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 = call_gro_receive(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         flush = 0;
 340
 341         for (p = *head; p; p = p->next) {
 342                 const struct guehdr *guehdr2;
 343
 344                 if (!NAPI_GRO_CB(p)->same_flow)
 345                         continue;
 346
 347                 guehdr2 = (struct guehdr *)(p->data + off);
 348
 349                 /* Compare base GUE header to be equal (covers
 350                  * hlen, version, proto_ctype, and flags.
 351                  */
 352                 if (guehdr->word != guehdr2->word) {
 353                         NAPI_GRO_CB(p)->same_flow = 0;
 354                         continue;
 355                 }
 356
 357                 /* Compare optional fields are the same. */
 358                 if (guehdr->hlen && memcmp(&guehdr[1], &guehdr2[1],
 359                                            guehdr->hlen << 2)) {
 360                         NAPI_GRO_CB(p)->same_flow = 0;
 361                         continue;
 362                 }
 363         }
 364
 365         /* We can clear the encap_mark for GUE as we are essentially doing
 366          * one of two possible things.  We are either adding an L4 tunnel
 367          * header to the outer L3 tunnel header, or we are are simply
 368          * treating the GRE tunnel header as though it is a UDP protocol
 369          * specific header such as VXLAN or GENEVE.
 370          */
 371         NAPI_GRO_CB(skb)->encap_mark = 0;
 372
 373         rcu_read_lock();
 374         offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
 375         ops = rcu_dereference(offloads[guehdr->proto_ctype]);
 376         if (WARN_ON_ONCE(!ops || !ops->callbacks.gro_receive))
 377                 goto out_unlock;
 378
 379         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
 380
 381 out_unlock:
 382         rcu_read_unlock();
 383 out:
 384         NAPI_GRO_CB(skb)->flush |= flush;
 385         skb_gro_remcsum_cleanup(skb, &grc);
 386
 387         return pp;
 388 }
 389
 390 static int gue_gro_complete(struct sk_buff *skb, int nhoff,
 391                             struct udp_offload *uoff)
 392 {
 393         const struct net_offload **offloads;
 394         struct guehdr *guehdr = (struct guehdr *)(skb->data + nhoff);
 395         const struct net_offload *ops;
 396         unsigned int guehlen;
 397         u8 proto;
 398         int err = -ENOENT;
 399
 400         proto = guehdr->proto_ctype;
 401
 402         guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
 403
 404         rcu_read_lock();
 405         offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
 406         ops = rcu_dereference(offloads[proto]);
 407         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 408                 goto out_unlock;
 409
 410         err = ops->callbacks.gro_complete(skb, nhoff + guehlen);
 411
 412 out_unlock:
 413         rcu_read_unlock();
 414         return err;
 415 }
 416
 417 static int fou_add_to_port_list(struct net *net, struct fou *fou)
 418 {
 419         struct fou_net *fn = net_generic(net, fou_net_id);
 420         struct fou *fout;
 421
 422         mutex_lock(&fn->fou_lock);
 423         list_for_each_entry(fout, &fn->fou_list, list) {
 424                 if (fou->port == fout->port) {
 425                         mutex_unlock(&fn->fou_lock);
 426                         return -EALREADY;
 427                 }
 428         }
 429
 430         list_add(&fou->list, &fn->fou_list);
 431         mutex_unlock(&fn->fou_lock);
 432
 433         return 0;
 434 }
 435
 436 static void fou_release(struct fou *fou)
 437 {
 438         struct socket *sock = fou->sock;
