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(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(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(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(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                 struct ipv6_txoptions *opt;
 223
 224                 dh->dccph_checksum = dccp_v6_csum_finish(skb,
 225                                                          &ireq->ir_v6_loc_addr,
 226                                                          &ireq->ir_v6_rmt_addr);
 227                 fl6.daddr = ireq->ir_v6_rmt_addr;
 228                 rcu_read_lock();
 229                 opt = ireq->ipv6_opt;
 230                 if (!opt)
 231                         opt = rcu_dereference(np->opt);
 232                 err = ip6_xmit(sk, skb, &fl6, sk->sk_mark, opt, np->tclass);
 233                 rcu_read_unlock();
 234                 err = net_xmit_eval(err);
 235         }
 236
 237 done:
 238         dst_release(dst);
 239         return err;
 240 }
 241
 242 static void dccp_v6_reqsk_destructor(struct request_sock *req)
 243 {
 244         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 245         kfree(inet_rsk(req)->ipv6_opt);
 246         kfree_skb(inet_rsk(req)->pktopts);
 247 }
 248
 249 static void dccp_v6_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 250 {
 251         const struct ipv6hdr *rxip6h;
 252         struct sk_buff *skb;
 253         struct flowi6 fl6;
 254         struct net *net = dev_net(skb_dst(rxskb)->dev);
 255         struct sock *ctl_sk = net->dccp.v6_ctl_sk;
 256         struct dst_entry *dst;
 257
 258         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 259                 return;
 260
 261         if (!ipv6_unicast_destination(rxskb))
 262                 return;
 263
 264         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 265         if (skb == NULL)
 266                 return;
 267
 268         rxip6h = ipv6_hdr(rxskb);
 269         dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr,
 270                                                             &rxip6h->daddr);
 271
 272         memset(&fl6, 0, sizeof(fl6));
 273         fl6.daddr = rxip6h->saddr;
 274         fl6.saddr = rxip6h->daddr;
 275
 276         fl6.flowi6_proto = IPPROTO_DCCP;
 277         fl6.flowi6_oif = inet6_iif(rxskb);
 278         fl6.fl6_dport = dccp_hdr(skb)->dccph_dport;
 279         fl6.fl6_sport = dccp_hdr(skb)->dccph_sport;
 280         security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6));
 281
 282         /* sk = NULL, but it is safe for now. RST socket required. */
 283         dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL);
 284         if (!IS_ERR(dst)) {
 285                 skb_dst_set(skb, dst);
 286                 ip6_xmit(ctl_sk, skb, &fl6, 0, NULL, 0);
 287                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 288                 DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 289                 return;
 290         }
 291
 292         kfree_skb(skb);
 293 }
 294
 295 static struct request_sock_ops dccp6_request_sock_ops = {
 296         .family         = AF_INET6,
 297         .obj_size       = sizeof(struct dccp6_request_sock),
 298         .rtx_syn_ack    = dccp_v6_send_response,
 299         .send_ack       = dccp_reqsk_send_ack,
 300         .destructor     = dccp_v6_reqsk_destructor,
 301         .send_reset     = dccp_v6_ctl_send_reset,
 302         .syn_ack_timeout = dccp_syn_ack_timeout,
 303 };
 304
 305 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 306 {
 307         struct request_sock *req;
 308         struct dccp_request_sock *dreq;
 309         struct inet_request_sock *ireq;
 310         struct ipv6_pinfo *np = inet6_sk(sk);
 311         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 312         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 313
 314         if (skb->protocol == htons(ETH_P_IP))
 315                 return dccp_v4_conn_request(sk, skb);
 316
 317         if (!ipv6_unicast_destination(skb))
 318                 return 0;       /* discard, don't send a reset here */
 319
 320         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 321                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 322                 return 0;
 323         }
 324
 325         if (dccp_bad_service_code(sk, service)) {
 326                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 327                 goto drop;
 328         }
 329         /*
 330          * There are no SYN attacks on IPv6, yet...
