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 = (struct dccp_hdr *)(skb->data + offset);
  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         if (skb->len < offset + sizeof(*dh) ||
  82             skb->len < offset + __dccp_basic_hdr_len(dh)) {
  83                 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
  84                                   ICMP6_MIB_INERRORS);
  85                 return;
  86         }
  87
  88         sk = __inet6_lookup_established(net, &dccp_hashinfo,
  89                                         &hdr->daddr, dh->dccph_dport,
  90                                         &hdr->saddr, ntohs(dh->dccph_sport),
  91                                         inet6_iif(skb));
  92
  93         if (!sk) {
  94                 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
  95                                   ICMP6_MIB_INERRORS);
  96                 return;
  97         }
  98
  99         if (sk->sk_state == DCCP_TIME_WAIT) {
 100                 inet_twsk_put(inet_twsk(sk));
 101                 return;
 102         }
 103         seq = dccp_hdr_seq(dh);
 104         if (sk->sk_state == DCCP_NEW_SYN_RECV)
 105                 return dccp_req_err(sk, seq);
 106
 107         bh_lock_sock(sk);
 108         if (sock_owned_by_user(sk))
 109                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 110
 111         if (sk->sk_state == DCCP_CLOSED)
 112                 goto out;
 113
 114         dp = dccp_sk(sk);
 115         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 116             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 117                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 118                 goto out;
 119         }
 120
 121         np = inet6_sk(sk);
 122
 123         if (type == NDISC_REDIRECT) {
 124                 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
 125
 126                 if (dst)
 127                         dst->ops->redirect(dst, sk, skb);
 128                 goto out;
 129         }
 130
 131         if (type == ICMPV6_PKT_TOOBIG) {
 132                 struct dst_entry *dst = NULL;
 133
 134                 if (!ip6_sk_accept_pmtu(sk))
 135                         goto out;
 136
 137                 if (sock_owned_by_user(sk))
 138                         goto out;
 139                 if ((1 << sk->sk_state) & (DCCPF_LISTEN | DCCPF_CLOSED))
 140                         goto out;
 141
 142                 dst = inet6_csk_update_pmtu(sk, ntohl(info));
 143                 if (!dst)
 144                         goto out;
 145
 146                 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst))
 147                         dccp_sync_mss(sk, dst_mtu(dst));
 148                 goto out;
 149         }
 150
 151         icmpv6_err_convert(type, code, &err);
 152
 153         /* Might be for an request_sock */
 154         switch (sk->sk_state) {
 155         case DCCP_REQUESTING:
 156         case DCCP_RESPOND:  /* Cannot happen.
 157                                It can, it SYNs are crossed. --ANK */
 158                 if (!sock_owned_by_user(sk)) {
 159                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 160                         sk->sk_err = err;
 161                         /*
 162                          * Wake people up to see the error
 163                          * (see connect in sock.c)
 164                          */
 165                         sk->sk_error_report(sk);
 166                         dccp_done(sk);
 167                 } else
 168                         sk->sk_err_soft = err;
 169                 goto out;
 170         }
 171
 172         if (!sock_owned_by_user(sk) && np->recverr) {
 173                 sk->sk_err = err;
 174                 sk->sk_error_report(sk);
 175         } else
 176                 sk->sk_err_soft = err;
 177
 178 out:
 179         bh_unlock_sock(sk);
 180         sock_put(sk);
 181 }
 182
 183
 184 static int dccp_v6_send_response(const struct sock *sk, struct request_sock *req)
 185 {
 186         struct inet_request_sock *ireq = inet_rsk(req);
 187         struct ipv6_pinfo *np = inet6_sk(sk);
 188         struct sk_buff *skb;
 189         struct in6_addr *final_p, final;
 190         struct flowi6 fl6;
 191         int err = -1;
 192         struct dst_entry *dst;
 193
 194         memset(&fl6, 0, sizeof(fl6));
 195         fl6.flowi6_proto = IPPROTO_DCCP;
 196         fl6.daddr = ireq->ir_v6_rmt_addr;
 197         fl6.saddr = ireq->ir_v6_loc_addr;
 198         fl6.flowlabel = 0;
 199         fl6.flowi6_oif = ireq->ir_iif;
