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 #include <net/secure_seq.h>
  30
  31 #include "ackvec.h"
  32 #include "ccid.h"
  33 #include "dccp.h"
  34 #include "feat.h"
  35
  36 /*
  37  * The per-net dccp.v4_ctl_sk socket is used for responding to
  38  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  39  * for this socket at the initialization time.
  40  */
  41
  42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  43 {
  44         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  45         struct inet_sock *inet = inet_sk(sk);
  46         struct dccp_sock *dp = dccp_sk(sk);
  47         __be16 orig_sport, orig_dport;
  48         __be32 daddr, nexthop;
  49         struct flowi4 *fl4;
  50         struct rtable *rt;
  51         int err;
  52         struct ip_options_rcu *inet_opt;
  53
  54         dp->dccps_role = DCCP_ROLE_CLIENT;
  55
  56         if (addr_len < sizeof(struct sockaddr_in))
  57                 return -EINVAL;
  58
  59         if (usin->sin_family != AF_INET)
  60                 return -EAFNOSUPPORT;
  61
  62         nexthop = daddr = usin->sin_addr.s_addr;
  63
  64         inet_opt = rcu_dereference_protected(inet->inet_opt,
  65                                              sock_owned_by_user(sk));
  66         if (inet_opt != NULL && inet_opt->opt.srr) {
  67                 if (daddr == 0)
  68                         return -EINVAL;
  69                 nexthop = inet_opt->opt.faddr;
  70         }
  71
  72         orig_sport = inet->inet_sport;
  73         orig_dport = usin->sin_port;
  74         fl4 = &inet->cork.fl.u.ip4;
  75         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
  76                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  77                               IPPROTO_DCCP,
  78                               orig_sport, orig_dport, sk);
  79         if (IS_ERR(rt))
  80                 return PTR_ERR(rt);
  81
  82         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  83                 ip_rt_put(rt);
  84                 return -ENETUNREACH;
  85         }
  86
  87         if (inet_opt == NULL || !inet_opt->opt.srr)
  88                 daddr = fl4->daddr;
  89
  90         if (inet->inet_saddr == 0)
  91                 inet->inet_saddr = fl4->saddr;
  92         sk_rcv_saddr_set(sk, inet->inet_saddr);
  93         inet->inet_dport = usin->sin_port;
  94         sk_daddr_set(sk, daddr);
  95
  96         inet_csk(sk)->icsk_ext_hdr_len = 0;
  97         if (inet_opt)
  98                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
  99         /*
 100          * Socket identity is still unknown (sport may be zero).
 101          * However we set state to DCCP_REQUESTING and not releasing socket
 102          * lock select source port, enter ourselves into the hash tables and
 103          * complete initialization after this.
 104          */
 105         dccp_set_state(sk, DCCP_REQUESTING);
 106         err = inet_hash_connect(&dccp_death_row, sk);
 107         if (err != 0)
 108                 goto failure;
 109
 110         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 111                                inet->inet_sport, inet->inet_dport, sk);
 112         if (IS_ERR(rt)) {
 113                 err = PTR_ERR(rt);
 114                 rt = NULL;
 115                 goto failure;
 116         }
 117         /* OK, now commit destination to socket.  */
 118         sk_setup_caps(sk, &rt->dst);
 119
 120         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 121                                                     inet->inet_daddr,
 122                                                     inet->inet_sport,
 123                                                     inet->inet_dport);
 124         inet->inet_id = dp->dccps_iss ^ jiffies;
 125
 126         err = dccp_connect(sk);
 127         rt = NULL;
 128         if (err != 0)
 129                 goto failure;
 130 out:
 131         return err;
 132 failure:
 133         /*
 134          * This unhashes the socket and releases the local port, if necessary.
 135          */
 136         dccp_set_state(sk, DCCP_CLOSED);
 137         ip_rt_put(rt);
 138         sk->sk_route_caps = 0;
 139         inet->inet_dport = 0;
 140         goto out;
 141 }
 142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 143
 144 /*
 145  * This routine does path mtu discovery as defined in RFC1191.
