net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/dccp.h>
  14 #include <linux/icmp.h>
  15 #include <linux/slab.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/random.h>
  19
  20 #include <net/icmp.h>
  21 #include <net/inet_common.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet_sock.h>
  24 #include <net/protocol.h>
  25 #include <net/sock.h>
  26 #include <net/timewait_sock.h>
  27 #include <net/tcp_states.h>
  28 #include <net/xfrm.h>
  29
  30 #include "ackvec.h"
  31 #include "ccid.h"
  32 #include "dccp.h"
  33 #include "feat.h"
  34
  35 /*
  36  * The per-net dccp.v4_ctl_sk socket is used for responding to
  37  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  38  * for this socket at the initialization time.
  39  */
  40
  41 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  42 {
  43         struct inet_sock *inet = inet_sk(sk);
  44         struct dccp_sock *dp = dccp_sk(sk);
  45         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  46         struct rtable *rt;
  47         __be32 daddr, nexthop;
  48         int tmp;
  49         int err;
  50
  51         dp->dccps_role = DCCP_ROLE_CLIENT;
  52
  53         if (addr_len < sizeof(struct sockaddr_in))
  54                 return -EINVAL;
  55
  56         if (usin->sin_family != AF_INET)
  57                 return -EAFNOSUPPORT;
  58
  59         nexthop = daddr = usin->sin_addr.s_addr;
  60         if (inet->opt != NULL && inet->opt->srr) {
  61                 if (daddr == 0)
  62                         return -EINVAL;
  63                 nexthop = inet->opt->faddr;
  64         }
  65
  66         tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
  67                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  68                                IPPROTO_DCCP,
  69                                inet->inet_sport, usin->sin_port, sk, 1);
  70         if (tmp < 0)
  71                 return tmp;
  72
  73         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  74                 ip_rt_put(rt);
  75                 return -ENETUNREACH;
  76         }
  77
  78         if (inet->opt == NULL || !inet->opt->srr)
  79                 daddr = rt->rt_dst;
  80
  81         if (inet->inet_saddr == 0)
  82                 inet->inet_saddr = rt->rt_src;
  83         inet->inet_rcv_saddr = inet->inet_saddr;
  84
  85         inet->inet_dport = usin->sin_port;
  86         inet->inet_daddr = daddr;
  87
  88         inet_csk(sk)->icsk_ext_hdr_len = 0;
  89         if (inet->opt != NULL)
  90                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
  91         /*
  92          * Socket identity is still unknown (sport may be zero).
  93          * However we set state to DCCP_REQUESTING and not releasing socket
  94          * lock select source port, enter ourselves into the hash tables and
  95          * complete initialization after this.
  96          */
  97         dccp_set_state(sk, DCCP_REQUESTING);
  98         err = inet_hash_connect(&dccp_death_row, sk);
  99         if (err != 0)
 100                 goto failure;
 101
 102         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->inet_sport,
 103                                 inet->inet_dport, sk);
 104         if (err != 0)
 105                 goto failure;
 106
 107         /* OK, now commit destination to socket.  */
 108         sk_setup_caps(sk, &rt->u.dst);
 109
 110         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 111                                                     inet->inet_daddr,
 112                                                     inet->inet_sport,
 113                                                     inet->inet_dport);
 114         inet->inet_id = dp->dccps_iss ^ jiffies;
 115
 116         err = dccp_connect(sk);
 117         rt = NULL;
 118         if (err != 0)
 119                 goto failure;
 120 out:
 121         return err;
 122 failure:
 123         /*
 124          * This unhashes the socket and releases the local port, if necessary.
 125          */
 126         dccp_set_state(sk, DCCP_CLOSED);
 127         ip_rt_put(rt);
 128         sk->sk_route_caps = 0;
 129         inet->inet_dport = 0;
 130         goto out;
 131 }
 132
 133 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 134
 135 /*
 136  * This routine does path mtu discovery as defined in RFC1191.
 137  */
 138 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 139                                           const struct iphdr *iph,
 140                                           u32 mtu)
 141 {
 142         struct dst_entry *dst;
 143         const struct inet_sock *inet = inet_sk(sk);
 144         const struct dccp_sock *dp = dccp_sk(sk);
 145
 146         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 147          * send out by Linux are always < 576bytes so they should go through
 148          * unfragmented).
