net/dccp/ipv4.c

   1 /*
   2  *  net/dccp/ipv4.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/icmp.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/random.h>
  19
  20 #include <net/icmp.h>
  21 #include <net/inet_hashtables.h>
  22 #include <net/sock.h>
  23 #include <net/tcp_states.h>
  24 #include <net/xfrm.h>
  25
  26 #include "ackvec.h"
  27 #include "ccid.h"
  28 #include "dccp.h"
  29
  30 struct inet_hashinfo __cacheline_aligned dccp_hashinfo = {
  31         .lhash_lock     = RW_LOCK_UNLOCKED,
  32         .lhash_users    = ATOMIC_INIT(0),
  33         .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(dccp_hashinfo.lhash_wait),
  34 };
  35
  36 EXPORT_SYMBOL_GPL(dccp_hashinfo);
  37
  38 static int dccp_v4_get_port(struct sock *sk, const unsigned short snum)
  39 {
  40         return inet_csk_get_port(&dccp_hashinfo, sk, snum,
  41                                  inet_csk_bind_conflict);
  42 }
  43
  44 static void dccp_v4_hash(struct sock *sk)
  45 {
  46         inet_hash(&dccp_hashinfo, sk);
  47 }
  48
  49 void dccp_unhash(struct sock *sk)
  50 {
  51         inet_unhash(&dccp_hashinfo, sk);
  52 }
  53
  54 EXPORT_SYMBOL_GPL(dccp_unhash);
  55
  56 /* called with local bh disabled */
  57 static int __dccp_v4_check_established(struct sock *sk, const __u16 lport,
  58                                       struct inet_timewait_sock **twp)
  59 {
  60         struct inet_sock *inet = inet_sk(sk);
  61         const u32 daddr = inet->rcv_saddr;
  62         const u32 saddr = inet->daddr;
  63         const int dif = sk->sk_bound_dev_if;
  64         INET_ADDR_COOKIE(acookie, saddr, daddr)
  65         const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
  66         unsigned int hash = inet_ehashfn(daddr, lport, saddr, inet->dport);
  67         struct inet_ehash_bucket *head = inet_ehash_bucket(&dccp_hashinfo, hash);
  68         const struct sock *sk2;
  69         const struct hlist_node *node;
  70         struct inet_timewait_sock *tw;
  71
  72         prefetch(head->chain.first);
  73         write_lock(&head->lock);
  74
  75         /* Check TIME-WAIT sockets first. */
  76         sk_for_each(sk2, node, &(head + dccp_hashinfo.ehash_size)->chain) {
  77                 tw = inet_twsk(sk2);
  78
  79                 if (INET_TW_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
  80                         goto not_unique;
  81         }
  82         tw = NULL;
  83
  84         /* And established part... */
  85         sk_for_each(sk2, node, &head->chain) {
  86                 if (INET_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
  87                         goto not_unique;
  88         }
  89
  90         /* Must record num and sport now. Otherwise we will see
  91          * in hash table socket with a funny identity. */
  92         inet->num = lport;
  93         inet->sport = htons(lport);
  94         sk->sk_hash = hash;
  95         BUG_TRAP(sk_unhashed(sk));
  96         __sk_add_node(sk, &head->chain);
  97         sock_prot_inc_use(sk->sk_prot);
  98         write_unlock(&head->lock);
  99
 100         if (twp != NULL) {
 101                 *twp = tw;
 102                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 103         } else if (tw != NULL) {
 104                 /* Silly. Should hash-dance instead... */
 105                 inet_twsk_deschedule(tw, &dccp_death_row);
 106                 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
 107
 108                 inet_twsk_put(tw);
 109         }
 110
 111         return 0;
 112
 113 not_unique:
 114         write_unlock(&head->lock);
 115         return -EADDRNOTAVAIL;
 116 }
 117
 118 /*
 119  * Bind a port for a connect operation and hash it.
 120  */
 121 static int dccp_v4_hash_connect(struct sock *sk)
 122 {
 123         const unsigned short snum = inet_sk(sk)->num;
 124         struct inet_bind_hashbucket *head;
 125         struct inet_bind_bucket *tb;
 126         int ret;
 127
 128         if (snum == 0) {
 129                 int low = sysctl_local_port_range[0];
 130                 int high = sysctl_local_port_range[1];
 131                 int remaining = (high - low) + 1;
 132                 int rover = net_random() % (high - low) + low;
 133                 struct hlist_node *node;
 134                 struct inet_timewait_sock *tw = NULL;
 135
 136                 local_bh_disable();
 137                 do {
 138                         head = &dccp_hashinfo.bhash[inet_bhashfn(rover,
 139                                                     dccp_hashinfo.bhash_size)];
 140                         spin_lock(&head->lock);
 141
 142                         /* Does not bother with rcv_saddr checks,
 143                          * because the established check is already
 144                          * unique enough.
