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 (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 317                 IP6_INC_STATS_BH(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 318                 return 0;
 319         }
 320
 321         if (dccp_bad_service_code(sk, service)) {
 322                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 323                 goto drop;
 324         }
 325         /*
 326          * There are no SYN attacks on IPv6, yet...
 327          */
 328         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 329         if (inet_csk_reqsk_queue_is_full(sk))
 330                 goto drop;
 331
 332         if (sk_acceptq_is_full(sk))
 333                 goto drop;
 334
 335         req = inet_reqsk_alloc(&dccp6_request_sock_ops, sk, true);
 336         if (req == NULL)
 337                 goto drop;
 338
 339         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 340                 goto drop_and_free;
 341
 342         dreq = dccp_rsk(req);
 343         if (dccp_parse_options(sk, dreq, skb))
 344                 goto drop_and_free;
 345
 346         if (security_inet_conn_request(sk, skb, req))
 347                 goto drop_and_free;
 348
 349         ireq = inet_rsk(req);
 350         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 351         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 352         ireq->ireq_family = AF_INET6;
 353         ireq->ir_mark = inet_request_mark(sk, skb);
 354
 355         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
 356             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 357             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 358                 atomic_inc(&skb->users);
 359                 ireq->pktopts = skb;
 360         }
 361         ireq->ir_iif = sk->sk_bound_dev_if;
 362
 363         /* So that link locals have meaning */
 364         if (!sk->sk_bound_dev_if &&
 365             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 366                 ireq->ir_iif = inet6_iif(skb);
 367
 368         /*
 369          * Step 3: Process LISTEN state
 370          *
 371          *   Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 372          *
 373          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 374          */
 375         dreq->dreq_isr     = dcb->dccpd_seq;
 376         dreq->dreq_gsr     = dreq->dreq_isr;
 377         dreq->dreq_iss     = dccp_v6_init_sequence(skb);
 378         dreq->dreq_gss     = dreq->dreq_iss;
 379         dreq->dreq_service = service;
 380
 381         if (dccp_v6_send_response(sk, req))
 382                 goto drop_and_free;
 383
 384         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 385         reqsk_put(req);
 386         return 0;
 387
 388 drop_and_free:
 389         reqsk_free(req);
 390 drop:
 391         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 392         return -1;
 393 }
 394
 395 static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,
 396                                               struct sk_buff *skb,
 397                                               struct request_sock *req,
 398                                               struct dst_entry *dst,
 399                                               struct request_sock *req_unhash,
 400                                               bool *own_req)
 401 {
 402         struct inet_request_sock *ireq = inet_rsk(req);
 403         struct ipv6_pinfo *newnp;
 404         const struct ipv6_pinfo *np = inet6_sk(sk);
 405         struct ipv6_txoptions *opt;
 406         struct inet_sock *newinet;
 407         struct dccp6_sock *newdp6;
 408         struct sock *newsk;
 409
 410         if (skb->protocol == htons(ETH_P_IP)) {
 411                 /*
 412                  *      v6 mapped
 413                  */
 414                 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst,
 415                                                   req_unhash, own_req);
 416                 if (newsk == NULL)
 417                         return NULL;
 418
 419                 newdp6 = (struct dccp6_sock *)newsk;
 420                 newinet = inet_sk(newsk);
 421                 newinet->pinet6 = &newdp6->inet6;
 422                 newnp = inet6_sk(newsk);
 423
 424                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 425
 426                 newnp->saddr = newsk->sk_v6_rcv_saddr;
 427
 428                 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
 429                 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
 430                 newnp->pktoptions  = NULL;
 431                 newnp->opt         = NULL;
 432                 newnp->ipv6_mc_list = NULL;
 433                 newnp->ipv6_ac_list = NULL;
 434                 newnp->ipv6_fl_list = NULL;
 435                 newnp->mcast_oif   = inet_iif(skb);
 436                 newnp->mcast_hops  = ip_hdr(skb)->ttl;
 437
 438                 /*
 439                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
 440                  * here, dccp_create_openreq_child now does this for us, see the comment in
 441                  * that function for the gory details. -acme
 442                  */
 443
 444                 /* It is tricky place. Until this moment IPv4 tcp
 445                    worked with IPv6 icsk.icsk_af_ops.
