net/ipv4/fou.c

   1 #include <linux/module.h>
   2 #include <linux/errno.h>
   3 #include <linux/socket.h>
   4 #include <linux/skbuff.h>
   5 #include <linux/ip.h>
   6 #include <linux/udp.h>
   7 #include <linux/types.h>
   8 #include <linux/kernel.h>
   9 #include <net/genetlink.h>
  10 #include <net/gue.h>
  11 #include <net/fou.h>
  12 #include <net/ip.h>
  13 #include <net/protocol.h>
  14 #include <net/udp.h>
  15 #include <net/udp_tunnel.h>
  16 #include <net/xfrm.h>
  17 #include <uapi/linux/fou.h>
  18 #include <uapi/linux/genetlink.h>
  19
  20 struct fou {
  21         struct socket *sock;
  22         u8 protocol;
  23         u8 flags;
  24         __be16 port;
  25         u8 family;
  26         u16 type;
  27         struct list_head list;
  28         struct rcu_head rcu;
  29 };
  30
  31 #define FOU_F_REMCSUM_NOPARTIAL BIT(0)
  32
  33 struct fou_cfg {
  34         u16 type;
  35         u8 protocol;
  36         u8 flags;
  37         struct udp_port_cfg udp_config;
  38 };
  39
  40 static unsigned int fou_net_id;
  41
  42 struct fou_net {
  43         struct list_head fou_list;
  44         struct mutex fou_lock;
  45 };
  46
  47 static inline struct fou *fou_from_sock(struct sock *sk)
  48 {
  49         return sk->sk_user_data;
  50 }
  51
  52 static int fou_recv_pull(struct sk_buff *skb, struct fou *fou, size_t len)
  53 {
  54         /* Remove 'len' bytes from the packet (UDP header and
  55          * FOU header if present).
  56          */
  57         if (fou->family == AF_INET)
  58                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
  59         else
  60                 ipv6_hdr(skb)->payload_len =
  61                     htons(ntohs(ipv6_hdr(skb)->payload_len) - len);
  62
  63         __skb_pull(skb, len);
  64         skb_postpull_rcsum(skb, udp_hdr(skb), len);
  65         skb_reset_transport_header(skb);
  66         return iptunnel_pull_offloads(skb);
  67 }
  68
  69 static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
  70 {
  71         struct fou *fou = fou_from_sock(sk);
  72
  73         if (!fou)
  74                 return 1;
  75
  76         if (fou_recv_pull(skb, fou, sizeof(struct udphdr)))
  77                 goto drop;
  78
  79         return -fou->protocol;
  80
  81 drop:
  82         kfree_skb(skb);
  83         return 0;
  84 }
  85
  86 static struct guehdr *gue_remcsum(struct sk_buff *skb, struct guehdr *guehdr,
  87                                   void *data, size_t hdrlen, u8 ipproto,
  88                                   bool nopartial)
  89 {
  90         __be16 *pd = data;
  91         size_t start = ntohs(pd[0]);
  92         size_t offset = ntohs(pd[1]);
  93         size_t plen = sizeof(struct udphdr) + hdrlen +
  94             max_t(size_t, offset + sizeof(u16), start);
  95
  96         if (skb->remcsum_offload)
  97                 return guehdr;
  98
  99         if (!pskb_may_pull(skb, plen))
 100                 return NULL;
 101         guehdr = (struct guehdr *)&udp_hdr(skb)[1];
 102
 103         skb_remcsum_process(skb, (void *)guehdr + hdrlen,
 104                             start, offset, nopartial);
 105
 106         return guehdr;
 107 }
 108
 109 static int gue_control_message(struct sk_buff *skb, struct guehdr *guehdr)
 110 {
 111         /* No support yet */
 112         kfree_skb(skb);
 113         return 0;
 114 }
 115
 116 static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
 117 {
 118         struct fou *fou = fou_from_sock(sk);
 119         size_t len, optlen, hdrlen;
 120         struct guehdr *guehdr;
 121         void *data;
 122         u16 doffset = 0;
 123
 124         if (!fou)
 125                 return 1;
 126
 127         len = sizeof(struct udphdr) + sizeof(struct guehdr);
 128         if (!pskb_may_pull(skb, len))
 129                 goto drop;
 130
 131         guehdr = (struct guehdr *)&udp_hdr(skb)[1];
 132
 133         switch (guehdr->version) {
 134         case 0: /* Full GUE header present */
 135                 break;
 136
 137         case 1: {
 138                 /* Direct encasulation of IPv4 or IPv6 */
 139
 140                 int prot;
 141
 142                 switch (((struct iphdr *)guehdr)->version) {
 143                 case 4:
 144                         prot = IPPROTO_IPIP;
 145                         break;
 146                 case 6:
 147                         prot = IPPROTO_IPV6;
 148                         break;
 149                 default:
 150                         goto drop;
 151                 }
 152
 153                 if (fou_recv_pull(skb, fou, sizeof(struct udphdr)))
 154                         goto drop;
 155
 156                 return -prot;
 157         }
 158
 159         default: /* Undefined version */
 160                 goto drop;
 161         }
 162
 163         optlen = guehdr->hlen << 2;
 164         len += optlen;
 165
 166         if (!pskb_may_pull(skb, len))
 167                 goto drop;
 168
 169         /* guehdr may change after pull */
 170         guehdr = (struct guehdr *)&udp_hdr(skb)[1];
 171
 172         hdrlen = sizeof(struct guehdr) + optlen;
 173
 174         if (guehdr->version != 0 || validate_gue_flags(guehdr, optlen))
 175                 goto drop;
 176
 177         hdrlen = sizeof(struct guehdr) + optlen;
 178
 179         if (fou->family == AF_INET)
 180                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
 181         else
 182                 ipv6_hdr(skb)->payload_len =
 183                     htons(ntohs(ipv6_hdr(skb)->payload_len) - len);
 184
 185         /* Pull csum through the guehdr now . This can be used if
 186          * there is a remote checksum offload.
