net/ipv4/tcp_ipv4.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Implementation of the Transmission Control Protocol(TCP).
   7  *
   8  * Version:     $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $
   9  *
  10  *              IPv4 specific functions
  11  *
  12  *
  13  *              code split from:
  14  *              linux/ipv4/tcp.c
  15  *              linux/ipv4/tcp_input.c
  16  *              linux/ipv4/tcp_output.c
  17  *
  18  *              See tcp.c for author information
  19  *
  20  *      This program is free software; you can redistribute it and/or
  21  *      modify it under the terms of the GNU General Public License
  22  *      as published by the Free Software Foundation; either version
  23  *      2 of the License, or (at your option) any later version.
  24  */
  25
  26 /*
  27  * Changes:
  28  *              David S. Miller :       New socket lookup architecture.
  29  *                                      This code is dedicated to John Dyson.
  30  *              David S. Miller :       Change semantics of established hash,
  31  *                                      half is devoted to TIME_WAIT sockets
  32  *                                      and the rest go in the other half.
  33  *              Andi Kleen :            Add support for syncookies and fixed
  34  *                                      some bugs: ip options weren't passed to
  35  *                                      the TCP layer, missed a check for an
  36  *                                      ACK bit.
  37  *              Andi Kleen :            Implemented fast path mtu discovery.
  38  *                                      Fixed many serious bugs in the
  39  *                                      request_sock handling and moved
  40  *                                      most of it into the af independent code.
  41  *                                      Added tail drop and some other bugfixes.
  42  *                                      Added new listen semantics.
  43  *              Mike McLagan    :       Routing by source
  44  *      Juan Jose Ciarlante:            ip_dynaddr bits
  45  *              Andi Kleen:             various fixes.
  46  *      Vitaly E. Lavrov        :       Transparent proxy revived after year
  47  *                                      coma.
  48  *      Andi Kleen              :       Fix new listen.
  49  *      Andi Kleen              :       Fix accept error reporting.
  50  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
  51  *      Alexey Kuznetsov                allow both IPv4 and IPv6 sockets to bind
  52  *                                      a single port at the same time.
  53  */
  54
  55 #include <linux/config.h>
  56
  57 #include <linux/types.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/module.h>
  60 #include <linux/random.h>
  61 #include <linux/cache.h>
  62 #include <linux/jhash.h>
  63 #include <linux/init.h>
  64 #include <linux/times.h>
  65
  66 #include <net/icmp.h>
  67 #include <net/inet_hashtables.h>
  68 #include <net/tcp.h>
  69 #include <net/transp_v6.h>
  70 #include <net/ipv6.h>
  71 #include <net/inet_common.h>
  72 #include <net/timewait_sock.h>
  73 #include <net/xfrm.h>
  74
  75 #include <linux/inet.h>
  76 #include <linux/ipv6.h>
  77 #include <linux/stddef.h>
  78 #include <linux/proc_fs.h>
  79 #include <linux/seq_file.h>
  80 #include <linux/vserver/debug.h>
  81
  82 int sysctl_tcp_tw_reuse;
  83 int sysctl_tcp_low_latency;
  84
  85 /* Check TCP sequence numbers in ICMP packets. */
  86 #define ICMP_MIN_LENGTH 8
  87
  88 /* Socket used for sending RSTs */
  89 static struct socket *tcp_socket;
  90
  91 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb);
  92
  93 struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {
  94         .lhash_lock     = RW_LOCK_UNLOCKED,
  95         .lhash_users    = ATOMIC_INIT(0),
  96         .lhash_wait     = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait),
  97 };
  98
  99 static int tcp_v4_get_port(struct sock *sk, unsigned short snum)
 100 {
 101         return inet_csk_get_port(&tcp_hashinfo, sk, snum,
 102                                  inet_csk_bind_conflict);
 103 }
 104
 105 static void tcp_v4_hash(struct sock *sk)
 106 {
 107         inet_hash(&tcp_hashinfo, sk);
 108 }
 109
 110 void tcp_unhash(struct sock *sk)
 111 {
 112         inet_unhash(&tcp_hashinfo, sk);
 113 }
 114
 115 static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb)
 116 {
 117         return secure_tcp_sequence_number(skb->nh.iph->daddr,
 118                                           skb->nh.iph->saddr,
 119                                           skb->h.th->dest,
 120                                           skb->h.th->source);
 121 }
 122
 123 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
 124 {
 125         const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
 126         struct tcp_sock *tp = tcp_sk(sk);
 127
 128         /* With PAWS, it is safe from the viewpoint
 129            of data integrity. Even without PAWS it is safe provided sequence
 130            spaces do not overlap i.e. at data rates <= 80Mbit/sec.
 131
 132            Actually, the idea is close to VJ's one, only timestamp cache is
 133            held not per host, but per port pair and TW bucket is used as state
 134            holder.
 135
 136            If TW bucket has been already destroyed we fall back to VJ's scheme
 137            and use initial timestamp retrieved from peer table.
 138          */
 139         if (tcptw->tw_ts_recent_stamp &&
 140             (twp == NULL || (sysctl_tcp_tw_reuse &&
 141                              xtime.tv_sec - tcptw->tw_ts_recent_stamp > 1))) {
 142                 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
 143                 if (tp->write_seq == 0)
 144                         tp->write_seq = 1;
 145                 tp->rx_opt.ts_recent       = tcptw->tw_ts_recent;
 146                 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 147                 sock_hold(sktw);
 148                 return 1;
 149         }
 150
 151         return 0;
 152 }
 153
 154 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
 155
 156 /* This will initiate an outgoing connection. */
 157 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 158 {
 159         struct inet_sock *inet = inet_sk(sk);
 160         struct tcp_sock *tp = tcp_sk(sk);
 161         struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 162         struct rtable *rt;
 163         u32 daddr, nexthop;
 164         int tmp;
 165         int err;
 166
 167         if (addr_len < sizeof(struct sockaddr_in))
 168                 return -EINVAL;
 169
 170         if (usin->sin_family != AF_INET)
 171                 return -EAFNOSUPPORT;
 172
 173         nexthop = daddr = usin->sin_addr.s_addr;
 174         if (inet->opt && inet->opt->srr) {
 175                 if (!daddr)
 176                         return -EINVAL;
 177                 nexthop = inet->opt->faddr;
 178         }
 179
 180         tmp = ip_route_connect(&rt, nexthop, inet->saddr,
 181                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 182                                IPPROTO_TCP,
 183                                inet->sport, usin->sin_port, sk);
 184         if (tmp < 0)
 185                 return tmp;
 186
 187         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 188                 ip_rt_put(rt);
 189                 return -ENETUNREACH;
 190         }
 191
 192         if (!inet->opt || !inet->opt->srr)
 193                 daddr = rt->rt_dst;
 194
 195         if (!inet->saddr)
 196                 inet->saddr = rt->rt_src;
 197         inet->rcv_saddr = inet->saddr;
 198
 199         if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) {
 200                 /* Reset inherited state */
 201                 tp->rx_opt.ts_recent       = 0;
 202                 tp->rx_opt.ts_recent_stamp = 0;
 203                 tp->write_seq              = 0;
 204         }
 205
 206         if (tcp_death_row.sysctl_tw_recycle &&
 207             !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
 208                 struct inet_peer *peer = rt_get_peer(rt);
 209
 210                 /* VJ's idea. We save last timestamp seen from
 211                  * the destination in peer table, when entering state TIME-WAIT
 212                  * and initialize rx_opt.ts_recent from it, when trying new connection.
 213                  */
 214
 215                 if (peer && peer->tcp_ts_stamp + TCP_PAWS_MSL >= xtime.tv_sec) {
 216                         tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
 217                         tp->rx_opt.ts_recent = peer->tcp_ts;
 218                 }
 219         }
 220
 221         inet->dport = usin->sin_port;
 222         inet->daddr = daddr;
 223
 224         inet_csk(sk)->icsk_ext_hdr_len = 0;
 225         if (inet->opt)
 226                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
 227
 228         tp->rx_opt.mss_clamp = 536;
 229
 230         /* Socket identity is still unknown (sport may be zero).
