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;
 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                            struct nft_set_ext **ext)
 131 {
 132         struct nft_hash *priv = nft_set_priv(set);
 133         struct nft_hash_elem *he = elem->priv;
 134         struct nft_hash_cmp_arg arg = {
 135                 .genmask = nft_genmask_next(net),
 136                 .set     = set,
 137                 .key     = elem->key.val.data,
 138         };
 139         struct nft_hash_elem *prev;
 140
 141         prev = rhashtable_lookup_get_insert_key(&priv->ht, &arg, &he->node,
 142                                                nft_hash_params);
 143         if (IS_ERR(prev))
 144                 return PTR_ERR(prev);
 145         if (prev) {
 146                 *ext = &prev->ext;
 147                 return -EEXIST;
 148         }
 149         return 0;
 150 }
 151
 152 static void nft_hash_activate(const struct net *net, const struct nft_set *set,
 153                               const struct nft_set_elem *elem)
 154 {
 155         struct nft_hash_elem *he = elem->priv;
 156
 157         nft_set_elem_change_active(net, set, &he->ext);
 158         nft_set_elem_clear_busy(&he->ext);
 159 }
 160
 161 static void *nft_hash_deactivate(const struct net *net,
 162                                  const struct nft_set *set,
 163                                  const struct nft_set_elem *elem)
 164 {
 165         struct nft_hash *priv = nft_set_priv(set);
 166         struct nft_hash_elem *he;
 167         struct nft_hash_cmp_arg arg = {
 168                 .genmask = nft_genmask_next(net),
 169                 .set     = set,
 170                 .key     = elem->key.val.data,
 171         };
 172
 173         rcu_read_lock();
 174         he = rhashtable_lookup_fast(&priv->ht, &arg, nft_hash_params);
 175         if (he != NULL) {
 176                 if (!nft_set_elem_mark_busy(&he->ext) ||
 177                     !nft_is_active(net, &he->ext))
 178                         nft_set_elem_change_active(net, set, &he->ext);
 179                 else
 180                         he = NULL;
 181         }
 182         rcu_read_unlock();
 183
 184         return he;
 185 }
 186
 187 static void nft_hash_remove(const struct nft_set *set,
 188                             const struct nft_set_elem *elem)
 189 {
 190         struct nft_hash *priv = nft_set_priv(set);
 191         struct nft_hash_elem *he = elem->priv;
 192
 193         rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
 194 }
 195
 196 static void nft_hash_walk(const struct nft_ctx *ctx, const struct nft_set *set,
 197                           struct nft_set_iter *iter)
 198 {
 199         struct nft_hash *priv = nft_set_priv(set);
 200         struct nft_hash_elem *he;
 201         struct rhashtable_iter hti;
 202         struct nft_set_elem elem;
 203         int err;
 204
 205         err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
 206         iter->err = err;
 207         if (err)
 208                 return;
 209
 210         err = rhashtable_walk_start(&hti);
 211         if (err && err != -EAGAIN) {
 212                 iter->err = err;
 213                 goto out;
 214         }
 215
 216         while ((he = rhashtable_walk_next(&hti))) {
 217                 if (IS_ERR(he)) {
 218                         err = PTR_ERR(he);
 219                         if (err != -EAGAIN) {
 220                                 iter->err = err;
 221                                 goto out;
 222                         }
 223
 224                         continue;
 225                 }
 226
 227                 if (iter->count < iter->skip)
 228                         goto cont;
 229                 if (nft_set_elem_expired(&he->ext))
 230                         goto cont;
 231                 if (!nft_set_elem_active(&he->ext, iter->genmask))
 232                         goto cont;
 233
 234                 elem.priv = he;
 235
 236                 iter->err = iter->fn(ctx, set, iter, &elem);
 237                 if (iter->err < 0)
 238                         goto out;
 239
 240 cont:
 241                 iter->count++;
 242         }
 243
 244 out:
 245         rhashtable_walk_stop(&hti);
 246         rhashtable_walk_exit(&hti);
 247 }
 248
 249 static void nft_hash_gc(struct work_struct *work)
 250 {
 251         struct nft_set *set;
 252         struct nft_hash_elem *he;
 253         struct nft_hash *priv;
 254         struct nft_set_gc_batch *gcb = NULL;
 255         struct rhashtable_iter hti;
 256         int err;
 257
 258         priv = container_of(work, struct nft_hash, gc_work.work);
 259         set  = nft_set_container_of(priv);
 260
 261         err = rhashtable_walk_init(&priv->ht, &hti, GFP_KERNEL);
 262         if (err)
 263                 goto schedule;
 264
 265         err = rhashtable_walk_start(&hti);
 266         if (err && err != -EAGAIN)
 267                 goto out;
 268
 269         while ((he = rhashtable_walk_next(&hti))) {
 270                 if (IS_ERR(he)) {
