net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/dccp.h>
  14 #include <linux/icmp.h>
  15 #include <linux/module.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/random.h>
  18
  19 #include <net/icmp.h>
  20 #include <net/inet_common.h>
  21 #include <net/inet_hashtables.h>
  22 #include <net/inet_sock.h>
  23 #include <net/protocol.h>
  24 #include <net/sock.h>
  25 #include <net/timewait_sock.h>
  26 #include <net/tcp_states.h>
  27 #include <net/xfrm.h>
  28
  29 #include "ackvec.h"
  30 #include "ccid.h"
  31 #include "dccp.h"
  32 #include "feat.h"
  33
  34 /*
  35  * This is the global socket data structure used for responding to
  36  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  37  * for this socket at the initialization time.
  38  */
  39 static struct socket *dccp_v4_ctl_socket;
  40
  41 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  42 {
  43         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  44                                  inet_csk_bind_conflict);
  45 }
  46
  47 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  48 {
  49         struct inet_sock *inet = inet_sk(sk);
  50         struct dccp_sock *dp = dccp_sk(sk);
  51         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  52         struct rtable *rt;
  53         __be32 daddr, nexthop;
  54         int tmp;
  55         int err;
  56
  57         dp->dccps_role = DCCP_ROLE_CLIENT;
  58
  59         if (addr_len < sizeof(struct sockaddr_in))
  60                 return -EINVAL;
  61
  62         if (usin->sin_family != AF_INET)
  63                 return -EAFNOSUPPORT;
  64
  65         nexthop = daddr = usin->sin_addr.s_addr;
  66         if (inet->opt != NULL && inet->opt->srr) {
  67                 if (daddr == 0)
  68                         return -EINVAL;
  69                 nexthop = inet->opt->faddr;
  70         }
  71
  72         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
  73                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  74                                IPPROTO_DCCP,
  75                                inet->sport, usin->sin_port, sk);
  76         if (tmp < 0)
  77                 return tmp;
  78
  79         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  80                 ip_rt_put(rt);
  81                 return -ENETUNREACH;
  82         }
  83
  84         if (inet->opt == NULL || !inet->opt->srr)
  85                 daddr = rt->rt_dst;
  86
  87         if (inet->saddr == 0)
  88                 inet->saddr = rt->rt_src;
  89         inet->rcv_saddr = inet->saddr;
  90
  91         inet->dport = usin->sin_port;
  92         inet->daddr = daddr;
  93
  94         inet_csk(sk)->icsk_ext_hdr_len = 0;
  95         if (inet->opt != NULL)
  96                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
  97         /*
  98          * Socket identity is still unknown (sport may be zero).
  99          * However we set state to DCCP_REQUESTING and not releasing socket
 100          * lock select source port, enter ourselves into the hash tables and
 101          * complete initialization after this.
 102          */
 103         dccp_set_state(sk, DCCP_REQUESTING);
 104         err = inet_hash_connect(&dccp_death_row, sk);
 105         if (err != 0)
 106                 goto failure;
 107
 108         err = ip_route_newports(&rt, IPPROTO_DCCP, inet->sport, inet->dport,
 109                                 sk);
 110         if (err != 0)
 111                 goto failure;
 112
 113         /* OK, now commit destination to socket.  */
 114         sk_setup_caps(sk, &rt->u.dst);
 115
 116         dp->dccps_gar =
 117                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 118                                                             inet->daddr,
 119                                                             inet->sport,
 120                                                             usin->sin_port);
 121         dccp_update_gss(sk, dp->dccps_iss);
 122
 123         inet->id = dp->dccps_iss ^ jiffies;
 124
 125         err = dccp_connect(sk);
 126         rt = NULL;
 127         if (err != 0)
 128                 goto failure;
 129 out:
 130         return err;
 131 failure:
 132         /*
 133          * This unhashes the socket and releases the local port, if necessary.
 134          */
 135         dccp_set_state(sk, DCCP_CLOSED);
 136         ip_rt_put(rt);
 137         sk->sk_route_caps = 0;
 138         inet->dport = 0;
 139         goto out;
 140 }
 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         /* We don't check in the destentry if pmtu discovery is forbidden
 163          * on this route. We just assume that no packet_to_big packets
 164          * are send back when pmtu discovery is not active.
