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
  56 #include <linux/types.h>
  57 #include <linux/fcntl.h>
  58 #include <linux/module.h>
  59 #include <linux/random.h>
  60 #include <linux/cache.h>
  61 #include <linux/jhash.h>
  62 #include <linux/init.h>
  63 #include <linux/times.h>
  64
  65 #include <net/icmp.h>
  66 #include <net/inet_hashtables.h>
  67 #include <net/tcp.h>
  68 #include <net/transp_v6.h>
  69 #include <net/ipv6.h>
  70 #include <net/inet_common.h>
  71 #include <net/timewait_sock.h>
  72 #include <net/xfrm.h>
  73 #include <net/netdma.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(tcp_hashinfo.lhash_lock),
  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->sk_gso_type = SKB_GSO_TCPV4;
 246         sk_setup_caps(sk, &rt->u.dst);
 247
 248         if (!tp->write_seq)
 249                 tp->write_seq = secure_tcp_sequence_number(inet->saddr,
 250                                                            inet->daddr,
 251                                                            inet->sport,
 252                                                            usin->sin_port);
 253
 254         inet->id = tp->write_seq ^ jiffies;
 255
 256         err = tcp_connect(sk);
 257         rt = NULL;
 258         if (err)
 259                 goto failure;
 260
 261         return 0;
 262
 263 failure:
 264         /* This unhashes the socket and releases the local port, if necessary. */
 265         tcp_set_state(sk, TCP_CLOSE);
 266         ip_rt_put(rt);
 267         sk->sk_route_caps = 0;
 268         inet->dport = 0;
 269         return err;
 270 }
 271
 272 /*
 273  * This routine does path mtu discovery as defined in RFC1191.
 274  */
 275 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
 276 {
 277         struct dst_entry *dst;
 278         struct inet_sock *inet = inet_sk(sk);
 279
 280         /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
 281          * send out by Linux are always <576bytes so they should go through
 282          * unfragmented).
 283          */
 284         if (sk->sk_state == TCP_LISTEN)
 285                 return;
 286
 287         /* We don't check in the destentry if pmtu discovery is forbidden
 288          * on this route. We just assume that no packet_to_big packets
 289          * are send back when pmtu discovery is not active.
 290          * There is a small race when the user changes this flag in the
 291          * route, but I think that's acceptable.
 292          */
 293         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 294                 return;
 295
 296         dst->ops->update_pmtu(dst, mtu);
 297
 298         /* Something is about to be wrong... Remember soft error
 299          * for the case, if this connection will not able to recover.
 300          */
 301         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 302                 sk->sk_err_soft = EMSGSIZE;
 303
 304         mtu = dst_mtu(dst);
 305
 306         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 307             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 308                 tcp_sync_mss(sk, mtu);
 309
 310                 /* Resend the TCP packet because it's
 311                  * clear that the old packet has been
 312                  * dropped. This is the new "fast" path mtu
 313                  * discovery.
 314                  */
 315                 tcp_simple_retransmit(sk);
 316         } /* else let the usual retransmit timer handle it */
 317 }
 318
 319 /*
 320  * This routine is called by the ICMP module when it gets some
 321  * sort of error condition.  If err < 0 then the socket should
 322  * be closed and the error returned to the user.  If err > 0
 323  * it's just the icmp type << 8 | icmp code.  After adjustment
 324  * header points to the first 8 bytes of the tcp header.  We need
 325  * to find the appropriate port.
 326  *
 327  * The locking strategy used here is very "optimistic". When
 328  * someone else accesses the socket the ICMP is just dropped
 329  * and for some paths there is no check at all.
 330  * A more general error queue to queue errors for later handling
 331  * is probably better.
 332  *
 333  */
 334
 335 void tcp_v4_err(struct sk_buff *skb, u32 info)
 336 {
 337         struct iphdr *iph = (struct iphdr *)skb->data;
 338         struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
 339         struct tcp_sock *tp;
 340         struct inet_sock *inet;
 341         int type = skb->h.icmph->type;
 342         int code = skb->h.icmph->code;
 343         struct sock *sk;
 344         __u32 seq;
 345         int err;
 346
 347         if (skb->len < (iph->ihl << 2) + 8) {
 348                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 349                 return;
 350         }
 351
 352         sk = inet_lookup(&tcp_hashinfo, iph->daddr, th->dest, iph->saddr,
 353                          th->source, inet_iif(skb));
 354         if (!sk) {
 355                 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
 356                 return;
 357         }
 358         if (sk->sk_state == TCP_TIME_WAIT) {
 359                 inet_twsk_put((struct inet_timewait_sock *)sk);
 360                 return;
 361         }
 362
 363         bh_lock_sock(sk);
 364         /* If too many ICMPs get dropped on busy
 365          * servers this needs to be solved differently.
