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->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_nonewsk;
 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         if (__inet_inherit_port(sk, newsk) < 0) {
 413                 sock_put(newsk);
 414                 goto exit;
 415         }
 416         __inet_hash_nolisten(newsk, NULL);
 417
 418         return newsk;
 419
 420 exit_overflow:
 421         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 422 exit_nonewsk:
 423         dst_release(dst);
 424 exit:
 425         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 426         return NULL;
 427 }
 428
 429 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 430
 431 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 432 {
 433         const struct dccp_hdr *dh = dccp_hdr(skb);
 434         const struct iphdr *iph = ip_hdr(skb);
 435         struct sock *nsk;
 436         struct request_sock **prev;
 437         /* Find possible connection requests. */
 438         struct request_sock *req = inet_csk_search_req(sk, &prev,
 439                                                        dh->dccph_sport,
 440                                                        iph->saddr, iph->daddr);
 441         if (req != NULL)
 442                 return dccp_check_req(sk, skb, req, prev);
 443
 444         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 445                                       iph->saddr, dh->dccph_sport,
 446                                       iph->daddr, dh->dccph_dport,
 447                                       inet_iif(skb));
 448         if (nsk != NULL) {
 449                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 450                         bh_lock_sock(nsk);
 451                         return nsk;
 452                 }
 453                 inet_twsk_put(inet_twsk(nsk));
 454                 return NULL;
 455         }
 456
 457         return sk;
 458 }
 459
 460 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 461                                            struct sk_buff *skb)
 462 {
 463         struct rtable *rt;
 464         struct flowi fl = { .oif = skb_rtable(skb)->rt_iif,
 465                             .nl_u = { .ip4_u =
 466                                       { .daddr = ip_hdr(skb)->saddr,
 467                                         .saddr = ip_hdr(skb)->daddr,
 468                                         .tos = RT_CONN_FLAGS(sk) } },
 469                             .proto = sk->sk_protocol,
 470                             .uli_u = { .ports =
 471                                        { .sport = dccp_hdr(skb)->dccph_dport,
 472                                          .dport = dccp_hdr(skb)->dccph_sport }
 473                                      }
 474                           };
 475
 476         security_skb_classify_flow(skb, &fl);
 477         if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
 478                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 479                 return NULL;
 480         }
 481
 482         return &rt->dst;
 483 }
 484
 485 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 486                                  struct request_values *rv_unused)
 487 {
 488         int err = -1;
 489         struct sk_buff *skb;
 490         struct dst_entry *dst;
 491
 492         dst = inet_csk_route_req(sk, req);
 493         if (dst == NULL)
 494                 goto out;
 495
 496         skb = dccp_make_response(sk, dst, req);
 497         if (skb != NULL) {
 498                 const struct inet_request_sock *ireq = inet_rsk(req);
 499                 struct dccp_hdr *dh = dccp_hdr(skb);
 500
 501                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 502                                                               ireq->rmt_addr);
 503                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 504                                             ireq->rmt_addr,
 505                                             ireq->opt);
 506                 err = net_xmit_eval(err);
 507         }
 508
 509 out:
 510         dst_release(dst);
 511         return err;
 512 }
 513
 514 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 515 {
 516         int err;
 517         const struct iphdr *rxiph;
 518         struct sk_buff *skb;
 519         struct dst_entry *dst;
 520         struct net *net = dev_net(skb_dst(rxskb)->dev);
 521         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 522
 523         /* Never send a reset in response to a reset. */
 524         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 525                 return;
 526
 527         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 528                 return;
 529
 530         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 531         if (dst == NULL)
 532                 return;
 533
 534         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 535         if (skb == NULL)
 536                 goto out;
 537
 538         rxiph = ip_hdr(rxskb);
 539         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 540                                                                  rxiph->daddr);
 541         skb_dst_set(skb, dst_clone(dst));
 542
 543         bh_lock_sock(ctl_sk);
 544         err = ip_build_and_send_pkt(skb, ctl_sk,
 545                                     rxiph->daddr, rxiph->saddr, NULL);
 546         bh_unlock_sock(ctl_sk);
 547
 548         if (net_xmit_eval(err) == 0) {
 549                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 550                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 551         }
 552 out:
 553          dst_release(dst);
 554 }
 555
 556 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 557 {
 558         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 559         kfree(inet_rsk(req)->opt);
 560 }
 561
 562 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 563         .family         = PF_INET,
 564         .obj_size       = sizeof(struct dccp_request_sock),
 565         .rtx_syn_ack    = dccp_v4_send_response,
 566         .send_ack       = dccp_reqsk_send_ack,
 567         .destructor     = dccp_v4_reqsk_destructor,
 568         .send_reset     = dccp_v4_ctl_send_reset,
 569 };
 570
 571 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 572 {
 573         struct inet_request_sock *ireq;
 574         struct request_sock *req;
 575         struct dccp_request_sock *dreq;
 576         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 577         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 578
 579         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 580         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 581                 return 0;       /* discard, don't send a reset here */
 582
 583         if (dccp_bad_service_code(sk, service)) {
 584                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 585                 goto drop;
 586         }
 587         /*
 588          * TW buckets are converted to open requests without
 589          * limitations, they conserve resources and peer is
 590          * evidently real one.
