net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/dccp.h>
  14 #include <linux/icmp.h>
  15 #include <linux/slab.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/random.h>
  19
  20 #include <net/icmp.h>
  21 #include <net/inet_common.h>
  22 #include <net/inet_hashtables.h>
  23 #include <net/inet_sock.h>
  24 #include <net/protocol.h>
  25 #include <net/sock.h>
  26 #include <net/timewait_sock.h>
  27 #include <net/tcp_states.h>
  28 #include <net/xfrm.h>
  29
  30 #include "ackvec.h"
  31 #include "ccid.h"
  32 #include "dccp.h"
  33 #include "feat.h"
  34
  35 /*
  36  * The per-net dccp.v4_ctl_sk socket is used for responding to
  37  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  38  * for this socket at the initialization time.
  39  */
  40
  41 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  42 {
  43         struct inet_sock *inet = inet_sk(sk);
  44         struct dccp_sock *dp = dccp_sk(sk);
  45         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  46         __be16 orig_sport, orig_dport;
  47         struct rtable *rt;
  48         __be32 daddr, nexthop;
  49         int tmp;
  50         int err;
  51
  52         dp->dccps_role = DCCP_ROLE_CLIENT;
  53
  54         if (addr_len < sizeof(struct sockaddr_in))
  55                 return -EINVAL;
  56
  57         if (usin->sin_family != AF_INET)
  58                 return -EAFNOSUPPORT;
  59
  60         nexthop = daddr = usin->sin_addr.s_addr;
  61         if (inet->opt != NULL && inet->opt->srr) {
  62                 if (daddr == 0)
  63                         return -EINVAL;
  64                 nexthop = inet->opt->faddr;
  65         }
  66
  67         orig_sport = inet->inet_sport;
  68         orig_dport = usin->sin_port;
  69         tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
  70                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  71                                IPPROTO_DCCP,
  72                                orig_sport, orig_dport, sk, 1);
  73         if (tmp < 0)
  74                 return tmp;
  75
  76         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  77                 ip_rt_put(rt);
  78                 return -ENETUNREACH;
  79         }
  80
  81         if (inet->opt == NULL || !inet->opt->srr)
  82                 daddr = rt->rt_dst;
  83
  84         if (inet->inet_saddr == 0)
  85                 inet->inet_saddr = rt->rt_src;
  86         inet->inet_rcv_saddr = inet->inet_saddr;
  87
  88         inet->inet_dport = usin->sin_port;
  89         inet->inet_daddr = daddr;
  90
  91         inet_csk(sk)->icsk_ext_hdr_len = 0;
  92         if (inet->opt != NULL)
  93                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
  94         /*
  95          * Socket identity is still unknown (sport may be zero).
  96          * However we set state to DCCP_REQUESTING and not releasing socket
  97          * lock select source port, enter ourselves into the hash tables and
  98          * complete initialization after this.
  99          */
 100         dccp_set_state(sk, DCCP_REQUESTING);
 101         err = inet_hash_connect(&dccp_death_row, sk);
 102         if (err != 0)
 103                 goto failure;
 104
 105         err = ip_route_newports(&rt, IPPROTO_DCCP,
 106                                 orig_sport, orig_dport,
 107                                 inet->inet_sport, inet->inet_dport, sk);
 108         if (err != 0)
 109                 goto failure;
 110
 111         /* OK, now commit destination to socket.  */
 112         sk_setup_caps(sk, &rt->dst);
 113
 114         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 115                                                     inet->inet_daddr,
 116                                                     inet->inet_sport,
 117                                                     inet->inet_dport);
 118         inet->inet_id = dp->dccps_iss ^ jiffies;
 119
 120         err = dccp_connect(sk);
 121         rt = NULL;
 122         if (err != 0)
 123                 goto failure;
 124 out:
 125         return err;
 126 failure:
 127         /*
 128          * This unhashes the socket and releases the local port, if necessary.
 129          */
 130         dccp_set_state(sk, DCCP_CLOSED);
 131         ip_rt_put(rt);
 132         sk->sk_route_caps = 0;
 133         inet->inet_dport = 0;
 134         goto out;
 135 }
 136
 137 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 138
 139 /*
 140  * This routine does path mtu discovery as defined in RFC1191.