 165          * There is a small race when the user changes this flag in the
 166          * route, but I think that's acceptable.
 167          */
 168         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 169                 return;
 170
 171         dst->ops->update_pmtu(dst, mtu);
 172
 173         /* Something is about to be wrong... Remember soft error
 174          * for the case, if this connection will not able to recover.
 175          */
 176         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 177                 sk->sk_err_soft = EMSGSIZE;
 178
 179         mtu = dst_mtu(dst);
 180
 181         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 182             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 183                 dccp_sync_mss(sk, mtu);
 184
 185                 /*
 186                  * From RFC 4340, sec. 14.1:
 187                  *
 188                  *      DCCP-Sync packets are the best choice for upward
 189                  *      probing, since DCCP-Sync probes do not risk application
 190                  *      data loss.
 191                  */
 192                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 193         } /* else let the usual retransmit timer handle it */
 194 }
 195
 196 /*
 197  * This routine is called by the ICMP module when it gets some sort of error
 198  * condition. If err < 0 then the socket should be closed and the error
 199  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 200  * After adjustment header points to the first 8 bytes of the tcp header. We
 201  * need to find the appropriate port.
 202  *
 203  * The locking strategy used here is very "optimistic". When someone else
 204  * accesses the socket the ICMP is just dropped and for some paths there is no
 205  * check at all. A more general error queue to queue errors for later handling
 206  * is probably better.
 207  */
 208 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 209 {
 210         const struct iphdr *iph = (struct iphdr *)skb->data;
 211         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 212                                                         (iph->ihl << 2));
 213         struct dccp_sock *dp;
 214         struct inet_sock *inet;
 215         const int type = skb->h.icmph->type;
 216         const int code = skb->h.icmph->code;
 217         struct sock *sk;
 218         __u64 seq;
 219         int err;
 220
 221         if (skb->len < (iph->ihl << 2) + 8) {
 222                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 223                 return;
 224         }
 225
 226         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 227                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 228         if (sk == NULL) {
 229                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 230                 return;
 231         }
 232
 233         if (sk->sk_state == DCCP_TIME_WAIT) {
 234                 inet_twsk_put(inet_twsk(sk));
 235                 return;
 236         }
 237
 238         bh_lock_sock(sk);
 239         /* If too many ICMPs get dropped on busy
 240          * servers this needs to be solved differently.
 241          */
 242         if (sock_owned_by_user(sk))
 243                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 244
 245         if (sk->sk_state == DCCP_CLOSED)
 246                 goto out;
 247
 248         dp = dccp_sk(sk);
 249         seq = dccp_hdr_seq(skb);
 250         if (sk->sk_state != DCCP_LISTEN &&
 251             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 252                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 253                 goto out;
 254         }
 255
 256         switch (type) {
 257         case ICMP_SOURCE_QUENCH:
 258                 /* Just silently ignore these. */
 259                 goto out;
 260         case ICMP_PARAMETERPROB:
 261                 err = EPROTO;
 262                 break;
 263         case ICMP_DEST_UNREACH:
 264                 if (code > NR_ICMP_UNREACH)
 265                         goto out;
 266
 267                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 268                         if (!sock_owned_by_user(sk))
 269                                 dccp_do_pmtu_discovery(sk, iph, info);
 270                         goto out;
 271                 }
 272
 273                 err = icmp_err_convert[code].errno;
 274                 break;
 275         case ICMP_TIME_EXCEEDED:
 276                 err = EHOSTUNREACH;
 277                 break;
 278         default:
 279                 goto out;
 280         }
 281
 282         switch (sk->sk_state) {
 283                 struct request_sock *req , **prev;
 284         case DCCP_LISTEN:
 285                 if (sock_owned_by_user(sk))
 286                         goto out;
 287                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 288                                           iph->daddr, iph->saddr);
 289                 if (!req)
 290                         goto out;
 291
 292                 /*
 293                  * ICMPs are not backlogged, hence we cannot get an established
 294                  * socket here.
 295                  */
 296                 BUG_TRAP(!req->sk);
 297
 298                 if (seq != dccp_rsk(req)->dreq_iss) {
 299                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 300                         goto out;
 301                 }
 302                 /*
 303                  * Still in RESPOND, just remove it silently.