 446                    Sync it now.
 447                  */
 448                 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
 449
 450                 return newsk;
 451         }
 452
 453
 454         if (sk_acceptq_is_full(sk))
 455                 goto out_overflow;
 456
 457         if (!dst) {
 458                 struct flowi6 fl6;
 459
 460                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_DCCP);
 461                 if (!dst)
 462                         goto out;
 463         }
 464
 465         newsk = dccp_create_openreq_child(sk, req, skb);
 466         if (newsk == NULL)
 467                 goto out_nonewsk;
 468
 469         /*
 470          * No need to charge this sock to the relevant IPv6 refcnt debug socks
 471          * count here, dccp_create_openreq_child now does this for us, see the
 472          * comment in that function for the gory details. -acme
 473          */
 474
 475         ip6_dst_store(newsk, dst, NULL, NULL);
 476         newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
 477                                                       NETIF_F_TSO);
 478         newdp6 = (struct dccp6_sock *)newsk;
 479         newinet = inet_sk(newsk);
 480         newinet->pinet6 = &newdp6->inet6;
 481         newnp = inet6_sk(newsk);
 482
 483         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 484
 485         newsk->sk_v6_daddr      = ireq->ir_v6_rmt_addr;
 486         newnp->saddr            = ireq->ir_v6_loc_addr;
 487         newsk->sk_v6_rcv_saddr  = ireq->ir_v6_loc_addr;
 488         newsk->sk_bound_dev_if  = ireq->ir_iif;
 489
 490         /* Now IPv6 options...
 491
 492            First: no IPv4 options.
 493          */
 494         newinet->inet_opt = NULL;
 495
 496         /* Clone RX bits */
 497         newnp->rxopt.all = np->rxopt.all;
 498
 499         newnp->ipv6_mc_list = NULL;
 500         newnp->ipv6_ac_list = NULL;
 501         newnp->ipv6_fl_list = NULL;
 502         newnp->pktoptions = NULL;
 503         newnp->opt        = NULL;
 504         newnp->mcast_oif  = inet6_iif(skb);
 505         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
 506
 507         /*
 508          * Clone native IPv6 options from listening socket (if any)
 509          *
 510          * Yes, keeping reference count would be much more clever, but we make
 511          * one more one thing there: reattach optmem to newsk.
 512          */
 513         opt = rcu_dereference(np->opt);
 514         if (opt) {
 515                 opt = ipv6_dup_options(newsk, opt);
 516                 RCU_INIT_POINTER(newnp->opt, opt);
 517         }
 518         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 519         if (opt)
 520                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
 521                                                     opt->opt_flen;
 522
 523         dccp_sync_mss(newsk, dst_mtu(dst));
 524
 525         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
 526         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
 527
 528         if (__inet_inherit_port(sk, newsk) < 0) {
 529                 inet_csk_prepare_forced_close(newsk);
 530                 dccp_done(newsk);
 531                 goto out;
 532         }
 533         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 534         /* Clone pktoptions received with SYN, if we own the req */
 535         if (*own_req && ireq->pktopts) {
 536                 newnp->pktoptions = skb_clone(ireq->pktopts, GFP_ATOMIC);
 537                 consume_skb(ireq->pktopts);
 538                 ireq->pktopts = NULL;
 539                 if (newnp->pktoptions)
 540                         skb_set_owner_r(newnp->pktoptions, newsk);
 541         }
 542
 543         return newsk;
 544
 545 out_overflow:
 546         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 547 out_nonewsk:
 548         dst_release(dst);
 549 out:
 550         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 551         return NULL;
 552 }
 553
 554 /* The socket must have it's spinlock held when we get
 555  * here.
 556  *
 557  * We have a potential double-lock case here, so even when
 558  * doing backlog processing we use the BH locking scheme.
 559  * This is because we cannot sleep with the original spinlock
 560  * held.
 561  */
 562 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 563 {
 564         struct ipv6_pinfo *np = inet6_sk(sk);
 565         struct sk_buff *opt_skb = NULL;
 566
 567         /* Imagine: socket is IPv6. IPv4 packet arrives,
 568            goes to IPv4 receive handler and backlogged.