 146  */
 147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 148                                           const struct iphdr *iph,
 149                                           u32 mtu)
 150 {
 151         struct dst_entry *dst;
 152         const struct inet_sock *inet = inet_sk(sk);
 153         const struct dccp_sock *dp = dccp_sk(sk);
 154
 155         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 156          * send out by Linux are always < 576bytes so they should go through
 157          * unfragmented).
 158          */
 159         if (sk->sk_state == DCCP_LISTEN)
 160                 return;
 161
 162         dst = inet_csk_update_pmtu(sk, mtu);
 163         if (!dst)
 164                 return;
 165
 166         /* Something is about to be wrong... Remember soft error
 167          * for the case, if this connection will not able to recover.
 168          */
 169         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 170                 sk->sk_err_soft = EMSGSIZE;
 171
 172         mtu = dst_mtu(dst);
 173
 174         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 175             ip_sk_accept_pmtu(sk) &&
 176             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 177                 dccp_sync_mss(sk, mtu);
 178
 179                 /*
 180                  * From RFC 4340, sec. 14.1:
 181                  *
 182                  *      DCCP-Sync packets are the best choice for upward
 183                  *      probing, since DCCP-Sync probes do not risk application
 184                  *      data loss.
 185                  */
 186                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 187         } /* else let the usual retransmit timer handle it */
 188 }
 189
 190 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 191 {
 192         struct dst_entry *dst = __sk_dst_check(sk, 0);
 193
 194         if (dst)
 195                 dst->ops->redirect(dst, sk, skb);
 196 }
 197
 198 void dccp_req_err(struct sock *sk, u64 seq)
 199         {
 200         struct request_sock *req = inet_reqsk(sk);
 201         struct net *net = sock_net(sk);
 202
 203         /*
 204          * ICMPs are not backlogged, hence we cannot get an established
 205          * socket here.
 206          */
 207         WARN_ON(req->sk);
 208
 209         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
 210                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 211                 reqsk_put(req);
 212         } else {
 213                 /*
 214                  * Still in RESPOND, just remove it silently.
 215                  * There is no good way to pass the error to the newly
 216                  * created socket, and POSIX does not want network
 217                  * errors returned from accept().
 218                  */
 219                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 220         }
 221 }
 222 EXPORT_SYMBOL(dccp_req_err);
 223
 224 /*
 225  * This routine is called by the ICMP module when it gets some sort of error
 226  * condition. If err < 0 then the socket should be closed and the error
 227  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 228  * After adjustment header points to the first 8 bytes of the tcp header. We
 229  * need to find the appropriate port.
 230  *
 231  * The locking strategy used here is very "optimistic". When someone else
 232  * accesses the socket the ICMP is just dropped and for some paths there is no
 233  * check at all. A more general error queue to queue errors for later handling
 234  * is probably better.
 235  */
 236 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 237 {
 238         const struct iphdr *iph = (struct iphdr *)skb->data;
 239         const u8 offset = iph->ihl << 2;
 240         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 241         struct dccp_sock *dp;
 242         struct inet_sock *inet;
 243         const int type = icmp_hdr(skb)->type;
 244         const int code = icmp_hdr(skb)->code;
 245         struct sock *sk;
 246         __u64 seq;
 247         int err;
 248         struct net *net = dev_net(skb->dev);
 249
 250         if (skb->len < offset + sizeof(*dh) ||
 251             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 252                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 253                 return;
 254         }
 255
 256         sk = __inet_lookup_established(net, &dccp_hashinfo,
 257                                        iph->daddr, dh->dccph_dport,
 258                                        iph->saddr, ntohs(dh->dccph_sport),
 259                                        inet_iif(skb));
 260         if (!sk) {
 261                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 262                 return;
 263         }
 264
 265         if (sk->sk_state == DCCP_TIME_WAIT) {
 266                 inet_twsk_put(inet_twsk(sk));
 267                 return;
 268         }
 269         seq = dccp_hdr_seq(dh);
 270         if (sk->sk_state == DCCP_NEW_SYN_RECV)
 271                 return dccp_req_err(sk, seq);
 272
 273         bh_lock_sock(sk);
 274         /* If too many ICMPs get dropped on busy
 275          * servers this needs to be solved differently.
