Home | History | Annotate | Line # | Download | only in netinet
tcp_subr.c revision 1.37
      1 /*	$NetBSD: tcp_subr.c,v 1.37 1997/12/31 03:31:26 thorpej Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by the University of
     18  *	California, Berkeley and its contributors.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  *
     35  *	@(#)tcp_subr.c	8.1 (Berkeley) 6/10/93
     36  */
     37 
     38 #include "rnd.h"
     39 
     40 #include <sys/param.h>
     41 #include <sys/proc.h>
     42 #include <sys/systm.h>
     43 #include <sys/malloc.h>
     44 #include <sys/mbuf.h>
     45 #include <sys/socket.h>
     46 #include <sys/socketvar.h>
     47 #include <sys/protosw.h>
     48 #include <sys/errno.h>
     49 #include <sys/kernel.h>
     50 #if NRND > 0
     51 #include <sys/rnd.h>
     52 #endif
     53 
     54 #include <net/route.h>
     55 #include <net/if.h>
     56 
     57 #include <netinet/in.h>
     58 #include <netinet/in_systm.h>
     59 #include <netinet/ip.h>
     60 #include <netinet/in_pcb.h>
     61 #include <netinet/ip_var.h>
     62 #include <netinet/ip_icmp.h>
     63 #include <netinet/tcp.h>
     64 #include <netinet/tcp_fsm.h>
     65 #include <netinet/tcp_seq.h>
     66 #include <netinet/tcp_timer.h>
     67 #include <netinet/tcp_var.h>
     68 #include <netinet/tcpip.h>
     69 
     70 /* patchable/settable parameters for tcp */
     71 int 	tcp_mssdflt = TCP_MSS;
     72 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
     73 int	tcp_do_rfc1323 = 1;
     74 int	tcp_init_win = 1;
     75 
     76 #ifndef TCBHASHSIZE
     77 #define	TCBHASHSIZE	128
     78 #endif
     79 int	tcbhashsize = TCBHASHSIZE;
     80 
     81 int	tcp_freeq __P((struct tcpcb *));
     82 
     83 /*
     84  * Tcp initialization
     85  */
     86 void
     87 tcp_init()
     88 {
     89 
     90 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
     91 	LIST_INIT(&tcp_delacks);
     92 	if (max_protohdr < sizeof(struct tcpiphdr))
     93 		max_protohdr = sizeof(struct tcpiphdr);
     94 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
     95 		panic("tcp_init");
     96 }
     97 
     98 /*
     99  * Create template to be used to send tcp packets on a connection.
    100  * Call after host entry created, allocates an mbuf and fills
    101  * in a skeletal tcp/ip header, minimizing the amount of work
    102  * necessary when the connection is used.
    103  */
    104 struct tcpiphdr *
    105 tcp_template(tp)
    106 	struct tcpcb *tp;
    107 {
    108 	register struct inpcb *inp = tp->t_inpcb;
    109 	register struct tcpiphdr *n;
    110 
    111 	if ((n = tp->t_template) == 0) {
    112 		MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
    113 		    M_MBUF, M_NOWAIT);
    114 		if (n == NULL)
    115 			return (0);
    116 	}
    117 	bzero(n->ti_x1, sizeof n->ti_x1);
    118 	n->ti_pr = IPPROTO_TCP;
    119 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
    120 	n->ti_src = inp->inp_laddr;
    121 	n->ti_dst = inp->inp_faddr;
    122 	n->ti_sport = inp->inp_lport;
    123 	n->ti_dport = inp->inp_fport;
    124 	n->ti_seq = 0;
    125 	n->ti_ack = 0;
    126 	n->ti_x2 = 0;
    127 	n->ti_off = 5;
    128 	n->ti_flags = 0;
    129 	n->ti_win = 0;
    130 	n->ti_sum = 0;
    131 	n->ti_urp = 0;
    132 	return (n);
    133 }
    134 
    135 /*
    136  * Send a single message to the TCP at address specified by
    137  * the given TCP/IP header.  If m == 0, then we make a copy
    138  * of the tcpiphdr at ti and send directly to the addressed host.
