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/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/icmp.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_hashtables.h>
  22 #include <net/sock.h>
  23 #include <net/tcp_states.h>
  24 #include <net/xfrm.h>
  25
  26 #include "ackvec.h"
  27 #include "ccid.h"
  28 #include "dccp.h"
  29
  30 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  31         .lhash_lock     = RW_LOCK_UNLOCKED,
  32         .lhash_users    = ATOMIC_INIT(0),
  33         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  34 };
  35
  36 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  37
  38 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  39 {
  40         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  41                                  inet_csk_bind_conflict);
  42 }
  43
  44 static void dccp_v4_hash(struct sock *sk)
  45 {
  46         inet_hash(&dccp_hashinfo, sk);
  47 }
  48
  49 static void dccp_v4_unhash(struct sock *sk)
  50 {
  51         inet_unhash(&dccp_hashinfo, sk);
  52 }
  53
  54 /* called with local bh disabled */
  55 static int __dccp_v4_check_established(struct sock *sk, const __u16 lport,
  56                                       struct inet_timewait_sock **twp)
  57 {
  58         struct inet_sock *inet = inet_sk(sk);
  59         const u32 daddr = inet->rcv_saddr;
  60         const u32 saddr = inet->daddr;
  61         const int dif = sk->sk_bound_dev_if;
  62         INET_ADDR_COOKIE(acookie, saddr, daddr)
  63         const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
  64         unsigned int hash = inet_ehashfn(daddr, lport, saddr, inet->dport);
  65         struct inet_ehash_bucket *head = inet_ehash_bucket(&dccp_hashinfo, hash);
  66         const struct sock *sk2;
  67         const struct hlist_node *node;
  68         struct inet_timewait_sock *tw;
  69
  70         prefetch(head->chain.first);
  71         write_lock(&head->lock);
  72
  73         /* Check TIME-WAIT sockets first. */
  74         sk_for_each(sk2, node, &(head + dccp_hashinfo.ehash_size)->chain) {
  75                 tw = inet_twsk(sk2);
  76
  77                 if (INET_TW_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
  78                         goto not_unique;
  79         }
  80         tw = NULL;
  81
  82         /* And established part... */
  83         sk_for_each(sk2, node, &head->chain) {
  84                 if (INET_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
  85                         goto not_unique;
  86         }
  87
  88         /* Must record num and sport now. Otherwise we will see
  89          * in hash table socket with a funny identity. */
  90         inet->num = lport;
  91         inet->sport = htons(lport);
  92         sk->sk_hash = hash;
  93         BUG_TRAP(sk_unhashed(sk));
  94         __sk_add_node(sk, &head->chain);
  95         sock_prot_inc_use(sk->sk_prot);
  96         write_unlock(&head->lock);
  97
  98         if (twp != NULL) {
  99                 *twp = tw;
 100                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 101         } else if (tw != NULL) {
 102                 /* Silly. Should hash-dance instead... */
 103                 inet_twsk_deschedule(tw, &dccp_death_row);
 104                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 105
 106                 inet_twsk_put(tw);
 107         }
 108
 109         return 0;
 110
 111 not_unique:
 112         write_unlock(&head->lock);
 113         return -EADDRNOTAVAIL;
 114 }
 115
 116 /*
 117  * Bind a port for a connect operation and hash it.
 118  */
 119 static int dccp_v4_hash_connect(struct sock *sk)
 120 {
 121         const unsigned short snum = inet_sk(sk)->num;
 122         struct inet_bind_hashbucket *head;
 123         struct inet_bind_bucket *tb;
 124         int ret;
 125
 126         if (snum == 0) {
 127                 int low = sysctl_local_port_range[0];
 128                 int high = sysctl_local_port_range[1];
 129                 int remaining = (high - low) + 1;
 130                 int rover = net_random() % (high - low) + low;
 131                 struct hlist_node *node;
 132                 struct inet_timewait_sock *tw = NULL;
 133
 134                 local_bh_disable();
 135                 do {
 136                         head = &dccp_hashinfo.bhash[inet_bhashfn(rover,
 137                                                     dccp_hashinfo.bhash_size)];
 138                         spin_lock(&head->lock);
 139
 140                         /* Does not bother with rcv_saddr checks,
 141                          * because the established check is already
 142                          * unique enough.
