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 <[email protected]>
   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_BH(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_BH(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_BH(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_BH(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_BH(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(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         final_p = fl6_update_dst(&fl6, np->opt, &final);
 206
 207         dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 208         if (IS_ERR(dst)) {
 209                 err = PTR_ERR(dst);
 210                 dst = NULL;
 211                 goto done;
 212         }
 213
 214         skb = dccp_make_response(sk, dst, req);
 215         if (skb != NULL) {
 216                 struct dccp_hdr *dh = dccp_hdr(skb);
 217
 218                 dh->dccph_checksum = dccp_v6_csum_finish(skb,
 219                                                          &ireq->ir_v6_loc_addr,
 220                                                          &ireq->ir_v6_rmt_addr);
 221                 fl6.daddr = ireq->ir_v6_rmt_addr;
 222                 err = ip6_xmit(sk, skb, &fl6, np->opt, np->tclass);
 223                 err = net_xmit_eval(err);
 224         }
 225
 226 done:
 227         dst_release(dst);
 228         return err;
 229 }
 230
 231 static void dccp_v6_reqsk_destructor(struct request_sock *req)
 232 {
 233         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 234         kfree_skb(inet_rsk(req)->pktopts);
 235 }
 236
 237 static void dccp_v6_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 238 {
 239         const struct ipv6hdr *rxip6h;
 240         struct sk_buff *skb;
 241         struct flowi6 fl6;
 242         struct net *net = dev_net(skb_dst(rxskb)->dev);
 243         struct sock *ctl_sk = net->dccp.v6_ctl_sk;
 244         struct dst_entry *dst;
 245
 246         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 247                 return;
 248
 249         if (!ipv6_unicast_destination(rxskb))
 250                 return;
 251
 252         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 253         if (skb == NULL)
 254                 return;
 255
 256         rxip6h = ipv6_hdr(rxskb);
 257         dccp_hdr(skb)->dccph_checksum = dccp_v6_csum_finish(skb, &rxip6h->saddr,
 258                                                             &rxip6h->daddr);
 259
 260         memset(&fl6, 0, sizeof(fl6));
 261         fl6.daddr = rxip6h->saddr;
 262         fl6.saddr = rxip6h->daddr;
 263
 264         fl6.flowi6_proto = IPPROTO_DCCP;
 265         fl6.flowi6_oif = inet6_iif(rxskb);
 266         fl6.fl6_dport = dccp_hdr(skb)->dccph_dport;
 267         fl6.fl6_sport = dccp_hdr(skb)->dccph_sport;
 268         security_skb_classify_flow(rxskb, flowi6_to_flowi(&fl6));
 269
 270         /* sk = NULL, but it is safe for now. RST socket required. */
 271         dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL);
 272         if (!IS_ERR(dst)) {
 273                 skb_dst_set(skb, dst);
 274                 ip6_xmit(ctl_sk, skb, &fl6, NULL, 0);
 275                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 276                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 277                 return;
 278         }
 279
 280         kfree_skb(skb);
 281 }
 282
 283 static struct request_sock_ops dccp6_request_sock_ops = {
 284         .family         = AF_INET6,
 285         .obj_size       = sizeof(struct dccp6_request_sock),
 286         .rtx_syn_ack    = dccp_v6_send_response,
 287         .send_ack       = dccp_reqsk_send_ack,
 288         .destructor     = dccp_v6_reqsk_destructor,
 289         .send_reset     = dccp_v6_ctl_send_reset,
 290         .syn_ack_timeout = dccp_syn_ack_timeout,
 291 };
 292
 293 static struct sock *dccp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
 294 {
 295         const struct dccp_hdr *dh = dccp_hdr(skb);
 296         const struct ipv6hdr *iph = ipv6_hdr(skb);
 297         struct request_sock *req;
 298         struct sock *nsk;
 299
 300         req = inet6_csk_search_req(sk, dh->dccph_sport, &iph->saddr,
 301                                    &iph->daddr, inet6_iif(skb));
 302         if (req) {
 303                 nsk = dccp_check_req(sk, skb, req);
 304                 if (!nsk)
 305                         reqsk_put(req);
 306                 return nsk;
 307         }
 308         nsk = __inet6_lookup_established(sock_net(sk), &dccp_hashinfo,
 309                                          &iph->saddr, dh->dccph_sport,
 310                                          &iph->daddr, ntohs(dh->dccph_dport),
 311                                          inet6_iif(skb));
 312         if (nsk != NULL) {
 313                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 314                         bh_lock_sock(nsk);
 315                         return nsk;
 316                 }
 317                 inet_twsk_put(inet_twsk(nsk));
 318                 return NULL;
 319         }
 320
 321         return sk;
 322 }
 323
 324 static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 325 {
