xref: /freebsd/sys/netinet/tcp_output.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)tcp_output.c	8.4 (Berkeley) 5/24/95
34  * $FreeBSD$
35  */
36 
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <stddef.h>
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/mbuf.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 
52 #include <net/route.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/ip.h>
57 #ifdef INET6
58 #include <netinet/ip6.h>
59 #endif
60 #include <netinet/in_pcb.h>
61 #ifdef INET6
62 #include <netinet6/in6_pcb.h>
63 #endif
64 #include <netinet/ip_var.h>
65 #ifdef INET6
66 #include <netinet6/ip6_var.h>
67 #endif
68 #include <netinet/tcp.h>
69 #define	TCPOUTFLAGS
70 #include <netinet/tcp_fsm.h>
71 #include <netinet/tcp_seq.h>
72 #include <netinet/tcp_timer.h>
73 #include <netinet/tcp_var.h>
74 #include <netinet/tcpip.h>
75 #ifdef TCPDEBUG
76 #include <netinet/tcp_debug.h>
77 #endif
78 
79 #ifdef IPSEC
80 #include <netinet6/ipsec.h>
81 #endif /*IPSEC*/
82 
83 #ifdef notyet
84 extern struct mbuf *m_copypack();
85 #endif
86 
87 static int path_mtu_discovery = 1;
88 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
89 	&path_mtu_discovery, 1, "Enable Path MTU Discovery");
90 
91 int ss_fltsz = 1;
92 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
93 	&ss_fltsz, 1, "Slow start flight size");
94 
95 int ss_fltsz_local = TCP_MAXWIN;               /* something large */
96 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
97 	&ss_fltsz_local, 1, "Slow start flight size for local networks");
98 
99 /*
100  * Tcp output routine: figure out what should be sent and send it.
101  */
102 int
103 tcp_output(tp)
104 	register struct tcpcb *tp;
105 {
106 	register struct socket *so = tp->t_inpcb->inp_socket;
107 	register long len, win;
108 	int off, flags, error;
109 	register struct mbuf *m;
110 	struct ip *ip = NULL;
111 	register struct ipovly *ipov = NULL;
112 #ifdef INET6
113 	struct ip6_hdr *ip6 = NULL;
114 #endif /* INET6 */
115 	register struct tcphdr *th;
116 	u_char opt[TCP_MAXOLEN];
117 	unsigned ipoptlen, optlen, hdrlen;
118 	int idle, sendalot;
119 	struct rmxp_tao *taop;
120 	struct rmxp_tao tao_noncached;
121 #ifdef INET6
122 	int isipv6;
123 #endif
124 
125 #ifdef INET6
126 	isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
127 #endif
128 
129 	/*
130 	 * Determine length of data that should be transmitted,
131 	 * and flags that will be used.
132 	 * If there is some data or critical controls (SYN, RST)
133 	 * to send, then transmit; otherwise, investigate further.
134 	 */
135 	idle = (tp->snd_max == tp->snd_una);
136 	if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
137 		/*
138 		 * We have been idle for "a while" and no acks are
139 		 * expected to clock out any data we send --
140 		 * slow start to get ack "clock" running again.
141 		 *
142 		 * Set the slow-start flight size depending on whether
143 		 * this is a local network or not.
144 		 */
145 		if (
146 #ifdef INET6
147 		    (isipv6 && in6_localaddr(&tp->t_inpcb->in6p_faddr)) ||
148 		    (!isipv6 &&
149 #endif
150 		     in_localaddr(tp->t_inpcb->inp_faddr)
151 #ifdef INET6
152 		     )
153 #endif
154 		    )
155 			tp->snd_cwnd = tp->t_maxseg * ss_fltsz_local;
156 		else
157 			tp->snd_cwnd = tp->t_maxseg * ss_fltsz;
158 	}
159 again:
160 	sendalot = 0;
161 	off = tp->snd_nxt - tp->snd_una;
162 	win = min(tp->snd_wnd, tp->snd_cwnd);
163 
164 	flags = tcp_outflags[tp->t_state];
165 	/*
166 	 * Get standard flags, and add SYN or FIN if requested by 'hidden'
167 	 * state flags.
168 	 */
169 	if (tp->t_flags & TF_NEEDFIN)
170 		flags |= TH_FIN;
171 	if (tp->t_flags & TF_NEEDSYN)
172 		flags |= TH_SYN;
173 
174 	/*
175 	 * If in persist timeout with window of 0, send 1 byte.
176 	 * Otherwise, if window is small but nonzero
177 	 * and timer expired, we will send what we can
178 	 * and go to transmit state.