 366          */
 367         if (sock_owned_by_user(sk))
 368                 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
 369
 370         if (sk->sk_state == TCP_CLOSE)
 371                 goto out;
 372
 373         tp = tcp_sk(sk);
 374         seq = ntohl(th->seq);
 375         if (sk->sk_state != TCP_LISTEN &&
 376             !between(seq, tp->snd_una, tp->snd_nxt)) {
 377                 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
 378                 goto out;
 379         }
 380
 381         switch (type) {
 382         case ICMP_SOURCE_QUENCH:
 383                 /* Just silently ignore these. */
 384                 goto out;
 385         case ICMP_PARAMETERPROB:
 386                 err = EPROTO;
 387                 break;
 388         case ICMP_DEST_UNREACH:
 389                 if (code > NR_ICMP_UNREACH)
 390                         goto out;
 391
 392                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 393                         if (!sock_owned_by_user(sk))
 394                                 do_pmtu_discovery(sk, iph, info);
 395                         goto out;
 396                 }
 397
 398                 err = icmp_err_convert[code].errno;
 399                 break;
 400         case ICMP_TIME_EXCEEDED:
 401                 err = EHOSTUNREACH;
 402                 break;
 403         default:
 404                 goto out;
 405         }
 406
 407         switch (sk->sk_state) {
 408                 struct request_sock *req, **prev;
 409         case TCP_LISTEN:
 410                 if (sock_owned_by_user(sk))
 411                         goto out;
 412
 413                 req = inet_csk_search_req(sk, &prev, th->dest,
 414                                           iph->daddr, iph->saddr);
 415                 if (!req)
 416                         goto out;
 417
 418                 /* ICMPs are not backlogged, hence we cannot get
 419                    an established socket here.
 420                  */
 421                 BUG_TRAP(!req->sk);
 422
 423                 if (seq != tcp_rsk(req)->snt_isn) {
 424                         NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
 425                         goto out;
 426                 }
 427
 428                 /*
 429                  * Still in SYN_RECV, just remove it silently.
 430                  * There is no good way to pass the error to the newly
 431                  * created socket, and POSIX does not want network
 432                  * errors returned from accept().
 433                  */
 434                 inet_csk_reqsk_queue_drop(sk, req, prev);
 435                 goto out;
 436
 437         case TCP_SYN_SENT:
 438         case TCP_SYN_RECV:  /* Cannot happen.
 439                                It can f.e. if SYNs crossed.
 440                              */
 441                 if (!sock_owned_by_user(sk)) {
 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         return 0;
 878 }
 879
 880
 881 /*
 882  * The three way handshake has completed - we got a valid synack -
 883  * now create the new socket.
 884  */
 885 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
 886                                   struct request_sock *req,
 887                                   struct dst_entry *dst)
 888 {
 889         struct inet_request_sock *ireq;
 890         struct inet_sock *newinet;
 891         struct tcp_sock *newtp;
 892         struct sock *newsk;
 893
 894         if (sk_acceptq_is_full(sk))
 895                 goto exit_overflow;
 896
 897         if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
 898                 goto exit;
 899
 900         newsk = tcp_create_openreq_child(sk, req, skb);
 901         if (!newsk)
 902                 goto exit;
 903
 904         newsk->sk_gso_type = SKB_GSO_TCPV4;
 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_nested(sk);
1111         ret = 0;
1112         if (!sock_owned_by_user(sk)) {
1113 #ifdef CONFIG_NET_DMA
1114                 struct tcp_sock *tp = tcp_sk(sk);
1115                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1116                         tp->ucopy.dma_chan = get_softnet_dma();
1117                 if (tp->ucopy.dma_chan)
1118                         ret = tcp_v4_do_rcv(sk, skb);
1119                 else
1120 #endif
1121                 {
1122                         if (!tcp_prequeue(sk, skb))
1123                         ret = tcp_v4_do_rcv(sk, skb);
1124                 }
1125         } else
1126                 sk_add_backlog(sk, skb);
1127         bh_unlock_sock(sk);
1128
1129         sock_put(sk);
1130
1131         return ret;
1132
1133 no_tcp_socket:
1134         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1135                 goto discard_it;
1136
1137         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1138 bad_packet:
1139                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1140         } else {
1141                 tcp_v4_send_reset(skb);
1142         }
