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 *other_dst = route->tuple[!dir].dst;
  32         struct dst_entry *dst = route->tuple[dir].dst;
  33
  34         ft->dir = dir;
  35
  36         switch (ctt->src.l3num) {
  37         case NFPROTO_IPV4:
  38                 ft->src_v4 = ctt->src.u3.in;
  39                 ft->dst_v4 = ctt->dst.u3.in;
  40                 ft->mtu = ip_dst_mtu_maybe_forward(dst, true);
  41                 break;
  42         case NFPROTO_IPV6:
  43                 ft->src_v6 = ctt->src.u3.in6;
  44                 ft->dst_v6 = ctt->dst.u3.in6;
  45                 ft->mtu = ip6_dst_mtu_forward(dst);
  46                 break;
  47         }
  48
  49         ft->l3proto = ctt->src.l3num;
  50         ft->l4proto = ctt->dst.protonum;
  51         ft->src_port = ctt->src.u.tcp.port;
  52         ft->dst_port = ctt->dst.u.tcp.port;
  53
  54         ft->iifidx = other_dst->dev->ifindex;
  55         ft->oifidx = dst->dev->ifindex;
  56         ft->dst_cache = dst;
  57 }
  58
  59 struct flow_offload *
  60 flow_offload_alloc(struct nf_conn *ct, struct nf_flow_route *route)
  61 {
  62         struct flow_offload_entry *entry;
  63         struct flow_offload *flow;
  64
  65         if (unlikely(nf_ct_is_dying(ct) ||
  66             !atomic_inc_not_zero(&ct->ct_general.use)))
  67                 return NULL;
  68
  69         entry = kzalloc(sizeof(*entry), GFP_ATOMIC);
  70         if (!entry)
  71                 goto err_ct_refcnt;
  72
  73         flow = &entry->flow;
  74
  75         if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst))
  76                 goto err_dst_cache_original;
  77
  78         if (!dst_hold_safe(route->tuple[FLOW_OFFLOAD_DIR_REPLY].dst))
  79                 goto err_dst_cache_reply;
  80
  81         entry->ct = ct;
  82
  83         flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_ORIGINAL);
  84         flow_offload_fill_dir(flow, ct, route, FLOW_OFFLOAD_DIR_REPLY);
  85
  86         if (ct->status & IPS_SRC_NAT)
  87                 flow->flags |= FLOW_OFFLOAD_SNAT;
  88         if (ct->status & IPS_DST_NAT)
  89                 flow->flags |= FLOW_OFFLOAD_DNAT;
  90
  91         return flow;
  92
  93 err_dst_cache_reply:
  94         dst_release(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst);
  95 err_dst_cache_original:
  96         kfree(entry);
  97 err_ct_refcnt:
  98         nf_ct_put(ct);
  99
 100         return NULL;
 101 }
 102 EXPORT_SYMBOL_GPL(flow_offload_alloc);
 103
 104 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
 105 {
 106         tcp->state = TCP_CONNTRACK_ESTABLISHED;
 107         tcp->seen[0].td_maxwin = 0;
 108         tcp->seen[1].td_maxwin = 0;
 109 }
 110
 111 #define NF_FLOWTABLE_TCP_PICKUP_TIMEOUT (120 * HZ)
 112 #define NF_FLOWTABLE_UDP_PICKUP_TIMEOUT (30 * HZ)
 113
 114 static void flow_offload_fixup_ct_state(struct nf_conn *ct)
 115 {
 116         const struct nf_conntrack_l4proto *l4proto;
 117         unsigned int timeout;
 118         int l4num;
 119
 120         l4num = nf_ct_protonum(ct);
 121         if (l4num == IPPROTO_TCP)
 122                 flow_offload_fixup_tcp(&ct->proto.tcp);
 123
 124         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), l4num);
 125         if (!l4proto)
 126                 return;
 127
 128         if (l4num == IPPROTO_TCP)
 129                 timeout = NF_FLOWTABLE_TCP_PICKUP_TIMEOUT;
 130         else if (l4num == IPPROTO_UDP)
 131                 timeout = NF_FLOWTABLE_UDP_PICKUP_TIMEOUT;
 132         else
 133                 return;
 134
 135         ct->timeout = nfct_time_stamp + timeout;
 136 }
 137
 138 void flow_offload_free(struct flow_offload *flow)
 139 {
 140         struct flow_offload_entry *e;
 141
 142         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
 143         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
 144         e = container_of(flow, struct flow_offload_entry, flow);
 145         if (flow->flags & FLOW_OFFLOAD_DYING)
 146                 nf_ct_delete(e->ct, 0, 0);
 147         nf_ct_put(e->ct);
 148         kfree_rcu(e, rcu_head);
 149 }
 150 EXPORT_SYMBOL_GPL(flow_offload_free);
 151
 152 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
 153 {
 154         const struct flow_offload_tuple *tuple = data;
 155
 156         return jhash(tuple, offsetof(struct flow_offload_tuple, dir), seed);
 157 }
 158
 159 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
 160 {
 161         const struct flow_offload_tuple_rhash *tuplehash = data;
 162
 163         return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, dir), seed);
