xref: /netbsd-src/sys/netinet/tcp_output.c (revision 7c7c171d130af9949261bc7dce2150a03c3d239c)
1 /*	$NetBSD: tcp_output.c,v 1.34 1998/04/13 21:18:19 kml Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
9  * Facility, NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
42  *	The Regents of the University of California.  All rights reserved.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. All advertising materials mentioning features or use of this software
53  *    must display the following acknowledgement:
54  *	This product includes software developed by the University of
55  *	California, Berkeley and its contributors.
56  * 4. Neither the name of the University nor the names of its contributors
57  *    may be used to endorse or promote products derived from this software
58  *    without specific prior written permission.
59  *
60  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
61  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
62  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
63  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
64  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
65  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
66  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
67  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
68  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
69  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
70  * SUCH DAMAGE.
71  *
72  *	@(#)tcp_output.c	8.4 (Berkeley) 5/24/95
73  */
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/malloc.h>
78 #include <sys/mbuf.h>
79 #include <sys/protosw.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/errno.h>
83 
84 #include <net/if.h>
85 #include <net/route.h>
86 
87 #include <netinet/in.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/ip.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/ip_var.h>
92 #include <netinet/tcp.h>
93 #define	TCPOUTFLAGS
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcpip.h>
99 #include <netinet/tcp_debug.h>
100 
101 #ifdef TUBA
102 #include <netiso/iso.h>
103 #include <netiso/tuba_table.h>
104 #endif
105 
106 #ifdef notyet
107 extern struct mbuf *m_copypack();
108 #endif
109 
110 #define MAX_TCPOPTLEN	32	/* max # bytes that go in options */
111 
112 /*
113  * Knob to enable Congestion Window Monitoring.
114  */
115 #ifdef TCP_CWM
116 int	tcp_cwm = 1;
117 #else
118 int	tcp_cwm = 0;
119 #endif
120 
121 static __inline void tcp_segsize __P((struct tcpcb *, int *, int *));
122 static __inline void
123 tcp_segsize(tp, txsegsizep, rxsegsizep)
124 	struct tcpcb *tp;
125 	int *txsegsizep, *rxsegsizep;
126 {
127 	struct inpcb *inp = tp->t_inpcb;
128 	struct rtentry *rt;
129 	struct ifnet *ifp;
130 	int size;
131 
132 	if ((rt = in_pcbrtentry(inp)) == NULL) {
133 		size = tcp_mssdflt;
134 		goto out;
135 	}
136 
137 	ifp = rt->rt_ifp;
138 
139 	if (rt->rt_rmx.rmx_mtu != 0)
140 		size = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
141 	else if (ip_mtudisc || in_localaddr(inp->inp_faddr) ||
142 		 ifp->if_flags & IFF_LOOPBACK)
143 		size = ifp->if_mtu - sizeof(struct tcpiphdr);
144 	else
145 		size = tcp_mssdflt;
146 	size -= (tcp_optlen(tp) + ip_optlen(tp->t_inpcb));
147 
148  out:
149 	*txsegsizep = min(tp->t_peermss, size);
150 	*rxsegsizep = min(tp->t_ourmss, size);
151 
152 	if (*txsegsizep != tp->t_segsz) {
153 	        /*
154 		 * XXX:  should we check to make sure these are all
155 		 *       at least txsegsize in length, now?
156 		 */
157 		tp->snd_cwnd = (tp->snd_cwnd / tp->t_segsz) * *txsegsizep;
158 		tp->snd_ssthresh = (tp->snd_ssthresh / tp->t_segsz) *
159 		    *txsegsizep;
160 		tp->t_segsz = *txsegsizep;
161 	}
162 }
163 
164 /*
165  * Tcp output routine: figure out what should be sent and send it.
166  */
167 int
168 tcp_output(tp)
169 	register struct tcpcb *tp;
170 {
171 	register struct socket *so = tp->t_inpcb->inp_socket;
172 	register long len, win;
173 	int off, flags, error;
174 	register struct mbuf *m;
175 	register struct tcpiphdr *ti;
176 	u_char opt[MAX_TCPOPTLEN];
177 	unsigned optlen, hdrlen;
178 	int idle, sendalot, txsegsize, rxsegsize;
179 
180 	tcp_segsize(tp, &txsegsize, &rxsegsize);
181 
182 	idle = (tp->snd_max == tp->snd_una);
183 
184 	if (tcp_cwm) {
185 		/*
186 		 * Hughes/Touch/Heidemann Congestion Window Monitoring.