 271                         if (PTR_ERR(he) != -EAGAIN)
 272                                 goto out;
 273                         continue;
 274                 }
 275
 276                 if (!nft_set_elem_expired(&he->ext))
 277                         continue;
 278                 if (nft_set_elem_mark_busy(&he->ext))
 279                         continue;
 280
 281                 gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
 282                 if (gcb == NULL)
 283                         goto out;
 284                 rhashtable_remove_fast(&priv->ht, &he->node, nft_hash_params);
 285                 atomic_dec(&set->nelems);
 286                 nft_set_gc_batch_add(gcb, he);
 287         }
 288 out:
 289         rhashtable_walk_stop(&hti);
 290         rhashtable_walk_exit(&hti);
 291
 292         nft_set_gc_batch_complete(gcb);
 293 schedule:
 294         queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
 295                            nft_set_gc_interval(set));
 296 }
 297
 298 static unsigned int nft_hash_privsize(const struct nlattr * const nla[])
 299 {
 300         return sizeof(struct nft_hash);
 301 }
 302
 303 static const struct rhashtable_params nft_hash_params = {
 304         .head_offset            = offsetof(struct nft_hash_elem, node),
 305         .hashfn                 = nft_hash_key,
 306         .obj_hashfn             = nft_hash_obj,
 307         .obj_cmpfn              = nft_hash_cmp,
 308         .automatic_shrinking    = true,
 309 };
 310
 311 static int nft_hash_init(const struct nft_set *set,
 312                          const struct nft_set_desc *desc,
 313                          const struct nlattr * const tb[])
 314 {
 315         struct nft_hash *priv = nft_set_priv(set);
 316         struct rhashtable_params params = nft_hash_params;
 317         int err;
 318
 319         params.nelem_hint = desc->size ?: NFT_HASH_ELEMENT_HINT;
 320         params.key_len    = set->klen;
 321
 322         err = rhashtable_init(&priv->ht, &params);
 323         if (err < 0)
 324                 return err;
 325
 326         INIT_DEFERRABLE_WORK(&priv->gc_work, nft_hash_gc);
 327         if (set->flags & NFT_SET_TIMEOUT)
 328                 queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
 329                                    nft_set_gc_interval(set));
 330         return 0;
 331 }
 332
 333 static void nft_hash_elem_destroy(void *ptr, void *arg)
 334 {
 335         nft_set_elem_destroy((const struct nft_set *)arg, ptr);
 336 }
 337
 338 static void nft_hash_destroy(const struct nft_set *set)
 339 {
 340         struct nft_hash *priv = nft_set_priv(set);
 341
 342         cancel_delayed_work_sync(&priv->gc_work);
 343         rhashtable_free_and_destroy(&priv->ht, nft_hash_elem_destroy,
 344                                     (void *)set);
 345 }
 346
 347 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
 348                               struct nft_set_estimate *est)
 349 {
 350         unsigned int esize;
 351
 352         esize = sizeof(struct nft_hash_elem);
 353         if (desc->size) {
 354                 est->size = sizeof(struct nft_hash) +
 355                             roundup_pow_of_two(desc->size * 4 / 3) *
 356                             sizeof(struct nft_hash_elem *) +
 357                             desc->size * esize;
 358         } else {
 359                 /* Resizing happens when the load drops below 30% or goes
 360                  * above 75%. The average of 52.5% load (approximated by 50%)
 361                  * is used for the size estimation of the hash buckets,
 362                  * meaning we calculate two buckets per element.
 363                  */
 364                 est->size = esize + 2 * sizeof(struct nft_hash_elem *);
 365         }
 366
 367         est->class = NFT_SET_CLASS_O_1;
 368
 369         return true;
 370 }
 371
 372 static struct nft_set_ops nft_hash_ops __read_mostly = {
 373         .privsize       = nft_hash_privsize,
 374         .elemsize       = offsetof(struct nft_hash_elem, ext),
 375         .estimate       = nft_hash_estimate,
 376         .init           = nft_hash_init,
 377         .destroy        = nft_hash_destroy,
 378         .insert         = nft_hash_insert,
 379         .activate       = nft_hash_activate,
 380         .deactivate     = nft_hash_deactivate,
 381         .remove         = nft_hash_remove,
 382         .lookup         = nft_hash_lookup,
 383         .update         = nft_hash_update,
 384         .walk           = nft_hash_walk,
 385         .features       = NFT_SET_MAP | NFT_SET_TIMEOUT,
 386         .owner          = THIS_MODULE,
 387 };
 388
 389 static int __init nft_hash_module_init(void)
 390 {
 391         return nft_register_set(&nft_hash_ops);
 392 }
 393
 394 static void __exit nft_hash_module_exit(void)
 395 {
 396         nft_unregister_set(&nft_hash_ops);
 397 }
 398
 399 module_init(nft_hash_module_init);
 400 module_exit(nft_hash_module_exit);
 401
 402 MODULE_LICENSE("GPL");
 403 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 404 MODULE_ALIAS_NFT_SET();