 326         struct request_sock *req;
 327         struct dccp_request_sock *dreq;
 328         struct inet_request_sock *ireq;
 329         struct ipv6_pinfo *np = inet6_sk(sk);
 330         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 331         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 332
 333         if (skb->protocol == htons(ETH_P_IP))
 334                 return dccp_v4_conn_request(sk, skb);
 335
 336         if (!ipv6_unicast_destination(skb))
 337                 return 0;       /* discard, don't send a reset here */
 338
 339         if (dccp_bad_service_code(sk, service)) {
 340                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 341                 goto drop;
 342         }
 343         /*
 344          * There are no SYN attacks on IPv6, yet...
 345          */
 346         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 347         if (inet_csk_reqsk_queue_is_full(sk))
 348                 goto drop;
 349
 350         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 351                 goto drop;
 352
 353         req = inet_reqsk_alloc(&dccp6_request_sock_ops, sk);
 354         if (req == NULL)
 355                 goto drop;
 356
 357         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 358                 goto drop_and_free;
 359
 360         dreq = dccp_rsk(req);
 361         if (dccp_parse_options(sk, dreq, skb))
 362                 goto drop_and_free;
 363
 364         if (security_inet_conn_request(sk, skb, req))
 365                 goto drop_and_free;
 366
 367         ireq = inet_rsk(req);
 368         ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
 369         ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
 370         ireq->ireq_family = AF_INET6;
 371
 372         if (ipv6_opt_accepted(sk, skb, IP6CB(skb)) ||
 373             np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
 374             np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
 375                 atomic_inc(&skb->users);
 376                 ireq->pktopts = skb;
 377         }
 378         ireq->ir_iif = sk->sk_bound_dev_if;
 379
 380         /* So that link locals have meaning */
 381         if (!sk->sk_bound_dev_if &&
 382             ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
 383                 ireq->ir_iif = inet6_iif(skb);
 384
 385         /*
 386          * Step 3: Process LISTEN state
 387          *
 388          *   Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 389          *
 390          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 391          */
 392         dreq->dreq_isr     = dcb->dccpd_seq;
 393         dreq->dreq_gsr     = dreq->dreq_isr;
 394         dreq->dreq_iss     = dccp_v6_init_sequence(skb);
 395         dreq->dreq_gss     = dreq->dreq_iss;
 396         dreq->dreq_service = service;
 397
 398         if (dccp_v6_send_response(sk, req))
 399                 goto drop_and_free;
 400
 401         inet6_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 402         return 0;
 403
 404 drop_and_free:
 405         reqsk_free(req);
 406 drop:
 407         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 408         return -1;
 409 }
 410
 411 static struct sock *dccp_v6_request_recv_sock(struct sock *sk,
 412                                               struct sk_buff *skb,
 413                                               struct request_sock *req,
 414                                               struct dst_entry *dst)
 415 {
 416         struct inet_request_sock *ireq = inet_rsk(req);
 417         struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
 418         struct inet_sock *newinet;
 419         struct dccp6_sock *newdp6;
 420         struct sock *newsk;
 421
 422         if (skb->protocol == htons(ETH_P_IP)) {
 423                 /*
 424                  *      v6 mapped
 425                  */
 426                 newsk = dccp_v4_request_recv_sock(sk, skb, req, dst);
 427                 if (newsk == NULL)
 428                         return NULL;
 429
 430                 newdp6 = (struct dccp6_sock *)newsk;
 431                 newinet = inet_sk(newsk);
 432                 newinet->pinet6 = &newdp6->inet6;
 433                 newnp = inet6_sk(newsk);
 434
 435                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 436
 437                 newnp->saddr = newsk->sk_v6_rcv_saddr;
 438
 439                 inet_csk(newsk)->icsk_af_ops = &dccp_ipv6_mapped;
 440                 newsk->sk_backlog_rcv = dccp_v4_do_rcv;
 441                 newnp->pktoptions  = NULL;
 442                 newnp->opt         = NULL;
 443                 newnp->mcast_oif   = inet6_iif(skb);
 444                 newnp->mcast_hops  = ipv6_hdr(skb)->hop_limit;
 445
 446                 /*
 447                  * No need to charge this sock to the relevant IPv6 refcnt debug socks count
 448                  * here, dccp_create_openreq_child now does this for us, see the comment in
 449                  * that function for the gory details. -acme
 450                  */
 451
 452                 /* It is tricky place. Until this moment IPv4 tcp
 453                    worked with IPv6 icsk.icsk_af_ops.
