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