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;
  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         /* Only need dccph_dport & dccph_sport which are the first
  82          * 4 bytes in dccp header.
  83          * Our caller (icmpv6_notify()) already pulled 8 bytes for us.
  84          */
  85         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
  86         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
  87         dh = (struct dccp_hdr *)(skb->data + offset);
  88
  89         sk = __inet6_lookup_established(net, &dccp_hashinfo,
  90                                         &hdr->daddr, dh->dccph_dport,
  91                                         &hdr->saddr, ntohs(dh->dccph_sport),
  92                                         inet6_iif(skb));
  93
  94         if (!sk) {
  95                 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
  96                                    ICMP6_MIB_INERRORS);
  97                 return;
  98         }
  99
 100         if (sk->sk_state == DCCP_TIME_WAIT) {
 101                 inet_twsk_put(inet_twsk(sk));
 102                 return;
 103         }
 104         seq = dccp_hdr_seq(dh);
 105         if (sk->sk_state == DCCP_NEW_SYN_RECV)
 106                 return dccp_req_err(sk, seq);
 107
 108         bh_lock_sock(sk);
 109         if (sock_owned_by_user(sk))
 110                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 111
 112         if (sk->sk_state == DCCP_CLOSED)
 113                 goto out;
 114
 115         dp = dccp_sk(sk);
 116         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 117             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 118                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 119                 goto out;
 120         }
 121
 122         np = inet6_sk(sk);
 123
 124         if (type == NDISC_REDIRECT) {
 125                 if (!sock_owned_by_user(sk)) {
 126                         struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 127
 128                         if (dst)
 129                                 dst->ops->redirect(dst, sk, skb);
 130                 }
 131                 goto out;
 132         }
 133
 134         if (type == ICMPV6_PKT_TOOBIG) {
 135                 struct dst_entry *dst = NULL;
 136
 137                 if (!ip6_sk_accept_pmtu(sk))
 138                         goto out;
 139
 140                 if (sock_owned_by_user(sk))
 141                         goto out;
 142                 if ((1 << sk->sk_state) & (DCCPF_LISTEN | DCCPF_CLOSED))
 143                         goto out;
 144
 145                 dst = inet6_csk_update_pmtu(sk, ntohl(info));
 146                 if (!dst)
 147                         goto out;
 148
 149                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst))
 150                         dccp_sync_mss(sk, dst_mtu(dst));
 151                 goto out;
 152         }
 153
 154         icmpv6_err_convert(type, code, &err);
 155
 156         /* Might be for an request_sock */
 157         switch (sk->sk_state) {
 158         case DCCP_REQUESTING:
 159         case DCCP_RESPOND:  /* Cannot happen.
 160                                It can, it SYNs are crossed. --ANK */
 161                 if (!sock_owned_by_user(sk)) {
 162                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 163                         sk->sk_err = err;
 164                         /*
 165                          * Wake people up to see the error
 166                          * (see connect in sock.c)
 167                          */
 168                         sk->sk_error_report(sk);
 169                         dccp_done(sk);
 170                 } else
 171                         sk->sk_err_soft = err;
 172                 goto out;
 173         }
 174
 175         if (!sock_owned_by_user(sk) && np->recverr) {
 176                 sk->sk_err = err;
 177                 sk->sk_error_report(sk);
 178         } else
 179                 sk->sk_err_soft = err;
 180
 181 out:
 182         bh_unlock_sock(sk);
 183         sock_put(sk);
 184 }
 185
 186
 187 static int dccp_v6_send_response(const struct sock *sk, struct request_sock *req)
 188 {
 189         struct inet_request_sock *ireq = inet_rsk(req);
 190         struct ipv6_pinfo *np = inet6_sk(sk);
 191         struct sk_buff *skb;
 192         struct in6_addr *final_p, final;
 193         struct flowi6 fl6;
 194         int err = -1;
 195         struct dst_entry *dst;
 196
 197         memset(&fl6, 0, sizeof(fl6));
 198         fl6.flowi6_proto = IPPROTO_DCCP;
 199         fl6.daddr = ireq->ir_v6_rmt_addr;
 200         fl6.saddr = ireq->ir_v6_loc_addr;