 187          */
 188         skb_postpull_rcsum(skb, udp_hdr(skb), len);
 189
 190         data = &guehdr[1];
 191
 192         if (guehdr->flags & GUE_FLAG_PRIV) {
 193                 __be32 flags = *(__be32 *)(data + doffset);
 194
 195                 doffset += GUE_LEN_PRIV;
 196
 197                 if (flags & GUE_PFLAG_REMCSUM) {
 198                         guehdr = gue_remcsum(skb, guehdr, data + doffset,
 199                                              hdrlen, guehdr->proto_ctype,
 200                                              !!(fou->flags &
 201                                                 FOU_F_REMCSUM_NOPARTIAL));
 202                         if (!guehdr)
 203                                 goto drop;
 204
 205                         data = &guehdr[1];
 206
 207                         doffset += GUE_PLEN_REMCSUM;
 208                 }
 209         }
 210
 211         if (unlikely(guehdr->control))
 212                 return gue_control_message(skb, guehdr);
 213
 214         __skb_pull(skb, sizeof(struct udphdr) + hdrlen);
 215         skb_reset_transport_header(skb);
 216
 217         if (iptunnel_pull_offloads(skb))
 218                 goto drop;
 219
 220         return -guehdr->proto_ctype;
 221
 222 drop:
 223         kfree_skb(skb);
 224         return 0;
 225 }
 226
 227 static struct sk_buff **fou_gro_receive(struct sock *sk,
 228                                         struct sk_buff **head,
 229                                         struct sk_buff *skb)
 230 {
 231         const struct net_offload *ops;
 232         struct sk_buff **pp = NULL;
 233         u8 proto = fou_from_sock(sk)->protocol;
 234         const struct net_offload **offloads;
 235
 236         /* We can clear the encap_mark for FOU as we are essentially doing
 237          * one of two possible things.  We are either adding an L4 tunnel
 238          * header to the outer L3 tunnel header, or we are are simply
 239          * treating the GRE tunnel header as though it is a UDP protocol
 240          * specific header such as VXLAN or GENEVE.