 439         struct sock *sk = sock->sk;
 440
 441         if (sk->sk_family == AF_INET)
 442                 udp_del_offload(&fou->udp_offloads);
 443         list_del(&fou->list);
 444         udp_tunnel_sock_release(sock);
 445
 446         kfree_rcu(fou, rcu);
 447 }
 448
 449 static int fou_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
 450 {
 451         udp_sk(sk)->encap_rcv = fou_udp_recv;
 452         fou->protocol = cfg->protocol;
 453         fou->udp_offloads.callbacks.gro_receive = fou_gro_receive;
 454         fou->udp_offloads.callbacks.gro_complete = fou_gro_complete;
 455         fou->udp_offloads.port = cfg->udp_config.local_udp_port;
 456         fou->udp_offloads.ipproto = cfg->protocol;
 457
 458         return 0;
 459 }
 460
 461 static int gue_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
 462 {
 463         udp_sk(sk)->encap_rcv = gue_udp_recv;
 464         fou->udp_offloads.callbacks.gro_receive = gue_gro_receive;
 465         fou->udp_offloads.callbacks.gro_complete = gue_gro_complete;
 466         fou->udp_offloads.port = cfg->udp_config.local_udp_port;
 467
 468         return 0;
 469 }
 470
 471 static int fou_create(struct net *net, struct fou_cfg *cfg,
 472                       struct socket **sockp)
 473 {
 474         struct socket *sock = NULL;
 475         struct fou *fou = NULL;
 476         struct sock *sk;
 477         int err;
 478
 479         /* Open UDP socket */
 480         err = udp_sock_create(net, &cfg->udp_config, &sock);
 481         if (err < 0)
 482                 goto error;
 483
 484         /* Allocate FOU port structure */
 485         fou = kzalloc(sizeof(*fou), GFP_KERNEL);
 486         if (!fou) {
 487                 err = -ENOMEM;
 488                 goto error;
 489         }
 490
 491         sk = sock->sk;
 492
 493         fou->flags = cfg->flags;
 494         fou->port = cfg->udp_config.local_udp_port;
 495
 496         /* Initial for fou type */
 497         switch (cfg->type) {
 498         case FOU_ENCAP_DIRECT:
 499                 err = fou_encap_init(sk, fou, cfg);
 500                 if (err)
 501                         goto error;
 502                 break;
 503         case FOU_ENCAP_GUE:
 504                 err = gue_encap_init(sk, fou, cfg);
 505                 if (err)
 506                         goto error;
 507                 break;
 508         default:
 509                 err = -EINVAL;
 510                 goto error;
 511         }
 512
 513         fou->type = cfg->type;
 514
 515         udp_sk(sk)->encap_type = 1;
 516         udp_encap_enable();
 517
 518         sk->sk_user_data = fou;
 519         fou->sock = sock;
 520
 521         inet_inc_convert_csum(sk);
 522
 523         sk->sk_allocation = GFP_ATOMIC;
 524
 525         if (cfg->udp_config.family == AF_INET) {
 526                 err = udp_add_offload(&fou->udp_offloads);
 527                 if (err)
 528                         goto error;
 529         }
 530
 531         err = fou_add_to_port_list(net, fou);
 532         if (err)
 533                 goto error;
 534
 535         if (sockp)
 536                 *sockp = sock;
 537
 538         return 0;
 539
 540 error:
 541         kfree(fou);
 542         if (sock)
 543                 udp_tunnel_sock_release(sock);
 544
 545         return err;
 546 }
 547
 548 static int fou_destroy(struct net *net, struct fou_cfg *cfg)
 549 {
 550         struct fou_net *fn = net_generic(net, fou_net_id);
 551         __be16 port = cfg->udp_config.local_udp_port;
 552         int err = -EINVAL;
 553         struct fou *fou;
 554
 555         mutex_lock(&fn->fou_lock);
 556         list_for_each_entry(fou, &fn->fou_list, list) {
 557                 if (fou->port == port) {
 558                         fou_release(fou);