 143                          */
 144                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 145                                 if (tb->port == rover) {
 146                                         BUG_TRAP(!hlist_empty(&tb->owners));
 147                                         if (tb->fastreuse >= 0)
 148                                                 goto next_port;
 149                                         if (!__dccp_v4_check_established(sk,
 150                                                                          rover,
 151                                                                          &tw))
 152                                                 goto ok;
 153                                         goto next_port;
 154                                 }
 155                         }
 156
 157                         tb = inet_bind_bucket_create(dccp_hashinfo.bind_bucket_cachep,
 158                                                      head, rover);
 159                         if (tb == NULL) {
 160                                 spin_unlock(&head->lock);
 161                                 break;
 162                         }
 163                         tb->fastreuse = -1;
 164                         goto ok;
 165
 166                 next_port:
 167                         spin_unlock(&head->lock);
 168                         if (++rover > high)
 169                                 rover = low;
 170                 } while (--remaining > 0);
 171
 172                 local_bh_enable();
 173
 174                 return -EADDRNOTAVAIL;
 175
 176 ok:
 177                 /* All locks still held and bhs disabled */
 178                 inet_bind_hash(sk, tb, rover);
 179                 if (sk_unhashed(sk)) {
 180                         inet_sk(sk)->sport = htons(rover);
 181                         __inet_hash(&dccp_hashinfo, sk, 0);
 182                 }
 183                 spin_unlock(&head->lock);
 184
 185                 if (tw != NULL) {
 186                         inet_twsk_deschedule(tw, &dccp_death_row);
 187                         inet_twsk_put(tw);
 188                 }
 189
 190                 ret = 0;
 191                 goto out;
 192         }
 193
 194         head = &dccp_hashinfo.bhash[inet_bhashfn(snum,
 195                                                  dccp_hashinfo.bhash_size)];
 196         tb   = inet_csk(sk)->icsk_bind_hash;
 197         spin_lock_bh(&head->lock);
 198         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 199                 __inet_hash(&dccp_hashinfo, sk, 0);
 200                 spin_unlock_bh(&head->lock);
 201                 return 0;
 202         } else {
 203                 spin_unlock(&head->lock);
 204                 /* No definite answer... Walk to established hash table */
 205                 ret = __dccp_v4_check_established(sk, snum, NULL);
 206 out:
 207                 local_bh_enable();
 208                 return ret;
 209         }
 210 }
 211
 212 static int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
 213                            int addr_len)
 214 {
 215         struct inet_sock *inet = inet_sk(sk);
 216         struct dccp_sock *dp = dccp_sk(sk);
 217         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 218         struct rtable *rt;
 219         u32 daddr, nexthop;
 220         int tmp;
 221         int err;
 222
 223         dp->dccps_role = DCCP_ROLE_CLIENT;
 224
 225         if (dccp_service_not_initialized(sk))
 226                 return -EPROTO;
 227
 228         if (addr_len < sizeof(struct sockaddr_in))
 229                 return -EINVAL;
 230
 231         if (usin->sin_family != AF_INET)
 232                 return -EAFNOSUPPORT;
 233
 234         nexthop = daddr = usin->sin_addr.s_addr;
 235         if (inet->opt != NULL && inet->opt->srr) {
 236                 if (daddr == 0)
 237                         return -EINVAL;
 238                 nexthop = inet->opt->faddr;
 239         }
 240
 241         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
 242                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 243                                IPPROTO_DCCP,
 244                                inet->sport, usin->sin_port, sk);
 245         if (tmp < 0)
 246                 return tmp;
 247
 248         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 249                 ip_rt_put(rt);
 250                 return -ENETUNREACH;
 251         }
 252
 253         if (inet->opt == NULL || !inet->opt->srr)
 254                 daddr = rt->rt_dst;
 255
 256         if (inet->saddr == 0)
 257                 inet->saddr = rt->rt_src;
 258         inet->rcv_saddr = inet->saddr;
 259
 260         inet->dport = usin->sin_port;
 261         inet->daddr = daddr;
 262
 263         dp->dccps_ext_header_len = 0;
 264         if (inet->opt != NULL)
 265                 dp->dccps_ext_header_len = inet->opt->optlen;
 266         /*
 267          * Socket identity is still unknown (sport may be zero).
 268          * However we set state to DCCP_REQUESTING and not releasing socket
 269          * lock select source port, enter ourselves into the hash tables and
 270          * complete initialization after this.
 271          */
 272         dccp_set_state(sk, DCCP_REQUESTING);
 273         err = dccp_v4_hash_connect(sk);
 274         if (err != 0)
 275                 goto failure;
 276
 277         err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
 278         if (err != 0)
 279                 goto failure;
 280
 281         /* OK, now commit destination to socket.  */
 282         sk_setup_caps(sk, &rt->u.dst);
 283
 284         dp->dccps_gar =
 285                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 286                                                             inet->daddr,
 287                                                             inet->sport,
 288                                                             usin->sin_port);
 289         dccp_update_gss(sk, dp->dccps_iss);
 290
 291         /*
 292          * SWL and AWL are initially adjusted so that they are not less than
 293          * the initial Sequence Numbers received and sent, respectively:
 294          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 295          *      AWL := max(GSS - W' + 1, ISS).
 296          * These adjustments MUST be applied only at the beginning of the
 297          * connection.
 298          */
 299         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 300
 301         inet->id = dp->dccps_iss ^ jiffies;
 302
 303         err = dccp_connect(sk);
 304         rt = NULL;
 305         if (err != 0)
 306                 goto failure;
 307 out:
 308         return err;
 309 failure:
 310         /*
 311          * This unhashes the socket and releases the local port, if necessary.
 312          */
 313         dccp_set_state(sk, DCCP_CLOSED);
 314         ip_rt_put(rt);
 315         sk->sk_route_caps = 0;
 316         inet->dport = 0;
 317         goto out;
 318 }
 319
 320 /*
 321  * This routine does path mtu discovery as defined in RFC1191.
 322  */
 323 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 324                                           const struct iphdr *iph,
 325                                           u32 mtu)
 326 {
 327         struct dst_entry *dst;
 328         const struct inet_sock *inet = inet_sk(sk);
 329         const struct dccp_sock *dp = dccp_sk(sk);
 330
 331         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 332          * send out by Linux are always < 576bytes so they should go through
 333          * unfragmented).