 454                    Sync it now.
 455                  */
 456                 dccp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
 457
 458                 return newsk;
 459         }
 460
 461
 462         if (sk_acceptq_is_full(sk))
 463                 goto out_overflow;
 464
 465         if (dst == NULL) {
 466                 struct in6_addr *final_p, final;
 467                 struct flowi6 fl6;
 468
 469                 memset(&fl6, 0, sizeof(fl6));
 470                 fl6.flowi6_proto = IPPROTO_DCCP;
 471                 fl6.daddr = ireq->ir_v6_rmt_addr;
 472                 final_p = fl6_update_dst(&fl6, np->opt, &final);
 473                 fl6.saddr = ireq->ir_v6_loc_addr;
 474                 fl6.flowi6_oif = sk->sk_bound_dev_if;
 475                 fl6.fl6_dport = ireq->ir_rmt_port;
 476                 fl6.fl6_sport = htons(ireq->ir_num);
 477                 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 478
 479                 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
 480                 if (IS_ERR(dst))
 481                         goto out;
 482         }
 483
 484         newsk = dccp_create_openreq_child(sk, req, skb);
 485         if (newsk == NULL)
 486                 goto out_nonewsk;
 487
 488         /*
 489          * No need to charge this sock to the relevant IPv6 refcnt debug socks
 490          * count here, dccp_create_openreq_child now does this for us, see the
 491          * comment in that function for the gory details. -acme
 492          */
 493
 494         __ip6_dst_store(newsk, dst, NULL, NULL);
 495         newsk->sk_route_caps = dst->dev->features & ~(NETIF_F_IP_CSUM |
 496                                                       NETIF_F_TSO);
 497         newdp6 = (struct dccp6_sock *)newsk;
 498         newinet = inet_sk(newsk);
 499         newinet->pinet6 = &newdp6->inet6;
 500         newnp = inet6_sk(newsk);
 501
 502         memcpy(newnp, np, sizeof(struct ipv6_pinfo));
 503
 504         newsk->sk_v6_daddr      = ireq->ir_v6_rmt_addr;
 505         newnp->saddr            = ireq->ir_v6_loc_addr;
 506         newsk->sk_v6_rcv_saddr  = ireq->ir_v6_loc_addr;
 507         newsk->sk_bound_dev_if  = ireq->ir_iif;
 508
 509         /* Now IPv6 options...
 510
 511            First: no IPv4 options.
 512          */
 513         newinet->inet_opt = NULL;
 514
 515         /* Clone RX bits */
 516         newnp->rxopt.all = np->rxopt.all;
 517
 518         /* Clone pktoptions received with SYN */
 519         newnp->pktoptions = NULL;
 520         if (ireq->pktopts != NULL) {
 521                 newnp->pktoptions = skb_clone(ireq->pktopts, GFP_ATOMIC);
 522                 consume_skb(ireq->pktopts);
 523                 ireq->pktopts = NULL;
 524                 if (newnp->pktoptions)
 525                         skb_set_owner_r(newnp->pktoptions, newsk);
 526         }
 527         newnp->opt        = NULL;
 528         newnp->mcast_oif  = inet6_iif(skb);
 529         newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
 530
 531         /*
 532          * Clone native IPv6 options from listening socket (if any)
 533          *
 534          * Yes, keeping reference count would be much more clever, but we make
 535          * one more one thing there: reattach optmem to newsk.