    139  * This is used to force keep alive messages out using the TCP
    140  * template for a connection tp->t_template.  If flags are given
    141  * then we send a message back to the TCP which originated the
    142  * segment ti, and discard the mbuf containing it and any other
    143  * attached mbufs.
    144  *
    145  * In any case the ack and sequence number of the transmitted
    146  * segment are as specified by the parameters.
    147  */
    148 int
    149 tcp_respond(tp, ti, m, ack, seq, flags)
    150 	struct tcpcb *tp;
    151 	register struct tcpiphdr *ti;
    152 	register struct mbuf *m;
    153 	tcp_seq ack, seq;
    154 	int flags;
    155 {
    156 	register int tlen;
    157 	int win = 0;
    158 	struct route *ro = 0;
    159 
    160 	if (tp) {
    161 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
    162 		ro = &tp->t_inpcb->inp_route;
    163 	}
    164 	if (m == 0) {
    165 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
    166 		if (m == NULL)
    167 			return (ENOBUFS);
    168 #ifdef TCP_COMPAT_42
    169 		tlen = 1;
    170 #else
    171 		tlen = 0;
    172 #endif
    173 		m->m_data += max_linkhdr;
    174 		*mtod(m, struct tcpiphdr *) = *ti;
    175 		ti = mtod(m, struct tcpiphdr *);
    176 		flags = TH_ACK;
    177 	} else {
    178 		m_freem(m->m_next);
    179 		m->m_next = 0;
    180 		m->m_data = (caddr_t)ti;
    181 		m->m_len = sizeof (struct tcpiphdr);
    182 		tlen = 0;
    183 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
    184 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
    185 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
    186 #undef xchg
    187 	}
    188 	bzero(ti->ti_x1, sizeof ti->ti_x1);
    189 	ti->ti_seq = htonl(seq);
    190 	ti->ti_ack = htonl(ack);
    191 	ti->ti_x2 = 0;
    192 	if ((flags & TH_SYN) == 0) {
    193 		if (tp)
    194 			ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
    195 		else
    196 			ti->ti_win = htons((u_int16_t)win);
    197 		ti->ti_off = sizeof (struct tcphdr) >> 2;
    198 		tlen += sizeof (struct tcphdr);
    199 	} else
    200 		tlen += ti->ti_off << 2;
    201 	ti->ti_len = htons((u_int16_t)tlen);
    202 	tlen += sizeof (struct ip);
    203 	m->m_len = tlen;
    204 	m->m_pkthdr.len = tlen;
    205 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
    206 	ti->ti_flags = flags;
    207 	ti->ti_urp = 0;
    208 	ti->ti_sum = 0;
    209 	ti->ti_sum = in_cksum(m, tlen);
    210 	((struct ip *)ti)->ip_len = tlen;
    211 	((struct ip *)ti)->ip_ttl = ip_defttl;
    212 	return ip_output(m, NULL, ro, 0, NULL);
    213 }
    214 
    215 /*
    216  * Create a new TCP control block, making an
    217  * empty reassembly queue and hooking it to the argument
    218  * protocol control block.
    219  */
    220 struct tcpcb *
    221 tcp_newtcpcb(inp)
    222 	struct inpcb *inp;
    223 {
    224 	register struct tcpcb *tp;
    225 
    226 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
    227 	if (tp == NULL)
    228 		return ((struct tcpcb *)0);
    229 	bzero((caddr_t)tp, sizeof(struct tcpcb));
    230 	LIST_INIT(&tp->segq);
    231 	tp->t_peermss = tcp_mssdflt;
    232 	tp->t_ourmss = tcp_mssdflt;
    233 	tp->t_segsz = tcp_mssdflt;
    234 
    235 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
    236 	tp->t_inpcb = inp;
    237 	/*
    238 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
    239 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
    240 	 * reasonable initial retransmit time.