 164 }
 165
 166 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
 167                                         const void *ptr)
 168 {
 169         const struct flow_offload_tuple *tuple = arg->key;
 170         const struct flow_offload_tuple_rhash *x = ptr;
 171
 172         if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, dir)))
 173                 return 1;
 174
 175         return 0;
 176 }
 177
 178 static const struct rhashtable_params nf_flow_offload_rhash_params = {
 179         .head_offset            = offsetof(struct flow_offload_tuple_rhash, node),
 180         .hashfn                 = flow_offload_hash,
 181         .obj_hashfn             = flow_offload_hash_obj,
 182         .obj_cmpfn              = flow_offload_hash_cmp,
 183         .automatic_shrinking    = true,
 184 };
 185
 186 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
 187 {
 188         int err;
 189
 190         flow->timeout = (u32)jiffies + NF_FLOW_TIMEOUT;
 191
 192         err = rhashtable_insert_fast(&flow_table->rhashtable,
 193                                      &flow->tuplehash[0].node,
 194                                      nf_flow_offload_rhash_params);
 195         if (err < 0)
 196                 return err;
 197
 198         err = rhashtable_insert_fast(&flow_table->rhashtable,
 199                                      &flow->tuplehash[1].node,
 200                                      nf_flow_offload_rhash_params);
 201         if (err < 0) {
 202                 rhashtable_remove_fast(&flow_table->rhashtable,
 203                                        &flow->tuplehash[0].node,
 204                                        nf_flow_offload_rhash_params);
 205                 return err;
 206         }
 207
 208         return 0;
 209 }
 210 EXPORT_SYMBOL_GPL(flow_offload_add);
 211
 212 static void flow_offload_del(struct nf_flowtable *flow_table,
 213                              struct flow_offload *flow)
 214 {
 215         struct flow_offload_entry *e;
 216
 217         rhashtable_remove_fast(&flow_table->rhashtable,
 218                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 219                                nf_flow_offload_rhash_params);
 220         rhashtable_remove_fast(&flow_table->rhashtable,
 221                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 222                                nf_flow_offload_rhash_params);
 223
 224         e = container_of(flow, struct flow_offload_entry, flow);
 225         clear_bit(IPS_OFFLOAD_BIT, &e->ct->status);
 226
 227         flow_offload_free(flow);
 228 }
 229
 230 void flow_offload_teardown(struct flow_offload *flow)
 231 {
 232         struct flow_offload_entry *e;
 233
 234         flow->flags |= FLOW_OFFLOAD_TEARDOWN;
 235
 236         e = container_of(flow, struct flow_offload_entry, flow);
 237         flow_offload_fixup_ct_state(e->ct);
 238 }
 239 EXPORT_SYMBOL_GPL(flow_offload_teardown);
 240
 241 struct flow_offload_tuple_rhash *
 242 flow_offload_lookup(struct nf_flowtable *flow_table,
 243                     struct flow_offload_tuple *tuple)
 244 {
 245         struct flow_offload_tuple_rhash *tuplehash;
 246         struct flow_offload *flow;
 247         int dir;
 248
 249         tuplehash = rhashtable_lookup_fast(&flow_table->rhashtable, tuple,
 250                                            nf_flow_offload_rhash_params);
 251         if (!tuplehash)
 252                 return NULL;
 253
 254         dir = tuplehash->tuple.dir;
 255         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 256         if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
 257                 return NULL;
 258
 259         return tuplehash;
 260 }
 261 EXPORT_SYMBOL_GPL(flow_offload_lookup);
 262
 263 int nf_flow_table_iterate(struct nf_flowtable *flow_table,
 264                           void (*iter)(struct flow_offload *flow, void *data),
 265                           void *data)
 266 {
 267         struct flow_offload_tuple_rhash *tuplehash;
 268         struct rhashtable_iter hti;
 269         struct flow_offload *flow;
 270         int err;
 271
 272         err = rhashtable_walk_init(&flow_table->rhashtable, &hti, GFP_KERNEL);
 273         if (err)
 274                 return err;
 275
 276         rhashtable_walk_start(&hti);
 277
 278         while ((tuplehash = rhashtable_walk_next(&hti))) {
 279                 if (IS_ERR(tuplehash)) {
 280                         err = PTR_ERR(tuplehash);
 281                         if (err != -EAGAIN)
 282                                 goto out;
