net/dccp/ipv6.c

   1 /*
   2  *      DCCP over IPv6
   3  *      Linux INET6 implementation
   4  *
   5  *      Based on net/dccp6/ipv6.c
   6  *
   7  *      Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  */
  14
  15 #include <linux/module.h>
  16 #include <linux/random.h>
  17 #include <linux/slab.h>
  18 #include <linux/xfrm.h>
  19
  20 #include <net/addrconf.h>
  21 #include <net/inet_common.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet_sock.h>
  24 #include <net/inet6_connection_sock.h>
  25 #include <net/inet6_hashtables.h>
  26 #include <net/ip6_route.h>
  27 #include <net/ipv6.h>
  28 #include <net/protocol.h>
  29 #include <net/transp_v6.h>
  30 #include <net/ip6_checksum.h>
  31 #include <net/xfrm.h>
  32 #include <net/secure_seq.h>
  33
  34 #include "dccp.h"
  35 #include "ipv6.h"
  36 #include "feat.h"
  37
  38 /* The per-net dccp.v6_ctl_sk is used for sending RSTs and ACKs */
  39
  40 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped;
  41 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops;
  42
  43 /* add pseudo-header to DCCP checksum stored in skb->csum */
  44 static inline __sum16 dccp_v6_csum_finish(struct sk_buff *skb,
  45                                       const struct in6_addr *saddr,
  46                                       const struct in6_addr *daddr)
  47 {
  48         return csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_DCCP, skb->csum);
  49 }
  50
  51 static inline void dccp_v6_send_check(struct sock *sk, struct sk_buff *skb)
  52 {
  53         struct ipv6_pinfo *np = inet6_sk(sk);
  54         struct dccp_hdr *dh = dccp_hdr(skb);
  55
  56         dccp_csum_outgoing(skb);
  57         dh->dccph_checksum = dccp_v6_csum_finish(skb, &np->saddr, &sk->sk_v6_daddr);
  58 }
  59
  60 static inline __u64 dccp_v6_init_sequence(struct sk_buff *skb)
  61 {
  62         return secure_dccpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
  63                                              ipv6_hdr(skb)->saddr.s6_addr32,
  64                                              dccp_hdr(skb)->dccph_dport,
  65                                              dccp_hdr(skb)->dccph_sport     );
  66
  67 }
  68
  69 static void dccp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
  70                         u8 type, u8 code, int offset, __be32 info)
  71 {
  72         const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
  73         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
  74         struct dccp_sock *dp;
  75         struct ipv6_pinfo *np;
  76         struct sock *sk;
  77         int err;
  78         __u64 seq;
  79         struct net *net = dev_net(skb->dev);
  80
  81         if (skb->len < offset + sizeof(*dh) ||
  82             skb->len < offset + __dccp_basic_hdr_len(dh)) {
  83                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
  84                                    ICMP6_MIB_INERRORS);
  85                 return;
  86         }
  87
  88         sk = __inet6_lookup_established(net, &dccp_hashinfo,
  89                                         &hdr->daddr, dh->dccph_dport,
  90                                         &hdr->saddr, ntohs(dh->dccph_sport),
  91                                         inet6_iif(skb));
  92
  93         if (!sk) {
  94                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
  95                                    ICMP6_MIB_INERRORS);
  96                 return;
  97         }
  98
  99         if (sk->sk_state == DCCP_TIME_WAIT) {
 100                 inet_twsk_put(inet_twsk(sk));
 101                 return;
 102         }
 103         seq = dccp_hdr_seq(dh);
 104         if (sk->sk_state == DCCP_NEW_SYN_RECV)
 105                 return dccp_req_err(sk, seq);
 106
 107         bh_lock_sock(sk);
 108         if (sock_owned_by_user(sk))
 109                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 110
 111         if (sk->sk_state == DCCP_CLOSED)
 112                 goto out;
 113
 114         dp = dccp_sk(sk);
 115         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 116             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 117                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 118                 goto out;
 119         }
 120
 121         np = inet6_sk(sk);
 122
 123         if (type == NDISC_REDIRECT) {
 124                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 125
 126                 if (dst)
 127                         dst->ops->redirect(dst, sk, skb);
 128                 goto out;
 129         }
 130
 131         if (type == ICMPV6_PKT_TOOBIG) {
 132                 struct dst_entry *dst = NULL;
 133
 134                 if (!ip6_sk_accept_pmtu(sk))
 135                         goto out;
 136
 137                 if (sock_owned_by_user(sk))
 138                         goto out;
 139                 if ((1 << sk->sk_state) & (DCCPF_LISTEN | DCCPF_CLOSED))
 140                         goto out;
 141
 142                 dst = inet6_csk_update_pmtu(sk, ntohl(info));
 143                 if (!dst)
 144                         goto out;
 145
 146                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst))
 147                         dccp_sync_mss(sk, dst_mtu(dst));
 148                 goto out;
 149         }
 150
 151         icmpv6_err_convert(type, code, &err);
 152
 153         /* Might be for an request_sock */
 154         switch (sk->sk_state) {
 155         case DCCP_REQUESTING:
 156         case DCCP_RESPOND:  /* Cannot happen.