 149          */
 150         if (sk->sk_state == DCCP_LISTEN)
 151                 return;
 152
 153         /* We don't check in the destentry if pmtu discovery is forbidden
 154          * on this route. We just assume that no packet_to_big packets
 155          * are send back when pmtu discovery is not active.
 156          * There is a small race when the user changes this flag in the
 157          * route, but I think that's acceptable.
 158          */
 159         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 160                 return;
 161
 162         dst->ops->update_pmtu(dst, mtu);
 163
 164         /* Something is about to be wrong... Remember soft error
 165          * for the case, if this connection will not able to recover.
 166          */
 167         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 168                 sk->sk_err_soft = EMSGSIZE;
 169
 170         mtu = dst_mtu(dst);
 171
 172         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 173             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 174                 dccp_sync_mss(sk, mtu);
 175
 176                 /*
 177                  * From RFC 4340, sec. 14.1:
 178                  *
 179                  *      DCCP-Sync packets are the best choice for upward
 180                  *      probing, since DCCP-Sync probes do not risk application
 181                  *      data loss.
 182                  */
 183                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 184         } /* else let the usual retransmit timer handle it */
 185 }
 186
 187 /*
 188  * This routine is called by the ICMP module when it gets some sort of error
 189  * condition. If err < 0 then the socket should be closed and the error
 190  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 191  * After adjustment header points to the first 8 bytes of the tcp header. We
 192  * need to find the appropriate port.
 193  *
 194  * The locking strategy used here is very "optimistic". When someone else
 195  * accesses the socket the ICMP is just dropped and for some paths there is no
 196  * check at all. A more general error queue to queue errors for later handling
 197  * is probably better.
 198  */
 199 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 200 {
 201         const struct iphdr *iph = (struct iphdr *)skb->data;
 202         const u8 offset = iph->ihl << 2;
 203         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 204         struct dccp_sock *dp;
 205         struct inet_sock *inet;
 206         const int type = icmp_hdr(skb)->type;
 207         const int code = icmp_hdr(skb)->code;
 208         struct sock *sk;
 209         __u64 seq;
 210         int err;
 211         struct net *net = dev_net(skb->dev);
 212
 213         if (skb->len < offset + sizeof(*dh) ||
 214             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 215                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 216                 return;
 217         }
 218
 219         sk = inet_lookup(net, &dccp_hashinfo,
 220                         iph->daddr, dh->dccph_dport,
 221                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 222         if (sk == NULL) {
 223                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 224                 return;
 225         }
 226
 227         if (sk->sk_state == DCCP_TIME_WAIT) {
 228                 inet_twsk_put(inet_twsk(sk));
 229                 return;
 230         }
 231
 232         bh_lock_sock(sk);
 233         /* If too many ICMPs get dropped on busy
 234          * servers this needs to be solved differently.
 235          */
 236         if (sock_owned_by_user(sk))
 237                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 238
 239         if (sk->sk_state == DCCP_CLOSED)
 240                 goto out;
 241
 242         dp = dccp_sk(sk);
 243         seq = dccp_hdr_seq(dh);
 244         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 245             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 246                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 247                 goto out;
 248         }
 249
 250         switch (type) {
 251         case ICMP_SOURCE_QUENCH:
 252                 /* Just silently ignore these. */
 253                 goto out;
 254         case ICMP_PARAMETERPROB:
 255                 err = EPROTO;
 256                 break;
 257         case ICMP_DEST_UNREACH:
 258                 if (code > NR_ICMP_UNREACH)
 259                         goto out;
 260
 261                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 262                         if (!sock_owned_by_user(sk))
 263                                 dccp_do_pmtu_discovery(sk, iph, info);
 264                         goto out;
 265                 }
 266
 267                 err = icmp_err_convert[code].errno;
 268                 break;
 269         case ICMP_TIME_EXCEEDED:
 270                 err = EHOSTUNREACH;
 271                 break;
 272         default:
 273                 goto out;
 274         }
 275
 276         switch (sk->sk_state) {
 277                 struct request_sock *req , **prev;
 278         case DCCP_LISTEN:
 279                 if (sock_owned_by_user(sk))
 280                         goto out;
 281                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 282                                           iph->daddr, iph->saddr);
 283                 if (!req)
 284                         goto out;
 285
 286                 /*
 287                  * ICMPs are not backlogged, hence we cannot get an established
 288                  * socket here.