 145                          */
 146                         inet_bind_bucket_for_each(tb, node, &head->chain) {
 147                                 if (tb->port == rover) {
 148                                         BUG_TRAP(!hlist_empty(&tb->owners));
 149                                         if (tb->fastreuse >= 0)
 150                                                 goto next_port;
 151                                         if (!__dccp_v4_check_established(sk,
 152                                                                          rover,
 153                                                                          &tw))
 154                                                 goto ok;
 155                                         goto next_port;
 156                                 }
 157                         }
 158
 159                         tb = inet_bind_bucket_create(dccp_hashinfo.bind_bucket_cachep,
 160                                                      head, rover);
 161                         if (tb == NULL) {
 162                                 spin_unlock(&head->lock);
 163                                 break;
 164                         }
 165                         tb->fastreuse = -1;
 166                         goto ok;
 167
 168                 next_port:
 169                         spin_unlock(&head->lock);
 170                         if (++rover > high)
 171                                 rover = low;
 172                 } while (--remaining > 0);
 173
 174                 local_bh_enable();
 175
 176                 return -EADDRNOTAVAIL;
 177
 178 ok:
 179                 /* All locks still held and bhs disabled */
 180                 inet_bind_hash(sk, tb, rover);
 181                 if (sk_unhashed(sk)) {
 182                         inet_sk(sk)->sport = htons(rover);
 183                         __inet_hash(&dccp_hashinfo, sk, 0);
 184                 }
 185                 spin_unlock(&head->lock);
 186
 187                 if (tw != NULL) {
 188                         inet_twsk_deschedule(tw, &dccp_death_row);
 189                         inet_twsk_put(tw);
 190                 }
 191
 192                 ret = 0;
 193                 goto out;
 194         }
 195
 196         head = &dccp_hashinfo.bhash[inet_bhashfn(snum,
 197                                                  dccp_hashinfo.bhash_size)];
 198         tb   = inet_csk(sk)->icsk_bind_hash;
 199         spin_lock_bh(&head->lock);
 200         if (sk_head(&tb->owners) == sk && sk->sk_bind_node.next == NULL) {
 201                 __inet_hash(&dccp_hashinfo, sk, 0);
 202                 spin_unlock_bh(&head->lock);
 203                 return 0;
 204         } else {
 205                 spin_unlock(&head->lock);
 206                 /* No definite answer... Walk to established hash table */
 207                 ret = __dccp_v4_check_established(sk, snum, NULL);
 208 out:
 209                 local_bh_enable();
 210                 return ret;
 211         }
 212 }
 213
 214 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 215 {
 216         struct inet_sock *inet = inet_sk(sk);
 217         struct dccp_sock *dp = dccp_sk(sk);
 218         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 219         struct rtable *rt;
 220         u32 daddr, nexthop;
 221         int tmp;
 222         int err;
 223
 224         dp->dccps_role = DCCP_ROLE_CLIENT;
 225
 226         if (dccp_service_not_initialized(sk))
 227                 return -EPROTO;
 228
 229         if (addr_len < sizeof(struct sockaddr_in))
 230                 return -EINVAL;
 231
 232         if (usin->sin_family != AF_INET)
 233                 return -EAFNOSUPPORT;
 234
 235         nexthop = daddr = usin->sin_addr.s_addr;
 236         if (inet->opt != NULL && inet->opt->srr) {
 237                 if (daddr == 0)
 238                         return -EINVAL;
 239                 nexthop = inet->opt->faddr;
 240         }
 241
 242         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
 243                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 244                                IPPROTO_DCCP,
 245                                inet->sport, usin->sin_port, sk);
 246         if (tmp < 0)
 247                 return tmp;
 248
 249         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 250                 ip_rt_put(rt);
 251                 return -ENETUNREACH;
 252         }
 253
 254         if (inet->opt == NULL || !inet->opt->srr)
 255                 daddr = rt->rt_dst;
 256
 257         if (inet->saddr == 0)
 258                 inet->saddr = rt->rt_src;
 259         inet->rcv_saddr = inet->saddr;
 260
 261         inet->dport = usin->sin_port;
 262         inet->daddr = daddr;
 263
 264         dp->dccps_ext_header_len = 0;
 265         if (inet->opt != NULL)
 266                 dp->dccps_ext_header_len = inet->opt->optlen;
 267         /*
 268          * Socket identity is still unknown (sport may be zero).
 269          * However we set state to DCCP_REQUESTING and not releasing socket
 270          * lock select source port, enter ourselves into the hash tables and
 271          * complete initialization after this.
 272          */
 273         dccp_set_state(sk, DCCP_REQUESTING);
 274         err = dccp_v4_hash_connect(sk);
 275         if (err != 0)
 276                 goto failure;
 277
 278         err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
 279         if (err != 0)
 280                 goto failure;
 281
 282         /* OK, now commit destination to socket.  */
 283         sk_setup_caps(sk, &rt->u.dst);
 284
 285         dp->dccps_gar =
 286                 dp->dccps_iss = secure_dccp_sequence_number(inet->saddr,
 287                                                             inet->daddr,
 288                                                             inet->sport,
 289                                                             usin->sin_port);
 290         dccp_update_gss(sk, dp->dccps_iss);
 291
 292         inet->id = dp->dccps_iss ^ jiffies;
 293
 294         err = dccp_connect(sk);
 295         rt = NULL;
 296         if (err != 0)
 297                 goto failure;
 298 out:
 299         return err;
 300 failure:
 301         /*
 302          * This unhashes the socket and releases the local port, if necessary.
 303          */
 304         dccp_set_state(sk, DCCP_CLOSED);
 305         ip_rt_put(rt);
 306         sk->sk_route_caps = 0;
 307         inet->dport = 0;
 308         goto out;
 309 }
 310
 311 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 312
 313 /*
 314  * This routine does path mtu discovery as defined in RFC1191.
 315  */
 316 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 317                                           const struct iphdr *iph,
 318                                           u32 mtu)
 319 {
 320         struct dst_entry *dst;
 321         const struct inet_sock *inet = inet_sk(sk);
 322         const struct dccp_sock *dp = dccp_sk(sk);
 323
 324         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 325          * send out by Linux are always < 576bytes so they should go through
 326          * unfragmented).
