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 "ccid.h"
  27 #include "dccp.h"
  28
  29 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  30         .lhash_lock     = RW_LOCK_UNLOCKED,
  31         .lhash_users    = ATOMIC_INIT(0),
  32         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  33         .portalloc_lock = SPIN_LOCK_UNLOCKED,
  34         .port_rover     = 1024 - 1,
  35 };
  36
  37 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  38
  39 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  40 {
  41         return inet_csk_get_port(&dccp_hashinfo, sk, snum);
  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         const int hash = inet_ehashfn(daddr, lport, saddr, inet->dport,
  65                                       dccp_hashinfo.ehash_size);
  66         struct inet_ehash_bucket *head = &dccp_hashinfo.ehash[hash];
  67         const struct sock *sk2;
  68         const struct hlist_node *node;
  69         struct inet_timewait_sock *tw;
  70
  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, 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, 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_hashent = 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 rover;
 128                 int low = sysctl_local_port_range[0];
 129                 int high = sysctl_local_port_range[1];
 130                 int remaining = (high - low) + 1;
 131                 struct hlist_node *node;
 132                 struct inet_timewait_sock *tw = NULL;
 133
 134                 local_bh_disable();
 135
 136                 /* TODO. Actually it is not so bad idea to remove
 137                  * dccp_hashinfo.portalloc_lock before next submission to
 138                  * Linus.
 139                  * As soon as we touch this place at all it is time to think.
 140                  *
 141                  * Now it protects single _advisory_ variable
 142                  * dccp_hashinfo.port_rover, hence it is mostly useless.
 143                  * Code will work nicely if we just delete it, but
 144                  * I am afraid in contented case it will work not better or
 145                  * even worse: another cpu just will hit the same bucket
 146                  * and spin there.
 147                  * So some cpu salt could remove both contention and
 148                  * memory pingpong. Any ideas how to do this in a nice way?
 149                  */
 150                 spin_lock(&dccp_hashinfo.portalloc_lock);
 151                 rover = dccp_hashinfo.port_rover;
 152
 153                 do {
 154                         rover++;
 155                         if ((rover < low) || (rover > high))
 156                                 rover = low;
 157                         head = &dccp_hashinfo.bhash[inet_bhashfn(rover,
 158                                                     dccp_hashinfo.bhash_size)];
 159                         spin_lock(&head->lock);
 160
 161                         /* Does not bother with rcv_saddr checks,
 162                          * because the established check is already
 163                          * unique enough.
 164                          */
 165                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 166                                 if (tb->port == rover) {
 167                                         BUG_TRAP(!hlist_empty(&tb->owners));
 168                                         if (tb->fastreuse >= 0)
 169                                                 goto next_port;
 170                                         if (!__dccp_v4_check_established(sk,
 171                                                                          rover,
 172                                                                          &tw))
 173                                                 goto ok;
 174                                         goto next_port;
 175                                 }
 176                         }
 177
 178                         tb = inet_bind_bucket_create(dccp_hashinfo.bind_bucket_cachep,
 179                                                      head, rover);
 180                         if (tb == NULL) {
 181                                 spin_unlock(&head->lock);
 182                                 break;
 183                         }
 184                         tb->fastreuse = -1;
 185                         goto ok;
 186
 187                 next_port:
 188                         spin_unlock(&head->lock);
 189                 } while (--remaining > 0);
 190                 dccp_hashinfo.port_rover = rover;
 191                 spin_unlock(&dccp_hashinfo.portalloc_lock);
 192
 193                 local_bh_enable();
 194
 195                 return -EADDRNOTAVAIL;
 196
 197 ok:
 198                 /* All locks still held and bhs disabled */
 199                 dccp_hashinfo.port_rover = rover;
 200                 spin_unlock(&dccp_hashinfo.portalloc_lock);
 201
 202                 inet_bind_hash(sk, tb, rover);
 203                 if (sk_unhashed(sk)) {
 204                         inet_sk(sk)->sport = htons(rover);
 205                         __inet_hash(&dccp_hashinfo, sk, 0);
 206                 }
 207                 spin_unlock(&head->lock);
 208
 209                 if (tw != NULL) {
 210                         inet_twsk_deschedule(tw, &dccp_death_row);
 211                         inet_twsk_put(tw);
 212                 }
 213
 214                 ret = 0;
 215                 goto out;
 216         }
 217
 218         head = &dccp_hashinfo.bhash[inet_bhashfn(snum,
 219                                                  dccp_hashinfo.bhash_size)];
 220         tb   = inet_csk(sk)->icsk_bind_hash;
 221         spin_lock_bh(&head->lock);
 222         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 223                 __inet_hash(&dccp_hashinfo, sk, 0);