 334          */
 335         if (sk->sk_state == DCCP_LISTEN)
 336                 return;
 337
 338         /* We don't check in the destentry if pmtu discovery is forbidden
 339          * on this route. We just assume that no packet_to_big packets
 340          * are send back when pmtu discovery is not active.
 341          * There is a small race when the user changes this flag in the
 342          * route, but I think that's acceptable.
 343          */
 344         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 345                 return;
 346
 347         dst->ops->update_pmtu(dst, mtu);
 348
 349         /* Something is about to be wrong... Remember soft error
 350          * for the case, if this connection will not able to recover.
 351          */
 352         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 353                 sk->sk_err_soft = EMSGSIZE;
 354
 355         mtu = dst_mtu(dst);
 356
 357         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 358             dp->dccps_pmtu_cookie > mtu) {
 359                 dccp_sync_mss(sk, mtu);
 360
 361                 /*
 362                  * From: draft-ietf-dccp-spec-11.txt
 363                  *
 364                  *      DCCP-Sync packets are the best choice for upward
 365                  *      probing, since DCCP-Sync probes do not risk application
 366                  *      data loss.
 367                  */
 368                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 369         } /* else let the usual retransmit timer handle it */
 370 }
 371
 372 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 373 {
 374         int err;
 375         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 376         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 377                                      sizeof(struct dccp_hdr_ext) +
 378                                      sizeof(struct dccp_hdr_ack_bits);
 379         struct sk_buff *skb;
 380
 381         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 382                 return;
 383
 384         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 385         if (skb == NULL)
 386                 return;
 387
 388         /* Reserve space for headers. */
 389         skb_reserve(skb, MAX_DCCP_HEADER);
 390
 391         skb->dst = dst_clone(rxskb->dst);
 392
 393         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 394         dh = dccp_hdr(skb);
 395         memset(dh, 0, dccp_hdr_ack_len);
 396
 397         /* Build DCCP header and checksum it. */
 398         dh->dccph_type     = DCCP_PKT_ACK;
 399         dh->dccph_sport    = rxdh->dccph_dport;
 400         dh->dccph_dport    = rxdh->dccph_sport;
 401         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 402         dh->dccph_x        = 1;
 403
 404         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 405         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 406                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 407
 408         bh_lock_sock(dccp_ctl_socket->sk);
 409         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 410                                     rxskb->nh.iph->daddr,
 411                                     rxskb->nh.iph->saddr, NULL);
 412         bh_unlock_sock(dccp_ctl_socket->sk);
 413
 414         if (err == NET_XMIT_CN || err == 0) {
 415                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 416                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 417         }
 418 }
 419
 420 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 421                                    struct request_sock *req)
 422 {
 423         dccp_v4_ctl_send_ack(skb);
 424 }
 425
 426 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 427                                  struct dst_entry *dst)
 428 {
 429         int err = -1;
 430         struct sk_buff *skb;
 431
 432         /* First, grab a route. */
 433
 434         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 435                 goto out;
 436
 437         skb = dccp_make_response(sk, dst, req);
 438         if (skb != NULL) {
 439                 const struct inet_request_sock *ireq = inet_rsk(req);
 440
 441                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 442                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 443                                             ireq->rmt_addr,
 444                                             ireq->opt);
 445                 if (err == NET_XMIT_CN)
 446                         err = 0;
 447         }
 448
 449 out:
 450         dst_release(dst);
 451         return err;
 452 }
 453
 454 /*
 455  * This routine is called by the ICMP module when it gets some sort of error
 456  * condition. If err < 0 then the socket should be closed and the error
 457  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 458  * After adjustment header points to the first 8 bytes of the tcp header. We
 459  * need to find the appropriate port.
 460  *
 461  * The locking strategy used here is very "optimistic". When someone else
 462  * accesses the socket the ICMP is just dropped and for some paths there is no
 463  * check at all. A more general error queue to queue errors for later handling
 464  * is probably better.
 465  */
 466 void dccp_v4_err(struct sk_buff *skb, u32 info)
 467 {
 468         const struct iphdr *iph = (struct iphdr *)skb->data;
 469         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 470                                                         (iph->ihl << 2));
 471         struct dccp_sock *dp;
 472         struct inet_sock *inet;
 473         const int type = skb->h.icmph->type;
 474         const int code = skb->h.icmph->code;
 475         struct sock *sk;
 476         __u64 seq;
 477         int err;
 478
 479         if (skb->len < (iph->ihl << 2) + 8) {
 480                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 481                 return;
 482         }
 483
 484         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 485                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 486         if (sk == NULL) {
 487                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 488                 return;
 489         }
 490
 491         if (sk->sk_state == DCCP_TIME_WAIT) {
 492                 inet_twsk_put((struct inet_timewait_sock *)sk);
 493                 return;
 494         }
 495
 496         bh_lock_sock(sk);
 497         /* If too many ICMPs get dropped on busy
 498          * servers this needs to be solved differently.