 304                  * There is no good way to pass the error to the newly
 305                  * created socket, and POSIX does not want network
 306                  * errors returned from accept().
 307                  */
 308                 inet_csk_reqsk_queue_drop(sk, req, prev);
 309                 goto out;
 310
 311         case DCCP_REQUESTING:
 312         case DCCP_RESPOND:
 313                 if (!sock_owned_by_user(sk)) {
 314                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 315                         sk->sk_err = err;
 316
 317                         sk->sk_error_report(sk);
 318
 319                         dccp_done(sk);
 320                 } else
 321                         sk->sk_err_soft = err;
 322                 goto out;
 323         }
 324
 325         /* If we've already connected we will keep trying
 326          * until we time out, or the user gives up.
 327          *
 328          * rfc1122 4.2.3.9 allows to consider as hard errors
 329          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 330          * but it is obsoleted by pmtu discovery).
 331          *
 332          * Note, that in modern internet, where routing is unreliable
 333          * and in each dark corner broken firewalls sit, sending random
 334          * errors ordered by their masters even this two messages finally lose
 335          * their original sense (even Linux sends invalid PORT_UNREACHs)
 336          *
 337          * Now we are in compliance with RFCs.
 338          *                                                      --ANK (980905)
 339          */
 340
 341         inet = inet_sk(sk);
 342         if (!sock_owned_by_user(sk) && inet->recverr) {
 343                 sk->sk_err = err;
 344                 sk->sk_error_report(sk);
 345         } else /* Only an error on timeout */
 346                 sk->sk_err_soft = err;
 347 out:
 348         bh_unlock_sock(sk);
 349         sock_put(sk);
 350 }
 351
 352 static inline u16 dccp_v4_csum_finish(struct sk_buff *skb,
 353                                       __be32 src, __be32 dst)
 354 {
 355         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 356 }
 357
 358 void dccp_v4_send_check(struct sock *sk, int unused, struct sk_buff *skb)
 359 {
 360         const struct inet_sock *inet = inet_sk(sk);
 361         struct dccp_hdr *dh = dccp_hdr(skb);
 362
 363         dccp_csum_outgoing(skb);
 364         dh->dccph_checksum = dccp_v4_csum_finish(skb, inet->saddr, inet->daddr);
 365 }
 366
 367 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 368
 369 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 370                                         const struct sk_buff *skb)
 371 {
 372         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 373                                            skb->nh.iph->saddr,
 374                                            dccp_hdr(skb)->dccph_dport,
 375                                            dccp_hdr(skb)->dccph_sport);
 376 }
 377
 378 /*
 379  * The three way handshake has completed - we got a valid ACK or DATAACK -
 380  * now create the new socket.
 381  *
 382  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 383  */
 384 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 385                                        struct request_sock *req,
 386                                        struct dst_entry *dst)
 387 {
 388         struct inet_request_sock *ireq;
 389         struct inet_sock *newinet;
 390         struct dccp_sock *newdp;
 391         struct sock *newsk;
 392
 393         if (sk_acceptq_is_full(sk))
 394                 goto exit_overflow;
 395
 396         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 397                 goto exit;
 398
 399         newsk = dccp_create_openreq_child(sk, req, skb);
 400         if (newsk == NULL)
 401                 goto exit;
 402
 403         sk_setup_caps(newsk, dst);
 404
 405         newdp              = dccp_sk(newsk);
 406         newinet            = inet_sk(newsk);
 407         ireq               = inet_rsk(req);
 408         newinet->daddr     = ireq->rmt_addr;
 409         newinet->rcv_saddr = ireq->loc_addr;
 410         newinet->saddr     = ireq->loc_addr;
 411         newinet->opt       = ireq->opt;
 412         ireq->opt          = NULL;
 413         newinet->mc_index  = inet_iif(skb);
 414         newinet->mc_ttl    = skb->nh.iph->ttl;
 415         newinet->id        = jiffies;
 416
 417         dccp_sync_mss(newsk, dst_mtu(dst));
 418
 419         __inet_hash(&dccp_hashinfo, newsk, 0);
 420         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 421
 422         return newsk;
 423
 424 exit_overflow:
 425         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 426 exit:
 427         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 428         dst_release(dst);
 429         return NULL;
 430 }
 431
 432 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 433
 434 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 435 {
 436         const struct dccp_hdr *dh = dccp_hdr(skb);
 437         const struct iphdr *iph = skb->nh.iph;
 438         struct sock *nsk;
 439         struct request_sock **prev;
 440         /* Find possible connection requests. */
 441         struct request_sock *req = inet_csk_search_req(sk, &prev,
 442                                                        dh->dccph_sport,
 443                                                        iph->saddr, iph->daddr);
 444         if (req != NULL)
 445                 return dccp_check_req(sk, skb, req, prev);
 446
 447         nsk = inet_lookup_established(&dccp_hashinfo,
 448                                       iph->saddr, dh->dccph_sport,
 449                                       iph->daddr, dh->dccph_dport,
 450                                       inet_iif(skb));
 451         if (nsk != NULL) {
 452                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 453                         bh_lock_sock(nsk);
 454                         return nsk;
 455                 }
 456                 inet_twsk_put(inet_twsk(nsk));
 457                 return NULL;
 458         }
 459
 460         return sk;
 461 }
 462
 463 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 464                                            struct sk_buff *skb)
 465 {
 466         struct rtable *rt;
 467         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 468                             .nl_u = { .ip4_u =
 469                                       { .daddr = skb->nh.iph->saddr,
 470                                         .saddr = skb->nh.iph->daddr,
 471                                         .tos = RT_CONN_FLAGS(sk) } },
 472                             .proto = sk->sk_protocol,
 473                             .uli_u = { .ports =
 474                                        { .sport = dccp_hdr(skb)->dccph_dport,
 475                                          .dport = dccp_hdr(skb)->dccph_sport }
 476                                      }
 477                           };
 478
 479         security_skb_classify_flow(skb, &fl);
 480         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 481                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 482                 return NULL;
 483         }
 484
 485         return &rt->u.dst;
 486 }
 487
 488 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 489                                  struct dst_entry *dst)
 490 {
 491         int err = -1;
 492         struct sk_buff *skb;
 493
 494         /* First, grab a route. */
 495
 496         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 497                 goto out;
 498
 499         skb = dccp_make_response(sk, dst, req);
 500         if (skb != NULL) {
 501                 const struct inet_request_sock *ireq = inet_rsk(req);
 502                 struct dccp_hdr *dh = dccp_hdr(skb);
 503
 504                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 505                                                               ireq->rmt_addr);
 506                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 507                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 508                                             ireq->rmt_addr,
 509                                             ireq->opt);
 510                 if (err == NET_XMIT_CN)
 511                         err = 0;
 512         }
 513
 514 out:
 515         dst_release(dst);
 516         return err;
 517 }
 518
 519 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 520 {
 521         int err;
 522         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 523         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 524                                        sizeof(struct dccp_hdr_ext) +
 525                                        sizeof(struct dccp_hdr_reset);
 526         struct sk_buff *skb;
 527         struct dst_entry *dst;
 528         u64 seqno;
 529
 530         /* Never send a reset in response to a reset. */
 531         if (rxdh->dccph_type == DCCP_PKT_RESET)
 532                 return;
 533
 534         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 535                 return;
 536
 537         dst = dccp_v4_route_skb(dccp_v4_ctl_socket->sk, rxskb);
 538         if (dst == NULL)
 539                 return;
 540
 541         skb = alloc_skb(dccp_v4_ctl_socket->sk->sk_prot->max_header,
 542                         GFP_ATOMIC);
 543         if (skb == NULL)