 141  */
 142 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 143                                           const struct iphdr *iph,
 144                                           u32 mtu)
 145 {
 146         struct dst_entry *dst;
 147         const struct inet_sock *inet = inet_sk(sk);
 148         const struct dccp_sock *dp = dccp_sk(sk);
 149
 150         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 151          * send out by Linux are always < 576bytes so they should go through
 152          * unfragmented).
 153          */
 154         if (sk->sk_state == DCCP_LISTEN)
 155                 return;
 156
 157         /* We don't check in the destentry if pmtu discovery is forbidden
 158          * on this route. We just assume that no packet_to_big packets
 159          * are send back when pmtu discovery is not active.
 160          * There is a small race when the user changes this flag in the
 161          * route, but I think that's acceptable.
 162          */
 163         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 164                 return;
 165
 166         dst->ops->update_pmtu(dst, mtu);
 167
 168         /* Something is about to be wrong... Remember soft error
 169          * for the case, if this connection will not able to recover.
 170          */
 171         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 172                 sk->sk_err_soft = EMSGSIZE;
 173
 174         mtu = dst_mtu(dst);
 175
 176         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 177             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 178                 dccp_sync_mss(sk, mtu);
 179
 180                 /*
 181                  * From RFC 4340, sec. 14.1:
 182                  *
 183                  *      DCCP-Sync packets are the best choice for upward
 184                  *      probing, since DCCP-Sync probes do not risk application
 185                  *      data loss.
 186                  */
 187                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 188         } /* else let the usual retransmit timer handle it */
 189 }
 190
 191 /*
 192  * This routine is called by the ICMP module when it gets some sort of error
 193  * condition. If err < 0 then the socket should be closed and the error
 194  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 195  * After adjustment header points to the first 8 bytes of the tcp header. We
 196  * need to find the appropriate port.
 197  *
 198  * The locking strategy used here is very "optimistic". When someone else
 199  * accesses the socket the ICMP is just dropped and for some paths there is no
 200  * check at all. A more general error queue to queue errors for later handling
 201  * is probably better.
 202  */
 203 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 204 {
 205         const struct iphdr *iph = (struct iphdr *)skb->data;
 206         const u8 offset = iph->ihl << 2;
 207         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 208         struct dccp_sock *dp;
 209         struct inet_sock *inet;
 210         const int type = icmp_hdr(skb)->type;
 211         const int code = icmp_hdr(skb)->code;
 212         struct sock *sk;
 213         __u64 seq;
 214         int err;
 215         struct net *net = dev_net(skb->dev);
 216
 217         if (skb->len < offset + sizeof(*dh) ||
 218             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 219                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 220                 return;
 221         }
 222
 223         sk = inet_lookup(net, &dccp_hashinfo,
 224                         iph->daddr, dh->dccph_dport,
 225                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 226         if (sk == NULL) {
 227                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 228                 return;
 229         }
 230
 231         if (sk->sk_state == DCCP_TIME_WAIT) {
 232                 inet_twsk_put(inet_twsk(sk));
 233                 return;
 234         }
 235
 236         bh_lock_sock(sk);
 237         /* If too many ICMPs get dropped on busy
 238          * servers this needs to be solved differently.
 239          */
 240         if (sock_owned_by_user(sk))
 241                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 242
 243         if (sk->sk_state == DCCP_CLOSED)
 244                 goto out;
 245
 246         dp = dccp_sk(sk);
 247         seq = dccp_hdr_seq(dh);
 248         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 249             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 250                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 251                 goto out;
 252         }
 253
 254         switch (type) {
 255         case ICMP_SOURCE_QUENCH:
 256                 /* Just silently ignore these. */
 257                 goto out;
 258         case ICMP_PARAMETERPROB:
 259                 err = EPROTO;
 260                 break;
 261         case ICMP_DEST_UNREACH:
 262                 if (code > NR_ICMP_UNREACH)
 263                         goto out;
 264
 265                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 266                         if (!sock_owned_by_user(sk))
 267                                 dccp_do_pmtu_discovery(sk, iph, info);
 268                         goto out;
 269                 }
 270
 271                 err = icmp_err_convert[code].errno;
 272                 break;
 273         case ICMP_TIME_EXCEEDED:
 274                 err = EHOSTUNREACH;
 275                 break;
 276         default:
 277                 goto out;
 278         }
 279
 280         switch (sk->sk_state) {
 281                 struct request_sock *req , **prev;
 282         case DCCP_LISTEN:
 283                 if (sock_owned_by_user(sk))
 284                         goto out;
 285                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 286                                           iph->daddr, iph->saddr);
 287                 if (!req)
 288                         goto out;
 289
 290                 /*
 291                  * ICMPs are not backlogged, hence we cannot get an established
 292                  * socket here.