179 	 */
180 	if (tp->t_force) {
181 		if (win == 0) {
182 			/*
183 			 * If we still have some data to send, then
184 			 * clear the FIN bit.  Usually this would
185 			 * happen below when it realizes that we
186 			 * aren't sending all the data.  However,
187 			 * if we have exactly 1 byte of unsent data,
188 			 * then it won't clear the FIN bit below,
189 			 * and if we are in persist state, we wind
190 			 * up sending the packet without recording
191 			 * that we sent the FIN bit.
192 			 *
193 			 * We can't just blindly clear the FIN bit,
194 			 * because if we don't have any more data
195 			 * to send then the probe will be the FIN
196 			 * itself.
197 			 */
198 			if (off < so->so_snd.sb_cc)
199 				flags &= ~TH_FIN;
200 			win = 1;
201 		} else {
202 			callout_stop(tp->tt_persist);
203 			tp->t_rxtshift = 0;
204 		}
205 	}
206 
207 	len = (long)ulmin(so->so_snd.sb_cc, win) - off;
208 
209 	if ((taop = tcp_gettaocache(tp->t_inpcb)) == NULL) {
210 		taop = &tao_noncached;
211 		bzero(taop, sizeof(*taop));
212 	}
213 
214 	/*
215 	 * Lop off SYN bit if it has already been sent.  However, if this
216 	 * is SYN-SENT state and if segment contains data and if we don't
217 	 * know that foreign host supports TAO, suppress sending segment.
218 	 */
219 	if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
220 		flags &= ~TH_SYN;
221 		off--, len++;
222 		if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
223 		    taop->tao_ccsent == 0)
224 			return 0;
225 	}
226 
227 	/*
228 	 * Be careful not to send data and/or FIN on SYN segments
229 	 * in cases when no CC option will be sent.
230 	 * This measure is needed to prevent interoperability problems
231 	 * with not fully conformant TCP implementations.
232 	 */
233 	if ((flags & TH_SYN) &&
234 	    ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) ||
235 	     ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) {
236 		len = 0;
237 		flags &= ~TH_FIN;
238 	}
239 
240 	if (len < 0) {
241 		/*
242 		 * If FIN has been sent but not acked,
243 		 * but we haven't been called to retransmit,
244 		 * len will be -1.  Otherwise, window shrank
245 		 * after we sent into it.  If window shrank to 0,
246 		 * cancel pending retransmit, pull snd_nxt back
247 		 * to (closed) window, and set the persist timer
248 		 * if it isn't already going.  If the window didn't
249 		 * close completely, just wait for an ACK.
250 		 */
251 		len = 0;
252 		if (win == 0) {
253 			callout_stop(tp->tt_rexmt);
254 			tp->t_rxtshift = 0;
255 			tp->snd_nxt = tp->snd_una;
256 			if (!callout_active(tp->tt_persist))
257 				tcp_setpersist(tp);
258 		}
259 	}
260 	if (len > tp->t_maxseg) {
261 		len = tp->t_maxseg;
262 		sendalot = 1;
263 	}
264 	if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
265 		flags &= ~TH_FIN;
266 
267 	win = sbspace(&so->so_rcv);
268 
269 	/*
270 	 * Sender silly window avoidance.  If connection is idle
271 	 * and can send all data, a maximum segment,
272 	 * at least a maximum default-size segment do it,
273 	 * or are forced, do it; otherwise don't bother.
274 	 * If peer's buffer is tiny, then send
275 	 * when window is at least half open.
276 	 * If retransmitting (possibly after persist timer forced us
277 	 * to send into a small window), then must resend.
278 	 */
279 	if (len) {
280 		if (len == tp->t_maxseg)
281 			goto send;
282 		if (!(tp->t_flags & TF_MORETOCOME) &&
283 		    (idle || tp->t_flags & TF_NODELAY) &&
284 		    (tp->t_flags & TF_NOPUSH) == 0 &&
285 		    len + off >= so->so_snd.sb_cc)
286 			goto send;
287 		if (tp->t_force)
288 			goto send;
289 		if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
290 			goto send;
291 		if (SEQ_LT(tp->snd_nxt, tp->snd_max))
292 			goto send;
293 	}
294 
295 	/*
296 	 * Compare available window to amount of window
297 	 * known to peer (as advertised window less
298 	 * next expected input).  If the difference is at least two
299 	 * max size segments, or at least 50% of the maximum possible
300 	 * window, then want to send a window update to peer.
301 	 */
302 	if (win > 0) {
303 		/*
304 		 * "adv" is the amount we can increase the window,
305 		 * taking into account that we are limited by
306 		 * TCP_MAXWIN << tp->rcv_scale.