 231          * However we set state to SYN-SENT and not releasing socket
 232          * lock select source port, enter ourselves into the hash tables and
 233          * complete initialization after this.
 234          */
 235         tcp_set_state(sk, TCP_SYN_SENT);
 236         err = inet_hash_connect(&tcp_death_row, sk);
 237         if (err)
 238                 goto failure;
 239
 240         err = ip_route_newports(&rt, IPPROTO_TCP, inet->sport, inet->dport, sk);
 241         if (err)
 242                 goto failure;
 243
 244         /* OK, now commit destination to socket.  */
 245         sk_setup_caps(sk, &rt->u.dst);
 246
 247         if (!tp->write_seq)
 248                 tp->write_seq = secure_tcp_sequence_number(inet->saddr,
 249                                                            inet->daddr,
 250                                                            inet->sport,
 251                                                            usin->sin_port);
 252
 253         inet->id = tp->write_seq ^ jiffies;
 254
 255         err = tcp_connect(sk);
 256         rt = NULL;
 257         if (err)
 258                 goto failure;
 259
 260         return 0;
 261
 262 failure:
 263         /* This unhashes the socket and releases the local port, if necessary. */
 264         tcp_set_state(sk, TCP_CLOSE);
 265         ip_rt_put(rt);
 266         sk->sk_route_caps = 0;
 267         inet->dport = 0;
 268         return err;
 269 }
 270
 271 /*
 272  * This routine does path mtu discovery as defined in RFC1191.
 273  */
 274 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
 275 {
 276         struct dst_entry *dst;
 277         struct inet_sock *inet = inet_sk(sk);
 278
 279         /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
 280          * send out by Linux are always <576bytes so they should go through
 281          * unfragmented).
 282          */
 283         if (sk->sk_state == TCP_LISTEN)
 284                 return;
 285
 286         /* We don't check in the destentry if pmtu discovery is forbidden
 287          * on this route. We just assume that no packet_to_big packets
 288          * are send back when pmtu discovery is not active.
 289          * There is a small race when the user changes this flag in the
 290          * route, but I think that's acceptable.
 291          */
 292         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 293                 return;
 294
 295         dst->ops->update_pmtu(dst, mtu);
 296
 297         /* Something is about to be wrong... Remember soft error
 298          * for the case, if this connection will not able to recover.
 299          */
 300         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 301                 sk->sk_err_soft = EMSGSIZE;
 302
 303         mtu = dst_mtu(dst);
 304
 305         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 306             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 307                 tcp_sync_mss(sk, mtu);
 308
 309                 /* Resend the TCP packet because it's
 310                  * clear that the old packet has been
 311                  * dropped. This is the new "fast" path mtu
 312                  * discovery.
 313                  */
 314                 tcp_simple_retransmit(sk);
 315         } /* else let the usual retransmit timer handle it */
 316 }
 317
 318 /*
 319  * This routine is called by the ICMP module when it gets some
 320  * sort of error condition.  If err < 0 then the socket should
 321  * be closed and the error returned to the user.  If err > 0
 322  * it's just the icmp type << 8 | icmp code.  After adjustment
 323  * header points to the first 8 bytes of the tcp header.  We need
 324  * to find the appropriate port.
 325  *
 326  * The locking strategy used here is very "optimistic". When
 327  * someone else accesses the socket the ICMP is just dropped
 328  * and for some paths there is no check at all.
 329  * A more general error queue to queue errors for later handling
 330  * is probably better.
 331  *
 332  */
 333
 334 void tcp_v4_err(struct sk_buff *skb, u32 info)
 335 {
 336         struct iphdr *iph = (struct iphdr *)skb->data;
 337         struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
 338         struct tcp_sock *tp;
 339         struct inet_sock *inet;
 340         int type = skb->h.icmph->type;
 341         int code = skb->h.icmph->code;
 342         struct sock *sk;
 343         __u32 seq;
 344         int err;
 345
 346         if (skb->len < (iph->ihl << 2) + 8) {
 347                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 348                 return;
 349         }
 350
 351         sk = inet_lookup(&tcp_hashinfo, iph->daddr, th->dest, iph->saddr,
 352                          th->source, inet_iif(skb));
 353         if (!sk) {
 354                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 355                 return;
 356         }
 357         if (sk->sk_state == TCP_TIME_WAIT) {
 358                 inet_twsk_put((struct inet_timewait_sock *)sk);
 359                 return;
 360         }
 361
 362         bh_lock_sock(sk);
 363         /* If too many ICMPs get dropped on busy
 364          * servers this needs to be solved differently.
 365          */
 366         if (sock_owned_by_user(sk))
 367                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 368
 369         if (sk->sk_state == TCP_CLOSE)
 370                 goto out;
 371
 372         tp = tcp_sk(sk);
 373         seq = ntohl(th->seq);
 374         if (sk->sk_state != TCP_LISTEN &&
 375             !between(seq, tp->snd_una, tp->snd_nxt)) {
 376                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 377                 goto out;
 378         }
 379
 380         switch (type) {
 381         case ICMP_SOURCE_QUENCH:
 382                 /* Just silently ignore these. */
 383                 goto out;
 384         case ICMP_PARAMETERPROB:
 385                 err = EPROTO;
 386                 break;
 387         case ICMP_DEST_UNREACH:
 388                 if (code > NR_ICMP_UNREACH)
 389                         goto out;
 390
 391                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 392                         if (!sock_owned_by_user(sk))
 393                                 do_pmtu_discovery(sk, iph, info);
 394                         goto out;
 395                 }
 396
 397                 err = icmp_err_convert[code].errno;
 398                 break;
 399         case ICMP_TIME_EXCEEDED:
 400                 err = EHOSTUNREACH;
 401                 break;
 402         default:
 403                 goto out;
 404         }
 405
 406         switch (sk->sk_state) {
 407                 struct request_sock *req, **prev;
 408         case TCP_LISTEN:
 409                 if (sock_owned_by_user(sk))
 410                         goto out;
 411
 412                 req = inet_csk_search_req(sk, &prev, th->dest,
 413                                           iph->daddr, iph->saddr);
 414                 if (!req)
 415                         goto out;
 416
 417                 /* ICMPs are not backlogged, hence we cannot get
 418                    an established socket here.
 419                  */
 420                 BUG_TRAP(!req->sk);
 421
 422                 if (seq != tcp_rsk(req)->snt_isn) {
 423                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 424                         goto out;
 425                 }
 426
 427                 /*
 428                  * Still in SYN_RECV, just remove it silently.
 429                  * There is no good way to pass the error to the newly
 430                  * created socket, and POSIX does not want network
 431                  * errors returned from accept().
 432                  */
 433                 inet_csk_reqsk_queue_drop(sk, req, prev);
 434                 goto out;
 435
 436         case TCP_SYN_SENT:
 437         case TCP_SYN_RECV:  /* Cannot happen.
 438                                It can f.e. if SYNs crossed.
 439                              */
 440                 if (!sock_owned_by_user(sk)) {
 441                         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
 442                         sk->sk_err = err;
 443
 444                         sk->sk_error_report(sk);
 445
 446                         tcp_done(sk);
 447                 } else {
 448                         sk->sk_err_soft = err;
 449                 }
 450                 goto out;
 451         }
 452
 453         /* If we've already connected we will keep trying
 454          * until we time out, or the user gives up.
 455          *
 456          * rfc1122 4.2.3.9 allows to consider as hard errors
 457          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 458          * but it is obsoleted by pmtu discovery).
 459          *
 460          * Note, that in modern internet, where routing is unreliable
 461          * and in each dark corner broken firewalls sit, sending random
 462          * errors ordered by their masters even this two messages finally lose
 463          * their original sense (even Linux sends invalid PORT_UNREACHs)
 464          *
 465          * Now we are in compliance with RFCs.