 241          */
 242         NAPI_GRO_CB(skb)->encap_mark = 0;
 243
 244         /* Flag this frame as already having an outer encap header */
 245         NAPI_GRO_CB(skb)->is_fou = 1;
 246
 247         rcu_read_lock();
 248         offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
 249         ops = rcu_dereference(offloads[proto]);
 250         if (!ops || !ops->callbacks.gro_receive)
 251                 goto out_unlock;
 252
 253         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 254
 255 out_unlock:
 256         rcu_read_unlock();
 257
 258         return pp;
 259 }
 260
 261 static int fou_gro_complete(struct sock *sk, struct sk_buff *skb,
 262                             int nhoff)
 263 {
 264         const struct net_offload *ops;
 265         u8 proto = fou_from_sock(sk)->protocol;
 266         int err = -ENOSYS;
 267         const struct net_offload **offloads;
 268
 269         rcu_read_lock();
 270         offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
 271         ops = rcu_dereference(offloads[proto]);
 272         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 273                 goto out_unlock;
 274
 275         err = ops->callbacks.gro_complete(skb, nhoff);
 276
 277         skb_set_inner_mac_header(skb, nhoff);
 278
 279 out_unlock:
 280         rcu_read_unlock();
 281
 282         return err;
 283 }
 284
 285 static struct guehdr *gue_gro_remcsum(struct sk_buff *skb, unsigned int off,
 286                                       struct guehdr *guehdr, void *data,
 287                                       size_t hdrlen, struct gro_remcsum *grc,
 288                                       bool nopartial)
 289 {
 290         __be16 *pd = data;
 291         size_t start = ntohs(pd[0]);
 292         size_t offset = ntohs(pd[1]);
 293
 294         if (skb->remcsum_offload)
 295                 return guehdr;
 296
 297         if (!NAPI_GRO_CB(skb)->csum_valid)
 298                 return NULL;
 299
 300         guehdr = skb_gro_remcsum_process(skb, (void *)guehdr, off, hdrlen,
 301                                          start, offset, grc, nopartial);
 302
 303         skb->remcsum_offload = 1;
 304
 305         return guehdr;
 306 }
 307
 308 static struct sk_buff **gue_gro_receive(struct sock *sk,
 309                                         struct sk_buff **head,
 310                                         struct sk_buff *skb)
 311 {
 312         const struct net_offload **offloads;
 313         const struct net_offload *ops;
 314         struct sk_buff **pp = NULL;
 315         struct sk_buff *p;
 316         struct guehdr *guehdr;
 317         size_t len, optlen, hdrlen, off;
 318         void *data;
 319         u16 doffset = 0;
 320         int flush = 1;
 321         struct fou *fou = fou_from_sock(sk);
 322         struct gro_remcsum grc;
 323         u8 proto;
 324
 325         skb_gro_remcsum_init(&grc);
 326
 327         off = skb_gro_offset(skb);
 328         len = off + sizeof(*guehdr);
 329
 330         guehdr = skb_gro_header_fast(skb, off);
 331         if (skb_gro_header_hard(skb, len)) {
 332                 guehdr = skb_gro_header_slow(skb, len, off);
 333                 if (unlikely(!guehdr))
 334                         goto out;
 335         }
 336
 337         switch (guehdr->version) {
 338         case 0:
 339                 break;
 340         case 1:
 341                 switch (((struct iphdr *)guehdr)->version) {
 342                 case 4:
 343                         proto = IPPROTO_IPIP;
 344                         break;
 345                 case 6:
 346                         proto = IPPROTO_IPV6;
 347                         break;
 348                 default:
 349                         goto out;
 350                 }
 351                 goto next_proto;
 352         default:
 353                 goto out;
 354         }
 355
 356         optlen = guehdr->hlen << 2;
 357         len += optlen;
 358
 359         if (skb_gro_header_hard(skb, len)) {
 360                 guehdr = skb_gro_header_slow(skb, len, off);
 361                 if (unlikely(!guehdr))
 362                         goto out;
 363         }
 364
 365         if (unlikely(guehdr->control) || guehdr->version != 0 ||
 366             validate_gue_flags(guehdr, optlen))
 367                 goto out;
 368
 369         hdrlen = sizeof(*guehdr) + optlen;
 370
 371         /* Adjust NAPI_GRO_CB(skb)->csum to account for guehdr,
 372          * this is needed if there is a remote checkcsum offload.
 373          */
 374         skb_gro_postpull_rcsum(skb, guehdr, hdrlen);
 375
 376         data = &guehdr[1];
 377
 378         if (guehdr->flags & GUE_FLAG_PRIV) {
 379                 __be32 flags = *(__be32 *)(data + doffset);
 380
 381                 doffset += GUE_LEN_PRIV;
 382
 383                 if (flags & GUE_PFLAG_REMCSUM) {
 384                         guehdr = gue_gro_remcsum(skb, off, guehdr,
 385                                                  data + doffset, hdrlen, &grc,
 386                                                  !!(fou->flags &
 387                                                     FOU_F_REMCSUM_NOPARTIAL));
 388
 389                         if (!guehdr)
 390                                 goto out;
 391
 392                         data = &guehdr[1];
 393
 394                         doffset += GUE_PLEN_REMCSUM;
 395                 }
 396         }
 397
 398         skb_gro_pull(skb, hdrlen);
 399
 400         for (p = *head; p; p = p->next) {
 401                 const struct guehdr *guehdr2;
 402
 403                 if (!NAPI_GRO_CB(p)->same_flow)
 404                         continue;
 405
 406                 guehdr2 = (struct guehdr *)(p->data + off);
 407
 408                 /* Compare base GUE header to be equal (covers
 409                  * hlen, version, proto_ctype, and flags.