1143
1144 discard_it:
1145         /* Discard frame. */
1146         kfree_skb(skb);
1147         return 0;
1148
1149 discard_and_relse:
1150         sock_put(sk);
1151         goto discard_it;
1152
1153 do_time_wait:
1154         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1155                 inet_twsk_put((struct inet_timewait_sock *) sk);
1156                 goto discard_it;
1157         }
1158
1159         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1160                 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1161                 inet_twsk_put((struct inet_timewait_sock *) sk);
1162                 goto discard_it;
1163         }
1164         switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk,
1165                                            skb, th)) {
1166         case TCP_TW_SYN: {
1167                 struct sock *sk2 = inet_lookup_listener(&tcp_hashinfo,
1168                                                         skb->nh.iph->daddr,
1169                                                         ntohs(th->dest),
1170                                                         inet_iif(skb));
1171                 if (sk2) {
1172                         inet_twsk_deschedule((struct inet_timewait_sock *)sk,
1173                                              &tcp_death_row);
1174                         inet_twsk_put((struct inet_timewait_sock *)sk);
1175                         sk = sk2;
1176                         goto process;
1177                 }
1178                 /* Fall through to ACK */
1179         }
1180         case TCP_TW_ACK:
1181                 tcp_v4_timewait_ack(sk, skb);
1182                 break;
1183         case TCP_TW_RST:
1184                 goto no_tcp_socket;
1185         case TCP_TW_SUCCESS:;
1186         }
1187         goto discard_it;
1188 }
1189
1190 /* VJ's idea. Save last timestamp seen from this destination
1191  * and hold it at least for normal timewait interval to use for duplicate
1192  * segment detection in subsequent connections, before they enter synchronized
1193  * state.
1194  */
1195
1196 int tcp_v4_remember_stamp(struct sock *sk)
1197 {
1198         struct inet_sock *inet = inet_sk(sk);
1199         struct tcp_sock *tp = tcp_sk(sk);
1200         struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1201         struct inet_peer *peer = NULL;
1202         int release_it = 0;
1203
1204         if (!rt || rt->rt_dst != inet->daddr) {
1205                 peer = inet_getpeer(inet->daddr, 1);
1206                 release_it = 1;
1207         } else {
1208                 if (!rt->peer)
1209                         rt_bind_peer(rt, 1);
1210                 peer = rt->peer;
1211         }
1212
1213         if (peer) {
1214                 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1215                     (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1216                      peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1217                         peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1218                         peer->tcp_ts = tp->rx_opt.ts_recent;
1219                 }
1220                 if (release_it)
1221                         inet_putpeer(peer);
1222                 return 1;
1223         }
1224
1225         return 0;
1226 }
1227
1228 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1229 {
1230         struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1231
1232         if (peer) {
1233                 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1234
1235                 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1236                     (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1237                      peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1238                         peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1239                         peer->tcp_ts       = tcptw->tw_ts_recent;
1240                 }
1241                 inet_putpeer(peer);
1242                 return 1;
1243         }
1244
1245         return 0;
1246 }
1247
1248 struct inet_connection_sock_af_ops ipv4_specific = {
1249         .queue_xmit        = ip_queue_xmit,
1250         .send_check        = tcp_v4_send_check,
1251         .rebuild_header    = inet_sk_rebuild_header,
1252         .conn_request      = tcp_v4_conn_request,
1253         .syn_recv_sock     = tcp_v4_syn_recv_sock,
1254         .remember_stamp    = tcp_v4_remember_stamp,
1255         .net_header_len    = sizeof(struct iphdr),
1256         .setsockopt        = ip_setsockopt,
1257         .getsockopt        = ip_getsockopt,
1258         .addr2sockaddr     = inet_csk_addr2sockaddr,
1259         .sockaddr_len      = sizeof(struct sockaddr_in),
1260 #ifdef CONFIG_COMPAT
1261         .compat_setsockopt = compat_ip_setsockopt,
1262         .compat_getsockopt = compat_ip_getsockopt,
1263 #endif
1264 };
1265
1266 /* NOTE: A lot of things set to zero explicitly by call to
1267  *       sk_alloc() so need not be done here.