 276          */
 277         if (sock_owned_by_user(sk))
 278                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 279
 280         if (sk->sk_state == DCCP_CLOSED)
 281                 goto out;
 282
 283         dp = dccp_sk(sk);
 284         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 285             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 286                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 287                 goto out;
 288         }
 289
 290         switch (type) {
 291         case ICMP_REDIRECT:
 292                 dccp_do_redirect(skb, sk);
 293                 goto out;
 294         case ICMP_SOURCE_QUENCH:
 295                 /* Just silently ignore these. */
 296                 goto out;
 297         case ICMP_PARAMETERPROB:
 298                 err = EPROTO;
 299                 break;
 300         case ICMP_DEST_UNREACH:
 301                 if (code > NR_ICMP_UNREACH)
 302                         goto out;
 303
 304                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 305                         if (!sock_owned_by_user(sk))
 306                                 dccp_do_pmtu_discovery(sk, iph, info);
 307                         goto out;
 308                 }
 309
 310                 err = icmp_err_convert[code].errno;
 311                 break;
 312         case ICMP_TIME_EXCEEDED:
 313                 err = EHOSTUNREACH;
 314                 break;
 315         default:
 316                 goto out;
 317         }
 318
 319         switch (sk->sk_state) {
 320         case DCCP_REQUESTING:
 321         case DCCP_RESPOND:
 322                 if (!sock_owned_by_user(sk)) {
 323                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 324                         sk->sk_err = err;
 325
 326                         sk->sk_error_report(sk);
 327
 328                         dccp_done(sk);
 329                 } else
 330                         sk->sk_err_soft = err;
 331                 goto out;
 332         }
 333
 334         /* If we've already connected we will keep trying
 335          * until we time out, or the user gives up.
 336          *
 337          * rfc1122 4.2.3.9 allows to consider as hard errors
 338          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 339          * but it is obsoleted by pmtu discovery).
 340          *
 341          * Note, that in modern internet, where routing is unreliable
 342          * and in each dark corner broken firewalls sit, sending random
 343          * errors ordered by their masters even this two messages finally lose
 344          * their original sense (even Linux sends invalid PORT_UNREACHs)
 345          *
 346          * Now we are in compliance with RFCs.
 347          *                                                      --ANK (980905)
 348          */
 349
 350         inet = inet_sk(sk);
 351         if (!sock_owned_by_user(sk) && inet->recverr) {
 352                 sk->sk_err = err;
 353                 sk->sk_error_report(sk);
 354         } else /* Only an error on timeout */
 355                 sk->sk_err_soft = err;
 356 out:
 357         bh_unlock_sock(sk);
 358         sock_put(sk);
 359 }
 360
 361 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 362                                       __be32 src, __be32 dst)
 363 {
 364         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 365 }
 366
 367 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 368 {
 369         const struct inet_sock *inet = inet_sk(sk);
 370         struct dccp_hdr *dh = dccp_hdr(skb);
 371
 372         dccp_csum_outgoing(skb);
 373         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 374                                                  inet->inet_saddr,
 375                                                  inet->inet_daddr);
 376 }
 377 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 378
 379 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 380 {
 381         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 382                                            ip_hdr(skb)->saddr,
 383                                            dccp_hdr(skb)->dccph_dport,
 384                                            dccp_hdr(skb)->dccph_sport);
 385 }
 386
 387 /*
 388  * The three way handshake has completed - we got a valid ACK or DATAACK -
 389  * now create the new socket.
