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                 flow->flags |= FLOW_OFFLOAD_SNAT;
  65         if (ct->status & IPS_DST_NAT)
  66                 flow->flags |= FLOW_OFFLOAD_DNAT;
  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 inline __s32 nf_flow_timeout_delta(unsigned int timeout)
 138 {
 139         return (__s32)(timeout - (u32)jiffies);
 140 }
 141
 142 static void flow_offload_fixup_ct_timeout(struct nf_conn *ct)
 143 {
 144         const struct nf_conntrack_l4proto *l4proto;
 145         int l4num = nf_ct_protonum(ct);
 146         unsigned int timeout;
 147
 148         l4proto = nf_ct_l4proto_find(l4num);
 149         if (!l4proto)
 150                 return;
 151
 152         if (l4num == IPPROTO_TCP)
 153                 timeout = NF_FLOWTABLE_TCP_PICKUP_TIMEOUT;
 154         else if (l4num == IPPROTO_UDP)
 155                 timeout = NF_FLOWTABLE_UDP_PICKUP_TIMEOUT;
 156         else
 157                 return;
 158
 159         if (nf_flow_timeout_delta(ct->timeout) > (__s32)timeout)
 160                 ct->timeout = nfct_time_stamp + timeout;
 161 }
 162
 163 static void flow_offload_fixup_ct_state(struct nf_conn *ct)
 164 {
 165         if (nf_ct_protonum(ct) == IPPROTO_TCP)
 166                 flow_offload_fixup_tcp(&ct->proto.tcp);
 167 }
 168
 169 static void flow_offload_fixup_ct(struct nf_conn *ct)
 170 {
 171         flow_offload_fixup_ct_state(ct);
 172         flow_offload_fixup_ct_timeout(ct);
 173 }
 174
 175 static void flow_offload_route_release(struct flow_offload *flow)
 176 {
 177         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_cache);
 178         dst_release(flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_cache);
 179 }
 180
 181 void flow_offload_free(struct flow_offload *flow)
 182 {
 183         switch (flow->type) {
 184         case NF_FLOW_OFFLOAD_ROUTE:
 185                 flow_offload_route_release(flow);
 186                 break;
 187         default:
 188                 break;
 189         }
 190         if (flow->flags & FLOW_OFFLOAD_DYING)
 191                 nf_ct_delete(flow->ct, 0, 0);
 192         nf_ct_put(flow->ct);
 193         kfree_rcu(flow, rcu_head);
 194 }
 195 EXPORT_SYMBOL_GPL(flow_offload_free);
 196
 197 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
 198 {
 199         const struct flow_offload_tuple *tuple = data;
 200
 201         return jhash(tuple, offsetof(struct flow_offload_tuple, dir), seed);
 202 }
 203
 204 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
 205 {
 206         const struct flow_offload_tuple_rhash *tuplehash = data;
 207
 208         return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, dir), seed);
 209 }
 210
 211 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
 212                                         const void *ptr)
 213 {
 214         const struct flow_offload_tuple *tuple = arg->key;
 215         const struct flow_offload_tuple_rhash *x = ptr;
 216
 217         if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, dir)))
 218                 return 1;
 219
 220         return 0;
 221 }
 222
 223 static const struct rhashtable_params nf_flow_offload_rhash_params = {
 224         .head_offset            = offsetof(struct flow_offload_tuple_rhash, node),
 225         .hashfn                 = flow_offload_hash,
 226         .obj_hashfn             = flow_offload_hash_obj,
 227         .obj_cmpfn              = flow_offload_hash_cmp,
 228         .automatic_shrinking    = true,
 229 };
 230
 231 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
 232 {
 233         int err;
 234
 235         flow->timeout = (u32)jiffies + NF_FLOW_TIMEOUT;
 236
 237         err = rhashtable_insert_fast(&flow_table->rhashtable,
 238                                      &flow->tuplehash[0].node,
 239                                      nf_flow_offload_rhash_params);
 240         if (err < 0)
 241                 return err;
 242
 243         err = rhashtable_insert_fast(&flow_table->rhashtable,
 244                                      &flow->tuplehash[1].node,
 245                                      nf_flow_offload_rhash_params);
