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         .portalloc_lock = SPIN_LOCK_UNLOCKED,
  35         .port_rover     = 1024 - 1,
  36 };
  37
  38 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  39
  40 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  41 {
  42         return inet_csk_get_port(&dccp_hashinfo, sk, snum);
  43 }
  44
  45 static void dccp_v4_hash(struct sock *sk)
  46 {
  47         inet_hash(&dccp_hashinfo, sk);
  48 }
  49
  50 static void dccp_v4_unhash(struct sock *sk)
  51 {
  52         inet_unhash(&dccp_hashinfo, sk);
  53 }
  54
  55 /* called with local bh disabled */
  56 static int __dccp_v4_check_established(struct sock *sk, const __u16 lport,
  57                                       struct inet_timewait_sock **twp)
  58 {
  59         struct inet_sock *inet = inet_sk(sk);
  60         const u32 daddr = inet->rcv_saddr;
  61         const u32 saddr = inet->daddr;
  62         const int dif = sk->sk_bound_dev_if;
  63         INET_ADDR_COOKIE(acookie, saddr, daddr)
  64         const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
  65         unsigned int hash = inet_ehashfn(daddr, lport, saddr, inet->dport);
  66         struct inet_ehash_bucket *head = inet_ehash_bucket(&dccp_hashinfo, hash);
  67         const struct sock *sk2;
  68         const struct hlist_node *node;
  69         struct inet_timewait_sock *tw;
  70
  71         prefetch(head->chain.first);
  72         write_lock(&head->lock);
  73
  74         /* Check TIME-WAIT sockets first. */
  75         sk_for_each(sk2, node, &(head + dccp_hashinfo.ehash_size)->chain) {
  76                 tw = inet_twsk(sk2);
  77
  78                 if (INET_TW_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
  79                         goto not_unique;
  80         }
  81         tw = NULL;
  82
  83         /* And established part... */
  84         sk_for_each(sk2, node, &head->chain) {
  85                 if (INET_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
  86                         goto not_unique;
  87         }
  88
  89         /* Must record num and sport now. Otherwise we will see
  90          * in hash table socket with a funny identity. */
  91         inet->num = lport;
  92         inet->sport = htons(lport);
  93         sk->sk_hash = hash;
  94         BUG_TRAP(sk_unhashed(sk));
  95         __sk_add_node(sk, &head->chain);
  96         sock_prot_inc_use(sk->sk_prot);
  97         write_unlock(&head->lock);
  98
  99         if (twp != NULL) {
 100                 *twp = tw;
 101                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 102         } else if (tw != NULL) {
 103                 /* Silly. Should hash-dance instead... */
 104                 inet_twsk_deschedule(tw, &dccp_death_row);
 105                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 106
 107                 inet_twsk_put(tw);
 108         }
 109
 110         return 0;
 111
 112 not_unique:
 113         write_unlock(&head->lock);
 114         return -EADDRNOTAVAIL;
 115 }
 116
 117 /*
 118  * Bind a port for a connect operation and hash it.
 119  */
 120 static int dccp_v4_hash_connect(struct sock *sk)
 121 {
 122         const unsigned short snum = inet_sk(sk)->num;
 123         struct inet_bind_hashbucket *head;
 124         struct inet_bind_bucket *tb;
 125         int ret;
 126
 127         if (snum == 0) {
 128                 int rover;
 129                 int low = sysctl_local_port_range[0];
 130                 int high = sysctl_local_port_range[1];
 131                 int remaining = (high - low) + 1;
 132                 struct hlist_node *node;
 133                 struct inet_timewait_sock *tw = NULL;
 134
 135                 local_bh_disable();
 136
 137                 /* TODO. Actually it is not so bad idea to remove
 138                  * dccp_hashinfo.portalloc_lock before next submission to
 139                  * Linus.
 140                  * As soon as we touch this place at all it is time to think.
 141                  *
 142                  * Now it protects single _advisory_ variable
 143                  * dccp_hashinfo.port_rover, hence it is mostly useless.
 144                  * Code will work nicely if we just delete it, but
 145                  * I am afraid in contented case it will work not better or
 146                  * even worse: another cpu just will hit the same bucket
 147                  * and spin there.
 148                  * So some cpu salt could remove both contention and
 149                  * memory pingpong. Any ideas how to do this in a nice way?
 150                  */
 151                 spin_lock(&dccp_hashinfo.portalloc_lock);
 152                 rover = dccp_hashinfo.port_rover;
 153
 154                 do {
 155                         rover++;
 156                         if ((rover < low) || (rover > high))
 157                                 rover = low;
 158                         head = &dccp_hashinfo.bhash[inet_bhashfn(rover,
 159                                                     dccp_hashinfo.bhash_size)];
 160                         spin_lock(&head->lock);
 161
 162                         /* Does not bother with rcv_saddr checks,
 163                          * because the established check is already
 164                          * unique enough.