 224                 spin_unlock_bh(&head->lock);
 225                 return 0;
 226         } else {
 227                 spin_unlock(&head->lock);
 228                 /* No definite answer... Walk to established hash table */
 229                 ret = __dccp_v4_check_established(sk, snum, NULL);
 230 out:
 231                 local_bh_enable();
 232                 return ret;
 233         }
 234 }
 235
 236 static int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
 237                            int addr_len)
 238 {
 239         struct inet_sock *inet = inet_sk(sk);
 240         struct dccp_sock *dp = dccp_sk(sk);
 241         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 242         struct rtable *rt;
 243         u32 daddr, nexthop;
 244         int tmp;
 245         int err;
 246
 247         dp->dccps_role = DCCP_ROLE_CLIENT;
 248
 249         if (addr_len < sizeof(struct sockaddr_in))
 250                 return -EINVAL;
 251
 252         if (usin->sin_family != AF_INET)
 253                 return -EAFNOSUPPORT;
 254
 255         nexthop = daddr = usin->sin_addr.s_addr;
 256         if (inet->opt != NULL && inet->opt->srr) {
 257                 if (daddr == 0)
 258                         return -EINVAL;
 259                 nexthop = inet->opt->faddr;
 260         }
 261
 262         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
 263                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 264                                IPPROTO_DCCP,
 265                                inet->sport, usin->sin_port, sk);
 266         if (tmp < 0)
 267                 return tmp;
 268
 269         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 270                 ip_rt_put(rt);
 271                 return -ENETUNREACH;
 272         }
 273
 274         if (inet->opt == NULL || !inet->opt->srr)
 275                 daddr = rt->rt_dst;
 276
 277         if (inet->saddr == 0)
 278                 inet->saddr = rt->rt_src;
 279         inet->rcv_saddr = inet->saddr;
 280
 281         inet->dport = usin->sin_port;
 282         inet->daddr = daddr;
 283
 284         dp->dccps_ext_header_len = 0;
 285         if (inet->opt != NULL)
 286                 dp->dccps_ext_header_len = inet->opt->optlen;
 287         /*
 288          * Socket identity is still unknown (sport may be zero).
 289          * However we set state to DCCP_REQUESTING and not releasing socket
 290          * lock select source port, enter ourselves into the hash tables and
 291          * complete initialization after this.
 292          */
 293         dccp_set_state(sk, DCCP_REQUESTING);
 294         err = dccp_v4_hash_connect(sk);
 295         if (err != 0)
 296                 goto failure;
 297
 298         err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
 299         if (err != 0)
 300                 goto failure;
 301
 302         /* OK, now commit destination to socket.  */
 303         sk_setup_caps(sk, &rt->u.dst);
 304
 305         dp->dccps_gar =
 306                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 307                                                             inet->daddr,
 308                                                             inet->sport,
 309                                                             usin->sin_port);
 310         dccp_update_gss(sk, dp->dccps_iss);
 311
 312         inet->id = dp->dccps_iss ^ jiffies;
 313
 314         err = dccp_connect(sk);
 315         rt = NULL;
 316         if (err != 0)
 317                 goto failure;
 318 out:
 319         return err;
 320 failure:
 321         /*
 322          * This unhashes the socket and releases the local port, if necessary.
 323          */
 324         dccp_set_state(sk, DCCP_CLOSED);
 325         ip_rt_put(rt);
 326         sk->sk_route_caps = 0;
 327         inet->dport = 0;
 328         goto out;
 329 }
 330
 331 /*
 332  * This routine does path mtu discovery as defined in RFC1191.
 333  */
 334 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 335                                           const struct iphdr *iph,
 336                                           u32 mtu)
 337 {
 338         struct dst_entry *dst;
 339         const struct inet_sock *inet = inet_sk(sk);
 340         const struct dccp_sock *dp = dccp_sk(sk);
 341
 342         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 343          * send out by Linux are always < 576bytes so they should go through
 344          * unfragmented).
 345          */
 346         if (sk->sk_state == DCCP_LISTEN)
 347                 return;
 348
 349         /* We don't check in the destentry if pmtu discovery is forbidden
 350          * on this route. We just assume that no packet_to_big packets
 351          * are send back when pmtu discovery is not active.
 352          * There is a small race when the user changes this flag in the
 353          * route, but I think that's acceptable.
 354          */
 355         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 356                 return;
 357
 358         dst->ops->update_pmtu(dst, mtu);
 359
 360         /* Something is about to be wrong... Remember soft error
 361          * for the case, if this connection will not able to recover.
 362          */
 363         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 364                 sk->sk_err_soft = EMSGSIZE;
 365
 366         mtu = dst_mtu(dst);
 367
 368         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 369             dp->dccps_pmtu_cookie > mtu) {
 370                 dccp_sync_mss(sk, mtu);
 371
 372                 /*
 373                  * From: draft-ietf-dccp-spec-11.txt
 374                  *
 375                  *      DCCP-Sync packets are the best choice for upward
 376                  *      probing, since DCCP-Sync probes do not risk application
 377                  *      data loss.