 201         fl6.flowlabel = 0;
 202         fl6.flowi6_oif = ireq->ir_iif;
 203         fl6.fl6_dport = ireq->ir_rmt_port;
 204         fl6.fl6_sport = htons(ireq->ir_num);
 205         security_req_classify_flow(req, flowi6_to_flowi(&fl6));
 206
 207
 208         rcu_read_lock();
 209         final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
 210         rcu_read_unlock();
 211
 212         dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
 213         if (IS_ERR(dst)) {
 214                 err = PTR_ERR(dst);
 215                 dst = NULL;
 216                 goto done;
 217         }
 218
 219         skb = dccp_make_response(sk, dst, req);
 220         if (skb != NULL) {
 221                 struct dccp_hdr *dh = dccp_hdr(skb);
 222
 223                 dh->dccph_checksum = dccp_v6_csum_finish(skb,
 224                                                          &ireq->ir_v6_loc_addr,
 225                                                          &ireq->ir_v6_rmt_addr);
 226                 fl6.daddr = ireq->ir_v6_rmt_addr;
 227                 rcu_read_lock();
 228                 err = ip6_xmit(sk, skb, &fl6, rcu_dereference(np->opt),
 229                                np->tclass);
 230                 rcu_read_unlock();
 231                 err = net_xmit_eval(err);
 232         }
 233
 234 done:
 235         dst_release(dst);
 236         return err;
 237 }
 238
 239 static void dccp_v6_reqsk_destructor(struct request_sock *req)
 240 {
 241         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 242         kfree_skb(inet_rsk(req)->pktopts);
 243 }
 244
 245 static void dccp_v6_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 246 {
 247         const struct ipv6hdr *rxip6h;
 248         struct sk_buff *skb;
 249         struct flowi6 fl6;
 250         struct net *net = dev_net(skb_dst(rxskb)->dev);
 251         struct sock *ctl_sk = net->dccp.v6_ctl_sk;
 252         struct dst_entry *dst;
 253
 254         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 255                 return;
 256
 257         if (!ipv6_unicast_destination(rxskb))
 258                 return;
 259
 260         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 261         if (skb == NULL)
 262                 return;
 263
 264         rxip6h = ipv6_hdr(rxskb);
 265         dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr,
 266                                                             &rxip6h->daddr);
 267
 268         memset(&fl6, 0, sizeof(fl6));
 269         fl6.daddr = rxip6h->saddr;
 270         fl6.saddr = rxip6h->daddr;
 271
 272         fl6.flowi6_proto = IPPROTO_DCCP;
 273         fl6.flowi6_oif = inet6_iif(rxskb);
 274         fl6.fl6_dport = dccp_hdr(skb)->dccph_dport;
 275         fl6.fl6_sport = dccp_hdr(skb)->dccph_sport;
 276         security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6));
 277
 278         /* sk = NULL, but it is safe for now. RST socket required. */
 279         dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL);
 280         if (!IS_ERR(dst)) {
 281                 skb_dst_set(skb, dst);
 282                 ip6_xmit(ctl_sk, skb, &fl6, NULL, 0);
 283                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 284                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 285                 return;
 286         }
 287
 288         kfree_skb(skb);
 289 }
 290
 291 static struct request_sock_ops dccp6_request_sock_ops = {
 292         .family         = AF_INET6,
 293         .obj_size       = sizeof(struct dccp6_request_sock),
 294         .rtx_syn_ack    = dccp_v6_send_response,
 295         .send_ack       = dccp_reqsk_send_ack,
 296         .destructor     = dccp_v6_reqsk_destructor,
 297         .send_reset     = dccp_v6_ctl_send_reset,
 298         .syn_ack_timeout = dccp_syn_ack_timeout,
 299 };
 300
 301 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 302 {
 303         struct request_sock *req;
 304         struct dccp_request_sock *dreq;
 305         struct inet_request_sock *ireq;
 306         struct ipv6_pinfo *np = inet6_sk(sk);
 307         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 308         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 309
 310         if (skb->protocol == htons(ETH_P_IP))
 311                 return dccp_v4_conn_request(sk, skb);
 312
 313         if (!ipv6_unicast_destination(skb))
 314                 return 0;       /* discard, don't send a reset here */
 315
 316         if (dccp_bad_service_code(sk, service)) {
 317                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 318                 goto drop;
 319         }
 320         /*
 321          * There are no SYN attacks on IPv6, yet...