 165                          */
 166                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 167                                 if (tb->port == rover) {
 168                                         BUG_TRAP(!hlist_empty(&tb->owners));
 169                                         if (tb->fastreuse >= 0)
 170                                                 goto next_port;
 171                                         if (!__dccp_v4_check_established(sk,
 172                                                                          rover,
 173                                                                          &tw))
 174                                                 goto ok;
 175                                         goto next_port;
 176                                 }
 177                         }
 178
 179                         tb = inet_bind_bucket_create(dccp_hashinfo.bind_bucket_cachep,
 180                                                      head, rover);
 181                         if (tb == NULL) {
 182                                 spin_unlock(&head->lock);
 183                                 break;
 184                         }
 185                         tb->fastreuse = -1;
 186                         goto ok;
 187
 188                 next_port:
 189                         spin_unlock(&head->lock);
 190                 } while (--remaining > 0);
 191                 dccp_hashinfo.port_rover = rover;
 192                 spin_unlock(&dccp_hashinfo.portalloc_lock);
 193
 194                 local_bh_enable();
 195
 196                 return -EADDRNOTAVAIL;
 197
 198 ok:
 199                 /* All locks still held and bhs disabled */
 200                 dccp_hashinfo.port_rover = rover;
 201                 spin_unlock(&dccp_hashinfo.portalloc_lock);
 202
 203                 inet_bind_hash(sk, tb, rover);
 204                 if (sk_unhashed(sk)) {
 205                         inet_sk(sk)->sport = htons(rover);
 206                         __inet_hash(&dccp_hashinfo, sk, 0);
 207                 }
 208                 spin_unlock(&head->lock);
 209
 210                 if (tw != NULL) {
 211                         inet_twsk_deschedule(tw, &dccp_death_row);
 212                         inet_twsk_put(tw);
 213                 }
 214
 215                 ret = 0;
 216                 goto out;
 217         }
 218
 219         head = &dccp_hashinfo.bhash[inet_bhashfn(snum,
 220                                                  dccp_hashinfo.bhash_size)];
 221         tb   = inet_csk(sk)->icsk_bind_hash;
 222         spin_lock_bh(&head->lock);
 223         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 224                 __inet_hash(&dccp_hashinfo, sk, 0);
 225                 spin_unlock_bh(&head->lock);
 226                 return 0;
 227         } else {
 228                 spin_unlock(&head->lock);
 229                 /* No definite answer... Walk to established hash table */
 230                 ret = __dccp_v4_check_established(sk, snum, NULL);
 231 out:
 232                 local_bh_enable();
 233                 return ret;
 234         }
 235 }
 236
 237 static int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
 238                            int addr_len)
 239 {
 240         struct inet_sock *inet = inet_sk(sk);
 241         struct dccp_sock *dp = dccp_sk(sk);
 242         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 243         struct rtable *rt;
 244         u32 daddr, nexthop;
 245         int tmp;
 246         int err;
 247
 248         dp->dccps_role = DCCP_ROLE_CLIENT;
 249
 250         if (dccp_service_not_initialized(sk))
 251                 return -EPROTO;
 252
 253         if (addr_len < sizeof(struct sockaddr_in))
 254                 return -EINVAL;
 255
 256         if (usin->sin_family != AF_INET)
 257                 return -EAFNOSUPPORT;
 258
 259         nexthop = daddr = usin->sin_addr.s_addr;
 260         if (inet->opt != NULL && inet->opt->srr) {
 261                 if (daddr == 0)
 262                         return -EINVAL;
 263                 nexthop = inet->opt->faddr;
 264         }
 265
 266         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
 267                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 268                                IPPROTO_DCCP,
 269                                inet->sport, usin->sin_port, sk);
 270         if (tmp < 0)
 271                 return tmp;
 272
 273         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 274                 ip_rt_put(rt);
 275                 return -ENETUNREACH;
 276         }
 277
 278         if (inet->opt == NULL || !inet->opt->srr)
 279                 daddr = rt->rt_dst;
 280
 281         if (inet->saddr == 0)
 282                 inet->saddr = rt->rt_src;
 283         inet->rcv_saddr = inet->saddr;
 284
 285         inet->dport = usin->sin_port;
 286         inet->daddr = daddr;
 287
 288         dp->dccps_ext_header_len = 0;
 289         if (inet->opt != NULL)
 290                 dp->dccps_ext_header_len = inet->opt->optlen;
 291         /*
 292          * Socket identity is still unknown (sport may be zero).
 293          * However we set state to DCCP_REQUESTING and not releasing socket
 294          * lock select source port, enter ourselves into the hash tables and
 295          * complete initialization after this.
 296          */
 297         dccp_set_state(sk, DCCP_REQUESTING);
 298         err = dccp_v4_hash_connect(sk);
 299         if (err != 0)
 300                 goto failure;
 301
 302         err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
 303         if (err != 0)
 304                 goto failure;
 305
 306         /* OK, now commit destination to socket.  */
 307         sk_setup_caps(sk, &rt->u.dst);
 308
 309         dp->dccps_gar =
 310                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 311                                                             inet->daddr,
 312                                                             inet->sport,
 313                                                             usin->sin_port);
 314         dccp_update_gss(sk, dp->dccps_iss);
 315
 316         /*
 317          * SWL and AWL are initially adjusted so that they are not less than
 318          * the initial Sequence Numbers received and sent, respectively:
 319          *      SWL := max(GSR + 1 - floor(W/4), ISR),
 320          *      AWL := max(GSS - W' + 1, ISS).
 321          * These adjustments MUST be applied only at the beginning of the
 322          * connection.