 569            From backlog it always goes here. Kerboom...
 570            Fortunately, dccp_rcv_established and rcv_established
 571            handle them correctly, but it is not case with
 572            dccp_v6_hnd_req and dccp_v6_ctl_send_reset().   --ANK
 573          */
 574
 575         if (skb->protocol == htons(ETH_P_IP))
 576                 return dccp_v4_do_rcv(sk, skb);
 577
 578         if (sk_filter(sk, skb))
 579                 goto discard;
 580
 581         /*
 582          * socket locking is here for SMP purposes as backlog rcv is currently
 583          * called with bh processing disabled.
 584          */
 585
 586         /* Do Stevens' IPV6_PKTOPTIONS.
 587
 588            Yes, guys, it is the only place in our code, where we
 589            may make it not affecting IPv4.
 590            The rest of code is protocol independent,
 591            and I do not like idea to uglify IPv4.
 592
 593            Actually, all the idea behind IPV6_PKTOPTIONS
 594            looks not very well thought. For now we latch
 595            options, received in the last packet, enqueued
 596            by tcp. Feel free to propose better solution.
 597                                                --ANK (980728)
 598          */
 599         if (np->rxopt.all)
 600         /*
 601          * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
 602          *        (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
 603          */
 604                 opt_skb = skb_clone(skb, GFP_ATOMIC);
 605
 606         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 607                 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
 608                         goto reset;
 609                 if (opt_skb) {
 610                         /* XXX This is where we would goto ipv6_pktoptions. */
 611                         __kfree_skb(opt_skb);
 612                 }
 613                 return 0;
 614         }
 615
 616         /*
 617          *  Step 3: Process LISTEN state
 618          *     If S.state == LISTEN,
 619          *       If P.type == Request or P contains a valid Init Cookie option,
 620          *            (* Must scan the packet's options to check for Init
 621          *               Cookies.  Only Init Cookies are processed here,
 622          *               however; other options are processed in Step 8.  This
 623          *               scan need only be performed if the endpoint uses Init
 624          *               Cookies *)
 625          *            (* Generate a new socket and switch to that socket *)
 626          *            Set S := new socket for this port pair
 627          *            S.state = RESPOND
 628          *            Choose S.ISS (initial seqno) or set from Init Cookies
 629          *            Initialize S.GAR := S.ISS
 630          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 631          *            Continue with S.state == RESPOND
 632          *            (* A Response packet will be generated in Step 11 *)
 633          *       Otherwise,
 634          *            Generate Reset(No Connection) unless P.type == Reset
 635          *            Drop packet and return
 636          *
 637          * NOTE: the check for the packet types is done in
 638          *       dccp_rcv_state_process
 639          */
 640
 641         if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
 642                 goto reset;
 643         if (opt_skb) {
 644                 /* XXX This is where we would goto ipv6_pktoptions. */
 645                 __kfree_skb(opt_skb);
 646         }
 647         return 0;
 648
 649 reset:
 650         dccp_v6_ctl_send_reset(sk, skb);
 651 discard:
 652         if (opt_skb != NULL)
 653                 __kfree_skb(opt_skb);
 654         kfree_skb(skb);
 655         return 0;
 656 }
 657
 658 static int dccp_v6_rcv(struct sk_buff *skb)
 659 {
 660         const struct dccp_hdr *dh;
 661         struct sock *sk;
 662         int min_cov;
 663
 664         /* Step 1: Check header basics */
 665
 666         if (dccp_invalid_packet(skb))
 667                 goto discard_it;
 668
 669         /* Step 1: If header checksum is incorrect, drop packet and return. */
 670         if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
 671                                      &ipv6_hdr(skb)->daddr)) {
 672                 DCCP_WARN("dropped packet with invalid checksum\n");
 673                 goto discard_it;
 674         }
 675
 676         dh = dccp_hdr(skb);
 677
 678         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 679         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 680
 681         if (dccp_packet_without_ack(skb))
 682                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 683         else
 684                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 685
 686 lookup:
 687         sk = __inet6_lookup_skb(&dccp_hashinfo, skb,
 688                                 dh->dccph_sport, dh->dccph_dport,