 157                                It can, it SYNs are crossed. --ANK */
 158                 if (!sock_owned_by_user(sk)) {
 159                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 160                         sk->sk_err = err;
 161                         /*
 162                          * Wake people up to see the error
 163                          * (see connect in sock.c)
 164                          */
 165                         sk->sk_error_report(sk);
 166                         dccp_done(sk);
 167                 } else
 168                         sk->sk_err_soft = err;
 169                 goto out;
 170         }
 171
 172         if (!sock_owned_by_user(sk) && np->recverr) {
 173                 sk->sk_err = err;
 174                 sk->sk_error_report(sk);
 175         } else
 176                 sk->sk_err_soft = err;
 177
 178 out:
 179         bh_unlock_sock(sk);
 180         sock_put(sk);
 181 }
 182
 183
 184 static int dccp_v6_send_response(const struct sock *sk, struct request_sock *req)
 185 {
 186         struct inet_request_sock *ireq = inet_rsk(req);
 187         struct ipv6_pinfo *np = inet6_sk(sk);
 188         struct sk_buff *skb;
 189         struct in6_addr *final_p, final;
 190         struct flowi6 fl6;
 191         int err = -1;
 192         struct dst_entry *dst;
 193
 194         memset(&fl6, 0, sizeof(fl6));
 195         fl6.flowi6_proto = IPPROTO_DCCP;
 196         fl6.daddr = ireq->ir_v6_rmt_addr;
 197         fl6.saddr = ireq->ir_v6_loc_addr;
 198         fl6.flowlabel = 0;
 199         fl6.flowi6_oif = ireq->ir_iif;
 200         fl6.fl6_dport = ireq->ir_rmt_port;
 201         fl6.fl6_sport = htons(ireq->ir_num);
 202         security_req_classify_flow(req, flowi6_to_flowi(&fl6));
 203
 204
 205         rcu_read_lock();
 206         final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
 207         rcu_read_unlock();
 208
 209         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 210         if (IS_ERR(dst)) {
 211                 err = PTR_ERR(dst);
 212                 dst = NULL;
 213                 goto done;
 214         }
 215
 216         skb = dccp_make_response(sk, dst, req);
 217         if (skb != NULL) {
 218                 struct dccp_hdr *dh = dccp_hdr(skb);
 219
 220                 dh->dccph_checksum = dccp_v6_csum_finish(skb,
 221                                                          &ireq->ir_v6_loc_addr,
 222                                                          &ireq->ir_v6_rmt_addr);
 223                 fl6.daddr = ireq->ir_v6_rmt_addr;
 224                 rcu_read_lock();
 225                 err = ip6_xmit(sk, skb, &fl6, rcu_dereference(np->opt),
 226                                np->tclass);
 227                 rcu_read_unlock();
 228                 err = net_xmit_eval(err);
 229         }
 230
 231 done:
 232         dst_release(dst);
 233         return err;
 234 }
 235
 236 static void dccp_v6_reqsk_destructor(struct request_sock *req)
 237 {
 238         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 239         kfree_skb(inet_rsk(req)->pktopts);
 240 }
 241
 242 static void dccp_v6_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 243 {
 244         const struct ipv6hdr *rxip6h;
 245         struct sk_buff *skb;
 246         struct flowi6 fl6;
 247         struct net *net = dev_net(skb_dst(rxskb)->dev);
 248         struct sock *ctl_sk = net->dccp.v6_ctl_sk;
 249         struct dst_entry *dst;
 250
 251         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 252                 return;
 253
 254         if (!ipv6_unicast_destination(rxskb))
 255                 return;
 256
 257         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 258         if (skb == NULL)
 259                 return;
 260
 261         rxip6h = ipv6_hdr(rxskb);
 262         dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr,
 263                                                             &rxip6h->daddr);
 264
 265         memset(&fl6, 0, sizeof(fl6));
 266         fl6.daddr = rxip6h->saddr;
 267         fl6.saddr = rxip6h->daddr;
 268
 269         fl6.flowi6_proto = IPPROTO_DCCP;
 270         fl6.flowi6_oif = inet6_iif(rxskb);
 271         fl6.fl6_dport = dccp_hdr(skb)->dccph_dport;
 272         fl6.fl6_sport = dccp_hdr(skb)->dccph_sport;
 273         security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6));
 274
 275         /* sk = NULL, but it is safe for now. RST socket required. */
 276         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 277         if (!IS_ERR(dst)) {
 278                 skb_dst_set(skb, dst);
 279                 ip6_xmit(ctl_sk, skb, &fl6, NULL, 0);
 280                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 281                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 282                 return;
 283         }
 284
 285         kfree_skb(skb);
 286 }
 287
 288 static struct request_sock_ops dccp6_request_sock_ops = {
 289         .family         = AF_INET6,
 290         .obj_size       = sizeof(struct dccp6_request_sock),
 291         .rtx_syn_ack    = dccp_v6_send_response,
 292         .send_ack       = dccp_reqsk_send_ack,
 293         .destructor     = dccp_v6_reqsk_destructor,
 294         .send_reset     = dccp_v6_ctl_send_reset,
 295         .syn_ack_timeout = dccp_syn_ack_timeout,
 296 };
 297
 298 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 299 {
 300         struct request_sock *req;
 301         struct dccp_request_sock *dreq;
 302         struct inet_request_sock *ireq;
 303         struct ipv6_pinfo *np = inet6_sk(sk);
 304         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 305         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 306
 307         if (skb->protocol == htons(ETH_P_IP))
 308                 return dccp_v4_conn_request(sk, skb);
 309
 310         if (!ipv6_unicast_destination(skb))
 311                 return 0;       /* discard, don't send a reset here */
 312
 313         if (dccp_bad_service_code(sk, service)) {
 314                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 315                 goto drop;
 316         }
 317         /*
 318          * There are no SYN attacks on IPv6, yet...
