net/netfilter/nf_flow_table_core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <linux/kernel.h>
   3 #include <linux/init.h>
   4 #include <linux/module.h>
   5 #include <linux/netfilter.h>
   6 #include <linux/rhashtable.h>
   7 #include <linux/netdevice.h>
   8 #include <net/ip.h>
   9 #include <net/ip6_route.h>
  10 #include <net/netfilter/nf_tables.h>
  11 #include <net/netfilter/nf_flow_table.h>
  12 #include <net/netfilter/nf_conntrack.h>
  13 #include <net/netfilter/nf_conntrack_core.h>
  14 #include <net/netfilter/nf_conntrack_l4proto.h>
  15 #include <net/netfilter/nf_conntrack_tuple.h>
  16
  17 static DEFINE_MUTEX(flowtable_lock);
  18 static LIST_HEAD(flowtables);
  19
  20 static void
  21 flow_offload_fill_dir(struct flow_offload *flow,
  22                       enum flow_offload_tuple_dir dir)
  23 {
  24         struct flow_offload_tuple *ft = &flow->tuplehash[dir].tuple;
  25         struct nf_conntrack_tuple *ctt = &flow->ct->tuplehash[dir].tuple;
  26
  27         ft->dir = dir;
  28
  29         switch (ctt->src.l3num) {
  30         case NFPROTO_IPV4:
  31                 ft->src_v4 = ctt->src.u3.in;
  32                 ft->dst_v4 = ctt->dst.u3.in;
  33                 break;
  34         case NFPROTO_IPV6:
  35                 ft->src_v6 = ctt->src.u3.in6;
  36                 ft->dst_v6 = ctt->dst.u3.in6;
  37                 break;
  38         }
  39
  40         ft->l3proto = ctt->src.l3num;
  41         ft->l4proto = ctt->dst.protonum;
  42         ft->src_port = ctt->src.u.tcp.port;
  43         ft->dst_port = ctt->dst.u.tcp.port;
  44 }
  45
  46 struct flow_offload *flow_offload_alloc(struct nf_conn *ct)
  47 {
  48         struct flow_offload *flow;
  49
  50         if (unlikely(nf_ct_is_dying(ct) ||
  51             !atomic_inc_not_zero(&ct->ct_general.use)))
  52                 return NULL;
  53
  54         flow = kzalloc(sizeof(*flow), GFP_ATOMIC);
  55         if (!flow)
  56                 goto err_ct_refcnt;
  57
  58         flow->ct = ct;
  59
  60         flow_offload_fill_dir(flow, FLOW_OFFLOAD_DIR_ORIGINAL);
  61         flow_offload_fill_dir(flow, FLOW_OFFLOAD_DIR_REPLY);
  62
  63         if (ct->status & IPS_SRC_NAT)
  64                 __set_bit(NF_FLOW_SNAT, &flow->flags);
  65         if (ct->status & IPS_DST_NAT)
  66                 __set_bit(NF_FLOW_DNAT, &flow->flags);
  67
  68         return flow;
  69
  70 err_ct_refcnt:
  71         nf_ct_put(ct);
  72
  73         return NULL;
  74 }
  75 EXPORT_SYMBOL_GPL(flow_offload_alloc);
  76
  77 static int flow_offload_fill_route(struct flow_offload *flow,
  78                                    const struct nf_flow_route *route,
  79                                    enum flow_offload_tuple_dir dir)
  80 {
  81         struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
  82         struct dst_entry *other_dst = route->tuple[!dir].dst;
  83         struct dst_entry *dst = route->tuple[dir].dst;
  84
  85         if (!dst_hold_safe(route->tuple[dir].dst))
  86                 return -1;
  87
  88         switch (flow_tuple->l3proto) {
  89         case NFPROTO_IPV4:
  90                 flow_tuple->mtu = ip_dst_mtu_maybe_forward(dst, true);
  91                 break;
  92         case NFPROTO_IPV6:
  93                 flow_tuple->mtu = ip6_dst_mtu_forward(dst);
  94                 break;
  95         }
  96
  97         flow_tuple->iifidx = other_dst->dev->ifindex;
  98         flow_tuple->dst_cache = dst;
  99
 100         return 0;
 101 }
 102
 103 int flow_offload_route_init(struct flow_offload *flow,
 104                             const struct nf_flow_route *route)
 105 {
 106         int err;
 107
 108         err = flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_ORIGINAL);
 109         if (err < 0)
 110                 return err;
 111
 112         err = flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_REPLY);
 113         if (err < 0)
 114                 goto err_route_reply;
 115
 116         flow->type = NF_FLOW_OFFLOAD_ROUTE;
 117
 118         return 0;
 119
 120 err_route_reply:
 121         dst_release(route->tuple[FLOW_OFFLOAD_DIR_ORIGINAL].dst);
 122
 123         return err;
 124 }
 125 EXPORT_SYMBOL_GPL(flow_offload_route_init);
 126