 378                  */
 379                 dccp_send_sync(sk, dp->dccps_gsr);
 380         } /* else let the usual retransmit timer handle it */
 381 }
 382
 383 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 384 {
 385         int err;
 386         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 387         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 388                                      sizeof(struct dccp_hdr_ext) +
 389                                      sizeof(struct dccp_hdr_ack_bits);
 390         struct sk_buff *skb;
 391
 392         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 393                 return;
 394
 395         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 396         if (skb == NULL)
 397                 return;
 398
 399         /* Reserve space for headers. */
 400         skb_reserve(skb, MAX_DCCP_HEADER);
 401
 402         skb->dst = dst_clone(rxskb->dst);
 403
 404         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 405         dh = dccp_hdr(skb);
 406         memset(dh, 0, dccp_hdr_ack_len);
 407
 408         /* Build DCCP header and checksum it. */
 409         dh->dccph_type     = DCCP_PKT_ACK;
 410         dh->dccph_sport    = rxdh->dccph_dport;
 411         dh->dccph_dport    = rxdh->dccph_sport;
 412         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 413         dh->dccph_x        = 1;
 414
 415         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 416         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 417                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 418
 419         bh_lock_sock(dccp_ctl_socket->sk);
 420         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 421                                     rxskb->nh.iph->daddr,
 422                                     rxskb->nh.iph->saddr, NULL);
 423         bh_unlock_sock(dccp_ctl_socket->sk);
 424
 425         if (err == NET_XMIT_CN || err == 0) {
 426                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 427                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 428         }
 429 }
 430
 431 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 432                                    struct request_sock *req)
 433 {
 434         dccp_v4_ctl_send_ack(skb);
 435 }
 436
 437 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 438                                  struct dst_entry *dst)
 439 {
 440         int err = -1;
 441         struct sk_buff *skb;
 442
 443         /* First, grab a route. */
 444
 445         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 446                 goto out;
 447
 448         skb = dccp_make_response(sk, dst, req);
 449         if (skb != NULL) {
 450                 const struct inet_request_sock *ireq = inet_rsk(req);
 451
 452                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 453                                             ireq->rmt_addr,
 454                                             ireq->opt);
 455                 if (err == NET_XMIT_CN)
 456                         err = 0;
 457         }
 458
 459 out:
 460         dst_release(dst);
 461         return err;
 462 }
 463
 464 /*
 465  * This routine is called by the ICMP module when it gets some sort of error
 466  * condition. If err < 0 then the socket should be closed and the error
 467  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 468  * After adjustment header points to the first 8 bytes of the tcp header. We
 469  * need to find the appropriate port.
 470  *
 471  * The locking strategy used here is very "optimistic". When someone else
 472  * accesses the socket the ICMP is just dropped and for some paths there is no
 473  * check at all. A more general error queue to queue errors for later handling
 474  * is probably better.
 475  */
 476 void dccp_v4_err(struct sk_buff *skb, u32 info)
 477 {
 478         const struct iphdr *iph = (struct iphdr *)skb->data;
 479         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 480                                                         (iph->ihl << 2));
 481         struct dccp_sock *dp;
 482         struct inet_sock *inet;
 483         const int type = skb->h.icmph->type;
 484         const int code = skb->h.icmph->code;
 485         struct sock *sk;
 486         __u64 seq;
 487         int err;
 488
 489         if (skb->len < (iph->ihl << 2) + 8) {
 490                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 491                 return;
 492         }
 493
 494         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 495                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 496         if (sk == NULL) {
 497                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 498                 return;
 499         }
 500
 501         if (sk->sk_state == DCCP_TIME_WAIT) {
 502                 inet_twsk_put((struct inet_timewait_sock *)sk);
 503                 return;
 504         }
 505
 506         bh_lock_sock(sk);
 507         /* If too many ICMPs get dropped on busy
 508          * servers this needs to be solved differently.
 509          */
 510         if (sock_owned_by_user(sk))
 511                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 512
 513         if (sk->sk_state == DCCP_CLOSED)
 514                 goto out;
 515
 516         dp = dccp_sk(sk);
 517         seq = dccp_hdr_seq(skb);
 518         if (sk->sk_state != DCCP_LISTEN &&
 519             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 520                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 521                 goto out;
 522         }
 523
 524         switch (type) {
 525         case ICMP_SOURCE_QUENCH:
 526                 /* Just silently ignore these. */
 527                 goto out;
 528         case ICMP_PARAMETERPROB:
 529                 err = EPROTO;
 530                 break;
 531         case ICMP_DEST_UNREACH:
 532                 if (code > NR_ICMP_UNREACH)
 533                         goto out;
 534
 535                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 536                         if (!sock_owned_by_user(sk))
 537                                 dccp_do_pmtu_discovery(sk, iph, info);
 538                         goto out;
 539                 }
 540
 541                 err = icmp_err_convert[code].errno;
 542                 break;
 543         case ICMP_TIME_EXCEEDED:
 544                 err = EHOSTUNREACH;
 545                 break;
 546         default:
 547                 goto out;
 548         }
 549
 550         switch (sk->sk_state) {
 551                 struct request_sock *req , **prev;
 552         case DCCP_LISTEN:
 553                 if (sock_owned_by_user(sk))
 554                         goto out;
 555                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 556                                           iph->daddr, iph->saddr);
 557                 if (!req)
 558                         goto out;
 559
 560                 /*
 561                  * ICMPs are not backlogged, hence we cannot get an established
 562                  * socket here.