 319          */
 320         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 321         if (inet_csk_reqsk_queue_is_full(sk))
 322                 goto drop;
 323
 324         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 325                 goto drop;
 326
 327         req = inet_reqsk_alloc(&dccp6_request_sock_ops, sk, true);
 328         if (req == NULL)
 329                 goto drop;
 330
 331         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 332                 goto drop_and_free;
 333
 334         dreq = dccp_rsk(req);
 335         if (dccp_parse_options(sk, dreq, skb))
 336                 goto drop_and_free;
 337
 338         if (security_inet_conn_request(sk, skb, req))
 339                 goto drop_and_free;
 340
 341         ireq = inet_rsk(req);
 342         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 343         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 344         ireq->ireq_family = AF_INET6;
 345
 346         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
 347             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 348             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 349                 atomic_inc(&skb->users);
 350                 ireq->pktopts = skb;
 351         }
 352         ireq->ir_iif = sk->sk_bound_dev_if;
 353
 354         /* So that link locals have meaning */
 355         if (!sk->sk_bound_dev_if &&
 356             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 357                 ireq->ir_iif = inet6_iif(skb);
 358
 359         /*
 360          * Step 3: Process LISTEN state
 361          *
 362          *   Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 363          *
 364          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 365          */
 366         dreq->dreq_isr     = dcb->dccpd_seq;
 367         dreq->dreq_gsr     = dreq->dreq_isr;
 368         dreq->dreq_iss     = dccp_v6_init_sequence(skb);
 369         dreq->dreq_gss     = dreq->dreq_iss;
 370         dreq->dreq_service = service;
 371
 372         if (dccp_v6_send_response(sk, req))
 373                 goto drop_and_free;
 374
 375         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 376         return 0;
 377
 378 drop_and_free:
 379         reqsk_free(req);
 380 drop:
 381         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 382         return -1;
 383 }
 384
 385 static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,
 386                                               struct sk_buff *skb,
 387                                               struct request_sock *req,
 388                                               struct dst_entry *dst,
 389                                               struct request_sock *req_unhash,
 390                                               bool *own_req)
 391 {
 392         struct inet_request_sock *ireq = inet_rsk(req);
 393         struct ipv6_pinfo *newnp;
 394         const struct ipv6_pinfo *np = inet6_sk(sk);
 395         struct ipv6_txoptions *opt;
 396         struct inet_sock *newinet;
 397         struct dccp6_sock *newdp6;
 398         struct sock *newsk;
 399
 400         if (skb->protocol == htons(ETH_P_IP)) {
 401                 /*
 402                  *      v6 mapped
 403                  */
 404                 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst,
 405                                                   req_unhash, own_req);
 406                 if (newsk == NULL)
 407                         return NULL;
 408
 409                 newdp6 = (struct dccp6_sock *)newsk;
 410                 newinet = inet_sk(newsk);
 411                 newinet->pinet6 = &newdp6->inet6;
 412                 newnp = inet6_sk(newsk);
 413
 414                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 415
 416                 newnp->saddr = newsk->sk_v6_rcv_saddr;
 417
 418                 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
 419                 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
 420                 newnp->pktoptions  = NULL;
 421                 newnp->opt         = NULL;
 422                 newnp->mcast_oif   = inet6_iif(skb);
 423                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
 424
 425                 /*
 426                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
 427                  * here, dccp_create_openreq_child now does this for us, see the comment in
 428                  * that function for the gory details. -acme
 429                  */
 430
 431                 /* It is tricky place. Until this moment IPv4 tcp
 432                    worked with IPv6 icsk.icsk_af_ops.