 544                 goto out;
 545
 546         /* Reserve space for headers. */
 547         skb_reserve(skb, dccp_v4_ctl_socket->sk->sk_prot->max_header);
 548         skb->dst = dst_clone(dst);
 549
 550         dh = dccp_zeroed_hdr(skb, dccp_hdr_reset_len);
 551
 552         /* Build DCCP header and checksum it. */
 553         dh->dccph_type     = DCCP_PKT_RESET;
 554         dh->dccph_sport    = rxdh->dccph_dport;
 555         dh->dccph_dport    = rxdh->dccph_sport;
 556         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 557         dh->dccph_x        = 1;
 558         dccp_hdr_reset(skb)->dccph_reset_code =
 559                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 560
 561         /* See "8.3.1. Abnormal Termination" in RFC 4340 */
 562         seqno = 0;
 563         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 564                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 565
 566         dccp_hdr_set_seq(dh, seqno);
 567         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 568                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 569
 570         dccp_csum_outgoing(skb);
 571         dh->dccph_checksum = dccp_v4_csum_finish(skb, rxskb->nh.iph->saddr,
 572                                                       rxskb->nh.iph->daddr);
 573
 574         bh_lock_sock(dccp_v4_ctl_socket->sk);
 575         err = ip_build_and_send_pkt(skb, dccp_v4_ctl_socket->sk,
 576                                     rxskb->nh.iph->daddr,
 577                                     rxskb->nh.iph->saddr, NULL);
 578         bh_unlock_sock(dccp_v4_ctl_socket->sk);
 579
 580         if (err == NET_XMIT_CN || err == 0) {
 581                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 582                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 583         }
 584 out:
 585          dst_release(dst);
 586 }
 587
 588 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 589 {
 590         kfree(inet_rsk(req)->opt);
 591 }
 592
 593 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 594         .family         = PF_INET,
 595         .obj_size       = sizeof(struct dccp_request_sock),
 596         .rtx_syn_ack    = dccp_v4_send_response,
 597         .send_ack       = dccp_reqsk_send_ack,
 598         .destructor     = dccp_v4_reqsk_destructor,
 599         .send_reset     = dccp_v4_ctl_send_reset,
 600 };
 601
 602 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 603 {
 604         struct inet_request_sock *ireq;
 605         struct request_sock *req;
 606         struct dccp_request_sock *dreq;
 607         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 608         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 609         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 610
 611         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 612         if (((struct rtable *)skb->dst)->rt_flags &
 613             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 614                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 615                 goto drop;
 616         }
 617
 618         if (dccp_bad_service_code(sk, service)) {
 619                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 620                 goto drop;
 621         }
 622         /*
 623          * TW buckets are converted to open requests without
 624          * limitations, they conserve resources and peer is
 625          * evidently real one.
 626          */
 627         if (inet_csk_reqsk_queue_is_full(sk))
 628                 goto drop;
 629
 630         /*
 631          * Accept backlog is full. If we have already queued enough
 632          * of warm entries in syn queue, drop request. It is better than
 633          * clogging syn queue with openreqs with exponentially increasing
 634          * timeout.
 635          */
 636         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 637                 goto drop;
 638
 639         req = reqsk_alloc(&dccp_request_sock_ops);
 640         if (req == NULL)
 641                 goto drop;
 642
 643         if (dccp_parse_options(sk, skb))
 644                 goto drop_and_free;
 645
 646         dccp_reqsk_init(req, skb);
 647
 648         if (security_inet_conn_request(sk, skb, req))
 649                 goto drop_and_free;
 650
 651         ireq = inet_rsk(req);
 652         ireq->loc_addr = skb->nh.iph->daddr;
 653         ireq->rmt_addr = skb->nh.iph->saddr;
 654         ireq->opt       = NULL;
 655
 656         /*
 657          * Step 3: Process LISTEN state
 658          *
 659          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 660          *
 661          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 662          * dccp_create_openreq_child.
