netfilter/nft_flow_offload.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <linux/kernel.h>
   3 #include <linux/module.h>
   4 #include <linux/init.h>
   5 #include <linux/netlink.h>
   6 #include <linux/netfilter.h>
   7 #include <linux/workqueue.h>
   8 #include <linux/spinlock.h>
   9 #include <linux/netfilter/nf_conntrack_common.h>
  10 #include <linux/netfilter/nf_tables.h>
  11 #include <net/ip.h> /* for ipv4 options. */
  12 #include <net/netfilter/nf_tables.h>
  13 #include <net/netfilter/nf_tables_core.h>
  14 #include <net/netfilter/nf_conntrack_core.h>
  15 #include <net/netfilter/nf_conntrack_extend.h>
  16 #include <net/netfilter/nf_flow_table.h>
  17
  18 struct nft_flow_offload {
  19         struct nft_flowtable    *flowtable;
  20 };
  21
  22 static enum flow_offload_xmit_type nft_xmit_type(struct dst_entry *dst)
  23 {
  24         if (dst_xfrm(dst))
  25                 return FLOW_OFFLOAD_XMIT_XFRM;
  26
  27         return FLOW_OFFLOAD_XMIT_NEIGH;
  28 }
  29
  30 static void nft_default_forward_path(struct nf_flow_route *route,
  31                                      struct dst_entry *dst_cache,
  32                                      enum ip_conntrack_dir dir)
  33 {
  34         route->tuple[!dir].in.ifindex   = dst_cache->dev->ifindex;
  35         route->tuple[dir].dst           = dst_cache;
  36         route->tuple[dir].xmit_type     = nft_xmit_type(dst_cache);
  37 }
  38
  39 static bool nft_is_valid_ether_device(const struct net_device *dev)
  40 {
  41         if (!dev || (dev->flags & IFF_LOOPBACK) || dev->type != ARPHRD_ETHER ||
  42             dev->addr_len != ETH_ALEN || !is_valid_ether_addr(dev->dev_addr))
  43                 return false;
  44
  45         return true;
  46 }
  47
  48 static int nft_dev_fill_forward_path(const struct nf_flow_route *route,
  49                                      const struct dst_entry *dst_cache,
  50                                      const struct nf_conn *ct,
  51                                      enum ip_conntrack_dir dir, u8 *ha,
  52                                      struct net_device_path_stack *stack)
  53 {
  54         const void *daddr = &ct->tuplehash[!dir].tuple.src.u3;
  55         struct net_device *dev = dst_cache->dev;
  56         struct neighbour *n;
  57         u8 nud_state;
  58
  59         if (!nft_is_valid_ether_device(dev))
  60                 goto out;
  61
  62         n = dst_neigh_lookup(dst_cache, daddr);
  63         if (!n)
  64                 return -1;
  65
  66         read_lock_bh(&n->lock);
  67         nud_state = n->nud_state;
  68         ether_addr_copy(ha, n->ha);
  69         read_unlock_bh(&n->lock);
  70         neigh_release(n);
  71
  72         if (!(nud_state & NUD_VALID))
  73                 return -1;
  74
  75 out:
  76         return dev_fill_forward_path(dev, ha, stack);
  77 }
  78
  79 struct nft_forward_info {
  80         const struct net_device *indev;
  81         const struct net_device *outdev;
  82         const struct net_device *hw_outdev;
  83         struct id {
  84                 __u16   id;
  85                 __be16  proto;
  86         } encap[NF_FLOW_TABLE_ENCAP_MAX];
  87         u8 num_encaps;
  88         u8 ingress_vlans;
  89         u8 h_source[ETH_ALEN];
  90         u8 h_dest[ETH_ALEN];
  91         enum flow_offload_xmit_type xmit_type;
  92 };
  93
  94 static void nft_dev_path_info(const struct net_device_path_stack *stack,
  95                               struct nft_forward_info *info,
  96                               unsigned char *ha, struct nf_flowtable *flowtable)
  97 {
  98         const struct net_device_path *path;
  99         int i;
 100
 101         memcpy(info->h_dest, ha, ETH_ALEN);
 102
 103         for (i = 0; i < stack->num_paths; i++) {
 104                 path = &stack->path[i];
 105                 switch (path->type) {
 106                 case DEV_PATH_ETHERNET:
 107                 case DEV_PATH_DSA:
 108                 case DEV_PATH_VLAN:
 109                 case DEV_PATH_PPPOE:
 110                         info->indev = path->dev;
 111                         if (is_zero_ether_addr(info->h_source))