 466          *                                                      --ANK (980905)
 467          */
 468
 469         inet = inet_sk(sk);
 470         if (!sock_owned_by_user(sk) && inet->recverr) {
 471                 sk->sk_err = err;
 472                 sk->sk_error_report(sk);
 473         } else  { /* Only an error on timeout */
 474                 sk->sk_err_soft = err;
 475         }
 476
 477 out:
 478         bh_unlock_sock(sk);
 479         sock_put(sk);
 480 }
 481
 482 /* This routine computes an IPv4 TCP checksum. */
 483 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
 484 {
 485         struct inet_sock *inet = inet_sk(sk);
 486         struct tcphdr *th = skb->h.th;
 487
 488         if (skb->ip_summed == CHECKSUM_HW) {
 489                 th->check = ~tcp_v4_check(th, len, inet->saddr, inet->daddr, 0);
 490                 skb->csum = offsetof(struct tcphdr, check);
 491         } else {
 492                 th->check = tcp_v4_check(th, len, inet->saddr, inet->daddr,
 493                                          csum_partial((char *)th,
 494                                                       th->doff << 2,
 495                                                       skb->csum));
 496         }
 497 }
 498
 499 int tcp_v4_gso_send_check(struct sk_buff *skb)
 500 {
 501         struct iphdr *iph;
 502         struct tcphdr *th;
 503
 504         if (!pskb_may_pull(skb, sizeof(*th)))
 505                 return -EINVAL;
 506
 507         iph = skb->nh.iph;
 508         th = skb->h.th;
 509
 510         th->check = 0;
 511         th->check = ~tcp_v4_check(th, skb->len, iph->saddr, iph->daddr, 0);
 512         skb->csum = offsetof(struct tcphdr, check);
 513         skb->ip_summed = CHECKSUM_HW;
 514         return 0;
 515 }
 516
 517 /*
 518  *      This routine will send an RST to the other tcp.
 519  *
 520  *      Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 521  *                    for reset.
 522  *      Answer: if a packet caused RST, it is not for a socket
 523  *              existing in our system, if it is matched to a socket,
 524  *              it is just duplicate segment or bug in other side's TCP.
 525  *              So that we build reply only basing on parameters
 526  *              arrived with segment.
 527  *      Exception: precedence violation. We do not implement it in any case.
 528  */
 529
 530 static void tcp_v4_send_reset(struct sk_buff *skb)
 531 {
 532         struct tcphdr *th = skb->h.th;
 533         struct tcphdr rth;
 534         struct ip_reply_arg arg;
 535
 536         /* Never send a reset in response to a reset. */
 537         if (th->rst)
 538                 return;
 539
 540         if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL)
 541                 return;
 542
 543         /* Swap the send and the receive. */
 544         memset(&rth, 0, sizeof(struct tcphdr));
 545         rth.dest   = th->source;
 546         rth.source = th->dest;
 547         rth.doff   = sizeof(struct tcphdr) / 4;
 548         rth.rst    = 1;
 549
 550         if (th->ack) {
 551                 rth.seq = th->ack_seq;
 552         } else {
 553                 rth.ack = 1;
 554                 rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 555                                     skb->len - (th->doff << 2));
 556         }
 557
 558         memset(&arg, 0, sizeof arg);
 559         arg.iov[0].iov_base = (unsigned char *)&rth;
 560         arg.iov[0].iov_len  = sizeof rth;
 561         arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
 562                                       skb->nh.iph->saddr, /*XXX*/
 563                                       sizeof(struct tcphdr), IPPROTO_TCP, 0);
 564         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 565
 566         ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth);
 567
 568         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
 569         TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
 570 }
 571
 572 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 573    outside socket context is ugly, certainly. What can I do?
 574  */
 575
 576 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 577                             u32 win, u32 ts)
 578 {
 579         struct tcphdr *th = skb->h.th;
 580         struct {
 581                 struct tcphdr th;
 582                 u32 tsopt[3];
 583         } rep;
 584         struct ip_reply_arg arg;
 585
 586         memset(&rep.th, 0, sizeof(struct tcphdr));
 587         memset(&arg, 0, sizeof arg);
 588
 589         arg.iov[0].iov_base = (unsigned char *)&rep;
 590         arg.iov[0].iov_len  = sizeof(rep.th);
 591         if (ts) {
 592                 rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 593                                      (TCPOPT_TIMESTAMP << 8) |
 594                                      TCPOLEN_TIMESTAMP);
 595                 rep.tsopt[1] = htonl(tcp_time_stamp);
 596                 rep.tsopt[2] = htonl(ts);
 597                 arg.iov[0].iov_len = sizeof(rep);
 598         }
 599
 600         /* Swap the send and the receive. */
 601         rep.th.dest    = th->source;
 602         rep.th.source  = th->dest;
 603         rep.th.doff    = arg.iov[0].iov_len / 4;
 604         rep.th.seq     = htonl(seq);
 605         rep.th.ack_seq = htonl(ack);
 606         rep.th.ack     = 1;
 607         rep.th.window  = htons(win);
 608
 609         arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
 610                                       skb->nh.iph->saddr, /*XXX*/
 611                                       arg.iov[0].iov_len, IPPROTO_TCP, 0);
 612         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 613
 614         ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len);
 615
 616         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
 617 }
 618
 619 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 620 {
 621         struct inet_timewait_sock *tw = inet_twsk(sk);
 622         const struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 623
 624         tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 625                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, tcptw->tw_ts_recent);
 626
 627         inet_twsk_put(tw);
 628 }
 629
 630 static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
 631 {
 632         tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
 633                         req->ts_recent);
 634 }
 635
 636 /*
 637  *      Send a SYN-ACK after having received an ACK.
 638  *      This still operates on a request_sock only, not on a big
 639  *      socket.
 640  */
 641 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
 642                               struct dst_entry *dst)
 643 {
 644         const struct inet_request_sock *ireq = inet_rsk(req);
 645         int err = -1;
 646         struct sk_buff * skb;
 647
 648         /* First, grab a route. */
 649         if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
 650                 goto out;
 651
 652         skb = tcp_make_synack(sk, dst, req);
 653
 654         if (skb) {
 655                 struct tcphdr *th = skb->h.th;
 656
 657                 th->check = tcp_v4_check(th, skb->len,
 658                                          ireq->loc_addr,
 659                                          ireq->rmt_addr,
 660                                          csum_partial((char *)th, skb->len,
 661                                                       skb->csum));
 662
 663                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 664                                             ireq->rmt_addr,
 665                                             ireq->opt);
 666                 if (err == NET_XMIT_CN)
 667                         err = 0;
 668         }
 669
 670 out:
 671         dst_release(dst);
 672         return err;
 673 }
 674
 675 /*
 676  *      IPv4 request_sock destructor.
 677  */
 678 static void tcp_v4_reqsk_destructor(struct request_sock *req)
 679 {
 680         kfree(inet_rsk(req)->opt);
 681 }
 682
 683 #ifdef CONFIG_SYN_COOKIES
 684 static void syn_flood_warning(struct sk_buff *skb)
 685 {
 686         static unsigned long warntime;
 687
 688         if (time_after(jiffies, (warntime + HZ * 60))) {
 689                 warntime = jiffies;
 690                 printk(KERN_INFO
 691                        "possible SYN flooding on port %d. Sending cookies.\n",
 692                        ntohs(skb->h.th->dest));
 693         }
 694 }
 695 #endif
 696
 697 /*
 698  * Save and compile IPv4 options into the request_sock if needed.