 293                  */
 294                 WARN_ON(req->sk);
 295
 296                 if (seq != dccp_rsk(req)->dreq_iss) {
 297                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 298                         goto out;
 299                 }
 300                 /*
 301                  * Still in RESPOND, just remove it silently.
 302                  * There is no good way to pass the error to the newly
 303                  * created socket, and POSIX does not want network
 304                  * errors returned from accept().
 305                  */
 306                 inet_csk_reqsk_queue_drop(sk, req, prev);
 307                 goto out;
 308
 309         case DCCP_REQUESTING:
 310         case DCCP_RESPOND:
 311                 if (!sock_owned_by_user(sk)) {
 312                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 313                         sk->sk_err = err;
 314
 315                         sk->sk_error_report(sk);
 316
 317                         dccp_done(sk);
 318                 } else
 319                         sk->sk_err_soft = err;
 320                 goto out;
 321         }
 322
 323         /* If we've already connected we will keep trying
 324          * until we time out, or the user gives up.
 325          *
 326          * rfc1122 4.2.3.9 allows to consider as hard errors
 327          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 328          * but it is obsoleted by pmtu discovery).
 329          *
 330          * Note, that in modern internet, where routing is unreliable
 331          * and in each dark corner broken firewalls sit, sending random
 332          * errors ordered by their masters even this two messages finally lose
 333          * their original sense (even Linux sends invalid PORT_UNREACHs)
 334          *
 335          * Now we are in compliance with RFCs.
 336          *                                                      --ANK (980905)
 337          */
 338
 339         inet = inet_sk(sk);
 340         if (!sock_owned_by_user(sk) && inet->recverr) {
 341                 sk->sk_err = err;
 342                 sk->sk_error_report(sk);
 343         } else /* Only an error on timeout */
 344                 sk->sk_err_soft = err;
 345 out:
 346         bh_unlock_sock(sk);
 347         sock_put(sk);
 348 }
 349
 350 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 351                                       __be32 src, __be32 dst)
 352 {
 353         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 354 }
 355
 356 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 357 {
 358         const struct inet_sock *inet = inet_sk(sk);
 359         struct dccp_hdr *dh = dccp_hdr(skb);
 360
 361         dccp_csum_outgoing(skb);
 362         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 363                                                  inet->inet_saddr,
 364                                                  inet->inet_daddr);
 365 }
 366
 367 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 368
 369 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 370 {
 371         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 372                                            ip_hdr(skb)->saddr,
 373                                            dccp_hdr(skb)->dccph_dport,
 374                                            dccp_hdr(skb)->dccph_sport);
 375 }
 376
 377 /*
 378  * The three way handshake has completed - we got a valid ACK or DATAACK -
 379  * now create the new socket.