 200         fl6.fl6_dport = ireq->ir_rmt_port;
 201         fl6.fl6_sport = htons(ireq->ir_num);
 202         security_req_classify_flow(req, flowi6_to_flowi(&fl6));
 203
 204
 205         rcu_read_lock();
 206         final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
 207         rcu_read_unlock();
 208
 209         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 210         if (IS_ERR(dst)) {
 211                 err = PTR_ERR(dst);
 212                 dst = NULL;
 213                 goto done;
 214         }
 215
 216         skb = dccp_make_response(sk, dst, req);
 217         if (skb != NULL) {
 218                 struct dccp_hdr *dh = dccp_hdr(skb);
 219
 220                 dh->dccph_checksum = dccp_v6_csum_finish(skb,
 221                                                          &ireq->ir_v6_loc_addr,
 222                                                          &ireq->ir_v6_rmt_addr);
 223                 fl6.daddr = ireq->ir_v6_rmt_addr;
 224                 rcu_read_lock();
 225                 err = ip6_xmit(sk, skb, &fl6, rcu_dereference(np->opt),
 226                                np->tclass);
 227                 rcu_read_unlock();
 228                 err = net_xmit_eval(err);
 229         }
 230
 231 done:
 232         dst_release(dst);
 233         return err;
 234 }
 235
 236 static void dccp_v6_reqsk_destructor(struct request_sock *req)
 237 {
 238         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 239         kfree_skb(inet_rsk(req)->pktopts);
 240 }
 241
 242 static void dccp_v6_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 243 {
 244         const struct ipv6hdr *rxip6h;
 245         struct sk_buff *skb;
 246         struct flowi6 fl6;
 247         struct net *net = dev_net(skb_dst(rxskb)->dev);
 248         struct sock *ctl_sk = net->dccp.v6_ctl_sk;
 249         struct dst_entry *dst;
 250
 251         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 252                 return;
 253
 254         if (!ipv6_unicast_destination(rxskb))
 255                 return;
 256
 257         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 258         if (skb == NULL)
 259                 return;
 260
 261         rxip6h = ipv6_hdr(rxskb);
 262         dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr,
 263                                                             &rxip6h->daddr);
 264
 265         memset(&fl6, 0, sizeof(fl6));
 266         fl6.daddr = rxip6h->saddr;
 267         fl6.saddr = rxip6h->daddr;
 268
 269         fl6.flowi6_proto = IPPROTO_DCCP;
 270         fl6.flowi6_oif = inet6_iif(rxskb);
 271         fl6.fl6_dport = dccp_hdr(skb)->dccph_dport;
 272         fl6.fl6_sport = dccp_hdr(skb)->dccph_sport;
 273         security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6));
 274
 275         /* sk = NULL, but it is safe for now. RST socket required. */
 276         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 277         if (!IS_ERR(dst)) {
 278                 skb_dst_set(skb, dst);
 279                 ip6_xmit(ctl_sk, skb, &fl6, NULL, 0);
 280                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 281                 DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 282                 return;
 283         }
 284
 285         kfree_skb(skb);
 286 }
 287
 288 static struct request_sock_ops dccp6_request_sock_ops = {
 289         .family         = AF_INET6,
 290         .obj_size       = sizeof(struct dccp6_request_sock),
 291         .rtx_syn_ack    = dccp_v6_send_response,
 292         .send_ack       = dccp_reqsk_send_ack,
 293         .destructor     = dccp_v6_reqsk_destructor,
 294         .send_reset     = dccp_v6_ctl_send_reset,
 295         .syn_ack_timeout = dccp_syn_ack_timeout,
 296 };
 297
 298 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 299 {
 300         struct request_sock *req;
 301         struct dccp_request_sock *dreq;
 302         struct inet_request_sock *ireq;
 303         struct ipv6_pinfo *np = inet6_sk(sk);
 304         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 305         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 306
 307         if (skb->protocol == htons(ETH_P_IP))
 308                 return dccp_v4_conn_request(sk, skb);
 309
 310         if (!ipv6_unicast_destination(skb))
 311                 return 0;       /* discard, don't send a reset here */
 312
 313         if (dccp_bad_service_code(sk, service)) {
 314                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 315                 goto drop;
 316         }
 317         /*
 318          * There are no SYN attacks on IPv6, yet...