 699  */
 700 static struct ip_options *tcp_v4_save_options(struct sock *sk,
 701                                               struct sk_buff *skb)
 702 {
 703         struct ip_options *opt = &(IPCB(skb)->opt);
 704         struct ip_options *dopt = NULL;
 705
 706         if (opt && opt->optlen) {
 707                 int opt_size = optlength(opt);
 708                 dopt = kmalloc(opt_size, GFP_ATOMIC);
 709                 if (dopt) {
 710                         if (ip_options_echo(dopt, skb)) {
 711                                 kfree(dopt);
 712                                 dopt = NULL;
 713                         }
 714                 }
 715         }
 716         return dopt;
 717 }
 718
 719 struct request_sock_ops tcp_request_sock_ops = {
 720         .family         =       PF_INET,
 721         .obj_size       =       sizeof(struct tcp_request_sock),
 722         .rtx_syn_ack    =       tcp_v4_send_synack,
 723         .send_ack       =       tcp_v4_reqsk_send_ack,
 724         .destructor     =       tcp_v4_reqsk_destructor,
 725         .send_reset     =       tcp_v4_send_reset,
 726 };
 727
 728 static struct timewait_sock_ops tcp_timewait_sock_ops = {
 729         .twsk_obj_size  = sizeof(struct tcp_timewait_sock),
 730         .twsk_unique    = tcp_twsk_unique,
 731 };
 732
 733 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 734 {
 735         struct inet_request_sock *ireq;
 736         struct tcp_options_received tmp_opt;
 737         struct request_sock *req;
 738         __u32 saddr = skb->nh.iph->saddr;
 739         __u32 daddr = skb->nh.iph->daddr;
 740         __u32 isn = TCP_SKB_CB(skb)->when;
 741         struct dst_entry *dst = NULL;
 742 #ifdef CONFIG_SYN_COOKIES
 743         int want_cookie = 0;
 744 #else
 745 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
 746 #endif
 747
 748         /* Never answer to SYNs send to broadcast or multicast */
 749         if (((struct rtable *)skb->dst)->rt_flags &
 750             (RTCF_BROADCAST | RTCF_MULTICAST))
 751                 goto drop;
 752
 753         /* TW buckets are converted to open requests without
 754          * limitations, they conserve resources and peer is
 755          * evidently real one.
 756          */
 757         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
 758 #ifdef CONFIG_SYN_COOKIES
 759                 if (sysctl_tcp_syncookies) {
 760                         want_cookie = 1;
 761                 } else
 762 #endif
 763                 goto drop;
 764         }
 765
 766         /* Accept backlog is full. If we have already queued enough
 767          * of warm entries in syn queue, drop request. It is better than
 768          * clogging syn queue with openreqs with exponentially increasing
 769          * timeout.
 770          */
 771         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 772                 goto drop;
 773
 774         req = reqsk_alloc(&tcp_request_sock_ops);
 775         if (!req)
 776                 goto drop;
 777
 778         tcp_clear_options(&tmp_opt);
 779         tmp_opt.mss_clamp = 536;
 780         tmp_opt.user_mss  = tcp_sk(sk)->rx_opt.user_mss;
 781
 782         tcp_parse_options(skb, &tmp_opt, 0);
 783
 784         if (want_cookie) {
 785                 tcp_clear_options(&tmp_opt);
 786                 tmp_opt.saw_tstamp = 0;
 787         }
 788
 789         if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) {
 790                 /* Some OSes (unknown ones, but I see them on web server, which
 791                  * contains information interesting only for windows'
 792                  * users) do not send their stamp in SYN. It is easy case.
 793                  * We simply do not advertise TS support.
 794                  */
 795                 tmp_opt.saw_tstamp = 0;
 796                 tmp_opt.tstamp_ok  = 0;
 797         }
 798         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
 799
 800         tcp_openreq_init(req, &tmp_opt, skb);
 801
 802         ireq = inet_rsk(req);
 803         ireq->loc_addr = daddr;
 804         ireq->rmt_addr = saddr;
 805         ireq->opt = tcp_v4_save_options(sk, skb);
 806         if (!want_cookie)
 807                 TCP_ECN_create_request(req, skb->h.th);
 808
 809         if (want_cookie) {
 810 #ifdef CONFIG_SYN_COOKIES
 811                 syn_flood_warning(skb);
 812 #endif
 813                 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
 814         } else if (!isn) {
 815                 struct inet_peer *peer = NULL;
 816
 817                 /* VJ's idea. We save last timestamp seen
 818                  * from the destination in peer table, when entering
 819                  * state TIME-WAIT, and check against it before
 820                  * accepting new connection request.
 821                  *
 822                  * If "isn" is not zero, this request hit alive
 823                  * timewait bucket, so that all the necessary checks
 824                  * are made in the function processing timewait state.
 825                  */
 826                 if (tmp_opt.saw_tstamp &&
 827                     tcp_death_row.sysctl_tw_recycle &&
 828                     (dst = inet_csk_route_req(sk, req)) != NULL &&
 829                     (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
 830                     peer->v4daddr == saddr) {
 831                         if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL &&
 832                             (s32)(peer->tcp_ts - req->ts_recent) >
 833                                                         TCP_PAWS_WINDOW) {
 834                                 NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED);
 835                                 dst_release(dst);
 836                                 goto drop_and_free;
 837                         }
 838                 }
 839                 /* Kill the following clause, if you dislike this way. */
 840                 else if (!sysctl_tcp_syncookies &&
 841                          (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
 842                           (sysctl_max_syn_backlog >> 2)) &&
 843                          (!peer || !peer->tcp_ts_stamp) &&
 844                          (!dst || !dst_metric(dst, RTAX_RTT))) {
 845                         /* Without syncookies last quarter of
 846                          * backlog is filled with destinations,
 847                          * proven to be alive.
 848                          * It means that we continue to communicate
 849                          * to destinations, already remembered
 850                          * to the moment of synflood.
 851                          */
 852                         LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open "
 853                                        "request from %u.%u.%u.%u/%u\n",
 854                                        NIPQUAD(saddr),
 855                                        ntohs(skb->h.th->source));
 856                         dst_release(dst);
 857                         goto drop_and_free;
 858                 }
 859
 860                 isn = tcp_v4_init_sequence(sk, skb);
 861         }
 862         tcp_rsk(req)->snt_isn = isn;
 863
 864         if (tcp_v4_send_synack(sk, req, dst))
 865                 goto drop_and_free;
 866
 867         if (want_cookie) {
 868                 reqsk_free(req);
 869         } else {
 870                 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
 871         }
 872         return 0;
 873
 874 drop_and_free:
 875         reqsk_free(req);
 876 drop:
 877         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
 878         return 0;
 879 }
 880
 881
 882 /*
 883  * The three way handshake has completed - we got a valid synack -
 884  * now create the new socket.
 885  */
 886 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
 887                                   struct request_sock *req,
 888                                   struct dst_entry *dst)
 889 {
 890         struct inet_request_sock *ireq;
 891         struct inet_sock *newinet;
 892         struct tcp_sock *newtp;
 893         struct sock *newsk;
 894
 895         if (sk_acceptq_is_full(sk))
 896                 goto exit_overflow;
 897
 898         if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
 899                 goto exit;
 900
 901         newsk = tcp_create_openreq_child(sk, req, skb);
 902         if (!newsk)
 903                 goto exit;
 904
 905         sk_setup_caps(newsk, dst);
 906
 907         newtp                 = tcp_sk(newsk);
 908         newinet               = inet_sk(newsk);
 909         ireq                  = inet_rsk(req);
 910         newinet->daddr        = ireq->rmt_addr;
 911         newinet->rcv_saddr    = ireq->loc_addr;
 912         newinet->saddr        = ireq->loc_addr;
 913         newinet->opt          = ireq->opt;
 914         ireq->opt             = NULL;
 915         newinet->mc_index     = inet_iif(skb);
 916         newinet->mc_ttl       = skb->nh.iph->ttl;
 917         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 918         if (newinet->opt)
 919                 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
 920         newinet->id = newtp->write_seq ^ jiffies;
 921
 922         tcp_mtup_init(newsk);
 923         tcp_sync_mss(newsk, dst_mtu(dst));
 924         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
 925         tcp_initialize_rcv_mss(newsk);
 926
 927         __inet_hash(&tcp_hashinfo, newsk, 0);
 928         __inet_inherit_port(&tcp_hashinfo, sk, newsk);
 929
 930         return newsk;
 931
 932 exit_overflow:
 933         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 934 exit:
 935         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 936         dst_release(dst);
 937         return NULL;
 938 }
 939
 940 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 941 {
 942         struct tcphdr *th = skb->h.th;
 943         struct iphdr *iph = skb->nh.iph;
 944         struct sock *nsk;
 945         struct request_sock **prev;
 946         /* Find possible connection requests. */
 947         struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
 948                                                        iph->saddr, iph->daddr);
 949         if (req)
 950                 return tcp_check_req(sk, skb, req, prev);
 951
 952         nsk = __inet_lookup_established(&tcp_hashinfo, skb->nh.iph->saddr,
 953                                         th->source, skb->nh.iph->daddr,
 954                                         ntohs(th->dest), inet_iif(skb));
 955
 956         if (nsk) {
 957                 if (nsk->sk_state != TCP_TIME_WAIT) {
 958                         bh_lock_sock(nsk);
 959                         return nsk;
 960                 }
 961                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 962                 return NULL;
 963         }
 964
 965 #ifdef CONFIG_SYN_COOKIES
 966         if (!th->rst && !th->syn && th->ack)
 967                 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
 968 #endif
 969         return sk;
 970 }
 971
 972 static int tcp_v4_checksum_init(struct sk_buff *skb)
 973 {
 974         if (skb->ip_summed == CHECKSUM_HW) {
 975                 if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr,
 976                                   skb->nh.iph->daddr, skb->csum)) {
 977                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 978                         return 0;
 979                 }
 980         }
 981
 982         skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr, skb->nh.iph->daddr,
 983                                        skb->len, IPPROTO_TCP, 0);
 984
 985         if (skb->len <= 76) {
 986                 return __skb_checksum_complete(skb);
 987         }
 988         return 0;
 989 }
 990
 991
 992 /* The socket must have it's spinlock held when we get
 993  * here.
