Home | History | Annotate | Line # | Download | only in netinet
tcp_subr.c revision 1.29
      1 /*	$NetBSD: tcp_subr.c,v 1.29 1997/10/10 01:51:09 explorer 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 <sys/param.h>
     39 #include <sys/proc.h>
     40 #include <sys/systm.h>
     41 #include <sys/malloc.h>
     42 #include <sys/mbuf.h>
     43 #include <sys/socket.h>
     44 #include <sys/socketvar.h>
     45 #include <sys/protosw.h>
     46 #include <sys/errno.h>
     47 #include <sys/kernel.h>
     48 #include <sys/rnd.h>
     49 
     50 #include <net/route.h>
     51 #include <net/if.h>
     52 
     53 #include <netinet/in.h>
     54 #include <netinet/in_systm.h>
     55 #include <netinet/ip.h>
     56 #include <netinet/in_pcb.h>
     57 #include <netinet/ip_var.h>
     58 #include <netinet/ip_icmp.h>
     59 #include <netinet/tcp.h>
     60 #include <netinet/tcp_fsm.h>
     61 #include <netinet/tcp_seq.h>
     62 #include <netinet/tcp_timer.h>
     63 #include <netinet/tcp_var.h>
     64 #include <netinet/tcpip.h>
     65 
     66 /* patchable/settable parameters for tcp */
     67 int 	tcp_mssdflt = TCP_MSS;
     68 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
     69 int	tcp_do_rfc1323 = 1;
     70 
     71 #ifndef TCBHASHSIZE
     72 #define	TCBHASHSIZE	128
     73 #endif
     74 int	tcbhashsize = TCBHASHSIZE;
     75 
     76 /*
     77  * Tcp initialization
     78  */
     79 void
     80 tcp_init()
     81 {
     82 
     83 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
     84 	if (max_protohdr < sizeof(struct tcpiphdr))
     85 		max_protohdr = sizeof(struct tcpiphdr);
     86 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
     87 		panic("tcp_init");
     88 }
     89 
     90 /*
     91  * Create template to be used to send tcp packets on a connection.
     92  * Call after host entry created, allocates an mbuf and fills
     93  * in a skeletal tcp/ip header, minimizing the amount of work
     94  * necessary when the connection is used.
     95  */
     96 struct tcpiphdr *
     97 tcp_template(tp)
     98 	struct tcpcb *tp;
     99 {
    100 	register struct inpcb *inp = tp->t_inpcb;
    101 	register struct tcpiphdr *n;
    102 
    103 	if ((n = tp->t_template) == 0) {
    104 		MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
    105 		    M_MBUF, M_NOWAIT);
    106 		if (n == NULL)
    107 			return (0);
    108 	}
    109 	bzero(n->ti_x1, sizeof n->ti_x1);
    110 	n->ti_pr = IPPROTO_TCP;
    111 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
    112 	n->ti_src = inp->inp_laddr;
    113 	n->ti_dst = inp->inp_faddr;
    114 	n->ti_sport = inp->inp_lport;
    115 	n->ti_dport = inp->inp_fport;
    116 	n->ti_seq = 0;
    117 	n->ti_ack = 0;
    118 	n->ti_x2 = 0;
    119 	n->ti_off = 5;
    120 	n->ti_flags = 0;
    121 	n->ti_win = 0;
    122 	n->ti_sum = 0;
    123 	n->ti_urp = 0;
    124 	return (n);
    125 }
    126 
    127 /*
    128  * Send a single message to the TCP at address specified by
    129  * the given TCP/IP header.  If m == 0, then we make a copy
    130  * of the tcpiphdr at ti and send directly to the addressed host.
    131  * This is used to force keep alive messages out using the TCP
    132  * template for a connection tp->t_template.  If flags are given
    133  * then we send a message back to the TCP which originated the
    134  * segment ti, and discard the mbuf containing it and any other
    135  * attached mbufs.
    136  *
    137  * In any case the ack and sequence number of the transmitted
    138  * segment are as specified by the parameters.