1268  */
1269 static int tcp_v4_init_sock(struct sock *sk)
1270 {
1271         struct inet_connection_sock *icsk = inet_csk(sk);
1272         struct tcp_sock *tp = tcp_sk(sk);
1273
1274         skb_queue_head_init(&tp->out_of_order_queue);
1275         tcp_init_xmit_timers(sk);
1276         tcp_prequeue_init(tp);
1277
1278         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1279         tp->mdev = TCP_TIMEOUT_INIT;
1280
1281         /* So many TCP implementations out there (incorrectly) count the
1282          * initial SYN frame in their delayed-ACK and congestion control
1283          * algorithms that we must have the following bandaid to talk
1284          * efficiently to them.  -DaveM
1285          */
1286         tp->snd_cwnd = 2;
1287
1288         /* See draft-stevens-tcpca-spec-01 for discussion of the
1289          * initialization of these values.
1290          */
1291         tp->snd_ssthresh = 0x7fffffff;  /* Infinity */
1292         tp->snd_cwnd_clamp = ~0;
1293         tp->mss_cache = 536;
1294
1295         tp->reordering = sysctl_tcp_reordering;
1296         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1297
1298         sk->sk_state = TCP_CLOSE;
1299
1300         sk->sk_write_space = sk_stream_write_space;
1301         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1302
1303         icsk->icsk_af_ops = &ipv4_specific;
1304         icsk->icsk_sync_mss = tcp_sync_mss;
1305
1306         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1307         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1308
1309         atomic_inc(&tcp_sockets_allocated);
1310
1311         return 0;
1312 }
1313
1314 int tcp_v4_destroy_sock(struct sock *sk)
1315 {
1316         struct tcp_sock *tp = tcp_sk(sk);
1317
1318         tcp_clear_xmit_timers(sk);
1319
1320         tcp_cleanup_congestion_control(sk);
1321
1322         /* Cleanup up the write buffer. */
1323         sk_stream_writequeue_purge(sk);
1324
1325         /* Cleans up our, hopefully empty, out_of_order_queue. */
1326         __skb_queue_purge(&tp->out_of_order_queue);
1327
1328 #ifdef CONFIG_NET_DMA
1329         /* Cleans up our sk_async_wait_queue */
1330         __skb_queue_purge(&sk->sk_async_wait_queue);
1331 #endif
1332
1333         /* Clean prequeue, it must be empty really */
1334         __skb_queue_purge(&tp->ucopy.prequeue);
1335
1336         /* Clean up a referenced TCP bind bucket. */
1337         if (inet_csk(sk)->icsk_bind_hash)
1338                 inet_put_port(&tcp_hashinfo, sk);
1339
1340         /*
1341          * If sendmsg cached page exists, toss it.
1342          */
1343         if (sk->sk_sndmsg_page) {
1344                 __free_page(sk->sk_sndmsg_page);
1345                 sk->sk_sndmsg_page = NULL;
1346         }
1347
1348         atomic_dec(&tcp_sockets_allocated);
1349
1350         return 0;
1351 }
1352
1353 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1354
1355 #ifdef CONFIG_PROC_FS
1356 /* Proc filesystem TCP sock list dumping. */
1357
1358 static inline struct inet_timewait_sock *tw_head(struct hlist_head *head)
1359 {
1360         return hlist_empty(head) ? NULL :
1361                 list_entry(head->first, struct inet_timewait_sock, tw_node);
1362 }
1363
1364 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1365 {
1366         return tw->tw_node.next ?
1367                 hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1368 }
1369
1370 static void *listening_get_next(struct seq_file *seq, void *cur)
1371 {
1372         struct inet_connection_sock *icsk;
1373         struct hlist_node *node;
1374         struct sock *sk = cur;
1375         struct tcp_iter_state* st = seq->private;
1376
1377         if (!sk) {
1378                 st->bucket = 0;
1379                 sk = sk_head(&tcp_hashinfo.listening_hash[0]);
1380                 goto get_sk;
1381         }
1382
1383         ++st->num;
1384
1385         if (st->state == TCP_SEQ_STATE_OPENREQ) {
1386                 struct request_sock *req = cur;
1387
1388                 icsk = inet_csk(st->syn_wait_sk);
1389                 req = req->dl_next;
1390                 while (1) {
1391                         while (req) {
1392                                 vxdprintk(VXD_CBIT(net, 6),
1393                                         "sk,req: %p [#%d] (from %d)", req->sk,
1394                                         (req->sk)?req->sk->sk_xid:0, vx_current_xid());
1395                                 if (req->sk &&
1396                                         !vx_check(req->sk->sk_xid, VX_IDENT|VX_WATCH))
1397                                         continue;
1398                                 if (req->rsk_ops->family == st->family) {
1399                                         cur = req;
1400                                         goto out;
1401                                 }
1402                                 req = req->dl_next;
1403                         }
1404                         if (++st->sbucket >= TCP_SYNQ_HSIZE)
1405                                 break;