 994  *
 995  * We have a potential double-lock case here, so even when
 996  * doing backlog processing we use the BH locking scheme.
 997  * This is because we cannot sleep with the original spinlock
 998  * held.
 999  */
1000 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1001 {
1002         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1003                 TCP_CHECK_TIMER(sk);
1004                 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
1005                         goto reset;
1006                 TCP_CHECK_TIMER(sk);
1007                 return 0;
1008         }
1009
1010         if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb))
1011                 goto csum_err;
1012
1013         if (sk->sk_state == TCP_LISTEN) {
1014                 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1015                 if (!nsk)
1016                         goto discard;
1017
1018                 if (nsk != sk) {
1019                         if (tcp_child_process(sk, nsk, skb))
1020                                 goto reset;
1021                         return 0;
1022                 }
1023         }
1024
1025         TCP_CHECK_TIMER(sk);
1026         if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1027                 goto reset;
1028         TCP_CHECK_TIMER(sk);
1029         return 0;
1030
1031 reset:
1032         tcp_v4_send_reset(skb);
1033 discard:
1034         kfree_skb(skb);
1035         /* Be careful here. If this function gets more complicated and
1036          * gcc suffers from register pressure on the x86, sk (in %ebx)
1037          * might be destroyed here. This current version compiles correctly,
1038          * but you have been warned.
1039          */
1040         return 0;
1041
1042 csum_err:
1043         TCP_INC_STATS_BH(TCP_MIB_INERRS);
1044         goto discard;
1045 }
1046
1047 /*
1048  *      From tcp_input.c
1049  */
1050
1051 int tcp_v4_rcv(struct sk_buff *skb)
1052 {
1053         struct tcphdr *th;
1054         struct sock *sk;
1055         int ret;
1056
1057         if (skb->pkt_type != PACKET_HOST)
1058                 goto discard_it;
1059
1060         /* Count it even if it's bad */
1061         TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1062
1063         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1064                 goto discard_it;
1065
1066         th = skb->h.th;
1067
1068         if (th->doff < sizeof(struct tcphdr) / 4)
1069                 goto bad_packet;
1070         if (!pskb_may_pull(skb, th->doff * 4))
1071                 goto discard_it;
1072
1073         /* An explanation is required here, I think.
1074          * Packet length and doff are validated by header prediction,
1075          * provided case of th->doff==0 is eliminated.
1076          * So, we defer the checks. */
1077         if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1078              tcp_v4_checksum_init(skb)))
1079                 goto bad_packet;
1080
1081         th = skb->h.th;
1082         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1083         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1084                                     skb->len - th->doff * 4);
1085         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1086         TCP_SKB_CB(skb)->when    = 0;
1087         TCP_SKB_CB(skb)->flags   = skb->nh.iph->tos;
1088         TCP_SKB_CB(skb)->sacked  = 0;
1089
1090         sk = __inet_lookup(&tcp_hashinfo, skb->nh.iph->saddr, th->source,
1091                            skb->nh.iph->daddr, ntohs(th->dest),
1092                            inet_iif(skb));
1093
1094         if (!sk)
1095                 goto no_tcp_socket;
1096
1097 process:
1098         if (sk->sk_state == TCP_TIME_WAIT)
1099                 goto do_time_wait;
1100
1101         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1102                 goto discard_and_relse;
1103         nf_reset(skb);
1104
1105         if (sk_filter(sk, skb, 0))
1106                 goto discard_and_relse;
1107
1108         skb->dev = NULL;
1109
1110         bh_lock_sock(sk);
1111         ret = 0;
1112         if (!sock_owned_by_user(sk)) {
1113                 if (!tcp_prequeue(sk, skb))
1114                         ret = tcp_v4_do_rcv(sk, skb);
1115         } else
1116                 sk_add_backlog(sk, skb);
1117         bh_unlock_sock(sk);
1118
1119         sock_put(sk);
1120
1121         return ret;
1122
1123 no_tcp_socket:
1124         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1125                 goto discard_it;
1126
1127         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1128 bad_packet:
1129                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1130         } else {
1131                 tcp_v4_send_reset(skb);
1132         }
1133
1134 discard_it:
1135         /* Discard frame. */
1136         kfree_skb(skb);
1137         return 0;
1138
1139 discard_and_relse:
1140         sock_put(sk);
1141         goto discard_it;
1142
1143 do_time_wait:
1144         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1145                 inet_twsk_put((struct inet_timewait_sock *) sk);
1146                 goto discard_it;
1147         }
1148
1149         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1150                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1151                 inet_twsk_put((struct inet_timewait_sock *) sk);
1152                 goto discard_it;
1153         }
1154         switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk,
1155                                            skb, th)) {
1156         case TCP_TW_SYN: {
1157                 struct sock *sk2 = inet_lookup_listener(&tcp_hashinfo,
1158                                                         skb->nh.iph->daddr,
1159                                                         ntohs(th->dest),
1160                                                         inet_iif(skb));
1161                 if (sk2) {
1162                         inet_twsk_deschedule((struct inet_timewait_sock *)sk,
1163                                              &tcp_death_row);
1164                         inet_twsk_put((struct inet_timewait_sock *)sk);
1165                         sk = sk2;
1166                         goto process;
1167                 }
1168                 /* Fall through to ACK */
1169         }
1170         case TCP_TW_ACK:
1171                 tcp_v4_timewait_ack(sk, skb);
1172                 break;
1173         case TCP_TW_RST:
1174                 goto no_tcp_socket;
1175         case TCP_TW_SUCCESS:;
1176         }
1177         goto discard_it;
1178 }
1179
1180 /* VJ's idea. Save last timestamp seen from this destination
1181  * and hold it at least for normal timewait interval to use for duplicate
1182  * segment detection in subsequent connections, before they enter synchronized
1183  * state.