 289                  */
 290                 WARN_ON(req->sk);
 291
 292                 if (seq != dccp_rsk(req)->dreq_iss) {
 293                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 294                         goto out;
 295                 }
 296                 /*
 297                  * Still in RESPOND, just remove it silently.
 298                  * There is no good way to pass the error to the newly
 299                  * created socket, and POSIX does not want network
 300                  * errors returned from accept().
 301                  */
 302                 inet_csk_reqsk_queue_drop(sk, req, prev);
 303                 goto out;
 304
 305         case DCCP_REQUESTING:
 306         case DCCP_RESPOND:
 307                 if (!sock_owned_by_user(sk)) {
 308                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 309                         sk->sk_err = err;
 310
 311                         sk->sk_error_report(sk);
 312
 313                         dccp_done(sk);
 314                 } else
 315                         sk->sk_err_soft = err;
 316                 goto out;
 317         }
 318
 319         /* If we've already connected we will keep trying
 320          * until we time out, or the user gives up.
 321          *
 322          * rfc1122 4.2.3.9 allows to consider as hard errors
 323          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 324          * but it is obsoleted by pmtu discovery).
 325          *
 326          * Note, that in modern internet, where routing is unreliable
 327          * and in each dark corner broken firewalls sit, sending random
 328          * errors ordered by their masters even this two messages finally lose
 329          * their original sense (even Linux sends invalid PORT_UNREACHs)
 330          *
 331          * Now we are in compliance with RFCs.
 332          *                                                      --ANK (980905)
 333          */
 334
 335         inet = inet_sk(sk);
 336         if (!sock_owned_by_user(sk) && inet->recverr) {
 337                 sk->sk_err = err;
 338                 sk->sk_error_report(sk);
 339         } else /* Only an error on timeout */
 340                 sk->sk_err_soft = err;
 341 out:
 342         bh_unlock_sock(sk);
 343         sock_put(sk);
 344 }
 345
 346 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 347                                       __be32 src, __be32 dst)
 348 {
 349         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 350 }
 351
 352 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 353 {
 354         const struct inet_sock *inet = inet_sk(sk);
 355         struct dccp_hdr *dh = dccp_hdr(skb);
 356
 357         dccp_csum_outgoing(skb);
 358         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 359                                                  inet->inet_saddr,
 360                                                  inet->inet_daddr);
 361 }
 362
 363 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 364
 365 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 366 {
 367         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 368                                            ip_hdr(skb)->saddr,
 369                                            dccp_hdr(skb)->dccph_dport,
 370                                            dccp_hdr(skb)->dccph_sport);
 371 }
 372
 373 /*
 374  * The three way handshake has completed - we got a valid ACK or DATAACK -
 375  * now create the new socket.