 322          */
 323         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 324         if (inet_csk_reqsk_queue_is_full(sk))
 325                 goto drop;
 326
 327         if (sk_acceptq_is_full(sk))
 328                 goto drop;
 329
 330         req = inet_reqsk_alloc(&dccp6_request_sock_ops, sk, true);
 331         if (req == NULL)
 332                 goto drop;
 333
 334         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 335                 goto drop_and_free;
 336
 337         dreq = dccp_rsk(req);
 338         if (dccp_parse_options(sk, dreq, skb))
 339                 goto drop_and_free;
 340
 341         if (security_inet_conn_request(sk, skb, req))
 342                 goto drop_and_free;
 343
 344         ireq = inet_rsk(req);
 345         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 346         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 347         ireq->ireq_family = AF_INET6;
 348         ireq->ir_mark = inet_request_mark(sk, skb);
 349
 350         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
 351             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 352             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 353                 atomic_inc(&skb->users);
 354                 ireq->pktopts = skb;
 355         }
 356         ireq->ir_iif = sk->sk_bound_dev_if;
 357
 358         /* So that link locals have meaning */
 359         if (!sk->sk_bound_dev_if &&
 360             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 361                 ireq->ir_iif = inet6_iif(skb);
 362
 363         /*
 364          * Step 3: Process LISTEN state
 365          *
 366          *   Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 367          *
 368          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 369          */
 370         dreq->dreq_isr     = dcb->dccpd_seq;
 371         dreq->dreq_gsr     = dreq->dreq_isr;
 372         dreq->dreq_iss     = dccp_v6_init_sequence(skb);
 373         dreq->dreq_gss     = dreq->dreq_iss;
 374         dreq->dreq_service = service;
 375
 376         if (dccp_v6_send_response(sk, req))
 377                 goto drop_and_free;
 378
 379         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 380         reqsk_put(req);
 381         return 0;
 382
 383 drop_and_free:
 384         reqsk_free(req);
 385 drop:
 386         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 387         return -1;
 388 }
 389
 390 static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,
 391                                               struct sk_buff *skb,
 392                                               struct request_sock *req,
 393                                               struct dst_entry *dst,
 394                                               struct request_sock *req_unhash,
 395                                               bool *own_req)
 396 {
 397         struct inet_request_sock *ireq = inet_rsk(req);
 398         struct ipv6_pinfo *newnp;
 399         const struct ipv6_pinfo *np = inet6_sk(sk);
 400         struct ipv6_txoptions *opt;
 401         struct inet_sock *newinet;
 402         struct dccp6_sock *newdp6;
 403         struct sock *newsk;
 404
 405         if (skb->protocol == htons(ETH_P_IP)) {
 406                 /*
 407                  *      v6 mapped
 408                  */
 409                 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst,
 410                                                   req_unhash, own_req);
 411                 if (newsk == NULL)
 412                         return NULL;
 413
 414                 newdp6 = (struct dccp6_sock *)newsk;
 415                 newinet = inet_sk(newsk);
 416                 newinet->pinet6 = &newdp6->inet6;
 417                 newnp = inet6_sk(newsk);
 418
 419                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 420
 421                 newnp->saddr = newsk->sk_v6_rcv_saddr;
 422
 423                 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
 424                 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
 425                 newnp->pktoptions  = NULL;
 426                 newnp->opt         = NULL;
 427                 newnp->ipv6_mc_list = NULL;
 428                 newnp->ipv6_ac_list = NULL;
 429                 newnp->ipv6_fl_list = NULL;
 430                 newnp->mcast_oif   = inet_iif(skb);
 431                 newnp->mcast_hops  = ip_hdr(skb)->ttl;
 432
 433                 /*
 434                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
 435                  * here, dccp_create_openreq_child now does this for us, see the comment in
 436                  * that function for the gory details. -acme
 437                  */
 438
 439                 /* It is tricky place. Until this moment IPv4 tcp
 440                    worked with IPv6 icsk.icsk_af_ops.
 441                    Sync it now.
 442                  */
 443                 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
 444
 445                 return newsk;
 446         }
 447
 448
 449         if (sk_acceptq_is_full(sk))
 450                 goto out_overflow;
 451
 452         if (!dst) {
 453                 struct flowi6 fl6;
 454
 455                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_DCCP);
 456                 if (!dst)
 457                         goto out;
 458         }
 459
 460         newsk = dccp_create_openreq_child(sk, req, skb);
 461         if (newsk == NULL)
 462                 goto out_nonewsk;
 463
 464         /*
 465          * No need to charge this sock to the relevant IPv6 refcnt debug socks
 466          * count here, dccp_create_openreq_child now does this for us, see the
 467          * comment in that function for the gory details. -acme
 468          */
 469
 470         ip6_dst_store(newsk, dst, NULL, NULL);
 471         newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
 472                                                       NETIF_F_TSO);
 473         newdp6 = (struct dccp6_sock *)newsk;
 474         newinet = inet_sk(newsk);
 475         newinet->pinet6 = &newdp6->inet6;
 476         newnp = inet6_sk(newsk);
 477
 478         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 479
 480         newsk->sk_v6_daddr      = ireq->ir_v6_rmt_addr;
 481         newnp->saddr            = ireq->ir_v6_loc_addr;
 482         newsk->sk_v6_rcv_saddr  = ireq->ir_v6_loc_addr;
 483         newsk->sk_bound_dev_if  = ireq->ir_iif;
 484
 485         /* Now IPv6 options...
