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