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