 563                  */
 564                 BUG_TRAP(!req->sk);
 565
 566                 if (seq != dccp_rsk(req)->dreq_iss) {
 567                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 568                         goto out;
 569                 }
 570                 /*
 571                  * Still in RESPOND, just remove it silently.
 572                  * There is no good way to pass the error to the newly
 573                  * created socket, and POSIX does not want network
 574                  * errors returned from accept().
 575                  */
 576                 inet_csk_reqsk_queue_drop(sk, req, prev);
 577                 goto out;
 578
 579         case DCCP_REQUESTING:
 580         case DCCP_RESPOND:
 581                 if (!sock_owned_by_user(sk)) {
 582                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 583                         sk->sk_err = err;
 584
 585                         sk->sk_error_report(sk);
 586
 587                         dccp_done(sk);
 588                 } else
 589                         sk->sk_err_soft = err;
 590                 goto out;
 591         }
 592
 593         /* If we've already connected we will keep trying
 594          * until we time out, or the user gives up.
 595          *
 596          * rfc1122 4.2.3.9 allows to consider as hard errors
 597          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 598          * but it is obsoleted by pmtu discovery).
 599          *
 600          * Note, that in modern internet, where routing is unreliable
 601          * and in each dark corner broken firewalls sit, sending random
 602          * errors ordered by their masters even this two messages finally lose
 603          * their original sense (even Linux sends invalid PORT_UNREACHs)
 604          *
 605          * Now we are in compliance with RFCs.
 606          *                                                      --ANK (980905)
 607          */
 608
 609         inet = inet_sk(sk);
 610         if (!sock_owned_by_user(sk) && inet->recverr) {
 611                 sk->sk_err = err;
 612                 sk->sk_error_report(sk);
 613         } else /* Only an error on timeout */
 614                 sk->sk_err_soft = err;
 615 out:
 616         bh_unlock_sock(sk);
 617         sock_put(sk);
 618 }
 619
 620 extern struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 621                                        enum dccp_reset_codes code);
 622
 623 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
 624 {
 625         struct sk_buff *skb;
 626         /*
 627          * FIXME: what if rebuild_header fails?
 628          * Should we be doing a rebuild_header here?
 629          */
 630         int err = inet_sk_rebuild_header(sk);
 631
 632         if (err != 0)
 633                 return err;
 634
 635         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
 636         if (skb != NULL) {
 637                 const struct dccp_sock *dp = dccp_sk(sk);
 638                 const struct inet_sock *inet = inet_sk(sk);
 639
 640                 err = ip_build_and_send_pkt(skb, sk,
 641                                             inet->saddr, inet->daddr, NULL);
 642                 if (err == NET_XMIT_CN)
 643                         err = 0;
 644
 645                 ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
 646                 ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
 647         }
 648
 649         return err;
 650 }
 651
 652 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 653                                         const struct sk_buff *skb)
 654 {
 655         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 656                                            skb->nh.iph->saddr,
 657                                            dccp_hdr(skb)->dccph_dport,
 658                                            dccp_hdr(skb)->dccph_sport);
 659 }
 660
 661 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 662 {
 663         struct inet_request_sock *ireq;
 664         struct dccp_sock dp;
 665         struct request_sock *req;
 666         struct dccp_request_sock *dreq;
 667         const __u32 saddr = skb->nh.iph->saddr;
 668         const __u32 daddr = skb->nh.iph->daddr;
 669         struct dst_entry *dst = NULL;
 670
 671         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 672         if (((struct rtable *)skb->dst)->rt_flags &
 673             (RTCF_BROADCAST | RTCF_MULTICAST))
 674                 goto drop;
 675
 676         /*
 677          * TW buckets are converted to open requests without
 678          * limitations, they conserve resources and peer is
 679          * evidently real one.
 680          */
 681         if (inet_csk_reqsk_queue_is_full(sk))
 682                 goto drop;
 683
 684         /*
 685          * Accept backlog is full. If we have already queued enough
 686          * of warm entries in syn queue, drop request. It is better than
 687          * clogging syn queue with openreqs with exponentially increasing
 688          * timeout.
 689          */
 690         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 691                 goto drop;
 692
 693         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 694         if (req == NULL)
 695                 goto drop;
 696
 697         /* FIXME: process options */
 698
 699         dccp_openreq_init(req, &dp, skb);
 700
 701         ireq = inet_rsk(req);
 702         ireq->loc_addr = daddr;
 703         ireq->rmt_addr = saddr;
 704         /* FIXME: Merge Aristeu's option parsing code when ready */
 705         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 706                                   right value */
 707         ireq->opt       = NULL;
 708
 709         /*
 710          * Step 3: Process LISTEN state
 711          *
 712          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 713          *
 714          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 715          * dccp_create_openreq_child.