 663          */
 664         dreq = dccp_rsk(req);
 665         dreq->dreq_isr     = dcb->dccpd_seq;
 666         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 667         dreq->dreq_service = service;
 668
 669         if (dccp_v4_send_response(sk, req, NULL))
 670                 goto drop_and_free;
 671
 672         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 673         return 0;
 674
 675 drop_and_free:
 676         reqsk_free(req);
 677 drop:
 678         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 679         dcb->dccpd_reset_code = reset_code;
 680         return -1;
 681 }
 682
 683 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 684
 685 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 686 {
 687         struct dccp_hdr *dh = dccp_hdr(skb);
 688
 689         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 690                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 691                         goto reset;
 692                 return 0;
 693         }
 694
 695         /*
 696          *  Step 3: Process LISTEN state
 697          *       If P.type == Request or P contains a valid Init Cookie option,
 698          *            (* Must scan the packet's options to check for Init
 699          *               Cookies.  Only Init Cookies are processed here,
 700          *               however; other options are processed in Step 8.  This
 701          *               scan need only be performed if the endpoint uses Init
 702          *               Cookies *)
 703          *            (* Generate a new socket and switch to that socket *)
 704          *            Set S := new socket for this port pair
 705          *            S.state = RESPOND
 706          *            Choose S.ISS (initial seqno) or set from Init Cookies
 707          *            Initialize S.GAR := S.ISS
 708          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 709          *            Continue with S.state == RESPOND
 710          *            (* A Response packet will be generated in Step 11 *)
 711          *       Otherwise,
 712          *            Generate Reset(No Connection) unless P.type == Reset
 713          *            Drop packet and return
 714          *
 715          * NOTE: the check for the packet types is done in
 716          *       dccp_rcv_state_process
 717          */
 718         if (sk->sk_state == DCCP_LISTEN) {
 719                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 720
 721                 if (nsk == NULL)
 722                         goto discard;
 723
 724                 if (nsk != sk) {
 725                         if (dccp_child_process(sk, nsk, skb))
 726                                 goto reset;
 727                         return 0;
 728                 }
 729         }
 730
 731         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 732                 goto reset;
 733         return 0;
 734
 735 reset:
 736         dccp_v4_ctl_send_reset(skb);
 737 discard:
 738         kfree_skb(skb);
 739         return 0;
 740 }
 741
 742 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 743
 744 int dccp_invalid_packet(struct sk_buff *skb)
 745 {
 746         const struct dccp_hdr *dh;
 747         unsigned int cscov;
 748
 749         if (skb->pkt_type != PACKET_HOST)
 750                 return 1;
 751
 752         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 753                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
 754                 return 1;
 755         }
 756
 757         dh = dccp_hdr(skb);
 758
 759         /* If the packet type is not understood, drop packet and return */
 760         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 761                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
 762                 return 1;
 763         }
 764
 765         /*
 766          * If P.Data Offset is too small for packet type, or too large for
 767          * packet, drop packet and return
 768          */
 769         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 770                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 771                                             "too small 1\n",
 772                                dh->dccph_doff);
 773                 return 1;
 774         }
 775
 776         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 777                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
 778                                             "too small 2\n",
 779                                dh->dccph_doff);
 780                 return 1;
 781         }
 782
 783         dh = dccp_hdr(skb);
 784
 785         /*
 786          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 787          * has short sequence numbers), drop packet and return
 788          */
 789         if (dh->dccph_x == 0 &&
 790             dh->dccph_type != DCCP_PKT_DATA &&
 791             dh->dccph_type != DCCP_PKT_ACK &&
 792             dh->dccph_type != DCCP_PKT_DATAACK) {
 793                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
 794                                             "nor DataAck and P.X == 0\n",
 795                                dccp_packet_name(dh->dccph_type));
 796                 return 1;
 797         }
 798
 799         /*
 800          * If P.CsCov is too large for the packet size, drop packet and return.
 801          * This must come _before_ checksumming (not as RFC 4340 suggests).
 802          */
 803         cscov = dccp_csum_coverage(skb);
 804         if (cscov > skb->len) {
 805                 LIMIT_NETDEBUG(KERN_WARNING
 806                                "DCCP: P.CsCov %u exceeds packet length %d\n",
 807                                dh->dccph_cscov, skb->len);
 808                 return 1;
 809         }
 810
 811         /* If header checksum is incorrect, drop packet and return.