187 		 * Count the number of packets currently pending
188 		 * acknowledgement, and limit our congestion window
189 		 * to the initial congestion window plus that count.
190 		 * This prevents bursting once all pending packets have
191 		 * been acknowledged (i.e. transmission is idle).
192 		 */
193 		tp->snd_cwnd = min(tp->snd_cwnd,
194 		    TCP_INITIAL_WINDOW(tcp_init_win, txsegsize) +
195 		    (tp->snd_nxt - tp->snd_una));
196 	} else {
197 		if (idle && tp->t_idle >= tp->t_rxtcur) {
198 			/*
199 			 * We have been idle for "a while" and no acks are
200 			 * expected to clock out any data we send --
201 			 * slow start to get ack "clock" running again.
202 			 */
203 			tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win,
204 			    txsegsize);
205 		}
206 	}
207 
208 again:
209 	/*
210 	 * Determine length of data that should be transmitted, and
211 	 * flags that should be used.  If there is some data or critical
212 	 * controls (SYN, RST) to send, then transmit; otherwise,
213 	 * investigate further.
214 	 */
215 	sendalot = 0;
216 	off = tp->snd_nxt - tp->snd_una;
217 	win = min(tp->snd_wnd, tp->snd_cwnd);
218 
219 	flags = tcp_outflags[tp->t_state];
220 	/*
221 	 * If in persist timeout with window of 0, send 1 byte.
222 	 * Otherwise, if window is small but nonzero
223 	 * and timer expired, we will send what we can
224 	 * and go to transmit state.
225 	 */
226 	if (tp->t_force) {
227 		if (win == 0) {
228 			/*
229 			 * If we still have some data to send, then
230 			 * clear the FIN bit.  Usually this would
231 			 * happen below when it realizes that we
232 			 * aren't sending all the data.  However,
233 			 * if we have exactly 1 byte of unset data,
234 			 * then it won't clear the FIN bit below,
235 			 * and if we are in persist state, we wind
236 			 * up sending the packet without recording
237 			 * that we sent the FIN bit.
238 			 *
239 			 * We can't just blindly clear the FIN bit,
240 			 * because if we don't have any more data
241 			 * to send then the probe will be the FIN
242 			 * itself.
243 			 */
244 			if (off < so->so_snd.sb_cc)
245 				flags &= ~TH_FIN;
246 			win = 1;
247 		} else {
248 			tp->t_timer[TCPT_PERSIST] = 0;
249 			tp->t_rxtshift = 0;
250 		}
251 	}
252 
253 	if (win < so->so_snd.sb_cc) {
254 		len = win - off;
255 		flags &= ~TH_FIN;
256 	} else
257 		len = so->so_snd.sb_cc - off;
258 
259 	if (len < 0) {
260 		/*
261 		 * If FIN has been sent but not acked,
262 		 * but we haven't been called to retransmit,
263 		 * len will be -1.  Otherwise, window shrank
264 		 * after we sent into it.  If window shrank to 0,
265 		 * cancel pending retransmit, pull snd_nxt back
266 		 * to (closed) window, and set the persist timer
267 		 * if it isn't already going.  If the window didn't
268 		 * close completely, just wait for an ACK.
269 		 */
270 		len = 0;
271 		if (win == 0) {
272 			tp->t_timer[TCPT_REXMT] = 0;
273 			tp->t_rxtshift = 0;
274 			tp->snd_nxt = tp->snd_una;
275 			if (tp->t_timer[TCPT_PERSIST] == 0)
276 				tcp_setpersist(tp);
277 		}
278 	}
279 	if (len > txsegsize) {
280 		len = txsegsize;
281 		flags &= ~TH_FIN;
282 		sendalot = 1;
283 	}
284 
285 	win = sbspace(&so->so_rcv);
286 
287 	/*
288 	 * Sender silly window avoidance.  If connection is idle
289 	 * and can send all data, a maximum segment,
290 	 * at least a maximum default-size segment do it,
291 	 * or are forced, do it; otherwise don't bother.
292 	 * If peer's buffer is tiny, then send
293 	 * when window is at least half open.
294 	 * If retransmitting (possibly after persist timer forced us
295 	 * to send into a small window), then must resend.
