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