 246         if (err < 0) {
 247                 rhashtable_remove_fast(&flow_table->rhashtable,
 248                                        &flow->tuplehash[0].node,
 249                                        nf_flow_offload_rhash_params);
 250                 return err;
 251         }
 252
 253         if (flow_table->flags & NF_FLOWTABLE_HW_OFFLOAD)
 254                 nf_flow_offload_add(flow_table, flow);
 255
 256         return 0;
 257 }
 258 EXPORT_SYMBOL_GPL(flow_offload_add);
 259
 260 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
 261 {
 262         return nf_flow_timeout_delta(flow->timeout) <= 0;
 263 }
 264
 265 static void flow_offload_del(struct nf_flowtable *flow_table,
 266                              struct flow_offload *flow)
 267 {
 268         rhashtable_remove_fast(&flow_table->rhashtable,
 269                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 270                                nf_flow_offload_rhash_params);
 271         rhashtable_remove_fast(&flow_table->rhashtable,
 272                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 273                                nf_flow_offload_rhash_params);
 274
 275         clear_bit(IPS_OFFLOAD_BIT, &flow->ct->status);
 276
 277         if (nf_flow_has_expired(flow))
 278                 flow_offload_fixup_ct(flow->ct);
 279         else if (flow->flags & FLOW_OFFLOAD_TEARDOWN)
 280                 flow_offload_fixup_ct_timeout(flow->ct);
 281
 282         flow_offload_free(flow);
 283 }
 284
 285 void flow_offload_teardown(struct flow_offload *flow)
 286 {
 287         flow->flags |= FLOW_OFFLOAD_TEARDOWN;
 288
 289         flow_offload_fixup_ct_state(flow->ct);
 290 }
 291 EXPORT_SYMBOL_GPL(flow_offload_teardown);
 292
 293 struct flow_offload_tuple_rhash *
 294 flow_offload_lookup(struct nf_flowtable *flow_table,
 295                     struct flow_offload_tuple *tuple)
 296 {
 297         struct flow_offload_tuple_rhash *tuplehash;
 298         struct flow_offload *flow;
 299         int dir;
 300
 301         tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
 302                                       nf_flow_offload_rhash_params);
 303         if (!tuplehash)
 304                 return NULL;
 305
 306         dir = tuplehash->tuple.dir;
 307         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 308         if (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))
 309                 return NULL;
 310
 311         if (unlikely(nf_ct_is_dying(flow->ct)))
 312                 return NULL;
 313
 314         return tuplehash;
 315 }
 316 EXPORT_SYMBOL_GPL(flow_offload_lookup);
 317
 318 static int
 319 nf_flow_table_iterate(struct nf_flowtable *flow_table,
 320                       void (*iter)(struct flow_offload *flow, void *data),
 321                       void *data)
 322 {
 323         struct flow_offload_tuple_rhash *tuplehash;
 324         struct rhashtable_iter hti;
 325         struct flow_offload *flow;
 326         int err = 0;
 327
 328         rhashtable_walk_enter(&flow_table->rhashtable, &hti);
 329         rhashtable_walk_start(&hti);
 330
 331         while ((tuplehash = rhashtable_walk_next(&hti))) {
 332                 if (IS_ERR(tuplehash)) {
 333                         if (PTR_ERR(tuplehash) != -EAGAIN) {
 334                                 err = PTR_ERR(tuplehash);
 335                                 break;
 336                         }
 337                         continue;
 338                 }
 339                 if (tuplehash->tuple.dir)
 340                         continue;
 341
 342                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 343
 344                 iter(flow, data);
 345         }
 346         rhashtable_walk_stop(&hti);
 347         rhashtable_walk_exit(&hti);
 348
 349         return err;
 350 }
 351
 352 static void nf_flow_offload_gc_step(struct flow_offload *flow, void *data)
 353 {
 354         struct nf_flowtable *flow_table = data;
 355
 356         if (flow->flags & FLOW_OFFLOAD_HW)
 357                 nf_flow_offload_stats(flow_table, flow);
 358