 559                         err = 0;
 560                         break;
 561                 }
 562         }
 563         mutex_unlock(&fn->fou_lock);
 564
 565         return err;
 566 }
 567
 568 static struct genl_family fou_nl_family = {
 569         .id             = GENL_ID_GENERATE,
 570         .hdrsize        = 0,
 571         .name           = FOU_GENL_NAME,
 572         .version        = FOU_GENL_VERSION,
 573         .maxattr        = FOU_ATTR_MAX,
 574         .netnsok        = true,
 575 };
 576
 577 static struct nla_policy fou_nl_policy[FOU_ATTR_MAX + 1] = {
 578         [FOU_ATTR_PORT] = { .type = NLA_U16, },
 579         [FOU_ATTR_AF] = { .type = NLA_U8, },
 580         [FOU_ATTR_IPPROTO] = { .type = NLA_U8, },
 581         [FOU_ATTR_TYPE] = { .type = NLA_U8, },
 582         [FOU_ATTR_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG, },
 583 };
 584
 585 static int parse_nl_config(struct genl_info *info,
 586                            struct fou_cfg *cfg)
 587 {
 588         memset(cfg, 0, sizeof(*cfg));
 589
 590         cfg->udp_config.family = AF_INET;
 591
 592         if (info->attrs[FOU_ATTR_AF]) {
 593                 u8 family = nla_get_u8(info->attrs[FOU_ATTR_AF]);
 594
 595                 if (family != AF_INET)
 596                         return -EINVAL;
 597
 598                 cfg->udp_config.family = family;
 599         }
 600
 601         if (info->attrs[FOU_ATTR_PORT]) {
 602                 __be16 port = nla_get_be16(info->attrs[FOU_ATTR_PORT]);
 603
 604                 cfg->udp_config.local_udp_port = port;
 605         }
 606
 607         if (info->attrs[FOU_ATTR_IPPROTO])
 608                 cfg->protocol = nla_get_u8(info->attrs[FOU_ATTR_IPPROTO]);
 609
 610         if (info->attrs[FOU_ATTR_TYPE])
 611                 cfg->type = nla_get_u8(info->attrs[FOU_ATTR_TYPE]);
 612
 613         if (info->attrs[FOU_ATTR_REMCSUM_NOPARTIAL])
 614                 cfg->flags |= FOU_F_REMCSUM_NOPARTIAL;
 615
 616         return 0;
 617 }
 618
 619 static int fou_nl_cmd_add_port(struct sk_buff *skb, struct genl_info *info)
 620 {
 621         struct net *net = genl_info_net(info);
 622         struct fou_cfg cfg;
 623         int err;
 624
 625         err = parse_nl_config(info, &cfg);
 626         if (err)
 627                 return err;
 628
 629         return fou_create(net, &cfg, NULL);
 630 }
 631
 632 static int fou_nl_cmd_rm_port(struct sk_buff *skb, struct genl_info *info)
 633 {
 634         struct net *net = genl_info_net(info);
 635         struct fou_cfg cfg;
 636         int err;
 637
 638         err = parse_nl_config(info, &cfg);
 639         if (err)
 640                 return err;
 641
 642         return fou_destroy(net, &cfg);
 643 }
 644
 645 static int fou_fill_info(struct fou *fou, struct sk_buff *msg)
 646 {
 647         if (nla_put_u8(msg, FOU_ATTR_AF, fou->sock->sk->sk_family) ||
 648             nla_put_be16(msg, FOU_ATTR_PORT, fou->port) ||
 649             nla_put_u8(msg, FOU_ATTR_IPPROTO, fou->protocol) ||
 650             nla_put_u8(msg, FOU_ATTR_TYPE, fou->type))
 651                 return -1;
 652
 653         if (fou->flags & FOU_F_REMCSUM_NOPARTIAL)
 654                 if (nla_put_flag(msg, FOU_ATTR_REMCSUM_NOPARTIAL))
 655                         return -1;
 656         return 0;
 657 }
 658
 659 static int fou_dump_info(struct fou *fou, u32 portid, u32 seq,
 660                          u32 flags, struct sk_buff *skb, u8 cmd)
 661 {
 662         void *hdr;
 663
 664         hdr = genlmsg_put(skb, portid, seq, &fou_nl_family, flags, cmd);
 665         if (!hdr)
 666                 return -ENOMEM;
 667
 668         if (fou_fill_info(fou, skb) < 0)
 669                 goto nla_put_failure;
 670
