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(struct sock *sk, struct sk_buff *skb,
 394                                        struct request_sock *req,
 395                                        struct dst_entry *dst)
 396 {
 397         struct inet_request_sock *ireq;
 398         struct inet_sock *newinet;
 399         struct sock *newsk;
 400
 401         if (sk_acceptq_is_full(sk))
 402                 goto exit_overflow;
 403
 404         newsk = dccp_create_openreq_child(sk, req, skb);
 405         if (newsk == NULL)
 406                 goto exit_nonewsk;
 407
 408         newinet            = inet_sk(newsk);
 409         ireq               = inet_rsk(req);
 410         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 411         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 412         newinet->inet_saddr     = ireq->ir_loc_addr;
 413         newinet->inet_opt       = ireq->opt;
 414         ireq->opt          = NULL;
 415         newinet->mc_index  = inet_iif(skb);
 416         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 417         newinet->inet_id   = jiffies;
 418
 419         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 420                 goto put_and_exit;
 421
 422         sk_setup_caps(newsk, dst);
 423
 424         dccp_sync_mss(newsk, dst_mtu(dst));
 425
 426         if (__inet_inherit_port(sk, newsk) < 0)
 427                 goto put_and_exit;
 428         __inet_hash_nolisten(newsk, NULL);
 429
 430         return newsk;
 431
 432 exit_overflow:
 433         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 434 exit_nonewsk:
 435         dst_release(dst);
 436 exit:
 437         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 438         return NULL;
 439 put_and_exit:
 440         inet_csk_prepare_forced_close(newsk);
 441         dccp_done(newsk);
 442         goto exit;
 443 }
 444 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 445
 446 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 447 {
 448         const struct dccp_hdr *dh = dccp_hdr(skb);
 449         const struct iphdr *iph = ip_hdr(skb);
 450         struct sock *nsk;
 451         /* Find possible connection requests. */
 452         struct request_sock *req = inet_csk_search_req(sk, dh->dccph_sport,
 453                                                        iph->saddr, iph->daddr);
 454         if (req) {
 455                 nsk = dccp_check_req(sk, skb, req);
 456                 reqsk_put(req);
 457                 return nsk;
 458         }
 459         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 460                                       iph->saddr, dh->dccph_sport,
 461                                       iph->daddr, dh->dccph_dport,
 462                                       inet_iif(skb));
 463         if (nsk != NULL) {
 464                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 465                         bh_lock_sock(nsk);
 466                         return nsk;
 467                 }
 468                 inet_twsk_put(inet_twsk(nsk));
 469                 return NULL;
 470         }
 471
 472         return sk;
 473 }
 474
 475 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 476                                            struct sk_buff *skb)
 477 {
 478         struct rtable *rt;
 479         const struct iphdr *iph = ip_hdr(skb);
 480         struct flowi4 fl4 = {
 481                 .flowi4_oif = inet_iif(skb),
 482                 .daddr = iph->saddr,
 483                 .saddr = iph->daddr,
 484                 .flowi4_tos = RT_CONN_FLAGS(sk),
 485                 .flowi4_proto = sk->sk_protocol,
 486                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 487                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 488         };
 489
 490         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 491         rt = ip_route_output_flow(net, &fl4, sk);
 492         if (IS_ERR(rt)) {
 493                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 494                 return NULL;
 495         }
 496
 497         return &rt->dst;
 498 }
 499
 500 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req)
 501 {
 502         int err = -1;
 503         struct sk_buff *skb;
 504         struct dst_entry *dst;
 505         struct flowi4 fl4;
 506
 507         dst = inet_csk_route_req(sk, &fl4, req);
 508         if (dst == NULL)
 509                 goto out;
 510
 511         skb = dccp_make_response(sk, dst, req);
 512         if (skb != NULL) {
 513                 const struct inet_request_sock *ireq = inet_rsk(req);
 514                 struct dccp_hdr *dh = dccp_hdr(skb);
 515
 516                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 517                                                               ireq->ir_rmt_addr);
 518                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 519                                             ireq->ir_rmt_addr,
 520                                             ireq->opt);
 521                 err = net_xmit_eval(err);
 522         }
 523
 524 out:
 525         dst_release(dst);
 526         return err;
 527 }
 528
 529 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 530 {
 531         int err;
 532         const struct iphdr *rxiph;
 533         struct sk_buff *skb;
 534         struct dst_entry *dst;
 535         struct net *net = dev_net(skb_dst(rxskb)->dev);
 536         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 537
 538         /* Never send a reset in response to a reset. */
 539         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 540                 return;
 541
 542         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 543                 return;
 544
 545         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 546         if (dst == NULL)
 547                 return;
 548
 549         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 550         if (skb == NULL)