 499          */
 500         if (sock_owned_by_user(sk))
 501                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 502
 503         if (sk->sk_state == DCCP_CLOSED)
 504                 goto out;
 505
 506         dp = dccp_sk(sk);
 507         seq = dccp_hdr_seq(skb);
 508         if (sk->sk_state != DCCP_LISTEN &&
 509             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 510                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 511                 goto out;
 512         }
 513
 514         switch (type) {
 515         case ICMP_SOURCE_QUENCH:
 516                 /* Just silently ignore these. */
 517                 goto out;
 518         case ICMP_PARAMETERPROB:
 519                 err = EPROTO;
 520                 break;
 521         case ICMP_DEST_UNREACH:
 522                 if (code > NR_ICMP_UNREACH)
 523                         goto out;
 524
 525                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 526                         if (!sock_owned_by_user(sk))
 527                                 dccp_do_pmtu_discovery(sk, iph, info);
 528                         goto out;
 529                 }
 530
 531                 err = icmp_err_convert[code].errno;
 532                 break;
 533         case ICMP_TIME_EXCEEDED:
 534                 err = EHOSTUNREACH;
 535                 break;
 536         default:
 537                 goto out;
 538         }
 539
 540         switch (sk->sk_state) {
 541                 struct request_sock *req , **prev;
 542         case DCCP_LISTEN:
 543                 if (sock_owned_by_user(sk))
 544                         goto out;
 545                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 546                                           iph->daddr, iph->saddr);
 547                 if (!req)
 548                         goto out;
 549
 550                 /*
 551                  * ICMPs are not backlogged, hence we cannot get an established
 552                  * socket here.
 553                  */
 554                 BUG_TRAP(!req->sk);
 555
 556                 if (seq != dccp_rsk(req)->dreq_iss) {
 557                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 558                         goto out;
 559                 }
 560                 /*
 561                  * Still in RESPOND, just remove it silently.
 562                  * There is no good way to pass the error to the newly
 563                  * created socket, and POSIX does not want network
 564                  * errors returned from accept().
 565                  */
 566                 inet_csk_reqsk_queue_drop(sk, req, prev);
 567                 goto out;
 568
 569         case DCCP_REQUESTING:
 570         case DCCP_RESPOND:
 571                 if (!sock_owned_by_user(sk)) {
 572                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 573                         sk->sk_err = err;
 574
 575                         sk->sk_error_report(sk);
 576
 577                         dccp_done(sk);
 578                 } else
 579                         sk->sk_err_soft = err;
 580                 goto out;
 581         }
 582
 583         /* If we've already connected we will keep trying
 584          * until we time out, or the user gives up.
 585          *
 586          * rfc1122 4.2.3.9 allows to consider as hard errors
 587          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 588          * but it is obsoleted by pmtu discovery).
 589          *
 590          * Note, that in modern internet, where routing is unreliable
 591          * and in each dark corner broken firewalls sit, sending random
 592          * errors ordered by their masters even this two messages finally lose
 593          * their original sense (even Linux sends invalid PORT_UNREACHs)
 594          *
 595          * Now we are in compliance with RFCs.
 596          *                                                      --ANK (980905)
 597          */
 598
 599         inet = inet_sk(sk);
 600         if (!sock_owned_by_user(sk) && inet->recverr) {
 601                 sk->sk_err = err;
 602                 sk->sk_error_report(sk);
 603         } else /* Only an error on timeout */
 604                 sk->sk_err_soft = err;
 605 out:
 606         bh_unlock_sock(sk);
 607         sock_put(sk);
 608 }
 609
 610 /* This routine computes an IPv4 DCCP checksum. */
 611 static void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 612 {
 613         const struct inet_sock *inet = inet_sk(sk);
 614         struct dccp_hdr *dh = dccp_hdr(skb);
 615
 616         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 617 }
 618
 619 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
 620 {
 621         struct sk_buff *skb;
 622         /*
 623          * FIXME: what if rebuild_header fails?
 624          * Should we be doing a rebuild_header here?
 625          */
 626         int err = inet_sk_rebuild_header(sk);
 627
 628         if (err != 0)
 629                 return err;
 630
 631         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
 632         if (skb != NULL) {
 633                 const struct inet_sock *inet = inet_sk(sk);
 634
 635                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 636                 err = ip_build_and_send_pkt(skb, sk,
 637                                             inet->saddr, inet->daddr, NULL);
 638                 if (err == NET_XMIT_CN)
 639                         err = 0;
 640         }
 641
 642         return err;
 643 }
 644
 645 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 646                                         const struct sk_buff *skb)
 647 {
 648         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 649                                            skb->nh.iph->saddr,
 650                                            dccp_hdr(skb)->dccph_dport,
 651                                            dccp_hdr(skb)->dccph_sport);
 652 }
 653
 654 static inline int dccp_bad_service_code(const struct sock *sk,
 655                                         const __u32 service)
 656 {
 657         const struct dccp_sock *dp = dccp_sk(sk);
 658
 659         if (dp->dccps_service == service)
 660                 return 0;
 661         return !dccp_list_has_service(dp->dccps_service_list, service);
 662 }
 663
 664 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 665 {
 666         struct inet_request_sock *ireq;
 667         struct dccp_sock dp;
 668         struct request_sock *req;
 669         struct dccp_request_sock *dreq;
 670         const __u32 saddr = skb->nh.iph->saddr;
 671         const __u32 daddr = skb->nh.iph->daddr;
 672         const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
 673         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 674         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 675         struct dst_entry *dst = NULL;
 676
 677         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 678         if (((struct rtable *)skb->dst)->rt_flags &
 679             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 680                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 681                 goto drop;
 682         }
 683
 684         if (dccp_bad_service_code(sk, service)) {
 685                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 686                 goto drop;
 687         }
 688         /*
 689          * TW buckets are converted to open requests without
 690          * limitations, they conserve resources and peer is
 691          * evidently real one.