 331          */
 332         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 333         if (inet_csk_reqsk_queue_is_full(sk))
 334                 goto drop;
 335
 336         if (sk_acceptq_is_full(sk))
 337                 goto drop;
 338
 339         req = inet_reqsk_alloc(&dccp6_request_sock_ops, sk, true);
 340         if (req == NULL)
 341                 goto drop;
 342
 343         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 344                 goto drop_and_free;
 345
 346         dreq = dccp_rsk(req);
 347         if (dccp_parse_options(sk, dreq, skb))
 348                 goto drop_and_free;
 349
 350         if (security_inet_conn_request(sk, skb, req))
 351                 goto drop_and_free;
 352
 353         ireq = inet_rsk(req);
 354         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 355         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 356         ireq->ireq_family = AF_INET6;
 357         ireq->ir_mark = inet_request_mark(sk, skb);
 358
 359         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
 360             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 361             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 362                 atomic_inc(&skb->users);
 363                 ireq->pktopts = skb;
 364         }
 365         ireq->ir_iif = sk->sk_bound_dev_if;
 366
 367         /* So that link locals have meaning */
 368         if (!sk->sk_bound_dev_if &&
 369             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 370                 ireq->ir_iif = inet6_iif(skb);
 371
 372         /*
 373          * Step 3: Process LISTEN state
 374          *
 375          *   Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 376          *
 377          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 378          */
 379         dreq->dreq_isr     = dcb->dccpd_seq;
 380         dreq->dreq_gsr     = dreq->dreq_isr;
 381         dreq->dreq_iss     = dccp_v6_init_sequence(skb);
 382         dreq->dreq_gss     = dreq->dreq_iss;
 383         dreq->dreq_service = service;
 384
 385         if (dccp_v6_send_response(sk, req))
 386                 goto drop_and_free;
 387
 388         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 389         reqsk_put(req);
 390         return 0;
 391
 392 drop_and_free:
 393         reqsk_free(req);
 394 drop:
 395         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 396         return -1;
 397 }
 398
 399 static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,
 400                                               struct sk_buff *skb,
 401                                               struct request_sock *req,
 402                                               struct dst_entry *dst,
 403                                               struct request_sock *req_unhash,
 404                                               bool *own_req)
 405 {
 406         struct inet_request_sock *ireq = inet_rsk(req);
 407         struct ipv6_pinfo *newnp;
 408         const struct ipv6_pinfo *np = inet6_sk(sk);
 409         struct ipv6_txoptions *opt;
 410         struct inet_sock *newinet;
 411         struct dccp6_sock *newdp6;
 412         struct sock *newsk;
 413
 414         if (skb->protocol == htons(ETH_P_IP)) {
 415                 /*
 416                  *      v6 mapped
 417                  */
 418                 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst,
 419                                                   req_unhash, own_req);
 420                 if (newsk == NULL)
 421                         return NULL;
 422
 423                 newdp6 = (struct dccp6_sock *)newsk;
 424                 newinet = inet_sk(newsk);
 425                 newinet->pinet6 = &newdp6->inet6;
 426                 newnp = inet6_sk(newsk);
 427
 428                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 429
 430                 newnp->saddr = newsk->sk_v6_rcv_saddr;
 431
 432                 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
 433                 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
 434                 newnp->pktoptions  = NULL;
 435                 newnp->opt         = NULL;
 436                 newnp->ipv6_mc_list = NULL;
 437                 newnp->ipv6_ac_list = NULL;
 438                 newnp->ipv6_fl_list = NULL;
 439                 newnp->mcast_oif   = inet_iif(skb);
 440                 newnp->mcast_hops  = ip_hdr(skb)->ttl;
 441
 442                 /*
 443                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
 444                  * here, dccp_create_openreq_child now does this for us, see the comment in
 445                  * that function for the gory details. -acme
 446                  */
 447
 448                 /* It is tricky place. Until this moment IPv4 tcp
 449                    worked with IPv6 icsk.icsk_af_ops.