    139  */
    140 int
    141 tcp_respond(tp, ti, m, ack, seq, flags)
    142 	struct tcpcb *tp;
    143 	register struct tcpiphdr *ti;
    144 	register struct mbuf *m;
    145 	tcp_seq ack, seq;
    146 	int flags;
    147 {
    148 	register int tlen;
    149 	int win = 0;
    150 	struct route *ro = 0;
    151 
    152 	if (tp) {
    153 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
    154 		ro = &tp->t_inpcb->inp_route;
    155 	}
    156 	if (m == 0) {
    157 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
    158 		if (m == NULL)
    159 			return (ENOBUFS);
    160 #ifdef TCP_COMPAT_42
    161 		tlen = 1;
    162 #else
    163 		tlen = 0;
    164 #endif
    165 		m->m_data += max_linkhdr;
    166 		*mtod(m, struct tcpiphdr *) = *ti;
    167 		ti = mtod(m, struct tcpiphdr *);
    168 		flags = TH_ACK;
    169 	} else {
    170 		m_freem(m->m_next);
    171 		m->m_next = 0;
    172 		m->m_data = (caddr_t)ti;
    173 		m->m_len = sizeof (struct tcpiphdr);
    174 		tlen = 0;
    175 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
    176 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
    177 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
    178 #undef xchg
    179 	}
    180 	bzero(ti->ti_x1, sizeof ti->ti_x1);
    181 	ti->ti_seq = htonl(seq);
    182 	ti->ti_ack = htonl(ack);
    183 	ti->ti_x2 = 0;
    184 	if ((flags & TH_SYN) == 0) {
    185 		if (tp)
    186 			ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
    187 		else
    188 			ti->ti_win = htons((u_int16_t)win);
    189 		ti->ti_off = sizeof (struct tcphdr) >> 2;
    190 		tlen += sizeof (struct tcphdr);
    191 	} else
    192 		tlen += ti->ti_off << 2;
    193 	ti->ti_len = htons((u_int16_t)tlen);
    194 	tlen += sizeof (struct ip);
    195 	m->m_len = tlen;
    196 	m->m_pkthdr.len = tlen;
    197 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
    198 	ti->ti_flags = flags;
    199 	ti->ti_urp = 0;
    200 	ti->ti_sum = 0;
    201 	ti->ti_sum = in_cksum(m, tlen);
    202 	((struct ip *)ti)->ip_len = tlen;
    203 	((struct ip *)ti)->ip_ttl = ip_defttl;
    204 	return ip_output(m, NULL, ro, 0, NULL);
    205 }
    206 
    207 /*
    208  * Create a new TCP control block, making an
    209  * empty reassembly queue and hooking it to the argument
    210  * protocol control block.
    211  */
    212 struct tcpcb *
    213 tcp_newtcpcb(inp)
    214 	struct inpcb *inp;
    215 {
    216 	register struct tcpcb *tp;
    217 
    218 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
    219 	if (tp == NULL)
    220 		return ((struct tcpcb *)0);
    221 	bzero((caddr_t)tp, sizeof(struct tcpcb));
    222 	LIST_INIT(&tp->segq);
    223 	tp->t_maxseg = tcp_mssdflt;
    224 	tp->t_ourmss = tcp_mssdflt;
    225 
    226 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
    227 	tp->t_inpcb = inp;
    228 	/*
    229 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
    230 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
    231 	 * reasonable initial retransmit time.
    232 	 */
    233 	tp->t_srtt = TCPTV_SRTTBASE;
    234 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
    235 	tp->t_rttmin = TCPTV_MIN;
    236 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
    237 	    TCPTV_MIN, TCPTV_REXMTMAX);
    238 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    239 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
    240 	inp->inp_ip.ip_ttl = ip_defttl;
    241 	inp->inp_ppcb = (caddr_t)tp;
    242 	return (tp);
    243 }
    244 
    245 /*
    246  * Drop a TCP connection, reporting
    247  * the specified error.  If connection is synchronized,
    248  * then send a RST to peer.