 716          */
 717         dreq = dccp_rsk(req);
 718         dreq->dreq_isr = DCCP_SKB_CB(skb)->dccpd_seq;
 719         dreq->dreq_iss = dccp_v4_init_sequence(sk, skb);
 720         dreq->dreq_service = dccp_hdr_request(skb)->dccph_req_service;
 721
 722         if (dccp_v4_send_response(sk, req, dst))
 723                 goto drop_and_free;
 724
 725         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 726         return 0;
 727
 728 drop_and_free:
 729         /*
 730          * FIXME: should be reqsk_free after implementing req->rsk_ops
 731          */
 732         __reqsk_free(req);
 733 drop:
 734         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 735         return -1;
 736 }
 737
 738 /*
 739  * The three way handshake has completed - we got a valid ACK or DATAACK -
 740  * now create the new socket.
 741  *
 742  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 743  */
 744 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 745                                        struct request_sock *req,
 746                                        struct dst_entry *dst)
 747 {
 748         struct inet_request_sock *ireq;
 749         struct inet_sock *newinet;
 750         struct dccp_sock *newdp;
 751         struct sock *newsk;
 752
 753         if (sk_acceptq_is_full(sk))
 754                 goto exit_overflow;
 755
 756         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 757                 goto exit;
 758
 759         newsk = dccp_create_openreq_child(sk, req, skb);
 760         if (newsk == NULL)
 761                 goto exit;
 762
 763         sk_setup_caps(newsk, dst);
 764
 765         newdp              = dccp_sk(newsk);
 766         newinet            = inet_sk(newsk);
 767         ireq               = inet_rsk(req);
 768         newinet->daddr     = ireq->rmt_addr;
 769         newinet->rcv_saddr = ireq->loc_addr;
 770         newinet->saddr     = ireq->loc_addr;
 771         newinet->opt       = ireq->opt;
 772         ireq->opt          = NULL;
 773         newinet->mc_index  = inet_iif(skb);
 774         newinet->mc_ttl    = skb->nh.iph->ttl;
 775         newinet->id        = jiffies;
 776
 777         dccp_sync_mss(newsk, dst_mtu(dst));
 778
 779         __inet_hash(&dccp_hashinfo, newsk, 0);
 780         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 781
 782         return newsk;
 783
 784 exit_overflow:
 785         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 786 exit:
 787         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 788         dst_release(dst);
 789         return NULL;
 790 }
 791
 792 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 793 {
 794         const struct dccp_hdr *dh = dccp_hdr(skb);
 795         const struct iphdr *iph = skb->nh.iph;
 796         struct sock *nsk;
 797         struct request_sock **prev;
 798         /* Find possible connection requests. */
 799         struct request_sock *req = inet_csk_search_req(sk, &prev,
 800                                                        dh->dccph_sport,
 801                                                        iph->saddr, iph->daddr);
 802         if (req != NULL)
 803                 return dccp_check_req(sk, skb, req, prev);
 804
 805         nsk = __inet_lookup_established(&dccp_hashinfo,
 806                                         iph->saddr, dh->dccph_sport,
 807                                         iph->daddr, ntohs(dh->dccph_dport),
 808                                         inet_iif(skb));
 809         if (nsk != NULL) {
 810                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 811                         bh_lock_sock(nsk);
 812                         return nsk;
 813                 }
 814                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 815                 return NULL;
 816         }
 817
 818         return sk;
 819 }
 820
 821 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
 822                      const u32 daddr)
 823 {
 824         const struct dccp_hdr* dh = dccp_hdr(skb);
 825         int checksum_len;
 826         u32 tmp;
 827
 828         if (dh->dccph_cscov == 0)
 829                 checksum_len = skb->len;
 830         else {
 831                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 832                 checksum_len = checksum_len < skb->len ? checksum_len :
 833                                                          skb->len;
 834         }
 835
 836         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 837         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 838                                  IPPROTO_DCCP, tmp);
 839 }
 840
 841 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 842                                    const u32 saddr, const u32 daddr)
 843 {
 844         struct dccp_hdr *dh = dccp_hdr(skb);
 845         int checksum_len;
 846         u32 tmp;
 847
 848         if (dh->dccph_cscov == 0)
 849                 checksum_len = skb->len;
 850         else {
 851                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 852                 checksum_len = checksum_len < skb->len ? checksum_len :
 853                                                          skb->len;
 854         }
 855         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 856         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 857                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 858 }
 859
 860 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 861                                            struct sk_buff *skb)
 862 {
 863         struct rtable *rt;
 864         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 865                             .nl_u = { .ip4_u =
 866                                       { .daddr = skb->nh.iph->saddr,
 867                                         .saddr = skb->nh.iph->daddr,
 868                                         .tos = RT_CONN_FLAGS(sk) } },
 869                             .proto = sk->sk_protocol,
 870                             .uli_u = { .ports =
 871                                        { .sport = dccp_hdr(skb)->dccph_dport,
 872                                          .dport = dccp_hdr(skb)->dccph_sport }
 873                                      }
 874                           };
 875
 876         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 877                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 878                 return NULL;
 879         }
 880
 881         return &rt->u.dst;
 882 }
 883
 884 void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 885 {
 886         int err;
 887         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 888         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 889                                        sizeof(struct dccp_hdr_ext) +
 890                                        sizeof(struct dccp_hdr_reset);
 891         struct sk_buff *skb;
 892         struct dst_entry *dst;
 893
 894         /* Never send a reset in response to a reset. */
 895         if (rxdh->dccph_type == DCCP_PKT_RESET)
 896                 return;
 897
 898         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 899                 return;
 900
 901         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 902         if (dst == NULL)
 903                 return;
 904
 905         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 906         if (skb == NULL)
