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
  43         switch (ctt->dst.protonum) {
  44         case IPPROTO_TCP:
  45         case IPPROTO_UDP:
  46                 ft->src_port = ctt->src.u.tcp.port;
  47                 ft->dst_port = ctt->dst.u.tcp.port;
  48                 break;
  49         }
  50 }
  51
  52 struct flow_offload *flow_offload_alloc(struct nf_conn *ct)
  53 {
  54         struct flow_offload *flow;
  55
  56         if (unlikely(nf_ct_is_dying(ct)))
  57                 return NULL;
  58
  59         flow = kzalloc(sizeof(*flow), GFP_ATOMIC);
  60         if (!flow)
  61                 return NULL;
  62
  63         refcount_inc(&ct->ct_general.use);
  64         flow->ct = ct;
  65
  66         flow_offload_fill_dir(flow, FLOW_OFFLOAD_DIR_ORIGINAL);
  67         flow_offload_fill_dir(flow, FLOW_OFFLOAD_DIR_REPLY);
  68
  69         if (ct->status & IPS_SRC_NAT)
  70                 __set_bit(NF_FLOW_SNAT, &flow->flags);
  71         if (ct->status & IPS_DST_NAT)
  72                 __set_bit(NF_FLOW_DNAT, &flow->flags);
  73
  74         return flow;
  75 }
  76 EXPORT_SYMBOL_GPL(flow_offload_alloc);
  77
  78 static u32 flow_offload_dst_cookie(struct flow_offload_tuple *flow_tuple)
  79 {
  80         const struct rt6_info *rt;
  81
  82         if (flow_tuple->l3proto == NFPROTO_IPV6) {
  83                 rt = (const struct rt6_info *)flow_tuple->dst_cache;
  84                 return rt6_get_cookie(rt);
  85         }
  86
  87         return 0;
  88 }
  89
  90 static struct dst_entry *nft_route_dst_fetch(struct nf_flow_route *route,
  91                                              enum flow_offload_tuple_dir dir)
  92 {
  93         struct dst_entry *dst = route->tuple[dir].dst;
  94
  95         route->tuple[dir].dst = NULL;
  96
  97         return dst;
  98 }
  99
 100 static int flow_offload_fill_route(struct flow_offload *flow,
 101                                    struct nf_flow_route *route,
 102                                    enum flow_offload_tuple_dir dir)
 103 {
 104         struct flow_offload_tuple *flow_tuple = &flow->tuplehash[dir].tuple;
 105         struct dst_entry *dst = nft_route_dst_fetch(route, dir);
 106         int i, j = 0;
 107
 108         switch (flow_tuple->l3proto) {
 109         case NFPROTO_IPV4:
 110                 flow_tuple->mtu = ip_dst_mtu_maybe_forward(dst, true);
 111                 break;
 112         case NFPROTO_IPV6:
 113                 flow_tuple->mtu = ip6_dst_mtu_maybe_forward(dst, true);
 114                 break;
 115         }
 116
 117         flow_tuple->iifidx = route->tuple[dir].in.ifindex;
 118         for (i = route->tuple[dir].in.num_encaps - 1; i >= 0; i--) {
 119                 flow_tuple->encap[j].id = route->tuple[dir].in.encap[i].id;
 120                 flow_tuple->encap[j].proto = route->tuple[dir].in.encap[i].proto;
 121                 if (route->tuple[dir].in.ingress_vlans & BIT(i))
 122                         flow_tuple->in_vlan_ingress |= BIT(j);
 123                 j++;
 124         }
 125         flow_tuple->encap_num = route->tuple[dir].in.num_encaps;
 126
 127         switch (route->tuple[dir].xmit_type) {
 128         case FLOW_OFFLOAD_XMIT_DIRECT:
 129                 memcpy(flow_tuple->out.h_dest, route->tuple[dir].out.h_dest,
 130                        ETH_ALEN);
 131                 memcpy(flow_tuple->out.h_source, route->tuple[dir].out.h_source,
 132                        ETH_ALEN);
 133                 flow_tuple->out.ifidx = route->tuple[dir].out.ifindex;
 134                 flow_tuple->out.hw_ifidx = route->tuple[dir].out.hw_ifindex;
 135                 dst_release(dst);
 136                 break;
 137         case FLOW_OFFLOAD_XMIT_XFRM:
 138         case FLOW_OFFLOAD_XMIT_NEIGH:
 139                 flow_tuple->dst_cache = dst;
 140                 flow_tuple->dst_cookie = flow_offload_dst_cookie(flow_tuple);
 141                 break;
 142         default:
 143                 WARN_ON_ONCE(1);
