net/core/lwt_bpf.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2016 Thomas Graf <tgraf@tgraf.ch>
   3  */
   4
   5 #include <linux/filter.h>
   6 #include <linux/kernel.h>
   7 #include <linux/module.h>
   8 #include <linux/skbuff.h>
   9 #include <linux/types.h>
  10 #include <linux/bpf.h>
  11 #include <net/lwtunnel.h>
  12 #include <net/gre.h>
  13 #include <net/ip6_route.h>
  14 #include <net/ipv6_stubs.h>
  15
  16 struct bpf_lwt_prog {
  17         struct bpf_prog *prog;
  18         char *name;
  19 };
  20
  21 struct bpf_lwt {
  22         struct bpf_lwt_prog in;
  23         struct bpf_lwt_prog out;
  24         struct bpf_lwt_prog xmit;
  25         int family;
  26 };
  27
  28 #define MAX_PROG_NAME 256
  29
  30 static inline struct bpf_lwt *bpf_lwt_lwtunnel(struct lwtunnel_state *lwt)
  31 {
  32         return (struct bpf_lwt *)lwt->data;
  33 }
  34
  35 #define NO_REDIRECT false
  36 #define CAN_REDIRECT true
  37
  38 static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
  39                        struct dst_entry *dst, bool can_redirect)
  40 {
  41         int ret;
  42
  43         /* Migration disable and BH disable are needed to protect per-cpu
  44          * redirect_info between BPF prog and skb_do_redirect().
  45          */
  46         migrate_disable();
  47         local_bh_disable();
  48         bpf_compute_data_pointers(skb);
  49         ret = bpf_prog_run_save_cb(lwt->prog, skb);
  50
  51         switch (ret) {
  52         case BPF_OK:
  53         case BPF_LWT_REROUTE:
  54                 break;
  55
  56         case BPF_REDIRECT:
  57                 if (unlikely(!can_redirect)) {
  58                         pr_warn_once("Illegal redirect return code in prog %s\n",
  59                                      lwt->name ? : "<unknown>");
  60                         ret = BPF_OK;
  61                 } else {
  62                         skb_reset_mac_header(skb);
  63                         ret = skb_do_redirect(skb);
  64                         if (ret == 0)
  65                                 ret = BPF_REDIRECT;
  66                 }
  67                 break;
  68
  69         case BPF_DROP:
  70                 kfree_skb(skb);
  71                 ret = -EPERM;
  72                 break;
  73
  74         default:
  75                 pr_warn_once("bpf-lwt: Illegal return value %u, expect packet loss\n", ret);
  76                 kfree_skb(skb);
  77                 ret = -EINVAL;
  78                 break;
  79         }
  80
  81         local_bh_enable();
  82         migrate_enable();
  83
  84         return ret;
  85 }
  86
  87 static int bpf_lwt_input_reroute(struct sk_buff *skb)
  88 {
  89         int err = -EINVAL;
  90
  91         if (skb->protocol == htons(ETH_P_IP)) {
  92                 struct net_device *dev = skb_dst(skb)->dev;
  93                 struct iphdr *iph = ip_hdr(skb);
  94
  95                 dev_hold(dev);
  96                 skb_dst_drop(skb);
  97                 err = ip_route_input_noref(skb, iph->daddr, iph->saddr,
  98                                            iph->tos, dev);
  99                 dev_put(dev);
 100         } else if (skb->protocol == htons(ETH_P_IPV6)) {
 101                 skb_dst_drop(skb);
 102                 err = ipv6_stub->ipv6_route_input(skb);
 103         } else {
 104                 err = -EAFNOSUPPORT;
 105         }
 106
 107         if (err)
 108                 goto err;
 109         return dst_input(skb);
 110
 111 err:
 112         kfree_skb(skb);
 113         return err;
 114 }
 115
 116 static int bpf_input(struct sk_buff *skb)
 117 {
 118         struct dst_entry *dst = skb_dst(skb);
 119         struct bpf_lwt *bpf;
 120         int ret;
 121
 122         bpf = bpf_lwt_lwtunnel(dst->lwtstate);
 123         if (bpf->in.prog) {
 124                 ret = run_lwt_bpf(skb, &bpf->in, dst, NO_REDIRECT);
 125                 if (ret < 0)
 126                         return ret;
 127                 if (ret == BPF_LWT_REROUTE)
 128                         return bpf_lwt_input_reroute(skb);
 129         }
 130
 131         if (unlikely(!dst->lwtstate->orig_input)) {
 132                 kfree_skb(skb);
 133                 return -EINVAL;
 134         }