    249  */
    250 struct tcpcb *
    251 tcp_drop(tp, errno)
    252 	register struct tcpcb *tp;
    253 	int errno;
    254 {
    255 	struct socket *so = tp->t_inpcb->inp_socket;
    256 
    257 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
    258 		tp->t_state = TCPS_CLOSED;
    259 		(void) tcp_output(tp);
    260 		tcpstat.tcps_drops++;
    261 	} else
    262 		tcpstat.tcps_conndrops++;
    263 	if (errno == ETIMEDOUT && tp->t_softerror)
    264 		errno = tp->t_softerror;
    265 	so->so_error = errno;
    266 	return (tcp_close(tp));
    267 }
    268 
    269 /*
    270  * Close a TCP control block:
    271  *	discard all space held by the tcp
    272  *	discard internet protocol block
    273  *	wake up any sleepers
    274  */
    275 struct tcpcb *
    276 tcp_close(tp)
    277 	register struct tcpcb *tp;
    278 {
    279 	register struct ipqent *qe;
    280 	struct inpcb *inp = tp->t_inpcb;
    281 	struct socket *so = inp->inp_socket;
    282 #ifdef RTV_RTT
    283 	register struct rtentry *rt;
    284 
    285 	/*
    286 	 * If we sent enough data to get some meaningful characteristics,
    287 	 * save them in the routing entry.  'Enough' is arbitrarily
    288 	 * defined as the sendpipesize (default 4K) * 16.  This would
    289 	 * give us 16 rtt samples assuming we only get one sample per
    290 	 * window (the usual case on a long haul net).  16 samples is
    291 	 * enough for the srtt filter to converge to within 5% of the correct
    292 	 * value; fewer samples and we could save a very bogus rtt.
    293 	 *
    294 	 * Don't update the default route's characteristics and don't
    295 	 * update anything that the user "locked".
    296 	 */
    297 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
    298 	    (rt = inp->inp_route.ro_rt) &&
    299 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
    300 		register u_long i = 0;
    301 
    302 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
    303 			i = tp->t_srtt *
    304 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    305 			if (rt->rt_rmx.rmx_rtt && i)
    306 				/*
    307 				 * filter this update to half the old & half
    308 				 * the new values, converting scale.
    309 				 * See route.h and tcp_var.h for a
    310 				 * description of the scaling constants.
    311 				 */
    312 				rt->rt_rmx.rmx_rtt =
    313 				    (rt->rt_rmx.rmx_rtt + i) / 2;
    314 			else
    315 				rt->rt_rmx.rmx_rtt = i;
    316 		}
    317 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
    318 			i = tp->t_rttvar *
    319 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
    320 			if (rt->rt_rmx.rmx_rttvar && i)
    321 				rt->rt_rmx.rmx_rttvar =
    322 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
    323 			else
    324 				rt->rt_rmx.rmx_rttvar = i;
    325 		}
    326 		/*
    327 		 * update the pipelimit (ssthresh) if it has been updated
    328 		 * already or if a pipesize was specified & the threshhold
    329 		 * got below half the pipesize.  I.e., wait for bad news
    330 		 * before we start updating, then update on both good
    331 		 * and bad news.
    332 		 */
    333 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
    334 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
    335 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
    336 			/*
    337 			 * convert the limit from user data bytes to
    338 			 * packets then to packet data bytes.