 144                 break;
 145         }
 146         flow_tuple->xmit_type = route->tuple[dir].xmit_type;
 147
 148         return 0;
 149 }
 150
 151 static void nft_flow_dst_release(struct flow_offload *flow,
 152                                  enum flow_offload_tuple_dir dir)
 153 {
 154         if (flow->tuplehash[dir].tuple.xmit_type == FLOW_OFFLOAD_XMIT_NEIGH ||
 155             flow->tuplehash[dir].tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)
 156                 dst_release(flow->tuplehash[dir].tuple.dst_cache);
 157 }
 158
 159 void flow_offload_route_init(struct flow_offload *flow,
 160                              struct nf_flow_route *route)
 161 {
 162         flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_ORIGINAL);
 163         flow_offload_fill_route(flow, route, FLOW_OFFLOAD_DIR_REPLY);
 164         flow->type = NF_FLOW_OFFLOAD_ROUTE;
 165 }
 166 EXPORT_SYMBOL_GPL(flow_offload_route_init);
 167
 168 static void flow_offload_fixup_tcp(struct ip_ct_tcp *tcp)
 169 {
 170         tcp->seen[0].td_maxwin = 0;
 171         tcp->seen[1].td_maxwin = 0;
 172 }
 173
 174 static void flow_offload_fixup_ct(struct nf_conn *ct)
 175 {
 176         struct net *net = nf_ct_net(ct);
 177         int l4num = nf_ct_protonum(ct);
 178         s32 timeout;
 179
 180         if (l4num == IPPROTO_TCP) {
 181                 struct nf_tcp_net *tn = nf_tcp_pernet(net);
 182
 183                 flow_offload_fixup_tcp(&ct->proto.tcp);
 184
 185                 timeout = tn->timeouts[ct->proto.tcp.state];
 186                 timeout -= tn->offload_timeout;
 187         } else if (l4num == IPPROTO_UDP) {
 188                 struct nf_udp_net *tn = nf_udp_pernet(net);
 189
 190                 timeout = tn->timeouts[UDP_CT_REPLIED];
 191                 timeout -= tn->offload_timeout;
 192         } else {
 193                 return;
 194         }
 195
 196         if (timeout < 0)
 197                 timeout = 0;
 198
 199         if (nf_flow_timeout_delta(READ_ONCE(ct->timeout)) > (__s32)timeout)
 200                 WRITE_ONCE(ct->timeout, nfct_time_stamp + timeout);
 201 }
 202
 203 static void flow_offload_route_release(struct flow_offload *flow)
 204 {
 205         nft_flow_dst_release(flow, FLOW_OFFLOAD_DIR_ORIGINAL);
 206         nft_flow_dst_release(flow, FLOW_OFFLOAD_DIR_REPLY);
 207 }
 208
 209 void flow_offload_free(struct flow_offload *flow)
 210 {
 211         switch (flow->type) {
 212         case NF_FLOW_OFFLOAD_ROUTE:
 213                 flow_offload_route_release(flow);
 214                 break;
 215         default:
 216                 break;
 217         }
 218         nf_ct_put(flow->ct);
 219         kfree_rcu(flow, rcu_head);
 220 }
 221 EXPORT_SYMBOL_GPL(flow_offload_free);
 222
 223 static u32 flow_offload_hash(const void *data, u32 len, u32 seed)
 224 {
 225         const struct flow_offload_tuple *tuple = data;
 226
 227         return jhash(tuple, offsetof(struct flow_offload_tuple, __hash), seed);
 228 }
 229
 230 static u32 flow_offload_hash_obj(const void *data, u32 len, u32 seed)
 231 {
 232         const struct flow_offload_tuple_rhash *tuplehash = data;
 233
 234         return jhash(&tuplehash->tuple, offsetof(struct flow_offload_tuple, __hash), seed);
 235 }
 236
 237 static int flow_offload_hash_cmp(struct rhashtable_compare_arg *arg,
 238                                         const void *ptr)
 239 {
 240         const struct flow_offload_tuple *tuple = arg->key;
 241         const struct flow_offload_tuple_rhash *x = ptr;
 242
 243         if (memcmp(&x->tuple, tuple, offsetof(struct flow_offload_tuple, __hash)))
 244                 return 1;
 245
 246         return 0;
 247 }
 248
 249 static const struct rhashtable_params nf_flow_offload_rhash_params = {
 250         .head_offset            = offsetof(struct flow_offload_tuple_rhash, node),
 251         .hashfn                 = flow_offload_hash,
 252         .obj_hashfn             = flow_offload_hash_obj,
 253         .obj_cmpfn              = flow_offload_hash_cmp,