 135
 136         return dst->lwtstate->orig_input(skb);
 137 }
 138
 139 static int bpf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 140 {
 141         struct dst_entry *dst = skb_dst(skb);
 142         struct bpf_lwt *bpf;
 143         int ret;
 144
 145         bpf = bpf_lwt_lwtunnel(dst->lwtstate);
 146         if (bpf->out.prog) {
 147                 ret = run_lwt_bpf(skb, &bpf->out, dst, NO_REDIRECT);
 148                 if (ret < 0)
 149                         return ret;
 150         }
 151
 152         if (unlikely(!dst->lwtstate->orig_output)) {
 153                 pr_warn_once("orig_output not set on dst for prog %s\n",
 154                              bpf->out.name);
 155                 kfree_skb(skb);
 156                 return -EINVAL;
 157         }
 158
 159         return dst->lwtstate->orig_output(net, sk, skb);
 160 }
 161
 162 static int xmit_check_hhlen(struct sk_buff *skb)
 163 {
 164         int hh_len = skb_dst(skb)->dev->hard_header_len;
 165
 166         if (skb_headroom(skb) < hh_len) {
 167                 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
 168
 169                 if (pskb_expand_head(skb, nhead, 0, GFP_ATOMIC))
 170                         return -ENOMEM;
 171         }
 172
 173         return 0;
 174 }
 175
 176 static int bpf_lwt_xmit_reroute(struct sk_buff *skb)
 177 {
 178         struct net_device *l3mdev = l3mdev_master_dev_rcu(skb_dst(skb)->dev);
 179         int oif = l3mdev ? l3mdev->ifindex : 0;
 180         struct dst_entry *dst = NULL;
 181         int err = -EAFNOSUPPORT;
 182         struct sock *sk;
 183         struct net *net;
 184         bool ipv4;
 185
 186         if (skb->protocol == htons(ETH_P_IP))
 187                 ipv4 = true;
 188         else if (skb->protocol == htons(ETH_P_IPV6))
 189                 ipv4 = false;
 190         else
 191                 goto err;
 192
 193         sk = sk_to_full_sk(skb->sk);
 194         if (sk) {
 195                 if (sk->sk_bound_dev_if)
 196                         oif = sk->sk_bound_dev_if;
 197                 net = sock_net(sk);
 198         } else {
 199                 net = dev_net(skb_dst(skb)->dev);
 200         }
 201
 202         if (ipv4) {
 203                 struct iphdr *iph = ip_hdr(skb);
 204                 struct flowi4 fl4 = {};
 205                 struct rtable *rt;
 206
 207                 fl4.flowi4_oif = oif;
 208                 fl4.flowi4_mark = skb->mark;
 209                 fl4.flowi4_uid = sock_net_uid(net, sk);
 210                 fl4.flowi4_tos = RT_TOS(iph->tos);
 211                 fl4.flowi4_flags = FLOWI_FLAG_ANYSRC;
 212                 fl4.flowi4_proto = iph->protocol;
 213                 fl4.daddr = iph->daddr;
 214                 fl4.saddr = iph->saddr;
 215
 216                 rt = ip_route_output_key(net, &fl4);
 217                 if (IS_ERR(rt)) {
 218                         err = PTR_ERR(rt);
 219                         goto err;
 220                 }
 221                 dst = &rt->dst;
 222         } else {
 223                 struct ipv6hdr *iph6 = ipv6_hdr(skb);
 224                 struct flowi6 fl6 = {};
 225
 226                 fl6.flowi6_oif = oif;
 227                 fl6.flowi6_mark = skb->mark;
 228                 fl6.flowi6_uid = sock_net_uid(net, sk);
 229                 fl6.flowlabel = ip6_flowinfo(iph6);
 230                 fl6.flowi6_proto = iph6->nexthdr;
 231                 fl6.daddr = iph6->daddr;
 232                 fl6.saddr = iph6->saddr;
 233
 234                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, skb->sk, &fl6, NULL);
 235                 if (IS_ERR(dst)) {
 236                         err = PTR_ERR(dst);
 237                         goto err;
 238                 }
 239         }
 240         if (unlikely(dst->error)) {
 241                 err = dst->error;
 242                 dst_release(dst);
 243                 goto err;
 244         }
 245
 246         /* Although skb header was reserved in bpf_lwt_push_ip_encap(), it
 247          * was done for the previous dst, so we are doing it here again, in
 248          * case the new dst needs much more space. The call below is a noop
 249          * if there is enough header space in skb.