 486
 487            First: no IPv4 options.
 488          */
 489         newinet->inet_opt = NULL;
 490
 491         /* Clone RX bits */
 492         newnp->rxopt.all = np->rxopt.all;
 493
 494         newnp->ipv6_mc_list = NULL;
 495         newnp->ipv6_ac_list = NULL;
 496         newnp->ipv6_fl_list = NULL;
 497         newnp->pktoptions = NULL;
 498         newnp->opt        = NULL;
 499         newnp->mcast_oif  = inet6_iif(skb);
 500         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
 501
 502         /*
 503          * Clone native IPv6 options from listening socket (if any)
 504          *
 505          * Yes, keeping reference count would be much more clever, but we make
 506          * one more one thing there: reattach optmem to newsk.
 507          */
 508         opt = rcu_dereference(np->opt);
 509         if (opt) {
 510                 opt = ipv6_dup_options(newsk, opt);
 511                 RCU_INIT_POINTER(newnp->opt, opt);
 512         }
 513         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 514         if (opt)
 515                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
 516                                                     opt->opt_flen;
 517
 518         dccp_sync_mss(newsk, dst_mtu(dst));
 519
 520         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
 521         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
 522
 523         if (__inet_inherit_port(sk, newsk) < 0) {
 524                 inet_csk_prepare_forced_close(newsk);
 525                 dccp_done(newsk);
 526                 goto out;
 527         }
 528         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 529         /* Clone pktoptions received with SYN, if we own the req */
 530         if (*own_req && ireq->pktopts) {
 531                 newnp->pktoptions = skb_clone(ireq->pktopts, GFP_ATOMIC);
 532                 consume_skb(ireq->pktopts);
 533                 ireq->pktopts = NULL;
 534                 if (newnp->pktoptions)
 535                         skb_set_owner_r(newnp->pktoptions, newsk);
 536         }
 537
 538         return newsk;
 539
 540 out_overflow:
 541         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 542 out_nonewsk:
 543         dst_release(dst);
 544 out:
 545         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 546         return NULL;
 547 }
 548
 549 /* The socket must have it's spinlock held when we get
 550  * here.
 551  *
 552  * We have a potential double-lock case here, so even when
 553  * doing backlog processing we use the BH locking scheme.
 554  * This is because we cannot sleep with the original spinlock
 555  * held.
 556  */
 557 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 558 {
 559         struct ipv6_pinfo *np = inet6_sk(sk);
 560         struct sk_buff *opt_skb = NULL;
 561
 562         /* Imagine: socket is IPv6. IPv4 packet arrives,
 563            goes to IPv4 receive handler and backlogged.
 564            From backlog it always goes here. Kerboom...
 565            Fortunately, dccp_rcv_established and rcv_established
 566            handle them correctly, but it is not case with
 567            dccp_v6_hnd_req and dccp_v6_ctl_send_reset().   --ANK
 568          */
 569
 570         if (skb->protocol == htons(ETH_P_IP))
 571                 return dccp_v4_do_rcv(sk, skb);
 572
 573         if (sk_filter(sk, skb))
 574                 goto discard;
 575
 576         /*
 577          * socket locking is here for SMP purposes as backlog rcv is currently
 578          * called with bh processing disabled.
 579          */
 580
 581         /* Do Stevens' IPV6_PKTOPTIONS.
 582
 583            Yes, guys, it is the only place in our code, where we
 584            may make it not affecting IPv4.
 585            The rest of code is protocol independent,
 586            and I do not like idea to uglify IPv4.
 587
 588            Actually, all the idea behind IPV6_PKTOPTIONS
 589            looks not very well thought. For now we latch
 590            options, received in the last packet, enqueued
 591            by tcp. Feel free to propose better solution.