 536          */
 537         if (np->opt != NULL)
 538                 newnp->opt = ipv6_dup_options(newsk, np->opt);
 539
 540         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 541         if (newnp->opt != NULL)
 542                 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
 543                                                      newnp->opt->opt_flen);
 544
 545         dccp_sync_mss(newsk, dst_mtu(dst));
 546
 547         newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
 548         newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
 549
 550         if (__inet_inherit_port(sk, newsk) < 0) {
 551                 inet_csk_prepare_forced_close(newsk);
 552                 dccp_done(newsk);
 553                 goto out;
 554         }
 555         __inet_hash(newsk, NULL);
 556
 557         return newsk;
 558
 559 out_overflow:
 560         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 561 out_nonewsk:
 562         dst_release(dst);
 563 out:
 564         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 565         return NULL;
 566 }
 567
 568 /* The socket must have it's spinlock held when we get
 569  * here.
 570  *
 571  * We have a potential double-lock case here, so even when
 572  * doing backlog processing we use the BH locking scheme.
 573  * This is because we cannot sleep with the original spinlock
 574  * held.
 575  */
 576 static int dccp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
 577 {
 578         struct ipv6_pinfo *np = inet6_sk(sk);
 579         struct sk_buff *opt_skb = NULL;
 580
 581         /* Imagine: socket is IPv6. IPv4 packet arrives,
 582            goes to IPv4 receive handler and backlogged.
 583            From backlog it always goes here. Kerboom...
 584            Fortunately, dccp_rcv_established and rcv_established
 585            handle them correctly, but it is not case with
 586            dccp_v6_hnd_req and dccp_v6_ctl_send_reset().   --ANK
 587          */
 588
 589         if (skb->protocol == htons(ETH_P_IP))
 590                 return dccp_v4_do_rcv(sk, skb);
 591
 592         if (sk_filter(sk, skb))
 593                 goto discard;
 594
 595         /*
 596          * socket locking is here for SMP purposes as backlog rcv is currently
 597          * called with bh processing disabled.
 598          */
 599
 600         /* Do Stevens' IPV6_PKTOPTIONS.
 601
 602            Yes, guys, it is the only place in our code, where we
 603            may make it not affecting IPv4.
 604            The rest of code is protocol independent,
 605            and I do not like idea to uglify IPv4.
 606
 607            Actually, all the idea behind IPV6_PKTOPTIONS
 608            looks not very well thought. For now we latch
 609            options, received in the last packet, enqueued
 610            by tcp. Feel free to propose better solution.
 611                                                --ANK (980728)
 612          */
 613         if (np->rxopt.all)
 614         /*
 615          * FIXME: Add handling of IPV6_PKTOPTIONS skb. See the comments below
 616          *        (wrt ipv6_pktopions) and net/ipv6/tcp_ipv6.c for an example.
 617          */
 618                 opt_skb = skb_clone(skb, GFP_ATOMIC);
 619
 620         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 621                 if (dccp_rcv_established(sk, skb, dccp_hdr(skb), skb->len))
 622                         goto reset;
 623                 if (opt_skb) {
 624                         /* XXX This is where we would goto ipv6_pktoptions. */
 625                         __kfree_skb(opt_skb);
 626                 }
 627                 return 0;
 628         }
 629
 630         /*
 631          *  Step 3: Process LISTEN state
 632          *     If S.state == LISTEN,
 633          *       If P.type == Request or P contains a valid Init Cookie option,
 634          *            (* Must scan the packet's options to check for Init
 635          *               Cookies.  Only Init Cookies are processed here,
 636          *               however; other options are processed in Step 8.  This
 637          *               scan need only be performed if the endpoint uses Init
 638          *               Cookies *)
 639          *            (* Generate a new socket and switch to that socket *)
 640          *            Set S := new socket for this port pair
 641          *            S.state = RESPOND
 642          *            Choose S.ISS (initial seqno) or set from Init Cookies
 643          *            Initialize S.GAR := S.ISS
 644          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 645          *            Continue with S.state == RESPOND
 646          *            (* A Response packet will be generated in Step 11 *)
 647          *       Otherwise,
 648          *            Generate Reset(No Connection) unless P.type == Reset
 649          *            Drop packet and return
 650          *
 651          * NOTE: the check for the packet types is done in
 652          *       dccp_rcv_state_process
 653          */
 654         if (sk->sk_state == DCCP_LISTEN) {
 655                 struct sock *nsk = dccp_v6_hnd_req(sk, skb);
 656
 657                 if (nsk == NULL)
 658                         goto discard;
 659                 /*
 660                  * Queue it on the new socket if the new socket is active,
 661                  * otherwise we just shortcircuit this and continue with
 662                  * the new socket..
