net/netfilter/nf_flow_table_core.c

   1 #include <linux/kernel.h>
   2 #include <linux/init.h>
   3 #include <linux/module.h>
   4 #include <linux/netfilter.h>
   5 #include <linux/rhashtable.h>
   6 #include <linux/netdevice.h>
   7 #include <net/ip.h>
   8 #include <net/ip6_route.h>
   9 #include <net/netfilter/nf_tables.h>
  10 #include <net/netfilter/nf_flow_table.h>
  11 #include <net/netfilter/nf_conntrack.h>
  12 #include <net/netfilter/nf_conntrack_core.h>
  13 #include <net/netfilter/nf_conntrack_tuple.h>
  14
  15 struct flow_offload_entry {
  16         struct flow_offload     flow;
  17         struct nf_conn          *ct;
  18         struct rcu_head         rcu_head;
  19 };
  20
  21 static DEFINE_MUTEX(flowtable_lock);
  22 static LIST_HEAD(flowtables);
  23
  24 static void
  25 flow_offload_fill_dir(struct flow_offload *flow, struct nf_conn *ct,
  26                       struct nf_flow_route *route,
  27                       enum flow_offload_tuple_dir dir)
  28 {
  29         struct flow_offload_tuple *ft = &flow->tuplehash[dir].tuple;
  30         struct nf_conntrack_tuple *ctt = &ct->tuplehash[dir].tuple;
  31         struct dst_entry *dst = route->tuple[dir].dst;
  32
  33         ft->dir = dir;
  34
  35         switch (ctt->src.l3num) {
  36         case NFPROTO_IPV4:
  37                 ft->src_v4 = ctt->src.u3.in;
  38                 ft->dst_v4 = ctt->dst.u3.in;
  39                 ft->mtu = ip_dst_mtu_maybe_forward(dst, true);
  40                 break;
  41         case NFPROTO_IPV6:
  42                 ft->src_v6 = ctt->src.u3.in6;
  43                 ft->dst_v6 = ctt->dst.u3.in6;
  44                 ft->mtu = ip6_dst_mtu_forward(dst);
  45                 break;
  46         }
  47
  48         ft->l3proto = ctt->src.l3num;
  49         ft->l4proto = ctt->dst.protonum;
  50         ft->src_port = ctt->src.u.tcp.port;
  51         ft->dst_port = ctt->dst.u.tcp.port;
  52
  53         ft->iifidx = route->tuple[dir].ifindex;
  54         ft->oifidx = route->tuple[!dir].ifindex;
  55         ft->dst_cache = dst;
  56 }
  57
  58 struct flow_offload *
  59 flow_offload_alloc(struct nf_conn *ct, struct nf_flow_route *route)
  60 {
  61         struct flow_offload_entry *entry;
  62         struct flow_offload *flow;
  63
  64         if (unlikely(nf_ct_is_dying(ct) ||
  65             !atomic_inc_not_zero(&ct->ct_general.use)))
  66                 return NULL;
  67
  68         entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
  69         if (!entry)
  70                 goto err_ct_refcnt;
  71
  72         flow = &entry->flow;
  73
  74         if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst))
  75                 goto err_dst_cache_original;
  76
  77         if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_REPLY].dst))
  78                 goto err_dst_cache_reply;
  79
  80         entry->ct = ct;
  81
  82         flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_ORIGINAL);
  83         flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_REPLY);
  84
  85         if (ct->status & IPS_SRC_NAT)
  86                 flow->flags |= FLOW_OFFLOAD_SNAT;
  87         if (ct->status & IPS_DST_NAT)
  88                 flow->flags |= FLOW_OFFLOAD_DNAT;
  89
  90         return flow;
  91
  92 err_dst_cache_reply:
  93         dst_release(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst);
  94 err_dst_cache_original:
  95         kfree(entry);
  96 err_ct_refcnt:
  97         nf_ct_put(ct);
  98
  99         return NULL;
 100 }
 101 EXPORT_SYMBOL_GPL(flow_offload_alloc);
 102
 103 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
 104 {
 105         tcp->state = TCP_CONNTRACK_ESTABLISHED;
 106         tcp->seen[0].td_maxwin = 0;
 107         tcp->seen[1].td_maxwin = 0;
 108 }
 109
 110 #define NF_FLOWTABLE_TCP_PICKUP_TIMEOUT (120 * HZ)
 111 #define NF_FLOWTABLE_UDP_PICKUP_TIMEOUT (30 * HZ)
 112
 113 static void flow_offload_fixup_ct_state(struct nf_conn *ct)
 114 {
 115         const struct nf_conntrack_l4proto *l4proto;
 116         unsigned int timeout;
 117         int l4num;
 118
 119         l4num = nf_ct_protonum(ct);
 120         if (l4num == IPPROTO_TCP)
 121                 flow_offload_fixup_tcp(&ct->proto.tcp);
 122
 123         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), l4num);
 124         if (!l4proto)
 125                 return;
 126
 127         if (l4num == IPPROTO_TCP)
 128                 timeout = NF_FLOWTABLE_TCP_PICKUP_TIMEOUT;
 129         else if (l4num == IPPROTO_UDP)
 130                 timeout = NF_FLOWTABLE_UDP_PICKUP_TIMEOUT;
 131         else
 132                 return;
 133
 134         ct->timeout = nfct_time_stamp + timeout;