 692          */
 693         if (inet_csk_reqsk_queue_is_full(sk))
 694                 goto drop;
 695
 696         /*
 697          * Accept backlog is full. If we have already queued enough
 698          * of warm entries in syn queue, drop request. It is better than
 699          * clogging syn queue with openreqs with exponentially increasing
 700          * timeout.
 701          */
 702         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 703                 goto drop;
 704
 705         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 706         if (req == NULL)
 707                 goto drop;
 708
 709         /* FIXME: process options */
 710
 711         dccp_openreq_init(req, &dp, skb);
 712
 713         ireq = inet_rsk(req);
 714         ireq->loc_addr = daddr;
 715         ireq->rmt_addr = saddr;
 716         /* FIXME: Merge Aristeu's option parsing code when ready */
 717         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 718                                   right value */
 719         ireq->opt       = NULL;
 720
 721         /*
 722          * Step 3: Process LISTEN state
 723          *
 724          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 725          *
 726          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 727          * dccp_create_openreq_child.
 728          */
 729         dreq = dccp_rsk(req);
 730         dreq->dreq_isr     = dcb->dccpd_seq;
 731         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 732         dreq->dreq_service = service;
 733
 734         if (dccp_v4_send_response(sk, req, dst))
 735                 goto drop_and_free;
 736
 737         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 738         return 0;
 739
 740 drop_and_free:
 741         /*
 742          * FIXME: should be reqsk_free after implementing req->rsk_ops
 743          */
 744         __reqsk_free(req);
 745 drop:
 746         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 747         dcb->dccpd_reset_code = reset_code;
 748         return -1;
 749 }
 750
 751 /*
 752  * The three way handshake has completed - we got a valid ACK or DATAACK -
 753  * now create the new socket.
 754  *
 755  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 756  */
 757 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 758                                        struct request_sock *req,
 759                                        struct dst_entry *dst)
 760 {
 761         struct inet_request_sock *ireq;
 762         struct inet_sock *newinet;
 763         struct dccp_sock *newdp;
 764         struct sock *newsk;
 765
 766         if (sk_acceptq_is_full(sk))
 767                 goto exit_overflow;
 768
 769         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 770                 goto exit;
 771
 772         newsk = dccp_create_openreq_child(sk, req, skb);
 773         if (newsk == NULL)
 774                 goto exit;
 775
 776         sk_setup_caps(newsk, dst);
 777
 778         newdp              = dccp_sk(newsk);
 779         newinet            = inet_sk(newsk);
 780         ireq               = inet_rsk(req);
 781         newinet->daddr     = ireq->rmt_addr;
 782         newinet->rcv_saddr = ireq->loc_addr;
 783         newinet->saddr     = ireq->loc_addr;
 784         newinet->opt       = ireq->opt;
 785         ireq->opt          = NULL;
 786         newinet->mc_index  = inet_iif(skb);
 787         newinet->mc_ttl    = skb->nh.iph->ttl;
 788         newinet->id        = jiffies;
 789
 790         dccp_sync_mss(newsk, dst_mtu(dst));
 791
 792         __inet_hash(&dccp_hashinfo, newsk, 0);
 793         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 794
 795         return newsk;
 796
 797 exit_overflow:
 798         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 799 exit:
 800         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 801         dst_release(dst);
 802         return NULL;
 803 }
 804
 805 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 806 {
 807         const struct dccp_hdr *dh = dccp_hdr(skb);
 808         const struct iphdr *iph = skb->nh.iph;
 809         struct sock *nsk;
 810         struct request_sock **prev;
 811         /* Find possible connection requests. */
 812         struct request_sock *req = inet_csk_search_req(sk, &prev,
 813                                                        dh->dccph_sport,
 814                                                        iph->saddr, iph->daddr);
 815         if (req != NULL)
 816                 return dccp_check_req(sk, skb, req, prev);
 817
 818         nsk = __inet_lookup_established(&dccp_hashinfo,
 819                                         iph->saddr, dh->dccph_sport,
 820                                         iph->daddr, ntohs(dh->dccph_dport),
 821                                         inet_iif(skb));
 822         if (nsk != NULL) {
 823                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 824                         bh_lock_sock(nsk);
 825                         return nsk;
 826                 }
 827                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 828                 return NULL;
 829         }
 830
 831         return sk;
 832 }
 833
 834 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
 835                      const u32 daddr)
 836 {
 837         const struct dccp_hdr* dh = dccp_hdr(skb);
 838         int checksum_len;
 839         u32 tmp;
 840
 841         if (dh->dccph_cscov == 0)
 842                 checksum_len = skb->len;
 843         else {
 844                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 845                 checksum_len = checksum_len < skb->len ? checksum_len :
 846                                                          skb->len;
 847         }
 848
 849         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 850         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 851                                  IPPROTO_DCCP, tmp);
 852 }
 853
 854 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 855                                    const u32 saddr, const u32 daddr)
 856 {
 857         struct dccp_hdr *dh = dccp_hdr(skb);
 858         int checksum_len;
 859         u32 tmp;
 860
 861         if (dh->dccph_cscov == 0)
 862                 checksum_len = skb->len;
 863         else {
 864                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 865                 checksum_len = checksum_len < skb->len ? checksum_len :
 866                                                          skb->len;