    241 	 */
    242 	tp->t_srtt = TCPTV_SRTTBASE;
    243 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
    244 	tp->t_rttmin = TCPTV_MIN;
    245 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
    246 	    TCPTV_MIN, TCPTV_REXMTMAX);
    247 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    248 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    249 	inp->inp_ip.ip_ttl = ip_defttl;
    250 	inp->inp_ppcb = (caddr_t)tp;
    251 	return (tp);
    252 }
    253 
    254 /*
    255  * Drop a TCP connection, reporting
    256  * the specified error.  If connection is synchronized,
    257  * then send a RST to peer.
    258  */
    259 struct tcpcb *
    260 tcp_drop(tp, errno)
    261 	register struct tcpcb *tp;
    262 	int errno;
    263 {
    264 	struct socket *so = tp->t_inpcb->inp_socket;
    265 
    266 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
    267 		tp->t_state = TCPS_CLOSED;
    268 		(void) tcp_output(tp);
    269 		tcpstat.tcps_drops++;
    270 	} else
    271 		tcpstat.tcps_conndrops++;
    272 	if (errno == ETIMEDOUT && tp->t_softerror)
    273 		errno = tp->t_softerror;
    274 	so->so_error = errno;
    275 	return (tcp_close(tp));
    276 }
    277 
    278 /*
    279  * Close a TCP control block:
    280  *	discard all space held by the tcp
    281  *	discard internet protocol block
    282  *	wake up any sleepers
    283  */
    284 struct tcpcb *
    285 tcp_close(tp)
    286 	register struct tcpcb *tp;
    287 {
    288 	struct inpcb *inp = tp->t_inpcb;
    289 	struct socket *so = inp->inp_socket;
    290 #ifdef RTV_RTT
    291 	register struct rtentry *rt;
    292 
    293 	/*
    294 	 * If we sent enough data to get some meaningful characteristics,
    295 	 * save them in the routing entry.  'Enough' is arbitrarily
    296 	 * defined as the sendpipesize (default 4K) * 16.  This would
    297 	 * give us 16 rtt samples assuming we only get one sample per
    298 	 * window (the usual case on a long haul net).  16 samples is
    299 	 * enough for the srtt filter to converge to within 5% of the correct
    300 	 * value; fewer samples and we could save a very bogus rtt.
    301 	 *
    302 	 * Don't update the default route's characteristics and don't
    303 	 * update anything that the user "locked".
    304 	 */
    305 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
    306 	    (rt = inp->inp_route.ro_rt) &&
    307 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
    308 		register u_long i = 0;
    309 
    310 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
    311 			i = tp->t_srtt *
    312 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    313 			if (rt->rt_rmx.rmx_rtt && i)
    314 				/*
    315 				 * filter this update to half the old & half
    316 				 * the new values, converting scale.
    317 				 * See route.h and tcp_var.h for a
    318 				 * description of the scaling constants.
    319 				 */
    320 				rt->rt_rmx.rmx_rtt =
    321 				    (rt->rt_rmx.rmx_rtt + i) / 2;
    322 			else
    323 				rt->rt_rmx.rmx_rtt = i;
    324 		}
    325 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
    326 			i = tp->t_rttvar *
    327 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
    328 			if (rt->rt_rmx.rmx_rttvar && i)
    329 				rt->rt_rmx.rmx_rttvar =
    330 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
    331 			else
    332 				rt->rt_rmx.rmx_rttvar = i;
    333 		}
    334 		/*
    335 		 * update the pipelimit (ssthresh) if it has been updated
    336 		 * already or if a pipesize was specified & the threshhold
    337 		 * got below half the pipesize.  I.e., wait for bad news
    338 		 * before we start updating, then update on both good
    339 		 * and bad news.