 359         if (nf_flow_has_expired(flow) || nf_ct_is_dying(flow->ct) ||
 360             (flow->flags & (FLOW_OFFLOAD_DYING | FLOW_OFFLOAD_TEARDOWN))) {
 361                 if (flow->flags & FLOW_OFFLOAD_HW) {
 362                         if (!(flow->flags & FLOW_OFFLOAD_HW_DYING))
 363                                 nf_flow_offload_del(flow_table, flow);
 364                         else if (flow->flags & FLOW_OFFLOAD_HW_DEAD)
 365                                 flow_offload_del(flow_table, flow);
 366                 } else {
 367                         flow_offload_del(flow_table, flow);
 368                 }
 369         }
 370 }
 371
 372 static void nf_flow_offload_work_gc(struct work_struct *work)
 373 {
 374         struct nf_flowtable *flow_table;
 375
 376         flow_table = container_of(work, struct nf_flowtable, gc_work.work);
 377         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 378         queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
 379 }
 380
 381 static int nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
 382                                 __be16 port, __be16 new_port)
 383 {
 384         struct tcphdr *tcph;
 385
 386         if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
 387             skb_try_make_writable(skb, thoff + sizeof(*tcph)))
 388                 return -1;
 389
 390         tcph = (void *)(skb_network_header(skb) + thoff);
 391         inet_proto_csum_replace2(&tcph->check, skb, port, new_port, true);
 392
 393         return 0;
 394 }
 395
 396 static int nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
 397                                 __be16 port, __be16 new_port)
 398 {
 399         struct udphdr *udph;
 400
 401         if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
 402             skb_try_make_writable(skb, thoff + sizeof(*udph)))
 403                 return -1;
 404
 405         udph = (void *)(skb_network_header(skb) + thoff);
 406         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 407                 inet_proto_csum_replace2(&udph->check, skb, port,
 408                                          new_port, true);
 409                 if (!udph->check)
 410                         udph->check = CSUM_MANGLED_0;
 411         }
 412
 413         return 0;
 414 }
 415
 416 static int nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
 417                             u8 protocol, __be16 port, __be16 new_port)
 418 {
 419         switch (protocol) {
 420         case IPPROTO_TCP:
 421                 if (nf_flow_nat_port_tcp(skb, thoff, port, new_port) < 0)
 422                         return NF_DROP;
 423                 break;
 424         case IPPROTO_UDP:
 425                 if (nf_flow_nat_port_udp(skb, thoff, port, new_port) < 0)
 426                         return NF_DROP;
 427                 break;
 428         }
 429
 430         return 0;
 431 }
 432
 433 int nf_flow_snat_port(const struct flow_offload *flow,
 434                       struct sk_buff *skb, unsigned int thoff,
 435                       u8 protocol, enum flow_offload_tuple_dir dir)
 436 {
 437         struct flow_ports *hdr;
 438         __be16 port, new_port;
 439
 440         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 441             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 442                 return -1;
 443
 444         hdr = (void *)(skb_network_header(skb) + thoff);
 445
 446         switch (dir) {
 447         case FLOW_OFFLOAD_DIR_ORIGINAL:
 448                 port = hdr->source;
 449                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
 450                 hdr->source = new_port;
 451                 break;
 452         case FLOW_OFFLOAD_DIR_REPLY:
 453                 port = hdr->dest;
 454                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
 455                 hdr->dest = new_port;
 456                 break;
 457         default:
 458                 return -1;
 459         }
 460
 461         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 462 }
 463 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
 464