 671         genlmsg_end(skb, hdr);
 672         return 0;
 673
 674 nla_put_failure:
 675         genlmsg_cancel(skb, hdr);
 676         return -EMSGSIZE;
 677 }
 678
 679 static int fou_nl_cmd_get_port(struct sk_buff *skb, struct genl_info *info)
 680 {
 681         struct net *net = genl_info_net(info);
 682         struct fou_net *fn = net_generic(net, fou_net_id);
 683         struct sk_buff *msg;
 684         struct fou_cfg cfg;
 685         struct fou *fout;
 686         __be16 port;
 687         int ret;
 688
 689         ret = parse_nl_config(info, &cfg);
 690         if (ret)
 691                 return ret;
 692         port = cfg.udp_config.local_udp_port;
 693         if (port == 0)
 694                 return -EINVAL;
 695
 696         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 697         if (!msg)
 698                 return -ENOMEM;
 699
 700         ret = -ESRCH;
 701         mutex_lock(&fn->fou_lock);
 702         list_for_each_entry(fout, &fn->fou_list, list) {
 703                 if (port == fout->port) {
 704                         ret = fou_dump_info(fout, info->snd_portid,
 705                                             info->snd_seq, 0, msg,
 706                                             info->genlhdr->cmd);
 707                         break;
 708                 }
 709         }
 710         mutex_unlock(&fn->fou_lock);
 711         if (ret < 0)
 712                 goto out_free;
 713
 714         return genlmsg_reply(msg, info);
 715
 716 out_free:
 717         nlmsg_free(msg);
 718         return ret;
 719 }
 720
 721 static int fou_nl_dump(struct sk_buff *skb, struct netlink_callback *cb)
 722 {
 723         struct net *net = sock_net(skb->sk);
 724         struct fou_net *fn = net_generic(net, fou_net_id);
 725         struct fou *fout;
 726         int idx = 0, ret;
 727
 728         mutex_lock(&fn->fou_lock);
 729         list_for_each_entry(fout, &fn->fou_list, list) {
 730                 if (idx++ < cb->args[0])
 731                         continue;
 732                 ret = fou_dump_info(fout, NETLINK_CB(cb->skb).portid,
 733                                     cb->nlh->nlmsg_seq, NLM_F_MULTI,
 734                                     skb, FOU_CMD_GET);
 735                 if (ret)
 736                         break;
 737         }
 738         mutex_unlock(&fn->fou_lock);
 739
 740         cb->args[0] = idx;
 741         return skb->len;
 742 }
 743
 744 static const struct genl_ops fou_nl_ops[] = {
 745         {
 746                 .cmd = FOU_CMD_ADD,
 747                 .doit = fou_nl_cmd_add_port,
 748                 .policy = fou_nl_policy,
 749                 .flags = GENL_ADMIN_PERM,
 750         },
 751         {
 752                 .cmd = FOU_CMD_DEL,
 753                 .doit = fou_nl_cmd_rm_port,
 754                 .policy = fou_nl_policy,
 755                 .flags = GENL_ADMIN_PERM,
 756         },
 757         {
 758                 .cmd = FOU_CMD_GET,
 759                 .doit = fou_nl_cmd_get_port,
 760                 .dumpit = fou_nl_dump,
 761                 .policy = fou_nl_policy,
 762         },
 763 };
 764
 765 size_t fou_encap_hlen(struct ip_tunnel_encap *e)
 766 {
 767         return sizeof(struct udphdr);
 768 }
 769 EXPORT_SYMBOL(fou_encap_hlen);
 770
 771 size_t gue_encap_hlen(struct ip_tunnel_encap *e)
 772 {
 773         size_t len;
 774         bool need_priv = false;
 775
 776         len = sizeof(struct udphdr) + sizeof(struct guehdr);
 777
 778         if (e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) {
 779                 len += GUE_PLEN_REMCSUM;
 780                 need_priv = true;
 781         }
 782
 783         len += need_priv ? GUE_LEN_PRIV : 0;
 784
 785         return len;