    340 		 */
    341 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
    342 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
    343 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
    344 			/*
    345 			 * convert the limit from user data bytes to
    346 			 * packets then to packet data bytes.
    347 			 */
    348 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
    349 			if (i < 2)
    350 				i = 2;
    351 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
    352 			if (rt->rt_rmx.rmx_ssthresh)
    353 				rt->rt_rmx.rmx_ssthresh =
    354 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
    355 			else
    356 				rt->rt_rmx.rmx_ssthresh = i;
    357 		}
    358 	}
    359 #endif /* RTV_RTT */
    360 	/* free the reassembly queue, if any */
    361 	(void) tcp_freeq(tp);
    362 
    363 	if (tp->t_template)
    364 		FREE(tp->t_template, M_MBUF);
    365 	free(tp, M_PCB);
    366 	inp->inp_ppcb = 0;
    367 	soisdisconnected(so);
    368 	in_pcbdetach(inp);
    369 	tcpstat.tcps_closed++;
    370 	return ((struct tcpcb *)0);
    371 }
    372 
    373 int
    374 tcp_freeq(tp)
    375 	struct tcpcb *tp;
    376 {
    377 	register struct ipqent *qe;
    378 	int rv = 0;
    379 
    380 	while ((qe = tp->segq.lh_first) != NULL) {
    381 		LIST_REMOVE(qe, ipqe_q);
    382 		m_freem(qe->ipqe_m);
    383 		FREE(qe, M_IPQ);
    384 		rv = 1;
    385 	}
    386 	return (rv);
    387 }
    388 
    389 /*
    390  * Protocol drain routine.  Called when memory is in short supply.
    391  */
    392 void
    393 tcp_drain()
    394 {
    395 	register struct inpcb *inp;
    396 	register struct tcpcb *tp;
    397 
    398 	/*
    399 	 * Free the sequence queue of all TCP connections.
    400 	 */
    401 	inp = tcbtable.inpt_queue.cqh_first;
    402 	if (inp)						/* XXX */
    403 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
    404 	    inp = inp->inp_queue.cqe_next) {
    405 		if ((tp = intotcpcb(inp)) != NULL) {
    406 			if (tcp_freeq(tp))
    407 				tcpstat.tcps_connsdrained++;
    408 		}
    409 	}
    410 }
    411 
    412 /*
    413  * Notify a tcp user of an asynchronous error;
    414  * store error as soft error, but wake up user
    415  * (for now, won't do anything until can select for soft error).
    416  */
    417 void
    418 tcp_notify(inp, error)
    419 	struct inpcb *inp;
    420 	int error;
    421 {
    422 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
    423 	register struct socket *so = inp->inp_socket;
    424 
    425 	/*
    426 	 * Ignore some errors if we are hooked up.
    427 	 * If connection hasn't completed, has retransmitted several times,
    428 	 * and receives a second error, give up now.  This is better
    429 	 * than waiting a long time to establish a connection that
    430 	 * can never complete.