 907                 goto out;
 908
 909         /* Reserve space for headers. */
 910         skb_reserve(skb, MAX_DCCP_HEADER);
 911         skb->dst = dst_clone(dst);
 912
 913         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 914         dh = dccp_hdr(skb);
 915         memset(dh, 0, dccp_hdr_reset_len);
 916
 917         /* Build DCCP header and checksum it. */
 918         dh->dccph_type     = DCCP_PKT_RESET;
 919         dh->dccph_sport    = rxdh->dccph_dport;
 920         dh->dccph_dport    = rxdh->dccph_sport;
 921         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 922         dh->dccph_x        = 1;
 923         dccp_hdr_reset(skb)->dccph_reset_code =
 924                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 925
 926         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 927         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 928                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 929
 930         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 931                                               rxskb->nh.iph->daddr);
 932
 933         bh_lock_sock(dccp_ctl_socket->sk);
 934         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 935                                     rxskb->nh.iph->daddr,
 936                                     rxskb->nh.iph->saddr, NULL);
 937         bh_unlock_sock(dccp_ctl_socket->sk);
 938
 939         if (err == NET_XMIT_CN || err == 0) {
 940                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 941                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 942         }
 943 out:
 944          dst_release(dst);
 945 }
 946
 947 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 948 {
 949         struct dccp_hdr *dh = dccp_hdr(skb);
 950
 951         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 952                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 953                         goto reset;
 954                 return 0;
 955         }
 956
 957         /*
 958          *  Step 3: Process LISTEN state
 959          *     If S.state == LISTEN,
 960          *        If P.type == Request or P contains a valid Init Cookie
 961          *              option,
 962          *           * Must scan the packet's options to check for an Init
 963          *              Cookie.  Only the Init Cookie is processed here,
 964          *              however; other options are processed in Step 8.  This
 965          *              scan need only be performed if the endpoint uses Init
 966          *              Cookies *
 967          *           * Generate a new socket and switch to that socket *
 968          *           Set S := new socket for this port pair
 969          *           S.state = RESPOND
 970          *           Choose S.ISS (initial seqno) or set from Init Cookie
 971          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 972          *           Continue with S.state == RESPOND
 973          *           * A Response packet will be generated in Step 11 *
 974          *        Otherwise,
 975          *           Generate Reset(No Connection) unless P.type == Reset
 976          *           Drop packet and return
 977          *
 978          * NOTE: the check for the packet types is done in
 979          *       dccp_rcv_state_process
 980          */
 981         if (sk->sk_state == DCCP_LISTEN) {
 982                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 983
 984                 if (nsk == NULL)
 985                         goto discard;
 986
 987                 if (nsk != sk) {
 988                         if (dccp_child_process(sk, nsk, skb))
 989                                 goto reset;
 990                         return 0;
 991                 }
 992         }
 993
 994         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 995                 goto reset;
 996         return 0;
 997
 998 reset:
 999         DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_NO_CONNECTION;
1000         dccp_v4_ctl_send_reset(skb);
1001 discard:
1002         kfree_skb(skb);
1003         return 0;
1004 }
1005
1006 static inline int dccp_invalid_packet(struct sk_buff *skb)
1007 {
1008         const struct dccp_hdr *dh;
1009
1010         if (skb->pkt_type != PACKET_HOST)
1011                 return 1;
1012
1013         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
1014                 dccp_pr_debug("pskb_may_pull failed\n");
1015                 return 1;
1016         }
1017
1018         dh = dccp_hdr(skb);
1019
1020         /* If the packet type is not understood, drop packet and return */
1021         if (dh->dccph_type >= DCCP_PKT_INVALID) {
1022                 dccp_pr_debug("invalid packet type\n");
1023                 return 1;
1024         }
1025
1026         /*
1027          * If P.Data Offset is too small for packet type, or too large for
1028          * packet, drop packet and return
1029          */
1030         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
1031                 dccp_pr_debug("Offset(%u) too small 1\n", dh->dccph_doff);
1032                 return 1;
1033         }
1034
1035         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
1036                 dccp_pr_debug("P.Data Offset(%u) too small 2\n",
1037                               dh->dccph_doff);
1038                 return 1;
1039         }
1040
1041         dh = dccp_hdr(skb);
1042
1043         /*
1044          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
1045          * has short sequence numbers), drop packet and return
1046          */
1047         if (dh->dccph_x == 0 &&
1048             dh->dccph_type != DCCP_PKT_DATA &&
1049             dh->dccph_type != DCCP_PKT_ACK &&
1050             dh->dccph_type != DCCP_PKT_DATAACK) {
1051                 dccp_pr_debug("P.type (%s) not Data, Ack nor DataAck and "
1052                               "P.X == 0\n", dccp_packet_name(dh->dccph_type));
1053                 return 1;
1054         }
1055
1056         /* If the header checksum is incorrect, drop packet and return */
1057         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
1058                                     skb->nh.iph->daddr) < 0) {
1059                 dccp_pr_debug("header checksum is incorrect\n");
1060                 return 1;
1061         }
1062
1063         return 0;
1064 }
1065
1066 /* this is called when real data arrives */
1067 int dccp_v4_rcv(struct sk_buff *skb)
1068 {
1069         const struct dccp_hdr *dh;
1070         struct sock *sk;
1071         int rc;
1072
1073         /* Step 1: Check header basics: */
1074
1075         if (dccp_invalid_packet(skb))
1076                 goto discard_it;
1077
1078         dh = dccp_hdr(skb);
1079 #if 0
1080         /*
1081          * Use something like this to simulate some DATA/DATAACK loss to test
1082          * dccp_ackpkts_add, you'll get something like this on a session that
1083          * sends 10 DATA/DATAACK packets:
1084          *
1085          * ackpkts_print: 281473596467422 |0,0|3,0|0,0|3,0|0,0|3,0|0,0|3,0|0,1|
1086          *
1087          * 0, 0 means: DCCP_ACKPKTS_STATE_RECEIVED, RLE == just this packet
1088          * 0, 1 means: DCCP_ACKPKTS_STATE_RECEIVED, RLE == two adjacent packets
1089          *                                                 with the same state
1090          * 3, 0 means: DCCP_ACKPKTS_STATE_NOT_RECEIVED, RLE == just this packet
1091          *
1092          * So...