 376  *
 377  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 378  */
 379 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 380                                        struct request_sock *req,
 381                                        struct dst_entry *dst)
 382 {
 383         struct inet_request_sock *ireq;
 384         struct inet_sock *newinet;
 385         struct sock *newsk;
 386
 387         if (sk_acceptq_is_full(sk))
 388                 goto exit_overflow;
 389
 390         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 391                 goto exit;
 392
 393         newsk = dccp_create_openreq_child(sk, req, skb);
 394         if (newsk == NULL)
 395                 goto exit;
 396
 397         sk_setup_caps(newsk, dst);
 398
 399         newinet            = inet_sk(newsk);
 400         ireq               = inet_rsk(req);
 401         newinet->inet_daddr     = ireq->rmt_addr;
 402         newinet->inet_rcv_saddr = ireq->loc_addr;
 403         newinet->inet_saddr     = ireq->loc_addr;
 404         newinet->opt       = ireq->opt;
 405         ireq->opt          = NULL;
 406         newinet->mc_index  = inet_iif(skb);
 407         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 408         newinet->inet_id   = jiffies;
 409
 410         dccp_sync_mss(newsk, dst_mtu(dst));
 411
 412         __inet_hash_nolisten(newsk, NULL);
 413         __inet_inherit_port(sk, newsk);
 414
 415         return newsk;
 416
 417 exit_overflow:
 418         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 419 exit:
 420         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 421         dst_release(dst);
 422         return NULL;
 423 }
 424
 425 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 426
 427 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 428 {
 429         const struct dccp_hdr *dh = dccp_hdr(skb);
 430         const struct iphdr *iph = ip_hdr(skb);
 431         struct sock *nsk;
 432         struct request_sock **prev;
 433         /* Find possible connection requests. */
 434         struct request_sock *req = inet_csk_search_req(sk, &prev,
 435                                                        dh->dccph_sport,
 436                                                        iph->saddr, iph->daddr);
 437         if (req != NULL)
 438                 return dccp_check_req(sk, skb, req, prev);
 439
 440         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 441                                       iph->saddr, dh->dccph_sport,
 442                                       iph->daddr, dh->dccph_dport,
 443                                       inet_iif(skb));
 444         if (nsk != NULL) {
 445                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 446                         bh_lock_sock(nsk);
 447                         return nsk;
 448                 }
 449                 inet_twsk_put(inet_twsk(nsk));
 450                 return NULL;
 451         }
 452
 453         return sk;
 454 }
 455
 456 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 457                                            struct sk_buff *skb)
 458 {
 459         struct rtable *rt;
 460         struct flowi fl = { .oif = skb_rtable(skb)->rt_iif,
 461                             .nl_u = { .ip4_u =
 462                                       { .daddr = ip_hdr(skb)->saddr,
 463                                         .saddr = ip_hdr(skb)->daddr,
 464                                         .tos = RT_CONN_FLAGS(sk) } },
 465                             .proto = sk->sk_protocol,
 466                             .uli_u = { .ports =
 467                                        { .sport = dccp_hdr(skb)->dccph_dport,
 468                                          .dport = dccp_hdr(skb)->dccph_sport }
 469                                      }
 470                           };
 471
 472         security_skb_classify_flow(skb, &fl);
 473         if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
 474                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 475                 return NULL;
 476         }
 477
 478         return &rt->u.dst;
 479 }
 480
 481 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 482                                  struct request_values *rv_unused)
 483 {
 484         int err = -1;
 485         struct sk_buff *skb;
 486         struct dst_entry *dst;
 487
 488         dst = inet_csk_route_req(sk, req);
 489         if (dst == NULL)
 490                 goto out;
 491
 492         skb = dccp_make_response(sk, dst, req);
 493         if (skb != NULL) {
 494                 const struct inet_request_sock *ireq = inet_rsk(req);
 495                 struct dccp_hdr *dh = dccp_hdr(skb);
 496
 497                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 498                                                               ireq->rmt_addr);
 499                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 500                                             ireq->rmt_addr,
 501                                             ireq->opt);
 502                 err = net_xmit_eval(err);
 503         }
 504
 505 out:
 506         dst_release(dst);
 507         return err;
 508 }
 509
 510 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 511 {
 512         int err;
 513         const struct iphdr *rxiph;
 514         struct sk_buff *skb;
 515         struct dst_entry *dst;
 516         struct net *net = dev_net(skb_dst(rxskb)->dev);
 517         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 518
 519         /* Never send a reset in response to a reset. */
 520         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 521                 return;
 522
 523         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 524                 return;
 525
 526         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 527         if (dst == NULL)
 528                 return;
 529
 530         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 531         if (skb == NULL)
 532                 goto out;
 533
 534         rxiph = ip_hdr(rxskb);
 535         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 536                                                                  rxiph->daddr);
 537         skb_dst_set(skb, dst_clone(dst));
 538
 539         bh_lock_sock(ctl_sk);
 540         err = ip_build_and_send_pkt(skb, ctl_sk,
 541                                     rxiph->daddr, rxiph->saddr, NULL);
 542         bh_unlock_sock(ctl_sk);
 543
 544         if (net_xmit_eval(err) == 0) {
 545                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 546                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 547         }
 548 out:
 549          dst_release(dst);
 550 }
 551
 552 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 553 {
 554         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 555         kfree(inet_rsk(req)->opt);
 556 }
 557
 558 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 559         .family         = PF_INET,
 560         .obj_size       = sizeof(struct dccp_request_sock),
 561         .rtx_syn_ack    = dccp_v4_send_response,
 562         .send_ack       = dccp_reqsk_send_ack,
 563         .destructor     = dccp_v4_reqsk_destructor,
 564         .send_reset     = dccp_v4_ctl_send_reset,
 565 };
 566
 567 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 568 {
 569         struct inet_request_sock *ireq;
 570         struct request_sock *req;
 571         struct dccp_request_sock *dreq;
 572         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 573         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 574
 575         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 576         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 577                 return 0;       /* discard, don't send a reset here */
 578
 579         if (dccp_bad_service_code(sk, service)) {
 580                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 581                 goto drop;
 582         }
 583         /*
 584          * TW buckets are converted to open requests without
 585          * limitations, they conserve resources and peer is
 586          * evidently real one.