 327          */
 328         if (sk->sk_state == DCCP_LISTEN)
 329                 return;
 330
 331         /* We don't check in the destentry if pmtu discovery is forbidden
 332          * on this route. We just assume that no packet_to_big packets
 333          * are send back when pmtu discovery is not active.
 334          * There is a small race when the user changes this flag in the
 335          * route, but I think that's acceptable.
 336          */
 337         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 338                 return;
 339
 340         dst->ops->update_pmtu(dst, mtu);
 341
 342         /* Something is about to be wrong... Remember soft error
 343          * for the case, if this connection will not able to recover.
 344          */
 345         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 346                 sk->sk_err_soft = EMSGSIZE;
 347
 348         mtu = dst_mtu(dst);
 349
 350         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 351             dp->dccps_pmtu_cookie > mtu) {
 352                 dccp_sync_mss(sk, mtu);
 353
 354                 /*
 355                  * From: draft-ietf-dccp-spec-11.txt
 356                  *
 357                  *      DCCP-Sync packets are the best choice for upward
 358                  *      probing, since DCCP-Sync probes do not risk application
 359                  *      data loss.
 360                  */
 361                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 362         } /* else let the usual retransmit timer handle it */
 363 }
 364
 365 static void dccp_v4_ctl_send_ack(struct sk_buff *rxskb)
 366 {
 367         int err;
 368         struct dccp_hdr *rxdh = dccp_hdr(rxskb), *dh;
 369         const int dccp_hdr_ack_len = sizeof(struct dccp_hdr) +
 370                                      sizeof(struct dccp_hdr_ext) +
 371                                      sizeof(struct dccp_hdr_ack_bits);
 372         struct sk_buff *skb;
 373
 374         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 375                 return;
 376
 377         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 378         if (skb == NULL)
 379                 return;
 380
 381         /* Reserve space for headers. */
 382         skb_reserve(skb, MAX_DCCP_HEADER);
 383
 384         skb->dst = dst_clone(rxskb->dst);
 385
 386         skb->h.raw = skb_push(skb, dccp_hdr_ack_len);
 387         dh = dccp_hdr(skb);
 388         memset(dh, 0, dccp_hdr_ack_len);
 389
 390         /* Build DCCP header and checksum it. */
 391         dh->dccph_type     = DCCP_PKT_ACK;
 392         dh->dccph_sport    = rxdh->dccph_dport;
 393         dh->dccph_dport    = rxdh->dccph_sport;
 394         dh->dccph_doff     = dccp_hdr_ack_len / 4;
 395         dh->dccph_x        = 1;
 396
 397         dccp_hdr_set_seq(dh, DCCP_SKB_CB(rxskb)->dccpd_ack_seq);
 398         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 399                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 400
 401         bh_lock_sock(dccp_ctl_socket->sk);
 402         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 403                                     rxskb->nh.iph->daddr,
 404                                     rxskb->nh.iph->saddr, NULL);
 405         bh_unlock_sock(dccp_ctl_socket->sk);
 406
 407         if (err == NET_XMIT_CN || err == 0) {
 408                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 409                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 410         }
 411 }
 412
 413 static void dccp_v4_reqsk_send_ack(struct sk_buff *skb,
 414                                    struct request_sock *req)
 415 {
 416         dccp_v4_ctl_send_ack(skb);
 417 }
 418
 419 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 420                                  struct dst_entry *dst)
 421 {
 422         int err = -1;
 423         struct sk_buff *skb;
 424
 425         /* First, grab a route. */
 426
 427         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 428                 goto out;
 429
 430         skb = dccp_make_response(sk, dst, req);
 431         if (skb != NULL) {
 432                 const struct inet_request_sock *ireq = inet_rsk(req);
 433
 434                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 435                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 436                                             ireq->rmt_addr,
 437                                             ireq->opt);
 438                 if (err == NET_XMIT_CN)
 439                         err = 0;
 440         }
 441
 442 out:
 443         dst_release(dst);
 444         return err;
 445 }
 446
 447 /*
 448  * This routine is called by the ICMP module when it gets some sort of error
 449  * condition. If err < 0 then the socket should be closed and the error
 450  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 451  * After adjustment header points to the first 8 bytes of the tcp header. We
 452  * need to find the appropriate port.
 453  *
 454  * The locking strategy used here is very "optimistic". When someone else
 455  * accesses the socket the ICMP is just dropped and for some paths there is no
 456  * check at all. A more general error queue to queue errors for later handling
 457  * is probably better.
 458  */
 459 void dccp_v4_err(struct sk_buff *skb, u32 info)
 460 {
 461         const struct iphdr *iph = (struct iphdr *)skb->data;
 462         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data +
 463                                                         (iph->ihl << 2));
 464         struct dccp_sock *dp;
 465         struct inet_sock *inet;
 466         const int type = skb->h.icmph->type;
 467         const int code = skb->h.icmph->code;
 468         struct sock *sk;
 469         __u64 seq;
 470         int err;
 471
 472         if (skb->len < (iph->ihl << 2) + 8) {
 473                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 474                 return;
 475         }
 476
 477         sk = inet_lookup(&dccp_hashinfo, iph->daddr, dh->dccph_dport,
 478                          iph->saddr, dh->dccph_sport, inet_iif(skb));
 479         if (sk == NULL) {
 480                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 481                 return;
 482         }
 483
 484         if (sk->sk_state == DCCP_TIME_WAIT) {
 485                 inet_twsk_put((struct inet_timewait_sock *)sk);
 486                 return;
 487         }
 488
 489         bh_lock_sock(sk);
 490         /* If too many ICMPs get dropped on busy
 491          * servers this needs to be solved differently.