 450                    Sync it now.
 451                  */
 452                 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
 453
 454                 return newsk;
 455         }
 456
 457
 458         if (sk_acceptq_is_full(sk))
 459                 goto out_overflow;
 460
 461         if (!dst) {
 462                 struct flowi6 fl6;
 463
 464                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_DCCP);
 465                 if (!dst)
 466                         goto out;
 467         }
 468
 469         newsk = dccp_create_openreq_child(sk, req, skb);
 470         if (newsk == NULL)
 471                 goto out_nonewsk;
 472
 473         /*
 474          * No need to charge this sock to the relevant IPv6 refcnt debug socks
 475          * count here, dccp_create_openreq_child now does this for us, see the
 476          * comment in that function for the gory details. -acme
 477          */
 478
 479         ip6_dst_store(newsk, dst, NULL, NULL);
 480         newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
 481                                                       NETIF_F_TSO);
 482         newdp6 = (struct dccp6_sock *)newsk;
 483         newinet = inet_sk(newsk);
 484         newinet->pinet6 = &newdp6->inet6;
 485         newnp = inet6_sk(newsk);
 486
 487         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 488
 489         newsk->sk_v6_daddr      = ireq->ir_v6_rmt_addr;
 490         newnp->saddr            = ireq->ir_v6_loc_addr;
 491         newsk->sk_v6_rcv_saddr  = ireq->ir_v6_loc_addr;
 492         newsk->sk_bound_dev_if  = ireq->ir_iif;
 493
 494         /* Now IPv6 options...
 495
 496            First: no IPv4 options.
 497          */
 498         newinet->inet_opt = NULL;
 499
 500         /* Clone RX bits */
 501         newnp->rxopt.all = np->rxopt.all;
 502
 503         newnp->ipv6_mc_list = NULL;
 504         newnp->ipv6_ac_list = NULL;
 505         newnp->ipv6_fl_list = NULL;
 506         newnp->pktoptions = NULL;
 507         newnp->opt        = NULL;
 508         newnp->mcast_oif  = inet6_iif(skb);
 509         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
 510
 511         /*
 512          * Clone native IPv6 options from listening socket (if any)
 513          *
 514          * Yes, keeping reference count would be much more clever, but we make
 515          * one more one thing there: reattach optmem to newsk.
 516          */
 517         opt = ireq->ipv6_opt;
 518         if (!opt)
 519                 opt = rcu_dereference(np->opt);
 520         if (opt) {
 521                 opt = ipv6_dup_options(newsk, opt);
 522                 RCU_INIT_POINTER(newnp->opt, opt);
 523         }
 524         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 525         if (opt)
 526                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
 527                                                     opt->opt_flen;
 528
 529         dccp_sync_mss(newsk, dst_mtu(dst));
 530
 531         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
 532         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
 533
 534         if (__inet_inherit_port(sk, newsk) < 0) {
 535                 inet_csk_prepare_forced_close(newsk);
 536                 dccp_done(newsk);
 537                 goto out;
 538         }
 539         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 540         /* Clone pktoptions received with SYN, if we own the req */
 541         if (*own_req && ireq->pktopts) {
 542                 newnp->pktoptions = skb_clone(ireq->pktopts, GFP_ATOMIC);
 543                 consume_skb(ireq->pktopts);
 544                 ireq->pktopts = NULL;
 545                 if (newnp->pktoptions)
 546                         skb_set_owner_r(newnp->pktoptions, newsk);
 547         }
 548
 549         return newsk;
 550
 551 out_overflow:
 552         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 553 out_nonewsk:
 554         dst_release(dst);
 555 out:
 556         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 557         return NULL;
 558 }
 559
 560 /* The socket must have it's spinlock held when we get
 561  * here.
 562  *
 563  * We have a potential double-lock case here, so even when
 564  * doing backlog processing we use the BH locking scheme.
 565  * This is because we cannot sleep with the original spinlock
 566  * held.
 567  */
 568 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 569 {
 570         struct ipv6_pinfo *np = inet6_sk(sk);
 571         struct sk_buff *opt_skb = NULL;
 572
 573         /* Imagine: socket is IPv6. IPv4 packet arrives,
 574            goes to IPv4 receive handler and backlogged.
