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 /*
 500  *      This routine will send an RST to the other tcp.
 501  *
 502  *      Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 503  *                    for reset.
 504  *      Answer: if a packet caused RST, it is not for a socket
 505  *              existing in our system, if it is matched to a socket,
 506  *              it is just duplicate segment or bug in other side's TCP.
 507  *              So that we build reply only basing on parameters
 508  *              arrived with segment.
 509  *      Exception: precedence violation. We do not implement it in any case.
 510  */
 511
 512 static void tcp_v4_send_reset(struct sk_buff *skb)
 513 {
 514         struct tcphdr *th = skb->h.th;
 515         struct tcphdr rth;
 516         struct ip_reply_arg arg;
 517
 518         /* Never send a reset in response to a reset. */
 519         if (th->rst)
 520                 return;
 521
 522         if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL)
 523                 return;
 524
 525         /* Swap the send and the receive. */
 526         memset(&rth, 0, sizeof(struct tcphdr));
 527         rth.dest   = th->source;
 528         rth.source = th->dest;
 529         rth.doff   = sizeof(struct tcphdr) / 4;
 530         rth.rst    = 1;
 531
 532         if (th->ack) {
 533                 rth.seq = th->ack_seq;
 534         } else {
 535                 rth.ack = 1;
 536                 rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 537                                     skb->len - (th->doff << 2));
 538         }
 539
 540         memset(&arg, 0, sizeof arg);
 541         arg.iov[0].iov_base = (unsigned char *)&rth;
 542         arg.iov[0].iov_len  = sizeof rth;
 543         arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
 544                                       skb->nh.iph->saddr, /*XXX*/
 545                                       sizeof(struct tcphdr), IPPROTO_TCP, 0);
 546         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 547
 548         ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth);
 549
 550         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
 551         TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
 552 }
 553
 554 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 555    outside socket context is ugly, certainly. What can I do?
 556  */
 557
 558 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 559                             u32 win, u32 ts)
 560 {
 561         struct tcphdr *th = skb->h.th;
 562         struct {
 563                 struct tcphdr th;
 564                 u32 tsopt[3];
 565         } rep;
 566         struct ip_reply_arg arg;
 567
 568         memset(&rep.th, 0, sizeof(struct tcphdr));
 569         memset(&arg, 0, sizeof arg);
 570
 571         arg.iov[0].iov_base = (unsigned char *)&rep;
 572         arg.iov[0].iov_len  = sizeof(rep.th);
 573         if (ts) {
 574                 rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 575                                      (TCPOPT_TIMESTAMP << 8) |
 576                                      TCPOLEN_TIMESTAMP);
 577                 rep.tsopt[1] = htonl(tcp_time_stamp);
 578                 rep.tsopt[2] = htonl(ts);
 579                 arg.iov[0].iov_len = sizeof(rep);
 580         }
 581
 582         /* Swap the send and the receive. */
 583         rep.th.dest    = th->source;
 584         rep.th.source  = th->dest;
 585         rep.th.doff    = arg.iov[0].iov_len / 4;
 586         rep.th.seq     = htonl(seq);
 587         rep.th.ack_seq = htonl(ack);
 588         rep.th.ack     = 1;
 589         rep.th.window  = htons(win);
 590
 591         arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
 592                                       skb->nh.iph->saddr, /*XXX*/
 593                                       arg.iov[0].iov_len, IPPROTO_TCP, 0);
 594         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 595
 596         ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len);
 597
 598         TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
 599 }
 600
 601 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 602 {
 603         struct inet_timewait_sock *tw = inet_twsk(sk);
 604         const struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 605
 606         tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 607                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, tcptw->tw_ts_recent);
 608
 609         inet_twsk_put(tw);
 610 }
 611
 612 static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
 613 {
 614         tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
 615                         req->ts_recent);
 616 }
 617
 618 /*
 619  *      Send a SYN-ACK after having received an ACK.
 620  *      This still operates on a request_sock only, not on a big
 621  *      socket.
 622  */
 623 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
 624                               struct dst_entry *dst)
 625 {
 626         const struct inet_request_sock *ireq = inet_rsk(req);
 627         int err = -1;
 628         struct sk_buff * skb;
 629
 630         /* First, grab a route. */
 631         if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
 632                 goto out;
 633
 634         skb = tcp_make_synack(sk, dst, req);
 635
 636         if (skb) {
 637                 struct tcphdr *th = skb->h.th;
 638
 639                 th->check = tcp_v4_check(th, skb->len,
 640                                          ireq->loc_addr,
 641                                          ireq->rmt_addr,
 642                                          csum_partial((char *)th, skb->len,
 643                                                       skb->csum));
 644
 645                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 646                                             ireq->rmt_addr,
 647                                             ireq->opt);
 648                 if (err == NET_XMIT_CN)
 649                         err = 0;
 650         }
 651
 652 out:
 653         dst_release(dst);
 654         return err;
 655 }
 656
 657 /*
 658  *      IPv4 request_sock destructor.
 659  */
 660 static void tcp_v4_reqsk_destructor(struct request_sock *req)
 661 {
 662         kfree(inet_rsk(req)->opt);
 663 }
 664
 665 #ifdef CONFIG_SYN_COOKIES
 666 static void syn_flood_warning(struct sk_buff *skb)
 667 {
 668         static unsigned long warntime;
 669
 670         if (time_after(jiffies, (warntime + HZ * 60))) {
 671                 warntime = jiffies;
 672                 printk(KERN_INFO
 673                        "possible SYN flooding on port %d. Sending cookies.\n",
 674                        ntohs(skb->h.th->dest));
 675         }
 676 }
 677 #endif
 678
 679 /*
 680  * Save and compile IPv4 options into the request_sock if needed.