    339 			 */
    340 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
    341 			if (i < 2)
    342 				i = 2;
    343 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
    344 			if (rt->rt_rmx.rmx_ssthresh)
    345 				rt->rt_rmx.rmx_ssthresh =
    346 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
    347 			else
    348 				rt->rt_rmx.rmx_ssthresh = i;
    349 		}
    350 	}
    351 #endif /* RTV_RTT */
    352 	/* free the reassembly queue, if any */
    353 	while ((qe = tp->segq.lh_first) != NULL) {
    354 		LIST_REMOVE(qe, ipqe_q);
    355 		m_freem(qe->ipqe_m);
    356 		FREE(qe, M_IPQ);
    357 	}
    358 	if (tp->t_template)
    359 		FREE(tp->t_template, M_MBUF);
    360 	free(tp, M_PCB);
    361 	inp->inp_ppcb = 0;
    362 	soisdisconnected(so);
    363 	in_pcbdetach(inp);
    364 	tcpstat.tcps_closed++;
    365 	return ((struct tcpcb *)0);
    366 }
    367 
    368 void
    369 tcp_drain()
    370 {
    371 
    372 }
    373 
    374 /*
    375  * Notify a tcp user of an asynchronous error;
    376  * store error as soft error, but wake up user
    377  * (for now, won't do anything until can select for soft error).
    378  */
    379 void
    380 tcp_notify(inp, error)
    381 	struct inpcb *inp;
    382 	int error;
    383 {
    384 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
    385 	register struct socket *so = inp->inp_socket;
    386 
    387 	/*
    388 	 * Ignore some errors if we are hooked up.
    389 	 * If connection hasn't completed, has retransmitted several times,
    390 	 * and receives a second error, give up now.  This is better
    391 	 * than waiting a long time to establish a connection that
    392 	 * can never complete.
    393 	 */
    394 	if (tp->t_state == TCPS_ESTABLISHED &&
    395 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
    396 	      error == EHOSTDOWN)) {
    397 		return;
    398 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
    399 	    tp->t_rxtshift > 3 && tp->t_softerror)
    400 		so->so_error = error;
    401 	else
    402 		tp->t_softerror = error;
    403 	wakeup((caddr_t) &so->so_timeo);
    404 	sorwakeup(so);
    405 	sowwakeup(so);
    406 }
    407 
    408 void *
    409 tcp_ctlinput(cmd, sa, v)
    410 	int cmd;
    411 	struct sockaddr *sa;
    412 	register void *v;
    413 {
    414 	register struct ip *ip = v;
    415 	register struct tcphdr *th;
    416 	extern int inetctlerrmap[];
    417 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
    418 	int errno;
    419 	int nmatch;
    420 
    421 	if ((unsigned)cmd >= PRC_NCMDS)
    422 		return NULL;
    423 	errno = inetctlerrmap[cmd];
    424 	if (cmd == PRC_QUENCH)
    425 		notify = tcp_quench;
    426 	else if (PRC_IS_REDIRECT(cmd))
    427 		notify = in_rtchange, ip = 0;
    428 	else if (cmd == PRC_HOSTDEAD)
    429 		ip = 0;
    430 	else if (errno == 0)
    431 		return NULL;
    432 	if (ip) {
    433 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
    434 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
    435 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
    436 		if (nmatch == 0 && syn_cache_count &&
    437 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
    438 		    inetctlerrmap[cmd] == ENETUNREACH ||
    439 		    inetctlerrmap[cmd] == EHOSTDOWN))
    440 			syn_cache_unreach(ip, th);
    441 	} else
    442 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
    443 		    notify);
    444 	return NULL;
    445 }
    446 
    447 /*
    448  * When a source quench is received, close congestion window
    449  * to one segment.  We will gradually open it again as we proceed.
    450  */
    451 void
    452 tcp_quench(inp, errno)
    453 	struct inpcb *inp;
    454 	int errno;
    455 {
    456 	struct tcpcb *tp = intotcpcb(inp);
    457 
    458 	if (tp)
    459 		tp->snd_cwnd = tp->t_maxseg;
    460 }
    461 
    462 /*
    463  * Compute the MSS to advertise to the peer.  Called only during
    464  * the 3-way handshake.  If we are the server (peer initiated
    465  * connection), we are called with the TCPCB for the listen
    466  * socket.  If we are the client (we initiated connection), we
    467  * are called witht he TCPCB for the actual connection.