 575            From backlog it always goes here. Kerboom...
 576            Fortunately, dccp_rcv_established and rcv_established
 577            handle them correctly, but it is not case with
 578            dccp_v6_hnd_req and dccp_v6_ctl_send_reset().   --ANK
 579          */
 580
 581         if (skb->protocol == htons(ETH_P_IP))
 582                 return dccp_v4_do_rcv(sk, skb);
 583
 584         if (sk_filter(sk, skb))
 585                 goto discard;
 586
 587         /*
 588          * socket locking is here for SMP purposes as backlog rcv is currently
 589          * called with bh processing disabled.
 590          */
 591
 592         /* Do Stevens' IPV6_PKTOPTIONS.
 593
 594            Yes, guys, it is the only place in our code, where we
 595            may make it not affecting IPv4.
 596            The rest of code is protocol independent,
 597            and I do not like idea to uglify IPv4.
 598
 599            Actually, all the idea behind IPV6_PKTOPTIONS
 600            looks not very well thought. For now we latch
 601            options, received in the last packet, enqueued
 602            by tcp. Feel free to propose better solution.
 603                                                --ANK (980728)
 604          */
 605         if (np->rxopt.all)
 606         /*
 607          * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
 608          *        (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
 609          */
 610                 opt_skb = skb_clone(skb, GFP_ATOMIC);
 611
 612         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 613                 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
 614                         goto reset;
 615                 if (opt_skb) {
 616                         /* XXX This is where we would goto ipv6_pktoptions. */
 617                         __kfree_skb(opt_skb);
 618                 }
 619                 return 0;
 620         }
 621
 622         /*
 623          *  Step 3: Process LISTEN state
 624          *     If S.state == LISTEN,
 625          *       If P.type == Request or P contains a valid Init Cookie option,
 626          *            (* Must scan the packet's options to check for Init
 627          *               Cookies.  Only Init Cookies are processed here,
 628          *               however; other options are processed in Step 8.  This
 629          *               scan need only be performed if the endpoint uses Init
 630          *               Cookies *)
 631          *            (* Generate a new socket and switch to that socket *)
 632          *            Set S := new socket for this port pair
 633          *            S.state = RESPOND
 634          *            Choose S.ISS (initial seqno) or set from Init Cookies
 635          *            Initialize S.GAR := S.ISS
 636          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 637          *            Continue with S.state == RESPOND
 638          *            (* A Response packet will be generated in Step 11 *)
 639          *       Otherwise,
 640          *            Generate Reset(No Connection) unless P.type == Reset
 641          *            Drop packet and return
 642          *
 643          * NOTE: the check for the packet types is done in
 644          *       dccp_rcv_state_process
 645          */
 646
 647         if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
 648                 goto reset;
 649         if (opt_skb) {
 650                 /* XXX This is where we would goto ipv6_pktoptions. */
 651                 __kfree_skb(opt_skb);
 652         }
 653         return 0;
 654
 655 reset:
 656         dccp_v6_ctl_send_reset(sk, skb);
 657 discard:
 658         if (opt_skb != NULL)
 659                 __kfree_skb(opt_skb);
 660         kfree_skb(skb);
 661         return 0;
 662 }
 663
 664 static int dccp_v6_rcv(struct sk_buff *skb)
 665 {
 666         const struct dccp_hdr *dh;
 667         bool refcounted;
 668         struct sock *sk;
 669         int min_cov;
 670
 671         /* Step 1: Check header basics */
 672
 673         if (dccp_invalid_packet(skb))
 674                 goto discard_it;
 675
 676         /* Step 1: If header checksum is incorrect, drop packet and return. */
 677         if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
 678                                      &ipv6_hdr(skb)->daddr)) {
 679                 DCCP_WARN("dropped packet with invalid checksum\n");
 680                 goto discard_it;
 681         }
 682
 683         dh = dccp_hdr(skb);
 684
 685         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 686         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 687
 688         if (dccp_packet_without_ack(skb))
 689                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 690         else
 691                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 692
 693 lookup:
 694         sk = __inet6_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 695                                 dh->dccph_sport, dh->dccph_dport,
 696                                 inet6_iif(skb), &refcounted);