 433                    Sync it now.
 434                  */
 435                 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
 436
 437                 return newsk;
 438         }
 439
 440
 441         if (sk_acceptq_is_full(sk))
 442                 goto out_overflow;
 443
 444         if (!dst) {
 445                 struct flowi6 fl6;
 446
 447                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_DCCP);
 448                 if (!dst)
 449                         goto out;
 450         }
 451
 452         newsk = dccp_create_openreq_child(sk, req, skb);
 453         if (newsk == NULL)
 454                 goto out_nonewsk;
 455
 456         /*
 457          * No need to charge this sock to the relevant IPv6 refcnt debug socks
 458          * count here, dccp_create_openreq_child now does this for us, see the
 459          * comment in that function for the gory details. -acme
 460          */
 461
 462         ip6_dst_store(newsk, dst, NULL, NULL);
 463         newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
 464                                                       NETIF_F_TSO);
 465         newdp6 = (struct dccp6_sock *)newsk;
 466         newinet = inet_sk(newsk);
 467         newinet->pinet6 = &newdp6->inet6;
 468         newnp = inet6_sk(newsk);
 469
 470         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 471
 472         newsk->sk_v6_daddr      = ireq->ir_v6_rmt_addr;
 473         newnp->saddr            = ireq->ir_v6_loc_addr;
 474         newsk->sk_v6_rcv_saddr  = ireq->ir_v6_loc_addr;
 475         newsk->sk_bound_dev_if  = ireq->ir_iif;
 476
 477         /* Now IPv6 options...
 478
 479            First: no IPv4 options.
 480          */
 481         newinet->inet_opt = NULL;
 482
 483         /* Clone RX bits */
 484         newnp->rxopt.all = np->rxopt.all;
 485
 486         newnp->pktoptions = NULL;
 487         newnp->opt        = NULL;
 488         newnp->mcast_oif  = inet6_iif(skb);
 489         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
 490
 491         /*
 492          * Clone native IPv6 options from listening socket (if any)
 493          *
 494          * Yes, keeping reference count would be much more clever, but we make
 495          * one more one thing there: reattach optmem to newsk.
 496          */
 497         opt = rcu_dereference(np->opt);
 498         if (opt) {
 499                 opt = ipv6_dup_options(newsk, opt);
 500                 RCU_INIT_POINTER(newnp->opt, opt);
 501         }
 502         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 503         if (opt)
 504                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
 505                                                     opt->opt_flen;
 506
 507         dccp_sync_mss(newsk, dst_mtu(dst));
 508
 509         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
 510         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
 511
 512         if (__inet_inherit_port(sk, newsk) < 0) {
 513                 inet_csk_prepare_forced_close(newsk);
 514                 dccp_done(newsk);
 515                 goto out;
 516         }
 517         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 518         /* Clone pktoptions received with SYN, if we own the req */
 519         if (*own_req && ireq->pktopts) {
 520                 newnp->pktoptions = skb_clone(ireq->pktopts, GFP_ATOMIC);
 521                 consume_skb(ireq->pktopts);
 522                 ireq->pktopts = NULL;
 523                 if (newnp->pktoptions)
 524                         skb_set_owner_r(newnp->pktoptions, newsk);
 525         }
 526
 527         return newsk;
 528
 529 out_overflow:
 530         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 531 out_nonewsk:
 532         dst_release(dst);
 533 out:
 534         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 535         return NULL;
 536 }
 537
 538 /* The socket must have it's spinlock held when we get
 539  * here.
 540  *
 541  * We have a potential double-lock case here, so even when
 542  * doing backlog processing we use the BH locking scheme.
 543  * This is because we cannot sleep with the original spinlock
 544  * held.
 545  */
 546 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 547 {
 548         struct ipv6_pinfo *np = inet6_sk(sk);
 549         struct sk_buff *opt_skb = NULL;
 550
 551         /* Imagine: socket is IPv6. IPv4 packet arrives,
 552            goes to IPv4 receive handler and backlogged.
 553            From backlog it always goes here. Kerboom...
 554            Fortunately, dccp_rcv_established and rcv_established
 555            handle them correctly, but it is not case with
 556            dccp_v6_hnd_req and dccp_v6_ctl_send_reset().   --ANK
 557          */
 558
 559         if (skb->protocol == htons(ETH_P_IP))
 560                 return dccp_v4_do_rcv(sk, skb);
 561
 562         if (sk_filter(sk, skb))
 563                 goto discard;
 564
 565         /*
 566          * socket locking is here for SMP purposes as backlog rcv is currently
 567          * called with bh processing disabled.
 568          */
 569
 570         /* Do Stevens' IPV6_PKTOPTIONS.
 571
 572            Yes, guys, it is the only place in our code, where we
 573            may make it not affecting IPv4.
 574            The rest of code is protocol independent,
 575            and I do not like idea to uglify IPv4.