 112                                 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 113
 114                         if (path->type == DEV_PATH_ETHERNET)
 115                                 break;
 116                         if (path->type == DEV_PATH_DSA) {
 117                                 i = stack->num_paths;
 118                                 break;
 119                         }
 120
 121                         /* DEV_PATH_VLAN and DEV_PATH_PPPOE */
 122                         if (info->num_encaps >= NF_FLOW_TABLE_ENCAP_MAX) {
 123                                 info->indev = NULL;
 124                                 break;
 125                         }
 126                         if (!info->outdev)
 127                                 info->outdev = path->dev;
 128                         info->encap[info->num_encaps].id = path->encap.id;
 129                         info->encap[info->num_encaps].proto = path->encap.proto;
 130                         info->num_encaps++;
 131                         if (path->type == DEV_PATH_PPPOE)
 132                                 memcpy(info->h_dest, path->encap.h_dest, ETH_ALEN);
 133                         break;
 134                 case DEV_PATH_BRIDGE:
 135                         if (is_zero_ether_addr(info->h_source))
 136                                 memcpy(info->h_source, path->dev->dev_addr, ETH_ALEN);
 137
 138                         switch (path->bridge.vlan_mode) {
 139                         case DEV_PATH_BR_VLAN_UNTAG_HW:
 140                                 info->ingress_vlans |= BIT(info->num_encaps - 1);
 141                                 break;
 142                         case DEV_PATH_BR_VLAN_TAG:
 143                                 info->encap[info->num_encaps].id = path->bridge.vlan_id;
 144                                 info->encap[info->num_encaps].proto = path->bridge.vlan_proto;
 145                                 info->num_encaps++;
 146                                 break;
 147                         case DEV_PATH_BR_VLAN_UNTAG:
 148                                 info->num_encaps--;
 149                                 break;
 150                         case DEV_PATH_BR_VLAN_KEEP:
 151                                 break;
 152                         }
 153                         info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
 154                         break;
 155                 default:
 156                         info->indev = NULL;
 157                         break;
 158                 }
 159         }
 160         if (!info->outdev)
 161                 info->outdev = info->indev;
 162
 163         info->hw_outdev = info->indev;
 164
 165         if (nf_flowtable_hw_offload(flowtable) &&
 166             nft_is_valid_ether_device(info->indev))
 167                 info->xmit_type = FLOW_OFFLOAD_XMIT_DIRECT;
 168 }
 169
 170 static bool nft_flowtable_find_dev(const struct net_device *dev,
 171                                    struct nft_flowtable *ft)
 172 {
 173         struct nft_hook *hook;
 174         bool found = false;
 175
 176         list_for_each_entry_rcu(hook, &ft->hook_list, list) {
 177                 if (hook->ops.dev != dev)
 178                         continue;
 179
 180                 found = true;
 181                 break;
 182         }
 183
 184         return found;
 185 }
 186
 187 static void nft_dev_forward_path(struct nf_flow_route *route,
 188                                  const struct nf_conn *ct,
 189                                  enum ip_conntrack_dir dir,
 190                                  struct nft_flowtable *ft)
 191 {
 192         const struct dst_entry *dst = route->tuple[dir].dst;
 193         struct net_device_path_stack stack;
 194         struct nft_forward_info info = {};
 195         unsigned char ha[ETH_ALEN];
 196         int i;
 197
 198         if (nft_dev_fill_forward_path(route, dst, ct, dir, ha, &stack) >= 0)
 199                 nft_dev_path_info(&stack, &info, ha, &ft->data);
 200
 201         if (!info.indev || !nft_flowtable_find_dev(info.indev, ft))
 202                 return;
 203
 204         route->tuple[!dir].in.ifindex = info.indev->ifindex;
 205         for (i = 0; i < info.num_encaps; i++) {