 681  */
 682 static struct ip_options *tcp_v4_save_options(struct sock *sk,
 683                                               struct sk_buff *skb)
 684 {
 685         struct ip_options *opt = &(IPCB(skb)->opt);
 686         struct ip_options *dopt = NULL;
 687
 688         if (opt && opt->optlen) {
 689                 int opt_size = optlength(opt);
 690                 dopt = kmalloc(opt_size, GFP_ATOMIC);
 691                 if (dopt) {
 692                         if (ip_options_echo(dopt, skb)) {
 693                                 kfree(dopt);
 694                                 dopt = NULL;
 695                         }
 696                 }
 697         }
 698         return dopt;
 699 }
 700
 701 struct request_sock_ops tcp_request_sock_ops = {
 702         .family         =       PF_INET,
 703         .obj_size       =       sizeof(struct tcp_request_sock),
 704         .rtx_syn_ack    =       tcp_v4_send_synack,
 705         .send_ack       =       tcp_v4_reqsk_send_ack,
 706         .destructor     =       tcp_v4_reqsk_destructor,
 707         .send_reset     =       tcp_v4_send_reset,
 708 };
 709
 710 static struct timewait_sock_ops tcp_timewait_sock_ops = {
 711         .twsk_obj_size  = sizeof(struct tcp_timewait_sock),
 712         .twsk_unique    = tcp_twsk_unique,
 713 };
 714
 715 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 716 {
 717         struct inet_request_sock *ireq;
 718         struct tcp_options_received tmp_opt;
 719         struct request_sock *req;
 720         __u32 saddr = skb->nh.iph->saddr;
 721         __u32 daddr = skb->nh.iph->daddr;
 722         __u32 isn = TCP_SKB_CB(skb)->when;
 723         struct dst_entry *dst = NULL;
 724 #ifdef CONFIG_SYN_COOKIES
 725         int want_cookie = 0;
 726 #else
 727 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
 728 #endif
 729
 730         /* Never answer to SYNs send to broadcast or multicast */
 731         if (((struct rtable *)skb->dst)->rt_flags &
 732             (RTCF_BROADCAST | RTCF_MULTICAST))
 733                 goto drop;
 734
 735         /* TW buckets are converted to open requests without
 736          * limitations, they conserve resources and peer is
 737          * evidently real one.
 738          */
 739         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
 740 #ifdef CONFIG_SYN_COOKIES
 741                 if (sysctl_tcp_syncookies) {
 742                         want_cookie = 1;
 743                 } else
 744 #endif
 745                 goto drop;
 746         }
 747
 748         /* Accept backlog is full. If we have already queued enough
 749          * of warm entries in syn queue, drop request. It is better than
 750          * clogging syn queue with openreqs with exponentially increasing
 751          * timeout.
 752          */
 753         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 754                 goto drop;
 755
 756         req = reqsk_alloc(&tcp_request_sock_ops);
 757         if (!req)
 758                 goto drop;
 759
 760         tcp_clear_options(&tmp_opt);
 761         tmp_opt.mss_clamp = 536;
 762         tmp_opt.user_mss  = tcp_sk(sk)->rx_opt.user_mss;
 763
 764         tcp_parse_options(skb, &tmp_opt, 0);
 765
 766         if (want_cookie) {
 767                 tcp_clear_options(&tmp_opt);
 768                 tmp_opt.saw_tstamp = 0;
 769         }
 770
 771         if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) {
 772                 /* Some OSes (unknown ones, but I see them on web server, which
 773                  * contains information interesting only for windows'
 774                  * users) do not send their stamp in SYN. It is easy case.
 775                  * We simply do not advertise TS support.
 776                  */
 777                 tmp_opt.saw_tstamp = 0;
 778                 tmp_opt.tstamp_ok  = 0;
 779         }
 780         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
 781
 782         tcp_openreq_init(req, &tmp_opt, skb);
 783
 784         ireq = inet_rsk(req);
 785         ireq->loc_addr = daddr;
 786         ireq->rmt_addr = saddr;
 787         ireq->opt = tcp_v4_save_options(sk, skb);
 788         if (!want_cookie)
 789                 TCP_ECN_create_request(req, skb->h.th);
 790
 791         if (want_cookie) {
 792 #ifdef CONFIG_SYN_COOKIES
 793                 syn_flood_warning(skb);
 794 #endif
 795                 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
 796         } else if (!isn) {
 797                 struct inet_peer *peer = NULL;
 798
 799                 /* VJ's idea. We save last timestamp seen
 800                  * from the destination in peer table, when entering
 801                  * state TIME-WAIT, and check against it before
 802                  * accepting new connection request.
 803                  *
 804                  * If "isn" is not zero, this request hit alive
 805                  * timewait bucket, so that all the necessary checks
 806                  * are made in the function processing timewait state.
 807                  */
 808                 if (tmp_opt.saw_tstamp &&
 809                     tcp_death_row.sysctl_tw_recycle &&
 810                     (dst = inet_csk_route_req(sk, req)) != NULL &&
 811                     (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
 812                     peer->v4daddr == saddr) {
 813                         if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL &&
 814                             (s32)(peer->tcp_ts - req->ts_recent) >
 815                                                         TCP_PAWS_WINDOW) {
 816                                 NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED);
 817                                 dst_release(dst);
 818                                 goto drop_and_free;
 819                         }
 820                 }
 821                 /* Kill the following clause, if you dislike this way. */
 822                 else if (!sysctl_tcp_syncookies &&
 823                          (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
 824                           (sysctl_max_syn_backlog >> 2)) &&
 825                          (!peer || !peer->tcp_ts_stamp) &&
 826                          (!dst || !dst_metric(dst, RTAX_RTT))) {
 827                         /* Without syncookies last quarter of
 828                          * backlog is filled with destinations,
 829                          * proven to be alive.
 830                          * It means that we continue to communicate
 831                          * to destinations, already remembered
 832                          * to the moment of synflood.
