGAIN) {
 231                 iter->err = err;
 232                 goto out;
 233         }
 234
 235         while ((he = rhashtable_walk_next(&hti))) {
 236                 if (IS_ERR(he)) {
 237                         err = PTR_ERR(he);
 238                         if (err != -EAGAIN) {
 239                                 iter->err = err;
 240                                 goto out;
 241                         }
 242
 243                         continue;
 244                 }
 245
 246                 if (iter->count < iter->skip)
 247                         goto cont;
 248                 if (nft_set_elem_expired(&he->ext))
 249                         goto cont;
 250                 if (!nft_set_elem_active(&he->ext, iter->genmask))
 251                         goto cont;
 252
 253                 elem.priv = he;
 254
 255                 iter->err = iter->fn(ctx, set, iter, &elem);
 256                 if (iter->err < 0)
 257                         goto out;
 258
 259 cont:
 260                 iter->count++;
 261         }
 262
 263 out:
 264         rhashtable_walk_stop(&hti);
 265         rhashtable_walk_exit(&hti);
 266 }
 267
 268 static void nft_hash_gc(struct work_struct *work)
 269 {
 270         struct nft_set *set;
 271         struct nft_hash_elem *he;
 272         struct nft_hash *priv;
 273         struct nft_set_gc_batch *gcb = NULL;
 274         struct rhashtable_iter hti;
 275         int err;
 276
 277         priv = container_of(work, struct nft_hash, gc_work.work);
 278         set  = nft_set_container_of(priv);
 279
 280         err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
 281         if (err)
 282                 goto schedule;
 283
 284         err = rhashtable_walk_start(&hti);
 285         if (err && err != -EAGAIN)
 286                 goto out;
 287
 288         while ((he = rhashtable_walk_next(&hti))) {
 289                 if (IS_ERR(he)) {
 290                         if (PTR_ERR(he) != -EAGAIN)
 291                                 goto out;
 292                         continue;
 293                 }
 294
 295                 if (!nft_set_elem_expired(&he->ext))
 296                         continue;
 297                 if (nft_set_elem_mark_busy(&he->ext))
 298                         continue;
 299
 300                 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
 301                 if (gcb == NULL)
 302                         goto out;
 303                 rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
 304                 atomic_dec(&set->nelems);
 305                 nft_set_gc_batch_add(gcb, he);
 306         }
 307 out:
 308         rhashtable_walk_stop(&hti);
 309         rhashtable_walk_exit(&hti);
 310
 311         nft_set_gc_batch_complete(gcb);
 312 schedule:
 313         queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
 314                            nft_set_gc_interval(set));
 315 }
 316
 317 static unsigned int nft_hash_privsize(const struct nlattr * const nla[])
 318 {
 319         return sizeof(struct nft_hash);
 320 }
 321
 322 static const struct rhashtable_params nft_hash_params = {
 323         .head_offset            = offsetof(struct nft_hash_elem, node),
 324         .hashfn                 = nft_hash_key,
 325         .obj_hashfn             = nft_hash_obj,
 326         .obj_cmpfn              = nft_hash_cmp,
 327         .automatic_shrinking    = true,
 328 };
 329
 330 static int nft_hash_init(const struct nft_set *set,
 331                          const struct nft_set_desc *desc,
 332                          const struct nlattr * const tb[])
 333 {
 334         struct nft_hash *priv = nft_set_priv(set);
 335         struct rhashtable_params params = nft_hash_params;
 336         int err;
 337
 338         params.nelem_hint = desc->size ?: NFT_HASH_ELEMENT_HINT;
 339         params.key_len    = set->klen;
 340
 341         err = rhashtable_init(&priv->ht, &params);
 342         if (err < 0)
 343                 return err;
 344
 345         INIT_DEFERRABLE_WORK(&priv->gc_work, nft_hash_gc);
 346         if (set->flags & NFT_SET_TIMEOUT)
 347                 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
 348                                    nft_set_gc_interval(set));
 349         return 0;
 350 }
 351
 352 static void nft_hash_elem_destroy(void *ptr, void *arg)
 353 {
 354         nft_set_elem_destroy((const struct nft_set *)arg, ptr, true);
 355 }
 356
 357 static void nft_hash_destroy(const struct nft_set *set)
 358 {
 359         struct nft_hash *priv = nft_set_priv(set);
 360
 361         cancel_delayed_work_sync(&priv->gc_work);
 362         rhashtable_free_and_destroy(&priv->ht, nft_hash_elem_destroy,
 363                                     (void *)set);
 364 }
 365
 366 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
 367                               struct nft_set_estimate *est)
 368 {
 369         unsigned int esize;
 370
 371         esize = sizeof(struct nft_hash_elem);
 372         if (desc->size) {
 373                 est->size = sizeof(struct nft_hash) +
 374                             roundup_pow_of_two(desc->size * 4 / 3) *
 375                             sizeof(struct nft_hash_elem *) +
 376                             desc->size * esize;
 377         } else {
 378                 /* Resizing happens when the load drops below 30% or goes
 379                  * above 75%. The average of 52.5% load (approximated by 50%)
 380                  * is used for the size estimation of the hash buckets,
 381                  * meaning we calculate two buckets per element.
 382                  */
 383                 est->size = esize + 2 * sizeof(struct nft_hash_elem *);
 384         }
 385
 386         est->class = NFT_SET_CLASS_O_1;
 387
 388         return true;
 389 }
 390
 391 static struct nft_set_ops nft_hash_ops __read_mostly = {
 392         .privsize       = nft_hash_privsize,
 393         .elemsize       = offsetof(struct nft_hash_elem, ext),
 394         .estimate       = nft_hash_estimate,
 395         .init           = nft_hash_init,
 396         .destroy        = nft_hash_destroy,
 397         .insert         = nft_hash_insert,
 398         .activate       = nft_hash_activate,
 399         .deactivate     = nft_hash_deactivate,
 400         .deactivate_one = nft_hash_deactivate_one,
 401         .remove         = nft_hash_remove,
 402         .lookup         = nft_hash_lookup,
 403         .update         = nft_hash_update,
 404         .walk           = nft_hash_walk,
 405         .features       = NFT_SET_MAP | NFT_SET_TIMEOUT,
 406         .owner          = THIS_MODULE,
 407 };
 408
 409 static int __init nft_hash_module_init(void)
 410 {
 411         return nft_register_set(&nft_hash_ops);
 412 }
 413
 414 static void __exit nft_hash_module_exit(void)
 415 {
 416         nft_unregister_set(&nft_hash_ops);
 417 }
 418
 419 module_init(nft_hash_module_init);
 420 module_exit(nft_hash_module_exit);
 421
 422 MODULE_LICENSE("GPL");
 423 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 424 MODULE_ALIAS_NFT_SET();