1093          *
1094          * 281473596467422 was received
1095          * 281473596467421 was not received
1096          * 281473596467420 was received
1097          * 281473596467419 was not received
1098          * 281473596467418 was received
1099          * 281473596467417 was not received
1100          * 281473596467416 was received
1101          * 281473596467415 was not received
1102          * 281473596467414 was received
1103          * 281473596467413 was received (this one was the 3way handshake
1104          *                               RESPONSE)
1105          *
1106          */
1107         if (dh->dccph_type == DCCP_PKT_DATA ||
1108             dh->dccph_type == DCCP_PKT_DATAACK) {
1109                 static int discard = 0;
1110
1111                 if (discard) {
1112                         discard = 0;
1113                         goto discard_it;
1114                 }
1115                 discard = 1;
1116         }
1117 #endif
1118         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
1119         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
1120
1121         dccp_pr_debug("%8.8s "
1122                       "src=%u.%u.%u.%u@%-5d "
1123                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
1124                       dccp_packet_name(dh->dccph_type),
1125                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
1126                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
1127                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
1128
1129         if (dccp_packet_without_ack(skb)) {
1130                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
1131                 dccp_pr_debug_cat("\n");
1132         } else {
1133                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
1134                 dccp_pr_debug_cat(", ack=%llu\n",
1135                                   (unsigned long long)
1136                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
1137         }
1138
1139         /* Step 2:
1140          *      Look up flow ID in table and get corresponding socket */
1141         sk = __inet_lookup(&dccp_hashinfo,
1142                            skb->nh.iph->saddr, dh->dccph_sport,
1143                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
1144                            inet_iif(skb));
1145
1146         /*
1147          * Step 2:
1148          *      If no socket ...
1149          *              Generate Reset(No Connection) unless P.type == Reset
1150          *              Drop packet and return
1151          */
1152         if (sk == NULL) {
1153                 dccp_pr_debug("failed to look up flow ID in table and "
1154                               "get corresponding socket\n");
1155                 goto no_dccp_socket;
1156         }
1157
1158         /*
1159          * Step 2:
1160          *      ... or S.state == TIMEWAIT,
1161          *              Generate Reset(No Connection) unless P.type == Reset
1162          *              Drop packet and return
1163          */
1164
1165         if (sk->sk_state == DCCP_TIME_WAIT) {
1166                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
1167                               "do_time_wait\n");
1168                 goto do_time_wait;
1169         }
1170
1171         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
1172                 dccp_pr_debug("xfrm4_policy_check failed\n");
1173                 goto discard_and_relse;
1174         }
1175
1176         if (sk_filter(sk, skb, 0)) {
1177                 dccp_pr_debug("sk_filter failed\n");
1178                 goto discard_and_relse;
1179         }
1180
1181         skb->dev = NULL;
1182
1183         bh_lock_sock(sk);
1184         rc = 0;
1185         if (!sock_owned_by_user(sk))
1186                 rc = dccp_v4_do_rcv(sk, skb);
1187         else
1188                 sk_add_backlog(sk, skb);
1189         bh_unlock_sock(sk);
1190
1191         sock_put(sk);
1192         return rc;
1193
1194 no_dccp_socket:
1195         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1196                 goto discard_it;
1197         /*
1198          * Step 2:
1199          *              Generate Reset(No Connection) unless P.type == Reset
1200          *              Drop packet and return
1201          */
1202         if (dh->dccph_type != DCCP_PKT_RESET) {
1203                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1204                                         DCCP_RESET_CODE_NO_CONNECTION;
1205                 dccp_v4_ctl_send_reset(skb);
1206         }
1207
1208 discard_it:
1209         /* Discard frame. */
1210         kfree_skb(skb);
1211         return 0;
1212
1213 discard_and_relse:
1214         sock_put(sk);
1215         goto discard_it;
1216
1217 do_time_wait:
1218         inet_twsk_put((struct inet_timewait_sock *)sk);
1219         goto no_dccp_socket;
1220 }
1221