 592                                                --ANK (980728)
 593          */
 594         if (np->rxopt.all)
 595         /*
 596          * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
 597          *        (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
 598          */
 599                 opt_skb = skb_clone(skb, GFP_ATOMIC);
 600
 601         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 602                 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
 603                         goto reset;
 604                 if (opt_skb) {
 605                         /* XXX This is where we would goto ipv6_pktoptions. */
 606                         __kfree_skb(opt_skb);
 607                 }
 608                 return 0;
 609         }
 610
 611         /*
 612          *  Step 3: Process LISTEN state
 613          *     If S.state == LISTEN,
 614          *       If P.type == Request or P contains a valid Init Cookie option,
 615          *            (* Must scan the packet's options to check for Init
 616          *               Cookies.  Only Init Cookies are processed here,
 617          *               however; other options are processed in Step 8.  This
 618          *               scan need only be performed if the endpoint uses Init
 619          *               Cookies *)
 620          *            (* Generate a new socket and switch to that socket *)
 621          *            Set S := new socket for this port pair
 622          *            S.state = RESPOND
 623          *            Choose S.ISS (initial seqno) or set from Init Cookies
 624          *            Initialize S.GAR := S.ISS
 625          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 626          *            Continue with S.state == RESPOND
 627          *            (* A Response packet will be generated in Step 11 *)
 628          *       Otherwise,
 629          *            Generate Reset(No Connection) unless P.type == Reset
 630          *            Drop packet and return
 631          *
 632          * NOTE: the check for the packet types is done in
 633          *       dccp_rcv_state_process
 634          */
 635
 636         if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
 637                 goto reset;
 638         if (opt_skb) {
 639                 /* XXX This is where we would goto ipv6_pktoptions. */
 640                 __kfree_skb(opt_skb);
 641         }
 642         return 0;
 643
 644 reset:
 645         dccp_v6_ctl_send_reset(sk, skb);
 646 discard:
 647         if (opt_skb != NULL)
 648                 __kfree_skb(opt_skb);
 649         kfree_skb(skb);
 650         return 0;
 651 }
 652
 653 static int dccp_v6_rcv(struct sk_buff *skb)
 654 {
 655         const struct dccp_hdr *dh;
 656         struct sock *sk;
 657         int min_cov;
 658
 659         /* Step 1: Check header basics */
 660
 661         if (dccp_invalid_packet(skb))
 662                 goto discard_it;
 663
 664         /* Step 1: If header checksum is incorrect, drop packet and return. */
 665         if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
 666                                      &ipv6_hdr(skb)->daddr)) {
 667                 DCCP_WARN("dropped packet with invalid checksum\n");
 668                 goto discard_it;
 669         }
 670
 671         dh = dccp_hdr(skb);
 672
 673         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 674         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 675
 676         if (dccp_packet_without_ack(skb))
 677                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 678         else
 679                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 680
 681 lookup:
 682         sk = __inet6_lookup_skb(&dccp_hashinfo, skb,
 683                                 dh->dccph_sport, dh->dccph_dport,
 684                                 inet6_iif(skb));
 685         if (!sk) {
 686                 dccp_pr_debug("failed to look up flow ID in table and "
 687                               "get corresponding socket\n");
 688                 goto no_dccp_socket;
 689         }
 690
 691         /*
 692          * Step 2:
 693          *      ... or S.state == TIMEWAIT,
 694          *              Generate Reset(No Connection) unless P.type == Reset
 695          *              Drop packet and return
 696          */
 697         if (sk->sk_state == DCCP_TIME_WAIT) {
 698                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 699                 inet_twsk_put(inet_twsk(sk));
 700                 goto no_dccp_socket;
 701         }
 702
 703         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 704                 struct request_sock *req = inet_reqsk(sk);
 705                 struct sock *nsk;
 706
 707                 sk = req->rsk_listener;
 708                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 709                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 710                         goto lookup;
 711                 }
 712                 sock_hold(sk);
 713                 nsk = dccp_check_req(sk, skb, req);
 714                 if (!nsk) {
 715                         reqsk_put(req);
 716                         goto discard_and_relse;
 717                 }
 718                 if (nsk == sk) {
 719                         reqsk_put(req);
 720                 } else if (dccp_child_process(sk, nsk, skb)) {
 721                         dccp_v6_ctl_send_reset(sk, skb);
 722                         goto discard_and_relse;
 723                 } else {
 724                         sock_put(sk);
 725                         return 0;
 726                 }
 727         }
 728         /*
 729          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 730          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 731          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 732          */
 733         min_cov = dccp_sk(sk)->dccps_pcrlen;
 734         if (dh->dccph_cscov  &&  (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 735                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 736                               dh->dccph_cscov, min_cov);
 737                 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
 738                 goto discard_and_relse;
 739         }
 740
 741         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
 742                 goto discard_and_relse;
 743
 744         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4) ? -1 : 0;
 745
 746 no_dccp_socket:
 747         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 748                 goto discard_it;
 749         /*
 750          * Step 2:
 751          *      If no socket ...