 867         }
 868         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 869         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 870                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 871 }
 872
 873 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 874                                            struct sk_buff *skb)
 875 {
 876         struct rtable *rt;
 877         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 878                             .nl_u = { .ip4_u =
 879                                       { .daddr = skb->nh.iph->saddr,
 880                                         .saddr = skb->nh.iph->daddr,
 881                                         .tos = RT_CONN_FLAGS(sk) } },
 882                             .proto = sk->sk_protocol,
 883                             .uli_u = { .ports =
 884                                        { .sport = dccp_hdr(skb)->dccph_dport,
 885                                          .dport = dccp_hdr(skb)->dccph_sport }
 886                                      }
 887                           };
 888
 889         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 890                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 891                 return NULL;
 892         }
 893
 894         return &rt->u.dst;
 895 }
 896
 897 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 898 {
 899         int err;
 900         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 901         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 902                                        sizeof(struct dccp_hdr_ext) +
 903                                        sizeof(struct dccp_hdr_reset);
 904         struct sk_buff *skb;
 905         struct dst_entry *dst;
 906         u64 seqno;
 907
 908         /* Never send a reset in response to a reset. */
 909         if (rxdh->dccph_type == DCCP_PKT_RESET)
 910                 return;
 911
 912         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 913                 return;
 914
 915         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 916         if (dst == NULL)
 917                 return;
 918
 919         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 920         if (skb == NULL)
 921                 goto out;
 922
 923         /* Reserve space for headers. */
 924         skb_reserve(skb, MAX_DCCP_HEADER);
 925         skb->dst = dst_clone(dst);
 926
 927         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 928         dh = dccp_hdr(skb);
 929         memset(dh, 0, dccp_hdr_reset_len);
 930
 931         /* Build DCCP header and checksum it. */
 932         dh->dccph_type     = DCCP_PKT_RESET;
 933         dh->dccph_sport    = rxdh->dccph_dport;
 934         dh->dccph_dport    = rxdh->dccph_sport;
 935         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 936         dh->dccph_x        = 1;
 937         dccp_hdr_reset(skb)->dccph_reset_code =
 938                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 939
 940         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 941         seqno = 0;
 942         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 943                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 944
 945         dccp_hdr_set_seq(dh, seqno);
 946         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 947                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 948
 949         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 950                                               rxskb->nh.iph->daddr);
 951
 952         bh_lock_sock(dccp_ctl_socket->sk);
 953         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 954                                     rxskb->nh.iph->daddr,
 955                                     rxskb->nh.iph->saddr, NULL);
 956         bh_unlock_sock(dccp_ctl_socket->sk);
 957
 958         if (err == NET_XMIT_CN || err == 0) {
 959                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 960                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 961         }
 962 out:
 963          dst_release(dst);
 964 }
 965
 966 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 967 {
 968         struct dccp_hdr *dh = dccp_hdr(skb);
 969
 970         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 971                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 972                         goto reset;
 973                 return 0;
 974         }
 975
 976         /*
 977          *  Step 3: Process LISTEN state
 978          *     If S.state == LISTEN,
 979          *        If P.type == Request or P contains a valid Init Cookie
 980          *              option,
 981          *           * Must scan the packet's options to check for an Init
 982          *              Cookie.  Only the Init Cookie is processed here,
 983          *              however; other options are processed in Step 8.  This
 984          *              scan need only be performed if the endpoint uses Init
 985          *              Cookies *
 986          *           * Generate a new socket and switch to that socket *
 987          *           Set S := new socket for this port pair
 988          *           S.state = RESPOND
 989          *           Choose S.ISS (initial seqno) or set from Init Cookie
 990          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 991          *           Continue with S.state == RESPOND
 992          *           * A Response packet will be generated in Step 11 *
 993          *        Otherwise,
 994          *           Generate Reset(No Connection) unless P.type == Reset
 995          *           Drop packet and return
 996          *
 997          * NOTE: the check for the packet types is done in
 998          *       dccp_rcv_state_process
 999          */
1000         if (sk->sk_state == DCCP_LISTEN) {
1001                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
1002
1003                 if (nsk == NULL)
1004                         goto discard;
1005
1006                 if (nsk != sk) {
1007                         if (dccp_child_process(sk, nsk, skb))
1008                                 goto reset;
1009                         return 0;
1010                 }
1011         }
1012
1013         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
1014                 goto reset;
1015         return 0;
1016
1017 reset:
1018         dccp_v4_ctl_send_reset(skb);
1019 discard:
1020         kfree_skb(skb);