296 	 */
297 	if (len) {
298 		if (len == txsegsize)
299 			goto send;
300 		if ((idle || tp->t_flags & TF_NODELAY) &&
301 		    len + off >= so->so_snd.sb_cc)
302 			goto send;
303 		if (tp->t_force)
304 			goto send;
305 		if (len >= tp->max_sndwnd / 2)
306 			goto send;
307 		if (SEQ_LT(tp->snd_nxt, tp->snd_max))
308 			goto send;
309 	}
310 
311 	/*
312 	 * Compare available window to amount of window known to peer
313 	 * (as advertised window less next expected input).  If the
314 	 * difference is at least twice the size of the largest segment
315 	 * we expect to receive (i.e. two segments) or at least 50% of
316 	 * the maximum possible window, then want to send a window update
317 	 * to peer.
318 	 */
319 	if (win > 0) {
320 		/*
321 		 * "adv" is the amount we can increase the window,
322 		 * taking into account that we are limited by
323 		 * TCP_MAXWIN << tp->rcv_scale.
324 		 */
325 		long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
326 			(tp->rcv_adv - tp->rcv_nxt);
327 
328 		if (adv >= (long) (2 * rxsegsize))
329 			goto send;
330 		if (2 * adv >= (long) so->so_rcv.sb_hiwat)
331 			goto send;
332 	}
333 
334 	/*
335 	 * Send if we owe peer an ACK.
336 	 */
337 	if (tp->t_flags & TF_ACKNOW)
338 		goto send;
339 	if (flags & (TH_SYN|TH_RST))
340 		goto send;
341 	if (SEQ_GT(tp->snd_up, tp->snd_una))
342 		goto send;
343 	/*
344 	 * If our state indicates that FIN should be sent
345 	 * and we have not yet done so, or we're retransmitting the FIN,
346 	 * then we need to send.
347 	 */
348 	if (flags & TH_FIN &&
349 	    ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
350 		goto send;
351 
352 	/*
353 	 * TCP window updates are not reliable, rather a polling protocol
354 	 * using ``persist'' packets is used to insure receipt of window
355 	 * updates.  The three ``states'' for the output side are:
356 	 *	idle			not doing retransmits or persists
357 	 *	persisting		to move a small or zero window
358 	 *	(re)transmitting	and thereby not persisting
359 	 *
360 	 * tp->t_timer[TCPT_PERSIST]
361 	 *	is set when we are in persist state.
362 	 * tp->t_force
363 	 *	is set when we are called to send a persist packet.
364 	 * tp->t_timer[TCPT_REXMT]
365 	 *	is set when we are retransmitting
366 	 * The output side is idle when both timers are zero.
367 	 *
368 	 * If send window is too small, there is data to transmit, and no
369 	 * retransmit or persist is pending, then go to persist state.
370 	 * If nothing happens soon, send when timer expires:
371 	 * if window is nonzero, transmit what we can,
372 	 * otherwise force out a byte.
373 	 */
374 	if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
375 	    tp->t_timer[TCPT_PERSIST] == 0) {
376 		tp->t_rxtshift = 0;
377 		tcp_setpersist(tp);
378 	}
379 
380 	/*
381 	 * No reason to send a segment, just return.
382 	 */
383 	return (0);
384 
385 send:
386 	/*
387 	 * Before ESTABLISHED, force sending of initial options
388 	 * unless TCP set not to do any options.
389 	 * NOTE: we assume that the IP/TCP header plus TCP options
390 	 * always fit in a single mbuf, leaving room for a maximum
391 	 * link header, i.e.
392 	 *	max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
393 	 */
394 	optlen = 0;
395 	hdrlen = sizeof (struct tcpiphdr);
396 	if (flags & TH_SYN) {
397 		struct rtentry *rt = in_pcbrtentry(tp->t_inpcb);
398 
399 		tp->snd_nxt = tp->iss;
400 		tp->t_ourmss = tcp_mss_to_advertise(rt != NULL ?