 663                  */
 664                 if (nsk != sk) {
 665                         if (dccp_child_process(sk, nsk, skb))
 666                                 goto reset;
 667                         if (opt_skb != NULL)
 668                                 __kfree_skb(opt_skb);
 669                         return 0;
 670                 }
 671         }
 672
 673         if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
 674                 goto reset;
 675         if (opt_skb) {
 676                 /* XXX This is where we would goto ipv6_pktoptions. */
 677                 __kfree_skb(opt_skb);
 678         }
 679         return 0;
 680
 681 reset:
 682         dccp_v6_ctl_send_reset(sk, skb);
 683 discard:
 684         if (opt_skb != NULL)
 685                 __kfree_skb(opt_skb);
 686         kfree_skb(skb);
 687         return 0;
 688 }
 689
 690 static int dccp_v6_rcv(struct sk_buff *skb)
 691 {
 692         const struct dccp_hdr *dh;
 693         struct sock *sk;
 694         int min_cov;
 695
 696         /* Step 1: Check header basics */
 697
 698         if (dccp_invalid_packet(skb))
 699                 goto discard_it;
 700
 701         /* Step 1: If header checksum is incorrect, drop packet and return. */
 702         if (dccp_v6_csum_finish(skb, &ipv6_hdr(skb)->saddr,
 703                                      &ipv6_hdr(skb)->daddr)) {
 704                 DCCP_WARN("dropped packet with invalid checksum\n");
 705                 goto discard_it;
 706         }
 707
 708         dh = dccp_hdr(skb);
 709
 710         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 711         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 712
 713         if (dccp_packet_without_ack(skb))
 714                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 715         else
 716                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 717
 718         /* Step 2:
 719          *      Look up flow ID in table and get corresponding socket */
 720         sk = __inet6_lookup_skb(&dccp_hashinfo, skb,
 721                                 dh->dccph_sport, dh->dccph_dport,
 722                                 inet6_iif(skb));
 723         /*
 724          * Step 2:
 725          *      If no socket ...
 726          */
 727         if (sk == NULL) {
 728                 dccp_pr_debug("failed to look up flow ID in table and "
 729                               "get corresponding socket\n");
 730                 goto no_dccp_socket;
 731         }
 732
 733         /*
 734          * Step 2:
 735          *      ... or S.state == TIMEWAIT,
 736          *              Generate Reset(No Connection) unless P.type == Reset
 737          *              Drop packet and return
 738          */
 739         if (sk->sk_state == DCCP_TIME_WAIT) {
 740                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 741                 inet_twsk_put(inet_twsk(sk));
 742                 goto no_dccp_socket;
 743         }
 744
 745         /*
 746          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 747          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 748          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 749          */
 750         min_cov = dccp_sk(sk)->dccps_pcrlen;
 751         if (dh->dccph_cscov  &&  (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 752                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 753                               dh->dccph_cscov, min_cov);
 754                 /* FIXME: send Data Dropped option (see also dccp_v4_rcv) */
 755                 goto discard_and_relse;
 756         }
 757
 758         if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
 759                 goto discard_and_relse;
 760
 761         return sk_receive_skb(sk, skb, 1) ? -1 : 0;
 762
 763 no_dccp_socket:
 764         if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 765                 goto discard_it;
 766         /*
 767          * Step 2:
 768          *      If no socket ...