 390  *
 391  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 392  */
 393 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 394                                        struct sk_buff *skb,
 395                                        struct request_sock *req,
 396                                        struct dst_entry *dst)
 397 {
 398         struct inet_request_sock *ireq;
 399         struct inet_sock *newinet;
 400         struct sock *newsk;
 401
 402         if (sk_acceptq_is_full(sk))
 403                 goto exit_overflow;
 404
 405         newsk = dccp_create_openreq_child(sk, req, skb);
 406         if (newsk == NULL)
 407                 goto exit_nonewsk;
 408
 409         newinet            = inet_sk(newsk);
 410         ireq               = inet_rsk(req);
 411         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 412         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 413         newinet->inet_saddr     = ireq->ir_loc_addr;
 414         newinet->inet_opt       = ireq->opt;
 415         ireq->opt          = NULL;
 416         newinet->mc_index  = inet_iif(skb);
 417         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 418         newinet->inet_id   = jiffies;
 419
 420         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 421                 goto put_and_exit;
 422
 423         sk_setup_caps(newsk, dst);
 424
 425         dccp_sync_mss(newsk, dst_mtu(dst));
 426
 427         if (__inet_inherit_port(sk, newsk) < 0)
 428                 goto put_and_exit;
 429         __inet_hash_nolisten(newsk, NULL);
 430
 431         return newsk;
 432
 433 exit_overflow:
 434         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 435 exit_nonewsk:
 436         dst_release(dst);
 437 exit:
 438         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 439         return NULL;
 440 put_and_exit:
 441         inet_csk_prepare_forced_close(newsk);
 442         dccp_done(newsk);
 443         goto exit;
 444 }
 445 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 446
 447 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 448 {
 449         const struct dccp_hdr *dh = dccp_hdr(skb);
 450         const struct iphdr *iph = ip_hdr(skb);
 451         struct sock *nsk;
 452         /* Find possible connection requests. */
 453         struct request_sock *req = inet_csk_search_req(sk, dh->dccph_sport,
 454                                                        iph->saddr, iph->daddr);
 455         if (req) {
 456                 nsk = dccp_check_req(sk, skb, req);
 457                 if (!nsk)
 458                         reqsk_put(req);
 459                 return nsk;
 460         }
 461         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 462                                       iph->saddr, dh->dccph_sport,
 463                                       iph->daddr, dh->dccph_dport,
 464                                       inet_iif(skb));
 465         if (nsk != NULL) {
 466                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 467                         bh_lock_sock(nsk);
 468                         return nsk;
 469                 }
 470                 inet_twsk_put(inet_twsk(nsk));
 471                 return NULL;
 472         }
 473
 474         return sk;
 475 }
 476
 477 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 478                                            struct sk_buff *skb)
 479 {
 480         struct rtable *rt;
 481         const struct iphdr *iph = ip_hdr(skb);
 482         struct flowi4 fl4 = {
 483                 .flowi4_oif = inet_iif(skb),
 484                 .daddr = iph->saddr,
 485                 .saddr = iph->daddr,
 486                 .flowi4_tos = RT_CONN_FLAGS(sk),
 487                 .flowi4_proto = sk->sk_protocol,
 488                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 489                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 490         };
 491
 492         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 493         rt = ip_route_output_flow(net, &fl4, sk);
 494         if (IS_ERR(rt)) {
 495                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 496                 return NULL;
 497         }
 498
 499         return &rt->dst;
 500 }
 501
 502 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 503 {
 504         int err = -1;
 505         struct sk_buff *skb;
 506         struct dst_entry *dst;
 507         struct flowi4 fl4;
 508
 509         dst = inet_csk_route_req(sk, &fl4, req);
 510         if (dst == NULL)
 511                 goto out;
 512
 513         skb = dccp_make_response(sk, dst, req);
 514         if (skb != NULL) {
 515                 const struct inet_request_sock *ireq = inet_rsk(req);
 516                 struct dccp_hdr *dh = dccp_hdr(skb);
 517
 518                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 519                                                               ireq->ir_rmt_addr);
 520                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 521                                             ireq->ir_rmt_addr,
 522                                             ireq->opt);
 523                 err = net_xmit_eval(err);
 524         }
 525
 526 out:
 527         dst_release(dst);
 528         return err;
 529 }
 530
 531 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 532 {
 533         int err;
 534         const struct iphdr *rxiph;
 535         struct sk_buff *skb;
 536         struct dst_entry *dst;
 537         struct net *net = dev_net(skb_dst(rxskb)->dev);
 538         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 539
 540         /* Never send a reset in response to a reset. */
 541         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 542                 return;
 543
 544         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 545                 return;
 546
 547         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 548         if (dst == NULL)
 549                 return;
 550
 551         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 552         if (skb == NULL)
 553                 goto out;
 554
 555         rxiph = ip_hdr(rxskb);
 556         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 557                                                                  rxiph->daddr);
 558         skb_dst_set(skb, dst_clone(dst));
 559
 560         bh_lock_sock(ctl_sk);
 561         err = ip_build_and_send_pkt(skb, ctl_sk,
 562                                     rxiph->daddr, rxiph->saddr, NULL);
 563         bh_unlock_sock(ctl_sk);
 564
 565         if (net_xmit_eval(err) == 0) {