 752          *              Generate Reset(No Connection) unless P.type == Reset
 753          *              Drop packet and return
 754          */
 755         if (dh->dccph_type != DCCP_PKT_RESET) {
 756                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 757                                         DCCP_RESET_CODE_NO_CONNECTION;
 758                 dccp_v6_ctl_send_reset(sk, skb);
 759         }
 760
 761 discard_it:
 762         kfree_skb(skb);
 763         return 0;
 764
 765 discard_and_relse:
 766         sock_put(sk);
 767         goto discard_it;
 768 }
 769
 770 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 771                            int addr_len)
 772 {
 773         struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr;
 774         struct inet_connection_sock *icsk = inet_csk(sk);
 775         struct inet_sock *inet = inet_sk(sk);
 776         struct ipv6_pinfo *np = inet6_sk(sk);
 777         struct dccp_sock *dp = dccp_sk(sk);
 778         struct in6_addr *saddr = NULL, *final_p, final;
 779         struct ipv6_txoptions *opt;
 780         struct flowi6 fl6;
 781         struct dst_entry *dst;
 782         int addr_type;
 783         int err;
 784
 785         dp->dccps_role = DCCP_ROLE_CLIENT;
 786
 787         if (addr_len < SIN6_LEN_RFC2133)
 788                 return -EINVAL;
 789
 790         if (usin->sin6_family != AF_INET6)
 791                 return -EAFNOSUPPORT;
 792
 793         memset(&fl6, 0, sizeof(fl6));
 794
 795         if (np->sndflow) {
 796                 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK;
 797                 IP6_ECN_flow_init(fl6.flowlabel);
 798                 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) {
 799                         struct ip6_flowlabel *flowlabel;
 800                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 801                         if (flowlabel == NULL)
 802                                 return -EINVAL;
 803                         fl6_sock_release(flowlabel);
 804                 }
 805         }
 806         /*
 807          * connect() to INADDR_ANY means loopback (BSD'ism).
 808          */
 809         if (ipv6_addr_any(&usin->sin6_addr))
 810                 usin->sin6_addr.s6_addr[15] = 1;
 811
 812         addr_type = ipv6_addr_type(&usin->sin6_addr);
 813
 814         if (addr_type & IPV6_ADDR_MULTICAST)
 815                 return -ENETUNREACH;
 816
 817         if (addr_type & IPV6_ADDR_LINKLOCAL) {
 818                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 819                     usin->sin6_scope_id) {
 820                         /* If interface is set while binding, indices
 821                          * must coincide.
 822                          */
 823                         if (sk->sk_bound_dev_if &&
 824                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 825                                 return -EINVAL;
 826
 827                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 828                 }
 829
 830                 /* Connect to link-local address requires an interface */
 831                 if (!sk->sk_bound_dev_if)
 832                         return -EINVAL;
 833         }
 834
 835         sk->sk_v6_daddr = usin->sin6_addr;
 836         np->flow_label = fl6.flowlabel;
 837
 838         /*
 839          * DCCP over IPv4
 840          */
 841         if (addr_type == IPV6_ADDR_MAPPED) {
 842                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 843                 struct sockaddr_in sin;
 844
 845                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 846
 847                 if (__ipv6_only_sock(sk))
 848                         return -ENETUNREACH;
 849
 850                 sin.sin_family = AF_INET;
 851                 sin.sin_port = usin->sin6_port;
 852                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 853
 854                 icsk->icsk_af_ops = &dccp_ipv6_mapped;
 855                 sk->sk_backlog_rcv = dccp_v4_do_rcv;
 856
 857                 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 858                 if (err) {
 859                         icsk->icsk_ext_hdr_len = exthdrlen;
 860                         icsk->icsk_af_ops = &dccp_ipv6_af_ops;
 861                         sk->sk_backlog_rcv = dccp_v6_do_rcv;
 862                         goto failure;
 863                 }
 864                 np->saddr = sk->sk_v6_rcv_saddr;
 865                 return err;
 866         }
 867
 868         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 869                 saddr = &sk->sk_v6_rcv_saddr;
 870
 871         fl6.flowi6_proto = IPPROTO_DCCP;
 872         fl6.daddr = sk->sk_v6_daddr;
 873         fl6.saddr = saddr ? *saddr : np->saddr;
 874         fl6.flowi6_oif = sk->sk_bound_dev_if;
 875         fl6.fl6_dport = usin->sin6_port;
 876         fl6.fl6_sport = inet->inet_sport;