 380  *
 381  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 382  */
 383 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 384                                        struct request_sock *req,
 385                                        struct dst_entry *dst)
 386 {
 387         struct inet_request_sock *ireq;
 388         struct inet_sock *newinet;
 389         struct sock *newsk;
 390
 391         if (sk_acceptq_is_full(sk))
 392                 goto exit_overflow;
 393
 394         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 395                 goto exit;
 396
 397         newsk = dccp_create_openreq_child(sk, req, skb);
 398         if (newsk == NULL)
 399                 goto exit_nonewsk;
 400
 401         sk_setup_caps(newsk, dst);
 402
 403         newinet            = inet_sk(newsk);
 404         ireq               = inet_rsk(req);
 405         newinet->inet_daddr     = ireq->rmt_addr;
 406         newinet->inet_rcv_saddr = ireq->loc_addr;
 407         newinet->inet_saddr     = ireq->loc_addr;
 408         newinet->opt       = ireq->opt;
 409         ireq->opt          = NULL;
 410         newinet->mc_index  = inet_iif(skb);
 411         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 412         newinet->inet_id   = jiffies;
 413
 414         dccp_sync_mss(newsk, dst_mtu(dst));
 415
 416         if (__inet_inherit_port(sk, newsk) < 0) {
 417                 sock_put(newsk);
 418                 goto exit;
 419         }
 420         __inet_hash_nolisten(newsk, NULL);
 421
 422         return newsk;
 423
 424 exit_overflow:
 425         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 426 exit_nonewsk:
 427         dst_release(dst);
 428 exit:
 429         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 430         return NULL;
 431 }
 432
 433 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 434
 435 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 436 {
 437         const struct dccp_hdr *dh = dccp_hdr(skb);
 438         const struct iphdr *iph = ip_hdr(skb);
 439         struct sock *nsk;
 440         struct request_sock **prev;
 441         /* Find possible connection requests. */
 442         struct request_sock *req = inet_csk_search_req(sk, &prev,
 443                                                        dh->dccph_sport,
 444                                                        iph->saddr, iph->daddr);
 445         if (req != NULL)
 446                 return dccp_check_req(sk, skb, req, prev);
 447
 448         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 449                                       iph->saddr, dh->dccph_sport,
 450                                       iph->daddr, dh->dccph_dport,
 451                                       inet_iif(skb));
 452         if (nsk != NULL) {
 453                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 454                         bh_lock_sock(nsk);
 455                         return nsk;
 456                 }
 457                 inet_twsk_put(inet_twsk(nsk));
 458                 return NULL;
 459         }
 460
 461         return sk;
 462 }
 463
 464 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 465                                            struct sk_buff *skb)
 466 {
 467         struct rtable *rt;
 468         struct flowi fl = { .oif = skb_rtable(skb)->rt_iif,
 469                             .fl4_dst = ip_hdr(skb)->saddr,
 470                             .fl4_src = ip_hdr(skb)->daddr,
 471                             .fl4_tos = RT_CONN_FLAGS(sk),
 472                             .proto = sk->sk_protocol,
 473                             .fl_ip_sport = dccp_hdr(skb)->dccph_dport,
 474                             .fl_ip_dport = dccp_hdr(skb)->dccph_sport
 475                           };
 476
 477         security_skb_classify_flow(skb, &fl);
 478         if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
 479                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 480                 return NULL;
 481         }
 482
 483         return &rt->dst;
 484 }
 485
 486 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 487                                  struct request_values *rv_unused)
 488 {
 489         int err = -1;
 490         struct sk_buff *skb;
 491         struct dst_entry *dst;
 492
 493         dst = inet_csk_route_req(sk, req);
 494         if (dst == NULL)
 495                 goto out;
 496
 497         skb = dccp_make_response(sk, dst, req);
 498         if (skb != NULL) {
 499                 const struct inet_request_sock *ireq = inet_rsk(req);
 500                 struct dccp_hdr *dh = dccp_hdr(skb);
 501
 502                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 503                                                               ireq->rmt_addr);
 504                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 505                                             ireq->rmt_addr,
 506                                             ireq->opt);
 507                 err = net_xmit_eval(err);
 508         }
 509
 510 out:
 511         dst_release(dst);
 512         return err;
 513 }
 514
 515 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 516 {
 517         int err;
 518         const struct iphdr *rxiph;
 519         struct sk_buff *skb;
 520         struct dst_entry *dst;
 521         struct net *net = dev_net(skb_dst(rxskb)->dev);
 522         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 523
 524         /* Never send a reset in response to a reset. */
 525         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 526                 return;
 527
 528         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 529                 return;
 530
 531         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 532         if (dst == NULL)
 533                 return;
 534
 535         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 536         if (skb == NULL)
 537                 goto out;
 538
 539         rxiph = ip_hdr(rxskb);
 540         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 541                                                                  rxiph->daddr);
 542         skb_dst_set(skb, dst_clone(dst));
 543
 544         bh_lock_sock(ctl_sk);
 545         err = ip_build_and_send_pkt(skb, ctl_sk,
 546                                     rxiph->daddr, rxiph->saddr, NULL);
 547         bh_unlock_sock(ctl_sk);
 548
