Home | History | Annotate | Line # | Download | only in netinet
tcp_subr.c revision 1.40
      1 /*	$NetBSD: tcp_subr.c,v 1.40 1998/01/30 08:42:11 mellon Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
      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	8.2 (Berkeley) 5/24/95
     36  */
     37 
     38 #include "opt_tcp_compat_42.h"
     39 #include "rnd.h"
     40 
     41 #include <sys/param.h>
     42 #include <sys/proc.h>
     43 #include <sys/systm.h>
     44 #include <sys/malloc.h>
     45 #include <sys/mbuf.h>
     46 #include <sys/socket.h>
     47 #include <sys/socketvar.h>
     48 #include <sys/protosw.h>
     49 #include <sys/errno.h>
     50 #include <sys/kernel.h>
     51 #if NRND > 0
     52 #include <sys/rnd.h>
     53 #endif
     54 
     55 #include <net/route.h>
     56 #include <net/if.h>
     57 
     58 #include <netinet/in.h>
     59 #include <netinet/in_systm.h>
     60 #include <netinet/ip.h>
     61 #include <netinet/in_pcb.h>
     62 #include <netinet/ip_var.h>
     63 #include <netinet/ip_icmp.h>
     64 #include <netinet/tcp.h>
     65 #include <netinet/tcp_fsm.h>
     66 #include <netinet/tcp_seq.h>
     67 #include <netinet/tcp_timer.h>
     68 #include <netinet/tcp_var.h>
     69 #include <netinet/tcpip.h>
     70 
     71 /* patchable/settable parameters for tcp */
     72 int 	tcp_mssdflt = TCP_MSS;
     73 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
     74 int	tcp_do_rfc1323 = 1;
     75 int	tcp_init_win = 1;
     76 
     77 #ifndef TCBHASHSIZE
     78 #define	TCBHASHSIZE	128
     79 #endif
     80 int	tcbhashsize = TCBHASHSIZE;
     81 
     82 int	tcp_freeq __P((struct tcpcb *));
     83 
     84 /*
     85  * Tcp initialization
     86  */
     87 void
     88 tcp_init()
     89 {
     90 
     91 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
     92 	LIST_INIT(&tcp_delacks);
     93 	if (max_protohdr < sizeof(struct tcpiphdr))
     94 		max_protohdr = sizeof(struct tcpiphdr);
     95 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
     96 		panic("tcp_init");
     97 }
     98 
     99 /*
    100  * Create template to be used to send tcp packets on a connection.
    101  * Call after host entry created, allocates an mbuf and fills
    102  * in a skeletal tcp/ip header, minimizing the amount of work
    103  * necessary when the connection is used.
    104  */
    105 struct tcpiphdr *
    106 tcp_template(tp)
    107 	struct tcpcb *tp;
    108 {
    109 	register struct inpcb *inp = tp->t_inpcb;
    110 	register struct tcpiphdr *n;
    111 
    112 	if ((n = tp->t_template) == 0) {
    113 		MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
    114 		    M_MBUF, M_NOWAIT);
    115 		if (n == NULL)
    116 			return (0);
    117 	}
    118 	bzero(n->ti_x1, sizeof n->ti_x1);
    119 	n->ti_pr = IPPROTO_TCP;
    120 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
    121 	n->ti_src = inp->inp_laddr;
    122 	n->ti_dst = inp->inp_faddr;
    123 	n->ti_sport = inp->inp_lport;
    124 	n->ti_dport = inp->inp_fport;
    125 	n->ti_seq = 0;
    126 	n->ti_ack = 0;
    127 	n->ti_x2 = 0;
    128 	n->ti_off = 5;
    129 	n->ti_flags = 0;
    130 	n->ti_win = 0;
    131 	n->ti_sum = 0;
    132 	n->ti_urp = 0;
    133 	return (n);
    134 }
    135 
    136 /*
    137  * Send a single message to the TCP at address specified by
    138  * the given TCP/IP header.  If m == 0, then we make a copy
    139  * of the tcpiphdr at ti and send directly to the addressed host.