307 		 */
308 		long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
309 			(tp->rcv_adv - tp->rcv_nxt);
310 
311 		if (adv >= (long) (2 * tp->t_maxseg))
312 			goto send;
313 		if (2 * adv >= (long) so->so_rcv.sb_hiwat)
314 			goto send;
315 	}
316 
317 	/*
318 	 * Send if we owe peer an ACK.
319 	 */
320 	if (tp->t_flags & TF_ACKNOW)
321 		goto send;
322 	if ((flags & TH_RST) ||
323 	    ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
324 		goto send;
325 	if (SEQ_GT(tp->snd_up, tp->snd_una))
326 		goto send;
327 	/*
328 	 * If our state indicates that FIN should be sent
329 	 * and we have not yet done so, or we're retransmitting the FIN,
330 	 * then we need to send.
331 	 */
332 	if (flags & TH_FIN &&
333 	    ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
334 		goto send;
335 
336 	/*
337 	 * TCP window updates are not reliable, rather a polling protocol
338 	 * using ``persist'' packets is used to insure receipt of window
339 	 * updates.  The three ``states'' for the output side are:
340 	 *	idle			not doing retransmits or persists
341 	 *	persisting		to move a small or zero window
342 	 *	(re)transmitting	and thereby not persisting
343 	 *
344 	 * callout_active(tp->tt_persist)
345 	 *	is true when we are in persist state.
346 	 * tp->t_force
347 	 *	is set when we are called to send a persist packet.
348 	 * callout_active(tp->tt_rexmt)
349 	 *	is set when we are retransmitting
350 	 * The output side is idle when both timers are zero.
351 	 *
352 	 * If send window is too small, there is data to transmit, and no
353 	 * retransmit or persist is pending, then go to persist state.
354 	 * If nothing happens soon, send when timer expires:
355 	 * if window is nonzero, transmit what we can,
356 	 * otherwise force out a byte.
357 	 */
358 	if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) &&
359 	    !callout_active(tp->tt_persist)) {
360 		tp->t_rxtshift = 0;
361 		tcp_setpersist(tp);
362 	}
363 
364 	/*
365 	 * No reason to send a segment, just return.
366 	 */
367 	return (0);
368 
369 send:
370 	/*
371 	 * Before ESTABLISHED, force sending of initial options
372 	 * unless TCP set not to do any options.
373 	 * NOTE: we assume that the IP/TCP header plus TCP options
374 	 * always fit in a single mbuf, leaving room for a maximum
375 	 * link header, i.e.
376 	 *	max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MHLEN
377 	 */
378 	optlen = 0;
379 #ifdef INET6
380 	if (isipv6)
381 		hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
382 	else
383 #endif
384 	hdrlen = sizeof (struct tcpiphdr);
385 	if (flags & TH_SYN) {
386 		tp->snd_nxt = tp->iss;
387 		if ((tp->t_flags & TF_NOOPT) == 0) {
388 			u_short mss;
389 
390 			opt[0] = TCPOPT_MAXSEG;
391 			opt[1] = TCPOLEN_MAXSEG;
392 			mss = htons((u_short) tcp_mssopt(tp));
393 			(void)memcpy(opt + 2, &mss, sizeof(mss));
394 			optlen = TCPOLEN_MAXSEG;
395 
396 			if ((tp->t_flags & TF_REQ_SCALE) &&
397 			    ((flags & TH_ACK) == 0 ||
398 			    (tp->t_flags & TF_RCVD_SCALE))) {
399 				*((u_int32_t *)(opt + optlen)) = htonl(
400 					TCPOPT_NOP << 24 |
401 					TCPOPT_WINDOW << 16 |
402 					TCPOLEN_WINDOW << 8 |
403 					tp->request_r_scale);
404 				optlen += 4;
405 			}
406 		}
407  	}
408 
409  	/*
410 	 * Send a timestamp and echo-reply if this is a SYN and our side
411 	 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
412 	 * and our peer have sent timestamps in our SYN's.
413  	 */
414  	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
415  	    (flags & TH_RST) == 0 &&
416 	    ((flags & TH_ACK) == 0 ||
417 	     (tp->t_flags & TF_RCVD_TSTMP))) {
418 		u_int32_t *lp = (u_int32_t *)(opt + optlen);
419 
420  		/* Form timestamp option as shown in appendix A of RFC 1323. */
421  		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
422  		*lp++ = htonl(ticks);
423  		*lp   = htonl(tp->ts_recent);
424  		optlen += TCPOLEN_TSTAMP_APPA;
425  	}
426 
427  	/*
428 	 * Send `CC-family' options if our side wants to use them (TF_REQ_CC),
429 	 * options are allowed (!TF_NOOPT) and it's not a RST.