 551                 goto out;
 552
 553         rxiph = ip_hdr(rxskb);
 554         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 555                                                                  rxiph->daddr);
 556         skb_dst_set(skb, dst_clone(dst));
 557
 558         bh_lock_sock(ctl_sk);
 559         err = ip_build_and_send_pkt(skb, ctl_sk,
 560                                     rxiph->daddr, rxiph->saddr, NULL);
 561         bh_unlock_sock(ctl_sk);
 562
 563         if (net_xmit_eval(err) == 0) {
 564                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 565                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 566         }
 567 out:
 568          dst_release(dst);
 569 }
 570
 571 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 572 {
 573         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 574         kfree(inet_rsk(req)->opt);
 575 }
 576
 577 void dccp_syn_ack_timeout(const struct request_sock *req)
 578 {
 579 }
 580 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 581
 582 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 583         .family         = PF_INET,
 584         .obj_size       = sizeof(struct dccp_request_sock),
 585         .rtx_syn_ack    = dccp_v4_send_response,
 586         .send_ack       = dccp_reqsk_send_ack,
 587         .destructor     = dccp_v4_reqsk_destructor,
 588         .send_reset     = dccp_v4_ctl_send_reset,
 589         .syn_ack_timeout = dccp_syn_ack_timeout,
 590 };
 591
 592 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 593 {
 594         struct inet_request_sock *ireq;
 595         struct request_sock *req;
 596         struct dccp_request_sock *dreq;
 597         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 598         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 599
 600         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 601         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 602                 return 0;       /* discard, don't send a reset here */
 603
 604         if (dccp_bad_service_code(sk, service)) {
 605                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 606                 goto drop;
 607         }
 608         /*
 609          * TW buckets are converted to open requests without
 610          * limitations, they conserve resources and peer is
 611          * evidently real one.
 612          */
 613         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 614         if (inet_csk_reqsk_queue_is_full(sk))
 615                 goto drop;
 616
 617         /*
 618          * Accept backlog is full. If we have already queued enough
 619          * of warm entries in syn queue, drop request. It is better than
 620          * clogging syn queue with openreqs with exponentially increasing
 621          * timeout.
 622          */
 623         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 624                 goto drop;
 625
 626         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk);
 627         if (req == NULL)
 628                 goto drop;
 629
 630         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 631                 goto drop_and_free;
 632
 633         dreq = dccp_rsk(req);
 634         if (dccp_parse_options(sk, dreq, skb))
 635                 goto drop_and_free;
 636
 637         if (security_inet_conn_request(sk, skb, req))
 638                 goto drop_and_free;
 639
 640         ireq = inet_rsk(req);
 641         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 642         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 643         ireq->ireq_family = AF_INET;
 644         ireq->ir_iif = sk->sk_bound_dev_if;
 645
 646         /*
 647          * Step 3: Process LISTEN state
 648          *
 649          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 650          *
 651          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 652          */
 653         dreq->dreq_isr     = dcb->dccpd_seq;
 654         dreq->dreq_gsr     = dreq->dreq_isr;
 655         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 656         dreq->dreq_gss     = dreq->dreq_iss;
 657         dreq->dreq_service = service;
 658
 659         if (dccp_v4_send_response(sk, req))
 660                 goto drop_and_free;
 661
 662         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 663         return 0;
 664
 665 drop_and_free:
 666         reqsk_free(req);
 667 drop:
 668         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 669         return -1;
 670 }
 671 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 672
 673 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 674 {
 675         struct dccp_hdr *dh = dccp_hdr(skb);
 676
 677         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 678                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 679                         goto reset;
 680                 return 0;
 681         }
 682
 683         /*
 684          *  Step 3: Process LISTEN state
 685          *       If P.type == Request or P contains a valid Init Cookie option,
 686          *            (* Must scan the packet's options to check for Init
 687          *               Cookies.  Only Init Cookies are processed here,
 688          *               however; other options are processed in Step 8.  This
 689          *               scan need only be performed if the endpoint uses Init
 690          *               Cookies *)
 691          *            (* Generate a new socket and switch to that socket *)
 692          *            Set S := new socket for this port pair
 693          *            S.state = RESPOND
 694          *            Choose S.ISS (initial seqno) or set from Init Cookies
 695          *            Initialize S.GAR := S.ISS
 696          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 697          *            Continue with S.state == RESPOND