 769          *              Generate Reset(No Connection) unless P.type == Reset
 770          *              Drop packet and return
 771          */
 772         if (dh->dccph_type != DCCP_PKT_RESET) {
 773                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 774                                         DCCP_RESET_CODE_NO_CONNECTION;
 775                 dccp_v6_ctl_send_reset(sk, skb);
 776         }
 777
 778 discard_it:
 779         kfree_skb(skb);
 780         return 0;
 781
 782 discard_and_relse:
 783         sock_put(sk);
 784         goto discard_it;
 785 }
 786
 787 static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 788                            int addr_len)
 789 {
 790         struct sockaddr_in6 *usin = (struct sockaddr_in6 *)uaddr;
 791         struct inet_connection_sock *icsk = inet_csk(sk);
 792         struct inet_sock *inet = inet_sk(sk);
 793         struct ipv6_pinfo *np = inet6_sk(sk);
 794         struct dccp_sock *dp = dccp_sk(sk);
 795         struct in6_addr *saddr = NULL, *final_p, final;
 796         struct flowi6 fl6;
 797         struct dst_entry *dst;
 798         int addr_type;
 799         int err;
 800
 801         dp->dccps_role = DCCP_ROLE_CLIENT;
 802
 803         if (addr_len < SIN6_LEN_RFC2133)
 804                 return -EINVAL;
 805
 806         if (usin->sin6_family != AF_INET6)
 807                 return -EAFNOSUPPORT;
 808
 809         memset(&fl6, 0, sizeof(fl6));
 810
 811         if (np->sndflow) {
 812                 fl6.flowlabel = usin->sin6_flowinfo & IPV6_FLOWINFO_MASK;
 813                 IP6_ECN_flow_init(fl6.flowlabel);
 814                 if (fl6.flowlabel & IPV6_FLOWLABEL_MASK) {
 815                         struct ip6_flowlabel *flowlabel;
 816                         flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 817                         if (flowlabel == NULL)
 818                                 return -EINVAL;
 819                         fl6_sock_release(flowlabel);
 820                 }
 821         }
 822         /*
 823          * connect() to INADDR_ANY means loopback (BSD'ism).
 824          */
 825         if (ipv6_addr_any(&usin->sin6_addr))
 826                 usin->sin6_addr.s6_addr[15] = 1;
 827
 828         addr_type = ipv6_addr_type(&usin->sin6_addr);
 829
 830         if (addr_type & IPV6_ADDR_MULTICAST)
 831                 return -ENETUNREACH;
 832
 833         if (addr_type & IPV6_ADDR_LINKLOCAL) {
 834                 if (addr_len >= sizeof(struct sockaddr_in6) &&
 835                     usin->sin6_scope_id) {
 836                         /* If interface is set while binding, indices
 837                          * must coincide.