 689                                 inet6_iif(skb));
 690         if (!sk) {
 691                 dccp_pr_debug("failed to look up flow ID in table and "
 692                               "get corresponding socket\n");
 693                 goto no_dccp_socket;
 694         }
 695
 696         /*
 697          * Step 2:
 698          *      ... or S.state == TIMEWAIT,
 699          *              Generate Reset(No Connection) unless P.type == Reset
 700          *              Drop packet and return
 701          */
 702         if (sk->sk_state == DCCP_TIME_WAIT) {
 703                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 704                 inet_twsk_put(inet_twsk(sk));
 705                 goto no_dccp_socket;
 706         }
 707
 708         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 709                 struct request_sock *req = inet_reqsk(sk);
 710                 struct sock *nsk;
 711
 712                 sk = req->rsk_listener;
 713                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 714                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 715                         goto lookup;
 716                 }
 717                 sock_hold(sk);
 718                 nsk = dccp_check_req(sk, skb, req);
 719                 if (!nsk) {
 720                         reqsk_put(req);
 721                         goto discard_and_relse;
 722                 }
 723                 if (nsk == sk) {
 724                         reqsk_put(req);
 725                 } else if (dccp_child_process(sk, nsk, skb)) {
 726                         dccp_v6_ctl_send_reset(sk, skb);
 727                         goto discard_and_relse;
 728                 } else {
 729                         sock_put(sk);
 730                         return 0;
 731                 }
 732         }
 733         /*
 734          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 735          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 736          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 737          */
 738         min_cov = dccp_sk(sk)->dccps_pcrlen;
 739         if (dh->dccph_cscov  &&  (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 740                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 741                               dh->dccph_cscov, min_cov);
 742                 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
 743                 goto discard_and_relse;
 744         }
 745
 746         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
 747                 goto discard_and_relse;
 748
 749         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4) ? -1 : 0;
 750
 751 no_dccp_socket:
 752         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 753                 goto discard_it;
 754         /*
 755          * Step 2:
 756          *      If no socket ...
 757          *              Generate Reset(No Connection) unless P.type == Reset
 758          *              Drop packet and return
 759          */
 760         if (dh->dccph_type != DCCP_PKT_RESET) {
 761                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 762                                         DCCP_RESET_CODE_NO_CONNECTION;
 763                 dccp_v6_ctl_send_reset(sk, skb);
 764         }
 765
 766 discard_it:
 767         kfree_skb(skb);
 768         return 0;
 769
 770 discard_and_relse:
 771         sock_put(sk);
 772         goto discard_it;
 773 }
 774
 775 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 776                            int addr_len)
 777 {
 778         struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr;
 779         struct inet_connection_sock *icsk = inet_csk(sk);
 780         struct inet_sock *inet = inet_sk(sk);
 781         struct ipv6_pinfo *np = inet6_sk(sk);
 782         struct dccp_sock *dp = dccp_sk(sk);
 783         struct in6_addr *saddr = NULL, *final_p, final;
 784         struct ipv6_txoptions *opt;
 785         struct flowi6 fl6;
 786         struct dst_entry *dst;
 787         int addr_type;
 788         int err;
 789
 790         dp->dccps_role = DCCP_ROLE_CLIENT;
 791
 792         if (addr_len < SIN6_LEN_RFC2133)
 793                 return -EINVAL;
 794
 795         if (usin->sin6_family != AF_INET6)
 796                 return -EAFNOSUPPORT;
 797
 798         memset(&fl6, 0, sizeof(fl6));
 799
 800         if (np->sndflow) {
 801                 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK;
 802                 IP6_ECN_flow_init(fl6.flowlabel);
 803                 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) {
 804                         struct ip6_flowlabel *flowlabel;
 805                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 806                         if (flowlabel == NULL)
 807                                 return -EINVAL;
 808                         fl6_sock_release(flowlabel);
 809                 }
 810         }
 811         /*
 812          * connect() to INADDR_ANY means loopback (BSD'ism).