 587          */
 588         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 589         if (inet_csk_reqsk_queue_is_full(sk))
 590                 goto drop;
 591
 592         /*
 593          * Accept backlog is full. If we have already queued enough
 594          * of warm entries in syn queue, drop request. It is better than
 595          * clogging syn queue with openreqs with exponentially increasing
 596          * timeout.
 597          */
 598         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 599                 goto drop;
 600
 601         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 602         if (req == NULL)
 603                 goto drop;
 604
 605         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 606                 goto drop_and_free;
 607
 608         dreq = dccp_rsk(req);
 609         if (dccp_parse_options(sk, dreq, skb))
 610                 goto drop_and_free;
 611
 612         if (security_inet_conn_request(sk, skb, req))
 613                 goto drop_and_free;
 614
 615         ireq = inet_rsk(req);
 616         ireq->loc_addr = ip_hdr(skb)->daddr;
 617         ireq->rmt_addr = ip_hdr(skb)->saddr;
 618
 619         /*
 620          * Step 3: Process LISTEN state
 621          *
 622          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 623          *
 624          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 625          * dccp_create_openreq_child.
 626          */
 627         dreq->dreq_isr     = dcb->dccpd_seq;
 628         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 629         dreq->dreq_service = service;
 630
 631         if (dccp_v4_send_response(sk, req, NULL))
 632                 goto drop_and_free;
 633
 634         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 635         return 0;
 636
 637 drop_and_free:
 638         reqsk_free(req);
 639 drop:
 640         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 641         return -1;
 642 }
 643
 644 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 645
 646 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 647 {
 648         struct dccp_hdr *dh = dccp_hdr(skb);
 649
 650         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 651                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 652                         goto reset;
 653                 return 0;
 654         }
 655
 656         /*
 657          *  Step 3: Process LISTEN state
 658          *       If P.type == Request or P contains a valid Init Cookie option,
 659          *            (* Must scan the packet's options to check for Init
 660          *               Cookies.  Only Init Cookies are processed here,
 661          *               however; other options are processed in Step 8.  This
 662          *               scan need only be performed if the endpoint uses Init
 663          *               Cookies *)
 664          *            (* Generate a new socket and switch to that socket *)
 665          *            Set S := new socket for this port pair
 666          *            S.state = RESPOND
 667          *            Choose S.ISS (initial seqno) or set from Init Cookies
 668          *            Initialize S.GAR := S.ISS
 669          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 670          *            Continue with S.state == RESPOND
 671          *            (* A Response packet will be generated in Step 11 *)
 672          *       Otherwise,
 673          *            Generate Reset(No Connection) unless P.type == Reset
 674          *            Drop packet and return
 675          *
 676          * NOTE: the check for the packet types is done in
 677          *       dccp_rcv_state_process
 678          */
 679         if (sk->sk_state == DCCP_LISTEN) {
 680                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 681
 682                 if (nsk == NULL)
 683                         goto discard;
 684
 685                 if (nsk != sk) {
 686                         if (dccp_child_process(sk, nsk, skb))
 687                                 goto reset;
 688                         return 0;
 689                 }
 690         }
 691
 692         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 693                 goto reset;
 694         return 0;
 695
 696 reset:
 697         dccp_v4_ctl_send_reset(sk, skb);
 698 discard:
 699         kfree_skb(skb);
 700         return 0;
 701 }
 702
 703 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 704
 705 /**
 706  *      dccp_invalid_packet  -  check for malformed packets
 707  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 708  *      Packets that fail these checks are ignored and do not receive Resets.