 135 }
 136
 137 void flow_offload_free(struct flow_offload *flow)
 138 {
 139         struct flow_offload_entry *e;
 140
 141         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
 142         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
 143         e = container_of(flow, struct flow_offload_entry, flow);
 144         if (flow->flags & FLOW_OFFLOAD_DYING)
 145                 nf_ct_delete(e->ct, 0, 0);
 146         nf_ct_put(e->ct);
 147         kfree_rcu(e, rcu_head);
 148 }
 149 EXPORT_SYMBOL_GPL(flow_offload_free);
 150
 151 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
 152 {
 153         const struct flow_offload_tuple *tuple = data;
 154
 155         return jhash(tuple, offsetof(struct flow_offload_tuple, dir), seed);
 156 }
 157
 158 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
 159 {
 160         const struct flow_offload_tuple_rhash *tuplehash = data;
 161
 162         return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, dir), seed);
 163 }
 164
 165 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
 166                                         const void *ptr)
 167 {
 168         const struct flow_offload_tuple *tuple = arg->key;
 169         const struct flow_offload_tuple_rhash *x = ptr;
 170
 171         if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, dir)))
 172                 return 1;
 173
 174         return 0;
 175 }
 176
 177 static const struct rhashtable_params nf_flow_offload_rhash_params = {
 178         .head_offset            = offsetof(struct flow_offload_tuple_rhash, node),
 179         .hashfn                 = flow_offload_hash,
 180         .obj_hashfn             = flow_offload_hash_obj,
 181         .obj_cmpfn              = flow_offload_hash_cmp,
 182         .automatic_shrinking    = true,
 183 };
 184
 185 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
 186 {
 187         flow->timeout = (u32)jiffies;
 188
 189         rhashtable_insert_fast(&flow_table->rhashtable,
 190                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 191                                nf_flow_offload_rhash_params);
 192         rhashtable_insert_fast(&flow_table->rhashtable,
 193                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 194                                nf_flow_offload_rhash_params);
 195         return 0;
 196 }
 197 EXPORT_SYMBOL_GPL(flow_offload_add);
 198
 199 static void flow_offload_del(struct nf_flowtable *flow_table,
 200                              struct flow_offload *flow)
 201 {
 202         struct flow_offload_entry *e;
 203
 204         rhashtable_remove_fast(&flow_table->rhashtable,
 205                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 206                                nf_flow_offload_rhash_params);
 207         rhashtable_remove_fast(&flow_table->rhashtable,
 208                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 209                                nf_flow_offload_rhash_params);
 210
 211         e = container_of(flow, struct flow_offload_entry, flow);
 212         clear_bit(IPS_OFFLOAD_BIT, &e->ct->status);
 213
 214         flow_offload_free(flow);
 215 }
 216
 217 void flow_offload_teardown(struct flow_offload *flow)
 218 {
 219         struct flow_offload_entry *e;
 220
 221         flow->flags |= FLOW_OFFLOAD_TEARDOWN;
 222
 223         e = container_of(flow, struct flow_offload_entry, flow);
 224         flow_offload_fixup_ct_state(e->ct);
 225 }
 226 EXPORT_SYMBOL_GPL(flow_offload_teardown);
 227
 228 struct flow_offload_tuple_rhash *
 229 flow_offload_lookup(struct nf_flowtable *flow_table,
 230                     struct flow_offload_tuple *tuple)
 231 {
 232         struct flow_offload_tuple_rhash *tuplehash;
 233         struct flow_offload *flow;
 234         int dir;
 235
 236         tuplehash = rhashtable_lookup_fast(&flow_table->rhashtable, tuple,
 237                                            nf_flow_offload_rhash_params);
 238         if (!tuplehash)
 239                 return NULL;
 240
 241         dir = tuplehash->tuple.dir;
 242         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 243         if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
 244                 return NULL;
 245
 246         return tuplehash;
 247 }
 248 EXPORT_SYMBOL_GPL(flow_offload_lookup);
 249
 250 int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 251                           void (*iter)(struct flow_offload *flow, void *data),
 252                           void *data)
 253 {
 254         struct flow_offload_tuple_rhash *tuplehash;
 255         struct rhashtable_iter hti;
 256         struct flow_offload *flow;
 257         int err;
 258