 410                  */
 411                 if (guehdr->word != guehdr2->word) {
 412                         NAPI_GRO_CB(p)->same_flow = 0;
 413                         continue;
 414                 }
 415
 416                 /* Compare optional fields are the same. */
 417                 if (guehdr->hlen && memcmp(&guehdr[1], &guehdr2[1],
 418                                            guehdr->hlen << 2)) {
 419                         NAPI_GRO_CB(p)->same_flow = 0;
 420                         continue;
 421                 }
 422         }
 423
 424         proto = guehdr->proto_ctype;
 425
 426 next_proto:
 427
 428         /* We can clear the encap_mark for GUE as we are essentially doing
 429          * one of two possible things.  We are either adding an L4 tunnel
 430          * header to the outer L3 tunnel header, or we are are simply
 431          * treating the GRE tunnel header as though it is a UDP protocol
 432          * specific header such as VXLAN or GENEVE.
 433          */
 434         NAPI_GRO_CB(skb)->encap_mark = 0;
 435
 436         /* Flag this frame as already having an outer encap header */
 437         NAPI_GRO_CB(skb)->is_fou = 1;
 438
 439         rcu_read_lock();
 440         offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
 441         ops = rcu_dereference(offloads[proto]);
 442         if (WARN_ON_ONCE(!ops || !ops->callbacks.gro_receive))
 443                 goto out_unlock;
 444
 445         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 446         flush = 0;
 447
 448 out_unlock:
 449         rcu_read_unlock();
 450 out:
 451         NAPI_GRO_CB(skb)->flush |= flush;
 452         skb_gro_remcsum_cleanup(skb, &grc);
 453         skb->remcsum_offload = 0;
 454
 455         return pp;
 456 }
 457
 458 static int gue_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 459 {
 460         const struct net_offload **offloads;
 461         struct guehdr *guehdr = (struct guehdr *)(skb->data + nhoff);
 462         const struct net_offload *ops;
 463         unsigned int guehlen = 0;
 464         u8 proto;
 465         int err = -ENOENT;
 466
 467         switch (guehdr->version) {
 468         case 0:
 469                 proto = guehdr->proto_ctype;
 470                 guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
 471                 break;
 472         case 1:
 473                 switch (((struct iphdr *)guehdr)->version) {
 474                 case 4:
 475                         proto = IPPROTO_IPIP;
 476                         break;
 477                 case 6:
 478                         proto = IPPROTO_IPV6;
 479                         break;
 480                 default:
 481                         return err;
 482                 }
 483                 break;
 484         default:
 485                 return err;
 486         }
 487
 488         rcu_read_lock();
 489         offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
 490         ops = rcu_dereference(offloads[proto]);
 491         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 492                 goto out_unlock;
 493
 494         err = ops->callbacks.gro_complete(skb, nhoff + guehlen);
 495
 496         skb_set_inner_mac_header(skb, nhoff + guehlen);
 497
 498 out_unlock:
 499         rcu_read_unlock();
 500         return err;
 501 }
 502
 503 static int fou_add_to_port_list(struct net *net, struct fou *fou)
 504 {
 505         struct fou_net *fn = net_generic(net, fou_net_id);
 506         struct fou *fout;
 507
 508         mutex_lock(&fn->fou_lock);
 509         list_for_each_entry(fout, &fn->fou_list, list) {
 510                 if (fou->port == fout->port &&
 511                     fou->family == fout->family) {
 512                         mutex_unlock(&fn->fou_lock);
 513                         return -EALREADY;
 514                 }
 515         }
 516
 517         list_add(&fou->list, &fn->fou_list);
 518         mutex_unlock(&fn->fou_lock);
 519
 520         return 0;
 521 }
 522
 523 static void fou_release(struct fou *fou)
 524 {
 525         struct socket *sock = fou->sock;
 526
 527         list_del(&fou->list);
 528         udp_tunnel_sock_release(sock);
 529
 530         kfree_rcu(fou, rcu);
 531 }
 532
 533 static int fou_create(struct net *net, struct fou_cfg *cfg,
 534                       struct socket **sockp)
 535 {
 536         struct socket *sock = NULL;
 537         struct fou *fou = NULL;
 538         struct sock *sk;
 539         struct udp_tunnel_sock_cfg tunnel_cfg;
 540         int err;
 541
 542         /* Open UDP socket */
 543         err = udp_sock_create(net, &cfg->udp_config, &sock);