 323          */
 324         dccp_set_seqno(&dp->dccps_awl, max48(dp->dccps_awl, dp->dccps_iss));
 325
 326         inet->id = dp->dccps_iss ^ jiffies;
 327
 328         err = dccp_connect(sk);
 329         rt = NULL;
 330         if (err != 0)
 331                 goto failure;
 332 out:
 333         return err;
 334 failure:
 335         /*
 336          * This unhashes the socket and releases the local port, if necessary.
 337          */
 338         dccp_set_state(sk, DCCP_CLOSED);
 339         ip_rt_put(rt);
 340         sk->sk_route_caps = 0;
 341         inet->dport = 0;
 342         goto out;
 343 }
 344
 345 /*
 346  * This routine does path mtu discovery as defined in RFC1191.
 347  */
 348 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 349                                           const struct iphdr *iph,
 350                                           u32 mtu)
 351 {
 352         struct dst_entry *dst;
 353         const struct inet_sock *inet = inet_sk(sk);
 354         const struct dccp_sock *dp = dccp_sk(sk);
 355
 356         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 357          * send out by Linux are always < 576bytes so they should go through
 358          * unfragmented).
 359          */
 360         if (sk->sk_state == DCCP_LISTEN)
 361                 return;
 362
 363         /* We don't check in the destentry if pmtu discovery is forbidden
 364          * on this route. We just assume that no packet_to_big packets
 365          * are send back when pmtu discovery is not active.
 366          * There is a small race when the user changes this flag in the
 367          * route, but I think that's acceptable.
 368          */
 369         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 370                 return;
 371
 372         dst->ops->update_pmtu(dst, mtu);
 373
 374         /* Something is about to be wrong... Remember soft error
 375          * for the case, if this connection will not able to recover.
 376          */
 377         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 378                 sk->sk_err_soft = EMSGSIZE;
 379
 380         mtu = dst_mtu(dst);
 381
 382         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 383             dp->dccps_pmtu_cookie > mtu) {
 384                 dccp_sync_mss(sk, mtu);
 385
 386                 /*
 387                  * From: draft-ietf-dccp-spec-11.txt
 388                  *
 389                  *      DCCP-Sync packets are the best choice for upward
 390                  *      probing, since DCCP-Sync probes do not risk application
 391                  *      data loss.
 392                  */
 393                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 394         } /* else let the usual retransmit timer handle it */
 395 }
 396
 397 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 398 {
 399         int err;
 400         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 401         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 402                                      sizeof(struct dccp_hdr_ext) +
 403                                      sizeof(struct dccp_hdr_ack_bits);
 404         struct sk_buff *skb;
 405
 406         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 407                 return;
 408
 409         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 410         if (skb == NULL)
 411                 return;
 412
 413         /* Reserve space for headers. */
 414         skb_reserve(skb, MAX_DCCP_HEADER);
 415
 416         skb->dst = dst_clone(rxskb->dst);
 417
 418         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 419         dh = dccp_hdr(skb);
 420         memset(dh, 0, dccp_hdr_ack_len);
 421
 422         /* Build DCCP header and checksum it. */
 423         dh->dccph_type     = DCCP_PKT_ACK;
 424         dh->dccph_sport    = rxdh->dccph_dport;
 425         dh->dccph_dport    = rxdh->dccph_sport;
 426         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 427         dh->dccph_x        = 1;
 428
 429         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 430         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 431                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 432
 433         bh_lock_sock(dccp_ctl_socket->sk);
 434         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 435                                     rxskb->nh.iph->daddr,
 436                                     rxskb->nh.iph->saddr, NULL);
 437         bh_unlock_sock(dccp_ctl_socket->sk);
 438
 439         if (err == NET_XMIT_CN || err == 0) {
 440                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 441                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 442         }
 443 }
 444
 445 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 446                                    struct request_sock *req)
 447 {
 448         dccp_v4_ctl_send_ack(skb);
 449 }
 450
 451 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 452                                  struct dst_entry *dst)
 453 {
 454         int err = -1;
 455         struct sk_buff *skb;
 456
 457         /* First, grab a route. */
 458
 459         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 460                 goto out;
 461
 462         skb = dccp_make_response(sk, dst, req);
 463         if (skb != NULL) {
 464                 const struct inet_request_sock *ireq = inet_rsk(req);
 465
 466                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 467                                             ireq->rmt_addr,
 468                                             ireq->opt);
 469                 if (err == NET_XMIT_CN)
 470                         err = 0;
 471         }
 472
 473 out:
 474         dst_release(dst);
 475         return err;
 476 }
 477
 478 /*
 479  * This routine is called by the ICMP module when it gets some sort of error
 480  * condition. If err < 0 then the socket should be closed and the error
 481  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 482  * After adjustment header points to the first 8 bytes of the tcp header. We
 483  * need to find the appropriate port.
 484  *
 485  * The locking strategy used here is very "optimistic". When someone else
 486  * accesses the socket the ICMP is just dropped and for some paths there is no
 487  * check at all. A more general error queue to queue errors for later handling
 488  * is probably better.