 549         if (net_xmit_eval(err) == 0) {
 550                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 551                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 552         }
 553 out:
 554          dst_release(dst);
 555 }
 556
 557 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 558 {
 559         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 560         kfree(inet_rsk(req)->opt);
 561 }
 562
 563 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 564         .family         = PF_INET,
 565         .obj_size       = sizeof(struct dccp_request_sock),
 566         .rtx_syn_ack    = dccp_v4_send_response,
 567         .send_ack       = dccp_reqsk_send_ack,
 568         .destructor     = dccp_v4_reqsk_destructor,
 569         .send_reset     = dccp_v4_ctl_send_reset,
 570 };
 571
 572 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 573 {
 574         struct inet_request_sock *ireq;
 575         struct request_sock *req;
 576         struct dccp_request_sock *dreq;
 577         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 578         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 579
 580         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 581         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 582                 return 0;       /* discard, don't send a reset here */
 583
 584         if (dccp_bad_service_code(sk, service)) {
 585                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 586                 goto drop;
 587         }
 588         /*
 589          * TW buckets are converted to open requests without
 590          * limitations, they conserve resources and peer is
 591          * evidently real one.
 592          */
 593         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 594         if (inet_csk_reqsk_queue_is_full(sk))
 595                 goto drop;
 596
 597         /*
 598          * Accept backlog is full. If we have already queued enough
 599          * of warm entries in syn queue, drop request. It is better than
 600          * clogging syn queue with openreqs with exponentially increasing
 601          * timeout.
 602          */
 603         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 604                 goto drop;
 605
 606         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 607         if (req == NULL)
 608                 goto drop;
 609
 610         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 611                 goto drop_and_free;
 612
 613         dreq = dccp_rsk(req);
 614         if (dccp_parse_options(sk, dreq, skb))
 615                 goto drop_and_free;
 616
 617         if (security_inet_conn_request(sk, skb, req))
 618                 goto drop_and_free;
 619
 620         ireq = inet_rsk(req);
 621         ireq->loc_addr = ip_hdr(skb)->daddr;
 622         ireq->rmt_addr = ip_hdr(skb)->saddr;
 623
 624         /*
 625          * Step 3: Process LISTEN state
 626          *
 627          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 628          *
 629          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 630          * dccp_create_openreq_child.
 631          */
 632         dreq->dreq_isr     = dcb->dccpd_seq;
 633         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 634         dreq->dreq_service = service;
 635
 636         if (dccp_v4_send_response(sk, req, NULL))
 637                 goto drop_and_free;
 638
 639         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 640         return 0;
 641
 642 drop_and_free:
 643         reqsk_free(req);
 644 drop:
 645         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 646         return -1;
 647 }
 648
 649 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 650
 651 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 652 {
 653         struct dccp_hdr *dh = dccp_hdr(skb);
 654
 655         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 656                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 657                         goto reset;
 658                 return 0;
 659         }
 660
 661         /*
 662          *  Step 3: Process LISTEN state
 663          *       If P.type == Request or P contains a valid Init Cookie option,
 664          *            (* Must scan the packet's options to check for Init
 665          *               Cookies.  Only Init Cookies are processed here,
 666          *               however; other options are processed in Step 8.  This
 667          *               scan need only be performed if the endpoint uses Init
 668          *               Cookies *)
 669          *            (* Generate a new socket and switch to that socket *)
 670          *            Set S := new socket for this port pair
 671          *            S.state = RESPOND
 672          *            Choose S.ISS (initial seqno) or set from Init Cookies
 673          *            Initialize S.GAR := S.ISS
 674          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 675          *            Continue with S.state == RESPOND
 676          *            (* A Response packet will be generated in Step 11 *)
 677          *       Otherwise,
 678          *            Generate Reset(No Connection) unless P.type == Reset
 679          *            Drop packet and return
 680          *
 681          * NOTE: the check for the packet types is done in
 682          *       dccp_rcv_state_process
 683          */
 684         if (sk->sk_state == DCCP_LISTEN) {
 685                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 686
 687                 if (nsk == NULL)
 688                         goto discard;
 689
 690                 if (nsk != sk) {
 691                         if (dccp_child_process(sk, nsk, skb))
 692                                 goto reset;
 693                         return 0;
 694                 }
 695         }
 696
 697         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 698                 goto reset;
 699         return 0;
 700
 701 reset:
 702         dccp_v4_ctl_send_reset(sk, skb);
 703 discard:
 704         kfree_skb(skb);
 705         return 0;
 706 }
 707
 708 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 709
 710 /**
 711  *      dccp_invalid_packet  -  check for malformed packets
 712  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 713  *      Packets that fail these checks are ignored and do not receive Resets.