1406 get_req:
1407                         req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1408                 }
1409                 sk        = sk_next(st->syn_wait_sk);
1410                 st->state = TCP_SEQ_STATE_LISTENING;
1411                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1412         } else {
1413                 icsk = inet_csk(sk);
1414                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1415                 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1416                         goto start_req;
1417                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1418                 sk = sk_next(sk);
1419         }
1420 get_sk:
1421         sk_for_each_from(sk, node) {
1422                 vxdprintk(VXD_CBIT(net, 6), "sk: %p [#%d] (from %d)",
1423                         sk, sk->sk_xid, vx_current_xid());
1424                 if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1425                         continue;
1426                 if (sk->sk_family == st->family) {
1427                         cur = sk;
1428                         goto out;
1429                 }
1430                 icsk = inet_csk(sk);
1431                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1432                 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1433 start_req:
1434                         st->uid         = sock_i_uid(sk);
1435                         st->syn_wait_sk = sk;
1436                         st->state       = TCP_SEQ_STATE_OPENREQ;
1437                         st->sbucket     = 0;
1438                         goto get_req;
1439                 }
1440                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1441         }
1442         if (++st->bucket < INET_LHTABLE_SIZE) {
1443                 sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]);
1444                 goto get_sk;
1445         }
1446         cur = NULL;
1447 out:
1448         return cur;
1449 }
1450
1451 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1452 {
1453         void *rc = listening_get_next(seq, NULL);
1454
1455         while (rc && *pos) {
1456                 rc = listening_get_next(seq, rc);
1457                 --*pos;
1458         }
1459         return rc;
1460 }
1461
1462 static void *established_get_first(struct seq_file *seq)
1463 {
1464         struct tcp_iter_state* st = seq->private;
1465         void *rc = NULL;
1466
1467         for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
1468                 struct sock *sk;
1469                 struct hlist_node *node;
1470                 struct inet_timewait_sock *tw;
1471
1472                 /* We can reschedule _before_ having picked the target: */
1473                 cond_resched_softirq();
1474
1475                 read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1476                 sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1477                         vxdprintk(VXD_CBIT(net, 6),
1478                                 "sk,egf: %p [#%d] (from %d)",
1479                                 sk, sk->sk_xid, vx_current_xid());
1480                         if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1481                                 continue;
1482                         if (sk->sk_family != st->family)
1483                                 continue;
1484                         rc = sk;
1485                         goto out;
1486                 }
1487                 st->state = TCP_SEQ_STATE_TIME_WAIT;
1488                 inet_twsk_for_each(tw, node,
1489                                    &tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain) {
1490                         vxdprintk(VXD_CBIT(net, 6),
1491                                 "tw: %p [#%d] (from %d)",
1492                                 tw, tw->tw_xid, vx_current_xid());
1493                         if (!vx_check(tw->tw_xid, VX_IDENT|VX_WATCH))
1494                                 continue;
1495                         if (tw->tw_family != st->family)
1496                                 continue;
1497                         rc = tw;
1498                         goto out;
1499                 }
1500                 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1501                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1502         }
1503 out:
1504         return rc;
1505 }
1506
1507 static void *established_get_next(struct seq_file *seq, void *cur)
1508 {
1509         struct sock *sk = cur;
1510         struct inet_timewait_sock *tw;
1511         struct hlist_node *node;
1512         struct tcp_iter_state* st = seq->private;
1513
1514         ++st->num;
1515
1516         if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
1517                 tw = cur;
1518                 tw = tw_next(tw);
1519 get_tw:
1520                 while (tw && (tw->tw_family != st->family ||
1521                         !vx_check(tw->tw_xid, VX_IDENT|VX_WATCH))) {