 833                          */
 834                         LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open "
 835                                        "request from %u.%u.%u.%u/%u\n",
 836                                        NIPQUAD(saddr),
 837                                        ntohs(skb->h.th->source));
 838                         dst_release(dst);
 839                         goto drop_and_free;
 840                 }
 841
 842                 isn = tcp_v4_init_sequence(sk, skb);
 843         }
 844         tcp_rsk(req)->snt_isn = isn;
 845
 846         if (tcp_v4_send_synack(sk, req, dst))
 847                 goto drop_and_free;
 848
 849         if (want_cookie) {
 850                 reqsk_free(req);
 851         } else {
 852                 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
 853         }
 854         return 0;
 855
 856 drop_and_free:
 857         reqsk_free(req);
 858 drop:
 859         TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
 860         return 0;
 861 }
 862
 863
 864 /*
 865  * The three way handshake has completed - we got a valid synack -
 866  * now create the new socket.
 867  */
 868 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
 869                                   struct request_sock *req,
 870                                   struct dst_entry *dst)
 871 {
 872         struct inet_request_sock *ireq;
 873         struct inet_sock *newinet;
 874         struct tcp_sock *newtp;
 875         struct sock *newsk;
 876
 877         if (sk_acceptq_is_full(sk))
 878                 goto exit_overflow;
 879
 880         if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
 881                 goto exit;
 882
 883         newsk = tcp_create_openreq_child(sk, req, skb);
 884         if (!newsk)
 885                 goto exit;
 886
 887         sk_setup_caps(newsk, dst);
 888
 889         newtp                 = tcp_sk(newsk);
 890         newinet               = inet_sk(newsk);
 891         ireq                  = inet_rsk(req);
 892         newinet->daddr        = ireq->rmt_addr;
 893         newinet->rcv_saddr    = ireq->loc_addr;
 894         newinet->saddr        = ireq->loc_addr;
 895         newinet->opt          = ireq->opt;
 896         ireq->opt             = NULL;
 897         newinet->mc_index     = inet_iif(skb);
 898         newinet->mc_ttl       = skb->nh.iph->ttl;
 899         inet_csk(newsk)->icsk_ext_hdr_len = 0;
 900         if (newinet->opt)
 901                 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
 902         newinet->id = newtp->write_seq ^ jiffies;
 903
 904         tcp_mtup_init(newsk);
 905         tcp_sync_mss(newsk, dst_mtu(dst));
 906         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
 907         tcp_initialize_rcv_mss(newsk);
 908
 909         __inet_hash(&tcp_hashinfo, newsk, 0);
 910         __inet_inherit_port(&tcp_hashinfo, sk, newsk);
 911
 912         return newsk;
 913
 914 exit_overflow:
 915         NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
 916 exit:
 917         NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
 918         dst_release(dst);
 919         return NULL;
 920 }
 921
 922 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 923 {
 924         struct tcphdr *th = skb->h.th;
 925         struct iphdr *iph = skb->nh.iph;
 926         struct sock *nsk;
 927         struct request_sock **prev;
 928         /* Find possible connection requests. */
 929         struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
 930                                                        iph->saddr, iph->daddr);
 931         if (req)
 932                 return tcp_check_req(sk, skb, req, prev);
 933
 934         nsk = __inet_lookup_established(&tcp_hashinfo, skb->nh.iph->saddr,
 935                                         th->source, skb->nh.iph->daddr,
 936                                         ntohs(th->dest), inet_iif(skb));
 937
 938         if (nsk) {
 939                 if (nsk->sk_state != TCP_TIME_WAIT) {
 940                         bh_lock_sock(nsk);
 941                         return nsk;
 942                 }
 943                 inet_twsk_put((struct inet_timewait_sock *)nsk);
 944                 return NULL;
 945         }
 946
 947 #ifdef CONFIG_SYN_COOKIES
 948         if (!th->rst && !th->syn && th->ack)
 949                 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
 950 #endif
 951         return sk;
 952 }
 953
 954 static int tcp_v4_checksum_init(struct sk_buff *skb)
 955 {
 956         if (skb->ip_summed == CHECKSUM_HW) {
 957                 if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr,
 958                                   skb->nh.iph->daddr, skb->csum)) {
 959                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 960                         return 0;
 961                 }
 962         }
 963
 964         skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr, skb->nh.iph->daddr,
 965                                        skb->len, IPPROTO_TCP, 0);
 966
 967         if (skb->len <= 76) {
 968                 return __skb_checksum_complete(skb);
 969         }
 970         return 0;
 971 }
 972
 973
 974 /* The socket must have it's spinlock held when we get
 975  * here.
 976  *
 977  * We have a potential double-lock case here, so even when
 978  * doing backlog processing we use the BH locking scheme.
 979  * This is because we cannot sleep with the original spinlock
 980  * held.
 981  */
 982 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 983 {
 984         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
 985                 TCP_CHECK_TIMER(sk);
 986                 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
 987                         goto reset;
 988                 TCP_CHECK_TIMER(sk);
 989                 return 0;
 990         }
 991
 992         if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb))
 993                 goto csum_err;
 994
 995         if (sk->sk_state == TCP_LISTEN) {
 996                 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
 997                 if (!nsk)
 998                         goto discard;
 999
1000                 if (nsk != sk) {
1001                         if (tcp_child_process(sk, nsk, skb))
1002                                 goto reset;
1003                         return 0;
1004                 }
1005         }
1006
1007         TCP_CHECK_TIMER(sk);
1008         if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1009                 goto reset;
1010         TCP_CHECK_TIMER(sk);
1011         return 0;
1012
1013 reset:
1014         tcp_v4_send_reset(skb);
1015 discard:
1016         kfree_skb(skb);
1017         /* Be careful here. If this function gets more complicated and
1018          * gcc suffers from register pressure on the x86, sk (in %ebx)
1019          * might be destroyed here. This current version compiles correctly,
1020          * but you have been warned.