 591          */
 592         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 593         if (inet_csk_reqsk_queue_is_full(sk))
 594                 goto drop;
 595
 596         /*
 597          * Accept backlog is full. If we have already queued enough
 598          * of warm entries in syn queue, drop request. It is better than
 599          * clogging syn queue with openreqs with exponentially increasing
 600          * timeout.
 601          */
 602         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 603                 goto drop;
 604
 605         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 606         if (req == NULL)
 607                 goto drop;
 608
 609         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 610                 goto drop_and_free;
 611
 612         dreq = dccp_rsk(req);
 613         if (dccp_parse_options(sk, dreq, skb))
 614                 goto drop_and_free;
 615
 616         if (security_inet_conn_request(sk, skb, req))
 617                 goto drop_and_free;
 618
 619         ireq = inet_rsk(req);
 620         ireq->loc_addr = ip_hdr(skb)->daddr;
 621         ireq->rmt_addr = ip_hdr(skb)->saddr;
 622
 623         /*
 624          * Step 3: Process LISTEN state
 625          *
 626          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 627          *
 628          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 629          * dccp_create_openreq_child.
 630          */
 631         dreq->dreq_isr     = dcb->dccpd_seq;
 632         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 633         dreq->dreq_service = service;
 634
 635         if (dccp_v4_send_response(sk, req, NULL))
 636                 goto drop_and_free;
 637
 638         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 639         return 0;
 640
 641 drop_and_free:
 642         reqsk_free(req);
 643 drop:
 644         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 645         return -1;
 646 }
 647
 648 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 649
 650 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 651 {
 652         struct dccp_hdr *dh = dccp_hdr(skb);
 653
 654         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 655                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 656                         goto reset;
 657                 return 0;
 658         }
 659
 660         /*
 661          *  Step 3: Process LISTEN state
 662          *       If P.type == Request or P contains a valid Init Cookie option,
 663          *            (* Must scan the packet's options to check for Init
 664          *               Cookies.  Only Init Cookies are processed here,
 665          *               however; other options are processed in Step 8.  This
 666          *               scan need only be performed if the endpoint uses Init
 667          *               Cookies *)
 668          *            (* Generate a new socket and switch to that socket *)
 669          *            Set S := new socket for this port pair
 670          *            S.state = RESPOND
 671          *            Choose S.ISS (initial seqno) or set from Init Cookies
 672          *            Initialize S.GAR := S.ISS
 673          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 674          *            Continue with S.state == RESPOND
 675          *            (* A Response packet will be generated in Step 11 *)
 676          *       Otherwise,
 677          *            Generate Reset(No Connection) unless P.type == Reset
 678          *            Drop packet and return
 679          *
 680          * NOTE: the check for the packet types is done in
 681          *       dccp_rcv_state_process
 682          */
 683         if (sk->sk_state == DCCP_LISTEN) {
 684                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 685
 686                 if (nsk == NULL)
 687                         goto discard;
 688
 689                 if (nsk != sk) {
 690                         if (dccp_child_process(sk, nsk, skb))
 691                                 goto reset;
 692                         return 0;
 693                 }
 694         }
 695
 696         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 697                 goto reset;
 698         return 0;
 699
 700 reset:
 701         dccp_v4_ctl_send_reset(sk, skb);
 702 discard:
 703         kfree_skb(skb);
 704         return 0;
 705 }
 706
 707 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 708
 709 /**
 710  *      dccp_invalid_packet  -  check for malformed packets
 711  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 712  *      Packets that fail these checks are ignored and do not receive Resets.