 250          */
 251         err = skb_cow_head(skb, LL_RESERVED_SPACE(dst->dev));
 252         if (unlikely(err))
 253                 goto err;
 254
 255         skb_dst_drop(skb);
 256         skb_dst_set(skb, dst);
 257
 258         err = dst_output(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 259         if (unlikely(err))
 260                 return err;
 261
 262         /* ip[6]_finish_output2 understand LWTUNNEL_XMIT_DONE */
 263         return LWTUNNEL_XMIT_DONE;
 264
 265 err:
 266         kfree_skb(skb);
 267         return err;
 268 }
 269
 270 static int bpf_xmit(struct sk_buff *skb)
 271 {
 272         struct dst_entry *dst = skb_dst(skb);
 273         struct bpf_lwt *bpf;
 274
 275         bpf = bpf_lwt_lwtunnel(dst->lwtstate);
 276         if (bpf->xmit.prog) {
 277                 __be16 proto = skb->protocol;
 278                 int ret;
 279
 280                 ret = run_lwt_bpf(skb, &bpf->xmit, dst, CAN_REDIRECT);
 281                 switch (ret) {
 282                 case BPF_OK:
 283                         /* If the header changed, e.g. via bpf_lwt_push_encap,
 284                          * BPF_LWT_REROUTE below should have been used if the
 285                          * protocol was also changed.
 286                          */
 287                         if (skb->protocol != proto) {
 288                                 kfree_skb(skb);
 289                                 return -EINVAL;
 290                         }
 291                         /* If the header was expanded, headroom might be too
 292                          * small for L2 header to come, expand as needed.
 293                          */
 294                         ret = xmit_check_hhlen(skb);
 295                         if (unlikely(ret))
 296                                 return ret;
 297
 298                         return LWTUNNEL_XMIT_CONTINUE;
 299                 case BPF_REDIRECT:
 300                         return LWTUNNEL_XMIT_DONE;
 301                 case BPF_LWT_REROUTE:
 302                         return bpf_lwt_xmit_reroute(skb);
 303                 default:
 304                         return ret;
 305                 }
 306         }
 307
 308         return LWTUNNEL_XMIT_CONTINUE;
 309 }
 310
 311 static void bpf_lwt_prog_destroy(struct bpf_lwt_prog *prog)
 312 {
 313         if (prog->prog)
 314                 bpf_prog_put(prog->prog);
 315
 316         kfree(prog->name);
 317 }
 318
 319 static void bpf_destroy_state(struct lwtunnel_state *lwt)
 320 {
 321         struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
 322
 323         bpf_lwt_prog_destroy(&bpf->in);
 324         bpf_lwt_prog_destroy(&bpf->out);
 325         bpf_lwt_prog_destroy(&bpf->xmit);
 326 }
 327
 328 static const struct nla_policy bpf_prog_policy[LWT_BPF_PROG_MAX + 1] = {
 329         [LWT_BPF_PROG_FD]   = { .type = NLA_U32, },
 330         [LWT_BPF_PROG_NAME] = { .type = NLA_NUL_STRING,
 331                                 .len = MAX_PROG_NAME },
 332 };
 333
 334 static int bpf_parse_prog(struct nlattr *attr, struct bpf_lwt_prog *prog,
 335                           enum bpf_prog_type type)
 336 {
 337         struct nlattr *tb[LWT_BPF_PROG_MAX + 1];
 338         struct bpf_prog *p;
 339         int ret;
 340         u32 fd;
 341
 342         ret = nla_parse_nested_deprecated(tb, LWT_BPF_PROG_MAX, attr,
 343                                           bpf_prog_policy, NULL);
 344         if (ret < 0)
 345                 return ret;
 346
 347         if (!tb[LWT_BPF_PROG_FD] || !tb[LWT_BPF_PROG_NAME])