 838                          */
 839                         if (sk->sk_bound_dev_if &&
 840                             sk->sk_bound_dev_if != usin->sin6_scope_id)
 841                                 return -EINVAL;
 842
 843                         sk->sk_bound_dev_if = usin->sin6_scope_id;
 844                 }
 845
 846                 /* Connect to link-local address requires an interface */
 847                 if (!sk->sk_bound_dev_if)
 848                         return -EINVAL;
 849         }
 850
 851         sk->sk_v6_daddr = usin->sin6_addr;
 852         np->flow_label = fl6.flowlabel;
 853
 854         /*
 855          * DCCP over IPv4
 856          */
 857         if (addr_type == IPV6_ADDR_MAPPED) {
 858                 u32 exthdrlen = icsk->icsk_ext_hdr_len;
 859                 struct sockaddr_in sin;
 860
 861                 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
 862
 863                 if (__ipv6_only_sock(sk))
 864                         return -ENETUNREACH;
 865
 866                 sin.sin_family = AF_INET;
 867                 sin.sin_port = usin->sin6_port;
 868                 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
 869
 870                 icsk->icsk_af_ops = &dccp_ipv6_mapped;
 871                 sk->sk_backlog_rcv = dccp_v4_do_rcv;
 872
 873                 err = dccp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
 874                 if (err) {
 875                         icsk->icsk_ext_hdr_len = exthdrlen;
 876                         icsk->icsk_af_ops = &dccp_ipv6_af_ops;
 877                         sk->sk_backlog_rcv = dccp_v6_do_rcv;
 878                         goto failure;
 879                 }
 880                 np->saddr = sk->sk_v6_rcv_saddr;
 881                 return err;
 882         }
 883
 884         if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
 885                 saddr = &sk->sk_v6_rcv_saddr;
 886
 887         fl6.flowi6_proto = IPPROTO_DCCP;
 888         fl6.daddr = sk->sk_v6_daddr;
 889         fl6.saddr = saddr ? *saddr : np->saddr;
 890         fl6.flowi6_oif = sk->sk_bound_dev_if;
 891         fl6.fl6_dport = usin->sin6_port;
 892         fl6.fl6_sport = inet->inet_sport;
 893         security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
 894
 895         final_p = fl6_update_dst(&fl6, np->opt, &final);
 896
 897         dst = ip6_dst_lookup_flow(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 (np->opt != NULL)
 916                 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
 917                                           np->opt->opt_nflen);
 918
 919         inet->inet_dport = usin->sin6_port;
 920
 921         dccp_set_state(sk, DCCP_REQUESTING);
 922         err = inet6_hash_connect(&dccp_death_row, sk);
 923         if (err)
 924                 goto late_failure;
 925
 926         dp->dccps_iss = secure_dccpv6_sequence_number(np->saddr.s6_addr32,
 927                                                       sk->sk_v6_daddr.s6_addr32,
 928                                                       inet->inet_sport,
 929                                                       inet->inet_dport);
 930         err = dccp_connect(sk);
 931         if (err)
 932                 goto late_failure;
 933
 934         return 0;
 935
 936 late_failure:
 937         dccp_set_state(sk, DCCP_CLOSED);
 938         __sk_dst_reset(sk);
 939 failure:
 940         inet->inet_dport = 0;
 941         sk->sk_route_caps = 0;
 942         return err;
 943 }
 944
 945 static const struct inet_connection_sock_af_ops dccp_ipv6_af_ops = {
 946         .queue_xmit        = inet6_csk_xmit,
 947         .send_check        = dccp_v6_send_check,
 948         .rebuild_header    = inet6_sk_rebuild_header,
 949         .conn_request      = dccp_v6_conn_request,
 950         .syn_recv_sock     = dccp_v6_request_recv_sock,
 951         .net_header_len    = sizeof(struct ipv6hdr),
 952         .setsockopt        = ipv6_setsockopt,
 953         .getsockopt        = ipv6_getsockopt,
 954         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 955         .sockaddr_len      = sizeof(struct sockaddr_in6),
 956         .bind_conflict     = inet6_csk_bind_conflict,
 957 #ifdef CONFIG_COMPAT
 958         .compat_setsockopt = compat_ipv6_setsockopt,
 959         .compat_getsockopt = compat_ipv6_getsockopt,
 960 #endif
 961 };
 962
 963 /*
 964  *      DCCP over IPv4 via INET6 API
 965  */
 966 static const struct inet_connection_sock_af_ops dccp_ipv6_mapped = {
 967         .queue_xmit        = ip_queue_xmit,
 968         .send_check        = dccp_v4_send_check,
 969         .rebuild_header    = inet_sk_rebuild_header,
 970         .conn_request      = dccp_v6_conn_request,
 971         .syn_recv_sock     = dccp_v6_request_recv_sock,
 972         .net_header_len    = sizeof(struct iphdr),
 973         .setsockopt        = ipv6_setsockopt,
 974         .getsockopt        = ipv6_getsockopt,
 975         .addr2sockaddr     = inet6_csk_addr2sockaddr,
 976         .sockaddr_len      = sizeof(struct sockaddr_in6),
 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              = inet_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 <[email protected]>");
1147 MODULE_DESCRIPTION("DCCPv6 - Datagram Congestion Controlled Protocol");