 786 }
 787 EXPORT_SYMBOL(gue_encap_hlen);
 788
 789 static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
 790                           struct flowi4 *fl4, u8 *protocol, __be16 sport)
 791 {
 792         struct udphdr *uh;
 793
 794         skb_push(skb, sizeof(struct udphdr));
 795         skb_reset_transport_header(skb);
 796
 797         uh = udp_hdr(skb);
 798
 799         uh->dest = e->dport;
 800         uh->source = sport;
 801         uh->len = htons(skb->len);
 802         uh->check = 0;
 803         udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
 804                      fl4->saddr, fl4->daddr, skb->len);
 805
 806         *protocol = IPPROTO_UDP;
 807 }
 808
 809 int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
 810                      u8 *protocol, struct flowi4 *fl4)
 811 {
 812         bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
 813         int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
 814         __be16 sport;
 815
 816         skb = iptunnel_handle_offloads(skb, csum, type);
 817
 818         if (IS_ERR(skb))
 819                 return PTR_ERR(skb);
 820
 821         sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
 822                                                skb, 0, 0, false);
 823         fou_build_udp(skb, e, fl4, protocol, sport);
 824
 825         return 0;
 826 }
 827 EXPORT_SYMBOL(fou_build_header);
 828
 829 int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
 830                      u8 *protocol, struct flowi4 *fl4)
 831 {
 832         bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
 833         int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
 834         struct guehdr *guehdr;
 835         size_t hdrlen, optlen = 0;
 836         __be16 sport;
 837         void *data;
 838         bool need_priv = false;
 839
 840         if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
 841             skb->ip_summed == CHECKSUM_PARTIAL) {
 842                 csum = false;
 843                 optlen += GUE_PLEN_REMCSUM;
 844                 type |= SKB_GSO_TUNNEL_REMCSUM;
 845                 need_priv = true;
 846         }
 847
 848         optlen += need_priv ? GUE_LEN_PRIV : 0;
 849
 850         skb = iptunnel_handle_offloads(skb, csum, type);
 851
 852         if (IS_ERR(skb))
 853                 return PTR_ERR(skb);
 854
 855         /* Get source port (based on flow hash) before skb_push */
 856         sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
 857                                                skb, 0, 0, false);
 858
 859         hdrlen = sizeof(struct guehdr) + optlen;
 860
 861         skb_push(skb, hdrlen);
 862
 863         guehdr = (struct guehdr *)skb->data;
 864
 865         guehdr->control = 0;
 866         guehdr->version = 0;
 867         guehdr->hlen = optlen >> 2;
 868         guehdr->flags = 0;
 869         guehdr->proto_ctype = *protocol;
 870
 871         data = &guehdr[1];
 872
 873         if (need_priv) {
 874                 __be32 *flags = data;
 875
 876                 guehdr->flags |= GUE_FLAG_PRIV;
 877                 *flags = 0;
 878                 data += GUE_LEN_PRIV;
 879
 880                 if (type & SKB_GSO_TUNNEL_REMCSUM) {
 881                         u16 csum_start = skb_checksum_start_offset(skb);
 882                         __be16 *pd = data;
 883
 884                         if (csum_start < hdrlen)
 885                                 return -EINVAL;
 886
 887                         csum_start -= hdrlen;
 888                         pd[0] = htons(csum_start);
 889                         pd[1] = htons(csum_start + skb->csum_offset);
 890
 891                         if (!skb_is_gso(skb)) {
 892                                 skb->ip_summed = CHECKSUM_NONE;