 319          */
 320         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 321         if (inet_csk_reqsk_queue_is_full(sk))
 322                 goto drop;
 323
 324         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 325                 goto drop;
 326
 327         req = inet_reqsk_alloc(&dccp6_request_sock_ops, sk, true);
 328         if (req == NULL)
 329                 goto drop;
 330
 331         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 332                 goto drop_and_free;
 333
 334         dreq = dccp_rsk(req);
 335         if (dccp_parse_options(sk, dreq, skb))
 336                 goto drop_and_free;
 337
 338         if (security_inet_conn_request(sk, skb, req))
 339                 goto drop_and_free;
 340
 341         ireq = inet_rsk(req);
 342         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 343         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 344         ireq->ireq_family = AF_INET6;
 345
 346         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
 347             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 348             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 349                 atomic_inc(&skb->users);
 350                 ireq->pktopts = skb;
 351         }
 352         ireq->ir_iif = sk->sk_bound_dev_if;
 353
 354         /* So that link locals have meaning */
 355         if (!sk->sk_bound_dev_if &&
 356             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 357                 ireq->ir_iif = inet6_iif(skb);
 358
 359         /*
 360          * Step 3: Process LISTEN state
 361          *
 362          *   Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 363          *
 364          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 365          */
 366         dreq->dreq_isr     = dcb->dccpd_seq;
 367         dreq->dreq_gsr     = dreq->dreq_isr;
 368         dreq->dreq_iss     = dccp_v6_init_sequence(skb);
 369         dreq->dreq_gss     = dreq->dreq_iss;
 370         dreq->dreq_service = service;
 371
 372         if (dccp_v6_send_response(sk, req))
 373                 goto drop_and_free;
 374
 375         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 376         return 0;
 377
 378 drop_and_free:
 379         reqsk_free(req);
 380 drop:
 381         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 382         return -1;
 383 }
 384
 385 static struct sock *dccp_v6_request_recv_sock(const struct sock *sk,
 386                                               struct sk_buff *skb,
 387                                               struct request_sock *req,
 388                                               struct dst_entry *dst,
 389                                               struct request_sock *req_unhash,
 390                                               bool *own_req)
 391 {
 392         struct inet_request_sock *ireq = inet_rsk(req);
 393         struct ipv6_pinfo *newnp;
 394         const struct ipv6_pinfo *np = inet6_sk(sk);
 395         struct ipv6_txoptions *opt;
 396         struct inet_sock *newinet;
 397         struct dccp6_sock *newdp6;
 398         struct sock *newsk;
 399
 400         if (skb->protocol == htons(ETH_P_IP)) {
 401                 /*
 402                  *      v6 mapped
 403                  */
 404                 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst,
 405                                                   req_unhash, own_req);
 406                 if (newsk == NULL)
 407                         return NULL;
 408
 409                 newdp6 = (struct dccp6_sock *)newsk;
 410                 newinet = inet_sk(newsk);
 411                 newinet->pinet6 = &newdp6->inet6;
 412                 newnp = inet6_sk(newsk);
 413
 414                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 415
 416                 newnp->saddr = newsk->sk_v6_rcv_saddr;
 417
 418                 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
 419                 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
 420                 newnp->pktoptions  = NULL;
 421                 newnp->opt         = NULL;
 422                 newnp->mcast_oif   = inet6_iif(skb);
 423                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
 424
 425                 /*
 426                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
 427                  * here, dccp_create_openreq_child now does this for us, see the comment in
 428                  * that function for the gory details. -acme
 429                  */
 430
 431                 /* It is tricky place. Until this moment IPv4 tcp
 432                    worked with IPv6 icsk.icsk_af_ops.