 348                 return -EINVAL;
 349
 350         prog->name = nla_memdup(tb[LWT_BPF_PROG_NAME], GFP_ATOMIC);
 351         if (!prog->name)
 352                 return -ENOMEM;
 353
 354         fd = nla_get_u32(tb[LWT_BPF_PROG_FD]);
 355         p = bpf_prog_get_type(fd, type);
 356         if (IS_ERR(p))
 357                 return PTR_ERR(p);
 358
 359         prog->prog = p;
 360
 361         return 0;
 362 }
 363
 364 static const struct nla_policy bpf_nl_policy[LWT_BPF_MAX + 1] = {
 365         [LWT_BPF_IN]            = { .type = NLA_NESTED, },
 366         [LWT_BPF_OUT]           = { .type = NLA_NESTED, },
 367         [LWT_BPF_XMIT]          = { .type = NLA_NESTED, },
 368         [LWT_BPF_XMIT_HEADROOM] = { .type = NLA_U32 },
 369 };
 370
 371 static int bpf_build_state(struct net *net, struct nlattr *nla,
 372                            unsigned int family, const void *cfg,
 373                            struct lwtunnel_state **ts,
 374                            struct netlink_ext_ack *extack)
 375 {
 376         struct nlattr *tb[LWT_BPF_MAX + 1];
 377         struct lwtunnel_state *newts;
 378         struct bpf_lwt *bpf;
 379         int ret;
 380
 381         if (family != AF_INET && family != AF_INET6)
 382                 return -EAFNOSUPPORT;
 383
 384         ret = nla_parse_nested_deprecated(tb, LWT_BPF_MAX, nla, bpf_nl_policy,
 385                                           extack);
 386         if (ret < 0)
 387                 return ret;
 388
 389         if (!tb[LWT_BPF_IN] && !tb[LWT_BPF_OUT] && !tb[LWT_BPF_XMIT])
 390                 return -EINVAL;
 391
 392         newts = lwtunnel_state_alloc(sizeof(*bpf));
 393         if (!newts)
 394                 return -ENOMEM;
 395
 396         newts->type = LWTUNNEL_ENCAP_BPF;
 397         bpf = bpf_lwt_lwtunnel(newts);
 398
 399         if (tb[LWT_BPF_IN]) {
 400                 newts->flags |= LWTUNNEL_STATE_INPUT_REDIRECT;
 401                 ret = bpf_parse_prog(tb[LWT_BPF_IN], &bpf->in,
 402                                      BPF_PROG_TYPE_LWT_IN);
 403                 if (ret  < 0)
 404                         goto errout;
 405         }
 406
 407         if (tb[LWT_BPF_OUT]) {
 408                 newts->flags |= LWTUNNEL_STATE_OUTPUT_REDIRECT;
 409                 ret = bpf_parse_prog(tb[LWT_BPF_OUT], &bpf->out,
 410                                      BPF_PROG_TYPE_LWT_OUT);
 411                 if (ret < 0)
 412                         goto errout;
 413         }
 414
 415         if (tb[LWT_BPF_XMIT]) {
 416                 newts->flags |= LWTUNNEL_STATE_XMIT_REDIRECT;
 417                 ret = bpf_parse_prog(tb[LWT_BPF_XMIT], &bpf->xmit,
 418                                      BPF_PROG_TYPE_LWT_XMIT);
 419                 if (ret < 0)
 420                         goto errout;
 421         }
 422
 423         if (tb[LWT_BPF_XMIT_HEADROOM]) {
 424                 u32 headroom = nla_get_u32(tb[LWT_BPF_XMIT_HEADROOM]);
 425
 426                 if (headroom > LWT_BPF_MAX_HEADROOM) {
 427                         ret = -ERANGE;
 428                         goto errout;
 429                 }
 430
 431                 newts->headroom = headroom;
 432         }
 433
 434         bpf->family = family;
 435         *ts = newts;
 436
 437         return 0;
 438
 439 errout:
 440         bpf_destroy_state(newts);
 441         kfree(newts);
 442         return ret;
 443 }
 444
 445 static int bpf_fill_lwt_prog(struct sk_buff *skb, int attr,
 446                              struct bpf_lwt_prog *prog)