    431 	 */
    432 	if (tp->t_state == TCPS_ESTABLISHED &&
    433 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
    434 	      error == EHOSTDOWN)) {
    435 		return;
    436 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
    437 	    tp->t_rxtshift > 3 && tp->t_softerror)
    438 		so->so_error = error;
    439 	else
    440 		tp->t_softerror = error;
    441 	wakeup((caddr_t) &so->so_timeo);
    442 	sorwakeup(so);
    443 	sowwakeup(so);
    444 }
    445 
    446 void *
    447 tcp_ctlinput(cmd, sa, v)
    448 	int cmd;
    449 	struct sockaddr *sa;
    450 	register void *v;
    451 {
    452 	register struct ip *ip = v;
    453 	register struct tcphdr *th;
    454 	extern int inetctlerrmap[];
    455 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
    456 	int errno;
    457 	int nmatch;
    458 
    459 	if ((unsigned)cmd >= PRC_NCMDS)
    460 		return NULL;
    461 	errno = inetctlerrmap[cmd];
    462 	if (cmd == PRC_QUENCH)
    463 		notify = tcp_quench;
    464 	else if (PRC_IS_REDIRECT(cmd))
    465 		notify = in_rtchange, ip = 0;
    466 	else if (cmd == PRC_MSGSIZE && ip_mtudisc)
    467 		notify = tcp_mtudisc, ip = 0;
    468 	else if (cmd == PRC_HOSTDEAD)
    469 		ip = 0;
    470 	else if (errno == 0)
    471 		return NULL;
    472 	if (ip) {
    473 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
    474 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
    475 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
    476 		if (nmatch == 0 && syn_cache_count &&
    477 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
    478 		    inetctlerrmap[cmd] == ENETUNREACH ||
    479 		    inetctlerrmap[cmd] == EHOSTDOWN))
    480 			syn_cache_unreach(ip, th);
    481 	} else
    482 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
    483 		    notify);
    484 	return NULL;
    485 }
    486 
    487 /*
    488  * When a source quench is received, close congestion window
    489  * to one segment.  We will gradually open it again as we proceed.
    490  */
    491 void
    492 tcp_quench(inp, errno)
    493 	struct inpcb *inp;
    494 	int errno;
    495 {
    496 	struct tcpcb *tp = intotcpcb(inp);
    497 
    498 	if (tp)
    499 		tp->snd_cwnd = tp->t_segsz;
    500 }
    501 
    502 /*
    503  * On receipt of path MTU corrections, flush old route and replace it
    504  * with the new one.  Retransmit all unacknowledged packets, to ensure
    505  * that all packets will be received.
    506  */
    507 void
    508 tcp_mtudisc(inp, errno)
    509 	struct inpcb *inp;
    510 	int errno;
    511 {
    512 	struct tcpcb *tp = intotcpcb(inp);
    513 	struct rtentry *rt = in_pcbrtentry(inp);
    514 
    515 	if (tp != 0) {
    516 		if (rt != 0) {
    517 			/*
    518 			 * If this was not a host route, remove and realloc.
    519 			 */
    520 			if ((rt->rt_flags & RTF_HOST) == 0) {
    521 				in_rtchange(inp, errno);
    522 				if ((rt = in_pcbrtentry(inp)) == 0)
    523 					return;
    524 			}
    525 
    526 			/*
    527 			 * Slow start out of the error condition.  We
    528 			 * use the MTU because we know it's smaller
    529 			 * than the previously transmitted segment.
    530 			 */
    531 			if (rt->rt_rmx.rmx_mtu != 0)
    532 				tp->snd_cwnd =
    533 				    TCP_INITIAL_WINDOW(rt->rt_rmx.rmx_mtu);
    534 		}
    535 
    536 		/*
    537 		 * Resend unacknowledged packets.
    538 		 */
    539 		tp->snd_nxt = tp->snd_una;
    540 		tcp_output(tp);
    541 	}
    542 }
    543 
    544 
    545 /*
    546  * Compute the MSS to advertise to the peer.  Called only during
    547  * the 3-way handshake.  If we are the server (peer initiated
    548  * connection), we are called with the TCPCB for the listen
    549  * socket.  If we are the client (we initiated connection), we
    550  * are called witht he TCPCB for the actual connection.
    551  */
    552 int
    553 tcp_mss_to_advertise(tp)
    554 	const struct tcpcb *tp;
    555 {
    556 	extern u_long in_maxmtu;
    557 	struct inpcb *inp;
    558 	struct socket *so;
    559 	int mss;
    560 
    561 	inp = tp->t_inpcb;
    562 	so = inp->inp_socket;
    563 
    564 	/*
    565 	 * In order to avoid defeating path MTU discovery on the peer,
    566 	 * we advertise the max MTU of all attached networks as our MSS,
    567 	 * per RFC 1191, section 3.1.