 433                    Sync it now.
 434                  */
 435                 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
 436
 437                 return newsk;
 438         }
 439
 440
 441         if (sk_acceptq_is_full(sk))
 442                 goto out_overflow;
 443
 444         if (!dst) {
 445                 struct flowi6 fl6;
 446
 447                 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_DCCP);
 448                 if (!dst)
 449                         goto out;
 450         }
 451
 452         newsk = dccp_create_openreq_child(sk, req, skb);
 453         if (newsk == NULL)
 454                 goto out_nonewsk;
 455
 456         /*
 457          * No need to charge this sock to the relevant IPv6 refcnt debug socks
 458          * count here, dccp_create_openreq_child now does this for us, see the
 459          * comment in that function for the gory details. -acme
 460          */
 461
 462         ip6_dst_store(newsk, dst, NULL, NULL);
 463         newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
 464                                                       NETIF_F_TSO);
 465         newdp6 = (struct dccp6_sock *)newsk;
 466         newinet = inet_sk(newsk);
 467         newinet->pinet6 = &newdp6->inet6;
 468         newnp = inet6_sk(newsk);
 469
 470         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 471
 472         newsk->sk_v6_daddr      = ireq->ir_v6_rmt_addr;
 473         newnp->saddr            = ireq->ir_v6_loc_addr;
 474         newsk->sk_v6_rcv_saddr  = ireq->ir_v6_loc_addr;
 475         newsk->sk_bound_dev_if  = ireq->ir_iif;
 476
 477         /* Now IPv6 options...
 478
 479            First: no IPv4 options.
 480          */
 481         newinet->inet_opt = NULL;
 482
 483         /* Clone RX bits */
 484         newnp->rxopt.all = np->rxopt.all;
 485
 486         newnp->pktoptions = NULL;
 487         newnp->opt        = NULL;
 488         newnp->mcast_oif  = inet6_iif(skb);
 489         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
 490
 491         /*
 492          * Clone native IPv6 options from listening socket (if any)
 493          *
 494          * Yes, keeping reference count would be much more clever, but we make
 495          * one more one thing there: reattach optmem to newsk.
 496          */
 497         opt = rcu_dereference(np->opt);
 498         if (opt) {
 499                 opt = ipv6_dup_options(newsk, opt);
 500                 RCU_INIT_POINTER(newnp->opt, opt);
 501         }
 502         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 503         if (opt)
 504                 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
 505                                                     opt->opt_flen;
 506
 507         dccp_sync_mss(newsk, dst_mtu(dst));
 508
 509         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
 510         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
 511
 512         if (__inet_inherit_port(sk, newsk) < 0) {
 513                 inet_csk_prepare_forced_close(newsk);
 514                 dccp_done(newsk);
 515                 goto out;
 516         }
 517         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 518         /* Clone pktoptions received with SYN, if we own the req */
 519         if (*own_req && ireq->pktopts) {
 520                 newnp->pktoptions = skb_clone(ireq->pktopts, GFP_ATOMIC);
 521                 consume_skb(ireq->pktopts);
 522                 ireq->pktopts = NULL;
 523                 if (newnp->pktoptions)
 524                         skb_set_owner_r(newnp->pktoptions, newsk);
 525         }
 526
 527         return newsk;
 528
 529 out_overflow:
 530         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 531 out_nonewsk:
 532         dst_release(dst);
 533 out:
 534         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 535         return NULL;
 536 }
 537
 538 /* The socket must have it's spinlock held when we get
 539  * here.
 540  *
 541  * We have a potential double-lock case here, so even when
 542  * doing backlog processing we use the BH locking scheme.
 543  * This is because we cannot sleep with the original spinlock
 544  * held.
 545  */
 546 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 547 {
 548         struct ipv6_pinfo *np = inet6_sk(sk);
 549         struct sk_buff *opt_skb = NULL;
 550
 551         /* Imagine: socket is IPv6. IPv4 packet arrives,
 552            goes to IPv4 receive handler and backlogged.
 553            From backlog it always goes here. Kerboom...
 554            Fortunately, dccp_rcv_established and rcv_established
 555            handle them correctly, but it is not case with
 556            dccp_v6_hnd_req and dccp_v6_ctl_send_reset().   --ANK
 557          */
 558
 559         if (skb->protocol == htons(ETH_P_IP))
 560                 return dccp_v4_do_rcv(sk, skb);
 561
 562         if (sk_filter(sk, skb))
 563                 goto discard;
 564
 565         /*
 566          * socket locking is here for SMP purposes as backlog rcv is currently
 567          * called with bh processing disabled.
 568          */
 569
 570         /* Do Stevens' IPV6_PKTOPTIONS.
 571
 572            Yes, guys, it is the only place in our code, where we
 573            may make it not affecting IPv4.
 574            The rest of code is protocol independent,
 575            and I do not like idea to uglify IPv4.