 447 {
 448         struct nlattr *nest;
 449
 450         if (!prog->prog)
 451                 return 0;
 452
 453         nest = nla_nest_start_noflag(skb, attr);
 454         if (!nest)
 455                 return -EMSGSIZE;
 456
 457         if (prog->name &&
 458             nla_put_string(skb, LWT_BPF_PROG_NAME, prog->name))
 459                 return -EMSGSIZE;
 460
 461         return nla_nest_end(skb, nest);
 462 }
 463
 464 static int bpf_fill_encap_info(struct sk_buff *skb, struct lwtunnel_state *lwt)
 465 {
 466         struct bpf_lwt *bpf = bpf_lwt_lwtunnel(lwt);
 467
 468         if (bpf_fill_lwt_prog(skb, LWT_BPF_IN, &bpf->in) < 0 ||
 469             bpf_fill_lwt_prog(skb, LWT_BPF_OUT, &bpf->out) < 0 ||
 470             bpf_fill_lwt_prog(skb, LWT_BPF_XMIT, &bpf->xmit) < 0)
 471                 return -EMSGSIZE;
 472
 473         return 0;
 474 }
 475
 476 static int bpf_encap_nlsize(struct lwtunnel_state *lwtstate)
 477 {
 478         int nest_len = nla_total_size(sizeof(struct nlattr)) +
 479                        nla_total_size(MAX_PROG_NAME) + /* LWT_BPF_PROG_NAME */
 480                        0;
 481
 482         return nest_len + /* LWT_BPF_IN */
 483                nest_len + /* LWT_BPF_OUT */
 484                nest_len + /* LWT_BPF_XMIT */
 485                0;
 486 }
 487
 488 static int bpf_lwt_prog_cmp(struct bpf_lwt_prog *a, struct bpf_lwt_prog *b)
 489 {
 490         /* FIXME:
 491          * The LWT state is currently rebuilt for delete requests which
 492          * results in a new bpf_prog instance. Comparing names for now.
 493          */
 494         if (!a->name && !b->name)
 495                 return 0;
 496
 497         if (!a->name || !b->name)
 498                 return 1;
 499
 500         return strcmp(a->name, b->name);
 501 }
 502
 503 static int bpf_encap_cmp(struct lwtunnel_state *a, struct lwtunnel_state *b)
 504 {
 505         struct bpf_lwt *a_bpf = bpf_lwt_lwtunnel(a);
 506         struct bpf_lwt *b_bpf = bpf_lwt_lwtunnel(b);
 507
 508         return bpf_lwt_prog_cmp(&a_bpf->in, &b_bpf->in) ||
 509                bpf_lwt_prog_cmp(&a_bpf->out, &b_bpf->out) ||
 510                bpf_lwt_prog_cmp(&a_bpf->xmit, &b_bpf->xmit);
 511 }
 512
 513 static const struct lwtunnel_encap_ops bpf_encap_ops = {
 514         .build_state    = bpf_build_state,
 515         .destroy_state  = bpf_destroy_state,
 516         .input          = bpf_input,
 517         .output         = bpf_output,
 518         .xmit           = bpf_xmit,
 519         .fill_encap     = bpf_fill_encap_info,
 520         .get_encap_size = bpf_encap_nlsize,
 521         .cmp_encap      = bpf_encap_cmp,
 522         .owner          = THIS_MODULE,
 523 };
 524
 525 static int handle_gso_type(struct sk_buff *skb, unsigned int gso_type,
 526                            int encap_len)
 527 {
 528         struct skb_shared_info *shinfo = skb_shinfo(skb);
 529
 530         gso_type |= SKB_GSO_DODGY;
 531         shinfo->gso_type |= gso_type;
 532         skb_decrease_gso_size(shinfo, encap_len);
 533         shinfo->gso_segs = 0;
 534         return 0;
 535 }
 536
 537 static int handle_gso_encap(struct sk_buff *skb, bool ipv4, int encap_len)
 538 {
 539         int next_hdr_offset;
 540         void *next_hdr;
 541         __u8 protocol;
 542
 543         /* SCTP and UDP_L4 gso need more nuanced handling than what
 544          * handle_gso_type() does above: skb_decrease_gso_size() is not enough.
 545          * So at the moment only TCP GSO packets are let through.