 544         if (err < 0)
 545                 goto error;
 546
 547         /* Allocate FOU port structure */
 548         fou = kzalloc(sizeof(*fou), GFP_KERNEL);
 549         if (!fou) {
 550                 err = -ENOMEM;
 551                 goto error;
 552         }
 553
 554         sk = sock->sk;
 555
 556         fou->port = cfg->udp_config.local_udp_port;
 557         fou->family = cfg->udp_config.family;
 558         fou->flags = cfg->flags;
 559         fou->type = cfg->type;
 560         fou->sock = sock;
 561
 562         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
 563         tunnel_cfg.encap_type = 1;
 564         tunnel_cfg.sk_user_data = fou;
 565         tunnel_cfg.encap_destroy = NULL;
 566
 567         /* Initial for fou type */
 568         switch (cfg->type) {
 569         case FOU_ENCAP_DIRECT:
 570                 tunnel_cfg.encap_rcv = fou_udp_recv;
 571                 tunnel_cfg.gro_receive = fou_gro_receive;
 572                 tunnel_cfg.gro_complete = fou_gro_complete;
 573                 fou->protocol = cfg->protocol;
 574                 break;
 575         case FOU_ENCAP_GUE:
 576                 tunnel_cfg.encap_rcv = gue_udp_recv;
 577                 tunnel_cfg.gro_receive = gue_gro_receive;
 578                 tunnel_cfg.gro_complete = gue_gro_complete;
 579                 break;
 580         default:
 581                 err = -EINVAL;
 582                 goto error;
 583         }
 584
 585         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 586
 587         sk->sk_allocation = GFP_ATOMIC;
 588
 589         err = fou_add_to_port_list(net, fou);
 590         if (err)
 591                 goto error;
 592
 593         if (sockp)
 594                 *sockp = sock;
 595
 596         return 0;
 597
 598 error:
 599         kfree(fou);
 600         if (sock)
 601                 udp_tunnel_sock_release(sock);
 602
 603         return err;
 604 }
 605
 606 static int fou_destroy(struct net *net, struct fou_cfg *cfg)
 607 {
 608         struct fou_net *fn = net_generic(net, fou_net_id);
 609         __be16 port = cfg->udp_config.local_udp_port;
 610         u8 family = cfg->udp_config.family;
 611         int err = -EINVAL;
 612         struct fou *fou;
 613
 614         mutex_lock(&fn->fou_lock);
 615         list_for_each_entry(fou, &fn->fou_list, list) {
 616                 if (fou->port == port && fou->family == family) {
 617                         fou_release(fou);
 618                         err = 0;
 619                         break;
 620                 }
 621         }
 622         mutex_unlock(&fn->fou_lock);
 623
 624         return err;
 625 }
 626
 627 static struct genl_family fou_nl_family;
 628
 629 static const struct nla_policy fou_nl_policy[FOU_ATTR_MAX + 1] = {
 630         [FOU_ATTR_PORT] = { .type = NLA_U16, },
 631         [FOU_ATTR_AF] = { .type = NLA_U8, },
 632         [FOU_ATTR_IPPROTO] = { .type = NLA_U8, },
 633         [FOU_ATTR_TYPE] = { .type = NLA_U8, },
 634         [FOU_ATTR_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG, },
 635 };
 636
 637 static int parse_nl_config(struct genl_info *info,
 638                            struct fou_cfg *cfg)
 639 {
 640         memset(cfg, 0, sizeof(*cfg));
 641
 642         cfg->udp_config.family = AF_INET;
 643
 644         if (info->attrs[FOU_ATTR_AF]) {
 645                 u8 family = nla_get_u8(info->attrs[FOU_ATTR_AF]);
 646
 647                 switch (family) {
 648                 case AF_INET:
 649                         break;
 650                 case AF_INET6:
 651                         cfg->udp_config.ipv6_v6only = 1;
 652                         break;
 653                 default:
 654                         return -EAFNOSUPPORT;
 655                 }
 656
 657                 cfg->udp_config.family = family;
 658         }
 659
 660         if (info->attrs[FOU_ATTR_PORT]) {
 661                 __be16 port = nla_get_be16(info->attrs[FOU_ATTR_PORT]);
 662
 663                 cfg->udp_config.local_udp_port = port;
 664         }
 665
 666         if (info->attrs[FOU_ATTR_IPPROTO])
 667                 cfg->protocol = nla_get_u8(info->attrs[FOU_ATTR_IPPROTO]);
 668
 669         if (info->attrs[FOU_ATTR_TYPE])
 670                 cfg->type = nla_get_u8(info->attrs[FOU_ATTR_TYPE]);
 671
 672         if (info->attrs[FOU_ATTR_REMCSUM_NOPARTIAL])
 673                 cfg->flags |= FOU_F_REMCSUM_NOPARTIAL;
 674
 675         return 0;
 676 }
 677
 678 static int fou_nl_cmd_add_port(struct sk_buff *skb, struct genl_info *info)
 679 {
 680         struct net *net = genl_info_net(info);
 681         struct fou_cfg cfg;
 682         int err;
 683