 206                 route->tuple[!dir].in.encap[i].id = info.encap[i].id;
 207                 route->tuple[!dir].in.encap[i].proto = info.encap[i].proto;
 208         }
 209         route->tuple[!dir].in.num_encaps = info.num_encaps;
 210         route->tuple[!dir].in.ingress_vlans = info.ingress_vlans;
 211
 212         if (info.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT) {
 213                 memcpy(route->tuple[dir].out.h_source, info.h_source, ETH_ALEN);
 214                 memcpy(route->tuple[dir].out.h_dest, info.h_dest, ETH_ALEN);
 215                 route->tuple[dir].out.ifindex = info.outdev->ifindex;
 216                 route->tuple[dir].out.hw_ifindex = info.hw_outdev->ifindex;
 217                 route->tuple[dir].xmit_type = info.xmit_type;
 218         }
 219 }
 220
 221 static int nft_flow_route(const struct nft_pktinfo *pkt,
 222                           const struct nf_conn *ct,
 223                           struct nf_flow_route *route,
 224                           enum ip_conntrack_dir dir,
 225                           struct nft_flowtable *ft)
 226 {
 227         struct dst_entry *this_dst = skb_dst(pkt->skb);
 228         struct dst_entry *other_dst = NULL;
 229         struct flowi fl;
 230
 231         memset(&fl, 0, sizeof(fl));
 232         switch (nft_pf(pkt)) {
 233         case NFPROTO_IPV4:
 234                 fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
 235                 fl.u.ip4.saddr = ct->tuplehash[!dir].tuple.src.u3.ip;
 236                 fl.u.ip4.flowi4_oif = nft_in(pkt)->ifindex;
 237                 fl.u.ip4.flowi4_iif = this_dst->dev->ifindex;
 238                 fl.u.ip4.flowi4_tos = RT_TOS(ip_hdr(pkt->skb)->tos);
 239                 fl.u.ip4.flowi4_mark = pkt->skb->mark;
 240                 fl.u.ip4.flowi4_flags = FLOWI_FLAG_ANYSRC;
 241                 break;
 242         case NFPROTO_IPV6:
 243                 fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
 244                 fl.u.ip6.saddr = ct->tuplehash[!dir].tuple.src.u3.in6;
 245                 fl.u.ip6.flowi6_oif = nft_in(pkt)->ifindex;
 246                 fl.u.ip6.flowi6_iif = this_dst->dev->ifindex;
 247                 fl.u.ip6.flowlabel = ip6_flowinfo(ipv6_hdr(pkt->skb));
 248                 fl.u.ip6.flowi6_mark = pkt->skb->mark;
 249                 fl.u.ip6.flowi6_flags = FLOWI_FLAG_ANYSRC;
 250                 break;
 251         }
 252
 253         if (!dst_hold_safe(this_dst))
 254                 return -ENOENT;
 255
 256         nf_route(nft_net(pkt), &other_dst, &fl, false, nft_pf(pkt));
 257         if (!other_dst) {
 258                 dst_release(this_dst);
 259                 return -ENOENT;
 260         }
 261
 262         nft_default_forward_path(route, this_dst, dir);
 263         nft_default_forward_path(route, other_dst, !dir);
 264
 265         if (route->tuple[dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH &&
 266             route->tuple[!dir].xmit_type == FLOW_OFFLOAD_XMIT_NEIGH) {
 267                 nft_dev_forward_path(route, ct, dir, ft);
 268                 nft_dev_forward_path(route, ct, !dir, ft);
 269         }
 270
 271         return 0;
 272 }
 273
 274 static bool nft_flow_offload_skip(struct sk_buff *skb, int family)
 275 {
 276         if (skb_sec_path(skb))
 277                 return true;
 278
 279         if (family == NFPROTO_IPV4) {
 280                 const struct ip_options *opt;
 281
 282                 opt = &(IPCB(skb)->opt);
 283
 284                 if (unlikely(opt->optlen))
 285                         return true;
 286         }
 287
 288         return false;
 289 }
 290
 291 static void nft_flow_offload_eval(const struct nft_expr *expr,
 292                                   struct nft_regs *regs,
 293                                   const struct nft_pktinfo *pkt)
 294 {
 295         struct nft_flow_offload *priv = nft_expr_priv(expr);
 296         struct nf_flowtable *flowtable = &priv->flowtable->data;
 297         struct tcphdr _tcph, *tcph = NULL;
 298         struct nf_flow_route route = {};
 299         enum ip_conntrack_info ctinfo;
 300         struct flow_offload *flow;
 301         enum ip_conntrack_dir dir;