 492          */
 493         if (sock_owned_by_user(sk))
 494                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 495
 496         if (sk->sk_state == DCCP_CLOSED)
 497                 goto out;
 498
 499         dp = dccp_sk(sk);
 500         seq = dccp_hdr_seq(skb);
 501         if (sk->sk_state != DCCP_LISTEN &&
 502             !between48(seq, dp->dccps_swl, dp->dccps_swh)) {
 503                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 504                 goto out;
 505         }
 506
 507         switch (type) {
 508         case ICMP_SOURCE_QUENCH:
 509                 /* Just silently ignore these. */
 510                 goto out;
 511         case ICMP_PARAMETERPROB:
 512                 err = EPROTO;
 513                 break;
 514         case ICMP_DEST_UNREACH:
 515                 if (code > NR_ICMP_UNREACH)
 516                         goto out;
 517
 518                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 519                         if (!sock_owned_by_user(sk))
 520                                 dccp_do_pmtu_discovery(sk, iph, info);
 521                         goto out;
 522                 }
 523
 524                 err = icmp_err_convert[code].errno;
 525                 break;
 526         case ICMP_TIME_EXCEEDED:
 527                 err = EHOSTUNREACH;
 528                 break;
 529         default:
 530                 goto out;
 531         }
 532
 533         switch (sk->sk_state) {
 534                 struct request_sock *req , **prev;
 535         case DCCP_LISTEN:
 536                 if (sock_owned_by_user(sk))
 537                         goto out;
 538                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 539                                           iph->daddr, iph->saddr);
 540                 if (!req)
 541                         goto out;
 542
 543                 /*
 544                  * ICMPs are not backlogged, hence we cannot get an established
 545                  * socket here.
 546                  */
 547                 BUG_TRAP(!req->sk);
 548
 549                 if (seq != dccp_rsk(req)->dreq_iss) {
 550                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 551                         goto out;
 552                 }
 553                 /*
 554                  * Still in RESPOND, just remove it silently.
 555                  * There is no good way to pass the error to the newly
 556                  * created socket, and POSIX does not want network
 557                  * errors returned from accept().
 558                  */
 559                 inet_csk_reqsk_queue_drop(sk, req, prev);
 560                 goto out;
 561
 562         case DCCP_REQUESTING:
 563         case DCCP_RESPOND:
 564                 if (!sock_owned_by_user(sk)) {
 565                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 566                         sk->sk_err = err;
 567
 568                         sk->sk_error_report(sk);
 569
 570                         dccp_done(sk);
 571                 } else
 572                         sk->sk_err_soft = err;
 573                 goto out;
 574         }
 575
 576         /* If we've already connected we will keep trying
 577          * until we time out, or the user gives up.
 578          *
 579          * rfc1122 4.2.3.9 allows to consider as hard errors
 580          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 581          * but it is obsoleted by pmtu discovery).
 582          *
 583          * Note, that in modern internet, where routing is unreliable
 584          * and in each dark corner broken firewalls sit, sending random
 585          * errors ordered by their masters even this two messages finally lose
 586          * their original sense (even Linux sends invalid PORT_UNREACHs)
 587          *
 588          * Now we are in compliance with RFCs.
 589          *                                                      --ANK (980905)
 590          */
 591
 592         inet = inet_sk(sk);
 593         if (!sock_owned_by_user(sk) && inet->recverr) {
 594                 sk->sk_err = err;
 595                 sk->sk_error_report(sk);
 596         } else /* Only an error on timeout */
 597                 sk->sk_err_soft = err;
 598 out:
 599         bh_unlock_sock(sk);
 600         sock_put(sk);
 601 }
 602
 603 /* This routine computes an IPv4 DCCP checksum. */
 604 void dccp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 605 {
 606         const struct inet_sock *inet = inet_sk(sk);
 607         struct dccp_hdr *dh = dccp_hdr(skb);
 608
 609         dh->dccph_checksum = dccp_v4_checksum(skb, inet->saddr, inet->daddr);
 610 }
 611
 612 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 613
 614 int dccp_v4_send_reset(struct sock *sk, enum dccp_reset_codes code)
 615 {
 616         struct sk_buff *skb;
 617         /*
 618          * FIXME: what if rebuild_header fails?
 619          * Should we be doing a rebuild_header here?
 620          */
 621         int err = inet_sk_rebuild_header(sk);
 622
 623         if (err != 0)
 624                 return err;
 625
 626         skb = dccp_make_reset(sk, sk->sk_dst_cache, code);
 627         if (skb != NULL) {
 628                 const struct inet_sock *inet = inet_sk(sk);
 629
 630                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 631                 err = ip_build_and_send_pkt(skb, sk,
 632                                             inet->saddr, inet->daddr, NULL);
 633                 if (err == NET_XMIT_CN)
 634                         err = 0;
 635         }
 636
 637         return err;
 638 }
 639
 640 static inline u64 dccp_v4_init_sequence(const struct sock *sk,
 641                                         const struct sk_buff *skb)
 642 {
 643         return secure_dccp_sequence_number(skb->nh.iph->daddr,
 644                                            skb->nh.iph->saddr,
 645                                            dccp_hdr(skb)->dccph_dport,
 646                                            dccp_hdr(skb)->dccph_sport);
 647 }
 648
 649 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 650 {
 651         struct inet_request_sock *ireq;
 652         struct dccp_sock dp;
 653         struct request_sock *req;
 654         struct dccp_request_sock *dreq;
 655         const __u32 saddr = skb->nh.iph->saddr;
 656         const __u32 daddr = skb->nh.iph->daddr;
 657         const __u32 service = dccp_hdr_request(skb)->dccph_req_service;
 658         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 659         __u8 reset_code = DCCP_RESET_CODE_TOO_BUSY;
 660
 661         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 662         if (((struct rtable *)skb->dst)->rt_flags &
 663             (RTCF_BROADCAST | RTCF_MULTICAST)) {
 664                 reset_code = DCCP_RESET_CODE_NO_CONNECTION;
 665                 goto drop;
 666         }
 667
 668         if (dccp_bad_service_code(sk, service)) {
 669                 reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 670                 goto drop;
 671         }
 672         /*
 673          * TW buckets are converted to open requests without
 674          * limitations, they conserve resources and peer is
 675          * evidently real one.