 684         err = parse_nl_config(info, &cfg);
 685         if (err)
 686                 return err;
 687
 688         return fou_create(net, &cfg, NULL);
 689 }
 690
 691 static int fou_nl_cmd_rm_port(struct sk_buff *skb, struct genl_info *info)
 692 {
 693         struct net *net = genl_info_net(info);
 694         struct fou_cfg cfg;
 695         int err;
 696
 697         err = parse_nl_config(info, &cfg);
 698         if (err)
 699                 return err;
 700
 701         return fou_destroy(net, &cfg);
 702 }
 703
 704 static int fou_fill_info(struct fou *fou, struct sk_buff *msg)
 705 {
 706         if (nla_put_u8(msg, FOU_ATTR_AF, fou->sock->sk->sk_family) ||
 707             nla_put_be16(msg, FOU_ATTR_PORT, fou->port) ||
 708             nla_put_u8(msg, FOU_ATTR_IPPROTO, fou->protocol) ||
 709             nla_put_u8(msg, FOU_ATTR_TYPE, fou->type))
 710                 return -1;
 711
 712         if (fou->flags & FOU_F_REMCSUM_NOPARTIAL)
 713                 if (nla_put_flag(msg, FOU_ATTR_REMCSUM_NOPARTIAL))
 714                         return -1;
 715         return 0;
 716 }
 717
 718 static int fou_dump_info(struct fou *fou, u32 portid, u32 seq,
 719                          u32 flags, struct sk_buff *skb, u8 cmd)
 720 {
 721         void *hdr;
 722
 723         hdr = genlmsg_put(skb, portid, seq, &fou_nl_family, flags, cmd);
 724         if (!hdr)
 725                 return -ENOMEM;
 726
 727         if (fou_fill_info(fou, skb) < 0)
 728                 goto nla_put_failure;
 729
 730         genlmsg_end(skb, hdr);
 731         return 0;
 732
 733 nla_put_failure:
 734         genlmsg_cancel(skb, hdr);
 735         return -EMSGSIZE;
 736 }
 737
 738 static int fou_nl_cmd_get_port(struct sk_buff *skb, struct genl_info *info)
 739 {
 740         struct net *net = genl_info_net(info);
 741         struct fou_net *fn = net_generic(net, fou_net_id);
 742         struct sk_buff *msg;
 743         struct fou_cfg cfg;
 744         struct fou *fout;
 745         __be16 port;
 746         u8 family;
 747         int ret;
 748
 749         ret = parse_nl_config(info, &cfg);
 750         if (ret)
 751                 return ret;
 752         port = cfg.udp_config.local_udp_port;
 753         if (port == 0)
 754                 return -EINVAL;
 755
 756         family = cfg.udp_config.family;
 757         if (family != AF_INET && family != AF_INET6)
 758                 return -EINVAL;
 759
 760         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 761         if (!msg)
 762                 return -ENOMEM;
 763
 764         ret = -ESRCH;
 765         mutex_lock(&fn->fou_lock);
 766         list_for_each_entry(fout, &fn->fou_list, list) {
 767                 if (port == fout->port && family == fout->family) {
 768                         ret = fou_dump_info(fout, info->snd_portid,
 769                                             info->snd_seq, 0, msg,
 770                                             info->genlhdr->cmd);
 771                         break;
 772                 }
 773         }
 774         mutex_unlock(&fn->fou_lock);
 775         if (ret < 0)
 776                 goto out_free;
 777
 778         return genlmsg_reply(msg, info);
 779
 780 out_free:
 781         nlmsg_free(msg);
 782         return ret;
 783 }
 784
 785 static int fou_nl_dump(struct sk_buff *skb, struct netlink_callback *cb)
 786 {
 787         struct net *net = sock_net(skb->sk);
 788         struct fou_net *fn = net_generic(net, fou_net_id);
 789         struct fou *fout;
 790         int idx = 0, ret;
 791
 792         mutex_lock(&fn->fou_lock);
 793         list_for_each_entry(fout, &fn->fou_list, list) {
 794                 if (idx++ < cb->args[0])
 795                         continue;
 796                 ret = fou_dump_info(fout, NETLINK_CB(cb->skb).portid,
 797                                     cb->nlh->nlmsg_seq, NLM_F_MULTI,
 798                                     skb, FOU_CMD_GET);
 799                 if (ret)
 800                         break;
 801         }
 802         mutex_unlock(&fn->fou_lock);
 803
 804         cb->args[0] = idx;
 805         return skb->len;
 806 }
 807
 808 static const struct genl_ops fou_nl_ops[] = {
 809         {
 810                 .cmd = FOU_CMD_ADD,
 811                 .doit = fou_nl_cmd_add_port,
 812                 .policy = fou_nl_policy,
 813                 .flags = GENL_ADMIN_PERM,
 814         },
 815         {
 816                 .cmd = FOU_CMD_DEL,
 817                 .doit = fou_nl_cmd_rm_port,
 818                 .policy = fou_nl_policy,
 819                 .flags = GENL_ADMIN_PERM,
 820         },
 821         {