1021          */
1022         return 0;
1023
1024 csum_err:
1025         TCP_INC_STATS_BH(TCP_MIB_INERRS);
1026         goto discard;
1027 }
1028
1029 /*
1030  *      From tcp_input.c
1031  */
1032
1033 int tcp_v4_rcv(struct sk_buff *skb)
1034 {
1035         struct tcphdr *th;
1036         struct sock *sk;
1037         int ret;
1038
1039         if (skb->pkt_type != PACKET_HOST)
1040                 goto discard_it;
1041
1042         /* Count it even if it's bad */
1043         TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1044
1045         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1046                 goto discard_it;
1047
1048         th = skb->h.th;
1049
1050         if (th->doff < sizeof(struct tcphdr) / 4)
1051                 goto bad_packet;
1052         if (!pskb_may_pull(skb, th->doff * 4))
1053                 goto discard_it;
1054
1055         /* An explanation is required here, I think.
1056          * Packet length and doff are validated by header prediction,
1057          * provided case of th->doff==0 is eliminated.
1058          * So, we defer the checks. */
1059         if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1060              tcp_v4_checksum_init(skb)))
1061                 goto bad_packet;
1062
1063         th = skb->h.th;
1064         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1065         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1066                                     skb->len - th->doff * 4);
1067         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1068         TCP_SKB_CB(skb)->when    = 0;
1069         TCP_SKB_CB(skb)->flags   = skb->nh.iph->tos;
1070         TCP_SKB_CB(skb)->sacked  = 0;
1071
1072         sk = __inet_lookup(&tcp_hashinfo, skb->nh.iph->saddr, th->source,
1073                            skb->nh.iph->daddr, ntohs(th->dest),
1074                            inet_iif(skb));
1075
1076         if (!sk)
1077                 goto no_tcp_socket;
1078
1079 process:
1080         if (sk->sk_state == TCP_TIME_WAIT)
1081                 goto do_time_wait;
1082
1083         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1084                 goto discard_and_relse;
1085         nf_reset(skb);
1086
1087         if (sk_filter(sk, skb, 0))
1088                 goto discard_and_relse;
1089
1090         skb->dev = NULL;
1091
1092         bh_lock_sock(sk);
1093         ret = 0;
1094         if (!sock_owned_by_user(sk)) {
1095                 if (!tcp_prequeue(sk, skb))
1096                         ret = tcp_v4_do_rcv(sk, skb);
1097         } else
1098                 sk_add_backlog(sk, skb);
1099         bh_unlock_sock(sk);
1100
1101         sock_put(sk);
1102
1103         return ret;
1104
1105 no_tcp_socket:
1106         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1107                 goto discard_it;
1108
1109         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1110 bad_packet:
1111                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1112         } else {
1113                 tcp_v4_send_reset(skb);
1114         }
1115
1116 discard_it:
1117         /* Discard frame. */
1118         kfree_skb(skb);
1119         return 0;
1120
1121 discard_and_relse:
1122         sock_put(sk);
1123         goto discard_it;
1124
1125 do_time_wait:
1126         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1127                 inet_twsk_put((struct inet_timewait_sock *) sk);
1128                 goto discard_it;
1129         }
1130
1131         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1132                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1133                 inet_twsk_put((struct inet_timewait_sock *) sk);
1134                 goto discard_it;
1135         }
1136         switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk,
1137                                            skb, th)) {
1138         case TCP_TW_SYN: {
1139                 struct sock *sk2 = inet_lookup_listener(&tcp_hashinfo,
1140                                                         skb->nh.iph->daddr,
1141                                                         ntohs(th->dest),
1142                                                         inet_iif(skb));
1143                 if (sk2) {
1144                         inet_twsk_deschedule((struct inet_timewait_sock *)sk,
1145                                              &tcp_death_row);
1146                         inet_twsk_put((struct inet_timewait_sock *)sk);
1147                         sk = sk2;
1148                         goto process;
1149                 }
1150                 /* Fall through to ACK */
1151         }
1152         case TCP_TW_ACK:
1153                 tcp_v4_timewait_ack(sk, skb);
1154                 break;
1155         case TCP_TW_RST:
1156                 goto no_tcp_socket;
1157         case TCP_TW_SUCCESS:;
1158         }
1159         goto discard_it;
1160 }
1161
1162 /* VJ's idea. Save last timestamp seen from this destination
1163  * and hold it at least for normal timewait interval to use for duplicate
1164  * segment detection in subsequent connections, before they enter synchronized
1165  * state.
1166  */
1167
1168 int tcp_v4_remember_stamp(struct sock *sk)
1169 {
1170         struct inet_sock *inet = inet_sk(sk);
1171         struct tcp_sock *tp = tcp_sk(sk);
1172         struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1173         struct inet_peer *peer = NULL;
1174         int release_it = 0;
1175
1176         if (!rt || rt->rt_dst != inet->daddr) {
1177                 peer = inet_getpeer(inet->daddr, 1);
1178                 release_it = 1;
1179         } else {
1180                 if (!rt->peer)
1181                         rt_bind_peer(rt, 1);
1182                 peer = rt->peer;
1183         }
1184
1185         if (peer) {
1186                 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1187                     (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1188                      peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1189                         peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1190                         peer->tcp_ts = tp->rx_opt.ts_recent;
1191                 }
1192                 if (release_it)
1193                         inet_putpeer(peer);
1194                 return 1;
1195         }
1196
1197         return 0;
1198 }
1199
1200 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1201 {
1202         struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1203
1204         if (peer) {
1205                 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1206
1207                 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1208                     (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1209                      peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1210                         peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1211                         peer->tcp_ts       = tcptw->tw_ts_recent;
1212                 }
1213                 inet_putpeer(peer);
1214                 return 1;
1215         }
1216
1217         return 0;
1218 }
1219
1220 struct inet_connection_sock_af_ops ipv4_specific = {
1221         .queue_xmit        = ip_queue_xmit,
1222         .send_check        = tcp_v4_send_check,
1223         .rebuild_header    = inet_sk_rebuild_header,
1224         .conn_request      = tcp_v4_conn_request,
1225         .syn_recv_sock     = tcp_v4_syn_recv_sock,
1226         .remember_stamp    = tcp_v4_remember_stamp,
1227         .net_header_len    = sizeof(struct iphdr),
1228         .setsockopt        = ip_setsockopt,
1229         .getsockopt        = ip_getsockopt,
1230         .addr2sockaddr     = inet_csk_addr2sockaddr,
1231         .sockaddr_len      = sizeof(struct sockaddr_in),
1232 #ifdef CONFIG_COMPAT
1233         .compat_setsockopt = compat_ip_setsockopt,
1234         .compat_getsockopt = compat_ip_getsockopt,
1235 #endif
1236 };
1237
1238 /* NOTE: A lot of things set to zero explicitly by call to
1239  *       sk_alloc() so need not be done here.