 576
 577            Actually, all the idea behind IPV6_PKTOPTIONS
 578            looks not very well thought. For now we latch
 579            options, received in the last packet, enqueued
 580            by tcp. Feel free to propose better solution.
 581                                                --ANK (980728)
 582          */
 583         if (np->rxopt.all)
 584         /*
 585          * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
 586          *        (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
 587          */
 588                 opt_skb = skb_clone(skb, GFP_ATOMIC);
 589
 590         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 591                 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
 592                         goto reset;
 593                 if (opt_skb) {
 594                         /* XXX This is where we would goto ipv6_pktoptions. */
 595                         __kfree_skb(opt_skb);
 596                 }
 597                 return 0;
 598         }
 599
 600         /*
 601          *  Step 3: Process LISTEN state
 602          *     If S.state == LISTEN,
 603          *       If P.type == Request or P contains a valid Init Cookie option,
 604          *            (* Must scan the packet's options to check for Init
 605          *               Cookies.  Only Init Cookies are processed here,
 606          *               however; other options are processed in Step 8.  This
 607          *               scan need only be performed if the endpoint uses Init
 608          *               Cookies *)
 609          *            (* Generate a new socket and switch to that socket *)
 610          *            Set S := new socket for this port pair
 611          *            S.state = RESPOND
 612          *            Choose S.ISS (initial seqno) or set from Init Cookies
 613          *            Initialize S.GAR := S.ISS
 614          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 615          *            Continue with S.state == RESPOND
 616          *            (* A Response packet will be generated in Step 11 *)
 617          *       Otherwise,
 618          *            Generate Reset(No Connection) unless P.type == Reset
 619          *            Drop packet and return
 620          *
 621          * NOTE: the check for the packet types is done in
 622          *       dccp_rcv_state_process
 623          */
 624
 625         if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
 626                 goto reset;
 627         if (opt_skb) {
 628                 /* XXX This is where we would goto ipv6_pktoptions. */
 629                 __kfree_skb(opt_skb);
 630         }
 631         return 0;
 632
 633 reset:
 634         dccp_v6_ctl_send_reset(sk, skb);
 635 discard:
 636         if (opt_skb != NULL)
 637                 __kfree_skb(opt_skb);
 638         kfree_skb(skb);
 639         return 0;
 640 }
 641
 642 static int dccp_v6_rcv(struct sk_buff *skb)
 643 {
 644         const struct dccp_hdr *dh;
 645         bool refcounted;
 646         struct sock *sk;
 647         int min_cov;
 648
 649         /* Step 1: Check header basics */
 650
 651         if (dccp_invalid_packet(skb))
 652                 goto discard_it;
 653
 654         /* Step 1: If header checksum is incorrect, drop packet and return. */
 655         if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
 656                                      &ipv6_hdr(skb)->daddr)) {
 657                 DCCP_WARN("dropped packet with invalid checksum\n");
 658                 goto discard_it;
 659         }
 660
 661         dh = dccp_hdr(skb);
 662
 663         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 664         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 665
 666         if (dccp_packet_without_ack(skb))
 667                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 668         else
 669                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 670
 671 lookup:
 672         sk = __inet6_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 673                                 dh->dccph_sport, dh->dccph_dport,
 674                                 inet6_iif(skb), &refcounted);
 675         if (!sk) {
 676                 dccp_pr_debug("failed to look up flow ID in table and "
 677                               "get corresponding socket\n");
 678                 goto no_dccp_socket;
 679         }
 680
 681         /*
 682          * Step 2:
 683          *      ... or S.state == TIMEWAIT,
 684          *              Generate Reset(No Connection) unless P.type == Reset
 685          *              Drop packet and return
 686          */
 687         if (sk->sk_state == DCCP_TIME_WAIT) {
 688                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 689                 inet_twsk_put(inet_twsk(sk));
 690                 goto no_dccp_socket;
 691         }
 692
 693         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 694                 struct request_sock *req = inet_reqsk(sk);
 695                 struct sock *nsk;
 696
 697                 sk = req->rsk_listener;
 698                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 699                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 700                         goto lookup;
 701                 }
 702                 sock_hold(sk);
 703                 refcounted = true;
 704                 nsk = dccp_check_req(sk, skb, req);
 705                 if (!nsk) {
 706                         reqsk_put(req);
 707                         goto discard_and_relse;
 708                 }
 709                 if (nsk == sk) {
 710                         reqsk_put(req);
 711                 } else if (dccp_child_process(sk, nsk, skb)) {
 712                         dccp_v6_ctl_send_reset(sk, skb);
 713                         goto discard_and_relse;
 714                 } else {
 715                         sock_put(sk);
 716                         return 0;
 717                 }
 718         }
 719         /*
 720          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 721          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 722          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 723          */
 724         min_cov = dccp_sk(sk)->dccps_pcrlen;
 725         if (dh->dccph_cscov  &&  (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 726                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 727                               dh->dccph_cscov, min_cov);
 728                 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
 729                 goto discard_and_relse;
 730         }
 731
 732         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
 733                 goto discard_and_relse;
 734
 735         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4) ? -1 : 0;
 736
 737 no_dccp_socket:
 738         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 739                 goto discard_it;
 740         /*
 741          * Step 2:
 742          *      If no socket ...