 254         .automatic_shrinking    = true,
 255 };
 256
 257 unsigned long flow_offload_get_timeout(struct flow_offload *flow)
 258 {
 259         unsigned long timeout = NF_FLOW_TIMEOUT;
 260         struct net *net = nf_ct_net(flow->ct);
 261         int l4num = nf_ct_protonum(flow->ct);
 262
 263         if (l4num == IPPROTO_TCP) {
 264                 struct nf_tcp_net *tn = nf_tcp_pernet(net);
 265
 266                 timeout = tn->offload_timeout;
 267         } else if (l4num == IPPROTO_UDP) {
 268                 struct nf_udp_net *tn = nf_udp_pernet(net);
 269
 270                 timeout = tn->offload_timeout;
 271         }
 272
 273         return timeout;
 274 }
 275
 276 int flow_offload_add(struct nf_flowtable *flow_table, struct flow_offload *flow)
 277 {
 278         int err;
 279
 280         flow->timeout = nf_flowtable_time_stamp + flow_offload_get_timeout(flow);
 281
 282         err = rhashtable_insert_fast(&flow_table->rhashtable,
 283                                      &flow->tuplehash[0].node,
 284                                      nf_flow_offload_rhash_params);
 285         if (err < 0)
 286                 return err;
 287
 288         err = rhashtable_insert_fast(&flow_table->rhashtable,
 289                                      &flow->tuplehash[1].node,
 290                                      nf_flow_offload_rhash_params);
 291         if (err < 0) {
 292                 rhashtable_remove_fast(&flow_table->rhashtable,
 293                                        &flow->tuplehash[0].node,
 294                                        nf_flow_offload_rhash_params);
 295                 return err;
 296         }
 297
 298         nf_ct_offload_timeout(flow->ct);
 299
 300         if (nf_flowtable_hw_offload(flow_table)) {
 301                 __set_bit(NF_FLOW_HW, &flow->flags);
 302                 nf_flow_offload_add(flow_table, flow);
 303         }
 304
 305         return 0;
 306 }
 307 EXPORT_SYMBOL_GPL(flow_offload_add);
 308
 309 void flow_offload_refresh(struct nf_flowtable *flow_table,
 310                           struct flow_offload *flow, bool force)
 311 {
 312         u32 timeout;
 313
 314         timeout = nf_flowtable_time_stamp + flow_offload_get_timeout(flow);
 315         if (force || timeout - READ_ONCE(flow->timeout) > HZ)
 316                 WRITE_ONCE(flow->timeout, timeout);
 317         else
 318                 return;
 319
 320         if (likely(!nf_flowtable_hw_offload(flow_table)))
 321                 return;
 322
 323         nf_flow_offload_add(flow_table, flow);
 324 }
 325 EXPORT_SYMBOL_GPL(flow_offload_refresh);
 326
 327 static inline bool nf_flow_has_expired(const struct flow_offload *flow)
 328 {
 329         return nf_flow_timeout_delta(flow->timeout) <= 0;
 330 }
 331
 332 static void flow_offload_del(struct nf_flowtable *flow_table,
 333                              struct flow_offload *flow)
 334 {
 335         rhashtable_remove_fast(&flow_table->rhashtable,
 336                                &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].node,
 337                                nf_flow_offload_rhash_params);
 338         rhashtable_remove_fast(&flow_table->rhashtable,
 339                                &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].node,
 340                                nf_flow_offload_rhash_params);
 341         flow_offload_free(flow);
 342 }
 343
 344 void flow_offload_teardown(struct flow_offload *flow)
 345 {
 346         clear_bit(IPS_OFFLOAD_BIT, &flow->ct->status);
 347         set_bit(NF_FLOW_TEARDOWN, &flow->flags);
 348         flow_offload_fixup_ct(flow->ct);
 349 }
 350 EXPORT_SYMBOL_GPL(flow_offload_teardown);
 351
 352 struct flow_offload_tuple_rhash *
 353 flow_offload_lookup(struct nf_flowtable *flow_table,
 354                     struct flow_offload_tuple *tuple)
 355 {
 356         struct flow_offload_tuple_rhash *tuplehash;
 357         struct flow_offload *flow;
 358         int dir;