 714  */
 715 int dccp_invalid_packet(struct sk_buff *skb)
 716 {
 717         const struct dccp_hdr *dh;
 718         unsigned int cscov;
 719
 720         if (skb->pkt_type != PACKET_HOST)
 721                 return 1;
 722
 723         /* If the packet is shorter than 12 bytes, drop packet and return */
 724         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 725                 DCCP_WARN("pskb_may_pull failed\n");
 726                 return 1;
 727         }
 728
 729         dh = dccp_hdr(skb);
 730
 731         /* If P.type is not understood, drop packet and return */
 732         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 733                 DCCP_WARN("invalid packet type\n");
 734                 return 1;
 735         }
 736
 737         /*
 738          * If P.Data Offset is too small for packet type, drop packet and return
 739          */
 740         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 741                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 742                 return 1;
 743         }
 744         /*
 745          * If P.Data Offset is too too large for packet, drop packet and return
 746          */
 747         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 748                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 749                 return 1;
 750         }
 751
 752         /*
 753          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 754          * has short sequence numbers), drop packet and return
 755          */
 756         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 757             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 758                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 759                           dccp_packet_name(dh->dccph_type));
 760                 return 1;
 761         }
 762
 763         /*
 764          * If P.CsCov is too large for the packet size, drop packet and return.
 765          * This must come _before_ checksumming (not as RFC 4340 suggests).
 766          */
 767         cscov = dccp_csum_coverage(skb);
 768         if (cscov > skb->len) {
 769                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 770                           dh->dccph_cscov, skb->len);
 771                 return 1;
 772         }
 773
 774         /* If header checksum is incorrect, drop packet and return.
 775          * (This step is completed in the AF-dependent functions.) */
 776         skb->csum = skb_checksum(skb, 0, cscov, 0);
 777
 778         return 0;
 779 }
 780
 781 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 782
 783 /* this is called when real data arrives */
 784 static int dccp_v4_rcv(struct sk_buff *skb)
 785 {
 786         const struct dccp_hdr *dh;
 787         const struct iphdr *iph;
 788         struct sock *sk;
 789         int min_cov;
 790
 791         /* Step 1: Check header basics */
 792
 793         if (dccp_invalid_packet(skb))
 794                 goto discard_it;
 795
 796         iph = ip_hdr(skb);
 797         /* Step 1: If header checksum is incorrect, drop packet and return */
 798         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 799                 DCCP_WARN("dropped packet with invalid checksum\n");
 800                 goto discard_it;
 801         }
 802
 803         dh = dccp_hdr(skb);
 804
 805         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 806         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 807
 808         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 809                       dccp_packet_name(dh->dccph_type),
 810                       &iph->saddr, ntohs(dh->dccph_sport),
 811                       &iph->daddr, ntohs(dh->dccph_dport),
 812                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 813
 814         if (dccp_packet_without_ack(skb)) {
 815                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 816                 dccp_pr_debug_cat("\n");
 817         } else {
 818                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 819                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 820                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 821         }
 822
 823         /* Step 2:
 824          *      Look up flow ID in table and get corresponding socket */
 825         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 826                                dh->dccph_sport, dh->dccph_dport);
 827         /*
 828          * Step 2:
 829          *      If no socket ...