 576
 577            Actually, all the idea behind IPV6_PKTOPTIONS
 578            looks not very well thought. For now we latch
 579            options, received in the last packet, enqueued
 580            by tcp. Feel free to propose better solution.
 581                                                --ANK (980728)
 582          */
 583         if (np->rxopt.all)
 584         /*
 585          * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
 586          *        (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
 587          */
 588                 opt_skb = skb_clone(skb, GFP_ATOMIC);
 589
 590         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 591                 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
 592                         goto reset;
 593                 if (opt_skb) {
 594                         /* XXX This is where we would goto ipv6_pktoptions. */
 595                         __kfree_skb(opt_skb);
 596                 }
 597                 return 0;
 598         }
 599
 600         /*
 601          *  Step 3: Process LISTEN state
 602          *     If S.state == LISTEN,
 603          *       If P.type == Request or P contains a valid Init Cookie option,
 604          *            (* Must scan the packet's options to check for Init
 605          *               Cookies.  Only Init Cookies are processed here,
 606          *               however; other options are processed in Step 8.  This
 607          *               scan need only be performed if the endpoint uses Init
 608          *               Cookies *)
 609          *            (* Generate a new socket and switch to that socket *)
 610          *            Set S := new socket for this port pair
 611          *            S.state = RESPOND
 612          *            Choose S.ISS (initial seqno) or set from Init Cookies
 613          *            Initialize S.GAR := S.ISS
 614          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 615          *            Continue with S.state == RESPOND
 616          *            (* A Response packet will be generated in Step 11 *)
 617          *       Otherwise,
 618          *            Generate Reset(No Connection) unless P.type == Reset
 619          *            Drop packet and return
 620          *
 621          * NOTE: the check for the packet types is done in
 622          *       dccp_rcv_state_process
 623          */
 624
 625         if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
 626                 goto reset;
 627         if (opt_skb) {
 628                 /* XXX This is where we would goto ipv6_pktoptions. */
 629                 __kfree_skb(opt_skb);
 630         }
 631         return 0;
 632
 633 reset:
 634         dccp_v6_ctl_send_reset(sk, skb);
 635 discard:
 636         if (opt_skb != NULL)
 637                 __kfree_skb(opt_skb);
 638         kfree_skb(skb);
 639         return 0;
 640 }
 641
 642 static int dccp_v6_rcv(struct sk_buff *skb)
 643 {
 644         const struct dccp_hdr *dh;
 645         struct sock *sk;
 646         int min_cov;
 647
 648         /* Step 1: Check header basics */
 649
 650         if (dccp_invalid_packet(skb))
 651                 goto discard_it;
 652
 653         /* Step 1: If header checksum is incorrect, drop packet and return. */
 654         if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
 655                                      &ipv6_hdr(skb)->daddr)) {
 656                 DCCP_WARN("dropped packet with invalid checksum\n");
 657                 goto discard_it;
 658         }
 659
 660         dh = dccp_hdr(skb);
 661
 662         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 663         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 664
 665         if (dccp_packet_without_ack(skb))
 666                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 667         else
 668                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 669
 670 lookup:
 671         sk = __inet6_lookup_skb(&dccp_hashinfo, skb,
 672                                 dh->dccph_sport, dh->dccph_dport,
 673                                 inet6_iif(skb));
 674         if (!sk) {
 675                 dccp_pr_debug("failed to look up flow ID in table and "
 676                               "get corresponding socket\n");
 677                 goto no_dccp_socket;
 678         }
 679
 680         /*
 681          * Step 2:
 682          *      ... or S.state == TIMEWAIT,
 683          *              Generate Reset(No Connection) unless P.type == Reset
 684          *              Drop packet and return
 685          */
 686         if (sk->sk_state == DCCP_TIME_WAIT) {
 687                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 688                 inet_twsk_put(inet_twsk(sk));
 689                 goto no_dccp_socket;
 690         }
 691
 692         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 693                 struct request_sock *req = inet_reqsk(sk);
 694                 struct sock *nsk = NULL;
 695
 696                 sk = req->rsk_listener;
 697                 if (likely(sk->sk_state == DCCP_LISTEN)) {
 698                         nsk = dccp_check_req(sk, skb, req);
 699                 } else {
 700                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 701                         goto lookup;
 702                 }
 703                 if (!nsk) {
 704                         reqsk_put(req);
 705                         goto discard_it;
 706                 }
 707                 if (nsk == sk) {
 708                         sock_hold(sk);
 709                         reqsk_put(req);
 710                 } else if (dccp_child_process(sk, nsk, skb)) {
 711                         dccp_v6_ctl_send_reset(sk, skb);
 712                         goto discard_it;
 713                 } else {
 714                         return 0;
 715                 }
 716         }
 717         /*
 718          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 719          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 720          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 721          */
 722         min_cov = dccp_sk(sk)->dccps_pcrlen;
 723         if (dh->dccph_cscov  &&  (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 724                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 725                               dh->dccph_cscov, min_cov);
 726                 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
 727                 goto discard_and_relse;
 728         }
 729
 730         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
 731                 goto discard_and_relse;
 732
 733         return sk_receive_skb(sk, skb, 1) ? -1 : 0;
 734
 735 no_dccp_socket:
 736         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 737                 goto discard_it;
 738         /*
 739          * Step 2:
 740          *      If no socket ...