401 						    rt->rt_ifp : NULL);
402 		if ((tp->t_flags & TF_NOOPT) == 0) {
403 			opt[0] = TCPOPT_MAXSEG;
404 			opt[1] = 4;
405 			opt[2] = (tp->t_ourmss >> 8) & 0xff;
406 			opt[3] = tp->t_ourmss & 0xff;
407 			optlen = 4;
408 
409 			if ((tp->t_flags & TF_REQ_SCALE) &&
410 			    ((flags & TH_ACK) == 0 ||
411 			    (tp->t_flags & TF_RCVD_SCALE))) {
412 				*((u_int32_t *) (opt + optlen)) = htonl(
413 					TCPOPT_NOP << 24 |
414 					TCPOPT_WINDOW << 16 |
415 					TCPOLEN_WINDOW << 8 |
416 					tp->request_r_scale);
417 				optlen += 4;
418 			}
419 		}
420  	}
421 
422  	/*
423 	 * Send a timestamp and echo-reply if this is a SYN and our side
424 	 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
425 	 * and our peer have sent timestamps in our SYN's.
426  	 */
427  	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
428  	     (flags & TH_RST) == 0 &&
429  	    ((flags & (TH_SYN|TH_ACK)) == TH_SYN ||
430 	     (tp->t_flags & TF_RCVD_TSTMP))) {
431 		u_int32_t *lp = (u_int32_t *)(opt + optlen);
432 
433  		/* Form timestamp option as shown in appendix A of RFC 1323. */
434  		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
435  		*lp++ = htonl(tcp_now);
436  		*lp   = htonl(tp->ts_recent);
437  		optlen += TCPOLEN_TSTAMP_APPA;
438  	}
439 
440  	hdrlen += optlen;
441 
442 #ifdef DIAGNOSTIC
443 	if (len > txsegsize)
444 		panic("tcp data to be sent is larger than segment");
445  	if (max_linkhdr + hdrlen > MHLEN)
446 		panic("tcphdr too big");
447 #endif
448 
449 	/*
450 	 * Grab a header mbuf, attaching a copy of data to
451 	 * be transmitted, and initialize the header from
452 	 * the template for sends on this connection.
453 	 */
454 	if (len) {
455 		if (tp->t_force && len == 1)
456 			tcpstat.tcps_sndprobe++;
457 		else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
458 			tcpstat.tcps_sndrexmitpack++;
459 			tcpstat.tcps_sndrexmitbyte += len;
460 		} else {
461 			tcpstat.tcps_sndpack++;
462 			tcpstat.tcps_sndbyte += len;
463 		}
464 #ifdef notyet
465 		if ((m = m_copypack(so->so_snd.sb_mb, off,
466 		    (int)len, max_linkhdr + hdrlen)) == 0) {
467 			error = ENOBUFS;
468 			goto out;
469 		}
470 		/*
471 		 * m_copypack left space for our hdr; use it.
472 		 */
473 		m->m_len += hdrlen;
474 		m->m_data -= hdrlen;
475 #else
476 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
477 		if (m == NULL) {
478 			error = ENOBUFS;
479 			goto out;
480 		}
481 		m->m_data += max_linkhdr;
482 		m->m_len = hdrlen;
483 		if (len <= MHLEN - hdrlen - max_linkhdr) {
484 			m_copydata(so->so_snd.sb_mb, off, (int) len,
485 			    mtod(m, caddr_t) + hdrlen);
486 			m->m_len += len;
487 		} else {
488 			m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
489 			if (m->m_next == 0) {
490 				(void) m_freem(m);
491 				error = ENOBUFS;
492 				goto out;
493 			}
494 		}
495 #endif
496 		/*
497 		 * If we're sending everything we've got, set PUSH.
498 		 * (This will keep happy those implementations which only
499 		 * give data to the user when a buffer fills or
500 		 * a PUSH comes in.)
501 		 */
502 		if (off + len == so->so_snd.sb_cc)
503 			flags |= TH_PUSH;
504 	} else {
505 		if (tp->t_flags & TF_ACKNOW)
506 			tcpstat.tcps_sndacks++;
507 		else if (flags & (TH_SYN|TH_FIN|TH_RST))
508 			tcpstat.tcps_sndctrl++;
509 		else if (SEQ_GT(tp->snd_up, tp->snd_una))
510 			tcpstat.tcps_sndurg++;
511 		else
512 			tcpstat.tcps_sndwinup++;
513 
514 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
515 		if (m == NULL) {
516 			error = ENOBUFS;
517 			goto out;
518 		}
519 		m->m_data += max_linkhdr;
520 		m->m_len = hdrlen;
521 	}
522 	m->m_pkthdr.rcvif = (struct ifnet *)0;
523 	ti = mtod(m, struct tcpiphdr *);
524 	if (tp->t_template == 0)
525 		panic("tcp_output");
526 	bcopy((caddr_t)tp->t_template, (caddr_t)ti, sizeof (struct tcpiphdr));
527 
528 	/*
529 	 * Fill in fields, remembering maximum advertised
530 	 * window for use in delaying messages about window sizes.