 709  */
 710 int dccp_invalid_packet(struct sk_buff *skb)
 711 {
 712         const struct dccp_hdr *dh;
 713         unsigned int cscov;
 714
 715         if (skb->pkt_type != PACKET_HOST)
 716                 return 1;
 717
 718         /* If the packet is shorter than 12 bytes, drop packet and return */
 719         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 720                 DCCP_WARN("pskb_may_pull failed\n");
 721                 return 1;
 722         }
 723
 724         dh = dccp_hdr(skb);
 725
 726         /* If P.type is not understood, drop packet and return */
 727         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 728                 DCCP_WARN("invalid packet type\n");
 729                 return 1;
 730         }
 731
 732         /*
 733          * If P.Data Offset is too small for packet type, drop packet and return
 734          */
 735         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 736                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 737                 return 1;
 738         }
 739         /*
 740          * If P.Data Offset is too too large for packet, drop packet and return
 741          */
 742         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 743                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 744                 return 1;
 745         }
 746
 747         /*
 748          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 749          * has short sequence numbers), drop packet and return
 750          */
 751         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 752             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 753                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 754                           dccp_packet_name(dh->dccph_type));
 755                 return 1;
 756         }
 757
 758         /*
 759          * If P.CsCov is too large for the packet size, drop packet and return.
 760          * This must come _before_ checksumming (not as RFC 4340 suggests).
 761          */
 762         cscov = dccp_csum_coverage(skb);
 763         if (cscov > skb->len) {
 764                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 765                           dh->dccph_cscov, skb->len);
 766                 return 1;
 767         }
 768
 769         /* If header checksum is incorrect, drop packet and return.
 770          * (This step is completed in the AF-dependent functions.) */
 771         skb->csum = skb_checksum(skb, 0, cscov, 0);
 772
 773         return 0;
 774 }
 775
 776 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 777
 778 /* this is called when real data arrives */
 779 static int dccp_v4_rcv(struct sk_buff *skb)
 780 {
 781         const struct dccp_hdr *dh;
 782         const struct iphdr *iph;
 783         struct sock *sk;
 784         int min_cov;
 785
 786         /* Step 1: Check header basics */
 787
 788         if (dccp_invalid_packet(skb))
 789                 goto discard_it;
 790
 791         iph = ip_hdr(skb);
 792         /* Step 1: If header checksum is incorrect, drop packet and return */
 793         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 794                 DCCP_WARN("dropped packet with invalid checksum\n");
 795                 goto discard_it;
 796         }
 797
 798         dh = dccp_hdr(skb);
 799
 800         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 801         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 802
 803         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 804                       dccp_packet_name(dh->dccph_type),
 805                       &iph->saddr, ntohs(dh->dccph_sport),
 806                       &iph->daddr, ntohs(dh->dccph_dport),
 807                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 808
 809         if (dccp_packet_without_ack(skb)) {
 810                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 811                 dccp_pr_debug_cat("\n");
 812         } else {
 813                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 814                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 815                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 816         }
 817
 818         /* Step 2:
 819          *      Look up flow ID in table and get corresponding socket */
 820         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 821                                dh->dccph_sport, dh->dccph_dport);
 822         /*
 823          * Step 2:
 824          *      If no socket ...