 741          *              Generate Reset(No Connection) unless P.type == Reset
 742          *              Drop packet and return
 743          */
 744         if (dh->dccph_type != DCCP_PKT_RESET) {
 745                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 746                                         DCCP_RESET_CODE_NO_CONNECTION;
 747                 dccp_v6_ctl_send_reset(sk, skb);
 748         }
 749
 750 discard_it:
 751         kfree_skb(skb);
 752         return 0;
 753
 754 discard_and_relse:
 755         sock_put(sk);
 756         goto discard_it;
 757 }
 758
 759 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 760                            int addr_len)
 761 {
 762         struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr;
 763         struct inet_connection_sock *icsk = inet_csk(sk);
 764         struct inet_sock *inet = inet_sk(sk);
 765         struct ipv6_pinfo *np = inet6_sk(sk);
 766         struct dccp_sock *dp = dccp_sk(sk);
 767         struct in6_addr *saddr = NULL, *final_p, final;
 768         struct ipv6_txoptions *opt;
 769         struct flowi6 fl6;
 770         struct dst_entry *dst;
 771         int addr_type;
 772         int err;
 773
 774         dp->dccps_role = DCCP_ROLE_CLIENT;
 775
 776         if (addr_len < SIN6_LEN_RFC2133)
 777                 return -EINVAL;
 778
 779         if (usin->sin6_family != AF_INET6)
 780                 return -EAFNOSUPPORT;
 781
 782         memset(&fl6, 0, sizeof(fl6));
 783
 784         if (np->sndflow) {
 785                 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK;
 786                 IP6_ECN_flow_init(fl6.flowlabel);
 787                 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) {
 788                         struct ip6_flowlabel *flowlabel;
 789                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 790                         if (flowlabel == NULL)
 791                                 return -EINVAL;
 792                         fl6_sock_release(flowlabel);
 793                 }
 794         }
 795         /*
 796          * connect() to INADDR_ANY means loopback (BSD'ism).
 797          */
 798         if (ipv6_addr_any(&usin->sin6_addr))
 799                 usin->sin6_addr.s6_addr[15] = 1;
 800
 801         addr_type = ipv6_addr_type(&usin->sin6_addr);
 802
 803         if (addr_type & IPV6_ADDR_MULTICAST)
 804                 return -ENETUNREACH;
 805
 806         if (addr_type & IPV6_ADDR_LINKLOCAL) {
 807                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 808                     usin->sin6_scope_id) {
 809                         /* If interface is set while binding, indices
 810                          * must coincide.
 811                          */
 812                         if (sk->sk_bound_dev_if &&
 813                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 814                                 return -EINVAL;
 815
 816                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 817                 }
 818
 819                 /* Connect to link-local address requires an interface */
 820                 if (!sk->sk_bound_dev_if)
 821                         return -EINVAL;
 822         }
 823
 824         sk->sk_v6_daddr = usin->sin6_addr;
 825         np->flow_label = fl6.flowlabel;
 826
 827         /*
 828          * DCCP over IPv4
 829          */
 830         if (addr_type == IPV6_ADDR_MAPPED) {
 831                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 832                 struct sockaddr_in sin;
 833
 834                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 835
 836                 if (__ipv6_only_sock(sk))
 837                         return -ENETUNREACH;
 838
 839                 sin.sin_family = AF_INET;
 840                 sin.sin_port = usin->sin6_port;
 841                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 842
 843                 icsk->icsk_af_ops = &dccp_ipv6_mapped;
 844                 sk->sk_backlog_rcv = dccp_v4_do_rcv;
 845
 846                 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 847                 if (err) {
 848                         icsk->icsk_ext_hdr_len = exthdrlen;
 849                         icsk->icsk_af_ops = &dccp_ipv6_af_ops;