 698          *            (* A Response packet will be generated in Step 11 *)
 699          *       Otherwise,
 700          *            Generate Reset(No Connection) unless P.type == Reset
 701          *            Drop packet and return
 702          *
 703          * NOTE: the check for the packet types is done in
 704          *       dccp_rcv_state_process
 705          */
 706         if (sk->sk_state == DCCP_LISTEN) {
 707                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 708
 709                 if (nsk == NULL)
 710                         goto discard;
 711
 712                 if (nsk != sk) {
 713                         if (dccp_child_process(sk, nsk, skb))
 714                                 goto reset;
 715                         return 0;
 716                 }
 717         }
 718
 719         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 720                 goto reset;
 721         return 0;
 722
 723 reset:
 724         dccp_v4_ctl_send_reset(sk, skb);
 725 discard:
 726         kfree_skb(skb);
 727         return 0;
 728 }
 729 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 730
 731 /**
 732  *      dccp_invalid_packet  -  check for malformed packets
 733  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 734  *      Packets that fail these checks are ignored and do not receive Resets.
 735  */
 736 int dccp_invalid_packet(struct sk_buff *skb)
 737 {
 738         const struct dccp_hdr *dh;
 739         unsigned int cscov;
 740
 741         if (skb->pkt_type != PACKET_HOST)
 742                 return 1;
 743
 744         /* If the packet is shorter than 12 bytes, drop packet and return */
 745         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 746                 DCCP_WARN("pskb_may_pull failed\n");
 747                 return 1;
 748         }
 749
 750         dh = dccp_hdr(skb);
 751
 752         /* If P.type is not understood, drop packet and return */
 753         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 754                 DCCP_WARN("invalid packet type\n");
 755                 return 1;
 756         }
 757
 758         /*
 759          * If P.Data Offset is too small for packet type, drop packet and return
 760          */
 761         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 762                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 763                 return 1;
 764         }
 765         /*
 766          * If P.Data Offset is too too large for packet, drop packet and return
 767          */
 768         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 769                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 770                 return 1;
 771         }
 772
 773         /*
 774          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 775          * has short sequence numbers), drop packet and return
 776          */
 777         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 778             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 779                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 780                           dccp_packet_name(dh->dccph_type));
 781                 return 1;
 782         }
 783
 784         /*
 785          * If P.CsCov is too large for the packet size, drop packet and return.
 786          * This must come _before_ checksumming (not as RFC 4340 suggests).
 787          */
 788         cscov = dccp_csum_coverage(skb);
 789         if (cscov > skb->len) {
 790                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 791                           dh->dccph_cscov, skb->len);
 792                 return 1;
 793         }
 794
 795         /* If header checksum is incorrect, drop packet and return.
 796          * (This step is completed in the AF-dependent functions.) */
 797         skb->csum = skb_checksum(skb, 0, cscov, 0);
 798
 799         return 0;
 800 }
 801 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 802
 803 /* this is called when real data arrives */
 804 static int dccp_v4_rcv(struct sk_buff *skb)
 805 {
 806         const struct dccp_hdr *dh;
 807         const struct iphdr *iph;
 808         struct sock *sk;
 809         int min_cov;
 810
 811         /* Step 1: Check header basics */
 812
 813         if (dccp_invalid_packet(skb))
 814                 goto discard_it;
 815
 816         iph = ip_hdr(skb);
 817         /* Step 1: If header checksum is incorrect, drop packet and return */
 818         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 819                 DCCP_WARN("dropped packet with invalid checksum\n");
 820                 goto discard_it;
 821         }
 822
 823         dh = dccp_hdr(skb);
 824
 825         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 826         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 827
 828         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 829                       dccp_packet_name(dh->dccph_type),
 830                       &iph->saddr, ntohs(dh->dccph_sport),
 831                       &iph->daddr, ntohs(dh->dccph_dport),
 832                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 833
 834         if (dccp_packet_without_ack(skb)) {
 835                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 836                 dccp_pr_debug_cat("\n");
 837         } else {
 838                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 839                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 840                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 841         }
 842
 843         /* Step 2:
 844          *      Look up flow ID in table and get corresponding socket */
 845         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 846                                dh->dccph_sport, dh->dccph_dport);
 847         /*
 848          * Step 2:
 849          *      If no socket ...