 489  */
 490 void dccp_v4_err(struct sk_buff *skb, u32 info)
 491 {
 492         const struct iphdr *iph = (struct iphdr *)skb->data;
 493         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 494                                                         (iph->ihl << 2));
 495         struct dccp_sock *dp;
 496         struct inet_sock *inet;
 497         const int type = skb->h.icmph->type;
 498         const int code = skb->h.icmph->code;
 499         struct sock *sk;
 500         __u64 seq;
 501         int err;
 502
 503         if (skb->len < (iph->ihl << 2) + 8) {
 504                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 505                 return;
 506         }
 507
 508         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 509                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 510         if (sk == NULL) {
 511                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 512                 return;
 513         }
 514
 515         if (sk->sk_state == DCCP_TIME_WAIT) {
 516                 inet_twsk_put((struct inet_timewait_sock *)sk);
 517                 return;
 518         }
 519
 520         bh_lock_sock(sk);
 521         /* If too many ICMPs get dropped on busy
 522          * servers this needs to be solved differently.
 523          */
 524         if (sock_owned_by_user(sk))
 525                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 526
 527         if (sk->sk_state == DCCP_CLOSED)
 528                 goto out;
 529
 530         dp = dccp_sk(sk);
 531         seq = dccp_hdr_seq(skb);
 532         if (sk->sk_state != DCCP_LISTEN &&
 533             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 534                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 535                 goto out;
 536         }
 537
 538         switch (type) {
 539         case ICMP_SOURCE_QUENCH:
 540                 /* Just silently ignore these. */
 541                 goto out;
 542         case ICMP_PARAMETERPROB:
 543                 err = EPROTO;
 544                 break;
 545         case ICMP_DEST_UNREACH:
 546                 if (code > NR_ICMP_UNREACH)
 547                         goto out;
 548
 549                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 550                         if (!sock_owned_by_user(sk))
 551                                 dccp_do_pmtu_discovery(sk, iph, info);
 552                         goto out;
 553                 }
 554
 555                 err = icmp_err_convert[code].errno;
 556                 break;
 557         case ICMP_TIME_EXCEEDED:
 558                 err = EHOSTUNREACH;
 559                 break;
 560         default:
 561                 goto out;
 562         }
 563
 564         switch (sk->sk_state) {
 565                 struct request_sock *req , **prev;
 566         case DCCP_LISTEN:
 567                 if (sock_owned_by_user(sk))
 568                         goto out;
 569                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 570                                           iph->daddr, iph->saddr);
 571                 if (!req)
 572                         goto out;
 573
 574                 /*
 575                  * ICMPs are not backlogged, hence we cannot get an established
 576                  * socket here.
 577                  */
 578                 BUG_TRAP(!req->sk);
 579
 580                 if (seq != dccp_rsk(req)->dreq_iss) {
 581                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 582                         goto out;
 583                 }
 584                 /*
 585                  * Still in RESPOND, just remove it silently.
 586                  * There is no good way to pass the error to the newly
 587                  * created socket, and POSIX does not want network
 588                  * errors returned from accept().
 589                  */
 590                 inet_csk_reqsk_queue_drop(sk, req, prev);
 591                 goto out;
 592
 593         case DCCP_REQUESTING:
 594         case DCCP_RESPOND:
 595                 if (!sock_owned_by_user(sk)) {
 596                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 597                         sk->sk_err = err;
 598
 599                         sk->sk_error_report(sk);
 600
 601                         dccp_done(sk);
 602                 } else
 603                         sk->sk_err_soft = err;
 604                 goto out;
 605         }
 606
 607         /* If we've already connected we will keep trying
 608          * until we time out, or the user gives up.
 609          *
 610          * rfc1122 4.2.3.9 allows to consider as hard errors
 611          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 612          * but it is obsoleted by pmtu discovery).
 613          *
 614          * Note, that in modern internet, where routing is unreliable
 615          * and in each dark corner broken firewalls sit, sending random
 616          * errors ordered by their masters even this two messages finally lose
 617          * their original sense (even Linux sends invalid PORT_UNREACHs)
 618          *
 619          * Now we are in compliance with RFCs.
 620          *                                                      --ANK (980905)
 621          */
 622
 623         inet = inet_sk(sk);
 624         if (!sock_owned_by_user(sk) && inet->recverr) {
 625                 sk->sk_err = err;
 626                 sk->sk_error_report(sk);
 627         } else /* Only an error on timeout */
 628                 sk->sk_err_soft = err;
 629 out:
 630         bh_unlock_sock(sk);
 631         sock_put(sk);
 632 }
 633
 634 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
 635 {
 636         struct sk_buff *skb;
 637         /*
 638          * FIXME: what if rebuild_header fails?
 639          * Should we be doing a rebuild_header here?