    140  * This is used to force keep alive messages out using the TCP
    141  * template for a connection tp->t_template.  If flags are given
    142  * then we send a message back to the TCP which originated the
    143  * segment ti, and discard the mbuf containing it and any other
    144  * attached mbufs.
    145  *
    146  * In any case the ack and sequence number of the transmitted
    147  * segment are as specified by the parameters.
    148  */
    149 int
    150 tcp_respond(tp, ti, m, ack, seq, flags)
    151 	struct tcpcb *tp;
    152 	register struct tcpiphdr *ti;
    153 	register struct mbuf *m;
    154 	tcp_seq ack, seq;
    155 	int flags;
    156 {
    157 	register int tlen;
    158 	int win = 0;
    159 	struct route *ro = 0;
    160 
    161 	if (tp) {
    162 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
    163 		ro = &tp->t_inpcb->inp_route;
    164 	}
    165 	if (m == 0) {
    166 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
    167 		if (m == NULL)
    168 			return (ENOBUFS);
    169 #ifdef TCP_COMPAT_42
    170 		tlen = 1;
    171 #else
    172 		tlen = 0;
    173 #endif
    174 		m->m_data += max_linkhdr;
    175 		*mtod(m, struct tcpiphdr *) = *ti;
    176 		ti = mtod(m, struct tcpiphdr *);
    177 		flags = TH_ACK;
    178 	} else {
    179 		m_freem(m->m_next);
    180 		m->m_next = 0;
    181 		m->m_data = (caddr_t)ti;
    182 		m->m_len = sizeof (struct tcpiphdr);
    183 		tlen = 0;
    184 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
    185 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
    186 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
    187 #undef xchg
    188 	}
    189 	bzero(ti->ti_x1, sizeof ti->ti_x1);
    190 	ti->ti_seq = htonl(seq);
    191 	ti->ti_ack = htonl(ack);
    192 	ti->ti_x2 = 0;
    193 	if ((flags & TH_SYN) == 0) {
    194 		if (tp)
    195 			ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
    196 		else
    197 			ti->ti_win = htons((u_int16_t)win);
    198 		ti->ti_off = sizeof (struct tcphdr) >> 2;
    199 		tlen += sizeof (struct tcphdr);
    200 	} else
    201 		tlen += ti->ti_off << 2;
    202 	ti->ti_len = htons((u_int16_t)tlen);
    203 	tlen += sizeof (struct ip);
    204 	m->m_len = tlen;
    205 	m->m_pkthdr.len = tlen;
    206 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
    207 	ti->ti_flags = flags;
    208 	ti->ti_urp = 0;
    209 	ti->ti_sum = 0;
    210 	ti->ti_sum = in_cksum(m, tlen);
    211 	((struct ip *)ti)->ip_len = tlen;
    212 	((struct ip *)ti)->ip_ttl = ip_defttl;
    213 	return ip_output(m, NULL, ro, 0, NULL);
    214 }
    215 
    216 /*
    217  * Create a new TCP control block, making an
    218  * empty reassembly queue and hooking it to the argument
    219  * protocol control block.
    220  */
    221 struct tcpcb *
    222 tcp_newtcpcb(inp)
    223 	struct inpcb *inp;
    224 {
    225 	register struct tcpcb *tp;
    226 
    227 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
    228 	if (tp == NULL)
    229 		return ((struct tcpcb *)0);
    230 	bzero((caddr_t)tp, sizeof(struct tcpcb));
    231 	LIST_INIT(&tp->segq);
    232 	tp->t_peermss = tcp_mssdflt;
    233 	tp->t_ourmss = tcp_mssdflt;
    234 	tp->t_segsz = tcp_mssdflt;
    235 
    236 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
    237 	tp->t_inpcb = inp;
    238 	/*
    239 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
    240 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
    241 	 * reasonable initial retransmit time.