1240  */
1241 static int tcp_v4_init_sock(struct sock *sk)
1242 {
1243         struct inet_connection_sock *icsk = inet_csk(sk);
1244         struct tcp_sock *tp = tcp_sk(sk);
1245
1246         skb_queue_head_init(&tp->out_of_order_queue);
1247         tcp_init_xmit_timers(sk);
1248         tcp_prequeue_init(tp);
1249
1250         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1251         tp->mdev = TCP_TIMEOUT_INIT;
1252
1253         /* So many TCP implementations out there (incorrectly) count the
1254          * initial SYN frame in their delayed-ACK and congestion control
1255          * algorithms that we must have the following bandaid to talk
1256          * efficiently to them.  -DaveM
1257          */
1258         tp->snd_cwnd = 2;
1259
1260         /* See draft-stevens-tcpca-spec-01 for discussion of the
1261          * initialization of these values.
1262          */
1263         tp->snd_ssthresh = 0x7fffffff;  /* Infinity */
1264         tp->snd_cwnd_clamp = ~0;
1265         tp->mss_cache = 536;
1266
1267         tp->reordering = sysctl_tcp_reordering;
1268         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1269
1270         sk->sk_state = TCP_CLOSE;
1271
1272         sk->sk_write_space = sk_stream_write_space;
1273         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1274
1275         icsk->icsk_af_ops = &ipv4_specific;
1276         icsk->icsk_sync_mss = tcp_sync_mss;
1277
1278         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1279         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1280
1281         atomic_inc(&tcp_sockets_allocated);
1282
1283         return 0;
1284 }
1285
1286 int tcp_v4_destroy_sock(struct sock *sk)
1287 {
1288         struct tcp_sock *tp = tcp_sk(sk);
1289
1290         tcp_clear_xmit_timers(sk);
1291
1292         tcp_cleanup_congestion_control(sk);
1293
1294         /* Cleanup up the write buffer. */
1295         sk_stream_writequeue_purge(sk);
1296
1297         /* Cleans up our, hopefully empty, out_of_order_queue. */
1298         __skb_queue_purge(&tp->out_of_order_queue);
1299
1300         /* Clean prequeue, it must be empty really */
1301         __skb_queue_purge(&tp->ucopy.prequeue);
1302
1303         /* Clean up a referenced TCP bind bucket. */
1304         if (inet_csk(sk)->icsk_bind_hash)
1305                 inet_put_port(&tcp_hashinfo, sk);
1306
1307         /*
1308          * If sendmsg cached page exists, toss it.
1309          */
1310         if (sk->sk_sndmsg_page) {
1311                 __free_page(sk->sk_sndmsg_page);
1312                 sk->sk_sndmsg_page = NULL;
1313         }
1314
1315         atomic_dec(&tcp_sockets_allocated);
1316
1317         return 0;
1318 }
1319
1320 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1321
1322 #ifdef CONFIG_PROC_FS
1323 /* Proc filesystem TCP sock list dumping. */
1324
1325 static inline struct inet_timewait_sock *tw_head(struct hlist_head *head)
1326 {
1327         return hlist_empty(head) ? NULL :
1328                 list_entry(head->first, struct inet_timewait_sock, tw_node);
1329 }
1330
1331 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1332 {
1333         return tw->tw_node.next ?
1334                 hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1335 }
1336
1337 static void *listening_get_next(struct seq_file *seq, void *cur)
1338 {
1339         struct inet_connection_sock *icsk;
1340         struct hlist_node *node;
1341         struct sock *sk = cur;
1342         struct tcp_iter_state* st = seq->private;
1343
1344         if (!sk) {
1345                 st->bucket = 0;
1346                 sk = sk_head(&tcp_hashinfo.listening_hash[0]);
1347                 goto get_sk;
1348         }
1349
1350         ++st->num;
1351
1352         if (st->state == TCP_SEQ_STATE_OPENREQ) {
1353                 struct request_sock *req = cur;
1354
1355                 icsk = inet_csk(st->syn_wait_sk);
1356                 req = req->dl_next;
1357                 while (1) {
1358                         while (req) {
1359                                 vxdprintk(VXD_CBIT(net, 6),
1360                                         "sk,req: %p [#%d] (from %d)", req->sk,
1361                                         (req->sk)?req->sk->sk_xid:0, vx_current_xid());
1362                                 if (req->sk &&
1363                                         !vx_check(req->sk->sk_xid, VX_IDENT|VX_WATCH))
1364                                         continue;
1365                                 if (req->rsk_ops->family == st->family) {
1366                                         cur = req;
1367                                         goto out;
1368                                 }
1369                                 req = req->dl_next;
1370                         }
1371                         if (++st->sbucket >= TCP_SYNQ_HSIZE)
1372                                 break;
1373 get_req:
1374                         req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1375                 }
1376                 sk        = sk_next(st->syn_wait_sk);
1377                 st->state = TCP_SEQ_STATE_LISTENING;
1378                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1379         } else {
1380                 icsk = inet_csk(sk);
1381                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1382                 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1383                         goto start_req;
1384                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1385                 sk = sk_next(sk);
1386         }
1387 get_sk:
1388         sk_for_each_from(sk, node) {
1389                 vxdprintk(VXD_CBIT(net, 6), "sk: %p [#%d] (from %d)",
1390                         sk, sk->sk_xid, vx_current_xid());
1391                 if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1392                         continue;
1393                 if (sk->sk_family == st->family) {
1394                         cur = sk;
1395                         goto out;
1396                 }
1397                 icsk = inet_csk(sk);
1398                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1399                 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1400 start_req:
1401                         st->uid         = sock_i_uid(sk);
1402                         st->syn_wait_sk = sk;
1403                         st->state       = TCP_SEQ_STATE_OPENREQ;
1404                         st->sbucket     = 0;
1405                         goto get_req;
1406                 }
1407                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1408         }
1409         if (++st->bucket < INET_LHTABLE_SIZE) {
1410                 sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]);
1411                 goto get_sk;
1412         }
1413         cur = NULL;
1414 out:
1415         return cur;
1416 }
1417
1418 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1419 {
1420         void *rc = listening_get_next(seq, NULL);
1421
1422         while (rc && *pos) {
1423                 rc = listening_get_next(seq, rc);
1424                 --*pos;
1425         }
1426         return rc;
1427 }
1428
1429 static void *established_get_first(struct seq_file *seq)
1430 {
1431         struct tcp_iter_state* st = seq->private;
1432         void *rc = NULL;
1433
1434         for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
1435                 struct sock *sk;
1436                 struct hlist_node *node;
1437                 struct inet_timewait_sock *tw;
1438
1439                 /* We can reschedule _before_ having picked the target: */
1440                 cond_resched_softirq();
1441
1442                 read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1443                 sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1444                         vxdprintk(VXD_CBIT(net, 6),
1445                                 "sk,egf: %p [#%d] (from %d)",
1446                                 sk, sk->sk_xid, vx_current_xid());
1447                         if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1448                                 continue;
1449                         if (sk->sk_family != st->family)
1450                                 continue;
1451                         rc = sk;
1452                         goto out;
1453                 }
1454                 st->state = TCP_SEQ_STATE_TIME_WAIT;
1455                 inet_twsk_for_each(tw, node,
1456                                    &tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain) {
1457                         vxdprintk(VXD_CBIT(net, 6),
1458                                 "tw: %p [#%d] (from %d)",
1459                                 tw, tw->tw_xid, vx_current_xid());
1460                         if (!vx_check(tw->tw_xid, VX_IDENT|VX_WATCH))
1461                                 continue;
1462                         if (tw->tw_family != st->family)
1463                                 continue;
1464                         rc = tw;
1465                         goto out;
1466                 }
1467                 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1468                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1469         }
1470 out:
1471         return rc;
1472 }
1473
1474 static void *established_get_next(struct seq_file *seq, void *cur)
1475 {
1476         struct sock *sk = cur;
1477         struct inet_timewait_sock *tw;
1478         struct hlist_node *node;
1479         struct tcp_iter_state* st = seq->private;
1480
1481         ++st->num;
1482
1483         if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
1484                 tw = cur;
1485                 tw = tw_next(tw);
1486 get_tw:
1487                 while (tw && (tw->tw_family != st->family ||
1488                         !vx_check(tw->tw_xid, VX_IDENT|VX_WATCH))) {
1489                         tw = tw_next(tw);
1490                 }
1491                 if (tw) {
1492                         cur = tw;
1493                         goto out;
1494                 }
1495                 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1496                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1497
1498                 /* We can reschedule between buckets: */
1499                 cond_resched_softirq();
1500
1501                 if (++st->bucket < tcp_hashinfo.ehash_size) {
1502                         read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1503                         sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain);
1504                 } else {
1505                         cur = NULL;
1506                         goto out;
1507                 }
1508         } else
1509                 sk = sk_next(sk);
1510
1511         sk_for_each_from(sk, node) {
1512                 vxdprintk(VXD_CBIT(net, 6),
1513                         "sk,egn: %p [#%d] (from %d)",
1514                         sk, sk->sk_xid, vx_current_xid());
1515                 if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1516                         continue;
1517                 if (sk->sk_family == st->family)
1518                         goto found;
1519         }
1520
1521         st->state = TCP_SEQ_STATE_TIME_WAIT;
1522         tw = tw_head(&tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain);
1523         goto get_tw;
1524 found:
1525         cur = sk;
1526 out:
1527         return cur;
1528 }
1529
1530 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1531 {
1532         void *rc = established_get_first(seq);
1533
1534         while (rc && pos) {
1535                 rc = established_get_next(seq, rc);
1536                 --pos;
1537         }
1538         return rc;
1539 }
1540
1541 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1542 {
1543         void *rc;
1544         struct tcp_iter_state* st = seq->private;
1545
1546         inet_listen_lock(&tcp_hashinfo);
1547         st->state = TCP_SEQ_STATE_LISTENING;
1548         rc        = listening_get_idx(seq, &pos);
1549
1550         if (!rc) {
1551                 inet_listen_unlock(&tcp_hashinfo);
1552                 local_bh_disable();
1553                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1554                 rc        = established_get_idx(seq, pos);
1555         }
1556
1557         return rc;
1558 }
1559
1560 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
1561 {
1562         struct tcp_iter_state* st = seq->private;