 713  */
 714 int dccp_invalid_packet(struct sk_buff *skb)
 715 {
 716         const struct dccp_hdr *dh;
 717         unsigned int cscov;
 718
 719         if (skb->pkt_type != PACKET_HOST)
 720                 return 1;
 721
 722         /* If the packet is shorter than 12 bytes, drop packet and return */
 723         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 724                 DCCP_WARN("pskb_may_pull failed\n");
 725                 return 1;
 726         }
 727
 728         dh = dccp_hdr(skb);
 729
 730         /* If P.type is not understood, drop packet and return */
 731         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 732                 DCCP_WARN("invalid packet type\n");
 733                 return 1;
 734         }
 735
 736         /*
 737          * If P.Data Offset is too small for packet type, drop packet and return
 738          */
 739         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 740                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 741                 return 1;
 742         }
 743         /*
 744          * If P.Data Offset is too too large for packet, drop packet and return
 745          */
 746         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 747                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 748                 return 1;
 749         }
 750
 751         /*
 752          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 753          * has short sequence numbers), drop packet and return
 754          */
 755         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 756             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 757                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 758                           dccp_packet_name(dh->dccph_type));
 759                 return 1;
 760         }
 761
 762         /*
 763          * If P.CsCov is too large for the packet size, drop packet and return.
 764          * This must come _before_ checksumming (not as RFC 4340 suggests).
 765          */
 766         cscov = dccp_csum_coverage(skb);
 767         if (cscov > skb->len) {
 768                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 769                           dh->dccph_cscov, skb->len);
 770                 return 1;
 771         }
 772
 773         /* If header checksum is incorrect, drop packet and return.
 774          * (This step is completed in the AF-dependent functions.) */
 775         skb->csum = skb_checksum(skb, 0, cscov, 0);
 776
 777         return 0;
 778 }
 779
 780 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 781
 782 /* this is called when real data arrives */
 783 static int dccp_v4_rcv(struct sk_buff *skb)
 784 {
 785         const struct dccp_hdr *dh;
 786         const struct iphdr *iph;
 787         struct sock *sk;
 788         int min_cov;
 789
 790         /* Step 1: Check header basics */
 791
 792         if (dccp_invalid_packet(skb))
 793                 goto discard_it;
 794
 795         iph = ip_hdr(skb);
 796         /* Step 1: If header checksum is incorrect, drop packet and return */
 797         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 798                 DCCP_WARN("dropped packet with invalid checksum\n");
 799                 goto discard_it;
 800         }
 801
 802         dh = dccp_hdr(skb);
 803
 804         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 805         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 806
 807         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 808                       dccp_packet_name(dh->dccph_type),
 809                       &iph->saddr, ntohs(dh->dccph_sport),
 810                       &iph->daddr, ntohs(dh->dccph_dport),
 811                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 812
 813         if (dccp_packet_without_ack(skb)) {
 814                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 815                 dccp_pr_debug_cat("\n");
 816         } else {
 817                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 818                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 819                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 820         }
 821
 822         /* Step 2:
 823          *      Look up flow ID in table and get corresponding socket */
 824         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 825                                dh->dccph_sport, dh->dccph_dport);
 826         /*
 827          * Step 2:
 828          *      If no socket ...