    242 	 */
    243 	tp->t_srtt = TCPTV_SRTTBASE;
    244 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
    245 	tp->t_rttmin = TCPTV_MIN;
    246 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
    247 	    TCPTV_MIN, TCPTV_REXMTMAX);
    248 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    249 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    250 	inp->inp_ip.ip_ttl = ip_defttl;
    251 	inp->inp_ppcb = (caddr_t)tp;
    252 	return (tp);
    253 }
    254 
    255 /*
    256  * Drop a TCP connection, reporting
    257  * the specified error.  If connection is synchronized,
    258  * then send a RST to peer.
    259  */
    260 struct tcpcb *
    261 tcp_drop(tp, errno)
    262 	register struct tcpcb *tp;
    263 	int errno;
    264 {
    265 	struct socket *so = tp->t_inpcb->inp_socket;
    266 
    267 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
    268 		tp->t_state = TCPS_CLOSED;
    269 		(void) tcp_output(tp);
    270 		tcpstat.tcps_drops++;
    271 	} else
    272 		tcpstat.tcps_conndrops++;
    273 	if (errno == ETIMEDOUT && tp->t_softerror)
    274 		errno = tp->t_softerror;
    275 	so->so_error = errno;
    276 	return (tcp_close(tp));
    277 }
    278 
    279 /*
    280  * Close a TCP control block:
    281  *	discard all space held by the tcp
    282  *	discard internet protocol block
    283  *	wake up any sleepers
    284  */
    285 struct tcpcb *
    286 tcp_close(tp)
    287 	register struct tcpcb *tp;
    288 {
    289 	struct inpcb *inp = tp->t_inpcb;
    290 	struct socket *so = inp->inp_socket;
    291 #ifdef RTV_RTT
    292 	register struct rtentry *rt;
    293 
    294 	/*
    295 	 * If we sent enough data to get some meaningful characteristics,
    296 	 * save them in the routing entry.  'Enough' is arbitrarily
    297 	 * defined as the sendpipesize (default 4K) * 16.  This would
    298 	 * give us 16 rtt samples assuming we only get one sample per
    299 	 * window (the usual case on a long haul net).  16 samples is
    300 	 * enough for the srtt filter to converge to within 5% of the correct
    301 	 * value; fewer samples and we could save a very bogus rtt.
    302 	 *
    303 	 * Don't update the default route's characteristics and don't
    304 	 * update anything that the user "locked".
    305 	 */
    306 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
    307 	    (rt = inp->inp_route.ro_rt) &&
    308 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
    309 		register u_long i = 0;
    310 
    311 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
    312 			i = tp->t_srtt *
    313 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    314 			if (rt->rt_rmx.rmx_rtt && i)
    315 				/*
    316 				 * filter this update to half the old & half
    317 				 * the new values, converting scale.
    318 				 * See route.h and tcp_var.h for a
    319 				 * description of the scaling constants.
    320 				 */
    321 				rt->rt_rmx.rmx_rtt =
    322 				    (rt->rt_rmx.rmx_rtt + i) / 2;
    323 			else
    324 				rt->rt_rmx.rmx_rtt = i;
    325 		}
    326 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
    327 			i = tp->t_rttvar *
    328 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
    329 			if (rt->rt_rmx.rmx_rttvar && i)
    330 				rt->rt_rmx.rmx_rttvar =
    331 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
    332 			else
    333 				rt->rt_rmx.rmx_rttvar = i;
    334 		}
    335 		/*
    336 		 * update the pipelimit (ssthresh) if it has been updated
    337 		 * already or if a pipesize was specified & the threshhold
    338 		 * got below half the pipesize.  I.e., wait for bad news
    339 		 * before we start updating, then update on both good
    340 		 * and bad news.