 850          */
 851         if (sk == NULL) {
 852                 dccp_pr_debug("failed to look up flow ID in table and "
 853                               "get corresponding socket\n");
 854                 goto no_dccp_socket;
 855         }
 856
 857         /*
 858          * Step 2:
 859          *      ... or S.state == TIMEWAIT,
 860          *              Generate Reset(No Connection) unless P.type == Reset
 861          *              Drop packet and return
 862          */
 863         if (sk->sk_state == DCCP_TIME_WAIT) {
 864                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 865                 inet_twsk_put(inet_twsk(sk));
 866                 goto no_dccp_socket;
 867         }
 868
 869         /*
 870          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 871          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 872          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 873          */
 874         min_cov = dccp_sk(sk)->dccps_pcrlen;
 875         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 876                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 877                               dh->dccph_cscov, min_cov);
 878                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 879                  *         options (Section 11.7) with Drop Code 0, Protocol
 880                  *         Constraints."                                     */
 881                 goto discard_and_relse;
 882         }
 883
 884         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 885                 goto discard_and_relse;
 886         nf_reset(skb);
 887
 888         return sk_receive_skb(sk, skb, 1);
 889
 890 no_dccp_socket:
 891         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 892                 goto discard_it;
 893         /*
 894          * Step 2:
 895          *      If no socket ...
 896          *              Generate Reset(No Connection) unless P.type == Reset
 897          *              Drop packet and return
 898          */
 899         if (dh->dccph_type != DCCP_PKT_RESET) {
 900                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 901                                         DCCP_RESET_CODE_NO_CONNECTION;
 902                 dccp_v4_ctl_send_reset(sk, skb);
 903         }
 904
 905 discard_it:
 906         kfree_skb(skb);
 907         return 0;
 908
 909 discard_and_relse:
 910         sock_put(sk);
 911         goto discard_it;
 912 }
 913
 914 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 915         .queue_xmit        = ip_queue_xmit,
 916         .send_check        = dccp_v4_send_check,
 917         .rebuild_header    = inet_sk_rebuild_header,
 918         .conn_request      = dccp_v4_conn_request,
 919         .syn_recv_sock     = dccp_v4_request_recv_sock,
 920         .net_header_len    = sizeof(struct iphdr),
 921         .setsockopt        = ip_setsockopt,
 922         .getsockopt        = ip_getsockopt,
 923         .addr2sockaddr     = inet_csk_addr2sockaddr,
 924         .sockaddr_len      = sizeof(struct sockaddr_in),
 925         .bind_conflict     = inet_csk_bind_conflict,
 926 #ifdef CONFIG_COMPAT
 927         .compat_setsockopt = compat_ip_setsockopt,
 928         .compat_getsockopt = compat_ip_getsockopt,
 929 #endif
 930 };
 931
 932 static int dccp_v4_init_sock(struct sock *sk)
 933 {
 934         static __u8 dccp_v4_ctl_sock_initialized;
 935         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 936
 937         if (err == 0) {
 938                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 939                         dccp_v4_ctl_sock_initialized = 1;
 940                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 941         }
 942
 943         return err;
 944 }
 945
 946 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 947         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 948 };
 949
 950 static struct proto dccp_v4_prot = {
 951         .name                   = "DCCP",
 952         .owner                  = THIS_MODULE,
 953         .close                  = dccp_close,
 954         .connect                = dccp_v4_connect,
 955         .disconnect             = dccp_disconnect,
 956         .ioctl                  = dccp_ioctl,
 957         .init                   = dccp_v4_init_sock,
 958         .setsockopt             = dccp_setsockopt,
 959         .getsockopt             = dccp_getsockopt,
 960         .sendmsg                = dccp_sendmsg,
 961         .recvmsg                = dccp_recvmsg,
 962         .backlog_rcv            = dccp_v4_do_rcv,
 963         .hash                   = inet_hash,
 964         .unhash                 = inet_unhash,
 965         .accept                 = inet_csk_accept,