 302         struct nf_conn *ct;
 303         int ret;
 304
 305         if (nft_flow_offload_skip(pkt->skb, nft_pf(pkt)))
 306                 goto out;
 307
 308         ct = nf_ct_get(pkt->skb, &ctinfo);
 309         if (!ct)
 310                 goto out;
 311
 312         switch (ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.protonum) {
 313         case IPPROTO_TCP:
 314                 tcph = skb_header_pointer(pkt->skb, nft_thoff(pkt),
 315                                           sizeof(_tcph), &_tcph);
 316                 if (unlikely(!tcph || tcph->fin || tcph->rst ||
 317                              !nf_conntrack_tcp_established(ct)))
 318                         goto out;
 319                 break;
 320         case IPPROTO_UDP:
 321                 break;
 322 #ifdef CONFIG_NF_CT_PROTO_GRE
 323         case IPPROTO_GRE: {
 324                 struct nf_conntrack_tuple *tuple;
 325
 326                 if (ct->status & IPS_NAT_MASK)
 327                         goto out;
 328                 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
 329                 /* No support for GRE v1 */
 330                 if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
 331                         goto out;
 332                 break;
 333         }
 334 #endif
 335         default:
 336                 goto out;
 337         }
 338
 339         if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
 340             ct->status & (IPS_SEQ_ADJUST | IPS_NAT_CLASH))
 341                 goto out;
 342
 343         if (!nf_ct_is_confirmed(ct))
 344                 goto out;
 345
 346         if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
 347                 goto out;
 348
 349         dir = CTINFO2DIR(ctinfo);
 350         if (nft_flow_route(pkt, ct, &route, dir, priv->flowtable) < 0)
 351                 goto err_flow_route;
 352
 353         flow = flow_offload_alloc(ct);
 354         if (!flow)
 355                 goto err_flow_alloc;
 356
 357         flow_offload_route_init(flow, &route);
 358
 359         if (tcph) {
 360                 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 361                 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
 362         }
 363
 364         __set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags);
 365         ret = flow_offload_add(flowtable, flow);
 366         if (ret < 0)
 367                 goto err_flow_add;
 368
 369         return;
 370
 371 err_flow_add:
 372         flow_offload_free(flow);
 373 err_flow_alloc:
 374         dst_release(route.tuple[dir].dst);
 375         dst_release(route.tuple[!dir].dst);
 376 err_flow_route:
 377         clear_bit(IPS_OFFLOAD_BIT, &ct->status);
 378 out:
 379         regs->verdict.code = NFT_BREAK;
 380 }
 381
 382 static int nft_flow_offload_validate(const struct nft_ctx *ctx,
 383                                      const struct nft_expr *expr,
 384                                      const struct nft_data **data)
 385 {
 386         unsigned int hook_mask = (1 << NF_INET_FORWARD);
 387
 388         if (ctx->family != NFPROTO_IPV4 &&
 389             ctx->family != NFPROTO_IPV6 &&
 390             ctx->family != NFPROTO_INET)
 391                 return -EOPNOTSUPP;
 392
 393         return nft_chain_validate_hooks(ctx->chain, hook_mask);
 394 }
 395
 396 static const struct nla_policy nft_flow_offload_policy[NFTA_FLOW_MAX + 1] = {
 397         [NFTA_FLOW_TABLE_NAME]  = { .type = NLA_STRING,
 398                                     .len = NFT_NAME_MAXLEN - 1 },
 399 };
 400
 401 static int nft_flow_offload_init(const struct nft_ctx *ctx,
 402                                  const struct nft_expr *expr,
 403                                  const struct nlattr * const tb[])
 404 {
 405         struct nft_flow_offload *priv = nft_expr_priv(expr);
 406         u8 genmask = nft_genmask_next(ctx->net);
 407         struct nft_flowtable *flowtable;
 408
 409         if (!tb[NFTA_FLOW_TABLE_NAME])
 410                 return -EINVAL;
 411
 412         flowtable = nft_flowtable_lookup(ctx->table, tb[NFTA_FLOW_TABLE_NAME],
 413                                          genmask);