1222 static int dccp_v4_init_sock(struct sock *sk)
1223 {
1224         struct dccp_sock *dp = dccp_sk(sk);
1225         static int dccp_ctl_socket_init = 1;
1226
1227         dccp_options_init(&dp->dccps_options);
1228
1229         if (dp->dccps_options.dccpo_send_ack_vector) {
1230                 dp->dccps_hc_rx_ackpkts =
1231                         dccp_ackpkts_alloc(DCCP_MAX_ACK_VECTOR_LEN,
1232                                            GFP_KERNEL);
1233
1234                 if (dp->dccps_hc_rx_ackpkts == NULL)
1235                         return -ENOMEM;
1236         }
1237
1238         /*
1239          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1240          * the listening (master) sock get CCID control blocks, which is not
1241          * necessary, but for now, to not mess with the test userspace apps,
1242          * lets leave it here, later the real solution is to do this in a
1243          * setsockopt(CCIDs-I-want/accept). -acme
1244          */
1245         if (likely(!dccp_ctl_socket_init)) {
1246                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_ccid,
1247                                                  sk);
1248                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_ccid,
1249                                                  sk);
1250                 if (dp->dccps_hc_rx_ccid == NULL ||
1251                     dp->dccps_hc_tx_ccid == NULL) {
1252                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1253                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1254                         dccp_ackpkts_free(dp->dccps_hc_rx_ackpkts);
1255                         dp->dccps_hc_rx_ackpkts = NULL;
1256                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1257                         return -ENOMEM;
1258                 }
1259         } else
1260                 dccp_ctl_socket_init = 0;
1261
1262         dccp_init_xmit_timers(sk);
1263         inet_csk(sk)->icsk_rto = DCCP_TIMEOUT_INIT;
1264         sk->sk_state = DCCP_CLOSED;
1265         dp->dccps_mss_cache = 536;
1266         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1267
1268         return 0;
1269 }
1270
1271 int dccp_v4_destroy_sock(struct sock *sk)
1272 {
1273         struct dccp_sock *dp = dccp_sk(sk);
1274
1275         /*
1276          * DCCP doesn't use sk_qrite_queue, just sk_send_head
1277          * for retransmissions
1278          */
1279         if (sk->sk_send_head != NULL) {
1280                 kfree_skb(sk->sk_send_head);
1281                 sk->sk_send_head = NULL;
1282         }
1283
1284         /* Clean up a referenced DCCP bind bucket. */
1285         if (inet_csk(sk)->icsk_bind_hash != NULL)
1286                 inet_put_port(&dccp_hashinfo, sk);
1287
1288         dccp_ackpkts_free(dp->dccps_hc_rx_ackpkts);
1289         dp->dccps_hc_rx_ackpkts = NULL;
1290         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1291         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1292         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1293
1294         return 0;
1295 }
1296
1297 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1298 {
1299         kfree(inet_rsk(req)->opt);
1300 }
1301
1302 static struct request_sock_ops dccp_request_sock_ops = {
1303         .family         = PF_INET,
1304         .obj_size       = sizeof(struct dccp_request_sock),
1305         .rtx_syn_ack    = dccp_v4_send_response,
1306         .send_ack       = dccp_v4_reqsk_send_ack,
1307         .destructor     = dccp_v4_reqsk_destructor,
1308         .send_reset     = dccp_v4_ctl_send_reset,
1309 };
1310
1311 struct proto dccp_v4_prot = {
1312         .name                   = "DCCP",
1313         .owner                  = THIS_MODULE,
1314         .close                  = dccp_close,
1315         .connect                = dccp_v4_connect,
1316         .disconnect             = dccp_disconnect,
1317         .ioctl                  = dccp_ioctl,
1318         .init                   = dccp_v4_init_sock,
1319         .setsockopt             = dccp_setsockopt,
1320         .getsockopt             = dccp_getsockopt,
1321         .sendmsg                = dccp_sendmsg,
1322         .recvmsg                = dccp_recvmsg,
1323         .backlog_rcv            = dccp_v4_do_rcv,
1324         .hash                   = dccp_v4_hash,
1325         .unhash                 = dccp_v4_unhash,
1326         .accept                 = inet_csk_accept,
1327         .get_port               = dccp_v4_get_port,
1328         .shutdown               = dccp_shutdown,
1329         .destroy                = dccp_v4_destroy_sock,
1330         .orphan_count           = &dccp_orphan_count,
1331         .max_header             = MAX_DCCP_HEADER,
1332         .obj_size               = sizeof(struct dccp_sock),
1333         .rsk_prot               = &dccp_request_sock_ops,
1334         .twsk_obj_size          = sizeof(struct inet_timewait_sock),
1335 };