1522                         tw = tw_next(tw);
1523                 }
1524                 if (tw) {
1525                         cur = tw;
1526                         goto out;
1527                 }
1528                 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1529                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1530
1531                 /* We can reschedule between buckets: */
1532                 cond_resched_softirq();
1533
1534                 if (++st->bucket < tcp_hashinfo.ehash_size) {
1535                         read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1536                         sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain);
1537                 } else {
1538                         cur = NULL;
1539                         goto out;
1540                 }
1541         } else
1542                 sk = sk_next(sk);
1543
1544         sk_for_each_from(sk, node) {
1545                 vxdprintk(VXD_CBIT(net, 6),
1546                         "sk,egn: %p [#%d] (from %d)",
1547                         sk, sk->sk_xid, vx_current_xid());
1548                 if (!vx_check(sk->sk_xid, VX_IDENT|VX_WATCH))
1549                         continue;
1550                 if (sk->sk_family == st->family)
1551                         goto found;
1552         }
1553
1554         st->state = TCP_SEQ_STATE_TIME_WAIT;
1555         tw = tw_head(&tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain);
1556         goto get_tw;
1557 found:
1558         cur = sk;
1559 out:
1560         return cur;
1561 }
1562
1563 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1564 {
1565         void *rc = established_get_first(seq);
1566
1567         while (rc && pos) {
1568                 rc = established_get_next(seq, rc);
1569                 --pos;
1570         }
1571         return rc;
1572 }
1573
1574 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1575 {
1576         void *rc;
1577         struct tcp_iter_state* st = seq->private;
1578
1579         inet_listen_lock(&tcp_hashinfo);
1580         st->state = TCP_SEQ_STATE_LISTENING;
1581         rc        = listening_get_idx(seq, &pos);
1582
1583         if (!rc) {
1584                 inet_listen_unlock(&tcp_hashinfo);
1585                 local_bh_disable();
1586                 st->state = TCP_SEQ_STATE_ESTABLISHED;
1587                 rc        = established_get_idx(seq, pos);
1588         }
1589
1590         return rc;
1591 }
1592
1593 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
1594 {
1595         struct tcp_iter_state* st = seq->private;
1596         st->state = TCP_SEQ_STATE_LISTENING;
1597         st->num = 0;
1598         return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1599 }
1600
1601 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1602 {
1603         void *rc = NULL;
1604         struct tcp_iter_state* st;
1605
1606         if (v == SEQ_START_TOKEN) {
1607                 rc = tcp_get_idx(seq, 0);
1608                 goto out;
1609         }
1610         st = seq->private;
1611
1612         switch (st->state) {
1613         case TCP_SEQ_STATE_OPENREQ:
1614         case TCP_SEQ_STATE_LISTENING:
1615                 rc = listening_get_next(seq, v);
1616                 if (!rc) {
1617                         inet_listen_unlock(&tcp_hashinfo);
1618                         local_bh_disable();
1619                         st->state = TCP_SEQ_STATE_ESTABLISHED;
1620                         rc        = established_get_first(seq);
1621                 }
1622                 break;
1623         case TCP_SEQ_STATE_ESTABLISHED:
1624         case TCP_SEQ_STATE_TIME_WAIT:
1625                 rc = established_get_next(seq, v);
1626                 break;
1627         }
1628 out:
1629         ++*pos;
1630         return rc;
1631 }
1632
1633 static void tcp_seq_stop(struct seq_file *seq, void *v)
1634 {
1635         struct tcp_iter_state* st = seq->private;
1636
1637         switch (st->state) {
1638         case TCP_SEQ_STATE_OPENREQ:
1639                 if (v) {
1640                         struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
1641                         read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1642                 }
1643         case TCP_SEQ_STATE_LISTENING:
1644                 if (v != SEQ_START_TOKEN)
1645                         inet_listen_unlock(&tcp_hashinfo);
1646                 break;
1647         case TCP_SEQ_STATE_TIME_WAIT:
1648         case TCP_SEQ_STATE_ESTABLISHED:
1649                 if (v)
1650                         read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1651                 local_bh_enable();
1652                 break;
1653         }
1654 }
1655
1656 static int tcp_seq_open(struct inode *inode, struct file *file)
1657 {
1658         struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
1659         struct seq_file *seq;
1660         struct tcp_iter_state *s;
1661         int rc;
1662
1663         if (unlikely(afinfo == NULL))
1664                 return -EINVAL;