 825          */
 826         if (sk == NULL) {
 827                 dccp_pr_debug("failed to look up flow ID in table and "
 828                               "get corresponding socket\n");
 829                 goto no_dccp_socket;
 830         }
 831
 832         /*
 833          * Step 2:
 834          *      ... or S.state == TIMEWAIT,
 835          *              Generate Reset(No Connection) unless P.type == Reset
 836          *              Drop packet and return
 837          */
 838         if (sk->sk_state == DCCP_TIME_WAIT) {
 839                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 840                 inet_twsk_put(inet_twsk(sk));
 841                 goto no_dccp_socket;
 842         }
 843
 844         /*
 845          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 846          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 847          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 848          */
 849         min_cov = dccp_sk(sk)->dccps_pcrlen;
 850         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 851                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 852                               dh->dccph_cscov, min_cov);
 853                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 854                  *         options (Section 11.7) with Drop Code 0, Protocol
 855                  *         Constraints."                                     */
 856                 goto discard_and_relse;
 857         }
 858
 859         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 860                 goto discard_and_relse;
 861         nf_reset(skb);
 862
 863         return sk_receive_skb(sk, skb, 1);
 864
 865 no_dccp_socket:
 866         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 867                 goto discard_it;
 868         /*
 869          * Step 2:
 870          *      If no socket ...
 871          *              Generate Reset(No Connection) unless P.type == Reset
 872          *              Drop packet and return
 873          */
 874         if (dh->dccph_type != DCCP_PKT_RESET) {
 875                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 876                                         DCCP_RESET_CODE_NO_CONNECTION;
 877                 dccp_v4_ctl_send_reset(sk, skb);
 878         }
 879
 880 discard_it:
 881         kfree_skb(skb);
 882         return 0;
 883
 884 discard_and_relse:
 885         sock_put(sk);
 886         goto discard_it;
 887 }
 888
 889 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 890         .queue_xmit        = ip_queue_xmit,
 891         .send_check        = dccp_v4_send_check,
 892         .rebuild_header    = inet_sk_rebuild_header,
 893         .conn_request      = dccp_v4_conn_request,
 894         .syn_recv_sock     = dccp_v4_request_recv_sock,
 895         .net_header_len    = sizeof(struct iphdr),
 896         .setsockopt        = ip_setsockopt,
 897         .getsockopt        = ip_getsockopt,
 898         .addr2sockaddr     = inet_csk_addr2sockaddr,
 899         .sockaddr_len      = sizeof(struct sockaddr_in),
 900         .bind_conflict     = inet_csk_bind_conflict,
 901 #ifdef CONFIG_COMPAT
 902         .compat_setsockopt = compat_ip_setsockopt,
 903         .compat_getsockopt = compat_ip_getsockopt,
 904 #endif
 905 };
 906
 907 static int dccp_v4_init_sock(struct sock *sk)
 908 {
 909         static __u8 dccp_v4_ctl_sock_initialized;
 910         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 911
 912         if (err == 0) {
 913                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 914                         dccp_v4_ctl_sock_initialized = 1;
 915                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 916         }
 917
 918         return err;
 919 }
 920
 921 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 922         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 923 };
 924
 925 static struct proto dccp_v4_prot = {
 926         .name                   = "DCCP",
 927         .owner                  = THIS_MODULE,
 928         .close                  = dccp_close,
 929         .connect                = dccp_v4_connect,
 930         .disconnect             = dccp_disconnect,
 931         .ioctl                  = dccp_ioctl,
 932         .init                   = dccp_v4_init_sock,
 933         .setsockopt             = dccp_setsockopt,
 934         .getsockopt             = dccp_getsockopt,
 935         .sendmsg                = dccp_sendmsg,
 936         .recvmsg                = dccp_recvmsg,
 937         .backlog_rcv            = dccp_v4_do_rcv,
 938         .hash                   = inet_hash,
 939         .unhash                 = inet_unhash,
 940         .accept                 = inet_csk_accept,