 465 int nf_flow_dnat_port(const struct flow_offload *flow,
 466                       struct sk_buff *skb, unsigned int thoff,
 467                       u8 protocol, enum flow_offload_tuple_dir dir)
 468 {
 469         struct flow_ports *hdr;
 470         __be16 port, new_port;
 471
 472         if (!pskb_may_pull(skb, thoff + sizeof(*hdr)) ||
 473             skb_try_make_writable(skb, thoff + sizeof(*hdr)))
 474                 return -1;
 475
 476         hdr = (void *)(skb_network_header(skb) + thoff);
 477
 478         switch (dir) {
 479         case FLOW_OFFLOAD_DIR_ORIGINAL:
 480                 port = hdr->dest;
 481                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
 482                 hdr->dest = new_port;
 483                 break;
 484         case FLOW_OFFLOAD_DIR_REPLY:
 485                 port = hdr->source;
 486                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
 487                 hdr->source = new_port;
 488                 break;
 489         default:
 490                 return -1;
 491         }
 492
 493         return nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 494 }
 495 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
 496
 497 int nf_flow_table_init(struct nf_flowtable *flowtable)
 498 {
 499         int err;
 500
 501         INIT_DEFERRABLE_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
 502         flow_block_init(&flowtable->flow_block);
 503
 504         err = rhashtable_init(&flowtable->rhashtable,
 505                               &nf_flow_offload_rhash_params);
 506         if (err < 0)
 507                 return err;
 508
 509         queue_delayed_work(system_power_efficient_wq,
 510                            &flowtable->gc_work, HZ);
 511
 512         mutex_lock(&flowtable_lock);
 513         list_add(&flowtable->list, &flowtables);
 514         mutex_unlock(&flowtable_lock);
 515
 516         return 0;
 517 }
 518 EXPORT_SYMBOL_GPL(nf_flow_table_init);
 519
 520 static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
 521 {
 522         struct net_device *dev = data;
 523
 524         if (!dev) {
 525                 flow_offload_teardown(flow);
 526                 return;
 527         }
 528
 529         if (net_eq(nf_ct_net(flow->ct), dev_net(dev)) &&
 530             (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
 531              flow->tuplehash[1].tuple.iifidx == dev->ifindex))
 532                 flow_offload_dead(flow);
 533 }
 534
 535 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
 536                                           struct net_device *dev)
 537 {
 538         nf_flow_table_offload_flush(flowtable);
 539         nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 540         flush_delayed_work(&flowtable->gc_work);
 541 }
 542
 543 void nf_flow_table_cleanup(struct net_device *dev)
 544 {
 545         struct nf_flowtable *flowtable;
 546
 547         mutex_lock(&flowtable_lock);
 548         list_for_each_entry(flowtable, &flowtables, list)
 549                 nf_flow_table_iterate_cleanup(flowtable, dev);
 550         mutex_unlock(&flowtable_lock);
 551 }
 552 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 553
 554 void nf_flow_table_free(struct nf_flowtable *flow_table)
 555 {
 556         mutex_lock(&flowtable_lock);
 557         list_del(&flow_table->list);
 558         mutex_unlock(&flowtable_lock);
 559         cancel_delayed_work_sync(&flow_table->gc_work);
 560         nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 561         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
 562         rhashtable_destroy(&flow_table->rhashtable);
 563 }
 564 EXPORT_SYMBOL_GPL(nf_flow_table_free);
 565
 566 static int __init nf_flow_table_module_init(void)
 567 {
 568         return nf_flow_table_offload_init();
 569 }
 570
 571 static void __exit nf_flow_table_module_exit(void)
 572 {
 573         nf_flow_table_offload_exit();
 574 }
 575
 576 module_init(nf_flow_table_module_init);
 577 module_exit(nf_flow_table_module_exit);
 578
 579 MODULE_LICENSE("GPL");
 580 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");