 893                                 skb->encapsulation = 0;
 894                         }
 895
 896                         *flags |= GUE_PFLAG_REMCSUM;
 897                         data += GUE_PLEN_REMCSUM;
 898                 }
 899
 900         }
 901
 902         fou_build_udp(skb, e, fl4, protocol, sport);
 903
 904         return 0;
 905 }
 906 EXPORT_SYMBOL(gue_build_header);
 907
 908 #ifdef CONFIG_NET_FOU_IP_TUNNELS
 909
 910 static const struct ip_tunnel_encap_ops fou_iptun_ops = {
 911         .encap_hlen = fou_encap_hlen,
 912         .build_header = fou_build_header,
 913 };
 914
 915 static const struct ip_tunnel_encap_ops gue_iptun_ops = {
 916         .encap_hlen = gue_encap_hlen,
 917         .build_header = gue_build_header,
 918 };
 919
 920 static int ip_tunnel_encap_add_fou_ops(void)
 921 {
 922         int ret;
 923
 924         ret = ip_tunnel_encap_add_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
 925         if (ret < 0) {
 926                 pr_err("can't add fou ops\n");
 927                 return ret;
 928         }
 929
 930         ret = ip_tunnel_encap_add_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
 931         if (ret < 0) {
 932                 pr_err("can't add gue ops\n");
 933                 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
 934                 return ret;
 935         }
 936
 937         return 0;
 938 }
 939
 940 static void ip_tunnel_encap_del_fou_ops(void)
 941 {
 942         ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
 943         ip_tunnel_encap_del_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
 944 }
 945
 946 #else
 947
 948 static int ip_tunnel_encap_add_fou_ops(void)
 949 {
 950         return 0;
 951 }
 952
 953 static void ip_tunnel_encap_del_fou_ops(void)
 954 {
 955 }
 956
 957 #endif
 958
 959 static __net_init int fou_init_net(struct net *net)
 960 {
 961         struct fou_net *fn = net_generic(net, fou_net_id);
 962
 963         INIT_LIST_HEAD(&fn->fou_list);
 964         mutex_init(&fn->fou_lock);
 965         return 0;
 966 }
 967
 968 static __net_exit void fou_exit_net(struct net *net)
 969 {
 970         struct fou_net *fn = net_generic(net, fou_net_id);
 971         struct fou *fou, *next;
 972
 973         /* Close all the FOU sockets */
 974         mutex_lock(&fn->fou_lock);
 975         list_for_each_entry_safe(fou, next, &fn->fou_list, list)
 976                 fou_release(fou);
 977         mutex_unlock(&fn->fou_lock);
 978 }
 979
 980 static struct pernet_operations fou_net_ops = {
 981         .init = fou_init_net,
 982         .exit = fou_exit_net,
 983         .id   = &fou_net_id,
 984         .size = sizeof(struct fou_net),
 985 };
 986
 987 static int __init fou_init(void)
 988 {
 989         int ret;
 990
 991         ret = register_pernet_device(&fou_net_ops);
 992         if (ret)
 993                 goto exit;
 994
 995         ret = genl_register_family_with_ops(&fou_nl_family,
 996                                             fou_nl_ops);
 997         if (ret < 0)
 998                 goto unregister;
 999
1000         ret = ip_tunnel_encap_add_fou_ops();
1001         if (ret == 0)
1002                 return 0;
1003
1004         genl_unregister_family(&fou_nl_family);
1005 unregister:
1006         unregister_pernet_device(&fou_net_ops);
1007 exit:
1008         return ret;
1009 }
1010
1011 static void __exit fou_fini(void)
1012 {
1013         ip_tunnel_encap_del_fou_ops();
1014         genl_unregister_family(&fou_nl_family);
1015         unregister_pernet_device(&fou_net_ops);
1016 }
1017
1018 module_init(fou_init);
1019 module_exit(fou_fini);
1020 MODULE_AUTHOR("Tom Herbert <therbert@google.com>");
1021 MODULE_LICENSE("GPL");