 640          */
 641         int err = inet_sk_rebuild_header(sk);
 642
 643         if (err != 0)
 644                 return err;
 645
 646         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
 647         if (skb != NULL) {
 648                 const struct inet_sock *inet = inet_sk(sk);
 649
 650                 err = ip_build_and_send_pkt(skb, sk,
 651                                             inet->saddr, inet->daddr, NULL);
 652                 if (err == NET_XMIT_CN)
 653                         err = 0;
 654         }
 655
 656         return err;
 657 }
 658
 659 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 660                                         const struct sk_buff *skb)
 661 {
 662         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 663                                            skb->nh.iph->saddr,
 664                                            dccp_hdr(skb)->dccph_dport,
 665                                            dccp_hdr(skb)->dccph_sport);
 666 }
 667
 668 static inline int dccp_bad_service_code(const struct sock *sk,
 669                                         const __u32 service)
 670 {
 671         const struct dccp_sock *dp = dccp_sk(sk);
 672
 673         if (dp->dccps_service == service)
 674                 return 0;
 675         return !dccp_list_has_service(dp->dccps_service_list, service);
 676 }
 677
 678 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 679 {
 680         struct inet_request_sock *ireq;
 681         struct dccp_sock dp;
 682         struct request_sock *req;
 683         struct dccp_request_sock *dreq;
 684         const __u32 saddr = skb->nh.iph->saddr;
 685         const __u32 daddr = skb->nh.iph->daddr;
 686         const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
 687         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 688         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 689         struct dst_entry *dst = NULL;
 690
 691         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 692         if (((struct rtable *)skb->dst)->rt_flags &
 693             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 694                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 695                 goto drop;
 696         }
 697
 698         if (dccp_bad_service_code(sk, service)) {
 699                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 700                 goto drop;
 701         }
 702         /*
 703          * TW buckets are converted to open requests without
 704          * limitations, they conserve resources and peer is
 705          * evidently real one.
 706          */
 707         if (inet_csk_reqsk_queue_is_full(sk))
 708                 goto drop;
 709
 710         /*
 711          * Accept backlog is full. If we have already queued enough
 712          * of warm entries in syn queue, drop request. It is better than
 713          * clogging syn queue with openreqs with exponentially increasing
 714          * timeout.
 715          */
 716         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 717                 goto drop;
 718
 719         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 720         if (req == NULL)
 721                 goto drop;
 722
 723         /* FIXME: process options */
 724
 725         dccp_openreq_init(req, &dp, skb);
 726
 727         ireq = inet_rsk(req);
 728         ireq->loc_addr = daddr;
 729         ireq->rmt_addr = saddr;
 730         /* FIXME: Merge Aristeu's option parsing code when ready */
 731         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 732                                   right value */
 733         ireq->opt       = NULL;
 734
 735         /*
 736          * Step 3: Process LISTEN state
 737          *
 738          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 739          *
 740          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 741          * dccp_create_openreq_child.
 742          */
 743         dreq = dccp_rsk(req);
 744         dreq->dreq_isr     = dcb->dccpd_seq;
 745         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 746         dreq->dreq_service = service;
 747
 748         if (dccp_v4_send_response(sk, req, dst))
 749                 goto drop_and_free;
 750
 751         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 752         return 0;
 753
 754 drop_and_free:
 755         /*
 756          * FIXME: should be reqsk_free after implementing req->rsk_ops
 757          */
 758         __reqsk_free(req);
 759 drop:
 760         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 761         dcb->dccpd_reset_code = reset_code;
 762         return -1;
 763 }
 764
 765 /*
 766  * The three way handshake has completed - we got a valid ACK or DATAACK -
 767  * now create the new socket.
 768  *
 769  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 770  */
 771 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 772                                        struct request_sock *req,
 773                                        struct dst_entry *dst)
 774 {
 775         struct inet_request_sock *ireq;
 776         struct inet_sock *newinet;
 777         struct dccp_sock *newdp;
 778         struct sock *newsk;
 779
 780         if (sk_acceptq_is_full(sk))
 781                 goto exit_overflow;
 782
 783         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 784                 goto exit;
 785
 786         newsk = dccp_create_openreq_child(sk, req, skb);
 787         if (newsk == NULL)
 788                 goto exit;
 789
 790         sk_setup_caps(newsk, dst);
 791
 792         newdp              = dccp_sk(newsk);
 793         newinet            = inet_sk(newsk);
 794         ireq               = inet_rsk(req);
 795         newinet->daddr     = ireq->rmt_addr;
 796         newinet->rcv_saddr = ireq->loc_addr;
 797         newinet->saddr     = ireq->loc_addr;
 798         newinet->opt       = ireq->opt;
 799         ireq->opt          = NULL;
 800         newinet->mc_index  = inet_iif(skb);
 801         newinet->mc_ttl    = skb->nh.iph->ttl;
 802         newinet->id        = jiffies;
 803
 804         dccp_sync_mss(newsk, dst_mtu(dst));
 805
 806         __inet_hash(&dccp_hashinfo, newsk, 0);
 807         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 808
 809         return newsk;
 810
 811 exit_overflow:
 812         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 813 exit:
 814         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 815         dst_release(dst);
 816         return NULL;
 817 }
 818
 819 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 820 {
 821         const struct dccp_hdr *dh = dccp_hdr(skb);
 822         const struct iphdr *iph = skb->nh.iph;