    568 	 *
    569 	 * XXX Should we allow room for the timestamp option if
    570 	 * XXX rfc1323 is enabled?
    571 	 */
    572 	mss = in_maxmtu - sizeof(struct tcpiphdr);
    573 
    574 	return (mss);
    575 }
    576 
    577 /*
    578  * Set connection variables based on the peer's advertised MSS.
    579  * We are passed the TCPCB for the actual connection.  If we
    580  * are the server, we are called by the compressed state engine
    581  * when the 3-way handshake is complete.  If we are the client,
    582  * we are called when we recieve the SYN,ACK from the server.
    583  *
    584  * NOTE: Our advertised MSS value must be initialized in the TCPCB
    585  * before this routine is called!
    586  */
    587 void
    588 tcp_mss_from_peer(tp, offer)
    589 	struct tcpcb *tp;
    590 	int offer;
    591 {
    592 	struct inpcb *inp = tp->t_inpcb;
    593 	struct socket *so = inp->inp_socket;
    594 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
    595 	struct rtentry *rt = in_pcbrtentry(inp);
    596 #endif
    597 	u_long bufsize;
    598 	int mss;
    599 
    600 	/*
    601 	 * Assume our MSS is the MSS of the peer, unless they sent us
    602 	 * an offer.  Do not accept offers less than 32 bytes.
    603 	 */
    604 	mss = tp->t_ourmss;
    605 	if (offer)
    606 		mss = offer;
    607 	mss = max(mss, 32);		/* sanity */
    608 
    609 	/*
    610 	 * If there's a pipesize, change the socket buffer to that size.
    611 	 * Make the socket buffer an integral number of MSS units.  If
    612 	 * the MSS is larger than the socket buffer, artificially decrease
    613 	 * the MSS.
    614 	 */
    615 #ifdef RTV_SPIPE
    616 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
    617 		bufsize = rt->rt_rmx.rmx_sendpipe;
    618 	else
    619 #endif
    620 		bufsize = so->so_snd.sb_hiwat;
    621 	if (bufsize < mss)
    622 		mss = bufsize;
    623 	else {
    624 		bufsize = roundup(bufsize, mss);
    625 		if (bufsize > sb_max)
    626 			bufsize = sb_max;
    627 		(void) sbreserve(&so->so_snd, bufsize);
    628 	}
    629 	tp->t_peermss = mss;
    630 	tp->t_segsz = mss;
    631 
    632 	/* Initialize the initial congestion window. */
    633 	tp->snd_cwnd = TCP_INITIAL_WINDOW(mss);
    634 
    635 #ifdef RTV_SSTHRESH
    636 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
    637 		/*
    638 		 * There's some sort of gateway or interface buffer
    639 		 * limit on the path.  Use this to set the slow
    640 		 * start threshold, but set the threshold to no less
    641 		 * than 2 * MSS.
    642 		 */
    643 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
    644 	}
    645 #endif
    646 }
    647 
    648 /*
    649  * Processing necessary when a TCP connection is established.
    650  */
    651 void
    652 tcp_established(tp)
    653 	struct tcpcb *tp;
    654 {
    655 	struct inpcb *inp = tp->t_inpcb;
    656 	struct socket *so = inp->inp_socket;
    657 #ifdef RTV_RPIPE
    658 	struct rtentry *rt = in_pcbrtentry(inp);
    659 #endif
    660 	u_long bufsize;
    661 
    662 	tp->t_state = TCPS_ESTABLISHED;
    663 	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
    664 
    665 #ifdef RTV_RPIPE
    666 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
    667 		bufsize = rt->rt_rmx.rmx_recvpipe;
    668 	else
    669 #endif
    670 		bufsize = so->so_rcv.sb_hiwat;
    671 	if (bufsize > tp->t_ourmss) {
    672 		bufsize = roundup(bufsize, tp->t_ourmss);
    673 		if (bufsize > sb_max)
    674 			bufsize = sb_max;
    675 		(void) sbreserve(&so->so_rcv, bufsize);
    676 	}
    677 }
    678 
    679 /*
    680  * Check if there's an initial rtt or rttvar.  Convert from the
    681  * route-table units to scaled multiples of the slow timeout timer.