 877         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 878
 879         opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
 880         final_p = fl6_update_dst(&fl6, opt, &final);
 881
 882         dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
 883         if (IS_ERR(dst)) {
 884                 err = PTR_ERR(dst);
 885                 goto failure;
 886         }
 887
 888         if (saddr == NULL) {
 889                 saddr = &fl6.saddr;
 890                 sk->sk_v6_rcv_saddr = *saddr;
 891         }
 892
 893         /* set the source address */
 894         np->saddr = *saddr;
 895         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 896
 897         ip6_dst_store(sk, dst, NULL, NULL);
 898
 899         icsk->icsk_ext_hdr_len = 0;
 900         if (opt)
 901                 icsk->icsk_ext_hdr_len = opt->opt_flen + opt->opt_nflen;
 902
 903         inet->inet_dport = usin->sin6_port;
 904
 905         dccp_set_state(sk, DCCP_REQUESTING);
 906         err = inet6_hash_connect(&dccp_death_row, sk);
 907         if (err)
 908                 goto late_failure;
 909
 910         dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32,
 911                                                       sk->sk_v6_daddr.s6_addr32,
 912                                                       inet->inet_sport,
 913                                                       inet->inet_dport);
 914         err = dccp_connect(sk);
 915         if (err)
 916                 goto late_failure;
 917
 918         return 0;
 919
 920 late_failure:
 921         dccp_set_state(sk, DCCP_CLOSED);
 922         __sk_dst_reset(sk);
 923 failure:
 924         inet->inet_dport = 0;
 925         sk->sk_route_caps = 0;
 926         return err;
 927 }
 928
 929 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = {
 930         .queue_xmit        = inet6_csk_xmit,
 931         .send_check        = dccp_v6_send_check,
 932         .rebuild_header    = inet6_sk_rebuild_header,
 933         .conn_request      = dccp_v6_conn_request,
 934         .syn_recv_sock     = dccp_v6_request_recv_sock,
 935         .net_header_len    = sizeof(struct ipv6hdr),
 936         .setsockopt        = ipv6_setsockopt,
 937         .getsockopt        = ipv6_getsockopt,
 938         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 939         .sockaddr_len      = sizeof(struct sockaddr_in6),
 940         .bind_conflict     = inet6_csk_bind_conflict,
 941 #ifdef CONFIG_COMPAT
 942         .compat_setsockopt = compat_ipv6_setsockopt,
 943         .compat_getsockopt = compat_ipv6_getsockopt,
 944 #endif
 945 };
 946
 947 /*
 948  *      DCCP over IPv4 via INET6 API
 949  */
 950 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = {
 951         .queue_xmit        = ip_queue_xmit,
 952         .send_check        = dccp_v4_send_check,
 953         .rebuild_header    = inet_sk_rebuild_header,
 954         .conn_request      = dccp_v6_conn_request,
 955         .syn_recv_sock     = dccp_v6_request_recv_sock,
 956         .net_header_len    = sizeof(struct iphdr),
 957         .setsockopt        = ipv6_setsockopt,
 958         .getsockopt        = ipv6_getsockopt,
 959         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 960         .sockaddr_len      = sizeof(struct sockaddr_in6),
 961         .bind_conflict     = inet6_csk_bind_conflict,
 962 #ifdef CONFIG_COMPAT
 963         .compat_setsockopt = compat_ipv6_setsockopt,
 964         .compat_getsockopt = compat_ipv6_getsockopt,
 965 #endif
 966 };
 967
 968 /* NOTE: A lot of things set to zero explicitly by call to
 969  *       sk_alloc() so need not be done here.
 970  */
 971 static int dccp_v6_init_sock(struct sock *sk)
 972 {
 973         static __u8 dccp_v6_ctl_sock_initialized;
 974         int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized);
 975
 976         if (err == 0) {
 977                 if (unlikely(!dccp_v6_ctl_sock_initialized))
 978                         dccp_v6_ctl_sock_initialized = 1;
 979                 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops;
 980         }
 981
 982         return err;
 983 }
 984
 985 static void dccp_v6_destroy_sock(struct sock *sk)
 986 {
 987         dccp_destroy_sock(sk);
 988         inet6_destroy_sock(sk);
 989 }
 990
 991 static struct timewait_sock_ops dccp6_timewait_sock_ops = {
 992         .twsk_obj_size  = sizeof(struct dccp6_timewait_sock),
 993 };
 994
 995 static struct proto dccp_v6_prot = {
 996         .name              = "DCCPv6",
 997         .owner             = THIS_MODULE,
 998         .close             = dccp_close,
 999         .connect           = dccp_v6_connect,
1000         .disconnect        = dccp_disconnect,
1001         .ioctl             = dccp_ioctl,
1002         .init              = dccp_v6_init_sock,
1003         .setsockopt        = dccp_setsockopt,
1004         .getsockopt        = dccp_getsockopt,