 359
 360         tuplehash = rhashtable_lookup(&flow_table->rhashtable, tuple,
 361                                       nf_flow_offload_rhash_params);
 362         if (!tuplehash)
 363                 return NULL;
 364
 365         dir = tuplehash->tuple.dir;
 366         flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 367         if (test_bit(NF_FLOW_TEARDOWN, &flow->flags))
 368                 return NULL;
 369
 370         if (unlikely(nf_ct_is_dying(flow->ct)))
 371                 return NULL;
 372
 373         return tuplehash;
 374 }
 375 EXPORT_SYMBOL_GPL(flow_offload_lookup);
 376
 377 static int
 378 nf_flow_table_iterate(struct nf_flowtable *flow_table,
 379                       void (*iter)(struct nf_flowtable *flowtable,
 380                                    struct flow_offload *flow, void *data),
 381                       void *data)
 382 {
 383         struct flow_offload_tuple_rhash *tuplehash;
 384         struct rhashtable_iter hti;
 385         struct flow_offload *flow;
 386         int err = 0;
 387
 388         rhashtable_walk_enter(&flow_table->rhashtable, &hti);
 389         rhashtable_walk_start(&hti);
 390
 391         while ((tuplehash = rhashtable_walk_next(&hti))) {
 392                 if (IS_ERR(tuplehash)) {
 393                         if (PTR_ERR(tuplehash) != -EAGAIN) {
 394                                 err = PTR_ERR(tuplehash);
 395                                 break;
 396                         }
 397                         continue;
 398                 }
 399                 if (tuplehash->tuple.dir)
 400                         continue;
 401
 402                 flow = container_of(tuplehash, struct flow_offload, tuplehash[0]);
 403
 404                 iter(flow_table, flow, data);
 405         }
 406         rhashtable_walk_stop(&hti);
 407         rhashtable_walk_exit(&hti);
 408
 409         return err;
 410 }
 411
 412 static bool nf_flow_custom_gc(struct nf_flowtable *flow_table,
 413                               const struct flow_offload *flow)
 414 {
 415         return flow_table->type->gc && flow_table->type->gc(flow);
 416 }
 417
 418 static void nf_flow_offload_gc_step(struct nf_flowtable *flow_table,
 419                                     struct flow_offload *flow, void *data)
 420 {
 421         if (nf_flow_has_expired(flow) ||
 422             nf_ct_is_dying(flow->ct) ||
 423             nf_flow_custom_gc(flow_table, flow))
 424                 flow_offload_teardown(flow);
 425
 426         if (test_bit(NF_FLOW_TEARDOWN, &flow->flags)) {
 427                 if (test_bit(NF_FLOW_HW, &flow->flags)) {
 428                         if (!test_bit(NF_FLOW_HW_DYING, &flow->flags))
 429                                 nf_flow_offload_del(flow_table, flow);
 430                         else if (test_bit(NF_FLOW_HW_DEAD, &flow->flags))
 431                                 flow_offload_del(flow_table, flow);
 432                 } else {
 433                         flow_offload_del(flow_table, flow);
 434                 }
 435         } else if (test_bit(NF_FLOW_HW, &flow->flags)) {
 436                 nf_flow_offload_stats(flow_table, flow);
 437         }
 438 }
 439
 440 void nf_flow_table_gc_run(struct nf_flowtable *flow_table)
 441 {
 442         nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, NULL);
 443 }
 444
 445 static void nf_flow_offload_work_gc(struct work_struct *work)
 446 {
 447         struct nf_flowtable *flow_table;
 448
 449         flow_table = container_of(work, struct nf_flowtable, gc_work.work);
 450         nf_flow_table_gc_run(flow_table);
 451         queue_delayed_work(system_power_efficient_wq, &flow_table->gc_work, HZ);
 452 }
 453
 454 static void nf_flow_nat_port_tcp(struct sk_buff *skb, unsigned int thoff,
 455                                  __be16 port, __be16 new_port)
 456 {
 457         struct tcphdr *tcph;
 458
 459         tcph = (void *)(skb_network_header(skb) + thoff);
 460         inet_proto_csum_replace2(&tcph->check, skb, port, new_port, false);
