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