 743          *              Generate Reset(No Connection) unless P.type == Reset
 744          *              Drop packet and return
 745          */
 746         if (dh->dccph_type != DCCP_PKT_RESET) {
 747                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 748                                         DCCP_RESET_CODE_NO_CONNECTION;
 749                 dccp_v6_ctl_send_reset(sk, skb);
 750         }
 751
 752 discard_it:
 753         kfree_skb(skb);
 754         return 0;
 755
 756 discard_and_relse:
 757         if (refcounted)
 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, lockdep_sock_is_held(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 #ifdef CONFIG_COMPAT
 954         .compat_setsockopt = compat_ipv6_setsockopt,
 955         .compat_getsockopt = compat_ipv6_getsockopt,
 956 #endif
 957 };
 958
 959 /* NOTE: A lot of things set to zero explicitly by call to
 960  *       sk_alloc() so need not be done here.
 961  */
 962 static int dccp_v6_init_sock(struct sock *sk)
 963 {
 964         static __u8 dccp_v6_ctl_sock_initialized;
 965         int err = dccp_init_sock(sk, dccp_v6_ctl_sock_initialized);
 966
 967         if (err == 0) {
 968                 if (unlikely(!dccp_v6_ctl_sock_initialized))
 969                         dccp_v6_ctl_sock_initialized = 1;
 970                 inet_csk(sk)->icsk_af_ops = &dccp_ipv6_af_ops;
 971         }
 972
 973         return err;
 974 }
 975
 976 static void dccp_v6_destroy_sock(struct sock *sk)
 977 {
 978         dccp_destroy_sock(sk);
 979         inet6_destroy_sock(sk);
 980 }
 981
 982 static struct timewait_sock_ops dccp6_timewait_sock_ops = {
 983         .twsk_obj_size  = sizeof(struct dccp6_timewait_sock),
 984 };
 985
 986 static struct proto dccp_v6_prot = {
 987         .name              = "DCCPv6",
 988         .owner             = THIS_MODULE,
 989         .close             = dccp_close,
 990         .connect           = dccp_v6_connect,
 991         .disconnect        = dccp_disconnect,
 992         .ioctl             = dccp_ioctl,
 993         .init              = dccp_v6_init_sock,
 994         .setsockopt        = dccp_setsockopt,
 995         .getsockopt        = dccp_getsockopt,
 996         .sendmsg           = dccp_sendmsg,
 997         .recvmsg           = dccp_recvmsg,
 998         .backlog_rcv       = dccp_v6_do_rcv,
 999         .hash              = inet6_hash,
1000         .unhash            = inet_unhash,
1001         .accept            = inet_csk_accept,
1002         .get_port          = inet_csk_get_port,
1003         .shutdown          = dccp_shutdown,
1004         .destroy           = dccp_v6_destroy_sock,
1005         .orphan_count      = &dccp_orphan_count,
1006         .max_header        = MAX_DCCP_HEADER,
1007         .obj_size          = sizeof(struct dccp6_sock),
1008         .slab_flags        = SLAB_DESTROY_BY_RCU,
1009         .rsk_prot          = &dccp6_request_sock_ops,
1010         .twsk_prot         = &dccp6_timewait_sock_ops,
1011         .h.hashinfo        = &dccp_hashinfo,
1012 #ifdef CONFIG_COMPAT
1013         .compat_setsockopt = compat_dccp_setsockopt,
1014         .compat_getsockopt = compat_dccp_getsockopt,
1015 #endif
1016 };
1017
1018 static const struct inet6_protocol dccp_v6_protocol = {
1019         .handler        = dccp_v6_rcv,
1020         .err_handler    = dccp_v6_err,
1021         .flags          = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1022 };
1023
1024 static const struct proto_ops inet6_dccp_ops = {
1025         .family            = PF_INET6,