 813          */
 814         if (ipv6_addr_any(&usin->sin6_addr))
 815                 usin->sin6_addr.s6_addr[15] = 1;
 816
 817         addr_type = ipv6_addr_type(&usin->sin6_addr);
 818
 819         if (addr_type & IPV6_ADDR_MULTICAST)
 820                 return -ENETUNREACH;
 821
 822         if (addr_type & IPV6_ADDR_LINKLOCAL) {
 823                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 824                     usin->sin6_scope_id) {
 825                         /* If interface is set while binding, indices
 826                          * must coincide.
 827                          */
 828                         if (sk->sk_bound_dev_if &&
 829                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 830                                 return -EINVAL;
 831
 832                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 833                 }
 834
 835                 /* Connect to link-local address requires an interface */
 836                 if (!sk->sk_bound_dev_if)
 837                         return -EINVAL;
 838         }
 839
 840         sk->sk_v6_daddr = usin->sin6_addr;
 841         np->flow_label = fl6.flowlabel;
 842
 843         /*
 844          * DCCP over IPv4
 845          */
 846         if (addr_type == IPV6_ADDR_MAPPED) {
 847                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 848                 struct sockaddr_in sin;
 849
 850                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 851
 852                 if (__ipv6_only_sock(sk))
 853                         return -ENETUNREACH;
 854
 855                 sin.sin_family = AF_INET;
 856                 sin.sin_port = usin->sin6_port;
 857                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 858
 859                 icsk->icsk_af_ops = &dccp_ipv6_mapped;
 860                 sk->sk_backlog_rcv = dccp_v4_do_rcv;
 861
 862                 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 863                 if (err) {
 864                         icsk->icsk_ext_hdr_len = exthdrlen;
 865                         icsk->icsk_af_ops = &dccp_ipv6_af_ops;
 866                         sk->sk_backlog_rcv = dccp_v6_do_rcv;
 867                         goto failure;
 868                 }
 869                 np->saddr = sk->sk_v6_rcv_saddr;
 870                 return err;
 871         }
 872
 873         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 874                 saddr = &sk->sk_v6_rcv_saddr;
 875
 876         fl6.flowi6_proto = IPPROTO_DCCP;
 877         fl6.daddr = sk->sk_v6_daddr;
 878         fl6.saddr = saddr ? *saddr : np->saddr;
 879         fl6.flowi6_oif = sk->sk_bound_dev_if;
 880         fl6.fl6_dport = usin->sin6_port;
 881         fl6.fl6_sport = inet->inet_sport;
 882         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 883
 884         opt = rcu_dereference_protected(np->opt, sock_owned_by_user(sk));
 885         final_p = fl6_update_dst(&fl6, opt, &final);
 886
 887         dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
 888         if (IS_ERR(dst)) {
 889                 err = PTR_ERR(dst);
 890                 goto failure;
 891         }
 892
 893         if (saddr == NULL) {
 894                 saddr = &fl6.saddr;
 895                 sk->sk_v6_rcv_saddr = *saddr;
 896         }
 897
 898         /* set the source address */
 899         np->saddr = *saddr;
 900         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 901
 902         ip6_dst_store(sk, dst, NULL, NULL);
 903
 904         icsk->icsk_ext_hdr_len = 0;
 905         if (opt)
 906                 icsk->icsk_ext_hdr_len = opt->opt_flen + opt->opt_nflen;
 907
 908         inet->inet_dport = usin->sin6_port;
 909
 910         dccp_set_state(sk, DCCP_REQUESTING);
 911         err = inet6_hash_connect(&dccp_death_row, sk);
 912         if (err)
 913                 goto late_failure;
 914
 915         dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32,
 916                                                       sk->sk_v6_daddr.s6_addr32,
 917                                                       inet->inet_sport,
 918                                                       inet->inet_dport);