 697         if (!sk) {
 698                 dccp_pr_debug("failed to look up flow ID in table and "
 699                               "get corresponding socket\n");
 700                 goto no_dccp_socket;
 701         }
 702
 703         /*
 704          * Step 2:
 705          *      ... or S.state == TIMEWAIT,
 706          *              Generate Reset(No Connection) unless P.type == Reset
 707          *              Drop packet and return
 708          */
 709         if (sk->sk_state == DCCP_TIME_WAIT) {
 710                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 711                 inet_twsk_put(inet_twsk(sk));
 712                 goto no_dccp_socket;
 713         }
 714
 715         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 716                 struct request_sock *req = inet_reqsk(sk);
 717                 struct sock *nsk;
 718
 719                 sk = req->rsk_listener;
 720                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 721                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 722                         goto lookup;
 723                 }
 724                 sock_hold(sk);
 725                 refcounted = true;
 726                 nsk = dccp_check_req(sk, skb, req);
 727                 if (!nsk) {
 728                         reqsk_put(req);
 729                         goto discard_and_relse;
 730                 }
 731                 if (nsk == sk) {
 732                         reqsk_put(req);
 733                 } else if (dccp_child_process(sk, nsk, skb)) {
 734                         dccp_v6_ctl_send_reset(sk, skb);
 735                         goto discard_and_relse;
 736                 } else {
 737                         sock_put(sk);
 738                         return 0;
 739                 }
 740         }
 741         /*
 742          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 743          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 744          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 745          */
 746         min_cov = dccp_sk(sk)->dccps_pcrlen;
 747         if (dh->dccph_cscov  &&  (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 748                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 749                               dh->dccph_cscov, min_cov);
 750                 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
 751                 goto discard_and_relse;
 752         }
 753
 754         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
 755                 goto discard_and_relse;
 756
 757         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4,
 758                                 refcounted) ? -1 : 0;
 759
 760 no_dccp_socket:
 761         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 762                 goto discard_it;
 763         /*
 764          * Step 2:
 765          *      If no socket ...
 766          *              Generate Reset(No Connection) unless P.type == Reset
 767          *              Drop packet and return
 768          */
 769         if (dh->dccph_type != DCCP_PKT_RESET) {
 770                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 771                                         DCCP_RESET_CODE_NO_CONNECTION;
 772                 dccp_v6_ctl_send_reset(sk, skb);
 773         }
 774
 775 discard_it:
 776         kfree_skb(skb);
 777         return 0;
 778
 779 discard_and_relse:
 780         if (refcounted)
 781                 sock_put(sk);
 782         goto discard_it;
 783 }
 784
 785 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 786                            int addr_len)
 787 {
 788         struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr;
 789         struct inet_connection_sock *icsk = inet_csk(sk);
 790         struct inet_sock *inet = inet_sk(sk);
 791         struct ipv6_pinfo *np = inet6_sk(sk);
 792         struct dccp_sock *dp = dccp_sk(sk);
 793         struct in6_addr *saddr = NULL, *final_p, final;
 794         struct ipv6_txoptions *opt;
 795         struct flowi6 fl6;
 796         struct dst_entry *dst;
 797         int addr_type;
 798         int err;
 799
 800         dp->dccps_role = DCCP_ROLE_CLIENT;
 801
 802         if (addr_len < SIN6_LEN_RFC2133)
 803                 return -EINVAL;
 804
 805         if (usin->sin6_family != AF_INET6)
 806                 return -EAFNOSUPPORT;
 807
 808         memset(&fl6, 0, sizeof(fl6));
 809
 810         if (np->sndflow) {
 811                 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK;
 812                 IP6_ECN_flow_init(fl6.flowlabel);
 813                 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) {
 814                         struct ip6_flowlabel *flowlabel;
 815                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 816                         if (flowlabel == NULL)
 817                                 return -EINVAL;
 818                         fl6_sock_release(flowlabel);
 819                 }
 820         }
 821         /*
 822          * connect() to INADDR_ANY means loopback (BSD'ism).