1026         .owner             = THIS_MODULE,
1027         .release           = inet6_release,
1028         .bind              = inet6_bind,
1029         .connect           = inet_stream_connect,
1030         .socketpair        = sock_no_socketpair,
1031         .accept            = inet_accept,
1032         .getname           = inet6_getname,
1033         .poll              = dccp_poll,
1034         .ioctl             = inet6_ioctl,
1035         .listen            = inet_dccp_listen,
1036         .shutdown          = inet_shutdown,
1037         .setsockopt        = sock_common_setsockopt,
1038         .getsockopt        = sock_common_getsockopt,
1039         .sendmsg           = inet_sendmsg,
1040         .recvmsg           = sock_common_recvmsg,
1041         .mmap              = sock_no_mmap,
1042         .sendpage          = sock_no_sendpage,
1043 #ifdef CONFIG_COMPAT
1044         .compat_setsockopt = compat_sock_common_setsockopt,
1045         .compat_getsockopt = compat_sock_common_getsockopt,
1046 #endif
1047 };
1048
1049 static struct inet_protosw dccp_v6_protosw = {
1050         .type           = SOCK_DCCP,
1051         .protocol       = IPPROTO_DCCP,
1052         .prot           = &dccp_v6_prot,
1053         .ops            = &inet6_dccp_ops,
1054         .flags          = INET_PROTOSW_ICSK,
1055 };
1056
1057 static int __net_init dccp_v6_init_net(struct net *net)
1058 {
1059         if (dccp_hashinfo.bhash == NULL)
1060                 return -ESOCKTNOSUPPORT;
1061
1062         return inet_ctl_sock_create(&net->dccp.v6_ctl_sk, PF_INET6,
1063                                     SOCK_DCCP, IPPROTO_DCCP, net);
1064 }
1065
1066 static void __net_exit dccp_v6_exit_net(struct net *net)
1067 {
1068         inet_ctl_sock_destroy(net->dccp.v6_ctl_sk);
1069 }
1070
1071 static struct pernet_operations dccp_v6_ops = {
1072         .init   = dccp_v6_init_net,
1073         .exit   = dccp_v6_exit_net,
1074 };
1075
1076 static int __init dccp_v6_init(void)
1077 {
1078         int err = proto_register(&dccp_v6_prot, 1);
1079
1080         if (err != 0)
1081                 goto out;
1082
1083         err = inet6_add_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1084         if (err != 0)
1085                 goto out_unregister_proto;
1086
1087         inet6_register_protosw(&dccp_v6_protosw);
1088
1089         err = register_pernet_subsys(&dccp_v6_ops);
1090         if (err != 0)
1091                 goto out_destroy_ctl_sock;
1092 out:
1093         return err;
1094
1095 out_destroy_ctl_sock:
1096         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1097         inet6_unregister_protosw(&dccp_v6_protosw);
1098 out_unregister_proto:
1099         proto_unregister(&dccp_v6_prot);
1100         goto out;
1101 }
1102
1103 static void __exit dccp_v6_exit(void)
1104 {
1105         unregister_pernet_subsys(&dccp_v6_ops);
1106         inet6_del_protocol(&dccp_v6_protocol, IPPROTO_DCCP);
1107         inet6_unregister_protosw(&dccp_v6_protosw);
1108         proto_unregister(&dccp_v6_prot);
1109 }
1110
1111 module_init(dccp_v6_init);
1112 module_exit(dccp_v6_exit);
1113
1114 /*
1115  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1116  * values directly, Also cover the case where the protocol is not specified,
1117  * i.e. net-pf-PF_INET6-proto-0-type-SOCK_DCCP
1118  */
1119 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 33, 6);
1120 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 0, 6);
1121 MODULE_LICENSE("GPL");
1122 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1123 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");