 919         err = dccp_connect(sk);
 920         if (err)
 921                 goto late_failure;
 922
 923         return 0;
 924
 925 late_failure:
 926         dccp_set_state(sk, DCCP_CLOSED);
 927         __sk_dst_reset(sk);
 928 failure:
 929         inet->inet_dport = 0;
 930         sk->sk_route_caps = 0;
 931         return err;
 932 }
 933
 934 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = {
 935         .queue_xmit        = inet6_csk_xmit,
 936         .send_check        = dccp_v6_send_check,
 937         .rebuild_header    = inet6_sk_rebuild_header,
 938         .conn_request      = dccp_v6_conn_request,
 939         .syn_recv_sock     = dccp_v6_request_recv_sock,
 940         .net_header_len    = sizeof(struct ipv6hdr),
 941         .setsockopt        = ipv6_setsockopt,
 942         .getsockopt        = ipv6_getsockopt,
 943         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 944         .sockaddr_len      = sizeof(struct sockaddr_in6),
 945         .bind_conflict     = inet6_csk_bind_conflict,
 946 #ifdef CONFIG_COMPAT
 947         .compat_setsockopt = compat_ipv6_setsockopt,
 948         .compat_getsockopt = compat_ipv6_getsockopt,
 949 #endif
 950 };
 951
 952 /*
 953  *      DCCP over IPv4 via INET6 API
 954  */
 955 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = {
 956         .queue_xmit        = ip_queue_xmit,
 957         .send_check        = dccp_v4_send_check,
 958         .rebuild_header    = inet_sk_rebuild_header,
 959         .conn_request      = dccp_v6_conn_request,
 960         .syn_recv_sock     = dccp_v6_request_recv_sock,
 961         .net_header_len    = sizeof(struct iphdr),
 962         .setsockopt        = ipv6_setsockopt,
 963         .getsockopt        = ipv6_getsockopt,
 964         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 965         .sockaddr_len      = sizeof(struct sockaddr_in6),
 966         .bind_conflict     = inet6_csk_bind_conflict,
 967 #ifdef CONFIG_COMPAT
 968         .compat_setsockopt = compat_ipv6_setsockopt,
 969         .compat_getsockopt = compat_ipv6_getsockopt,
 970 #endif
 971 };
 972
 973 /* NOTE: A lot of things set to zero explicitly by call to
 974  *       sk_alloc() so need not be done here.
 975  */
 976 static int dccp_v6_init_sock(struct sock *sk)
 977 {
 978         static __u8 dccp_v6_ctl_sock_initialized;
 979         int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized);
 980
 981         if (err == 0) {
 982                 if (unlikely(!dccp_v6_ctl_sock_initialized))
 983                         dccp_v6_ctl_sock_initialized = 1;
 984                 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops;
 985         }
 986
 987         return err;
 988 }
 989
 990 static void dccp_v6_destroy_sock(struct sock *sk)
 991 {
 992         dccp_destroy_sock(sk);
 993         inet6_destroy_sock(sk);
 994 }
 995
 996 static struct timewait_sock_ops dccp6_timewait_sock_ops = {
 997         .twsk_obj_size  = sizeof(struct dccp6_timewait_sock),
 998 };
 999
1000 static struct proto dccp_v6_prot = {
1001         .name              = "DCCPv6",
1002         .owner             = THIS_MODULE,
1003         .close             = dccp_close,
1004         .connect           = dccp_v6_connect,
1005         .disconnect        = dccp_disconnect,
1006         .ioctl             = dccp_ioctl,
1007         .init              = dccp_v6_init_sock,
1008         .setsockopt        = dccp_setsockopt,
1009         .getsockopt        = dccp_getsockopt,
1010         .sendmsg           = dccp_sendmsg,
1011         .recvmsg           = dccp_recvmsg,
1012         .backlog_rcv       = dccp_v6_do_rcv,
1013         .hash              = inet_hash,
1014         .unhash            = inet_unhash,
1015         .accept            = inet_csk_accept,
1016         .get_port          = inet_csk_get_port,
1017         .shutdown          = dccp_shutdown,
1018         .destroy           = dccp_v6_destroy_sock,
1019         .orphan_count      = &dccp_orphan_count,
1020         .max_header        = MAX_DCCP_HEADER,
1021         .obj_size          = sizeof(struct dccp6_sock),
1022         .slab_flags        = SLAB_DESTROY_BY_RCU,
1023         .rsk_prot          = &dccp6_request_sock_ops,