 823          */
 824         if (ipv6_addr_any(&usin->sin6_addr))
 825                 usin->sin6_addr.s6_addr[15] = 1;
 826
 827         addr_type = ipv6_addr_type(&usin->sin6_addr);
 828
 829         if (addr_type & IPV6_ADDR_MULTICAST)
 830                 return -ENETUNREACH;
 831
 832         if (addr_type & IPV6_ADDR_LINKLOCAL) {
 833                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 834                     usin->sin6_scope_id) {
 835                         /* If interface is set while binding, indices
 836                          * must coincide.
 837                          */
 838                         if (sk->sk_bound_dev_if &&
 839                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 840                                 return -EINVAL;
 841
 842                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 843                 }
 844
 845                 /* Connect to link-local address requires an interface */
 846                 if (!sk->sk_bound_dev_if)
 847                         return -EINVAL;
 848         }
 849
 850         sk->sk_v6_daddr = usin->sin6_addr;
 851         np->flow_label = fl6.flowlabel;
 852
 853         /*
 854          * DCCP over IPv4
 855          */
 856         if (addr_type == IPV6_ADDR_MAPPED) {
 857                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 858                 struct sockaddr_in sin;
 859
 860                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 861
 862                 if (__ipv6_only_sock(sk))
 863                         return -ENETUNREACH;
 864
 865                 sin.sin_family = AF_INET;
 866                 sin.sin_port = usin->sin6_port;
 867                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 868
 869                 icsk->icsk_af_ops = &dccp_ipv6_mapped;
 870                 sk->sk_backlog_rcv = dccp_v4_do_rcv;
 871
 872                 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 873                 if (err) {
 874                         icsk->icsk_ext_hdr_len = exthdrlen;
 875                         icsk->icsk_af_ops = &dccp_ipv6_af_ops;
 876                         sk->sk_backlog_rcv = dccp_v6_do_rcv;
 877                         goto failure;
 878                 }
 879                 np->saddr = sk->sk_v6_rcv_saddr;
 880                 return err;
 881         }
 882
 883         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 884                 saddr = &sk->sk_v6_rcv_saddr;
 885
 886         fl6.flowi6_proto = IPPROTO_DCCP;
 887         fl6.daddr = sk->sk_v6_daddr;
 888         fl6.saddr = saddr ? *saddr : np->saddr;
 889         fl6.flowi6_oif = sk->sk_bound_dev_if;
 890         fl6.fl6_dport = usin->sin6_port;
 891         fl6.fl6_sport = inet->inet_sport;
 892         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 893
 894         opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
 895         final_p = fl6_update_dst(&fl6, opt, &final);
 896
 897         dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
 898         if (IS_ERR(dst)) {
 899                 err = PTR_ERR(dst);
 900                 goto failure;
 901         }
 902
 903         if (saddr == NULL) {
 904                 saddr = &fl6.saddr;
 905                 sk->sk_v6_rcv_saddr = *saddr;
 906         }
 907
 908         /* set the source address */
 909         np->saddr = *saddr;
 910         inet->inet_rcv_saddr = LOOPBACK4_IPV6;
 911
 912         ip6_dst_store(sk, dst, NULL, NULL);
 913
 914         icsk->icsk_ext_hdr_len = 0;
 915         if (opt)
 916                 icsk->icsk_ext_hdr_len = opt->opt_flen + opt->opt_nflen;
 917
 918         inet->inet_dport = usin->sin6_port;
 919
 920         dccp_set_state(sk, DCCP_REQUESTING);
 921         err = inet6_hash_connect(&dccp_death_row, sk);
 922         if (err)
 923                 goto late_failure;
 924
 925         dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32,
 926                                                       sk->sk_v6_daddr.s6_addr32,
 927                                                       inet->inet_sport,
 928                                                       inet->inet_dport);