    341 		 */
    342 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
    343 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
    344 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
    345 			/*
    346 			 * convert the limit from user data bytes to
    347 			 * packets then to packet data bytes.
    348 			 */
    349 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
    350 			if (i < 2)
    351 				i = 2;
    352 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
    353 			if (rt->rt_rmx.rmx_ssthresh)
    354 				rt->rt_rmx.rmx_ssthresh =
    355 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
    356 			else
    357 				rt->rt_rmx.rmx_ssthresh = i;
    358 		}
    359 	}
    360 #endif /* RTV_RTT */
    361 	/* free the reassembly queue, if any */
    362 	(void) tcp_freeq(tp);
    363 	TCP_CLEAR_DELACK(tp);
    364 
    365 	if (tp->t_template)
    366 		FREE(tp->t_template, M_MBUF);
    367 	free(tp, M_PCB);
    368 	inp->inp_ppcb = 0;
    369 	soisdisconnected(so);
    370 	in_pcbdetach(inp);
    371 	tcpstat.tcps_closed++;
    372 	return ((struct tcpcb *)0);
    373 }
    374 
    375 int
    376 tcp_freeq(tp)
    377 	struct tcpcb *tp;
    378 {
    379 	register struct ipqent *qe;
    380 	int rv = 0;
    381 
    382 	while ((qe = tp->segq.lh_first) != NULL) {
    383 		LIST_REMOVE(qe, ipqe_q);
    384 		m_freem(qe->ipqe_m);
    385 		FREE(qe, M_IPQ);
    386 		rv = 1;
    387 	}
    388 	return (rv);
    389 }
    390 
    391 /*
    392  * Protocol drain routine.  Called when memory is in short supply.
    393  */
    394 void
    395 tcp_drain()
    396 {
    397 	register struct inpcb *inp;
    398 	register struct tcpcb *tp;
    399 
    400 	/*
    401 	 * Free the sequence queue of all TCP connections.
    402 	 */
    403 	inp = tcbtable.inpt_queue.cqh_first;
    404 	if (inp)						/* XXX */
    405 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
    406 	    inp = inp->inp_queue.cqe_next) {
    407 		if ((tp = intotcpcb(inp)) != NULL) {
    408 			if (tcp_freeq(tp))
    409 				tcpstat.tcps_connsdrained++;
    410 		}
    411 	}
    412 }
    413 
    414 /*
    415  * Notify a tcp user of an asynchronous error;
    416  * store error as soft error, but wake up user
    417  * (for now, won't do anything until can select for soft error).
    418  */
    419 void
    420 tcp_notify(inp, error)
    421 	struct inpcb *inp;
    422 	int error;
    423 {
    424 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
    425 	register struct socket *so = inp->inp_socket;
    426 
    427 	/*
    428 	 * Ignore some errors if we are hooked up.
    429 	 * If connection hasn't completed, has retransmitted several times,
    430 	 * and receives a second error, give up now.  This is better
    431 	 * than waiting a long time to establish a connection that
    432 	 * can never complete.