 812          * (This step is completed in the AF-dependent functions.) */
 813         skb->csum = skb_checksum(skb, 0, cscov, 0);
 814
 815         return 0;
 816 }
 817
 818 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 819
 820 /* this is called when real data arrives */
 821 static int dccp_v4_rcv(struct sk_buff *skb)
 822 {
 823         const struct dccp_hdr *dh;
 824         struct sock *sk;
 825         int min_cov;
 826
 827         /* Step 1: Check header basics */
 828
 829         if (dccp_invalid_packet(skb))
 830                 goto discard_it;
 831
 832         /* Step 1: If header checksum is incorrect, drop packet and return */
 833         if (dccp_v4_csum_finish(skb, skb->nh.iph->saddr, skb->nh.iph->daddr)) {
 834                 LIMIT_NETDEBUG(KERN_WARNING
 835                                "%s: dropped packet with invalid checksum\n",
 836                                __FUNCTION__);
 837                 goto discard_it;
 838         }
 839
 840         dh = dccp_hdr(skb);
 841
 842         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
 843         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 844
 845         dccp_pr_debug("%8.8s "
 846                       "src=%u.%u.%u.%u@%-5d "
 847                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
 848                       dccp_packet_name(dh->dccph_type),
 849                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
 850                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
 851                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 852
 853         if (dccp_packet_without_ack(skb)) {
 854                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 855                 dccp_pr_debug_cat("\n");
 856         } else {
 857                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 858                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 859                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 860         }
 861
 862         /* Step 2:
 863          *      Look up flow ID in table and get corresponding socket */
 864         sk = __inet_lookup(&dccp_hashinfo,
 865                            skb->nh.iph->saddr, dh->dccph_sport,
 866                            skb->nh.iph->daddr, dh->dccph_dport,
 867                            inet_iif(skb));
 868
 869         /*
 870          * Step 2:
 871          *      If no socket ...
 872          */
 873         if (sk == NULL) {
 874                 dccp_pr_debug("failed to look up flow ID in table and "
 875                               "get corresponding socket\n");
 876                 goto no_dccp_socket;
 877         }
 878
 879         /*
 880          * Step 2:
 881          *      ... or S.state == TIMEWAIT,
 882          *              Generate Reset(No Connection) unless P.type == Reset
 883          *              Drop packet and return
 884          */
 885         if (sk->sk_state == DCCP_TIME_WAIT) {
 886                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 887                 inet_twsk_put(inet_twsk(sk));
 888                 goto no_dccp_socket;
 889         }
 890
 891         /*
 892          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 893          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 894          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 895          */
 896         min_cov = dccp_sk(sk)->dccps_pcrlen;
 897         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 898                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 899                               dh->dccph_cscov, min_cov);
 900                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 901                  *         options (Section 11.7) with Drop Code 0, Protocol
 902                  *         Constraints."                                     */
 903                 goto discard_and_relse;
 904         }
 905
 906         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 907                 goto discard_and_relse;
 908         nf_reset(skb);
 909
 910         return sk_receive_skb(sk, skb);
 911
 912 no_dccp_socket:
 913         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 914                 goto discard_it;
 915         /*
 916          * Step 2:
 917          *      If no socket ...
 918          *              Generate Reset(No Connection) unless P.type == Reset
 919          *              Drop packet and return
 920          */
 921         if (dh->dccph_type != DCCP_PKT_RESET) {
 922                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 923                                         DCCP_RESET_CODE_NO_CONNECTION;
 924                 dccp_v4_ctl_send_reset(skb);
 925         }
 926
 927 discard_it:
 928         kfree_skb(skb);
 929         return 0;
 930
 931 discard_and_relse:
 932         sock_put(sk);
 933         goto discard_it;
 934 }
 935
 936 static struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 937         .queue_xmit        = ip_queue_xmit,
 938         .send_check        = dccp_v4_send_check,
 939         .rebuild_header    = inet_sk_rebuild_header,
 940         .conn_request      = dccp_v4_conn_request,
 941         .syn_recv_sock     = dccp_v4_request_recv_sock,