 823         struct sock *nsk;
 824         struct request_sock **prev;
 825         /* Find possible connection requests. */
 826         struct request_sock *req = inet_csk_search_req(sk, &prev,
 827                                                        dh->dccph_sport,
 828                                                        iph->saddr, iph->daddr);
 829         if (req != NULL)
 830                 return dccp_check_req(sk, skb, req, prev);
 831
 832         nsk = __inet_lookup_established(&dccp_hashinfo,
 833                                         iph->saddr, dh->dccph_sport,
 834                                         iph->daddr, ntohs(dh->dccph_dport),
 835                                         inet_iif(skb));
 836         if (nsk != NULL) {
 837                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 838                         bh_lock_sock(nsk);
 839                         return nsk;
 840                 }
 841                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 842                 return NULL;
 843         }
 844
 845         return sk;
 846 }
 847
 848 int dccp_v4_checksum(const struct sk_buff *skb, const u32 saddr,
 849                      const u32 daddr)
 850 {
 851         const struct dccp_hdr* dh = dccp_hdr(skb);
 852         int checksum_len;
 853         u32 tmp;
 854
 855         if (dh->dccph_cscov == 0)
 856                 checksum_len = skb->len;
 857         else {
 858                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 859                 checksum_len = checksum_len < skb->len ? checksum_len :
 860                                                          skb->len;
 861         }
 862
 863         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 864         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 865                                  IPPROTO_DCCP, tmp);
 866 }
 867
 868 static int dccp_v4_verify_checksum(struct sk_buff *skb,
 869                                    const u32 saddr, const u32 daddr)
 870 {
 871         struct dccp_hdr *dh = dccp_hdr(skb);
 872         int checksum_len;
 873         u32 tmp;
 874
 875         if (dh->dccph_cscov == 0)
 876                 checksum_len = skb->len;
 877         else {
 878                 checksum_len = (dh->dccph_cscov + dh->dccph_x) * sizeof(u32);
 879                 checksum_len = checksum_len < skb->len ? checksum_len :
 880                                                          skb->len;
 881         }
 882         tmp = csum_partial((unsigned char *)dh, checksum_len, 0);
 883         return csum_tcpudp_magic(saddr, daddr, checksum_len,
 884                                  IPPROTO_DCCP, tmp) == 0 ? 0 : -1;
 885 }
 886
 887 static struct dst_entry* dccp_v4_route_skb(struct sock *sk,
 888                                            struct sk_buff *skb)
 889 {
 890         struct rtable *rt;
 891         struct flowi fl = { .oif = ((struct rtable *)skb->dst)->rt_iif,
 892                             .nl_u = { .ip4_u =
 893                                       { .daddr = skb->nh.iph->saddr,
 894                                         .saddr = skb->nh.iph->daddr,
 895                                         .tos = RT_CONN_FLAGS(sk) } },
 896                             .proto = sk->sk_protocol,
 897                             .uli_u = { .ports =
 898                                        { .sport = dccp_hdr(skb)->dccph_dport,
 899                                          .dport = dccp_hdr(skb)->dccph_sport }
 900                                      }
 901                           };
 902
 903         if (ip_route_output_flow(&rt, &fl, sk, 0)) {
 904                 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
 905                 return NULL;
 906         }
 907
 908         return &rt->u.dst;
 909 }
 910
 911 static void dccp_v4_ctl_send_reset(struct sk_buff *rxskb)
 912 {
 913         int err;
 914         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 915         const int dccp_hdr_reset_len = sizeof(struct dccp_hdr) +
 916                                        sizeof(struct dccp_hdr_ext) +
 917                                        sizeof(struct dccp_hdr_reset);
 918         struct sk_buff *skb;
 919         struct dst_entry *dst;
 920         u64 seqno;
 921
 922         /* Never send a reset in response to a reset. */
 923         if (rxdh->dccph_type == DCCP_PKT_RESET)
 924                 return;
 925
 926         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 927                 return;
 928
 929         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 930         if (dst == NULL)
 931                 return;
 932
 933         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 934         if (skb == NULL)
 935                 goto out;
 936
 937         /* Reserve space for headers. */
 938         skb_reserve(skb, MAX_DCCP_HEADER);
 939         skb->dst = dst_clone(dst);
 940
 941         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 942         dh = dccp_hdr(skb);
 943         memset(dh, 0, dccp_hdr_reset_len);
 944
 945         /* Build DCCP header and checksum it. */
 946         dh->dccph_type     = DCCP_PKT_RESET;
 947         dh->dccph_sport    = rxdh->dccph_dport;
 948         dh->dccph_dport    = rxdh->dccph_sport;
 949         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 950         dh->dccph_x        = 1;
 951         dccp_hdr_reset(skb)->dccph_reset_code =
 952                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 953
 954         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 955         seqno = 0;
 956         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 957                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 958
 959         dccp_hdr_set_seq(dh, seqno);
 960         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 961                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 962
 963         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 964                                               rxskb->nh.iph->daddr);
 965
 966         bh_lock_sock(dccp_ctl_socket->sk);
 967         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 968                                     rxskb->nh.iph->daddr,
 969                                     rxskb->nh.iph->saddr, NULL);
 970         bh_unlock_sock(dccp_ctl_socket->sk);
 971
 972         if (err == NET_XMIT_CN || err == 0) {
 973                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 974                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 975         }
 976 out:
 977          dst_release(dst);
 978 }
 979
 980 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 981 {