 822                 .cmd = FOU_CMD_GET,
 823                 .doit = fou_nl_cmd_get_port,
 824                 .dumpit = fou_nl_dump,
 825                 .policy = fou_nl_policy,
 826         },
 827 };
 828
 829 static struct genl_family fou_nl_family __ro_after_init = {
 830         .hdrsize        = 0,
 831         .name           = FOU_GENL_NAME,
 832         .version        = FOU_GENL_VERSION,
 833         .maxattr        = FOU_ATTR_MAX,
 834         .netnsok        = true,
 835         .module         = THIS_MODULE,
 836         .ops            = fou_nl_ops,
 837         .n_ops          = ARRAY_SIZE(fou_nl_ops),
 838 };
 839
 840 size_t fou_encap_hlen(struct ip_tunnel_encap *e)
 841 {
 842         return sizeof(struct udphdr);
 843 }
 844 EXPORT_SYMBOL(fou_encap_hlen);
 845
 846 size_t gue_encap_hlen(struct ip_tunnel_encap *e)
 847 {
 848         size_t len;
 849         bool need_priv = false;
 850
 851         len = sizeof(struct udphdr) + sizeof(struct guehdr);
 852
 853         if (e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) {
 854                 len += GUE_PLEN_REMCSUM;
 855                 need_priv = true;
 856         }
 857
 858         len += need_priv ? GUE_LEN_PRIV : 0;
 859
 860         return len;
 861 }
 862 EXPORT_SYMBOL(gue_encap_hlen);
 863
 864 int __fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
 865                        u8 *protocol, __be16 *sport, int type)
 866 {
 867         int err;
 868
 869         err = iptunnel_handle_offloads(skb, type);
 870         if (err)
 871                 return err;
 872
 873         *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
 874                                                 skb, 0, 0, false);
 875
 876         return 0;
 877 }
 878 EXPORT_SYMBOL(__fou_build_header);
 879
 880 int __gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
 881                        u8 *protocol, __be16 *sport, int type)
 882 {
 883         struct guehdr *guehdr;
 884         size_t hdrlen, optlen = 0;
 885         void *data;
 886         bool need_priv = false;
 887         int err;
 888
 889         if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
 890             skb->ip_summed == CHECKSUM_PARTIAL) {
 891                 optlen += GUE_PLEN_REMCSUM;
 892                 type |= SKB_GSO_TUNNEL_REMCSUM;
 893                 need_priv = true;
 894         }
 895
 896         optlen += need_priv ? GUE_LEN_PRIV : 0;
 897
 898         err = iptunnel_handle_offloads(skb, type);
 899         if (err)
 900                 return err;
 901
 902         /* Get source port (based on flow hash) before skb_push */
 903         *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
 904                                                 skb, 0, 0, false);
 905
 906         hdrlen = sizeof(struct guehdr) + optlen;
 907
 908         skb_push(skb, hdrlen);
 909
 910         guehdr = (struct guehdr *)skb->data;
 911
 912         guehdr->control = 0;
 913         guehdr->version = 0;
 914         guehdr->hlen = optlen >> 2;
 915         guehdr->flags = 0;
 916         guehdr->proto_ctype = *protocol;
 917
 918         data = &guehdr[1];
 919
 920         if (need_priv) {
 921                 __be32 *flags = data;
 922
 923                 guehdr->flags |= GUE_FLAG_PRIV;
 924                 *flags = 0;
 925                 data += GUE_LEN_PRIV;
 926
 927                 if (type & SKB_GSO_TUNNEL_REMCSUM) {
 928                         u16 csum_start = skb_checksum_start_offset(skb);
 929                         __be16 *pd = data;
 930
 931                         if (csum_start < hdrlen)
 932                                 return -EINVAL;
 933
 934                         csum_start -= hdrlen;
 935                         pd[0] = htons(csum_start);
 936                         pd[1] = htons(csum_start + skb->csum_offset);
 937
 938                         if (!skb_is_gso(skb)) {
 939                                 skb->ip_summed = CHECKSUM_NONE;
 940                                 skb->encapsulation = 0;
 941                         }
 942
 943                         *flags |= GUE_PFLAG_REMCSUM;
 944                         data += GUE_PLEN_REMCSUM;
 945                 }
 946
 947         }
 948
 949         return 0;
 950 }
 951 EXPORT_SYMBOL(__gue_build_header);
 952
 953 #ifdef CONFIG_NET_FOU_IP_TUNNELS
 954
 955 static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
 956                           struct flowi4 *fl4, u8 *protocol, __be16 sport)
 957 {
 958         struct udphdr *uh;
 959
 960         skb_push(skb, sizeof(struct udphdr));
 961         skb_reset_transport_header(skb);
 962
 963         uh = udp_hdr(skb);
