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