 830          */
 831         if (sk == NULL) {
 832                 dccp_pr_debug("failed to look up flow ID in table and "
 833                               "get corresponding socket\n");
 834                 goto no_dccp_socket;
 835         }
 836
 837         /*
 838          * Step 2:
 839          *      ... or S.state == TIMEWAIT,
 840          *              Generate Reset(No Connection) unless P.type == Reset
 841          *              Drop packet and return
 842          */
 843         if (sk->sk_state == DCCP_TIME_WAIT) {
 844                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 845                 inet_twsk_put(inet_twsk(sk));
 846                 goto no_dccp_socket;
 847         }
 848
 849         /*
 850          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 851          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 852          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 853          */
 854         min_cov = dccp_sk(sk)->dccps_pcrlen;
 855         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 856                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 857                               dh->dccph_cscov, min_cov);
 858                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 859                  *         options (Section 11.7) with Drop Code 0, Protocol
 860                  *         Constraints."                                     */
 861                 goto discard_and_relse;
 862         }
 863
 864         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 865                 goto discard_and_relse;
 866         nf_reset(skb);
 867
 868         return sk_receive_skb(sk, skb, 1);
 869
 870 no_dccp_socket:
 871         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 872                 goto discard_it;
 873         /*
 874          * Step 2:
 875          *      If no socket ...
 876          *              Generate Reset(No Connection) unless P.type == Reset
 877          *              Drop packet and return
 878          */
 879         if (dh->dccph_type != DCCP_PKT_RESET) {
 880                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 881                                         DCCP_RESET_CODE_NO_CONNECTION;
 882                 dccp_v4_ctl_send_reset(sk, skb);
 883         }
 884
 885 discard_it:
 886         kfree_skb(skb);
 887         return 0;
 888
 889 discard_and_relse:
 890         sock_put(sk);
 891         goto discard_it;
 892 }
 893
 894 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 895         .queue_xmit        = ip_queue_xmit,
 896         .send_check        = dccp_v4_send_check,
 897         .rebuild_header    = inet_sk_rebuild_header,
 898         .conn_request      = dccp_v4_conn_request,
 899         .syn_recv_sock     = dccp_v4_request_recv_sock,
 900         .net_header_len    = sizeof(struct iphdr),
 901         .setsockopt        = ip_setsockopt,
 902         .getsockopt        = ip_getsockopt,
 903         .addr2sockaddr     = inet_csk_addr2sockaddr,
 904         .sockaddr_len      = sizeof(struct sockaddr_in),
 905         .bind_conflict     = inet_csk_bind_conflict,
 906 #ifdef CONFIG_COMPAT
 907         .compat_setsockopt = compat_ip_setsockopt,
 908         .compat_getsockopt = compat_ip_getsockopt,
 909 #endif
 910 };
 911
 912 static int dccp_v4_init_sock(struct sock *sk)
 913 {
 914         static __u8 dccp_v4_ctl_sock_initialized;
 915         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 916
 917         if (err == 0) {
 918                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 919                         dccp_v4_ctl_sock_initialized = 1;
 920                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 921         }
 922
 923         return err;
 924 }
 925
 926 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 927         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 928 };
 929
 930 static struct proto dccp_v4_prot = {
 931         .name                   = "DCCP",
 932         .owner                  = THIS_MODULE,
 933         .close                  = dccp_close,
 934         .connect                = dccp_v4_connect,
 935         .disconnect             = dccp_disconnect,
 936         .ioctl                  = dccp_ioctl,
 937         .init                   = dccp_v4_init_sock,
 938         .setsockopt             = dccp_setsockopt,
 939         .getsockopt             = dccp_getsockopt,
 940         .sendmsg                = dccp_sendmsg,
 941         .recvmsg                = dccp_recvmsg,
 942         .backlog_rcv            = dccp_v4_do_rcv,
 943         .hash                   = inet_hash,
 944         .unhash                 = inet_unhash,
 945         .accept                 = inet_csk_accept,