 929         err = dccp_connect(sk);
 930         if (err)
 931                 goto late_failure;
 932
 933         return 0;
 934
 935 late_failure:
 936         dccp_set_state(sk, DCCP_CLOSED);
 937         __sk_dst_reset(sk);
 938 failure:
 939         inet->inet_dport = 0;
 940         sk->sk_route_caps = 0;
 941         return err;
 942 }
 943
 944 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = {
 945         .queue_xmit        = inet6_csk_xmit,
 946         .send_check        = dccp_v6_send_check,
 947         .rebuild_header    = inet6_sk_rebuild_header,
 948         .conn_request      = dccp_v6_conn_request,
 949         .syn_recv_sock     = dccp_v6_request_recv_sock,
 950         .net_header_len    = sizeof(struct ipv6hdr),
 951         .setsockopt        = ipv6_setsockopt,
 952         .getsockopt        = ipv6_getsockopt,
 953         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 954         .sockaddr_len      = sizeof(struct sockaddr_in6),
 955         .bind_conflict     = inet6_csk_bind_conflict,
 956 #ifdef CONFIG_COMPAT
 957         .compat_setsockopt = compat_ipv6_setsockopt,
 958         .compat_getsockopt = compat_ipv6_getsockopt,
 959 #endif
 960 };
 961
 962 /*
 963  *      DCCP over IPv4 via INET6 API
 964  */
 965 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = {
 966         .queue_xmit        = ip_queue_xmit,
 967         .send_check        = dccp_v4_send_check,
 968         .rebuild_header    = inet_sk_rebuild_header,
 969         .conn_request      = dccp_v6_conn_request,
 970         .syn_recv_sock     = dccp_v6_request_recv_sock,
 971         .net_header_len    = sizeof(struct iphdr),
 972         .setsockopt        = ipv6_setsockopt,
 973         .getsockopt        = ipv6_getsockopt,
 974         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 975         .sockaddr_len      = sizeof(struct sockaddr_in6),
 976         .bind_conflict     = inet6_csk_bind_conflict,
 977 #ifdef CONFIG_COMPAT
 978         .compat_setsockopt = compat_ipv6_setsockopt,
 979         .compat_getsockopt = compat_ipv6_getsockopt,
 980 #endif
 981 };
 982
 983 /* NOTE: A lot of things set to zero explicitly by call to
 984  *       sk_alloc() so need not be done here.
 985  */
 986 static int dccp_v6_init_sock(struct sock *sk)
 987 {
 988         static __u8 dccp_v6_ctl_sock_initialized;
 989         int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized);
 990
 991         if (err == 0) {
 992                 if (unlikely(!dccp_v6_ctl_sock_initialized))
 993                         dccp_v6_ctl_sock_initialized = 1;
 994                 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops;
 995         }
 996
 997         return err;
 998 }
 999
1000 static void dccp_v6_destroy_sock(struct sock *sk)
1001 {
1002         dccp_destroy_sock(sk);
1003         inet6_destroy_sock(sk);
1004 }
1005
1006 static struct timewait_sock_ops dccp6_timewait_sock_ops = {
1007         .twsk_obj_size  = sizeof(struct dccp6_timewait_sock),
1008 };
1009
1010 static struct proto dccp_v6_prot = {
1011         .name              = "DCCPv6",
1012         .owner             = THIS_MODULE,
1013         .close             = dccp_close,
1014         .connect           = dccp_v6_connect,
1015         .disconnect        = dccp_disconnect,
1016         .ioctl             = dccp_ioctl,
1017         .init              = dccp_v6_init_sock,
1018         .setsockopt        = dccp_setsockopt,
1019         .getsockopt        = dccp_getsockopt,
1020         .sendmsg           = dccp_sendmsg,
1021         .recvmsg           = dccp_recvmsg,
1022         .backlog_rcv       = dccp_v6_do_rcv,
1023         .hash              = inet6_hash,
1024         .unhash            = inet_unhash,
1025         .accept            = inet_csk_accept,
1026         .get_port          = inet_csk_get_port,
1027         .shutdown          = dccp_shutdown,
1028         .destroy           = dccp_v6_destroy_sock,
1029         .orphan_count      = &dccp_orphan_count,
1030         .max_header        = MAX_DCCP_HEADER,
1031         .obj_size          = sizeof(struct dccp6_sock),
1032         .slab_flags        = SLAB_DESTROY_BY_RCU,
1033         .rsk_prot          = &dccp6_request_sock_ops,