 676          */
 677         if (inet_csk_reqsk_queue_is_full(sk))
 678                 goto drop;
 679
 680         /*
 681          * Accept backlog is full. If we have already queued enough
 682          * of warm entries in syn queue, drop request. It is better than
 683          * clogging syn queue with openreqs with exponentially increasing
 684          * timeout.
 685          */
 686         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 687                 goto drop;
 688
 689         req = reqsk_alloc(sk->sk_prot->rsk_prot);
 690         if (req == NULL)
 691                 goto drop;
 692
 693         /* FIXME: process options */
 694
 695         dccp_openreq_init(req, &dp, skb);
 696
 697         ireq = inet_rsk(req);
 698         ireq->loc_addr = daddr;
 699         ireq->rmt_addr = saddr;
 700         req->rcv_wnd    = 100; /* Fake, option parsing will get the
 701                                   right value */
 702         ireq->opt       = NULL;
 703
 704         /*
 705          * Step 3: Process LISTEN state
 706          *
 707          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 708          *
 709          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 710          * dccp_create_openreq_child.
 711          */
 712         dreq = dccp_rsk(req);
 713         dreq->dreq_isr     = dcb->dccpd_seq;
 714         dreq->dreq_iss     = dccp_v4_init_sequence(sk, skb);
 715         dreq->dreq_service = service;
 716
 717         if (dccp_v4_send_response(sk, req, NULL))
 718                 goto drop_and_free;
 719
 720         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 721         return 0;
 722
 723 drop_and_free:
 724         /*
 725          * FIXME: should be reqsk_free after implementing req->rsk_ops
 726          */
 727         __reqsk_free(req);
 728 drop:
 729         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 730         dcb->dccpd_reset_code = reset_code;
 731         return -1;
 732 }
 733
 734 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 735
 736 /*
 737  * The three way handshake has completed - we got a valid ACK or DATAACK -
 738  * now create the new socket.
 739  *
 740  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 741  */
 742 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 743                                        struct request_sock *req,
 744                                        struct dst_entry *dst)
 745 {
 746         struct inet_request_sock *ireq;
 747         struct inet_sock *newinet;
 748         struct dccp_sock *newdp;
 749         struct sock *newsk;
 750
 751         if (sk_acceptq_is_full(sk))
 752                 goto exit_overflow;
 753
 754         if (dst == NULL && (dst = inet_csk_route_req(sk, req)) == NULL)
 755                 goto exit;
 756
 757         newsk = dccp_create_openreq_child(sk, req, skb);
 758         if (newsk == NULL)
 759                 goto exit;
 760
 761         sk_setup_caps(newsk, dst);
 762
 763         newdp              = dccp_sk(newsk);
 764         newinet            = inet_sk(newsk);
 765         ireq               = inet_rsk(req);
 766         newinet->daddr     = ireq->rmt_addr;
 767         newinet->rcv_saddr = ireq->loc_addr;
 768         newinet->saddr     = ireq->loc_addr;
 769         newinet->opt       = ireq->opt;
 770         ireq->opt          = NULL;
 771         newinet->mc_index  = inet_iif(skb);
 772         newinet->mc_ttl    = skb->nh.iph->ttl;
 773         newinet->id        = jiffies;
 774
 775         dccp_sync_mss(newsk, dst_mtu(dst));
 776
 777         __inet_hash(&dccp_hashinfo, newsk, 0);
 778         __inet_inherit_port(&dccp_hashinfo, sk, newsk);
 779
 780         return newsk;
 781
 782 exit_overflow:
 783         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 784 exit:
 785         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 786         dst_release(dst);
 787         return NULL;
 788 }
 789
 790 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 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 static 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         u64 seqno;
 894
 895         /* Never send a reset in response to a reset. */
 896         if (rxdh->dccph_type == DCCP_PKT_RESET)
 897                 return;
 898
 899         if (((struct rtable *)rxskb->dst)->rt_type != RTN_LOCAL)
 900                 return;
 901
 902         dst = dccp_v4_route_skb(dccp_ctl_socket->sk, rxskb);
 903         if (dst == NULL)
 904                 return;
 905
 906         skb = alloc_skb(MAX_DCCP_HEADER + 15, GFP_ATOMIC);
 907         if (skb == NULL)
 908                 goto out;
 909
 910         /* Reserve space for headers. */
 911         skb_reserve(skb, MAX_DCCP_HEADER);
 912         skb->dst = dst_clone(dst);
 913
 914         skb->h.raw = skb_push(skb, dccp_hdr_reset_len);
 915         dh = dccp_hdr(skb);
 916         memset(dh, 0, dccp_hdr_reset_len);
 917
 918         /* Build DCCP header and checksum it. */
 919         dh->dccph_type     = DCCP_PKT_RESET;
 920         dh->dccph_sport    = rxdh->dccph_dport;
 921         dh->dccph_dport    = rxdh->dccph_sport;
 922         dh->dccph_doff     = dccp_hdr_reset_len / 4;
 923         dh->dccph_x        = 1;
 924         dccp_hdr_reset(skb)->dccph_reset_code =
 925                                 DCCP_SKB_CB(rxskb)->dccpd_reset_code;
 926
 927         /* See "8.3.1. Abnormal Termination" in draft-ietf-dccp-spec-11 */
 928         seqno = 0;
 929         if (DCCP_SKB_CB(rxskb)->dccpd_ack_seq != DCCP_PKT_WITHOUT_ACK_SEQ)