 461 }
 462
 463 static void nf_flow_nat_port_udp(struct sk_buff *skb, unsigned int thoff,
 464                                  __be16 port, __be16 new_port)
 465 {
 466         struct udphdr *udph;
 467
 468         udph = (void *)(skb_network_header(skb) + thoff);
 469         if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
 470                 inet_proto_csum_replace2(&udph->check, skb, port,
 471                                          new_port, false);
 472                 if (!udph->check)
 473                         udph->check = CSUM_MANGLED_0;
 474         }
 475 }
 476
 477 static void nf_flow_nat_port(struct sk_buff *skb, unsigned int thoff,
 478                              u8 protocol, __be16 port, __be16 new_port)
 479 {
 480         switch (protocol) {
 481         case IPPROTO_TCP:
 482                 nf_flow_nat_port_tcp(skb, thoff, port, new_port);
 483                 break;
 484         case IPPROTO_UDP:
 485                 nf_flow_nat_port_udp(skb, thoff, port, new_port);
 486                 break;
 487         }
 488 }
 489
 490 void nf_flow_snat_port(const struct flow_offload *flow,
 491                        struct sk_buff *skb, unsigned int thoff,
 492                        u8 protocol, enum flow_offload_tuple_dir dir)
 493 {
 494         struct flow_ports *hdr;
 495         __be16 port, new_port;
 496
 497         hdr = (void *)(skb_network_header(skb) + thoff);
 498
 499         switch (dir) {
 500         case FLOW_OFFLOAD_DIR_ORIGINAL:
 501                 port = hdr->source;
 502                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_port;
 503                 hdr->source = new_port;
 504                 break;
 505         case FLOW_OFFLOAD_DIR_REPLY:
 506                 port = hdr->dest;
 507                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_port;
 508                 hdr->dest = new_port;
 509                 break;
 510         }
 511
 512         nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 513 }
 514 EXPORT_SYMBOL_GPL(nf_flow_snat_port);
 515
 516 void nf_flow_dnat_port(const struct flow_offload *flow, struct sk_buff *skb,
 517                        unsigned int thoff, u8 protocol,
 518                        enum flow_offload_tuple_dir dir)
 519 {
 520         struct flow_ports *hdr;
 521         __be16 port, new_port;
 522
 523         hdr = (void *)(skb_network_header(skb) + thoff);
 524
 525         switch (dir) {
 526         case FLOW_OFFLOAD_DIR_ORIGINAL:
 527                 port = hdr->dest;
 528                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_port;
 529                 hdr->dest = new_port;
 530                 break;
 531         case FLOW_OFFLOAD_DIR_REPLY:
 532                 port = hdr->source;
 533                 new_port = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_port;
 534                 hdr->source = new_port;
 535                 break;
 536         }
 537
 538         nf_flow_nat_port(skb, thoff, protocol, port, new_port);
 539 }
 540 EXPORT_SYMBOL_GPL(nf_flow_dnat_port);
 541
 542 int nf_flow_table_init(struct nf_flowtable *flowtable)
 543 {
 544         int err;
 545
 546         INIT_DELAYED_WORK(&flowtable->gc_work, nf_flow_offload_work_gc);
 547         flow_block_init(&flowtable->flow_block);
 548         init_rwsem(&flowtable->flow_block_lock);
 549
 550         err = rhashtable_init(&flowtable->rhashtable,
 551                               &nf_flow_offload_rhash_params);
 552         if (err < 0)
 553                 return err;
 554
 555         queue_delayed_work(system_power_efficient_wq,
 556                            &flowtable->gc_work, HZ);
 557
 558         mutex_lock(&flowtable_lock);
 559         list_add(&flowtable->list, &flowtables);
 560         mutex_unlock(&flowtable_lock);
 561
 562         return 0;
 563 }
 564 EXPORT_SYMBOL_GPL(nf_flow_table_init);
 565
 566 static void nf_flow_table_do_cleanup(struct nf_flowtable *flow_table,