 127 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
 128 {
 129         tcp->state = TCP_CONNTRACK_ESTABLISHED;
 130         tcp->seen[0].td_maxwin = 0;
 131         tcp->seen[1].td_maxwin = 0;
 132 }
 133
 134 #define NF_FLOWTABLE_TCP_PICKUP_TIMEOUT (120 * HZ)
 135 #define NF_FLOWTABLE_UDP_PICKUP_TIMEOUT (30 * HZ)
 136
 137 static void flow_offload_fixup_ct_timeout(struct nf_conn *ct)
 138 {
 139         const struct nf_conntrack_l4proto *l4proto;
 140         int l4num = nf_ct_protonum(ct);
 141         unsigned int timeout;
 142
 143         l4proto = nf_ct_l4proto_find(l4num);
 144         if (!l4proto)
 145                 return;
 146
 147         if (l4num == IPPROTO_TCP)
 148                 timeout = NF_FLOWTABLE_TCP_PICKUP_TIMEOUT;
 149         else if (l4num == IPPROTO_UDP)
 150                 timeout = NF_FLOWTABLE_UDP_PICKUP_TIMEOUT;
 151         else
 152                 return;
 153
 154         if (nf_flow_timeout_delta(ct->timeout) > (__s32)timeout)
 155                 ct->timeout = nfct_time_stamp + timeout;
 156 }
 157
 158 static void flow_offload_fixup_ct_state(struct nf_conn *ct)
 159 {
 160         if (nf_ct_protonum(ct) == IPPROTO_TCP)
 161                 flow_offload_fixup_tcp(&ct->proto.tcp);
 162 }
 163
 164 static void flow_offload_fixup_ct(struct nf_conn *ct)
 165 {
 166         flow_offload_fixup_ct_state(ct);
 167         flow_offload_fixup_ct_timeout(ct);
 168 }
 169
 170 static void flow_offload_route_release(struct flow_offload *flow)
 171 {
 172         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
 173         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
 174 }
 175
 176 void flow_offload_free(struct flow_offload *flow)
 177 {
 178         switch (flow->type) {
 179         case NF_FLOW_OFFLOAD_ROUTE:
 180                 flow_offload_route_release(flow);
 181                 break;
 182         default:
 183                 break;
 184         }
 185         nf_ct_put(flow->ct);
 186         kfree_rcu(flow, rcu_head);
 187 }
 188 EXPORT_SYMBOL_GPL(flow_offload_free);
 189
 190 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
 191 {
 192         const struct flow_offload_tuple *tuple = data;
 193
 194         return jhash(tuple, offsetof(struct flow_offload_tuple, __hash), seed);
 195 }
 196
 197 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
 198 {
 199         const struct flow_offload_tuple_rhash *tuplehash = data;
 200
 201         return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, __hash), seed);
 202 }
 203
 204 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
 205                                         const void *ptr)
 206 {
 207         const struct flow_offload_tuple *tuple = arg->key;
 208         const struct flow_offload_tuple_rhash *x = ptr;
 209
 210         if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, __hash)))
 211                 return 1;
 212
 213         return 0;
 214 }
 215
 216 static const struct rhashtable_params nf_flow_offload_rhash_params = {
 217         .head_offset            = offsetof(struct flow_offload_tuple_rhash, node),
 218         .hashfn                 = flow_offload_hash,
 219         .obj_hashfn             = flow_offload_hash_obj,
 220         .obj_cmpfn              = flow_offload_hash_cmp,
 221         .automatic_shrinking    = true,
 222 };
 223
 224 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
 225 {
 226         int err;
 227
 228         flow->timeout = nf_flowtable_time_stamp + NF_FLOW_TIMEOUT;
 229
 230         err = rhashtable_insert_fast(&flow_table->rhashtable,
 231                                      &flow->tuplehash[0].node,
 232                                      nf_flow_offload_rhash_params);
 233         if (err < 0)
 234                 return err;
 235
 236         err = rhashtable_insert_fast(&flow_table->rhashtable,
 237                                      &flow->tuplehash[1].node,
 238                                      nf_flow_offload_rhash_params);
 239         if (err < 0) {
 240                 rhashtable_remove_fast(&flow_table->rhashtable,
 241                                        &flow->tuplehash[0].node,
 242                                        nf_flow_offload_rhash_params);
 243                 return err;