 930                 dccp_set_seqno(&seqno, DCCP_SKB_CB(rxskb)->dccpd_ack_seq + 1);
 931
 932         dccp_hdr_set_seq(dh, seqno);
 933         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb),
 934                          DCCP_SKB_CB(rxskb)->dccpd_seq);
 935
 936         dh->dccph_checksum = dccp_v4_checksum(skb, rxskb->nh.iph->saddr,
 937                                               rxskb->nh.iph->daddr);
 938
 939         bh_lock_sock(dccp_ctl_socket->sk);
 940         err = ip_build_and_send_pkt(skb, dccp_ctl_socket->sk,
 941                                     rxskb->nh.iph->daddr,
 942                                     rxskb->nh.iph->saddr, NULL);
 943         bh_unlock_sock(dccp_ctl_socket->sk);
 944
 945         if (err == NET_XMIT_CN || err == 0) {
 946                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 947                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 948         }
 949 out:
 950          dst_release(dst);
 951 }
 952
 953 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 954 {
 955         struct dccp_hdr *dh = dccp_hdr(skb);
 956
 957         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 958                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 959                         goto reset;
 960                 return 0;
 961         }
 962
 963         /*
 964          *  Step 3: Process LISTEN state
 965          *     If S.state == LISTEN,
 966          *        If P.type == Request or P contains a valid Init Cookie
 967          *              option,
 968          *           * Must scan the packet's options to check for an Init
 969          *              Cookie.  Only the Init Cookie is processed here,
 970          *              however; other options are processed in Step 8.  This
 971          *              scan need only be performed if the endpoint uses Init
 972          *              Cookies *
 973          *           * Generate a new socket and switch to that socket *
 974          *           Set S := new socket for this port pair
 975          *           S.state = RESPOND
 976          *           Choose S.ISS (initial seqno) or set from Init Cookie
 977          *           Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 978          *           Continue with S.state == RESPOND
 979          *           * A Response packet will be generated in Step 11 *
 980          *        Otherwise,
 981          *           Generate Reset(No Connection) unless P.type == Reset
 982          *           Drop packet and return
 983          *
 984          * NOTE: the check for the packet types is done in
 985          *       dccp_rcv_state_process
 986          */
 987         if (sk->sk_state == DCCP_LISTEN) {
 988                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 989
 990                 if (nsk == NULL)
 991                         goto discard;
 992
 993                 if (nsk != sk) {
 994                         if (dccp_child_process(sk, nsk, skb))
 995                                 goto reset;
 996                         return 0;
 997                 }
 998         }
 999
1000         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
1001                 goto reset;
1002         return 0;
1003
1004 reset:
1005         dccp_v4_ctl_send_reset(skb);
1006 discard:
1007         kfree_skb(skb);
1008         return 0;
1009 }
1010
1011 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
1012
1013 int dccp_invalid_packet(struct sk_buff *skb)
1014 {
1015         const struct dccp_hdr *dh;
1016
1017         if (skb->pkt_type != PACKET_HOST)
1018                 return 1;
1019
1020         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
1021                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: pskb_may_pull failed\n");
1022                 return 1;
1023         }
1024
1025         dh = dccp_hdr(skb);
1026
1027         /* If the packet type is not understood, drop packet and return */
1028         if (dh->dccph_type >= DCCP_PKT_INVALID) {
1029                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: invalid packet type\n");
1030                 return 1;
1031         }
1032
1033         /*
1034          * If P.Data Offset is too small for packet type, or too large for
1035          * packet, drop packet and return
1036          */
1037         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
1038                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1039                                             "too small 1\n",
1040                                dh->dccph_doff);
1041                 return 1;
1042         }
1043
1044         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
1045                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.Data Offset(%u) "
1046                                             "too small 2\n",
1047                                dh->dccph_doff);
1048                 return 1;
1049         }
1050
1051         dh = dccp_hdr(skb);
1052
1053         /*
1054          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
1055          * has short sequence numbers), drop packet and return
1056          */
1057         if (dh->dccph_x == 0 &&
1058             dh->dccph_type != DCCP_PKT_DATA &&
1059             dh->dccph_type != DCCP_PKT_ACK &&
1060             dh->dccph_type != DCCP_PKT_DATAACK) {
1061                 LIMIT_NETDEBUG(KERN_WARNING "DCCP: P.type (%s) not Data, Ack "
1062                                             "nor DataAck and P.X == 0\n",
1063                                dccp_packet_name(dh->dccph_type));