 567                                      struct flow_offload *flow, void *data)
 568 {
 569         struct net_device *dev = data;
 570
 571         if (!dev) {
 572                 flow_offload_teardown(flow);
 573                 return;
 574         }
 575
 576         if (net_eq(nf_ct_net(flow->ct), dev_net(dev)) &&
 577             (flow->tuplehash[0].tuple.iifidx == dev->ifindex ||
 578              flow->tuplehash[1].tuple.iifidx == dev->ifindex))
 579                 flow_offload_teardown(flow);
 580 }
 581
 582 void nf_flow_table_gc_cleanup(struct nf_flowtable *flowtable,
 583                               struct net_device *dev)
 584 {
 585         nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 586         flush_delayed_work(&flowtable->gc_work);
 587         nf_flow_table_offload_flush(flowtable);
 588 }
 589
 590 void nf_flow_table_cleanup(struct net_device *dev)
 591 {
 592         struct nf_flowtable *flowtable;
 593
 594         mutex_lock(&flowtable_lock);
 595         list_for_each_entry(flowtable, &flowtables, list)
 596                 nf_flow_table_gc_cleanup(flowtable, dev);
 597         mutex_unlock(&flowtable_lock);
 598 }
 599 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 600
 601 void nf_flow_table_free(struct nf_flowtable *flow_table)
 602 {
 603         mutex_lock(&flowtable_lock);
 604         list_del(&flow_table->list);
 605         mutex_unlock(&flowtable_lock);
 606
 607         cancel_delayed_work_sync(&flow_table->gc_work);
 608         nf_flow_table_offload_flush(flow_table);
 609         /* ... no more pending work after this stage ... */
 610         nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 611         nf_flow_table_gc_run(flow_table);
 612         nf_flow_table_offload_flush_cleanup(flow_table);
 613         rhashtable_destroy(&flow_table->rhashtable);
 614 }
 615 EXPORT_SYMBOL_GPL(nf_flow_table_free);
 616
 617 static int nf_flow_table_init_net(struct net *net)
 618 {
 619         net->ft.stat = alloc_percpu(struct nf_flow_table_stat);
 620         return net->ft.stat ? 0 : -ENOMEM;
 621 }
 622
 623 static void nf_flow_table_fini_net(struct net *net)
 624 {
 625         free_percpu(net->ft.stat);
 626 }
 627
 628 static int nf_flow_table_pernet_init(struct net *net)
 629 {
 630         int ret;
 631
 632         ret = nf_flow_table_init_net(net);
 633         if (ret < 0)
 634                 return ret;
 635
 636         ret = nf_flow_table_init_proc(net);
 637         if (ret < 0)
 638                 goto out_proc;
 639
 640         return 0;
 641
 642 out_proc:
 643         nf_flow_table_fini_net(net);
 644         return ret;
 645 }
 646
 647 static void nf_flow_table_pernet_exit(struct list_head *net_exit_list)
 648 {
 649         struct net *net;
 650
 651         list_for_each_entry(net, net_exit_list, exit_list) {
 652                 nf_flow_table_fini_proc(net);
 653                 nf_flow_table_fini_net(net);
 654         }
 655 }
 656
 657 static struct pernet_operations nf_flow_table_net_ops = {
 658         .init = nf_flow_table_pernet_init,
 659         .exit_batch = nf_flow_table_pernet_exit,
 660 };
 661
 662 static int __init nf_flow_table_module_init(void)
 663 {
 664         int ret;
 665
 666         ret = register_pernet_subsys(&nf_flow_table_net_ops);
 667         if (ret < 0)
 668                 return ret;
 669
 670         ret = nf_flow_table_offload_init();
 671         if (ret)
 672                 goto out_offload;
 673
 674         return 0;
 675
 676 out_offload:
 677         unregister_pernet_subsys(&nf_flow_table_net_ops);
 678         return ret;
 679 }
 680
 681 static void __exit nf_flow_table_module_exit(void)
 682 {
 683         nf_flow_table_offload_exit();
 684         unregister_pernet_subsys(&nf_flow_table_net_ops);
 685 }
 686
 687 module_init(nf_flow_table_module_init);
 688 module_exit(nf_flow_table_module_exit);
 689
 690 MODULE_LICENSE("GPL");
 691 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
 692 MODULE_DESCRIPTION("Netfilter flow table module");