    433 	 */
    434 	if (tp->t_state == TCPS_ESTABLISHED &&
    435 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
    436 	      error == EHOSTDOWN)) {
    437 		return;
    438 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
    439 	    tp->t_rxtshift > 3 && tp->t_softerror)
    440 		so->so_error = error;
    441 	else
    442 		tp->t_softerror = error;
    443 	wakeup((caddr_t) &so->so_timeo);
    444 	sorwakeup(so);
    445 	sowwakeup(so);
    446 }
    447 
    448 void *
    449 tcp_ctlinput(cmd, sa, v)
    450 	int cmd;
    451 	struct sockaddr *sa;
    452 	register void *v;
    453 {
    454 	register struct ip *ip = v;
    455 	register struct tcphdr *th;
    456 	extern int inetctlerrmap[];
    457 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
    458 	int errno;
    459 	int nmatch;
    460 
    461 	if ((unsigned)cmd >= PRC_NCMDS)
    462 		return NULL;
    463 	errno = inetctlerrmap[cmd];
    464 	if (cmd == PRC_QUENCH)
    465 		notify = tcp_quench;
    466 	else if (PRC_IS_REDIRECT(cmd))
    467 		notify = in_rtchange, ip = 0;
    468 	else if (cmd == PRC_MSGSIZE && ip_mtudisc)
    469 		notify = tcp_mtudisc, ip = 0;
    470 	else if (cmd == PRC_HOSTDEAD)
    471 		ip = 0;
    472 	else if (errno == 0)
    473 		return NULL;
    474 	if (ip) {
    475 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
    476 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
    477 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
    478 		if (nmatch == 0 && syn_cache_count &&
    479 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
    480 		    inetctlerrmap[cmd] == ENETUNREACH ||
    481 		    inetctlerrmap[cmd] == EHOSTDOWN))
    482 			syn_cache_unreach(ip, th);
    483 	} else
    484 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
    485 		    notify);
    486 	return NULL;
    487 }
    488 
    489 /*
    490  * When a source quench is received, close congestion window
    491  * to one segment.  We will gradually open it again as we proceed.
    492  */
    493 void
    494 tcp_quench(inp, errno)
    495 	struct inpcb *inp;
    496 	int errno;
    497 {
    498 	struct tcpcb *tp = intotcpcb(inp);
    499 
    500 	if (tp)
    501 		tp->snd_cwnd = tp->t_segsz;
    502 }
    503 
    504 /*
    505  * On receipt of path MTU corrections, flush old route and replace it
    506  * with the new one.  Retransmit all unacknowledged packets, to ensure
    507  * that all packets will be received.
    508  */
    509 void
    510 tcp_mtudisc(inp, errno)
    511 	struct inpcb *inp;
    512 	int errno;
    513 {
    514 	struct tcpcb *tp = intotcpcb(inp);
    515 	struct rtentry *rt = in_pcbrtentry(inp);
    516 
    517 	if (tp != 0) {
    518 		if (rt != 0) {
    519 			/*
    520 			 * If this was not a host route, remove and realloc.
    521 			 */
    522 			if ((rt->rt_flags & RTF_HOST) == 0) {
    523 				in_rtchange(inp, errno);
    524 				if ((rt = in_pcbrtentry(inp)) == 0)
    525 					return;
    526 			}
    527 
    528 			/*
    529 			 * Slow start out of the error condition.  We
    530 			 * use the MTU because we know it's smaller
    531 			 * than the previously transmitted segment.
    532 			 */
    533 			if (rt->rt_rmx.rmx_mtu != 0)
    534 				tp->snd_cwnd =
    535 				    TCP_INITIAL_WINDOW(rt->rt_rmx.rmx_mtu);
    536 		}
    537 
    538 		/*
    539 		 * Resend unacknowledged packets.
    540 		 */
    541 		tp->snd_nxt = tp->snd_una;
    542 		tcp_output(tp);
    543 	}
    544 }
    545 
    546 
    547 /*
    548  * Compute the MSS to advertise to the peer.  Called only during
    549  * the 3-way handshake.  If we are the server (peer initiated
    550  * connection), we are called with the TCPCB for the listen
    551  * socket.  If we are the client (we initiated connection), we
    552  * are called witht he TCPCB for the actual connection.
    553  */
    554 int
    555 tcp_mss_to_advertise(tp)
    556 	const struct tcpcb *tp;
    557 {
    558 	extern u_long in_maxmtu;
    559 	struct inpcb *inp;
    560 	struct socket *so;
    561 	int mss;
    562 
    563 	inp = tp->t_inpcb;
    564 	so = inp->inp_socket;
    565 
    566 	/*
    567 	 * In order to avoid defeating path MTU discovery on the peer,
    568 	 * we advertise the max MTU of all attached networks as our MSS,
    569 	 * per RFC 1191, section 3.1.