1005         .sendmsg           = dccp_sendmsg,
1006         .recvmsg           = dccp_recvmsg,
1007         .backlog_rcv       = dccp_v6_do_rcv,
1008         .hash              = inet_hash,
1009         .unhash            = inet_unhash,
1010         .accept            = inet_csk_accept,
1011         .get_port          = inet_csk_get_port,
1012         .shutdown          = dccp_shutdown,
1013         .destroy           = dccp_v6_destroy_sock,
1014         .orphan_count      = &dccp_orphan_count,
1015         .max_header        = MAX_DCCP_HEADER,
1016         .obj_size          = sizeof(struct dccp6_sock),
1017         .slab_flags        = SLAB_DESTROY_BY_RCU,
1018         .rsk_prot          = &dccp6_request_sock_ops,
1019         .twsk_prot         = &dccp6_timewait_sock_ops,
1020         .h.hashinfo        = &dccp_hashinfo,
1021 #ifdef CONFIG_COMPAT
1022         .compat_setsockopt = compat_dccp_setsockopt,
1023         .compat_getsockopt = compat_dccp_getsockopt,
1024 #endif
1025 };
1026
1027 static const struct inet6_protocol dccp_v6_protocol = {
1028         .handler        = dccp_v6_rcv,
1029         .err_handler    = dccp_v6_err,
1030         .flags          = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1031 };
1032
1033 static const struct proto_ops inet6_dccp_ops = {
1034         .family            = PF_INET6,
1035         .owner             = THIS_MODULE,
1036         .release           = inet6_release,
1037         .bind              = inet6_bind,
1038         .connect           = inet_stream_connect,
1039         .socketpair        = sock_no_socketpair,
1040         .accept            = inet_accept,
1041         .getname           = inet6_getname,
1042         .poll              = dccp_poll,
1043         .ioctl             = inet6_ioctl,
1044         .listen            = inet_dccp_listen,
1045         .shutdown          = inet_shutdown,
1046         .setsockopt        = sock_common_setsockopt,
1047         .getsockopt        = sock_common_getsockopt,
1048         .sendmsg           = inet_sendmsg,
1049         .recvmsg           = sock_common_recvmsg,
1050         .mmap              = sock_no_mmap,
1051         .sendpage          = sock_no_sendpage,
1052 #ifdef CONFIG_COMPAT
1053         .compat_setsockopt = compat_sock_common_setsockopt,
1054         .compat_getsockopt = compat_sock_common_getsockopt,
1055 #endif
1056 };
1057
1058 static struct inet_protosw dccp_v6_protosw = {
1059         .type           = SOCK_DCCP,
1060         .protocol       = IPPROTO_DCCP,
1061         .prot           = &dccp_v6_prot,
1062         .ops            = &inet6_dccp_ops,
1063         .flags          = INET_PROTOSW_ICSK,
1064 };
1065
1066 static int __net_init dccp_v6_init_net(struct net *net)
1067 {
1068         if (dccp_hashinfo.bhash == NULL)
1069                 return -ESOCKTNOSUPPORT;
1070
1071         return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6,
1072                                     SOCK_DCCP, IPPROTO_DCCP, net);
1073 }
1074
1075 static void __net_exit dccp_v6_exit_net(struct net *net)
1076 {
1077         inet_ctl_sock_destroy(net->dccp.v6_ctl_sk);
1078 }
1079
1080 static struct pernet_operations dccp_v6_ops = {
1081         .init   = dccp_v6_init_net,
1082         .exit   = dccp_v6_exit_net,
1083 };
1084
1085 static int __init dccp_v6_init(void)
1086 {
1087         int err = proto_register(&dccp_v6_prot, 1);
1088
1089         if (err != 0)
1090                 goto out;
1091
1092         err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1093         if (err != 0)
1094                 goto out_unregister_proto;
1095
1096         inet6_register_protosw(&dccp_v6_protosw);
1097
1098         err = register_pernet_subsys(&dccp_v6_ops);
1099         if (err != 0)
1100                 goto out_destroy_ctl_sock;
1101 out:
1102         return err;
1103
1104 out_destroy_ctl_sock:
1105         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1106         inet6_unregister_protosw(&dccp_v6_protosw);
1107 out_unregister_proto:
1108         proto_unregister(&dccp_v6_prot);
1109         goto out;
1110 }
1111
1112 static void __exit dccp_v6_exit(void)
1113 {
1114         unregister_pernet_subsys(&dccp_v6_ops);
1115         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1116         inet6_unregister_protosw(&dccp_v6_protosw);
1117         proto_unregister(&dccp_v6_prot);
1118 }
1119
1120 module_init(dccp_v6_init);
1121 module_exit(dccp_v6_exit);
1122
1123 /*
1124  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1125  * values directly, Also cover the case where the protocol is not specified,
1126  * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1127  */
1128 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6);
1129 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6);
1130 MODULE_LICENSE("GPL");
1131 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1132 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");