 546          */
 547         if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
 548                 return -ENOTSUPP;
 549
 550         if (ipv4) {
 551                 protocol = ip_hdr(skb)->protocol;
 552                 next_hdr_offset = sizeof(struct iphdr);
 553                 next_hdr = skb_network_header(skb) + next_hdr_offset;
 554         } else {
 555                 protocol = ipv6_hdr(skb)->nexthdr;
 556                 next_hdr_offset = sizeof(struct ipv6hdr);
 557                 next_hdr = skb_network_header(skb) + next_hdr_offset;
 558         }
 559
 560         switch (protocol) {
 561         case IPPROTO_GRE:
 562                 next_hdr_offset += sizeof(struct gre_base_hdr);
 563                 if (next_hdr_offset > encap_len)
 564                         return -EINVAL;
 565
 566                 if (((struct gre_base_hdr *)next_hdr)->flags & GRE_CSUM)
 567                         return handle_gso_type(skb, SKB_GSO_GRE_CSUM,
 568                                                encap_len);
 569                 return handle_gso_type(skb, SKB_GSO_GRE, encap_len);
 570
 571         case IPPROTO_UDP:
 572                 next_hdr_offset += sizeof(struct udphdr);
 573                 if (next_hdr_offset > encap_len)
 574                         return -EINVAL;
 575
 576                 if (((struct udphdr *)next_hdr)->check)
 577                         return handle_gso_type(skb, SKB_GSO_UDP_TUNNEL_CSUM,
 578                                                encap_len);
 579                 return handle_gso_type(skb, SKB_GSO_UDP_TUNNEL, encap_len);
 580
 581         case IPPROTO_IP:
 582         case IPPROTO_IPV6:
 583                 if (ipv4)
 584                         return handle_gso_type(skb, SKB_GSO_IPXIP4, encap_len);
 585                 else
 586                         return handle_gso_type(skb, SKB_GSO_IPXIP6, encap_len);
 587
 588         default:
 589                 return -EPROTONOSUPPORT;
 590         }
 591 }
 592
 593 int bpf_lwt_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, bool ingress)
 594 {
 595         struct iphdr *iph;
 596         bool ipv4;
 597         int err;
 598
 599         if (unlikely(len < sizeof(struct iphdr) || len > LWT_BPF_MAX_HEADROOM))
 600                 return -EINVAL;
 601
 602         /* validate protocol and length */
 603         iph = (struct iphdr *)hdr;
 604         if (iph->version == 4) {
 605                 ipv4 = true;
 606                 if (unlikely(len < iph->ihl * 4))
 607                         return -EINVAL;
 608         } else if (iph->version == 6) {
 609                 ipv4 = false;
 610                 if (unlikely(len < sizeof(struct ipv6hdr)))
 611                         return -EINVAL;
 612         } else {
 613                 return -EINVAL;
 614         }
 615
 616         if (ingress)
 617                 err = skb_cow_head(skb, len + skb->mac_len);
 618         else
 619                 err = skb_cow_head(skb,
 620                                    len + LL_RESERVED_SPACE(skb_dst(skb)->dev));
 621         if (unlikely(err))
 622                 return err;
 623
 624         /* push the encap headers and fix pointers */
 625         skb_reset_inner_headers(skb);
 626         skb_reset_inner_mac_header(skb);  /* mac header is not yet set */
 627         skb_set_inner_protocol(skb, skb->protocol);
 628         skb->encapsulation = 1;
 629         skb_push(skb, len);
 630         if (ingress)
 631                 skb_postpush_rcsum(skb, iph, len);
 632         skb_reset_network_header(skb);
 633         memcpy(skb_network_header(skb), hdr, len);
 634         bpf_compute_data_pointers(skb);
 635         skb_clear_hash(skb);
 636
 637         if (ipv4) {
 638                 skb->protocol = htons(ETH_P_IP);
 639                 iph = ip_hdr(skb);
 640
 641                 if (!iph->check)
 642                         iph->check = ip_fast_csum((unsigned char *)iph,
 643                                                   iph->ihl);
 644         } else {
 645                 skb->protocol = htons(ETH_P_IPV6);
 646         }
 647
 648         if (skb_is_gso(skb))
 649                 return handle_gso_encap(skb, ipv4, len);
 650
 651         return 0;
 652 }
 653
 654 static int __init bpf_lwt_init(void)
 655 {
 656         return lwtunnel_encap_add_ops(&bpf_encap_ops, LWTUNNEL_ENCAP_BPF);
 657 }
 658
 659 subsys_initcall(bpf_lwt_init)