1665
1666         s = kzalloc(sizeof(*s), GFP_KERNEL);
1667         if (!s)
1668                 return -ENOMEM;
1669         s->family               = afinfo->family;
1670         s->seq_ops.start        = tcp_seq_start;
1671         s->seq_ops.next         = tcp_seq_next;
1672         s->seq_ops.show         = afinfo->seq_show;
1673         s->seq_ops.stop         = tcp_seq_stop;
1674
1675         rc = seq_open(file, &s->seq_ops);
1676         if (rc)
1677                 goto out_kfree;
1678         seq          = file->private_data;
1679         seq->private = s;
1680 out:
1681         return rc;
1682 out_kfree:
1683         kfree(s);
1684         goto out;
1685 }
1686
1687 int tcp_proc_register(struct tcp_seq_afinfo *afinfo)
1688 {
1689         int rc = 0;
1690         struct proc_dir_entry *p;
1691
1692         if (!afinfo)
1693                 return -EINVAL;
1694         afinfo->seq_fops->owner         = afinfo->owner;
1695         afinfo->seq_fops->open          = tcp_seq_open;
1696         afinfo->seq_fops->read          = seq_read;
1697         afinfo->seq_fops->llseek        = seq_lseek;
1698         afinfo->seq_fops->release       = seq_release_private;
1699
1700         p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1701         if (p)
1702                 p->data = afinfo;
1703         else
1704                 rc = -ENOMEM;
1705         return rc;
1706 }
1707
1708 void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo)
1709 {
1710         if (!afinfo)
1711                 return;
1712         proc_net_remove(afinfo->name);
1713         memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1714 }
1715
1716 static void get_openreq4(struct sock *sk, struct request_sock *req,
1717                          char *tmpbuf, int i, int uid)
1718 {
1719         const struct inet_request_sock *ireq = inet_rsk(req);
1720         int ttd = req->expires - jiffies;
1721
1722         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1723                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p",
1724                 i,
1725                 ireq->loc_addr,
1726                 ntohs(inet_sk(sk)->sport),
1727                 ireq->rmt_addr,
1728                 ntohs(ireq->rmt_port),
1729                 TCP_SYN_RECV,
1730                 0, 0, /* could print option size, but that is af dependent. */
1731                 1,    /* timers active (only the expire timer) */
1732                 jiffies_to_clock_t(ttd),
1733                 req->retrans,
1734                 uid,
1735                 0,  /* non standard timer */
1736                 0, /* open_requests have no inode */
1737                 atomic_read(&sk->sk_refcnt),
1738                 req);
1739 }
1740
1741 static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i)
1742 {
1743         int timer_active;
1744         unsigned long timer_expires;
1745         struct tcp_sock *tp = tcp_sk(sp);
1746         const struct inet_connection_sock *icsk = inet_csk(sp);
1747         struct inet_sock *inet = inet_sk(sp);
1748         unsigned int dest = inet->daddr;
1749         unsigned int src = inet->rcv_saddr;
1750         __u16 destp = ntohs(inet->dport);
1751         __u16 srcp = ntohs(inet->sport);
1752
1753         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1754                 timer_active    = 1;
1755                 timer_expires   = icsk->icsk_timeout;
1756         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1757                 timer_active    = 4;
1758                 timer_expires   = icsk->icsk_timeout;
1759         } else if (timer_pending(&sp->sk_timer)) {
1760                 timer_active    = 2;
1761                 timer_expires   = sp->sk_timer.expires;
1762         } else {
1763                 timer_active    = 0;
1764                 timer_expires = jiffies;
1765         }
1766
1767         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
1768                         "%08X %5d %8d %lu %d %p %u %u %u %u %d",
1769                 i, src, srcp, dest, destp, sp->sk_state,
1770                 tp->write_seq - tp->snd_una,
1771                 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1772                 timer_active,
1773                 jiffies_to_clock_t(timer_expires - jiffies),
1774                 icsk->icsk_retransmits,
1775                 sock_i_uid(sp),
1776                 icsk->icsk_probes_out,
1777                 sock_i_ino(sp),
1778                 atomic_read(&sp->sk_refcnt), sp,
1779                 icsk->icsk_rto,
1780                 icsk->icsk_ack.ato,
1781                 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1782                 tp->snd_cwnd,
1783                 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh);
1784 }
1785
1786 static void get_timewait4_sock(struct inet_timewait_sock *tw, char *tmpbuf, int i)
1787 {
1788         unsigned int dest, src;
1789         __u16 destp, srcp;
1790         int ttd = tw->tw_ttd - jiffies;
1791
1792         if (ttd < 0)