531 	 * If resending a FIN, be sure not to use a new sequence number.
532 	 */
533 	if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
534 	    tp->snd_nxt == tp->snd_max)
535 		tp->snd_nxt--;
536 	/*
537 	 * If we are doing retransmissions, then snd_nxt will
538 	 * not reflect the first unsent octet.  For ACK only
539 	 * packets, we do not want the sequence number of the
540 	 * retransmitted packet, we want the sequence number
541 	 * of the next unsent octet.  So, if there is no data
542 	 * (and no SYN or FIN), use snd_max instead of snd_nxt
543 	 * when filling in ti_seq.  But if we are in persist
544 	 * state, snd_max might reflect one byte beyond the
545 	 * right edge of the window, so use snd_nxt in that
546 	 * case, since we know we aren't doing a retransmission.
547 	 * (retransmit and persist are mutually exclusive...)
548 	 */
549 	if (len || (flags & (TH_SYN|TH_FIN)) || tp->t_timer[TCPT_PERSIST])
550 		ti->ti_seq = htonl(tp->snd_nxt);
551 	else
552 		ti->ti_seq = htonl(tp->snd_max);
553 	ti->ti_ack = htonl(tp->rcv_nxt);
554 	if (optlen) {
555 		bcopy((caddr_t)opt, (caddr_t)(ti + 1), optlen);
556 		ti->ti_off = (sizeof (struct tcphdr) + optlen) >> 2;
557 	}
558 	ti->ti_flags = flags;
559 	/*
560 	 * Calculate receive window.  Don't shrink window,
561 	 * but avoid silly window syndrome.
562 	 */
563 	if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)rxsegsize)
564 		win = 0;
565 	if (win > (long)TCP_MAXWIN << tp->rcv_scale)
566 		win = (long)TCP_MAXWIN << tp->rcv_scale;
567 	if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
568 		win = (long)(tp->rcv_adv - tp->rcv_nxt);
569 	ti->ti_win = htons((u_int16_t) (win>>tp->rcv_scale));
570 	if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
571 		u_int32_t urp = tp->snd_up - tp->snd_nxt;
572 		if (urp > IP_MAXPACKET)
573 			urp = IP_MAXPACKET;
574 		ti->ti_urp = htons((u_int16_t)urp);
575 		ti->ti_flags |= TH_URG;
576 	} else
577 		/*
578 		 * If no urgent pointer to send, then we pull
579 		 * the urgent pointer to the left edge of the send window
580 		 * so that it doesn't drift into the send window on sequence
581 		 * number wraparound.
582 		 */
583 		tp->snd_up = tp->snd_una;		/* drag it along */
584 
585 	/*
586 	 * Put TCP length in extended header, and then
587 	 * checksum extended header and data.
588 	 */
589 	if (len + optlen)
590 		ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) +
591 		    optlen + len));
592 	ti->ti_sum = in_cksum(m, (int)(hdrlen + len));
593 
594 	/*
595 	 * In transmit state, time the transmission and arrange for
596 	 * the retransmit.  In persist state, just set snd_max.
597 	 */
598 	if (tp->t_force == 0 || tp->t_timer[TCPT_PERSIST] == 0) {
599 		tcp_seq startseq = tp->snd_nxt;
600 
601 		/*
602 		 * Advance snd_nxt over sequence space of this segment.
603 		 */
604 		if (flags & (TH_SYN|TH_FIN)) {
605 			if (flags & TH_SYN)
606 				tp->snd_nxt++;
607 			if (flags & TH_FIN) {
608 				tp->snd_nxt++;
609 				tp->t_flags |= TF_SENTFIN;
610 			}
611 		}
612 		tp->snd_nxt += len;
613 		if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
614 			tp->snd_max = tp->snd_nxt;
615 			/*
616 			 * Time this transmission if not a retransmission and
617 			 * not currently timing anything.