    570 	 *
    571 	 * XXX Should we allow room for the timestamp option if
    572 	 * XXX rfc1323 is enabled?
    573 	 */
    574 	mss = in_maxmtu - sizeof(struct tcpiphdr);
    575 
    576 	return (mss);
    577 }
    578 
    579 /*
    580  * Set connection variables based on the peer's advertised MSS.
    581  * We are passed the TCPCB for the actual connection.  If we
    582  * are the server, we are called by the compressed state engine
    583  * when the 3-way handshake is complete.  If we are the client,
    584  * we are called when we recieve the SYN,ACK from the server.
    585  *
    586  * NOTE: Our advertised MSS value must be initialized in the TCPCB
    587  * before this routine is called!
    588  */
    589 void
    590 tcp_mss_from_peer(tp, offer)
    591 	struct tcpcb *tp;
    592 	int offer;
    593 {
    594 	struct inpcb *inp = tp->t_inpcb;
    595 	struct socket *so = inp->inp_socket;
    596 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
    597 	struct rtentry *rt = in_pcbrtentry(inp);
    598 #endif
    599 	u_long bufsize;
    600 	int mss;
    601 
    602 	/*
    603 	 * Assume our MSS is the MSS of the peer, unless they sent us
    604 	 * an offer.  Do not accept offers less than 32 bytes.
    605 	 */
    606 	mss = tp->t_ourmss;
    607 	if (offer)
    608 		mss = offer;
    609 	mss = max(mss, 32);		/* sanity */
    610 
    611 	/*
    612 	 * If there's a pipesize, change the socket buffer to that size.
    613 	 * Make the socket buffer an integral number of MSS units.  If
    614 	 * the MSS is larger than the socket buffer, artificially decrease
    615 	 * the MSS.
    616 	 */
    617 #ifdef RTV_SPIPE
    618 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
    619 		bufsize = rt->rt_rmx.rmx_sendpipe;
    620 	else
    621 #endif
    622 		bufsize = so->so_snd.sb_hiwat;
    623 	if (bufsize < mss)
    624 		mss = bufsize;
    625 	else {
    626 		bufsize = roundup(bufsize, mss);
    627 		if (bufsize > sb_max)
    628 			bufsize = sb_max;
    629 		(void) sbreserve(&so->so_snd, bufsize);
    630 	}
    631 	tp->t_peermss = mss;
    632 	tp->t_segsz = mss;
    633 
    634 	/* Initialize the initial congestion window. */
    635 	tp->snd_cwnd = TCP_INITIAL_WINDOW(mss);
    636 
    637 #ifdef RTV_SSTHRESH
    638 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
    639 		/*
    640 		 * There's some sort of gateway or interface buffer
    641 		 * limit on the path.  Use this to set the slow
    642 		 * start threshold, but set the threshold to no less
    643 		 * than 2 * MSS.
    644 		 */
    645 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
    646 	}
    647 #endif
    648 }
    649 
    650 /*
    651  * Processing necessary when a TCP connection is established.
    652  */
    653 void
    654 tcp_established(tp)
    655 	struct tcpcb *tp;
    656 {
    657 	struct inpcb *inp = tp->t_inpcb;
    658 	struct socket *so = inp->inp_socket;
    659 #ifdef RTV_RPIPE
    660 	struct rtentry *rt = in_pcbrtentry(inp);
    661 #endif
    662 	u_long bufsize;
    663 
    664 	tp->t_state = TCPS_ESTABLISHED;
    665 	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
    666 
    667 #ifdef RTV_RPIPE
    668 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
    669 		bufsize = rt->rt_rmx.rmx_recvpipe;
    670 	else
    671 #endif
    672 		bufsize = so->so_rcv.sb_hiwat;
    673 	if (bufsize > tp->t_ourmss) {
    674 		bufsize = roundup(bufsize, tp->t_ourmss);
    675 		if (bufsize > sb_max)
    676 			bufsize = sb_max;
    677 		(void) sbreserve(&so->so_rcv, bufsize);
    678 	}
    679 }
    680 
    681 /*
    682  * Check if there's an initial rtt or rttvar.  Convert from the
    683  * route-table units to scaled multiples of the slow timeout timer.