 850                         sk->sk_backlog_rcv = dccp_v6_do_rcv;
 851                         goto failure;
 852                 }
 853                 np->saddr = sk->sk_v6_rcv_saddr;
 854                 return err;
 855         }
 856
 857         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 858                 saddr = &sk->sk_v6_rcv_saddr;
 859
 860         fl6.flowi6_proto = IPPROTO_DCCP;
 861         fl6.daddr = sk->sk_v6_daddr;
 862         fl6.saddr = saddr ? *saddr : np->saddr;
 863         fl6.flowi6_oif = sk->sk_bound_dev_if;
 864         fl6.fl6_dport = usin->sin6_port;
 865         fl6.fl6_sport = inet->inet_sport;
 866         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 867
 868         opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
 869         final_p = fl6_update_dst(&fl6, opt, &final);
 870
 871         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 872         if (IS_ERR(dst)) {
 873                 err = PTR_ERR(dst);
 874                 goto failure;
 875         }
 876
 877         if (saddr == NULL) {
 878                 saddr = &fl6.saddr;
 879                 sk->sk_v6_rcv_saddr = *saddr;
 880         }
 881
 882         /* set the source address */
 883         np->saddr = *saddr;
 884         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 885
 886         ip6_dst_store(sk, dst, NULL, NULL);
 887
 888         icsk->icsk_ext_hdr_len = 0;
 889         if (opt)
 890                 icsk->icsk_ext_hdr_len = opt->opt_flen + opt->opt_nflen;
 891
 892         inet->inet_dport = usin->sin6_port;
 893
 894         dccp_set_state(sk, DCCP_REQUESTING);
 895         err = inet6_hash_connect(&dccp_death_row, sk);
 896         if (err)
 897                 goto late_failure;
 898
 899         dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32,
 900                                                       sk->sk_v6_daddr.s6_addr32,
 901                                                       inet->inet_sport,
 902                                                       inet->inet_dport);
 903         err = dccp_connect(sk);
 904         if (err)
 905                 goto late_failure;
 906
 907         return 0;
 908
 909 late_failure:
 910         dccp_set_state(sk, DCCP_CLOSED);
 911         __sk_dst_reset(sk);
 912 failure:
 913         inet->inet_dport = 0;
 914         sk->sk_route_caps = 0;
 915         return err;
 916 }
 917
 918 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = {
 919         .queue_xmit        = inet6_csk_xmit,
 920         .send_check        = dccp_v6_send_check,
 921         .rebuild_header    = inet6_sk_rebuild_header,
 922         .conn_request      = dccp_v6_conn_request,
 923         .syn_recv_sock     = dccp_v6_request_recv_sock,
 924         .net_header_len    = sizeof(struct ipv6hdr),
 925         .setsockopt        = ipv6_setsockopt,
 926         .getsockopt        = ipv6_getsockopt,
 927         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 928         .sockaddr_len      = sizeof(struct sockaddr_in6),
 929         .bind_conflict     = inet6_csk_bind_conflict,
 930 #ifdef CONFIG_COMPAT
 931         .compat_setsockopt = compat_ipv6_setsockopt,
 932         .compat_getsockopt = compat_ipv6_getsockopt,
 933 #endif
 934 };
 935
 936 /*
 937  *      DCCP over IPv4 via INET6 API
 938  */
 939 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = {
 940         .queue_xmit        = ip_queue_xmit,
 941         .send_check        = dccp_v4_send_check,
 942         .rebuild_header    = inet_sk_rebuild_header,
 943         .conn_request      = dccp_v6_conn_request,
 944         .syn_recv_sock     = dccp_v6_request_recv_sock,
 945         .net_header_len    = sizeof(struct iphdr),
 946         .setsockopt        = ipv6_setsockopt,
 947         .getsockopt        = ipv6_getsockopt,
 948         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 949         .sockaddr_len      = sizeof(struct sockaddr_in6),
 950 #ifdef CONFIG_COMPAT
 951         .compat_setsockopt = compat_ipv6_setsockopt,
 952         .compat_getsockopt = compat_ipv6_getsockopt,
 953 #endif
 954 };
 955
 956 /* NOTE: A lot of things set to zero explicitly by call to
 957  *       sk_alloc() so need not be done here.