    468  */
    469 int
    470 tcp_mss_to_advertise(tp)
    471 	const struct tcpcb *tp;
    472 {
    473 	extern u_long in_maxmtu;
    474 	struct inpcb *inp;
    475 	struct socket *so;
    476 	int mss;
    477 
    478 	inp = tp->t_inpcb;
    479 	so = inp->inp_socket;
    480 
    481 	/*
    482 	 * In order to avoid defeating path MTU discovery on the peer,
    483 	 * we advertise the max MTU of all attached networks as our MSS,
    484 	 * per RFC 1191, section 3.1.
    485 	 *
    486 	 * XXX Should we allow room for the timestamp option if
    487 	 * XXX rfc1323 is enabled?
    488 	 */
    489 	mss = in_maxmtu - sizeof(struct tcpiphdr);
    490 
    491 	return (mss);
    492 }
    493 
    494 /*
    495  * Set connection variables based on the peer's advertised MSS.
    496  * We are passed the TCPCB for the actual connection.  If we
    497  * are the server, we are called by the compressed state engine
    498  * when the 3-way handshake is complete.  If we are the client,
    499  * we are called when we recieve the SYN,ACK from the server.
    500  *
    501  * NOTE: Our advertised MSS value must be initialized in the TCPCB
    502  * before this routine is called!
    503  */
    504 void
    505 tcp_mss_from_peer(tp, offer)
    506 	struct tcpcb *tp;
    507 	int offer;
    508 {
    509 	struct inpcb *inp = tp->t_inpcb;
    510 	struct socket *so = inp->inp_socket;
    511 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
    512 	struct rtentry *rt = in_pcbrtentry(inp);
    513 #endif
    514 	u_long bufsize;
    515 	int mss;
    516 
    517 	/*
    518 	 * Assume our MSS is the MSS of the peer, unless they sent us
    519 	 * an offer.  Do not accept offers less than 32 bytes.
    520 	 */
    521 	mss = tp->t_ourmss;
    522 	if (offer)
    523 		mss = offer;
    524 	mss = max(mss, 32);		/* sanity */
    525 
    526 	/*
    527 	 * If there's a pipesize, change the socket buffer to that size.
    528 	 * Make the socket buffer an integral number of MSS units.  If
    529 	 * the MSS is larger than the socket buffer, artificially decrease
    530 	 * the MSS.
    531 	 */
    532 #ifdef RTV_SPIPE
    533 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
    534 		bufsize = rt->rt_rmx.rmx_sendpipe;
    535 	else
    536 #endif
    537 		bufsize = so->so_snd.sb_hiwat;
    538 	if (bufsize < mss)
    539 		mss = bufsize;
    540 	else {
    541 		bufsize = roundup(bufsize, mss);
    542 		if (bufsize > sb_max)
    543 			bufsize = sb_max;
    544 		(void) sbreserve(&so->so_snd, bufsize);
    545 	}
    546 	tp->t_maxseg = mss;
    547 
    548 	/* Initialize the initial congestion window. */
    549 	tp->snd_cwnd = mss;
    550 
    551 #ifdef RTV_SSTHRESH
    552 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
    553 		/*
    554 		 * There's some sort of gateway or interface buffer
    555 		 * limit on the path.  Use this to set the slow
    556 		 * start threshold, but set the threshold to no less
    557 		 * than 2 * MSS.
    558 		 */
    559 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
    560 	}
    561 #endif
    562 }
    563 
    564 /*
    565  * Processing necessary when a TCP connection is established.
    566  */
    567 void
    568 tcp_established(tp)
    569 	struct tcpcb *tp;
    570 {
    571 	struct inpcb *inp = tp->t_inpcb;
    572 	struct socket *so = inp->inp_socket;
    573 #ifdef RTV_RPIPE
    574 	struct rtentry *rt = in_pcbrtentry(inp);
    575 #endif
    576 	u_long bufsize;
    577 
    578 	tp->t_state = TCPS_ESTABLISHED;
    579 	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
    580 
    581 #ifdef RTV_RPIPE
    582 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
    583 		bufsize = rt->rt_rmx.rmx_recvpipe;
    584 	else
    585 #endif
    586 		bufsize = so->so_rcv.sb_hiwat;
    587 	if (bufsize > tp->t_ourmss) {
    588 		bufsize = roundup(bufsize, tp->t_ourmss);
    589 		if (bufsize > sb_max)
    590 			bufsize = sb_max;
    591 		(void) sbreserve(&so->so_rcv, bufsize);
    592 	}
    593 }
    594 
    595 /*
    596  * Check if there's an initial rtt or rttvar.  Convert from the
    597  * route-table units to scaled multiples of the slow timeout timer.