 982         struct dccp_hdr *dh = dccp_hdr(skb);
 983
 984         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 985                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 986                         goto reset;
 987                 return 0;
 988         }
 989
 990         /*
 991          *  Step 3: Process LISTEN state
 992          *     If S.state == LISTEN,
 993          *        If P.type == Request or P contains a valid Init Cookie
 994          *              option,
 995          *           * Must scan the packet's options to check for an Init
 996          *              Cookie.  Only the Init Cookie is processed here,
 997          *              however; other options are processed in Step 8.  This
 998          *              scan need only be performed if the endpoint uses Init
 999          *              Cookies *
1000          *           * Generate a new socket and switch to that socket *
1001          *           Set S := new socket for this port pair
1002          *           S.state = RESPOND
1003          *           Choose S.ISS (initial seqno) or set from Init Cookie
1004          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
1005          *           Continue with S.state == RESPOND
1006          *           * A Response packet will be generated in Step 11 *
1007          *        Otherwise,
1008          *           Generate Reset(No Connection) unless P.type == Reset
1009          *           Drop packet and return
1010          *
1011          * NOTE: the check for the packet types is done in
1012          *       dccp_rcv_state_process
1013          */
1014         if (sk->sk_state == DCCP_LISTEN) {
1015                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
1016
1017                 if (nsk == NULL)
1018                         goto discard;
1019
1020                 if (nsk != sk) {
1021                         if (dccp_child_process(sk, nsk, skb))
1022                                 goto reset;
1023                         return 0;
1024                 }
1025         }
1026
1027         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
1028                 goto reset;
1029         return 0;
1030
1031 reset:
1032         dccp_v4_ctl_send_reset(skb);
1033 discard:
1034         kfree_skb(skb);
1035         return 0;
1036 }
1037
1038 static inline int dccp_invalid_packet(struct sk_buff *skb)
1039 {
1040         const struct dccp_hdr *dh;
1041
1042         if (skb->pkt_type != PACKET_HOST)
1043                 return 1;
1044
1045         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
1046                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
1047                 return 1;
1048         }
1049
1050         dh = dccp_hdr(skb);
1051
1052         /* If the packet type is not understood, drop packet and return */
1053         if (dh->dccph_type >= DCCP_PKT_INVALID) {
1054                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
1055                 return 1;
1056         }
1057
1058         /*
1059          * If P.Data Offset is too small for packet type, or too large for
1060          * packet, drop packet and return
1061          */
1062         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
1063                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1064                                             "too small 1\n",
1065                                dh->dccph_doff);
1066                 return 1;
1067         }
1068
1069         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
1070                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1071                                             "too small 2\n",
1072                                dh->dccph_doff);
1073                 return 1;
1074         }
1075
1076         dh = dccp_hdr(skb);
1077
1078         /*
1079          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
1080          * has short sequence numbers), drop packet and return
1081          */
1082         if (dh->dccph_x == 0 &&
1083             dh->dccph_type != DCCP_PKT_DATA &&
1084             dh->dccph_type != DCCP_PKT_ACK &&
1085             dh->dccph_type != DCCP_PKT_DATAACK) {
1086                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
1087                                             "nor DataAck and P.X == 0\n",
1088                                dccp_packet_name(dh->dccph_type));
1089                 return 1;
1090         }
1091
1092         /* If the header checksum is incorrect, drop packet and return */
1093         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
1094                                     skb->nh.iph->daddr) < 0) {
1095                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: header checksum is "
1096                                             "incorrect\n");
1097                 return 1;
1098         }
1099
1100         return 0;
1101 }
1102
1103 /* this is called when real data arrives */
1104 int dccp_v4_rcv(struct sk_buff *skb)
1105 {
1106         const struct dccp_hdr *dh;
1107         struct sock *sk;
1108         int rc;
1109
1110         /* Step 1: Check header basics: */
1111
1112         if (dccp_invalid_packet(skb))
1113                 goto discard_it;
1114
1115         dh = dccp_hdr(skb);
1116
1117         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
1118         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
1119
1120         dccp_pr_debug("%8.8s "
1121                       "src=%u.%u.%u.%u@%-5d "
1122                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
1123                       dccp_packet_name(dh->dccph_type),
1124                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
1125                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
1126                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
1127
1128         if (dccp_packet_without_ack(skb)) {
1129                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
1130                 dccp_pr_debug_cat("\n");
1131         } else {
1132                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
1133                 dccp_pr_debug_cat(", ack=%llu\n",
1134                                   (unsigned long long)
1135                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
1136         }
1137
1138         /* Step 2:
1139          *      Look up flow ID in table and get corresponding socket */
1140         sk = __inet_lookup(&dccp_hashinfo,
1141                            skb->nh.iph->saddr, dh->dccph_sport,
1142                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
1143                            inet_iif(skb));
1144
1145         /*
1146          * Step 2:
1147          *      If no socket ...