1184  */
1185
1186 int tcp_v4_remember_stamp(struct sock *sk)
1187 {
1188         struct inet_sock *inet = inet_sk(sk);
1189         struct tcp_sock *tp = tcp_sk(sk);
1190         struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1191         struct inet_peer *peer = NULL;
1192         int release_it = 0;
1193
1194         if (!rt || rt->rt_dst != inet->daddr) {
1195                 peer = inet_getpeer(inet->daddr, 1);
1196                 release_it = 1;
1197         } else {
1198                 if (!rt->peer)
1199                         rt_bind_peer(rt, 1);
1200                 peer = rt->peer;
1201         }
1202
1203         if (peer) {
1204                 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1205                     (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1206                      peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1207                         peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1208                         peer->tcp_ts = tp->rx_opt.ts_recent;
1209                 }
1210                 if (release_it)
1211                         inet_putpeer(peer);
1212                 return 1;
1213         }
1214
1215         return 0;
1216 }
1217
1218 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1219 {
1220         struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1221
1222         if (peer) {
1223                 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1224
1225                 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1226                     (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1227                      peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1228                         peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1229                         peer->tcp_ts       = tcptw->tw_ts_recent;
1230                 }
1231                 inet_putpeer(peer);
1232                 return 1;
1233         }
1234
1235         return 0;
1236 }
1237
1238 struct inet_connection_sock_af_ops ipv4_specific = {
1239         .queue_xmit        = ip_queue_xmit,
1240         .send_check        = tcp_v4_send_check,
1241         .rebuild_header    = inet_sk_rebuild_header,
1242         .conn_request      = tcp_v4_conn_request,
1243         .syn_recv_sock     = tcp_v4_syn_recv_sock,
1244         .remember_stamp    = tcp_v4_remember_stamp,
1245         .net_header_len    = sizeof(struct iphdr),
1246         .setsockopt        = ip_setsockopt,
1247         .getsockopt        = ip_getsockopt,
1248         .addr2sockaddr     = inet_csk_addr2sockaddr,
1249         .sockaddr_len      = sizeof(struct sockaddr_in),
1250 #ifdef CONFIG_COMPAT
1251         .compat_setsockopt = compat_ip_setsockopt,
1252         .compat_getsockopt = compat_ip_getsockopt,
1253 #endif
1254 };
1255
1256 /* NOTE: A lot of things set to zero explicitly by call to
1257  *       sk_alloc() so need not be done here.
1258  */
1259 static int tcp_v4_init_sock(struct sock *sk)
1260 {
1261         struct inet_connection_sock *icsk = inet_csk(sk);
1262         struct tcp_sock *tp = tcp_sk(sk);
1263
1264         skb_queue_head_init(&tp->out_of_order_queue);
1265         tcp_init_xmit_timers(sk);
1266         tcp_prequeue_init(tp);
1267
1268         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1269         tp->mdev = TCP_TIMEOUT_INIT;
1270
1271         /* So many TCP implementations out there (incorrectly) count the
1272          * initial SYN frame in their delayed-ACK and congestion control
1273          * algorithms that we must have the following bandaid to talk
1274          * efficiently to them.  -DaveM
1275          */
1276         tp->snd_cwnd = 2;
1277
1278         /* See draft-stevens-tcpca-spec-01 for discussion of the
1279          * initialization of these values.
1280          */
1281         tp->snd_ssthresh = 0x7fffffff;  /* Infinity */
1282         tp->snd_cwnd_clamp = ~0;
1283         tp->mss_cache = 536;
1284
1285         tp->reordering = sysctl_tcp_reordering;
1286         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1287
1288         sk->sk_state = TCP_CLOSE;
1289
1290         sk->sk_write_space = sk_stream_write_space;
1291         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1292
1293         icsk->icsk_af_ops = &ipv4_specific;
1294         icsk->icsk_sync_mss = tcp_sync_mss;
1295
1296         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1297         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1298
1299         atomic_inc(&tcp_sockets_allocated);
1300
1301         return 0;
1302 }
1303
1304 int tcp_v4_destroy_sock(struct sock *sk)
1305 {
1306         struct tcp_sock *tp = tcp_sk(sk);
1307
1308         tcp_clear_xmit_timers(sk);
1309
1310         tcp_cleanup_congestion_control(sk);
1311
1312         /* Cleanup up the write buffer. */
1313         sk_stream_writequeue_purge(sk);
1314
1315         /* Cleans up our, hopefully empty, out_of_order_queue. */
1316         __skb_queue_purge(&tp->out_of_order_queue);
1317
1318         /* Clean prequeue, it must be empty really */
1319         __skb_queue_purge(&tp->ucopy.prequeue);
1320
1321         /* Clean up a referenced TCP bind bucket. */
1322         if (inet_csk(sk)->icsk_bind_hash)
1323                 inet_put_port(&tcp_hashinfo, sk);
1324
1325         /*
1326          * If sendmsg cached page exists, toss it.
1327          */
1328         if (sk->sk_sndmsg_page) {
1329                 __free_page(sk->sk_sndmsg_page);
1330                 sk->sk_sndmsg_page = NULL;
1331         }
1332
1333         atomic_dec(&tcp_sockets_allocated);
1334
1335         return 0;
1336 }
1337
1338 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1339
1340 #ifdef CONFIG_PROC_FS
1341 /* Proc filesystem TCP sock list dumping. */
1342
1343 static inline struct inet_timewait_sock *tw_head(struct hlist_head *head)
1344 {
1345         return hlist_empty(head) ? NULL :
1346                 list_entry(head->first, struct inet_timewait_sock, tw_node);
1347 }
1348
1349 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1350 {
1351         return tw->tw_node.next ?
1352                 hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1353 }
1354
1355 static void *listening_get_next(struct seq_file *seq, void *cur)
1356 {
1357         struct inet_connection_sock *icsk;
1358         struct hlist_node *node;
1359         struct sock *sk = cur;
1360         struct tcp_iter_state* st = seq->private;
1361
1362         if (!sk) {
1363                 st->bucket = 0;
1364                 sk = sk_head(&tcp_hashinfo.listening_hash[0]);
1365                 goto get_sk;
1366         }
1367
1368         ++st->num;
1369
1370         if (st->state == TCP_SEQ_STATE_OPENREQ) {
1371                 struct request_sock *req = cur;
1372
1373                 icsk = inet_csk(st->syn_wait_sk);
1374                 req = req->dl_next;
1375                 while (1) {
1376                         while (req) {
1377                                 vxdprintk(VXD_CBIT(net, 6),
1378                                         "sk,req: %p [#%d] (from %d)", req->sk,
1379                                         (req->sk)?req->sk->sk_xid:0, vx_current_xid());
1380                                 if (req->sk &&
1381                                         !vx_check(req->sk->sk_xid, VX_IDENT|VX_WATCH))
1382                                         continue;
1383                                 if (req->rsk_ops->family == st->family) {
1384                                         cur = req;
1385                                         goto out;
1386                                 }
1387                                 req = req->dl_next;
1388                         }
1389                         if (++st->sbucket >= TCP_SYNQ_HSIZE)
1390                                 break;
1391 get_req:
1392                         req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1393                 }
1394                 sk        = sk_next(st->syn_wait_sk);
1395                 st->state = TCP_SEQ_STATE_LISTENING;
1396                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1397         } else {
1398                 icsk = inet_csk(sk);
1399                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1400                 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1401                         goto start_req;
1402                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1403                 sk = sk_next(sk);
1404         }
1405 get_sk:
1406         sk_for_each_from(sk, node) {
1407                 vxdprintk(VXD_CBIT(net, 6), "sk: %p [#%d] (from %d)",
1408                         sk, sk->sk_xid, vx_current_xid());
1409                 if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1410                         continue;
1411                 if (sk->sk_family == st->family) {
1412                         cur = sk;
1413                         goto out;
1414                 }
1415                 icsk = inet_csk(sk);
1416                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1417                 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1418 start_req:
1419                         st->uid         = sock_i_uid(sk);
1420                         st->syn_wait_sk = sk;