1024         .twsk_prot         = &dccp6_timewait_sock_ops,
1025         .h.hashinfo        = &dccp_hashinfo,
1026 #ifdef CONFIG_COMPAT
1027         .compat_setsockopt = compat_dccp_setsockopt,
1028         .compat_getsockopt = compat_dccp_getsockopt,
1029 #endif
1030 };
1031
1032 static const struct inet6_protocol dccp_v6_protocol = {
1033         .handler        = dccp_v6_rcv,
1034         .err_handler    = dccp_v6_err,
1035         .flags          = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1036 };
1037
1038 static const struct proto_ops inet6_dccp_ops = {
1039         .family            = PF_INET6,
1040         .owner             = THIS_MODULE,
1041         .release           = inet6_release,
1042         .bind              = inet6_bind,
1043         .connect           = inet_stream_connect,
1044         .socketpair        = sock_no_socketpair,
1045         .accept            = inet_accept,
1046         .getname           = inet6_getname,
1047         .poll              = dccp_poll,
1048         .ioctl             = inet6_ioctl,
1049         .listen            = inet_dccp_listen,
1050         .shutdown          = inet_shutdown,
1051         .setsockopt        = sock_common_setsockopt,
1052         .getsockopt        = sock_common_getsockopt,
1053         .sendmsg           = inet_sendmsg,
1054         .recvmsg           = sock_common_recvmsg,
1055         .mmap              = sock_no_mmap,
1056         .sendpage          = sock_no_sendpage,
1057 #ifdef CONFIG_COMPAT
1058         .compat_setsockopt = compat_sock_common_setsockopt,
1059         .compat_getsockopt = compat_sock_common_getsockopt,
1060 #endif
1061 };
1062
1063 static struct inet_protosw dccp_v6_protosw = {
1064         .type           = SOCK_DCCP,
1065         .protocol       = IPPROTO_DCCP,
1066         .prot           = &dccp_v6_prot,
1067         .ops            = &inet6_dccp_ops,
1068         .flags          = INET_PROTOSW_ICSK,
1069 };
1070
1071 static int __net_init dccp_v6_init_net(struct net *net)
1072 {
1073         if (dccp_hashinfo.bhash == NULL)
1074                 return -ESOCKTNOSUPPORT;
1075
1076         return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6,
1077                                     SOCK_DCCP, IPPROTO_DCCP, net);
1078 }
1079
1080 static void __net_exit dccp_v6_exit_net(struct net *net)
1081 {
1082         inet_ctl_sock_destroy(net->dccp.v6_ctl_sk);
1083 }
1084
1085 static struct pernet_operations dccp_v6_ops = {
1086         .init   = dccp_v6_init_net,
1087         .exit   = dccp_v6_exit_net,
1088 };
1089
1090 static int __init dccp_v6_init(void)
1091 {
1092         int err = proto_register(&dccp_v6_prot, 1);
1093
1094         if (err != 0)
1095                 goto out;
1096
1097         err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1098         if (err != 0)
1099                 goto out_unregister_proto;
1100
1101         inet6_register_protosw(&dccp_v6_protosw);
1102
1103         err = register_pernet_subsys(&dccp_v6_ops);
1104         if (err != 0)
1105                 goto out_destroy_ctl_sock;
1106 out:
1107         return err;
1108
1109 out_destroy_ctl_sock:
1110         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1111         inet6_unregister_protosw(&dccp_v6_protosw);
1112 out_unregister_proto:
1113         proto_unregister(&dccp_v6_prot);
1114         goto out;
1115 }
1116
1117 static void __exit dccp_v6_exit(void)
1118 {
1119         unregister_pernet_subsys(&dccp_v6_ops);
1120         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1121         inet6_unregister_protosw(&dccp_v6_protosw);
1122         proto_unregister(&dccp_v6_prot);
1123 }
1124
1125 module_init(dccp_v6_init);
1126 module_exit(dccp_v6_exit);
1127
1128 /*
1129  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1130  * values directly, Also cover the case where the protocol is not specified,
1131  * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1132  */
1133 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6);
1134 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6);
1135 MODULE_LICENSE("GPL");
1136 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1137 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");