 566                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 567                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 568         }
 569 out:
 570          dst_release(dst);
 571 }
 572
 573 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 574 {
 575         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 576         kfree(inet_rsk(req)->opt);
 577 }
 578
 579 void dccp_syn_ack_timeout(const struct request_sock *req)
 580 {
 581 }
 582 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 583
 584 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 585         .family         = PF_INET,
 586         .obj_size       = sizeof(struct dccp_request_sock),
 587         .rtx_syn_ack    = dccp_v4_send_response,
 588         .send_ack       = dccp_reqsk_send_ack,
 589         .destructor     = dccp_v4_reqsk_destructor,
 590         .send_reset     = dccp_v4_ctl_send_reset,
 591         .syn_ack_timeout = dccp_syn_ack_timeout,
 592 };
 593
 594 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 595 {
 596         struct inet_request_sock *ireq;
 597         struct request_sock *req;
 598         struct dccp_request_sock *dreq;
 599         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 600         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 601
 602         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 603         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 604                 return 0;       /* discard, don't send a reset here */
 605
 606         if (dccp_bad_service_code(sk, service)) {
 607                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 608                 goto drop;
 609         }
 610         /*
 611          * TW buckets are converted to open requests without
 612          * limitations, they conserve resources and peer is
 613          * evidently real one.
 614          */
 615         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 616         if (inet_csk_reqsk_queue_is_full(sk))
 617                 goto drop;
 618
 619         /*
 620          * Accept backlog is full. If we have already queued enough
 621          * of warm entries in syn queue, drop request. It is better than
 622          * clogging syn queue with openreqs with exponentially increasing
 623          * timeout.
 624          */
 625         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 626                 goto drop;
 627
 628         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk);
 629         if (req == NULL)
 630                 goto drop;
 631
 632         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 633                 goto drop_and_free;
 634
 635         dreq = dccp_rsk(req);
 636         if (dccp_parse_options(sk, dreq, skb))
 637                 goto drop_and_free;
 638
 639         if (security_inet_conn_request(sk, skb, req))
 640                 goto drop_and_free;
 641
 642         ireq = inet_rsk(req);
 643         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 644         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 645         ireq->ireq_family = AF_INET;
 646         ireq->ir_iif = sk->sk_bound_dev_if;
 647
 648         /*
 649          * Step 3: Process LISTEN state
 650          *
 651          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 652          *
 653          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 654          */
 655         dreq->dreq_isr     = dcb->dccpd_seq;
 656         dreq->dreq_gsr     = dreq->dreq_isr;
 657         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 658         dreq->dreq_gss     = dreq->dreq_iss;
 659         dreq->dreq_service = service;
 660
 661         if (dccp_v4_send_response(sk, req))
 662                 goto drop_and_free;
 663
 664         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 665         return 0;
 666
 667 drop_and_free:
 668         reqsk_free(req);
 669 drop:
 670         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 671         return -1;
 672 }
 673 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 674
 675 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 676 {
 677         struct dccp_hdr *dh = dccp_hdr(skb);
 678
 679         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 680                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 681                         goto reset;
 682                 return 0;
 683         }
 684
 685         /*
 686          *  Step 3: Process LISTEN state
 687          *       If P.type == Request or P contains a valid Init Cookie option,
 688          *            (* Must scan the packet's options to check for Init
 689          *               Cookies.  Only Init Cookies are processed here,
 690          *               however; other options are processed in Step 8.  This
 691          *               scan need only be performed if the endpoint uses Init
 692          *               Cookies *)
 693          *            (* Generate a new socket and switch to that socket *)
 694          *            Set S := new socket for this port pair
 695          *            S.state = RESPOND
 696          *            Choose S.ISS (initial seqno) or set from Init Cookies
 697          *            Initialize S.GAR := S.ISS
 698          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 699          *            Continue with S.state == RESPOND
 700          *            (* A Response packet will be generated in Step 11 *)
 701          *       Otherwise,
 702          *            Generate Reset(No Connection) unless P.type == Reset
 703          *            Drop packet and return
 704          *
 705          * NOTE: the check for the packet types is done in
 706          *       dccp_rcv_state_process
 707          */
 708         if (sk->sk_state == DCCP_LISTEN) {
 709                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 710
 711                 if (nsk == NULL)
 712                         goto discard;
 713
 714                 if (nsk != sk) {
 715                         if (dccp_child_process(sk, nsk, skb))
 716                                 goto reset;
 717                         return 0;
 718                 }
 719         }
 720
 721         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 722                 goto reset;
 723         return 0;
 724
 725 reset:
 726         dccp_v4_ctl_send_reset(sk, skb);
 727 discard:
 728         kfree_skb(skb);
 729         return 0;
 730 }
 731 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 732
 733 /**
 734  *      dccp_invalid_packet  -  check for malformed packets
 735  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 736  *      Packets that fail these checks are ignored and do not receive Resets.