1563         st->state = TCP_SEQ_STATE_LISTENING;
1564         st->num = 0;
1565         return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1566 }
1567
1568 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1569 {
1570         void *rc = NULL;
1571         struct tcp_iter_state* st;
1572
1573         if (v == SEQ_START_TOKEN) {
1574                 rc = tcp_get_idx(seq, 0);
1575                 goto out;
1576         }
1577         st = seq->private;
1578
1579         switch (st->state) {
1580         case TCP_SEQ_STATE_OPENREQ:
1581         case TCP_SEQ_STATE_LISTENING:
1582                 rc = listening_get_next(seq, v);
1583                 if (!rc) {
1584                         inet_listen_unlock(&tcp_hashinfo);
1585                         local_bh_disable();
1586                         st->state = TCP_SEQ_STATE_ESTABLISHED;
1587                         rc        = established_get_first(seq);
1588                 }
1589                 break;
1590         case TCP_SEQ_STATE_ESTABLISHED:
1591         case TCP_SEQ_STATE_TIME_WAIT:
1592                 rc = established_get_next(seq, v);
1593                 break;
1594         }
1595 out:
1596         ++*pos;
1597         return rc;
1598 }
1599
1600 static void tcp_seq_stop(struct seq_file *seq, void *v)
1601 {
1602         struct tcp_iter_state* st = seq->private;
1603
1604         switch (st->state) {
1605         case TCP_SEQ_STATE_OPENREQ:
1606                 if (v) {
1607                         struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
1608                         read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1609                 }
1610         case TCP_SEQ_STATE_LISTENING:
1611                 if (v != SEQ_START_TOKEN)
1612                         inet_listen_unlock(&tcp_hashinfo);
1613                 break;
1614         case TCP_SEQ_STATE_TIME_WAIT:
1615         case TCP_SEQ_STATE_ESTABLISHED:
1616                 if (v)
1617                         read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1618                 local_bh_enable();
1619                 break;
1620         }
1621 }
1622
1623 static int tcp_seq_open(struct inode *inode, struct file *file)
1624 {
1625         struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
1626         struct seq_file *seq;
1627         struct tcp_iter_state *s;
1628         int rc;
1629
1630         if (unlikely(afinfo == NULL))
1631                 return -EINVAL;
1632
1633         s = kmalloc(sizeof(*s), GFP_KERNEL);
1634         if (!s)
1635                 return -ENOMEM;
1636         memset(s, 0, sizeof(*s));
1637         s->family               = afinfo->family;
1638         s->seq_ops.start        = tcp_seq_start;
1639         s->seq_ops.next         = tcp_seq_next;
1640         s->seq_ops.show         = afinfo->seq_show;
1641         s->seq_ops.stop         = tcp_seq_stop;
1642
1643         rc = seq_open(file, &s->seq_ops);
1644         if (rc)
1645                 goto out_kfree;
1646         seq          = file->private_data;
1647         seq->private = s;
1648 out:
1649         return rc;
1650 out_kfree:
1651         kfree(s);
1652         goto out;
1653 }
1654
1655 int tcp_proc_register(struct tcp_seq_afinfo *afinfo)
1656 {
1657         int rc = 0;
1658         struct proc_dir_entry *p;
1659
1660         if (!afinfo)
1661                 return -EINVAL;
1662         afinfo->seq_fops->owner         = afinfo->owner;
1663         afinfo->seq_fops->open          = tcp_seq_open;
1664         afinfo->seq_fops->read          = seq_read;
1665         afinfo->seq_fops->llseek        = seq_lseek;
1666         afinfo->seq_fops->release       = seq_release_private;
1667
1668         p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1669         if (p)
1670                 p->data = afinfo;
1671         else
1672                 rc = -ENOMEM;
1673         return rc;
1674 }
1675
1676 void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo)
1677 {
1678         if (!afinfo)
1679                 return;
1680         proc_net_remove(afinfo->name);
1681         memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1682 }
1683
1684 static void get_openreq4(struct sock *sk, struct request_sock *req,
1685                          char *tmpbuf, int i, int uid)
1686 {
1687         const struct inet_request_sock *ireq = inet_rsk(req);
1688         int ttd = req->expires - jiffies;
1689
1690         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1691                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p",
1692                 i,
1693                 ireq->loc_addr,
1694                 ntohs(inet_sk(sk)->sport),
1695                 ireq->rmt_addr,
1696                 ntohs(ireq->rmt_port),
1697                 TCP_SYN_RECV,
1698                 0, 0, /* could print option size, but that is af dependent. */
1699                 1,    /* timers active (only the expire timer) */
1700                 jiffies_to_clock_t(ttd),
1701                 req->retrans,
1702                 uid,
1703                 0,  /* non standard timer */
1704                 0, /* open_requests have no inode */
1705                 atomic_read(&sk->sk_refcnt),
1706                 req);
1707 }
1708
1709 static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i)
1710 {
1711         int timer_active;
1712         unsigned long timer_expires;
1713         struct tcp_sock *tp = tcp_sk(sp);
1714         const struct inet_connection_sock *icsk = inet_csk(sp);
1715         struct inet_sock *inet = inet_sk(sp);
1716         unsigned int dest = inet->daddr;
1717         unsigned int src = inet->rcv_saddr;
1718         __u16 destp = ntohs(inet->dport);
1719         __u16 srcp = ntohs(inet->sport);
1720
1721         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1722                 timer_active    = 1;
1723                 timer_expires   = icsk->icsk_timeout;
1724         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1725                 timer_active    = 4;
1726                 timer_expires   = icsk->icsk_timeout;
1727         } else if (timer_pending(&sp->sk_timer)) {
1728                 timer_active    = 2;
1729                 timer_expires   = sp->sk_timer.expires;
1730         } else {