1021         return 0;
1022 }
1023
1024 static inline int dccp_invalid_packet(struct sk_buff *skb)
1025 {
1026         const struct dccp_hdr *dh;
1027
1028         if (skb->pkt_type != PACKET_HOST)
1029                 return 1;
1030
1031         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
1032                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
1033                 return 1;
1034         }
1035
1036         dh = dccp_hdr(skb);
1037
1038         /* If the packet type is not understood, drop packet and return */
1039         if (dh->dccph_type >= DCCP_PKT_INVALID) {
1040                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
1041                 return 1;
1042         }
1043
1044         /*
1045          * If P.Data Offset is too small for packet type, or too large for
1046          * packet, drop packet and return
1047          */
1048         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
1049                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1050                                             "too small 1\n",
1051                                dh->dccph_doff);
1052                 return 1;
1053         }
1054
1055         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
1056                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1057                                             "too small 2\n",
1058                                dh->dccph_doff);
1059                 return 1;
1060         }
1061
1062         dh = dccp_hdr(skb);
1063
1064         /*
1065          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
1066          * has short sequence numbers), drop packet and return
1067          */
1068         if (dh->dccph_x == 0 &&
1069             dh->dccph_type != DCCP_PKT_DATA &&
1070             dh->dccph_type != DCCP_PKT_ACK &&
1071             dh->dccph_type != DCCP_PKT_DATAACK) {
1072                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
1073                                             "nor DataAck and P.X == 0\n",
1074                                dccp_packet_name(dh->dccph_type));
1075                 return 1;
1076         }
1077
1078         /* If the header checksum is incorrect, drop packet and return */
1079         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
1080                                     skb->nh.iph->daddr) < 0) {
1081                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: header checksum is "
1082                                             "incorrect\n");
1083                 return 1;
1084         }
1085
1086         return 0;
1087 }
1088
1089 /* this is called when real data arrives */
1090 int dccp_v4_rcv(struct sk_buff *skb)
1091 {
1092         const struct dccp_hdr *dh;
1093         struct sock *sk;
1094         int rc;
1095
1096         /* Step 1: Check header basics: */
1097
1098         if (dccp_invalid_packet(skb))
1099                 goto discard_it;
1100
1101         dh = dccp_hdr(skb);
1102
1103         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
1104         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
1105
1106         dccp_pr_debug("%8.8s "
1107                       "src=%u.%u.%u.%u@%-5d "
1108                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
1109                       dccp_packet_name(dh->dccph_type),
1110                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
1111                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
1112                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
1113
1114         if (dccp_packet_without_ack(skb)) {
1115                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
1116                 dccp_pr_debug_cat("\n");
1117         } else {
1118                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
1119                 dccp_pr_debug_cat(", ack=%llu\n",
1120                                   (unsigned long long)
1121                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
1122         }
1123
1124         /* Step 2:
1125          *      Look up flow ID in table and get corresponding socket */
1126         sk = __inet_lookup(&dccp_hashinfo,
1127                            skb->nh.iph->saddr, dh->dccph_sport,
1128                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
1129                            inet_iif(skb));
1130
1131         /*
1132          * Step 2:
1133          *      If no socket ...
1134          *              Generate Reset(No Connection) unless P.type == Reset
1135          *              Drop packet and return
1136          */
1137         if (sk == NULL) {
1138                 dccp_pr_debug("failed to look up flow ID in table and "
1139                               "get corresponding socket\n");
1140                 goto no_dccp_socket;
1141         }
1142
1143         /*
1144          * Step 2:
1145          *      ... or S.state == TIMEWAIT,
1146          *              Generate Reset(No Connection) unless P.type == Reset
1147          *              Drop packet and return
1148          */
1149
1150         if (sk->sk_state == DCCP_TIME_WAIT) {
1151                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
1152                               "do_time_wait\n");
1153                 goto do_time_wait;
1154         }
1155
1156         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
1157                 dccp_pr_debug("xfrm4_policy_check failed\n");
1158                 goto discard_and_relse;
1159         }
1160
1161         if (sk_filter(sk, skb, 0)) {
1162                 dccp_pr_debug("sk_filter failed\n");
1163                 goto discard_and_relse;
1164         }
1165
1166         skb->dev = NULL;
1167
1168         bh_lock_sock(sk);
1169         rc = 0;
1170         if (!sock_owned_by_user(sk))
1171                 rc = dccp_v4_do_rcv(sk, skb);
1172         else
1173                 sk_add_backlog(sk, skb);
1174         bh_unlock_sock(sk);
1175
1176         sock_put(sk);
1177         return rc;
1178
1179 no_dccp_socket:
1180         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1181                 goto discard_it;
1182         /*
1183          * Step 2:
1184          *              Generate Reset(No Connection) unless P.type == Reset
1185          *              Drop packet and return
1186          */
1187         if (dh->dccph_type != DCCP_PKT_RESET) {
1188                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1189                                         DCCP_RESET_CODE_NO_CONNECTION;