 283
 284                         continue;
 285                 }
 286                 if (tuplehash->tuple.dir)
 287                         continue;
 288
 289                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 290
 291                 iter(flow, data);
 292         }
 293 out:
 294         rhashtable_walk_stop(&hti);
 295         rhashtable_walk_exit(&hti);
 296
 297         return err;
 298 }
 299 EXPORT_SYMBOL_GPL(nf_flow_table_iterate);
 300
 301 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
 302 {
 303         return (__s32)(flow->timeout - (u32)jiffies) <= 0;
 304 }
 305
 306 static int nf_flow_offload_gc_step(struct nf_flowtable *flow_table)
 307 {
 308         struct flow_offload_tuple_rhash *tuplehash;
 309         struct rhashtable_iter hti;
 310         struct flow_offload *flow;
 311         int err;
 312
 313         err = rhashtable_walk_init(&flow_table->rhashtable, &hti, GFP_KERNEL);
 314         if (err)
 315                 return 0;
 316
 317         rhashtable_walk_start(&hti);
 318
 319         while ((tuplehash = rhashtable_walk_next(&hti))) {
 320                 if (IS_ERR(tuplehash)) {
 321                         err = PTR_ERR(tuplehash);
 322                         if (err != -EAGAIN)
 323                                 goto out;
 324
 325                         continue;
 326                 }
 327                 if (tuplehash->tuple.dir)
 328                         continue;
 329
 330                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 331
 332                 if (nf_flow_has_expired(flow) ||
 333                     (flow->flags & (FLOW_OFFLOAD_DYING |
 334                                     FLOW_OFFLOAD_TEARDOWN)))
 335                         flow_offload_del(flow_table, flow);
 336         }
 337 out:
 338         rhashtable_walk_stop(&hti);
 339         rhashtable_walk_exit(&hti);
 340
 341         return 1;
 342 }
 343
 344 static void nf_flow_offload_work_gc(struct work_struct *work)
 345 {
 346         struct nf_flowtable *flow_table;
 347
 348         flow_table = container_of(work, struct nf_flowtable, gc_work.work);
 349         nf_flow_offload_gc_step(flow_table);
 350         queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
 351 }
 352
 353 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
 354                                 __be16 port, __be16 new_port)
 355 {
 356         struct tcphdr *tcph;
 357
 358         if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
 359             skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 360                 return -1;
 361
 362         tcph = (void *)(skb_network_header(skb) + thoff);
 363         inet_proto_csum_replace2(&tcph->check, skb, port, new_port, false);
 364
 365         return 0;
 366 }
 367
 368 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
 369                                 __be16 port, __be16 new_port)
 370 {
 371         struct udphdr *udph;
 372
 373         if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
 374             skb_try_make_writable(skb, thoff + sizeof(*udph)))
 375                 return -1;
 376
 377         udph = (void *)(skb_network_header(skb) + thoff);
 378         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 379                 inet_proto_csum_replace2(&udph->check, skb, port,
 380                                          new_port, false);
 381                 if (!udph->check)
 382                         udph->check = CSUM_MANGLED_0;
 383         }
 384
 385         return 0;
 386 }
 387
 388 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
 389                             u8 protocol, __be16 port, __be16 new_port)
 390 {
 391         switch (protocol) {
 392         case IPPROTO_TCP:
 393                 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
 394                         return NF_DROP;
 395                 break;
 396         case IPPROTO_UDP:
 397                 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
 398                         return NF_DROP;
 399                 break;
 400         }
 401
 402         return 0;
 403 }
 404
 405 int nf_flow_snat_port(const struct flow_offload *flow,
 406                       struct sk_buff *skb, unsigned int thoff,
 407                       u8 protocol, enum flow_offload_tuple_dir dir)
 408 {
 409         struct flow_ports *hdr;
 410         __be16 port, new_port;
 411
 412         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 413             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 414                 return -1;