 244         }
 245
 246         nf_ct_offload_timeout(flow->ct);
 247
 248         if (nf_flowtable_hw_offload(flow_table)) {
 249                 __set_bit(NF_FLOW_HW, &flow->flags);
 250                 nf_flow_offload_add(flow_table, flow);
 251         }
 252
 253         return 0;
 254 }
 255 EXPORT_SYMBOL_GPL(flow_offload_add);
 256
 257 void flow_offload_refresh(struct nf_flowtable *flow_table,
 258                           struct flow_offload *flow)
 259 {
 260         flow->timeout = nf_flowtable_time_stamp + NF_FLOW_TIMEOUT;
 261
 262         if (likely(!nf_flowtable_hw_offload(flow_table) ||
 263                    !test_and_clear_bit(NF_FLOW_HW_REFRESH, &flow->flags)))
 264                 return;
 265
 266         nf_flow_offload_add(flow_table, flow);
 267 }
 268 EXPORT_SYMBOL_GPL(flow_offload_refresh);
 269
 270 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
 271 {
 272         return nf_flow_timeout_delta(flow->timeout) <= 0;
 273 }
 274
 275 static void flow_offload_del(struct nf_flowtable *flow_table,
 276                              struct flow_offload *flow)
 277 {
 278         rhashtable_remove_fast(&flow_table->rhashtable,
 279                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 280                                nf_flow_offload_rhash_params);
 281         rhashtable_remove_fast(&flow_table->rhashtable,
 282                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 283                                nf_flow_offload_rhash_params);
 284
 285         clear_bit(IPS_OFFLOAD_BIT, &flow->ct->status);
 286
 287         if (nf_flow_has_expired(flow))
 288                 flow_offload_fixup_ct(flow->ct);
 289         else
 290                 flow_offload_fixup_ct_timeout(flow->ct);
 291
 292         flow_offload_free(flow);
 293 }
 294
 295 void flow_offload_teardown(struct flow_offload *flow)
 296 {
 297         set_bit(NF_FLOW_TEARDOWN, &flow->flags);
 298
 299         flow_offload_fixup_ct_state(flow->ct);
 300 }
 301 EXPORT_SYMBOL_GPL(flow_offload_teardown);
 302
 303 struct flow_offload_tuple_rhash *
 304 flow_offload_lookup(struct nf_flowtable *flow_table,
 305                     struct flow_offload_tuple *tuple)
 306 {
 307         struct flow_offload_tuple_rhash *tuplehash;
 308         struct flow_offload *flow;
 309         int dir;
 310
 311         tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
 312                                       nf_flow_offload_rhash_params);
 313         if (!tuplehash)
 314                 return NULL;
 315
 316         dir = tuplehash->tuple.dir;
 317         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 318         if (test_bit(NF_FLOW_TEARDOWN, &flow->flags))
 319                 return NULL;
 320
 321         if (unlikely(nf_ct_is_dying(flow->ct)))
 322                 return NULL;
 323
 324         return tuplehash;
 325 }
 326 EXPORT_SYMBOL_GPL(flow_offload_lookup);
 327
 328 static int
 329 nf_flow_table_iterate(struct nf_flowtable *flow_table,
 330                       void (*iter)(struct flow_offload *flow, void *data),
 331                       void *data)
 332 {
 333         struct flow_offload_tuple_rhash *tuplehash;
 334         struct rhashtable_iter hti;
 335         struct flow_offload *flow;
 336         int err = 0;
 337
 338         rhashtable_walk_enter(&flow_table->rhashtable, &hti);
 339         rhashtable_walk_start(&hti);
 340
 341         while ((tuplehash = rhashtable_walk_next(&hti))) {
 342                 if (IS_ERR(tuplehash)) {
 343                         if (PTR_ERR(tuplehash) != -EAGAIN) {
 344                                 err = PTR_ERR(tuplehash);
 345                                 break;
 346                         }
 347                         continue;
 348                 }
 349                 if (tuplehash->tuple.dir)
 350                         continue;
 351
 352                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 353
 354                 iter(flow, data);
 355         }
 356         rhashtable_walk_stop(&hti);
 357         rhashtable_walk_exit(&hti);
 358
 359         return err;
 360 }
 361
 362 static void nf_flow_offload_gc_step(struct flow_offload *flow, void *data)
 363 {
 364         struct nf_flowtable *flow_table = data;