 958  */
 959 static int dccp_v6_init_sock(struct sock *sk)
 960 {
 961         static __u8 dccp_v6_ctl_sock_initialized;
 962         int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized);
 963
 964         if (err == 0) {
 965                 if (unlikely(!dccp_v6_ctl_sock_initialized))
 966                         dccp_v6_ctl_sock_initialized = 1;
 967                 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops;
 968         }
 969
 970         return err;
 971 }
 972
 973 static void dccp_v6_destroy_sock(struct sock *sk)
 974 {
 975         dccp_destroy_sock(sk);
 976         inet6_destroy_sock(sk);
 977 }
 978
 979 static struct timewait_sock_ops dccp6_timewait_sock_ops = {
 980         .twsk_obj_size  = sizeof(struct dccp6_timewait_sock),
 981 };
 982
 983 static struct proto dccp_v6_prot = {
 984         .name              = "DCCPv6",
 985         .owner             = THIS_MODULE,
 986         .close             = dccp_close,
 987         .connect           = dccp_v6_connect,
 988         .disconnect        = dccp_disconnect,
 989         .ioctl             = dccp_ioctl,
 990         .init              = dccp_v6_init_sock,
 991         .setsockopt        = dccp_setsockopt,
 992         .getsockopt        = dccp_getsockopt,
 993         .sendmsg           = dccp_sendmsg,
 994         .recvmsg           = dccp_recvmsg,
 995         .backlog_rcv       = dccp_v6_do_rcv,
 996         .hash              = inet6_hash,
 997         .unhash            = inet_unhash,
 998         .accept            = inet_csk_accept,
 999         .get_port          = inet_csk_get_port,
1000         .shutdown          = dccp_shutdown,
1001         .destroy           = dccp_v6_destroy_sock,
1002         .orphan_count      = &dccp_orphan_count,
1003         .max_header        = MAX_DCCP_HEADER,
1004         .obj_size          = sizeof(struct dccp6_sock),
1005         .slab_flags        = SLAB_DESTROY_BY_RCU,
1006         .rsk_prot          = &dccp6_request_sock_ops,
1007         .twsk_prot         = &dccp6_timewait_sock_ops,
1008         .h.hashinfo        = &dccp_hashinfo,
1009 #ifdef CONFIG_COMPAT
1010         .compat_setsockopt = compat_dccp_setsockopt,
1011         .compat_getsockopt = compat_dccp_getsockopt,
1012 #endif
1013 };
1014
1015 static const struct inet6_protocol dccp_v6_protocol = {
1016         .handler        = dccp_v6_rcv,
1017         .err_handler    = dccp_v6_err,
1018         .flags          = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1019 };
1020
1021 static const struct proto_ops inet6_dccp_ops = {
1022         .family            = PF_INET6,
1023         .owner             = THIS_MODULE,
1024         .release           = inet6_release,
1025         .bind              = inet6_bind,
1026         .connect           = inet_stream_connect,
1027         .socketpair        = sock_no_socketpair,
1028         .accept            = inet_accept,
1029         .getname           = inet6_getname,
1030         .poll              = dccp_poll,
1031         .ioctl             = inet6_ioctl,
1032         .listen            = inet_dccp_listen,
1033         .shutdown          = inet_shutdown,
1034         .setsockopt        = sock_common_setsockopt,
1035         .getsockopt        = sock_common_getsockopt,
1036         .sendmsg           = inet_sendmsg,
1037         .recvmsg           = sock_common_recvmsg,
1038         .mmap              = sock_no_mmap,
1039         .sendpage          = sock_no_sendpage,
1040 #ifdef CONFIG_COMPAT
1041         .compat_setsockopt = compat_sock_common_setsockopt,
1042         .compat_getsockopt = compat_sock_common_getsockopt,
1043 #endif
1044 };
1045
1046 static struct inet_protosw dccp_v6_protosw = {
1047         .type           = SOCK_DCCP,
1048         .protocol       = IPPROTO_DCCP,
1049         .prot           = &dccp_v6_prot,
1050         .ops            = &inet6_dccp_ops,
1051         .flags          = INET_PROTOSW_ICSK,
1052 };
1053
1054 static int __net_init dccp_v6_init_net(struct net *net)
1055 {
1056         if (dccp_hashinfo.bhash == NULL)
1057                 return -ESOCKTNOSUPPORT;
1058
1059         return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6,
1060                                     SOCK_DCCP, IPPROTO_DCCP, net);
1061 }
1062
1063 static void __net_exit dccp_v6_exit_net(struct net *net)
1064 {
1065         inet_ctl_sock_destroy(net->dccp.v6_ctl_sk);
1066 }
1067
1068 static struct pernet_operations dccp_v6_ops = {
1069         .init   = dccp_v6_init_net,
1070         .exit   = dccp_v6_exit_net,
1071 };
1072
1073 static int __init dccp_v6_init(void)
1074 {
1075         int err = proto_register(&dccp_v6_prot, 1);
1076
1077         if (err != 0)
1078                 goto out;
1079
1080         err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1081         if (err != 0)
1082                 goto out_unregister_proto;
1083
1084         inet6_register_protosw(&dccp_v6_protosw);
1085
1086         err = register_pernet_subsys(&dccp_v6_ops);
1087         if (err != 0)
1088                 goto out_destroy_ctl_sock;
1089 out:
1090         return err;
1091
1092 out_destroy_ctl_sock:
1093         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1094         inet6_unregister_protosw(&dccp_v6_protosw);
1095 out_unregister_proto:
1096         proto_unregister(&dccp_v6_prot);
1097         goto out;
1098 }
1099
1100 static void __exit dccp_v6_exit(void)
1101 {
1102         unregister_pernet_subsys(&dccp_v6_ops);
1103         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1104         inet6_unregister_protosw(&dccp_v6_protosw);
1105         proto_unregister(&dccp_v6_prot);
1106 }
1107
1108 module_init(dccp_v6_init);
1109 module_exit(dccp_v6_exit);
1110
1111 /*
1112  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1113  * values directly, Also cover the case where the protocol is not specified,
1114  * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1115  */
1116 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6);
1117 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6);
1118 MODULE_LICENSE("GPL");
1119 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1120 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");