1421                         st->state       = TCP_SEQ_STATE_OPENREQ;
1422                         st->sbucket     = 0;
1423                         goto get_req;
1424                 }
1425                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1426         }
1427         if (++st->bucket < INET_LHTABLE_SIZE) {
1428                 sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]);
1429                 goto get_sk;
1430         }
1431         cur = NULL;
1432 out:
1433         return cur;
1434 }
1435
1436 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1437 {
1438         void *rc = listening_get_next(seq, NULL);
1439
1440         while (rc && *pos) {
1441                 rc = listening_get_next(seq, rc);
1442                 --*pos;
1443         }
1444         return rc;
1445 }
1446
1447 static void *established_get_first(struct seq_file *seq)
1448 {
1449         struct tcp_iter_state* st = seq->private;
1450         void *rc = NULL;
1451
1452         for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
1453                 struct sock *sk;
1454                 struct hlist_node *node;
1455                 struct inet_timewait_sock *tw;
1456
1457                 /* We can reschedule _before_ having picked the target: */
1458                 cond_resched_softirq();
1459
1460                 read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1461                 sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1462                         vxdprintk(VXD_CBIT(net, 6),
1463                                 "sk,egf: %p [#%d] (from %d)",
1464                                 sk, sk->sk_xid, vx_current_xid());
1465                         if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1466                                 continue;
1467                         if (sk->sk_family != st->family)
1468                                 continue;
1469                         rc = sk;
1470                         goto out;
1471                 }
1472                 st->state = TCP_SEQ_STATE_TIME_WAIT;
1473                 inet_twsk_for_each(tw, node,
1474                                    &tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain) {
1475                         vxdprintk(VXD_CBIT(net, 6),
1476                                 "tw: %p [#%d] (from %d)",
1477                                 tw, tw->tw_xid, vx_current_xid());
1478                         if (!vx_check(tw->tw_xid, VX_IDENT|VX_WATCH))
1479                                 continue;
1480                         if (tw->tw_family != st->family)
1481                                 continue;
1482                         rc = tw;
1483                         goto out;
1484                 }
1485                 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1486                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1487         }
1488 out:
1489         return rc;
1490 }
1491
1492 static void *established_get_next(struct seq_file *seq, void *cur)
1493 {
1494         struct sock *sk = cur;
1495         struct inet_timewait_sock *tw;
1496         struct hlist_node *node;
1497         struct tcp_iter_state* st = seq->private;
1498
1499         ++st->num;
1500
1501         if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
1502                 tw = cur;
1503                 tw = tw_next(tw);
1504 get_tw:
1505                 while (tw && (tw->tw_family != st->family ||
1506                         !vx_check(tw->tw_xid, VX_IDENT|VX_WATCH))) {
1507                         tw = tw_next(tw);
1508                 }
1509                 if (tw) {
1510                         cur = tw;
1511                         goto out;
1512                 }
1513                 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1514                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1515
1516                 /* We can reschedule between buckets: */
1517                 cond_resched_softirq();
1518
1519                 if (++st->bucket < tcp_hashinfo.ehash_size) {
1520                         read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1521                         sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain);
1522                 } else {
1523                         cur = NULL;
1524                         goto out;
1525                 }
1526         } else
1527                 sk = sk_next(sk);
1528
1529         sk_for_each_from(sk, node) {
1530                 vxdprintk(VXD_CBIT(net, 6),
1531                         "sk,egn: %p [#%d] (from %d)",
1532                         sk, sk->sk_xid, vx_current_xid());
1533                 if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1534                         continue;
1535                 if (sk->sk_family == st->family)
1536                         goto found;
1537         }
1538
1539         st->state = TCP_SEQ_STATE_TIME_WAIT;
1540         tw = tw_head(&tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain);
1541         goto get_tw;
1542 found:
1543         cur = sk;
1544 out:
1545         return cur;
1546 }
1547
1548 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1549 {
1550         void *rc = established_get_first(seq);
1551
1552         while (rc && pos) {
1553                 rc = established_get_next(seq, rc);
1554                 --pos;
1555         }
1556         return rc;
1557 }
1558
1559 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1560 {
1561         void *rc;
1562         struct tcp_iter_state* st = seq->private;
1563
1564         inet_listen_lock(&tcp_hashinfo);
1565         st->state = TCP_SEQ_STATE_LISTENING;
1566         rc        = listening_get_idx(seq, &pos);
1567
1568         if (!rc) {
1569                 inet_listen_unlock(&tcp_hashinfo);
1570                 local_bh_disable();
1571                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1572                 rc        = established_get_idx(seq, pos);
1573         }
1574
1575         return rc;
1576 }
1577
1578 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
1579 {
1580         struct tcp_iter_state* st = seq->private;
1581         st->state = TCP_SEQ_STATE_LISTENING;
1582         st->num = 0;
1583         return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1584 }
1585
1586 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1587 {
1588         void *rc = NULL;
1589         struct tcp_iter_state* st;
1590
1591         if (v == SEQ_START_TOKEN) {
1592                 rc = tcp_get_idx(seq, 0);
1593                 goto out;
1594         }
1595         st = seq->private;
1596
1597         switch (st->state) {
1598         case TCP_SEQ_STATE_OPENREQ:
1599         case TCP_SEQ_STATE_LISTENING:
1600                 rc = listening_get_next(seq, v);
1601                 if (!rc) {
1602                         inet_listen_unlock(&tcp_hashinfo);
1603                         local_bh_disable();
1604                         st->state = TCP_SEQ_STATE_ESTABLISHED;
1605                         rc        = established_get_first(seq);
1606                 }
1607                 break;
1608         case TCP_SEQ_STATE_ESTABLISHED:
1609         case TCP_SEQ_STATE_TIME_WAIT:
1610                 rc = established_get_next(seq, v);
1611                 break;
1612         }
1613 out:
1614         ++*pos;
1615         return rc;
1616 }
1617
1618 static void tcp_seq_stop(struct seq_file *seq, void *v)
1619 {
1620         struct tcp_iter_state* st = seq->private;
1621
1622         switch (st->state) {
1623         case TCP_SEQ_STATE_OPENREQ:
1624                 if (v) {
1625                         struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
1626                         read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1627                 }
1628         case TCP_SEQ_STATE_LISTENING:
1629                 if (v != SEQ_START_TOKEN)
1630                         inet_listen_unlock(&tcp_hashinfo);
1631                 break;
1632         case TCP_SEQ_STATE_TIME_WAIT:
1633         case TCP_SEQ_STATE_ESTABLISHED:
1634                 if (v)
1635                         read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1636                 local_bh_enable();
1637                 break;
1638         }
1639 }
1640
1641 static int tcp_seq_open(struct inode *inode, struct file *file)
1642 {
1643         struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
1644         struct seq_file *seq;
1645         struct tcp_iter_state *s;
1646         int rc;
1647
1648         if (unlikely(afinfo == NULL))
1649                 return -EINVAL;
1650
1651         s = kmalloc(sizeof(*s), GFP_KERNEL);
1652         if (!s)
1653                 return -ENOMEM;
1654         memset(s, 0, sizeof(*s));
1655         s->family               = afinfo->family;
1656         s->seq_ops.start        = tcp_seq_start;
1657         s->seq_ops.next         = tcp_seq_next;
1658         s->seq_ops.show         = afinfo->seq_show;
1659         s->seq_ops.stop         = tcp_seq_stop;
1660
1661         rc = seq_open(file, &s->seq_ops);
1662         if (rc)
1663                 goto out_kfree;
1664         seq          = file->private_data;
1665         seq->private = s;
1666 out:
1667         return rc;
1668 out_kfree:
1669         kfree(s);
1670         goto out;
1671 }
1672