1064                 return 1;
1065         }
1066
1067         return 0;
1068 }
1069
1070 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
1071
1072 /* this is called when real data arrives */
1073 int dccp_v4_rcv(struct sk_buff *skb)
1074 {
1075         const struct dccp_hdr *dh;
1076         struct sock *sk;
1077         int rc;
1078
1079         /* Step 1: Check header basics: */
1080
1081         if (dccp_invalid_packet(skb))
1082                 goto discard_it;
1083
1084         /* If the header checksum is incorrect, drop packet and return */
1085         if (dccp_v4_verify_checksum(skb, skb->nh.iph->saddr,
1086                                     skb->nh.iph->daddr) < 0) {
1087                 LIMIT_NETDEBUG(KERN_WARNING "%s: incorrect header checksum\n",
1088                                __FUNCTION__);
1089                 goto discard_it;
1090         }
1091
1092         dh = dccp_hdr(skb);
1093
1094         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(skb);
1095         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
1096
1097         dccp_pr_debug("%8.8s "
1098                       "src=%u.%u.%u.%u@%-5d "
1099                       "dst=%u.%u.%u.%u@%-5d seq=%llu",
1100                       dccp_packet_name(dh->dccph_type),
1101                       NIPQUAD(skb->nh.iph->saddr), ntohs(dh->dccph_sport),
1102                       NIPQUAD(skb->nh.iph->daddr), ntohs(dh->dccph_dport),
1103                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
1104
1105         if (dccp_packet_without_ack(skb)) {
1106                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
1107                 dccp_pr_debug_cat("\n");
1108         } else {
1109                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
1110                 dccp_pr_debug_cat(", ack=%llu\n",
1111                                   (unsigned long long)
1112                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
1113         }
1114
1115         /* Step 2:
1116          *      Look up flow ID in table and get corresponding socket */
1117         sk = __inet_lookup(&dccp_hashinfo,
1118                            skb->nh.iph->saddr, dh->dccph_sport,
1119                            skb->nh.iph->daddr, ntohs(dh->dccph_dport),
1120                            inet_iif(skb));
1121
1122         /*
1123          * Step 2:
1124          *      If no socket ...
1125          *              Generate Reset(No Connection) unless P.type == Reset
1126          *              Drop packet and return
1127          */
1128         if (sk == NULL) {
1129                 dccp_pr_debug("failed to look up flow ID in table and "
1130                               "get corresponding socket\n");
1131                 goto no_dccp_socket;
1132         }
1133
1134         /*
1135          * Step 2:
1136          *      ... or S.state == TIMEWAIT,
1137          *              Generate Reset(No Connection) unless P.type == Reset
1138          *              Drop packet and return
1139          */
1140
1141         if (sk->sk_state == DCCP_TIME_WAIT) {
1142                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: "
1143                               "do_time_wait\n");
1144                 goto do_time_wait;
1145         }
1146
1147         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
1148                 dccp_pr_debug("xfrm4_policy_check failed\n");
1149                 goto discard_and_relse;
1150         }
1151
1152         if (sk_filter(sk, skb, 0)) {
1153                 dccp_pr_debug("sk_filter failed\n");
1154                 goto discard_and_relse;
1155         }
1156
1157         skb->dev = NULL;
1158
1159         bh_lock_sock(sk);
1160         rc = 0;
1161         if (!sock_owned_by_user(sk))
1162                 rc = dccp_v4_do_rcv(sk, skb);
1163         else
1164                 sk_add_backlog(sk, skb);
1165         bh_unlock_sock(sk);
1166
1167         sock_put(sk);
1168         return rc;
1169
1170 no_dccp_socket:
1171         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1172                 goto discard_it;
1173         /*
1174          * Step 2:
1175          *              Generate Reset(No Connection) unless P.type == Reset
1176          *              Drop packet and return
1177          */
1178         if (dh->dccph_type != DCCP_PKT_RESET) {
1179                 DCCP_SKB_CB(skb)->dccpd_reset_code =
1180                                         DCCP_RESET_CODE_NO_CONNECTION;
1181                 dccp_v4_ctl_send_reset(skb);
1182         }
1183
1184 discard_it:
1185         /* Discard frame. */
1186         kfree_skb(skb);
1187         return 0;
1188
1189 discard_and_relse:
1190         sock_put(sk);
1191         goto discard_it;
1192
1193 do_time_wait:
1194         inet_twsk_put((struct inet_timewait_sock *)sk);
1195         goto no_dccp_socket;
1196 }
1197
1198 struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
1199         .queue_xmit     = ip_queue_xmit,
1200         .send_check     = dccp_v4_send_check,
1201         .rebuild_header = inet_sk_rebuild_header,
1202         .conn_request   = dccp_v4_conn_request,
1203         .syn_recv_sock  = dccp_v4_request_recv_sock,
1204         .net_header_len = sizeof(struct iphdr),
1205         .setsockopt     = ip_setsockopt,
1206         .getsockopt     = ip_getsockopt,
1207         .addr2sockaddr  = inet_csk_addr2sockaddr,
1208         .sockaddr_len   = sizeof(struct sockaddr_in),
1209 };
1210
1211 int dccp_v4_init_sock(struct sock *sk)
1212 {
1213         struct dccp_sock *dp = dccp_sk(sk);