 737  */
 738 int dccp_invalid_packet(struct sk_buff *skb)
 739 {
 740         const struct dccp_hdr *dh;
 741         unsigned int cscov;
 742
 743         if (skb->pkt_type != PACKET_HOST)
 744                 return 1;
 745
 746         /* If the packet is shorter than 12 bytes, drop packet and return */
 747         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 748                 DCCP_WARN("pskb_may_pull failed\n");
 749                 return 1;
 750         }
 751
 752         dh = dccp_hdr(skb);
 753
 754         /* If P.type is not understood, drop packet and return */
 755         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 756                 DCCP_WARN("invalid packet type\n");
 757                 return 1;
 758         }
 759
 760         /*
 761          * If P.Data Offset is too small for packet type, drop packet and return
 762          */
 763         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 764                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 765                 return 1;
 766         }
 767         /*
 768          * If P.Data Offset is too too large for packet, drop packet and return
 769          */
 770         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 771                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 772                 return 1;
 773         }
 774
 775         /*
 776          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 777          * has short sequence numbers), drop packet and return
 778          */
 779         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 780             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 781                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 782                           dccp_packet_name(dh->dccph_type));
 783                 return 1;
 784         }
 785
 786         /*
 787          * If P.CsCov is too large for the packet size, drop packet and return.
 788          * This must come _before_ checksumming (not as RFC 4340 suggests).
 789          */
 790         cscov = dccp_csum_coverage(skb);
 791         if (cscov > skb->len) {
 792                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 793                           dh->dccph_cscov, skb->len);
 794                 return 1;
 795         }
 796
 797         /* If header checksum is incorrect, drop packet and return.
 798          * (This step is completed in the AF-dependent functions.) */
 799         skb->csum = skb_checksum(skb, 0, cscov, 0);
 800
 801         return 0;
 802 }
 803 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 804
 805 /* this is called when real data arrives */
 806 static int dccp_v4_rcv(struct sk_buff *skb)
 807 {
 808         const struct dccp_hdr *dh;
 809         const struct iphdr *iph;
 810         struct sock *sk;
 811         int min_cov;
 812
 813         /* Step 1: Check header basics */
 814
 815         if (dccp_invalid_packet(skb))
 816                 goto discard_it;
 817
 818         iph = ip_hdr(skb);
 819         /* Step 1: If header checksum is incorrect, drop packet and return */
 820         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 821                 DCCP_WARN("dropped packet with invalid checksum\n");
 822                 goto discard_it;
 823         }
 824
 825         dh = dccp_hdr(skb);
 826
 827         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 828         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 829
 830         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 831                       dccp_packet_name(dh->dccph_type),
 832                       &iph->saddr, ntohs(dh->dccph_sport),
 833                       &iph->daddr, ntohs(dh->dccph_dport),
 834                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 835
 836         if (dccp_packet_without_ack(skb)) {
 837                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 838                 dccp_pr_debug_cat("\n");
 839         } else {
 840                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 841                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 842                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 843         }
 844
 845         /* Step 2:
 846          *      Look up flow ID in table and get corresponding socket */
 847         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 848                                dh->dccph_sport, dh->dccph_dport);
 849         /*
 850          * Step 2:
 851          *      If no socket ...