618 			 */
619 			if (tp->t_rtt == 0) {
620 				tp->t_rtt = 1;
621 				tp->t_rtseq = startseq;
622 				tcpstat.tcps_segstimed++;
623 			}
624 		}
625 
626 		/*
627 		 * Set retransmit timer if not currently set,
628 		 * and not doing an ack or a keep-alive probe.
629 		 * Initial value for retransmit timer is smoothed
630 		 * round-trip time + 2 * round-trip time variance.
631 		 * Initialize shift counter which is used for backoff
632 		 * of retransmit time.
633 		 */
634 		if (tp->t_timer[TCPT_REXMT] == 0 &&
635 		    tp->snd_nxt != tp->snd_una) {
636 			tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
637 			if (tp->t_timer[TCPT_PERSIST]) {
638 				tp->t_timer[TCPT_PERSIST] = 0;
639 				tp->t_rxtshift = 0;
640 			}
641 		}
642 	} else
643 		if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
644 			tp->snd_max = tp->snd_nxt + len;
645 
646 	/*
647 	 * Trace.
648 	 */
649 	if (so->so_options & SO_DEBUG)
650 		tcp_trace(TA_OUTPUT, tp->t_state, tp, ti, 0);
651 
652 	/*
653 	 * Fill in IP length and desired time to live and
654 	 * send to IP level.  There should be a better way
655 	 * to handle ttl and tos; we could keep them in
656 	 * the template, but need a way to checksum without them.
657 	 */
658 	m->m_pkthdr.len = hdrlen + len;
659 #ifdef TUBA
660 	if (tp->t_tuba_pcb)
661 		error = tuba_output(m, tp);
662 	else
663 #endif
664     {
665 	struct rtentry *rt;
666 
667 	((struct ip *)ti)->ip_len = m->m_pkthdr.len;
668 	((struct ip *)ti)->ip_ttl = tp->t_inpcb->inp_ip.ip_ttl;	/* XXX */
669 	((struct ip *)ti)->ip_tos = tp->t_inpcb->inp_ip.ip_tos;	/* XXX */
670 
671 	if (ip_mtudisc && (rt = in_pcbrtentry(tp->t_inpcb)) != 0 &&
672 	    (rt->rt_rmx.rmx_locks & RTV_MTU) == 0)
673 		((struct ip *)ti)->ip_off |= IP_DF;
674 
675 #if BSD >= 43
676 	error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
677 	    so->so_options & SO_DONTROUTE, 0);
678 #else
679 	error = ip_output(m, (struct mbuf *)0, &tp->t_inpcb->inp_route,
680 	    so->so_options & SO_DONTROUTE);
681 #endif
682     }
683 	if (error) {
684 out:
685 		if (error == ENOBUFS) {
686 			tcp_quench(tp->t_inpcb, 0);
687 			return (0);
688 		}
689 		if ((error == EHOSTUNREACH || error == ENETDOWN)
690 		    && TCPS_HAVERCVDSYN(tp->t_state)) {
691 			tp->t_softerror = error;
692 			return (0);
693 		}
694 		return (error);
695 	}
696 	tcpstat.tcps_sndtotal++;
697 	if (tp->t_flags & TF_DELACK)
698 		tcpstat.tcps_delack++;
699 
700 	/*
701 	 * Data sent (as far as we can tell).
702 	 * If this advertises a larger window than any other segment,
703 	 * then remember the size of the advertised window.
704 	 * Any pending ACK has now been sent.
705 	 */
706 	if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
707 		tp->rcv_adv = tp->rcv_nxt + win;
708 	tp->last_ack_sent = tp->rcv_nxt;
709 	tp->t_flags &= ~TF_ACKNOW;
710 	TCP_CLEAR_DELACK(tp);
711 	if (sendalot)
712 		goto again;
713 	return (0);
714 }
715 
716 void
717 tcp_setpersist(tp)
718 	register struct tcpcb *tp;
719 {
720 	register int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2);
721 
722 	if (tp->t_timer[TCPT_REXMT])
723 		panic("tcp_output REXMT");
724 	/*
725 	 * Start/restart persistance timer.
726 	 */
727 	if (t < tp->t_rttmin)
728 		t = tp->t_rttmin;
729 	TCPT_RANGESET(tp->t_timer[TCPT_PERSIST],
730 	    t * tcp_backoff[tp->t_rxtshift],
731 	    TCPTV_PERSMIN, TCPTV_PERSMAX);
732 	if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
733 		tp->t_rxtshift++;
734 }
735