 942         .net_header_len    = sizeof(struct iphdr),
 943         .setsockopt        = ip_setsockopt,
 944         .getsockopt        = ip_getsockopt,
 945         .addr2sockaddr     = inet_csk_addr2sockaddr,
 946         .sockaddr_len      = sizeof(struct sockaddr_in),
 947 #ifdef CONFIG_COMPAT
 948         .compat_setsockopt = compat_ip_setsockopt,
 949         .compat_getsockopt = compat_ip_getsockopt,
 950 #endif
 951 };
 952
 953 static int dccp_v4_init_sock(struct sock *sk)
 954 {
 955         static __u8 dccp_v4_ctl_sock_initialized;
 956         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 957
 958         if (err == 0) {
 959                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 960                         dccp_v4_ctl_sock_initialized = 1;
 961                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 962         }
 963
 964         return err;
 965 }
 966
 967 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 968         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 969 };
 970
 971 static struct proto dccp_v4_prot = {
 972         .name                   = "DCCP",
 973         .owner                  = THIS_MODULE,
 974         .close                  = dccp_close,
 975         .connect                = dccp_v4_connect,
 976         .disconnect             = dccp_disconnect,
 977         .ioctl                  = dccp_ioctl,
 978         .init                   = dccp_v4_init_sock,
 979         .setsockopt             = dccp_setsockopt,
 980         .getsockopt             = dccp_getsockopt,
 981         .sendmsg                = dccp_sendmsg,
 982         .recvmsg                = dccp_recvmsg,
 983         .backlog_rcv            = dccp_v4_do_rcv,
 984         .hash                   = dccp_hash,
 985         .unhash                 = dccp_unhash,
 986         .accept                 = inet_csk_accept,
 987         .get_port               = dccp_v4_get_port,
 988         .shutdown               = dccp_shutdown,
 989         .destroy                = dccp_destroy_sock,
 990         .orphan_count           = &dccp_orphan_count,
 991         .max_header             = MAX_DCCP_HEADER,
 992         .obj_size               = sizeof(struct dccp_sock),
 993         .rsk_prot               = &dccp_request_sock_ops,
 994         .twsk_prot              = &dccp_timewait_sock_ops,
 995 #ifdef CONFIG_COMPAT
 996         .compat_setsockopt      = compat_dccp_setsockopt,
 997         .compat_getsockopt      = compat_dccp_getsockopt,
 998 #endif
 999 };
1000
1001 static struct net_protocol dccp_v4_protocol = {
1002         .handler        = dccp_v4_rcv,
1003         .err_handler    = dccp_v4_err,
1004         .no_policy      = 1,
1005 };
1006
1007 static const struct proto_ops inet_dccp_ops = {
1008         .family            = PF_INET,
1009         .owner             = THIS_MODULE,
1010         .release           = inet_release,
1011         .bind              = inet_bind,
1012         .connect           = inet_stream_connect,
1013         .socketpair        = sock_no_socketpair,
1014         .accept            = inet_accept,
1015         .getname           = inet_getname,
1016         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1017         .poll              = dccp_poll,
1018         .ioctl             = inet_ioctl,
1019         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1020         .listen            = inet_dccp_listen,
1021         .shutdown          = inet_shutdown,
1022         .setsockopt        = sock_common_setsockopt,
1023         .getsockopt        = sock_common_getsockopt,
1024         .sendmsg           = inet_sendmsg,
1025         .recvmsg           = sock_common_recvmsg,
1026         .mmap              = sock_no_mmap,
1027         .sendpage          = sock_no_sendpage,
1028 #ifdef CONFIG_COMPAT
1029         .compat_setsockopt = compat_sock_common_setsockopt,
1030         .compat_getsockopt = compat_sock_common_getsockopt,
1031 #endif
1032 };
1033
1034 static struct inet_protosw dccp_v4_protosw = {
1035         .type           = SOCK_DCCP,
1036         .protocol       = IPPROTO_DCCP,
1037         .prot           = &dccp_v4_prot,
1038         .ops            = &inet_dccp_ops,
1039         .capability     = -1,
1040         .no_check       = 0,
1041         .flags          = INET_PROTOSW_ICSK,
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 = inet_csk_ctl_sock_create(&dccp_v4_ctl_socket, PF_INET,
1058                                        SOCK_DCCP, IPPROTO_DCCP);
1059         if (err)
1060                 goto out_unregister_protosw;
1061 out:
1062         return err;
1063 out_unregister_protosw:
1064         inet_unregister_protosw(&dccp_v4_protosw);
1065         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1066 out_proto_unregister:
1067         proto_unregister(&dccp_v4_prot);
1068         goto out;
1069 }
1070
1071 static void __exit dccp_v4_exit(void)
1072 {
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-" __stringify(PF_INET) "-proto-33-type-6");
1087 MODULE_ALIAS("net-pf-" __stringify(PF_INET) "-proto-0-type-6");
1088 MODULE_LICENSE("GPL");
1089 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1090 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");