 415
 416         hdr = (void *)(skb_network_header(skb) + thoff);
 417
 418         switch (dir) {
 419         case FLOW_OFFLOAD_DIR_ORIGINAL:
 420                 port = hdr->source;
 421                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
 422                 hdr->source = new_port;
 423                 break;
 424         case FLOW_OFFLOAD_DIR_REPLY:
 425                 port = hdr->dest;
 426                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
 427                 hdr->dest = new_port;
 428                 break;
 429         default:
 430                 return -1;
 431         }
 432
 433         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 434 }
 435 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
 436
 437 int nf_flow_dnat_port(const struct flow_offload *flow,
 438                       struct sk_buff *skb, unsigned int thoff,
 439                       u8 protocol, enum flow_offload_tuple_dir dir)
 440 {
 441         struct flow_ports *hdr;
 442         __be16 port, new_port;
 443
 444         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 445             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 446                 return -1;
 447
 448         hdr = (void *)(skb_network_header(skb) + thoff);
 449
 450         switch (dir) {
 451         case FLOW_OFFLOAD_DIR_ORIGINAL:
 452                 port = hdr->dest;
 453                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
 454                 hdr->dest = new_port;
 455                 break;
 456         case FLOW_OFFLOAD_DIR_REPLY:
 457                 port = hdr->source;
 458                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
 459                 hdr->source = new_port;
 460                 break;
 461         default:
 462                 return -1;
 463         }
 464
 465         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 466 }
 467 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
 468
 469 int nf_flow_table_init(struct nf_flowtable *flowtable)
 470 {
 471         int err;
 472
 473         INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
 474
 475         err = rhashtable_init(&flowtable->rhashtable,
 476                               &nf_flow_offload_rhash_params);
 477         if (err < 0)
 478                 return err;
 479
 480         queue_delayed_work(system_power_efficient_wq,
 481                            &flowtable->gc_work, HZ);
 482
 483         mutex_lock(&flowtable_lock);
 484         list_add(&flowtable->list, &flowtables);
 485         mutex_unlock(&flowtable_lock);
 486
 487         return 0;
 488 }
 489 EXPORT_SYMBOL_GPL(nf_flow_table_init);
 490
 491 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
 492 {
 493         struct net_device *dev = data;
 494         struct flow_offload_entry *e;
 495
 496         e = container_of(flow, struct flow_offload_entry, flow);
 497
 498         if (!dev) {
 499                 flow_offload_teardown(flow);
 500                 return;
 501         }
 502         if (net_eq(nf_ct_net(e->ct), dev_net(dev)) &&
 503             (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
 504              flow->tuplehash[1].tuple.iifidx == dev->ifindex))
 505                 flow_offload_dead(flow);
 506 }
 507
 508 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
 509                                           struct net_device *dev)
 510 {
 511         nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 512         flush_delayed_work(&flowtable->gc_work);
 513 }
 514
 515 void nf_flow_table_cleanup(struct net *net, struct net_device *dev)
 516 {
 517         struct nf_flowtable *flowtable;
 518
 519         mutex_lock(&flowtable_lock);
 520         list_for_each_entry(flowtable, &flowtables, list)
 521                 nf_flow_table_iterate_cleanup(flowtable, dev);
 522         mutex_unlock(&flowtable_lock);
 523 }
 524 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 525
 526 void nf_flow_table_free(struct nf_flowtable *flow_table)
 527 {
 528         mutex_lock(&flowtable_lock);
 529         list_del(&flow_table->list);
 530         mutex_unlock(&flowtable_lock);
 531         cancel_delayed_work_sync(&flow_table->gc_work);
 532         nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 533         WARN_ON(!nf_flow_offload_gc_step(flow_table));
 534         rhashtable_destroy(&flow_table->rhashtable);
 535 }
 536 EXPORT_SYMBOL_GPL(nf_flow_table_free);
 537
 538 MODULE_LICENSE("GPL");
 539 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");