    684  * Called only during the 3-way handshake.
    685  */
    686 void
    687 tcp_rmx_rtt(tp)
    688 	struct tcpcb *tp;
    689 {
    690 #ifdef RTV_RTT
    691 	struct rtentry *rt;
    692 	int rtt;
    693 
    694 	if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
    695 		return;
    696 
    697 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
    698 		/*
    699 		 * XXX The lock bit for MTU indicates that the value
    700 		 * is also a minimum value; this is subject to time.
    701 		 */
    702 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
    703 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
    704 		tp->t_srtt = rtt /
    705 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    706 		if (rt->rt_rmx.rmx_rttvar) {
    707 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
    708 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
    709 				(TCP_RTTVAR_SHIFT + 2));
    710 		} else {
    711 			/* Default variation is +- 1 rtt */
    712 			tp->t_rttvar =
    713 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
    714 		}
    715 		TCPT_RANGESET(tp->t_rxtcur,
    716 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
    717 		    tp->t_rttmin, TCPTV_REXMTMAX);
    718 	}
    719 #endif
    720 }
    721 
    722 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
    723 
    724 /*
    725  * Get a new sequence value given a tcp control block
    726  */
    727 tcp_seq
    728 tcp_new_iss(tp, len, addin)
    729 	void            *tp;
    730 	u_long           len;
    731 	tcp_seq		 addin;
    732 {
    733 	tcp_seq          tcp_iss;
    734 
    735 	/*
    736 	 * add randomness about this connection, but do not estimate
    737 	 * entropy from the timing, since the physical device driver would
    738 	 * have done that for us.
    739 	 */
    740 #if NRND > 0
    741 	if (tp != NULL)
    742 		rnd_add_data(NULL, tp, len, 0);
    743 #endif
    744 
    745 	/*
    746 	 * randomize.
    747 	 */
    748 #if NRND > 0
    749 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
    750 #else
    751 	tcp_iss = random();
    752 #endif
    753 
    754 	/*
    755 	 * If we were asked to add some amount to a known value,
    756 	 * we will take a random value obtained above, mask off the upper
    757 	 * bits, and add in the known value.  We also add in a constant to
    758 	 * ensure that we are at least a certain distance from the original
    759 	 * value.
    760 	 *
    761 	 * This is used when an old connection is in timed wait
    762 	 * and we have a new one coming in, for instance.
    763 	 */
    764 	if (addin != 0) {
    765 #ifdef TCPISS_DEBUG
    766 		printf("Random %08x, ", tcp_iss);
    767 #endif
    768 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    769 		tcp_iss = tcp_iss + addin + TCP_ISSINCR;
    770 		tcp_iss_seq += TCP_ISSINCR;
    771 		tcp_iss += tcp_iss_seq;
    772 #ifdef TCPISS_DEBUG
    773 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
    774 #endif
    775 	} else {
    776 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    777 		tcp_iss_seq += TCP_ISSINCR;
    778 		tcp_iss += tcp_iss_seq;
    779 #ifdef TCPISS_DEBUG
    780 		printf("ISS %08x\n", tcp_iss);
    781 #endif
    782 	}
    783 
    784 #ifdef TCP_COMPAT_42
    785 	/*
    786 	 * limit it to the positive range for really old TCP implementations
    787 	 */
    788 	if ((int)tcp_iss < 0)
    789 		tcp_iss &= 0x7fffffff;		/* XXX */
    790 #endif
    791 
    792 	return tcp_iss;
    793 }
    794