 829          */
 830         if (sk == NULL) {
 831                 dccp_pr_debug("failed to look up flow ID in table and "
 832                               "get corresponding socket\n");
 833                 goto no_dccp_socket;
 834         }
 835
 836         /*
 837          * Step 2:
 838          *      ... or S.state == TIMEWAIT,
 839          *              Generate Reset(No Connection) unless P.type == Reset
 840          *              Drop packet and return
 841          */
 842         if (sk->sk_state == DCCP_TIME_WAIT) {
 843                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 844                 inet_twsk_put(inet_twsk(sk));
 845                 goto no_dccp_socket;
 846         }
 847
 848         /*
 849          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 850          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 851          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 852          */
 853         min_cov = dccp_sk(sk)->dccps_pcrlen;
 854         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 855                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 856                               dh->dccph_cscov, min_cov);
 857                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 858                  *         options (Section 11.7) with Drop Code 0, Protocol
 859                  *         Constraints."                                     */
 860                 goto discard_and_relse;
 861         }
 862
 863         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 864                 goto discard_and_relse;
 865         nf_reset(skb);
 866
 867         return sk_receive_skb(sk, skb, 1);
 868
 869 no_dccp_socket:
 870         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 871                 goto discard_it;
 872         /*
 873          * Step 2:
 874          *      If no socket ...
 875          *              Generate Reset(No Connection) unless P.type == Reset
 876          *              Drop packet and return
 877          */
 878         if (dh->dccph_type != DCCP_PKT_RESET) {
 879                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 880                                         DCCP_RESET_CODE_NO_CONNECTION;
 881                 dccp_v4_ctl_send_reset(sk, skb);
 882         }
 883
 884 discard_it:
 885         kfree_skb(skb);
 886         return 0;
 887
 888 discard_and_relse:
 889         sock_put(sk);
 890         goto discard_it;
 891 }
 892
 893 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 894         .queue_xmit        = ip_queue_xmit,
 895         .send_check        = dccp_v4_send_check,
 896         .rebuild_header    = inet_sk_rebuild_header,
 897         .conn_request      = dccp_v4_conn_request,
 898         .syn_recv_sock     = dccp_v4_request_recv_sock,
 899         .net_header_len    = sizeof(struct iphdr),
 900         .setsockopt        = ip_setsockopt,
 901         .getsockopt        = ip_getsockopt,
 902         .addr2sockaddr     = inet_csk_addr2sockaddr,
 903         .sockaddr_len      = sizeof(struct sockaddr_in),
 904         .bind_conflict     = inet_csk_bind_conflict,
 905 #ifdef CONFIG_COMPAT
 906         .compat_setsockopt = compat_ip_setsockopt,
 907         .compat_getsockopt = compat_ip_getsockopt,
 908 #endif
 909 };
 910
 911 static int dccp_v4_init_sock(struct sock *sk)
 912 {
 913         static __u8 dccp_v4_ctl_sock_initialized;
 914         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 915
 916         if (err == 0) {
 917                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 918                         dccp_v4_ctl_sock_initialized = 1;
 919                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 920         }
 921
 922         return err;
 923 }
 924
 925 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 926         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 927 };
 928
 929 static struct proto dccp_v4_prot = {
 930         .name                   = "DCCP",
 931         .owner                  = THIS_MODULE,
 932         .close                  = dccp_close,
 933         .connect                = dccp_v4_connect,
 934         .disconnect             = dccp_disconnect,
 935         .ioctl                  = dccp_ioctl,
 936         .init                   = dccp_v4_init_sock,
 937         .setsockopt             = dccp_setsockopt,
 938         .getsockopt             = dccp_getsockopt,
 939         .sendmsg                = dccp_sendmsg,
 940         .recvmsg                = dccp_recvmsg,
 941         .backlog_rcv            = dccp_v4_do_rcv,
 942         .hash                   = inet_hash,
 943         .unhash                 = inet_unhash,
 944         .accept                 = inet_csk_accept,