1034         .twsk_prot         = &dccp6_timewait_sock_ops,
1035         .h.hashinfo        = &dccp_hashinfo,
1036 #ifdef CONFIG_COMPAT
1037         .compat_setsockopt = compat_dccp_setsockopt,
1038         .compat_getsockopt = compat_dccp_getsockopt,
1039 #endif
1040 };
1041
1042 static const struct inet6_protocol dccp_v6_protocol = {
1043         .handler        = dccp_v6_rcv,
1044         .err_handler    = dccp_v6_err,
1045         .flags          = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1046 };
1047
1048 static const struct proto_ops inet6_dccp_ops = {
1049         .family            = PF_INET6,
1050         .owner             = THIS_MODULE,
1051         .release           = inet6_release,
1052         .bind              = inet6_bind,
1053         .connect           = inet_stream_connect,
1054         .socketpair        = sock_no_socketpair,
1055         .accept            = inet_accept,
1056         .getname           = inet6_getname,
1057         .poll              = dccp_poll,
1058         .ioctl             = inet6_ioctl,
1059         .listen            = inet_dccp_listen,
1060         .shutdown          = inet_shutdown,
1061         .setsockopt        = sock_common_setsockopt,
1062         .getsockopt        = sock_common_getsockopt,
1063         .sendmsg           = inet_sendmsg,
1064         .recvmsg           = sock_common_recvmsg,
1065         .mmap              = sock_no_mmap,
1066         .sendpage          = sock_no_sendpage,
1067 #ifdef CONFIG_COMPAT
1068         .compat_setsockopt = compat_sock_common_setsockopt,
1069         .compat_getsockopt = compat_sock_common_getsockopt,
1070 #endif
1071 };
1072
1073 static struct inet_protosw dccp_v6_protosw = {
1074         .type           = SOCK_DCCP,
1075         .protocol       = IPPROTO_DCCP,
1076         .prot           = &dccp_v6_prot,
1077         .ops            = &inet6_dccp_ops,
1078         .flags          = INET_PROTOSW_ICSK,
1079 };
1080
1081 static int __net_init dccp_v6_init_net(struct net *net)
1082 {
1083         if (dccp_hashinfo.bhash == NULL)
1084                 return -ESOCKTNOSUPPORT;
1085
1086         return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6,
1087                                     SOCK_DCCP, IPPROTO_DCCP, net);
1088 }
1089
1090 static void __net_exit dccp_v6_exit_net(struct net *net)
1091 {
1092         inet_ctl_sock_destroy(net->dccp.v6_ctl_sk);
1093 }
1094
1095 static struct pernet_operations dccp_v6_ops = {
1096         .init   = dccp_v6_init_net,
1097         .exit   = dccp_v6_exit_net,
1098 };
1099
1100 static int __init dccp_v6_init(void)
1101 {
1102         int err = proto_register(&dccp_v6_prot, 1);
1103
1104         if (err != 0)
1105                 goto out;
1106
1107         err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1108         if (err != 0)
1109                 goto out_unregister_proto;
1110
1111         inet6_register_protosw(&dccp_v6_protosw);
1112
1113         err = register_pernet_subsys(&dccp_v6_ops);
1114         if (err != 0)
1115                 goto out_destroy_ctl_sock;
1116 out:
1117         return err;
1118
1119 out_destroy_ctl_sock:
1120         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1121         inet6_unregister_protosw(&dccp_v6_protosw);
1122 out_unregister_proto:
1123         proto_unregister(&dccp_v6_prot);
1124         goto out;
1125 }
1126
1127 static void __exit dccp_v6_exit(void)
1128 {
1129         unregister_pernet_subsys(&dccp_v6_ops);
1130         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1131         inet6_unregister_protosw(&dccp_v6_protosw);
1132         proto_unregister(&dccp_v6_prot);
1133 }
1134
1135 module_init(dccp_v6_init);
1136 module_exit(dccp_v6_exit);
1137
1138 /*
1139  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1140  * values directly, Also cover the case where the protocol is not specified,
1141  * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1142  */
1143 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6);
1144 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6);
1145 MODULE_LICENSE("GPL");
1146 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1147 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");