 966         .get_port               = inet_csk_get_port,
 967         .shutdown               = dccp_shutdown,
 968         .destroy                = dccp_destroy_sock,
 969         .orphan_count           = &dccp_orphan_count,
 970         .max_header             = MAX_DCCP_HEADER,
 971         .obj_size               = sizeof(struct dccp_sock),
 972         .slab_flags             = SLAB_DESTROY_BY_RCU,
 973         .rsk_prot               = &dccp_request_sock_ops,
 974         .twsk_prot              = &dccp_timewait_sock_ops,
 975         .h.hashinfo             = &dccp_hashinfo,
 976 #ifdef CONFIG_COMPAT
 977         .compat_setsockopt      = compat_dccp_setsockopt,
 978         .compat_getsockopt      = compat_dccp_getsockopt,
 979 #endif
 980 };
 981
 982 static const struct net_protocol dccp_v4_protocol = {
 983         .handler        = dccp_v4_rcv,
 984         .err_handler    = dccp_v4_err,
 985         .no_policy      = 1,
 986         .netns_ok       = 1,
 987         .icmp_strict_tag_validation = 1,
 988 };
 989
 990 static const struct proto_ops inet_dccp_ops = {
 991         .family            = PF_INET,
 992         .owner             = THIS_MODULE,
 993         .release           = inet_release,
 994         .bind              = inet_bind,
 995         .connect           = inet_stream_connect,
 996         .socketpair        = sock_no_socketpair,
 997         .accept            = inet_accept,
 998         .getname           = inet_getname,
 999         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1000         .poll              = dccp_poll,
1001         .ioctl             = inet_ioctl,
1002         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1003         .listen            = inet_dccp_listen,
1004         .shutdown          = inet_shutdown,
1005         .setsockopt        = sock_common_setsockopt,
1006         .getsockopt        = sock_common_getsockopt,
1007         .sendmsg           = inet_sendmsg,
1008         .recvmsg           = sock_common_recvmsg,
1009         .mmap              = sock_no_mmap,
1010         .sendpage          = sock_no_sendpage,
1011 #ifdef CONFIG_COMPAT
1012         .compat_setsockopt = compat_sock_common_setsockopt,
1013         .compat_getsockopt = compat_sock_common_getsockopt,
1014 #endif
1015 };
1016
1017 static struct inet_protosw dccp_v4_protosw = {
1018         .type           = SOCK_DCCP,
1019         .protocol       = IPPROTO_DCCP,
1020         .prot           = &dccp_v4_prot,
1021         .ops            = &inet_dccp_ops,
1022         .flags          = INET_PROTOSW_ICSK,
1023 };
1024
1025 static int __net_init dccp_v4_init_net(struct net *net)
1026 {
1027         if (dccp_hashinfo.bhash == NULL)
1028                 return -ESOCKTNOSUPPORT;
1029
1030         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1031                                     SOCK_DCCP, IPPROTO_DCCP, net);
1032 }
1033
1034 static void __net_exit dccp_v4_exit_net(struct net *net)
1035 {
1036         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1037 }
1038
1039 static struct pernet_operations dccp_v4_ops = {
1040         .init   = dccp_v4_init_net,
1041         .exit   = dccp_v4_exit_net,
1042 };
1043
1044 static int __init dccp_v4_init(void)
1045 {
1046         int err = proto_register(&dccp_v4_prot, 1);
1047
1048         if (err != 0)
1049                 goto out;
1050
1051         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1052         if (err != 0)
1053                 goto out_proto_unregister;
1054
1055         inet_register_protosw(&dccp_v4_protosw);
1056
1057         err = register_pernet_subsys(&dccp_v4_ops);
1058         if (err)
1059                 goto out_destroy_ctl_sock;
1060 out:
1061         return err;
1062 out_destroy_ctl_sock:
1063         inet_unregister_protosw(&dccp_v4_protosw);
1064         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1065 out_proto_unregister:
1066         proto_unregister(&dccp_v4_prot);
1067         goto out;
1068 }
1069
1070 static void __exit dccp_v4_exit(void)
1071 {
1072         unregister_pernet_subsys(&dccp_v4_ops);
1073         inet_unregister_protosw(&dccp_v4_protosw);
1074         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1075         proto_unregister(&dccp_v4_prot);
1076 }
1077
1078 module_init(dccp_v4_init);
1079 module_exit(dccp_v4_exit);
1080
1081 /*
1082  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1083  * values directly, Also cover the case where the protocol is not specified,
1084  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1085  */
1086 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1087 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1088 MODULE_LICENSE("GPL");
1089 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1090 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");