1731                 timer_active    = 0;
1732                 timer_expires = jiffies;
1733         }
1734
1735         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
1736                         "%08X %5d %8d %lu %d %p %u %u %u %u %d",
1737                 i, src, srcp, dest, destp, sp->sk_state,
1738                 tp->write_seq - tp->snd_una, tp->rcv_nxt - tp->copied_seq,
1739                 timer_active,
1740                 jiffies_to_clock_t(timer_expires - jiffies),
1741                 icsk->icsk_retransmits,
1742                 sock_i_uid(sp),
1743                 icsk->icsk_probes_out,
1744                 sock_i_ino(sp),
1745                 atomic_read(&sp->sk_refcnt), sp,
1746                 icsk->icsk_rto,
1747                 icsk->icsk_ack.ato,
1748                 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1749                 tp->snd_cwnd,
1750                 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh);
1751 }
1752
1753 static void get_timewait4_sock(struct inet_timewait_sock *tw, char *tmpbuf, int i)
1754 {
1755         unsigned int dest, src;
1756         __u16 destp, srcp;
1757         int ttd = tw->tw_ttd - jiffies;
1758
1759         if (ttd < 0)
1760                 ttd = 0;
1761
1762         dest  = tw->tw_daddr;
1763         src   = tw->tw_rcv_saddr;
1764         destp = ntohs(tw->tw_dport);
1765         srcp  = ntohs(tw->tw_sport);
1766
1767         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1768                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p",
1769                 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
1770                 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1771                 atomic_read(&tw->tw_refcnt), tw);
1772 }
1773
1774 #define TMPSZ 150
1775
1776 static int tcp4_seq_show(struct seq_file *seq, void *v)
1777 {
1778         struct tcp_iter_state* st;
1779         char tmpbuf[TMPSZ + 1];
1780
1781         if (v == SEQ_START_TOKEN) {
1782                 seq_printf(seq, "%-*s\n", TMPSZ - 1,
1783                            "  sl  local_address rem_address   st tx_queue "
1784                            "rx_queue tr tm->when retrnsmt   uid  timeout "
1785                            "inode");
1786                 goto out;
1787         }
1788         st = seq->private;
1789
1790         switch (st->state) {
1791         case TCP_SEQ_STATE_LISTENING:
1792         case TCP_SEQ_STATE_ESTABLISHED:
1793                 get_tcp4_sock(v, tmpbuf, st->num);
1794                 break;
1795         case TCP_SEQ_STATE_OPENREQ:
1796                 get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid);
1797                 break;
1798         case TCP_SEQ_STATE_TIME_WAIT:
1799                 get_timewait4_sock(v, tmpbuf, st->num);
1800                 break;
1801         }
1802         seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf);
1803 out:
1804         return 0;
1805 }
1806
1807 static struct file_operations tcp4_seq_fops;
1808 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
1809         .owner          = THIS_MODULE,
1810         .name           = "tcp",
1811         .family         = AF_INET,
1812         .seq_show       = tcp4_seq_show,
1813         .seq_fops       = &tcp4_seq_fops,
1814 };
1815
1816 int __init tcp4_proc_init(void)
1817 {
1818         return tcp_proc_register(&tcp4_seq_afinfo);
1819 }
1820
1821 void tcp4_proc_exit(void)
1822 {
1823         tcp_proc_unregister(&tcp4_seq_afinfo);
1824 }
1825 #endif /* CONFIG_PROC_FS */
1826
1827 struct proto tcp_prot = {
1828         .name                   = "TCP",
1829         .owner                  = THIS_MODULE,
1830         .close                  = tcp_close,
1831         .connect                = tcp_v4_connect,
1832         .disconnect             = tcp_disconnect,
1833         .accept                 = inet_csk_accept,
1834         .ioctl                  = tcp_ioctl,
1835         .init                   = tcp_v4_init_sock,
1836         .destroy                = tcp_v4_destroy_sock,
1837         .shutdown               = tcp_shutdown,
1838         .setsockopt             = tcp_setsockopt,
1839         .getsockopt             = tcp_getsockopt,
1840         .sendmsg                = tcp_sendmsg,
1841         .recvmsg                = tcp_recvmsg,
1842         .backlog_rcv            = tcp_v4_do_rcv,
1843         .hash                   = tcp_v4_hash,
1844         .unhash                 = tcp_unhash,
1845         .get_port               = tcp_v4_get_port,
1846         .enter_memory_pressure  = tcp_enter_memory_pressure,
1847         .sockets_allocated      = &tcp_sockets_allocated,
1848         .orphan_count           = &tcp_orphan_count,
1849         .memory_allocated       = &tcp_memory_allocated,
1850         .memory_pressure        = &tcp_memory_pressure,
1851         .sysctl_mem             = sysctl_tcp_mem,
1852         .sysctl_wmem            = sysctl_tcp_wmem,
1853         .sysctl_rmem            = sysctl_tcp_rmem,
1854         .max_header             = MAX_TCP_HEADER,
1855         .obj_size               = sizeof(struct tcp_sock),
1856         .twsk_prot              = &tcp_timewait_sock_ops,
1857         .rsk_prot               = &tcp_request_sock_ops,
1858 #ifdef CONFIG_COMPAT
1859         .compat_setsockopt      = compat_tcp_setsockopt,
1860         .compat_getsockopt      = compat_tcp_getsockopt,
1861 #endif
1862 };
1863
1864 void __init tcp_v4_init(struct net_proto_family *ops)
1865 {
1866         if (inet_csk_ctl_sock_create(&tcp_socket, PF_INET, SOCK_RAW, IPPROTO_TCP) < 0)
1867                 panic("Failed to create the TCP control socket.\n");
1868 }
1869
1870 EXPORT_SYMBOL(ipv4_specific);
1871 EXPORT_SYMBOL(tcp_hashinfo);
1872 EXPORT_SYMBOL(tcp_prot);
1873 EXPORT_SYMBOL(tcp_unhash);
1874 EXPORT_SYMBOL(tcp_v4_conn_request);
1875 EXPORT_SYMBOL(tcp_v4_connect);
1876 EXPORT_SYMBOL(tcp_v4_do_rcv);
1877 EXPORT_SYMBOL(tcp_v4_remember_stamp);
1878 EXPORT_SYMBOL(tcp_v4_send_check);
1879 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1880
1881 #ifdef CONFIG_PROC_FS
1882 EXPORT_SYMBOL(tcp_proc_register);
1883 EXPORT_SYMBOL(tcp_proc_unregister);
1884 #endif
1885 EXPORT_SYMBOL(sysctl_local_port_range);
1886 EXPORT_SYMBOL(sysctl_tcp_low_latency);
1887