 945         .get_port               = inet_csk_get_port,
 946         .shutdown               = dccp_shutdown,
 947         .destroy                = dccp_destroy_sock,
 948         .orphan_count           = &dccp_orphan_count,
 949         .max_header             = MAX_DCCP_HEADER,
 950         .obj_size               = sizeof(struct dccp_sock),
 951         .slab_flags             = SLAB_DESTROY_BY_RCU,
 952         .rsk_prot               = &dccp_request_sock_ops,
 953         .twsk_prot              = &dccp_timewait_sock_ops,
 954         .h.hashinfo             = &dccp_hashinfo,
 955 #ifdef CONFIG_COMPAT
 956         .compat_setsockopt      = compat_dccp_setsockopt,
 957         .compat_getsockopt      = compat_dccp_getsockopt,
 958 #endif
 959 };
 960
 961 static const struct net_protocol dccp_v4_protocol = {
 962         .handler        = dccp_v4_rcv,
 963         .err_handler    = dccp_v4_err,
 964         .no_policy      = 1,
 965         .netns_ok       = 1,
 966 };
 967
 968 static const struct proto_ops inet_dccp_ops = {
 969         .family            = PF_INET,
 970         .owner             = THIS_MODULE,
 971         .release           = inet_release,
 972         .bind              = inet_bind,
 973         .connect           = inet_stream_connect,
 974         .socketpair        = sock_no_socketpair,
 975         .accept            = inet_accept,
 976         .getname           = inet_getname,
 977         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 978         .poll              = dccp_poll,
 979         .ioctl             = inet_ioctl,
 980         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 981         .listen            = inet_dccp_listen,
 982         .shutdown          = inet_shutdown,
 983         .setsockopt        = sock_common_setsockopt,
 984         .getsockopt        = sock_common_getsockopt,
 985         .sendmsg           = inet_sendmsg,
 986         .recvmsg           = sock_common_recvmsg,
 987         .mmap              = sock_no_mmap,
 988         .sendpage          = sock_no_sendpage,
 989 #ifdef CONFIG_COMPAT
 990         .compat_setsockopt = compat_sock_common_setsockopt,
 991         .compat_getsockopt = compat_sock_common_getsockopt,
 992 #endif
 993 };
 994
 995 static struct inet_protosw dccp_v4_protosw = {
 996         .type           = SOCK_DCCP,
 997         .protocol       = IPPROTO_DCCP,
 998         .prot           = &dccp_v4_prot,
 999         .ops            = &inet_dccp_ops,
1000         .no_check       = 0,
1001         .flags          = INET_PROTOSW_ICSK,
1002 };
1003
1004 static int __net_init dccp_v4_init_net(struct net *net)
1005 {
1006         if (dccp_hashinfo.bhash == NULL)
1007                 return -ESOCKTNOSUPPORT;
1008
1009         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1010                                     SOCK_DCCP, IPPROTO_DCCP, net);
1011 }
1012
1013 static void __net_exit dccp_v4_exit_net(struct net *net)
1014 {
1015         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1016 }
1017
1018 static struct pernet_operations dccp_v4_ops = {
1019         .init   = dccp_v4_init_net,
1020         .exit   = dccp_v4_exit_net,
1021 };
1022
1023 static int __init dccp_v4_init(void)
1024 {
1025         int err = proto_register(&dccp_v4_prot, 1);
1026
1027         if (err != 0)
1028                 goto out;
1029
1030         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1031         if (err != 0)
1032                 goto out_proto_unregister;
1033
1034         inet_register_protosw(&dccp_v4_protosw);
1035
1036         err = register_pernet_subsys(&dccp_v4_ops);
1037         if (err)
1038                 goto out_destroy_ctl_sock;
1039 out:
1040         return err;
1041 out_destroy_ctl_sock:
1042         inet_unregister_protosw(&dccp_v4_protosw);
1043         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1044 out_proto_unregister:
1045         proto_unregister(&dccp_v4_prot);
1046         goto out;
1047 }
1048
1049 static void __exit dccp_v4_exit(void)
1050 {
1051         unregister_pernet_subsys(&dccp_v4_ops);
1052         inet_unregister_protosw(&dccp_v4_protosw);
1053         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1054         proto_unregister(&dccp_v4_prot);
1055 }
1056
1057 module_init(dccp_v4_init);
1058 module_exit(dccp_v4_exit);
1059
1060 /*
1061  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1062  * values directly, Also cover the case where the protocol is not specified,
1063  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1064  */
1065 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1066 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1067 MODULE_LICENSE("GPL");
1068 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1069 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");