 365
 366         if (nf_flow_has_expired(flow) || nf_ct_is_dying(flow->ct))
 367                 set_bit(NF_FLOW_TEARDOWN, &flow->flags);
 368
 369         if (test_bit(NF_FLOW_TEARDOWN, &flow->flags)) {
 370                 if (test_bit(NF_FLOW_HW, &flow->flags)) {
 371                         if (!test_bit(NF_FLOW_HW_DYING, &flow->flags))
 372                                 nf_flow_offload_del(flow_table, flow);
 373                         else if (test_bit(NF_FLOW_HW_DEAD, &flow->flags))
 374                                 flow_offload_del(flow_table, flow);
 375                 } else {
 376                         flow_offload_del(flow_table, flow);
 377                 }
 378         } else if (test_bit(NF_FLOW_HW, &flow->flags)) {
 379                 nf_flow_offload_stats(flow_table, flow);
 380         }
 381 }
 382
 383 static void nf_flow_offload_work_gc(struct work_struct *work)
 384 {
 385         struct nf_flowtable *flow_table;
 386
 387         flow_table = container_of(work, struct nf_flowtable, gc_work.work);
 388         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 389         queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
 390 }
 391
 392
 393 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
 394                                 __be16 port, __be16 new_port)
 395 {
 396         struct tcphdr *tcph;
 397
 398         if (skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 399                 return -1;
 400
 401         tcph = (void *)(skb_network_header(skb) + thoff);
 402         inet_proto_csum_replace2(&tcph->check, skb, port, new_port, true);
 403
 404         return 0;
 405 }
 406
 407 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
 408                                 __be16 port, __be16 new_port)
 409 {
 410         struct udphdr *udph;
 411
 412         if (skb_try_make_writable(skb, thoff + sizeof(*udph)))
 413                 return -1;
 414
 415         udph = (void *)(skb_network_header(skb) + thoff);
 416         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 417                 inet_proto_csum_replace2(&udph->check, skb, port,
 418                                          new_port, true);
 419                 if (!udph->check)
 420                         udph->check = CSUM_MANGLED_0;
 421         }
 422
 423         return 0;
 424 }
 425
 426 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
 427                             u8 protocol, __be16 port, __be16 new_port)
 428 {
 429         switch (protocol) {
 430         case IPPROTO_TCP:
 431                 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
 432                         return NF_DROP;
 433                 break;
 434         case IPPROTO_UDP:
 435                 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
 436                         return NF_DROP;
 437                 break;
 438         }
 439
 440         return 0;
 441 }
 442
 443 int nf_flow_snat_port(const struct flow_offload *flow,
 444                       struct sk_buff *skb, unsigned int thoff,
 445                       u8 protocol, enum flow_offload_tuple_dir dir)
 446 {
 447         struct flow_ports *hdr;
 448         __be16 port, new_port;
 449
 450         if (skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 451                 return -1;
 452
 453         hdr = (void *)(skb_network_header(skb) + thoff);
 454
 455         switch (dir) {
 456         case FLOW_OFFLOAD_DIR_ORIGINAL:
 457                 port = hdr->source;
 458                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
 459                 hdr->source = new_port;
 460                 break;
 461         case FLOW_OFFLOAD_DIR_REPLY:
 462                 port = hdr->dest;
 463                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
 464                 hdr->dest = new_port;
 465                 break;
 466         default:
 467                 return -1;
 468         }
 469
 470         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 471 }
 472 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
 473
 474 int nf_flow_dnat_port(const struct flow_offload *flow,
 475                       struct sk_buff *skb, unsigned int thoff,