1214         static int dccp_ctl_socket_init = 1;
1215
1216         dccp_options_init(&dp->dccps_options);
1217         do_gettimeofday(&dp->dccps_epoch);
1218
1219         if (dp->dccps_options.dccpo_send_ack_vector) {
1220                 dp->dccps_hc_rx_ackvec = dccp_ackvec_alloc(DCCP_MAX_ACKVEC_LEN,
1221                                                            GFP_KERNEL);
1222                 if (dp->dccps_hc_rx_ackvec == NULL)
1223                         return -ENOMEM;
1224         }
1225
1226         /*
1227          * FIXME: We're hardcoding the CCID, and doing this at this point makes
1228          * the listening (master) sock get CCID control blocks, which is not
1229          * necessary, but for now, to not mess with the test userspace apps,
1230          * lets leave it here, later the real solution is to do this in a
1231          * setsockopt(CCIDs-I-want/accept). -acme
1232          */
1233         if (likely(!dccp_ctl_socket_init)) {
1234                 dp->dccps_hc_rx_ccid = ccid_init(dp->dccps_options.dccpo_rx_ccid,
1235                                                  sk);
1236                 dp->dccps_hc_tx_ccid = ccid_init(dp->dccps_options.dccpo_tx_ccid,
1237                                                  sk);
1238                 if (dp->dccps_hc_rx_ccid == NULL ||
1239                     dp->dccps_hc_tx_ccid == NULL) {
1240                         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1241                         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1242                         if (dp->dccps_options.dccpo_send_ack_vector) {
1243                                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1244                                 dp->dccps_hc_rx_ackvec = NULL;
1245                         }
1246                         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1247                         return -ENOMEM;
1248                 }
1249         } else
1250                 dccp_ctl_socket_init = 0;
1251
1252         dccp_init_xmit_timers(sk);
1253         inet_csk(sk)->icsk_rto = DCCP_TIMEOUT_INIT;
1254         sk->sk_state = DCCP_CLOSED;
1255         sk->sk_write_space = dccp_write_space;
1256         inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
1257         dp->dccps_mss_cache = 536;
1258         dp->dccps_role = DCCP_ROLE_UNDEFINED;
1259         dp->dccps_service = DCCP_SERVICE_INVALID_VALUE;
1260
1261         return 0;
1262 }
1263
1264 EXPORT_SYMBOL_GPL(dccp_v4_init_sock);
1265
1266 int dccp_v4_destroy_sock(struct sock *sk)
1267 {
1268         struct dccp_sock *dp = dccp_sk(sk);
1269
1270         /*
1271          * DCCP doesn't use sk_write_queue, just sk_send_head
1272          * for retransmissions
1273          */
1274         if (sk->sk_send_head != NULL) {
1275                 kfree_skb(sk->sk_send_head);
1276                 sk->sk_send_head = NULL;
1277         }
1278
1279         /* Clean up a referenced DCCP bind bucket. */
1280         if (inet_csk(sk)->icsk_bind_hash != NULL)
1281                 inet_put_port(&dccp_hashinfo, sk);
1282
1283         kfree(dp->dccps_service_list);
1284         dp->dccps_service_list = NULL;
1285
1286         ccid_hc_rx_exit(dp->dccps_hc_rx_ccid, sk);
1287         ccid_hc_tx_exit(dp->dccps_hc_tx_ccid, sk);
1288         if (dp->dccps_options.dccpo_send_ack_vector) {
1289                 dccp_ackvec_free(dp->dccps_hc_rx_ackvec);
1290                 dp->dccps_hc_rx_ackvec = NULL;
1291         }
1292         ccid_exit(dp->dccps_hc_rx_ccid, sk);
1293         ccid_exit(dp->dccps_hc_tx_ccid, sk);
1294         dp->dccps_hc_rx_ccid = dp->dccps_hc_tx_ccid = NULL;
1295
1296         return 0;
1297 }
1298
1299 EXPORT_SYMBOL_GPL(dccp_v4_destroy_sock);
1300
1301 static void dccp_v4_reqsk_destructor(struct request_sock *req)
1302 {
1303         kfree(inet_rsk(req)->opt);
1304 }
1305
1306 static struct request_sock_ops dccp_request_sock_ops = {
1307         .family         = PF_INET,
1308         .obj_size       = sizeof(struct dccp_request_sock),
1309         .rtx_syn_ack    = dccp_v4_send_response,
1310         .send_ack       = dccp_v4_reqsk_send_ack,
1311         .destructor     = dccp_v4_reqsk_destructor,
1312         .send_reset     = dccp_v4_ctl_send_reset,
1313 };
1314
1315 struct proto dccp_prot = {
1316         .name                   = "DCCP",
1317         .owner                  = THIS_MODULE,
1318         .close                  = dccp_close,
1319         .connect                = dccp_v4_connect,
1320         .disconnect             = dccp_disconnect,
1321         .ioctl                  = dccp_ioctl,
1322         .init                   = dccp_v4_init_sock,
1323         .setsockopt             = dccp_setsockopt,
1324         .getsockopt             = dccp_getsockopt,
1325         .sendmsg                = dccp_sendmsg,
1326         .recvmsg                = dccp_recvmsg,
1327         .backlog_rcv            = dccp_v4_do_rcv,
1328         .hash                   = dccp_v4_hash,
1329         .unhash                 = dccp_unhash,
1330         .accept                 = inet_csk_accept,
1331         .get_port               = dccp_v4_get_port,
1332         .shutdown               = dccp_shutdown,
1333         .destroy                = dccp_v4_destroy_sock,
1334         .orphan_count           = &dccp_orphan_count,
1335         .max_header             = MAX_DCCP_HEADER,
1336         .obj_size               = sizeof(struct dccp_sock),
1337         .rsk_prot               = &dccp_request_sock_ops,
1338         .twsk_obj_size          = sizeof(struct inet_timewait_sock),
1339 };