 964
 965         uh->dest = e->dport;
 966         uh->source = sport;
 967         uh->len = htons(skb->len);
 968         udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
 969                      fl4->saddr, fl4->daddr, skb->len);
 970
 971         *protocol = IPPROTO_UDP;
 972 }
 973
 974 static int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
 975                             u8 *protocol, struct flowi4 *fl4)
 976 {
 977         int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
 978                                                        SKB_GSO_UDP_TUNNEL;
 979         __be16 sport;
 980         int err;
 981
 982         err = __fou_build_header(skb, e, protocol, &sport, type);
 983         if (err)
 984                 return err;
 985
 986         fou_build_udp(skb, e, fl4, protocol, sport);
 987
 988         return 0;
 989 }
 990
 991 static int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
 992                             u8 *protocol, struct flowi4 *fl4)
 993 {
 994         int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
 995                                                        SKB_GSO_UDP_TUNNEL;
 996         __be16 sport;
 997         int err;
 998
 999         err = __gue_build_header(skb, e, protocol, &sport, type);
1000         if (err)
1001                 return err;
1002
1003         fou_build_udp(skb, e, fl4, protocol, sport);
1004
1005         return 0;
1006 }
1007
1008
1009 static const struct ip_tunnel_encap_ops fou_iptun_ops = {
1010         .encap_hlen = fou_encap_hlen,
1011         .build_header = fou_build_header,
1012 };
1013
1014 static const struct ip_tunnel_encap_ops gue_iptun_ops = {
1015         .encap_hlen = gue_encap_hlen,
1016         .build_header = gue_build_header,
1017 };
1018
1019 static int ip_tunnel_encap_add_fou_ops(void)
1020 {
1021         int ret;
1022
1023         ret = ip_tunnel_encap_add_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1024         if (ret < 0) {
1025                 pr_err("can't add fou ops\n");
1026                 return ret;
1027         }
1028
1029         ret = ip_tunnel_encap_add_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1030         if (ret < 0) {
1031                 pr_err("can't add gue ops\n");
1032                 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1033                 return ret;
1034         }
1035
1036         return 0;
1037 }
1038
1039 static void ip_tunnel_encap_del_fou_ops(void)
1040 {
1041         ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1042         ip_tunnel_encap_del_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1043 }
1044
1045 #else
1046
1047 static int ip_tunnel_encap_add_fou_ops(void)
1048 {
1049         return 0;
1050 }
1051
1052 static void ip_tunnel_encap_del_fou_ops(void)
1053 {
1054 }
1055
1056 #endif
1057
1058 static __net_init int fou_init_net(struct net *net)
1059 {
1060         struct fou_net *fn = net_generic(net, fou_net_id);
1061
1062         INIT_LIST_HEAD(&fn->fou_list);
1063         mutex_init(&fn->fou_lock);
1064         return 0;
1065 }
1066
1067 static __net_exit void fou_exit_net(struct net *net)
1068 {
1069         struct fou_net *fn = net_generic(net, fou_net_id);
1070         struct fou *fou, *next;
1071
1072         /* Close all the FOU sockets */
1073         mutex_lock(&fn->fou_lock);
1074         list_for_each_entry_safe(fou, next, &fn->fou_list, list)
1075                 fou_release(fou);
1076         mutex_unlock(&fn->fou_lock);
1077 }
1078
1079 static struct pernet_operations fou_net_ops = {
1080         .init = fou_init_net,
1081         .exit = fou_exit_net,
1082         .id   = &fou_net_id,
1083         .size = sizeof(struct fou_net),
1084 };
1085
1086 static int __init fou_init(void)
1087 {
1088         int ret;
1089
1090         ret = register_pernet_device(&fou_net_ops);
1091         if (ret)
1092                 goto exit;
1093
1094         ret = genl_register_family(&fou_nl_family);
1095         if (ret < 0)
1096                 goto unregister;
1097
1098         ret = ip_tunnel_encap_add_fou_ops();
1099         if (ret == 0)
1100                 return 0;
1101
1102         genl_unregister_family(&fou_nl_family);
1103 unregister:
1104         unregister_pernet_device(&fou_net_ops);
1105 exit:
1106         return ret;
1107 }
1108
1109 static void __exit fou_fini(void)
1110 {
1111         ip_tunnel_encap_del_fou_ops();
1112         genl_unregister_family(&fou_nl_family);
1113         unregister_pernet_device(&fou_net_ops);
1114 }
1115
1116 module_init(fou_init);
1117 module_exit(fou_fini);
1118 MODULE_AUTHOR("Tom Herbert <therbert@google.com>");
1119 MODULE_LICENSE("GPL");