 852          */
 853         if (sk == NULL) {
 854                 dccp_pr_debug("failed to look up flow ID in table and "
 855                               "get corresponding socket\n");
 856                 goto no_dccp_socket;
 857         }
 858
 859         /*
 860          * Step 2:
 861          *      ... or S.state == TIMEWAIT,
 862          *              Generate Reset(No Connection) unless P.type == Reset
 863          *              Drop packet and return
 864          */
 865         if (sk->sk_state == DCCP_TIME_WAIT) {
 866                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 867                 inet_twsk_put(inet_twsk(sk));
 868                 goto no_dccp_socket;
 869         }
 870
 871         /*
 872          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 873          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 874          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 875          */
 876         min_cov = dccp_sk(sk)->dccps_pcrlen;
 877         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 878                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 879                               dh->dccph_cscov, min_cov);
 880                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 881                  *         options (Section 11.7) with Drop Code 0, Protocol
 882                  *         Constraints."                                     */
 883                 goto discard_and_relse;
 884         }
 885
 886         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 887                 goto discard_and_relse;
 888         nf_reset(skb);
 889
 890         return sk_receive_skb(sk, skb, 1);
 891
 892 no_dccp_socket:
 893         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 894                 goto discard_it;
 895         /*
 896          * Step 2:
 897          *      If no socket ...
 898          *              Generate Reset(No Connection) unless P.type == Reset
 899          *              Drop packet and return
 900          */
 901         if (dh->dccph_type != DCCP_PKT_RESET) {
 902                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 903                                         DCCP_RESET_CODE_NO_CONNECTION;
 904                 dccp_v4_ctl_send_reset(sk, skb);
 905         }
 906
 907 discard_it:
 908         kfree_skb(skb);
 909         return 0;
 910
 911 discard_and_relse:
 912         sock_put(sk);
 913         goto discard_it;
 914 }
 915
 916 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 917         .queue_xmit        = ip_queue_xmit,
 918         .send_check        = dccp_v4_send_check,
 919         .rebuild_header    = inet_sk_rebuild_header,
 920         .conn_request      = dccp_v4_conn_request,
 921         .syn_recv_sock     = dccp_v4_request_recv_sock,
 922         .net_header_len    = sizeof(struct iphdr),
 923         .setsockopt        = ip_setsockopt,
 924         .getsockopt        = ip_getsockopt,
 925         .addr2sockaddr     = inet_csk_addr2sockaddr,
 926         .sockaddr_len      = sizeof(struct sockaddr_in),
 927         .bind_conflict     = inet_csk_bind_conflict,
 928 #ifdef CONFIG_COMPAT
 929         .compat_setsockopt = compat_ip_setsockopt,
 930         .compat_getsockopt = compat_ip_getsockopt,
 931 #endif
 932 };
 933
 934 static int dccp_v4_init_sock(struct sock *sk)
 935 {
 936         static __u8 dccp_v4_ctl_sock_initialized;
 937         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 938
 939         if (err == 0) {
 940                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 941                         dccp_v4_ctl_sock_initialized = 1;
 942                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 943         }
 944
 945         return err;
 946 }
 947
 948 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 949         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 950 };
 951
 952 static struct proto dccp_v4_prot = {
 953         .name                   = "DCCP",
 954         .owner                  = THIS_MODULE,
 955         .close                  = dccp_close,
 956         .connect                = dccp_v4_connect,
 957         .disconnect             = dccp_disconnect,
 958         .ioctl                  = dccp_ioctl,
 959         .init                   = dccp_v4_init_sock,
 960         .setsockopt             = dccp_setsockopt,
 961         .getsockopt             = dccp_getsockopt,
 962         .sendmsg                = dccp_sendmsg,
 963         .recvmsg                = dccp_recvmsg,
 964         .backlog_rcv            = dccp_v4_do_rcv,
 965         .hash                   = inet_hash,
 966         .unhash                 = inet_unhash,
 967         .accept                 = inet_csk_accept,