 941         .get_port               = inet_csk_get_port,
 942         .shutdown               = dccp_shutdown,
 943         .destroy                = dccp_destroy_sock,
 944         .orphan_count           = &dccp_orphan_count,
 945         .max_header             = MAX_DCCP_HEADER,
 946         .obj_size               = sizeof(struct dccp_sock),
 947         .slab_flags             = SLAB_DESTROY_BY_RCU,
 948         .rsk_prot               = &dccp_request_sock_ops,
 949         .twsk_prot              = &dccp_timewait_sock_ops,
 950         .h.hashinfo             = &dccp_hashinfo,
 951 #ifdef CONFIG_COMPAT
 952         .compat_setsockopt      = compat_dccp_setsockopt,
 953         .compat_getsockopt      = compat_dccp_getsockopt,
 954 #endif
 955 };
 956
 957 static const struct net_protocol dccp_v4_protocol = {
 958         .handler        = dccp_v4_rcv,
 959         .err_handler    = dccp_v4_err,
 960         .no_policy      = 1,
 961         .netns_ok       = 1,
 962 };
 963
 964 static const struct proto_ops inet_dccp_ops = {
 965         .family            = PF_INET,
 966         .owner             = THIS_MODULE,
 967         .release           = inet_release,
 968         .bind              = inet_bind,
 969         .connect           = inet_stream_connect,
 970         .socketpair        = sock_no_socketpair,
 971         .accept            = inet_accept,
 972         .getname           = inet_getname,
 973         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 974         .poll              = dccp_poll,
 975         .ioctl             = inet_ioctl,
 976         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 977         .listen            = inet_dccp_listen,
 978         .shutdown          = inet_shutdown,
 979         .setsockopt        = sock_common_setsockopt,
 980         .getsockopt        = sock_common_getsockopt,
 981         .sendmsg           = inet_sendmsg,
 982         .recvmsg           = sock_common_recvmsg,
 983         .mmap              = sock_no_mmap,
 984         .sendpage          = sock_no_sendpage,
 985 #ifdef CONFIG_COMPAT
 986         .compat_setsockopt = compat_sock_common_setsockopt,
 987         .compat_getsockopt = compat_sock_common_getsockopt,
 988 #endif
 989 };
 990
 991 static struct inet_protosw dccp_v4_protosw = {
 992         .type           = SOCK_DCCP,
 993         .protocol       = IPPROTO_DCCP,
 994         .prot           = &dccp_v4_prot,
 995         .ops            = &inet_dccp_ops,
 996         .no_check       = 0,
 997         .flags          = INET_PROTOSW_ICSK,
 998 };
 999
1000 static int __net_init dccp_v4_init_net(struct net *net)
1001 {
1002         if (dccp_hashinfo.bhash == NULL)
1003                 return -ESOCKTNOSUPPORT;
1004
1005         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1006                                     SOCK_DCCP, IPPROTO_DCCP, net);
1007 }
1008
1009 static void __net_exit dccp_v4_exit_net(struct net *net)
1010 {
1011         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1012 }
1013
1014 static struct pernet_operations dccp_v4_ops = {
1015         .init   = dccp_v4_init_net,
1016         .exit   = dccp_v4_exit_net,
1017 };
1018
1019 static int __init dccp_v4_init(void)
1020 {
1021         int err = proto_register(&dccp_v4_prot, 1);
1022
1023         if (err != 0)
1024                 goto out;
1025
1026         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1027         if (err != 0)
1028                 goto out_proto_unregister;
1029
1030         inet_register_protosw(&dccp_v4_protosw);
1031
1032         err = register_pernet_subsys(&dccp_v4_ops);
1033         if (err)
1034                 goto out_destroy_ctl_sock;
1035 out:
1036         return err;
1037 out_destroy_ctl_sock:
1038         inet_unregister_protosw(&dccp_v4_protosw);
1039         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1040 out_proto_unregister:
1041         proto_unregister(&dccp_v4_prot);
1042         goto out;
1043 }
1044
1045 static void __exit dccp_v4_exit(void)
1046 {
1047         unregister_pernet_subsys(&dccp_v4_ops);
1048         inet_unregister_protosw(&dccp_v4_protosw);
1049         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1050         proto_unregister(&dccp_v4_prot);
1051 }
1052
1053 module_init(dccp_v4_init);
1054 module_exit(dccp_v4_exit);
1055
1056 /*
1057  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1058  * values directly, Also cover the case where the protocol is not specified,
1059  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1060  */
1061 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1062 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1063 MODULE_LICENSE("GPL");
1064 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1065 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");