 476                       u8 protocol, enum flow_offload_tuple_dir dir)
 477 {
 478         struct flow_ports *hdr;
 479         __be16 port, new_port;
 480
 481         if (skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 482                 return -1;
 483
 484         hdr = (void *)(skb_network_header(skb) + thoff);
 485
 486         switch (dir) {
 487         case FLOW_OFFLOAD_DIR_ORIGINAL:
 488                 port = hdr->dest;
 489                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
 490                 hdr->dest = new_port;
 491                 break;
 492         case FLOW_OFFLOAD_DIR_REPLY:
 493                 port = hdr->source;
 494                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
 495                 hdr->source = new_port;
 496                 break;
 497         default:
 498                 return -1;
 499         }
 500
 501         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 502 }
 503 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
 504
 505 int nf_flow_table_init(struct nf_flowtable *flowtable)
 506 {
 507         int err;
 508
 509         INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
 510         flow_block_init(&flowtable->flow_block);
 511         init_rwsem(&flowtable->flow_block_lock);
 512
 513         err = rhashtable_init(&flowtable->rhashtable,
 514                               &nf_flow_offload_rhash_params);
 515         if (err < 0)
 516                 return err;
 517
 518         queue_delayed_work(system_power_efficient_wq,
 519                            &flowtable->gc_work, HZ);
 520
 521         mutex_lock(&flowtable_lock);
 522         list_add(&flowtable->list, &flowtables);
 523         mutex_unlock(&flowtable_lock);
 524
 525         return 0;
 526 }
 527 EXPORT_SYMBOL_GPL(nf_flow_table_init);
 528
 529 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
 530 {
 531         struct net_device *dev = data;
 532
 533         if (!dev) {
 534                 flow_offload_teardown(flow);
 535                 return;
 536         }
 537
 538         if (net_eq(nf_ct_net(flow->ct), dev_net(dev)) &&
 539             (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
 540              flow->tuplehash[1].tuple.iifidx == dev->ifindex))
 541                 flow_offload_teardown(flow);
 542 }
 543
 544 void nf_flow_table_gc_cleanup(struct nf_flowtable *flowtable,
 545                               struct net_device *dev)
 546 {
 547         nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 548         flush_delayed_work(&flowtable->gc_work);
 549         nf_flow_table_offload_flush(flowtable);
 550 }
 551
 552 void nf_flow_table_cleanup(struct net_device *dev)
 553 {
 554         struct nf_flowtable *flowtable;
 555
 556         mutex_lock(&flowtable_lock);
 557         list_for_each_entry(flowtable, &flowtables, list)
 558                 nf_flow_table_gc_cleanup(flowtable, dev);
 559         mutex_unlock(&flowtable_lock);
 560 }
 561 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 562
 563 void nf_flow_table_free(struct nf_flowtable *flow_table)
 564 {
 565         mutex_lock(&flowtable_lock);
 566         list_del(&flow_table->list);
 567         mutex_unlock(&flowtable_lock);
 568
 569         cancel_delayed_work_sync(&flow_table->gc_work);
 570         nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 571         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 572         nf_flow_table_offload_flush(flow_table);
 573         if (nf_flowtable_hw_offload(flow_table))
 574                 nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step,
 575                                       flow_table);
 576         rhashtable_destroy(&flow_table->rhashtable);
 577 }
 578 EXPORT_SYMBOL_GPL(nf_flow_table_free);
 579
 580 static int __init nf_flow_table_module_init(void)
 581 {
 582         return nf_flow_table_offload_init();
 583 }
 584
 585 static void __exit nf_flow_table_module_exit(void)
 586 {
 587         nf_flow_table_offload_exit();
 588 }
 589
 590 module_init(nf_flow_table_module_init);
 591 module_exit(nf_flow_table_module_exit);
 592
 593 MODULE_LICENSE("GPL");
 594 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
 595 MODULE_DESCRIPTION("Netfilter flow table module");