 414         if (IS_ERR(flowtable))
 415                 return PTR_ERR(flowtable);
 416
 417         if (!nft_use_inc(&flowtable->use))
 418                 return -EMFILE;
 419
 420         priv->flowtable = flowtable;
 421
 422         return nf_ct_netns_get(ctx->net, ctx->family);
 423 }
 424
 425 static void nft_flow_offload_deactivate(const struct nft_ctx *ctx,
 426                                         const struct nft_expr *expr,
 427                                         enum nft_trans_phase phase)
 428 {
 429         struct nft_flow_offload *priv = nft_expr_priv(expr);
 430
 431         nf_tables_deactivate_flowtable(ctx, priv->flowtable, phase);
 432 }
 433
 434 static void nft_flow_offload_activate(const struct nft_ctx *ctx,
 435                                       const struct nft_expr *expr)
 436 {
 437         struct nft_flow_offload *priv = nft_expr_priv(expr);
 438
 439         nft_use_inc_restore(&priv->flowtable->use);
 440 }
 441
 442 static void nft_flow_offload_destroy(const struct nft_ctx *ctx,
 443                                      const struct nft_expr *expr)
 444 {
 445         nf_ct_netns_put(ctx->net, ctx->family);
 446 }
 447
 448 static int nft_flow_offload_dump(struct sk_buff *skb,
 449                                  const struct nft_expr *expr, bool reset)
 450 {
 451         struct nft_flow_offload *priv = nft_expr_priv(expr);
 452
 453         if (nla_put_string(skb, NFTA_FLOW_TABLE_NAME, priv->flowtable->name))
 454                 goto nla_put_failure;
 455
 456         return 0;
 457
 458 nla_put_failure:
 459         return -1;
 460 }
 461
 462 static struct nft_expr_type nft_flow_offload_type;
 463 static const struct nft_expr_ops nft_flow_offload_ops = {
 464         .type           = &nft_flow_offload_type,
 465         .size           = NFT_EXPR_SIZE(sizeof(struct nft_flow_offload)),
 466         .eval           = nft_flow_offload_eval,
 467         .init           = nft_flow_offload_init,
 468         .activate       = nft_flow_offload_activate,
 469         .deactivate     = nft_flow_offload_deactivate,
 470         .destroy        = nft_flow_offload_destroy,
 471         .validate       = nft_flow_offload_validate,
 472         .dump           = nft_flow_offload_dump,
 473         .reduce         = NFT_REDUCE_READONLY,
 474 };
 475
 476 static struct nft_expr_type nft_flow_offload_type __read_mostly = {
 477         .name           = "flow_offload",
 478         .ops            = &nft_flow_offload_ops,
 479         .policy         = nft_flow_offload_policy,
 480         .maxattr        = NFTA_FLOW_MAX,
 481         .owner          = THIS_MODULE,
 482 };
 483
 484 static int flow_offload_netdev_event(struct notifier_block *this,
 485                                      unsigned long event, void *ptr)
 486 {
 487         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 488
 489         if (event != NETDEV_DOWN)
 490                 return NOTIFY_DONE;
 491
 492         nf_flow_table_cleanup(dev);
 493
 494         return NOTIFY_DONE;
 495 }
 496
 497 static struct notifier_block flow_offload_netdev_notifier = {
 498         .notifier_call  = flow_offload_netdev_event,
 499 };
 500
 501 static int __init nft_flow_offload_module_init(void)
 502 {
 503         int err;
 504
 505         err = register_netdevice_notifier(&flow_offload_netdev_notifier);
 506         if (err)
 507                 goto err;
 508
 509         err = nft_register_expr(&nft_flow_offload_type);
 510         if (err < 0)
 511                 goto register_expr;
 512
 513         return 0;
 514
 515 register_expr:
 516         unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 517 err:
 518         return err;
 519 }
 520
 521 static void __exit nft_flow_offload_module_exit(void)
 522 {
 523         nft_unregister_expr(&nft_flow_offload_type);
 524         unregister_netdevice_notifier(&flow_offload_netdev_notifier);
 525 }
 526
 527 module_init(nft_flow_offload_module_init);
 528 module_exit(nft_flow_offload_module_exit);
 529
 530 MODULE_LICENSE("GPL");
 531 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
 532 MODULE_ALIAS_NFT_EXPR("flow_offload");
 533 MODULE_DESCRIPTION("nftables hardware flow offload module");