    682  * Called only during the 3-way handshake.
    683  */
    684 void
    685 tcp_rmx_rtt(tp)
    686 	struct tcpcb *tp;
    687 {
    688 #ifdef RTV_RTT
    689 	struct rtentry *rt;
    690 	int rtt;
    691 
    692 	if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
    693 		return;
    694 
    695 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
    696 		/*
    697 		 * XXX The lock bit for MTU indicates that the value
    698 		 * is also a minimum value; this is subject to time.
    699 		 */
    700 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
    701 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
    702 		tp->t_srtt = rtt /
    703 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    704 		if (rt->rt_rmx.rmx_rttvar) {
    705 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
    706 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
    707 				(TCP_RTTVAR_SHIFT + 2));
    708 		} else {
    709 			/* Default variation is +- 1 rtt */
    710 			tp->t_rttvar =
    711 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
    712 		}
    713 		TCPT_RANGESET(tp->t_rxtcur,
    714 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
    715 		    tp->t_rttmin, TCPTV_REXMTMAX);
    716 	}
    717 #endif
    718 }
    719 
    720 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
    721 
    722 /*
    723  * Get a new sequence value given a tcp control block
    724  */
    725 tcp_seq
    726 tcp_new_iss(tp, len, addin)
    727 	void            *tp;
    728 	u_long           len;
    729 	tcp_seq		 addin;
    730 {
    731 	tcp_seq          tcp_iss;
    732 
    733 	/*
    734 	 * add randomness about this connection, but do not estimate
    735 	 * entropy from the timing, since the physical device driver would
    736 	 * have done that for us.
    737 	 */
    738 #if NRND > 0
    739 	if (tp != NULL)
    740 		rnd_add_data(NULL, tp, len, 0);
    741 #endif
    742 
    743 	/*
    744 	 * randomize.
    745 	 */
    746 #if NRND > 0
    747 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
    748 #else
    749 	tcp_iss = random();
    750 #endif
    751 
    752 	/*
    753 	 * If we were asked to add some amount to a known value,
    754 	 * we will take a random value obtained above, mask off the upper
    755 	 * bits, and add in the known value.  We also add in a constant to
    756 	 * ensure that we are at least a certain distance from the original
    757 	 * value.
    758 	 *
    759 	 * This is used when an old connection is in timed wait
    760 	 * and we have a new one coming in, for instance.
    761 	 */
    762 	if (addin != 0) {
    763 #ifdef TCPISS_DEBUG
    764 		printf("Random %08x, ", tcp_iss);
    765 #endif
    766 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    767 		tcp_iss = tcp_iss + addin + TCP_ISSINCR;
    768 		tcp_iss_seq += TCP_ISSINCR;
    769 		tcp_iss += tcp_iss_seq;
    770 #ifdef TCPISS_DEBUG
    771 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
    772 #endif
    773 	} else {
    774 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    775 		tcp_iss_seq += TCP_ISSINCR;
    776 		tcp_iss += tcp_iss_seq;
    777 #ifdef TCPISS_DEBUG
    778 		printf("ISS %08x\n", tcp_iss);
    779 #endif
    780 	}
    781 
    782 #ifdef TCP_COMPAT_42
    783 	/*
    784 	 * limit it to the positive range for really old TCP implementations
    785 	 */
    786 	if ((int)tcp_iss < 0)
    787 		tcp_iss &= 0x7fffffff;		/* XXX */
    788 #endif
    789 
    790 	return tcp_iss;
    791 }
    792