1793                 ttd = 0;
1794
1795         dest  = tw->tw_daddr;
1796         src   = tw->tw_rcv_saddr;
1797         destp = ntohs(tw->tw_dport);
1798         srcp  = ntohs(tw->tw_sport);
1799
1800         sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1801                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p",
1802                 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
1803                 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1804                 atomic_read(&tw->tw_refcnt), tw);
1805 }
1806
1807 #define TMPSZ 150
1808
1809 static int tcp4_seq_show(struct seq_file *seq, void *v)
1810 {
1811         struct tcp_iter_state* st;
1812         char tmpbuf[TMPSZ + 1];
1813
1814         if (v == SEQ_START_TOKEN) {
1815                 seq_printf(seq, "%-*s\n", TMPSZ - 1,
1816                            "  sl  local_address rem_address   st tx_queue "
1817                            "rx_queue tr tm->when retrnsmt   uid  timeout "
1818                            "inode");
1819                 goto out;
1820         }
1821         st = seq->private;
1822
1823         switch (st->state) {
1824         case TCP_SEQ_STATE_LISTENING:
1825         case TCP_SEQ_STATE_ESTABLISHED:
1826                 get_tcp4_sock(v, tmpbuf, st->num);
1827                 break;
1828         case TCP_SEQ_STATE_OPENREQ:
1829                 get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid);
1830                 break;
1831         case TCP_SEQ_STATE_TIME_WAIT:
1832                 get_timewait4_sock(v, tmpbuf, st->num);
1833                 break;
1834         }
1835         seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf);
1836 out:
1837         return 0;
1838 }
1839
1840 static struct file_operations tcp4_seq_fops;
1841 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
1842         .owner          = THIS_MODULE,
1843         .name           = "tcp",
1844         .family         = AF_INET,
1845         .seq_show       = tcp4_seq_show,
1846         .seq_fops       = &tcp4_seq_fops,
1847 };
1848
1849 int __init tcp4_proc_init(void)
1850 {
1851         return tcp_proc_register(&tcp4_seq_afinfo);
1852 }
1853
1854 void tcp4_proc_exit(void)
1855 {
1856         tcp_proc_unregister(&tcp4_seq_afinfo);
1857 }
1858 #endif /* CONFIG_PROC_FS */
1859
1860 struct proto tcp_prot = {
1861         .name                   = "TCP",
1862         .owner                  = THIS_MODULE,
1863         .close                  = tcp_close,
1864         .connect                = tcp_v4_connect,
1865         .disconnect             = tcp_disconnect,
1866         .accept                 = inet_csk_accept,
1867         .ioctl                  = tcp_ioctl,
1868         .init                   = tcp_v4_init_sock,
1869         .destroy                = tcp_v4_destroy_sock,
1870         .shutdown               = tcp_shutdown,
1871         .setsockopt             = tcp_setsockopt,
1872         .getsockopt             = tcp_getsockopt,
1873         .sendmsg                = tcp_sendmsg,
1874         .recvmsg                = tcp_recvmsg,
1875         .backlog_rcv            = tcp_v4_do_rcv,
1876         .hash                   = tcp_v4_hash,
1877         .unhash                 = tcp_unhash,
1878         .get_port               = tcp_v4_get_port,
1879         .enter_memory_pressure  = tcp_enter_memory_pressure,
1880         .sockets_allocated      = &tcp_sockets_allocated,
1881         .orphan_count           = &tcp_orphan_count,
1882         .memory_allocated       = &tcp_memory_allocated,
1883         .memory_pressure        = &tcp_memory_pressure,
1884         .sysctl_mem             = sysctl_tcp_mem,
1885         .sysctl_wmem            = sysctl_tcp_wmem,
1886         .sysctl_rmem            = sysctl_tcp_rmem,
1887         .max_header             = MAX_TCP_HEADER,
1888         .obj_size               = sizeof(struct tcp_sock),
1889         .twsk_prot              = &tcp_timewait_sock_ops,
1890         .rsk_prot               = &tcp_request_sock_ops,
1891 #ifdef CONFIG_COMPAT
1892         .compat_setsockopt      = compat_tcp_setsockopt,
1893         .compat_getsockopt      = compat_tcp_getsockopt,
1894 #endif
1895 };
1896
1897 void __init tcp_v4_init(struct net_proto_family *ops)
1898 {
1899         if (inet_csk_ctl_sock_create(&tcp_socket, PF_INET, SOCK_RAW, IPPROTO_TCP) < 0)
1900                 panic("Failed to create the TCP control socket.\n");
1901 }
1902
1903 EXPORT_SYMBOL(ipv4_specific);
1904 EXPORT_SYMBOL(tcp_hashinfo);
1905 EXPORT_SYMBOL(tcp_prot);
1906 EXPORT_SYMBOL(tcp_unhash);
1907 EXPORT_SYMBOL(tcp_v4_conn_request);
1908 EXPORT_SYMBOL(tcp_v4_connect);
1909 EXPORT_SYMBOL(tcp_v4_do_rcv);
1910 EXPORT_SYMBOL(tcp_v4_remember_stamp);
1911 EXPORT_SYMBOL(tcp_v4_send_check);
1912 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1913
1914 #ifdef CONFIG_PROC_FS
1915 EXPORT_SYMBOL(tcp_proc_register);
1916 EXPORT_SYMBOL(tcp_proc_unregister);
1917 #endif
1918 EXPORT_SYMBOL(sysctl_local_port_range);
1919 EXPORT_SYMBOL(sysctl_tcp_low_latency);
1920