 259         err = rhashtable_walk_init(&flow_table->rhashtable, &hti, GFP_KERNEL);
 260         if (err)
 261                 return err;
 262
 263         rhashtable_walk_start(&hti);
 264
 265         while ((tuplehash = rhashtable_walk_next(&hti))) {
 266                 if (IS_ERR(tuplehash)) {
 267                         err = PTR_ERR(tuplehash);
 268                         if (err != -EAGAIN)
 269                                 goto out;
 270
 271                         continue;
 272                 }
 273                 if (tuplehash->tuple.dir)
 274                         continue;
 275
 276                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 277
 278                 iter(flow, data);
 279         }
 280 out:
 281         rhashtable_walk_stop(&hti);
 282         rhashtable_walk_exit(&hti);
 283
 284         return err;
 285 }
 286 EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
 287
 288 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
 289 {
 290         return (__s32)(flow->timeout - (u32)jiffies) <= 0;
 291 }
 292
 293 static int nf_flow_offload_gc_step(struct nf_flowtable *flow_table)
 294 {
 295         struct flow_offload_tuple_rhash *tuplehash;
 296         struct rhashtable_iter hti;
 297         struct flow_offload *flow;
 298         int err;
 299
 300         err = rhashtable_walk_init(&flow_table->rhashtable, &hti, GFP_KERNEL);
 301         if (err)
 302                 return 0;
 303
 304         rhashtable_walk_start(&hti);
 305
 306         while ((tuplehash = rhashtable_walk_next(&hti))) {
 307                 if (IS_ERR(tuplehash)) {
 308                         err = PTR_ERR(tuplehash);
 309                         if (err != -EAGAIN)
 310                                 goto out;
 311
 312                         continue;
 313                 }
 314                 if (tuplehash->tuple.dir)
 315                         continue;
 316
 317                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 318
 319                 if (nf_flow_has_expired(flow) ||
 320                     (flow->flags & (FLOW_OFFLOAD_DYING |
 321                                     FLOW_OFFLOAD_TEARDOWN)))
 322                         flow_offload_del(flow_table, flow);
 323         }
 324 out:
 325         rhashtable_walk_stop(&hti);
 326         rhashtable_walk_exit(&hti);
 327
 328         return 1;
 329 }
 330
 331 static void nf_flow_offload_work_gc(struct work_struct *work)
 332 {
 333         struct nf_flowtable *flow_table;
 334
 335         flow_table = container_of(work, struct nf_flowtable, gc_work.work);
 336         nf_flow_offload_gc_step(flow_table);
 337         queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
 338 }
 339
 340 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
 341                                 __be16 port, __be16 new_port)
 342 {
 343         struct tcphdr *tcph;
 344
 345         if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
 346             skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 347                 return -1;
 348
 349         tcph = (void *)(skb_network_header(skb) + thoff);
 350         inet_proto_csum_replace2(&tcph->check, skb, port, new_port, true);
 351
 352         return 0;
 353 }
 354
 355 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
 356                                 __be16 port, __be16 new_port)
 357 {
 358         struct udphdr *udph;
 359
 360         if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
 361             skb_try_make_writable(skb, thoff + sizeof(*udph)))
 362                 return -1;
 363
 364         udph = (void *)(skb_network_header(skb) + thoff);
 365         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 366                 inet_proto_csum_replace2(&udph->check, skb, port,
 367                                          new_port, true);
 368                 if (!udph->check)
 369                         udph->check = CSUM_MANGLED_0;
 370         }
 371
 372         return 0;
 373 }
 374
 375 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
 376                             u8 protocol, __be16 port, __be16 new_port)
 377 {
 378         switch (protocol) {
 379         case IPPROTO_TCP:
 380                 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
 381                         return NF_DROP;
 382                 break;
 383         case IPPROTO_UDP:
 384                 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
 385                         return NF_DROP;
 386                 break;
 387         }
 388
 389         return 0;
 390 }
 391
 392 int nf_flow_snat_port(const struct flow_offload *flow,
 393                       struct sk_buff *skb, unsigned int thoff,