1148          *              Generate Reset(No Connection) unless P.type == Reset
1149          *              Drop packet and return
1150          */
1151         if (sk == NULL) {
1152                 dccp_pr_debug("failed to look up flow ID in table and "
1153                               "get corresponding socket\n");
1154                 goto no_dccp_socket;
1155         }
1156
1157         /*
1158          * Step 2:
1159          *      ... or S.state == TIMEWAIT,
1160          *              Generate Reset(No Connection) unless P.type == Reset
1161          *              Drop packet and return
1162          */
1163
1164         if (sk->sk_state == DCCP_TIME_WAIT) {
1165                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
1166                               "do_time_wait\n");
1167                 goto do_time_wait;
1168         }
1169
1170         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
1171                 dccp_pr_debug("xfrm4_policy_check failed\n");
1172                 goto discard_and_relse;
1173         }
1174
1175         if (sk_filter(sk, skb, 0)) {
1176                 dccp_pr_debug("sk_filter failed\n");
1177                 goto discard_and_relse;
1178         }
1179
1180         skb->dev = NULL;
1181
1182         bh_lock_sock(sk);
1183         rc = 0;
1184         if (!sock_owned_by_user(sk))
1185                 rc = dccp_v4_do_rcv(sk, skb);
1186         else
1187                 sk_add_backlog(sk, skb);
1188         bh_unlock_sock(sk);
1189
1190         sock_put(sk);
1191         return rc;
1192
1193 no_dccp_socket:
1194         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1195                 goto discard_it;
1196         /*
1197          * Step 2:
1198          *              Generate Reset(No Connection) unless P.type == Reset
1199          *              Drop packet and return
1200          */
1201         if (dh->dccph_type != DCCP_PKT_RESET) {
1202                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1203                                         DCCP_RESET_CODE_NO_CONNECTION;
1204                 dccp_v4_ctl_send_reset(skb);
1205         }
1206
1207 discard_it:
1208         /* Discard frame. */
1209         kfree_skb(skb);
1210         return 0;
1211
1212 discard_and_relse:
1213         sock_put(sk);
1214         goto discard_it;
1215
1216 do_time_wait:
1217         inet_twsk_put((struct inet_timewait_sock *)sk);
1218         goto no_dccp_socket;
1219 }
1220
1221 static int dccp_v4_init_sock(struct sock *sk)
1222 {
1223         struct dccp_sock *dp = dccp_sk(sk);
1224         static int dccp_ctl_socket_init = 1;
1225
1226         dccp_options_init(&dp->dccps_options);
1227         do_gettimeofday(&dp->dccps_epoch);
1228
1229         if (dp->dccps_options.dccpo_send_ack_vector) {
1230                 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
1231                                                            GFP_KERNEL);
1232                 if (dp->dccps_hc_rx_ackvec == NULL)
1233                         return -ENOMEM;
1234         }
1235
1236         /*
1237          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1238          * the listening (master) sock get CCID control blocks, which is not
1239          * necessary, but for now, to not mess with the test userspace apps,
1240          * lets leave it here, later the real solution is to do this in a
1241          * setsockopt(CCIDs-I-want/accept). -acme
1242          */
1243         if (likely(!dccp_ctl_socket_init)) {
1244                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1245                                                  sk);
1246                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1247                                                  sk);
1248                 if (dp->dccps_hc_rx_ccid == NULL ||
1249                     dp->dccps_hc_tx_ccid == NULL) {
1250                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1251                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1252                         if (dp->dccps_options.dccpo_send_ack_vector) {
1253                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1254                                 dp->dccps_hc_rx_ackvec = NULL;
1255                         }
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         sk->sk_write_space = dccp_write_space;
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_qrite_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         if (dp->dccps_service_list != NULL) {
1291                 kfree(dp->dccps_service_list);
1292                 dp->dccps_service_list = NULL;
1293         }
1294
1295         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1296         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1297         if (dp->dccps_options.dccpo_send_ack_vector) {
1298                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1299                 dp->dccps_hc_rx_ackvec = NULL;
1300         }
1301         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1302         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1303         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1304
1305         return 0;
1306 }
1307
1308 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1309 {
1310         kfree(inet_rsk(req)->opt);
1311 }
1312
1313 static struct request_sock_ops dccp_request_sock_ops = {
1314         .family         = PF_INET,
1315         .obj_size       = sizeof(struct dccp_request_sock),
1316         .rtx_syn_ack    = dccp_v4_send_response,
1317         .send_ack       = dccp_v4_reqsk_send_ack,
1318         .destructor     = dccp_v4_reqsk_destructor,
1319         .send_reset     = dccp_v4_ctl_send_reset,
1320 };
1321
1322 struct proto dccp_v4_prot = {
1323         .name                   = "DCCP",
1324         .owner                  = THIS_MODULE,
1325         .close                  = dccp_close,
1326         .connect                = dccp_v4_connect,
1327         .disconnect             = dccp_disconnect,
1328         .ioctl                  = dccp_ioctl,
1329         .init                   = dccp_v4_init_sock,
1330         .setsockopt             = dccp_setsockopt,
1331         .getsockopt             = dccp_getsockopt,
1332         .sendmsg                = dccp_sendmsg,
1333         .recvmsg                = dccp_recvmsg,
1334         .backlog_rcv            = dccp_v4_do_rcv,
1335         .hash                   = dccp_v4_hash,
1336         .unhash                 = dccp_v4_unhash,
1337         .accept                 = inet_csk_accept,
1338         .get_port               = dccp_v4_get_port,
1339         .shutdown               = dccp_shutdown,
1340         .destroy                = dccp_v4_destroy_sock,
1341         .orphan_count           = &dccp_orphan_count,
1342         .max_header             = MAX_DCCP_HEADER,
1343         .obj_size               = sizeof(struct dccp_sock),
1344         .rsk_prot               = &dccp_request_sock_ops,
1345         .twsk_obj_size          = sizeof(struct inet_timewait_sock),
1346 };