1190                 dccp_v4_ctl_send_reset(skb);
1191         }
1192
1193 discard_it:
1194         /* Discard frame. */
1195         kfree_skb(skb);
1196         return 0;
1197
1198 discard_and_relse:
1199         sock_put(sk);
1200         goto discard_it;
1201
1202 do_time_wait:
1203         inet_twsk_put((struct inet_timewait_sock *)sk);
1204         goto no_dccp_socket;
1205 }
1206
1207 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
1208         .queue_xmit     = ip_queue_xmit,
1209         .send_check     = dccp_v4_send_check,
1210         .rebuild_header = inet_sk_rebuild_header,
1211         .conn_request   = dccp_v4_conn_request,
1212         .syn_recv_sock  = dccp_v4_request_recv_sock,
1213         .net_header_len = sizeof(struct iphdr),
1214         .setsockopt     = ip_setsockopt,
1215         .getsockopt     = ip_getsockopt,
1216         .addr2sockaddr  = inet_csk_addr2sockaddr,
1217         .sockaddr_len   = sizeof(struct sockaddr_in),
1218 };
1219
1220 static int dccp_v4_init_sock(struct sock *sk)
1221 {
1222         struct dccp_sock *dp = dccp_sk(sk);
1223         static int dccp_ctl_socket_init = 1;
1224
1225         dccp_options_init(&dp->dccps_options);
1226         do_gettimeofday(&dp->dccps_epoch);
1227
1228         if (dp->dccps_options.dccpo_send_ack_vector) {
1229                 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
1230                                                            GFP_KERNEL);
1231                 if (dp->dccps_hc_rx_ackvec == NULL)
1232                         return -ENOMEM;
1233         }
1234
1235         /*
1236          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1237          * the listening (master) sock get CCID control blocks, which is not
1238          * necessary, but for now, to not mess with the test userspace apps,
1239          * lets leave it here, later the real solution is to do this in a
1240          * setsockopt(CCIDs-I-want/accept). -acme
1241          */
1242         if (likely(!dccp_ctl_socket_init)) {
1243                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1244                                                  sk);
1245                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1246                                                  sk);
1247                 if (dp->dccps_hc_rx_ccid == NULL ||
1248                     dp->dccps_hc_tx_ccid == NULL) {
1249                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1250                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1251                         if (dp->dccps_options.dccpo_send_ack_vector) {
1252                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1253                                 dp->dccps_hc_rx_ackvec = NULL;
1254                         }
1255                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1256                         return -ENOMEM;
1257                 }
1258         } else
1259                 dccp_ctl_socket_init = 0;
1260
1261         dccp_init_xmit_timers(sk);
1262         inet_csk(sk)->icsk_rto = DCCP_TIMEOUT_INIT;
1263         sk->sk_state = DCCP_CLOSED;
1264         sk->sk_write_space = dccp_write_space;
1265         inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1266         dp->dccps_mss_cache = 536;
1267         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1268         dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1269
1270         return 0;
1271 }
1272
1273 static int dccp_v4_destroy_sock(struct sock *sk)
1274 {
1275         struct dccp_sock *dp = dccp_sk(sk);
1276
1277         /*
1278          * DCCP doesn't use sk_write_queue, just sk_send_head
1279          * for retransmissions
1280          */
1281         if (sk->sk_send_head != NULL) {
1282                 kfree_skb(sk->sk_send_head);
1283                 sk->sk_send_head = NULL;
1284         }
1285
1286         /* Clean up a referenced DCCP bind bucket. */
1287         if (inet_csk(sk)->icsk_bind_hash != NULL)
1288                 inet_put_port(&dccp_hashinfo, sk);
1289
1290         kfree(dp->dccps_service_list);
1291         dp->dccps_service_list = NULL;
1292
1293         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1294         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1295         if (dp->dccps_options.dccpo_send_ack_vector) {
1296                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1297                 dp->dccps_hc_rx_ackvec = NULL;
1298         }
1299         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1300         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1301         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1302
1303         return 0;
1304 }
1305
1306 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1307 {
1308         kfree(inet_rsk(req)->opt);
1309 }
1310
1311 static struct request_sock_ops dccp_request_sock_ops = {
1312         .family         = PF_INET,
1313         .obj_size       = sizeof(struct dccp_request_sock),
1314         .rtx_syn_ack    = dccp_v4_send_response,
1315         .send_ack       = dccp_v4_reqsk_send_ack,
1316         .destructor     = dccp_v4_reqsk_destructor,
1317         .send_reset     = dccp_v4_ctl_send_reset,
1318 };
1319
1320 struct proto dccp_v4_prot = {
1321         .name                   = "DCCP",
1322         .owner                  = THIS_MODULE,
1323         .close                  = dccp_close,
1324         .connect                = dccp_v4_connect,
1325         .disconnect             = dccp_disconnect,
1326         .ioctl                  = dccp_ioctl,
1327         .init                   = dccp_v4_init_sock,
1328         .setsockopt             = dccp_setsockopt,
1329         .getsockopt             = dccp_getsockopt,
1330         .sendmsg                = dccp_sendmsg,
1331         .recvmsg                = dccp_recvmsg,
1332         .backlog_rcv            = dccp_v4_do_rcv,
1333         .hash                   = dccp_v4_hash,
1334         .unhash                 = dccp_v4_unhash,
1335         .accept                 = inet_csk_accept,
1336         .get_port               = dccp_v4_get_port,
1337         .shutdown               = dccp_shutdown,
1338         .destroy                = dccp_v4_destroy_sock,
1339         .orphan_count           = &dccp_orphan_count,
1340         .max_header             = MAX_DCCP_HEADER,
1341         .obj_size               = sizeof(struct dccp_sock),
1342         .rsk_prot               = &dccp_request_sock_ops,
1343         .twsk_obj_size          = sizeof(struct inet_timewait_sock),
1344 };