    598  * Called only during the 3-way handshake.
    599  */
    600 void
    601 tcp_rmx_rtt(tp)
    602 	struct tcpcb *tp;
    603 {
    604 #ifdef RTV_RTT
    605 	struct rtentry *rt;
    606 	int rtt;
    607 
    608 	if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
    609 		return;
    610 
    611 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
    612 		/*
    613 		 * XXX The lock bit for MTU indicates that the value
    614 		 * is also a minimum value; this is subject to time.
    615 		 */
    616 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
    617 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
    618 		tp->t_srtt = rtt /
    619 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
    620 		if (rt->rt_rmx.rmx_rttvar) {
    621 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
    622 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
    623 				(TCP_RTTVAR_SHIFT + 2));
    624 		} else {
    625 			/* Default variation is +- 1 rtt */
    626 			tp->t_rttvar =
    627 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
    628 		}
    629 		TCPT_RANGESET(tp->t_rxtcur,
    630 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
    631 		    tp->t_rttmin, TCPTV_REXMTMAX);
    632 	}
    633 #endif
    634 }
    635 
    636 /*
    637  * Get a new sequence value given a tcp control block
    638  */
    639 tcp_seq
    640 tcp_new_iss(tp, len, addin)
    641 	void            *tp;
    642 	u_long           len;
    643 	tcp_seq		 addin;
    644 {
    645 	tcp_seq          tcp_iss;
    646 	static tcp_seq	 tcp_iss_seq = 0;
    647 
    648 #define TCP_ISS_RANDOM_MASK 0x003fffff
    649 #define TCP_ISS_INCR        0x00400000
    650 
    651 	/*
    652 	 * add randomness about this connection, but do not estimate
    653 	 * entropy from the timing, since the physical device driver would
    654 	 * have done that for us.
    655 	 */
    656 	if (tp != NULL)
    657 		rnd_add_data(NULL, tp, len, 0);
    658 
    659 	/*
    660 	 * randomize.
    661 	 */
    662 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
    663 
    664 	/*
    665 	 * If we were asked to add some amount to a known value,
    666 	 * we will take a random value obtained above, mask off the upper
    667 	 * bits, and add in the known value.  We also add in a constant to
    668 	 * ensure that we are at least a certain distance from the original
    669 	 * value.
    670 	 *
    671 	 * This is used when an old connection is in timed wait
    672 	 * and we have a new one coming in, for instance.
    673 	 */
    674 	if (addin != 0) {
    675 #ifdef TCPISS_DEBUG
    676 		printf("Random %08x, ", tcp_iss);
    677 #endif
    678 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    679 		tcp_iss = tcp_iss + addin + TCP_ISS_INCR;
    680 #ifdef TCPISS_DEBUG
    681 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
    682 #endif
    683 	} else {
    684 		tcp_iss &= TCP_ISS_RANDOM_MASK;
    685 		tcp_iss_seq += TCP_ISS_INCR;
    686 		tcp_iss += tcp_iss_seq;
    687 #ifdef TCPISS_DEBUG
    688 		printf("ISS %08x\n", tcp_iss);
    689 #endif
    690 	}
    691 
    692 #ifdef TCP_COMPAT_42
    693 	/*
    694 	 * limit it to the positive range for really old TCP implementations
    695 	 */
    696 	if ((int)tcp_iss < 0)
    697 		tcp_iss &= 0x7fffffff;		/* XXX */
    698 #endif
    699 
    700 	return tcp_iss;
    701 }
    702