 968         .get_port               = inet_csk_get_port,
 969         .shutdown               = dccp_shutdown,
 970         .destroy                = dccp_destroy_sock,
 971         .orphan_count           = &dccp_orphan_count,
 972         .max_header             = MAX_DCCP_HEADER,
 973         .obj_size               = sizeof(struct dccp_sock),
 974         .slab_flags             = SLAB_DESTROY_BY_RCU,
 975         .rsk_prot               = &dccp_request_sock_ops,
 976         .twsk_prot              = &dccp_timewait_sock_ops,
 977         .h.hashinfo             = &dccp_hashinfo,
 978 #ifdef CONFIG_COMPAT
 979         .compat_setsockopt      = compat_dccp_setsockopt,
 980         .compat_getsockopt      = compat_dccp_getsockopt,
 981 #endif
 982 };
 983
 984 static const struct net_protocol dccp_v4_protocol = {
 985         .handler        = dccp_v4_rcv,
 986         .err_handler    = dccp_v4_err,
 987         .no_policy      = 1,
 988         .netns_ok       = 1,
 989         .icmp_strict_tag_validation = 1,
 990 };
 991
 992 static const struct proto_ops inet_dccp_ops = {
 993         .family            = PF_INET,
 994         .owner             = THIS_MODULE,
 995         .release           = inet_release,
 996         .bind              = inet_bind,
 997         .connect           = inet_stream_connect,
 998         .socketpair        = sock_no_socketpair,
 999         .accept            = inet_accept,
1000         .getname           = inet_getname,
1001         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1002         .poll              = dccp_poll,
1003         .ioctl             = inet_ioctl,
1004         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1005         .listen            = inet_dccp_listen,
1006         .shutdown          = inet_shutdown,
1007         .setsockopt        = sock_common_setsockopt,
1008         .getsockopt        = sock_common_getsockopt,
1009         .sendmsg           = inet_sendmsg,
1010         .recvmsg           = sock_common_recvmsg,
1011         .mmap              = sock_no_mmap,
1012         .sendpage          = sock_no_sendpage,
1013 #ifdef CONFIG_COMPAT
1014         .compat_setsockopt = compat_sock_common_setsockopt,
1015         .compat_getsockopt = compat_sock_common_getsockopt,
1016 #endif
1017 };
1018
1019 static struct inet_protosw dccp_v4_protosw = {
1020         .type           = SOCK_DCCP,
1021         .protocol       = IPPROTO_DCCP,
1022         .prot           = &dccp_v4_prot,
1023         .ops            = &inet_dccp_ops,
1024         .flags          = INET_PROTOSW_ICSK,
1025 };
1026
1027 static int __net_init dccp_v4_init_net(struct net *net)
1028 {
1029         if (dccp_hashinfo.bhash == NULL)
1030                 return -ESOCKTNOSUPPORT;
1031
1032         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1033                                     SOCK_DCCP, IPPROTO_DCCP, net);
1034 }
1035
1036 static void __net_exit dccp_v4_exit_net(struct net *net)
1037 {
1038         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1039 }
1040
1041 static struct pernet_operations dccp_v4_ops = {
1042         .init   = dccp_v4_init_net,
1043         .exit   = dccp_v4_exit_net,
1044 };
1045
1046 static int __init dccp_v4_init(void)
1047 {
1048         int err = proto_register(&dccp_v4_prot, 1);
1049
1050         if (err != 0)
1051                 goto out;
1052
1053         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1054         if (err != 0)
1055                 goto out_proto_unregister;
1056
1057         inet_register_protosw(&dccp_v4_protosw);
1058
1059         err = register_pernet_subsys(&dccp_v4_ops);
1060         if (err)
1061                 goto out_destroy_ctl_sock;
1062 out:
1063         return err;
1064 out_destroy_ctl_sock:
1065         inet_unregister_protosw(&dccp_v4_protosw);
1066         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1067 out_proto_unregister:
1068         proto_unregister(&dccp_v4_prot);
1069         goto out;
1070 }
1071
1072 static void __exit dccp_v4_exit(void)
1073 {
1074         unregister_pernet_subsys(&dccp_v4_ops);
1075         inet_unregister_protosw(&dccp_v4_protosw);
1076         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1077         proto_unregister(&dccp_v4_prot);
1078 }
1079
1080 module_init(dccp_v4_init);
1081 module_exit(dccp_v4_exit);
1082
1083 /*
1084  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1085  * values directly, Also cover the case where the protocol is not specified,
1086  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1087  */
1088 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1089 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1090 MODULE_LICENSE("GPL");
1091 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1092 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");