 394                       u8 protocol, enum flow_offload_tuple_dir dir)
 395 {
 396         struct flow_ports *hdr;
 397         __be16 port, new_port;
 398
 399         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 400             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 401                 return -1;
 402
 403         hdr = (void *)(skb_network_header(skb) + thoff);
 404
 405         switch (dir) {
 406         case FLOW_OFFLOAD_DIR_ORIGINAL:
 407                 port = hdr->source;
 408                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
 409                 hdr->source = new_port;
 410                 break;
 411         case FLOW_OFFLOAD_DIR_REPLY:
 412                 port = hdr->dest;
 413                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
 414                 hdr->dest = new_port;
 415                 break;
 416         default:
 417                 return -1;
 418         }
 419
 420         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 421 }
 422 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
 423
 424 int nf_flow_dnat_port(const struct flow_offload *flow,
 425                       struct sk_buff *skb, unsigned int thoff,
 426                       u8 protocol, enum flow_offload_tuple_dir dir)
 427 {
 428         struct flow_ports *hdr;
 429         __be16 port, new_port;
 430
 431         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 432             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 433                 return -1;
 434
 435         hdr = (void *)(skb_network_header(skb) + thoff);
 436
 437         switch (dir) {
 438         case FLOW_OFFLOAD_DIR_ORIGINAL:
 439                 port = hdr->dest;
 440                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
 441                 hdr->dest = new_port;
 442                 break;
 443         case FLOW_OFFLOAD_DIR_REPLY:
 444                 port = hdr->source;
 445                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
 446                 hdr->source = new_port;
 447                 break;
 448         default:
 449                 return -1;
 450         }
 451
 452         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 453 }
 454 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
 455
 456 int nf_flow_table_init(struct nf_flowtable *flowtable)
 457 {
 458         int err;
 459
 460         INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
 461
 462         err = rhashtable_init(&flowtable->rhashtable,
 463                               &nf_flow_offload_rhash_params);
 464         if (err < 0)
 465                 return err;
 466
 467         queue_delayed_work(system_power_efficient_wq,
 468                            &flowtable->gc_work, HZ);
 469
 470         mutex_lock(&flowtable_lock);
 471         list_add(&flowtable->list, &flowtables);
 472         mutex_unlock(&flowtable_lock);
 473
 474         return 0;
 475 }
 476 EXPORT_SYMBOL_GPL(nf_flow_table_init);
 477
 478 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
 479 {
 480         struct net_device *dev = data;
 481
 482         if (!dev) {
 483                 flow_offload_teardown(flow);
 484                 return;
 485         }
 486
 487         if (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
 488             flow->tuplehash[1].tuple.iifidx == dev->ifindex)
 489                 flow_offload_dead(flow);
 490 }
 491
 492 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
 493                                           struct net_device *dev)
 494 {
 495         nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 496         flush_delayed_work(&flowtable->gc_work);
 497 }
 498
 499 void nf_flow_table_cleanup(struct net *net, struct net_device *dev)
 500 {
 501         struct nf_flowtable *flowtable;
 502
 503         mutex_lock(&flowtable_lock);
 504         list_for_each_entry(flowtable, &flowtables, list)
 505                 nf_flow_table_iterate_cleanup(flowtable, dev);
 506         mutex_unlock(&flowtable_lock);
 507 }
 508 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 509
 510 void nf_flow_table_free(struct nf_flowtable *flow_table)
 511 {
 512         mutex_lock(&flowtable_lock);
 513         list_del(&flow_table->list);
 514         mutex_unlock(&flowtable_lock);
 515         cancel_delayed_work_sync(&flow_table->gc_work);
 516         nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 517         WARN_ON(!nf_flow_offload_gc_step(flow_table));
 518         rhashtable_destroy(&flow_table->rhashtable);
 519 }
 520 EXPORT_SYMBOL_GPL(nf_flow_table_free);
 521
 522 MODULE_LICENSE("GPL");
 523 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");