 946         .get_port               = inet_csk_get_port,
 947         .shutdown               = dccp_shutdown,
 948         .destroy                = dccp_destroy_sock,
 949         .orphan_count           = &dccp_orphan_count,
 950         .max_header             = MAX_DCCP_HEADER,
 951         .obj_size               = sizeof(struct dccp_sock),
 952         .slab_flags             = SLAB_DESTROY_BY_RCU,
 953         .rsk_prot               = &dccp_request_sock_ops,
 954         .twsk_prot              = &dccp_timewait_sock_ops,
 955         .h.hashinfo             = &dccp_hashinfo,
 956 #ifdef CONFIG_COMPAT
 957         .compat_setsockopt      = compat_dccp_setsockopt,
 958         .compat_getsockopt      = compat_dccp_getsockopt,
 959 #endif
 960 };
 961
 962 static const struct net_protocol dccp_v4_protocol = {
 963         .handler        = dccp_v4_rcv,
 964         .err_handler    = dccp_v4_err,
 965         .no_policy      = 1,
 966         .netns_ok       = 1,
 967 };
 968
 969 static const struct proto_ops inet_dccp_ops = {
 970         .family            = PF_INET,
 971         .owner             = THIS_MODULE,
 972         .release           = inet_release,
 973         .bind              = inet_bind,
 974         .connect           = inet_stream_connect,
 975         .socketpair        = sock_no_socketpair,
 976         .accept            = inet_accept,
 977         .getname           = inet_getname,
 978         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 979         .poll              = dccp_poll,
 980         .ioctl             = inet_ioctl,
 981         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 982         .listen            = inet_dccp_listen,
 983         .shutdown          = inet_shutdown,
 984         .setsockopt        = sock_common_setsockopt,
 985         .getsockopt        = sock_common_getsockopt,
 986         .sendmsg           = inet_sendmsg,
 987         .recvmsg           = sock_common_recvmsg,
 988         .mmap              = sock_no_mmap,
 989         .sendpage          = sock_no_sendpage,
 990 #ifdef CONFIG_COMPAT
 991         .compat_setsockopt = compat_sock_common_setsockopt,
 992         .compat_getsockopt = compat_sock_common_getsockopt,
 993 #endif
 994 };
 995
 996 static struct inet_protosw dccp_v4_protosw = {
 997         .type           = SOCK_DCCP,
 998         .protocol       = IPPROTO_DCCP,
 999         .prot           = &dccp_v4_prot,
1000         .ops            = &inet_dccp_ops,
1001         .no_check       = 0,
1002         .flags          = INET_PROTOSW_ICSK,
1003 };
1004
1005 static int __net_init dccp_v4_init_net(struct net *net)
1006 {
1007         if (dccp_hashinfo.bhash == NULL)
1008                 return -ESOCKTNOSUPPORT;
1009
1010         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1011                                     SOCK_DCCP, IPPROTO_DCCP, net);
1012 }
1013
1014 static void __net_exit dccp_v4_exit_net(struct net *net)
1015 {
1016         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1017 }
1018
1019 static struct pernet_operations dccp_v4_ops = {
1020         .init   = dccp_v4_init_net,
1021         .exit   = dccp_v4_exit_net,
1022 };
1023
1024 static int __init dccp_v4_init(void)
1025 {
1026         int err = proto_register(&dccp_v4_prot, 1);
1027
1028         if (err != 0)
1029                 goto out;
1030
1031         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1032         if (err != 0)
1033                 goto out_proto_unregister;
1034
1035         inet_register_protosw(&dccp_v4_protosw);
1036
1037         err = register_pernet_subsys(&dccp_v4_ops);
1038         if (err)
1039                 goto out_destroy_ctl_sock;
1040 out:
1041         return err;
1042 out_destroy_ctl_sock:
1043         inet_unregister_protosw(&dccp_v4_protosw);
1044         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1045 out_proto_unregister:
1046         proto_unregister(&dccp_v4_prot);
1047         goto out;
1048 }
1049
1050 static void __exit dccp_v4_exit(void)
1051 {
1052         unregister_pernet_subsys(&dccp_v4_ops);
1053         inet_unregister_protosw(&dccp_v4_protosw);
1054         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1055         proto_unregister(&dccp_v4_prot);
1056 }
1057
1058 module_init(dccp_v4_init);
1059 module_exit(dccp_v4_exit);
1060
1061 /*
1062  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1063  * values directly, Also cover the case where the protocol is not specified,
1064  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1065  */
1066 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1067 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1068 MODULE_LICENSE("GPL");
1069 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1070 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");