1673 int tcp_proc_register(struct tcp_seq_afinfo *afinfo)
1674 {
1675         int rc = 0;
1676         struct proc_dir_entry *p;
1677
1678         if (!afinfo)
1679                 return -EINVAL;
1680         afinfo->seq_fops->owner         = afinfo->owner;
1681         afinfo->seq_fops->open          = tcp_seq_open;
1682         afinfo->seq_fops->read          = seq_read;
1683         afinfo->seq_fops->llseek        = seq_lseek;
1684         afinfo->seq_fops->release       = seq_release_private;
1685
1686         p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1687         if (p)
1688                 p->data = afinfo;
1689         else
1690                 rc = -ENOMEM;
1691         return rc;
1692 }
1693
1694 void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo)
1695 {
1696         if (!afinfo)
1697                 return;
1698         proc_net_remove(afinfo->name);
1699         memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1700 }
1701
1702 static void get_openreq4(struct sock *sk, struct request_sock *req,
1703                          char *tmpbuf, int i, int uid)
1704 {
1705         const struct inet_request_sock *ireq = inet_rsk(req);
1706         int ttd = req->expires - jiffies;
1707
1708         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1709                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p",
1710                 i,
1711                 ireq->loc_addr,
1712                 ntohs(inet_sk(sk)->sport),
1713                 ireq->rmt_addr,
1714                 ntohs(ireq->rmt_port),
1715                 TCP_SYN_RECV,
1716                 0, 0, /* could print option size, but that is af dependent. */
1717                 1,    /* timers active (only the expire timer) */
1718                 jiffies_to_clock_t(ttd),
1719                 req->retrans,
1720                 uid,
1721                 0,  /* non standard timer */
1722                 0, /* open_requests have no inode */
1723                 atomic_read(&sk->sk_refcnt),
1724                 req);
1725 }
1726
1727 static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i)
1728 {
1729         int timer_active;
1730         unsigned long timer_expires;
1731         struct tcp_sock *tp = tcp_sk(sp);
1732         const struct inet_connection_sock *icsk = inet_csk(sp);
1733         struct inet_sock *inet = inet_sk(sp);
1734         unsigned int dest = inet->daddr;
1735         unsigned int src = inet->rcv_saddr;
1736         __u16 destp = ntohs(inet->dport);
1737         __u16 srcp = ntohs(inet->sport);
1738
1739         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1740                 timer_active    = 1;
1741                 timer_expires   = icsk->icsk_timeout;
1742         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1743                 timer_active    = 4;
1744                 timer_expires   = icsk->icsk_timeout;
1745         } else if (timer_pending(&sp->sk_timer)) {
1746                 timer_active    = 2;
1747                 timer_expires   = sp->sk_timer.expires;
1748         } else {
1749                 timer_active    = 0;
1750                 timer_expires = jiffies;
1751         }
1752
1753         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
1754                         "%08X %5d %8d %lu %d %p %u %u %u %u %d",
1755                 i, src, srcp, dest, destp, sp->sk_state,
1756                 tp->write_seq - tp->snd_una, tp->rcv_nxt - tp->copied_seq,
1757                 timer_active,
1758                 jiffies_to_clock_t(timer_expires - jiffies),
1759                 icsk->icsk_retransmits,
1760                 sock_i_uid(sp),
1761                 icsk->icsk_probes_out,
1762                 sock_i_ino(sp),
1763                 atomic_read(&sp->sk_refcnt), sp,
1764                 icsk->icsk_rto,
1765                 icsk->icsk_ack.ato,
1766                 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1767                 tp->snd_cwnd,
1768                 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh);
1769 }
1770
1771 static void get_timewait4_sock(struct inet_timewait_sock *tw, char *tmpbuf, int i)
1772 {
1773         unsigned int dest, src;
1774         __u16 destp, srcp;
1775         int ttd = tw->tw_ttd - jiffies;
1776
1777         if (ttd < 0)
1778                 ttd = 0;
1779
1780         dest  = tw->tw_daddr;
1781         src   = tw->tw_rcv_saddr;
1782         destp = ntohs(tw->tw_dport);
1783         srcp  = ntohs(tw->tw_sport);
1784
1785         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1786                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p",
1787                 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
1788                 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1789                 atomic_read(&tw->tw_refcnt), tw);
1790 }
1791
1792 #define TMPSZ 150
1793
1794 static int tcp4_seq_show(struct seq_file *seq, void *v)
1795 {
1796         struct tcp_iter_state* st;
1797         char tmpbuf[TMPSZ + 1];
1798
1799         if (v == SEQ_START_TOKEN) {
1800                 seq_printf(seq, "%-*s\n", TMPSZ - 1,
1801                            "  sl  local_address rem_address   st tx_queue "
1802                            "rx_queue tr tm->when retrnsmt   uid  timeout "
1803                            "inode");
1804                 goto out;
1805         }
1806         st = seq->private;
1807
1808         switch (st->state) {
1809         case TCP_SEQ_STATE_LISTENING:
1810         case TCP_SEQ_STATE_ESTABLISHED:
1811                 get_tcp4_sock(v, tmpbuf, st->num);
1812                 break;
1813         case TCP_SEQ_STATE_OPENREQ:
1814                 get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid);
1815                 break;
1816         case TCP_SEQ_STATE_TIME_WAIT:
1817                 get_timewait4_sock(v, tmpbuf, st->num);
1818                 break;
1819         }
1820         seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf);
1821 out:
1822         return 0;
1823 }
1824
1825 static struct file_operations tcp4_seq_fops;
1826 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
1827         .owner          = THIS_MODULE,
1828         .name           = "tcp",
1829         .family         = AF_INET,
1830         .seq_show       = tcp4_seq_show,
1831         .seq_fops       = &tcp4_seq_fops,
1832 };
1833
1834 int __init tcp4_proc_init(void)
1835 {
1836         return tcp_proc_register(&tcp4_seq_afinfo);
1837 }
1838
1839 void tcp4_proc_exit(void)
1840 {
1841         tcp_proc_unregister(&tcp4_seq_afinfo);
1842 }
1843 #endif /* CONFIG_PROC_FS */
1844
1845 struct proto tcp_prot = {
1846         .name                   = "TCP",
1847         .owner                  = THIS_MODULE,
1848         .close                  = tcp_close,
1849         .connect                = tcp_v4_connect,
1850         .disconnect             = tcp_disconnect,
1851         .accept                 = inet_csk_accept,
1852         .ioctl                  = tcp_ioctl,
1853         .init                   = tcp_v4_init_sock,
1854         .destroy                = tcp_v4_destroy_sock,
1855         .shutdown               = tcp_shutdown,
1856         .setsockopt             = tcp_setsockopt,
1857         .getsockopt             = tcp_getsockopt,
1858         .sendmsg                = tcp_sendmsg,
1859         .recvmsg                = tcp_recvmsg,
1860         .backlog_rcv            = tcp_v4_do_rcv,
1861         .hash                   = tcp_v4_hash,
1862         .unhash                 = tcp_unhash,
1863         .get_port               = tcp_v4_get_port,
1864         .enter_memory_pressure  = tcp_enter_memory_pressure,
1865         .sockets_allocated      = &tcp_sockets_allocated,
1866         .orphan_count           = &tcp_orphan_count,
1867         .memory_allocated       = &tcp_memory_allocated,
1868         .memory_pressure        = &tcp_memory_pressure,
1869         .sysctl_mem             = sysctl_tcp_mem,
1870         .sysctl_wmem            = sysctl_tcp_wmem,
1871         .sysctl_rmem            = sysctl_tcp_rmem,
1872         .max_header             = MAX_TCP_HEADER,
1873         .obj_size               = sizeof(struct tcp_sock),
1874         .twsk_prot              = &tcp_timewait_sock_ops,
1875         .rsk_prot               = &tcp_request_sock_ops,
1876 #ifdef CONFIG_COMPAT
1877         .compat_setsockopt      = compat_tcp_setsockopt,
1878         .compat_getsockopt      = compat_tcp_getsockopt,
1879 #endif
1880 };
1881
1882 void __init tcp_v4_init(struct net_proto_family *ops)
1883 {
1884         if (inet_csk_ctl_sock_create(&tcp_socket, PF_INET, SOCK_RAW, IPPROTO_TCP) < 0)
1885                 panic("Failed to create the TCP control socket.\n");
1886 }
1887
1888 EXPORT_SYMBOL(ipv4_specific);
1889 EXPORT_SYMBOL(tcp_hashinfo);
1890 EXPORT_SYMBOL(tcp_prot);
1891 EXPORT_SYMBOL(tcp_unhash);
1892 EXPORT_SYMBOL(tcp_v4_conn_request);
1893 EXPORT_SYMBOL(tcp_v4_connect);
1894 EXPORT_SYMBOL(tcp_v4_do_rcv);
1895 EXPORT_SYMBOL(tcp_v4_remember_stamp);
1896 EXPORT_SYMBOL(tcp_v4_send_check);
1897 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1898
1899 #ifdef CONFIG_PROC_FS
1900 EXPORT_SYMBOL(tcp_proc_register);
1901 EXPORT_SYMBOL(tcp_proc_unregister);
1902 #endif
1903 EXPORT_SYMBOL(sysctl_local_port_range);
1904 EXPORT_SYMBOL(sysctl_tcp_low_latency);
1905