430  	 */
431  	if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
432  	     (flags & TH_RST) == 0) {
433 		switch (flags & (TH_SYN|TH_ACK)) {
434 		/*
435 		 * This is a normal ACK, send CC if we received CC before
436 		 * from our peer.
437 		 */
438 		case TH_ACK:
439 			if (!(tp->t_flags & TF_RCVD_CC))
440 				break;
441 			/*FALLTHROUGH*/
442 
443 		/*
444 		 * We can only get here in T/TCP's SYN_SENT* state, when
445 		 * we're a sending a non-SYN segment without waiting for
446 		 * the ACK of our SYN.  A check above assures that we only
447 		 * do this if our peer understands T/TCP.
448 		 */
449 		case 0:
450 			opt[optlen++] = TCPOPT_NOP;
451 			opt[optlen++] = TCPOPT_NOP;
452 			opt[optlen++] = TCPOPT_CC;
453 			opt[optlen++] = TCPOLEN_CC;
454 			*(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
455 
456 			optlen += 4;
457 			break;
458 
459 		/*
460 		 * This is our initial SYN, check whether we have to use
461 		 * CC or CC.new.
462 		 */
463 		case TH_SYN:
464 			opt[optlen++] = TCPOPT_NOP;
465 			opt[optlen++] = TCPOPT_NOP;
466 			opt[optlen++] = tp->t_flags & TF_SENDCCNEW ?
467 						TCPOPT_CCNEW : TCPOPT_CC;
468 			opt[optlen++] = TCPOLEN_CC;
469 			*(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
470  			optlen += 4;
471 			break;
472 
473 		/*
474 		 * This is a SYN,ACK; send CC and CC.echo if we received
475 		 * CC from our peer.
476 		 */
477 		case (TH_SYN|TH_ACK):
478 			if (tp->t_flags & TF_RCVD_CC) {
479 				opt[optlen++] = TCPOPT_NOP;
480 				opt[optlen++] = TCPOPT_NOP;
481 				opt[optlen++] = TCPOPT_CC;
482 				opt[optlen++] = TCPOLEN_CC;
483 				*(u_int32_t *)&opt[optlen] =
484 					htonl(tp->cc_send);
485 				optlen += 4;
486 				opt[optlen++] = TCPOPT_NOP;
487 				opt[optlen++] = TCPOPT_NOP;
488 				opt[optlen++] = TCPOPT_CCECHO;
489 				opt[optlen++] = TCPOLEN_CC;
490 				*(u_int32_t *)&opt[optlen] =
491 					htonl(tp->cc_recv);
492 				optlen += 4;
493 			}
494 			break;
495 		}
496  	}
497 
498  	hdrlen += optlen;
499 
500 #ifdef INET6
501 	if (isipv6)
502 		ipoptlen = ip6_optlen(tp->t_inpcb);
503 	else
504 #endif
505       {
506 	if (tp->t_inpcb->inp_options) {
507 		ipoptlen = tp->t_inpcb->inp_options->m_len -
508 				offsetof(struct ipoption, ipopt_list);
509 	} else {
510 		ipoptlen = 0;
511 	}
512       }
513 #ifdef IPSEC
514 	ipoptlen += ipsec_hdrsiz_tcp(tp);
515 #endif
516 
517 	/*
518 	 * Adjust data length if insertion of options will
519 	 * bump the packet length beyond the t_maxopd length.
520 	 * Clear the FIN bit because we cut off the tail of
521 	 * the segment.
522 	 */
523 	if (len + optlen + ipoptlen > tp->t_maxopd) {
524 		/*
525 		 * If there is still more to send, don't close the connection.
526 		 */
527 		flags &= ~TH_FIN;
528 		len = tp->t_maxopd - optlen - ipoptlen;
529 		sendalot = 1;
530 	}
531 
532 /*#ifdef DIAGNOSTIC*/
533 #ifdef INET6
534  	if (max_linkhdr + hdrlen > MCLBYTES)
535 		panic("tcphdr too big");
536 #else
537  	if (max_linkhdr + hdrlen > MHLEN)
538 		panic("tcphdr too big");
539 #endif
540 /*#endif*/
541 
542 	/*
543 	 * Grab a header mbuf, attaching a copy of data to
544 	 * be transmitted, and initialize the header from
545 	 * the template for sends on this connection.
546 	 */
547 	if (len) {
548 		if (tp->t_force && len == 1)
549 			tcpstat.tcps_sndprobe++;
550 		else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
551 			tcpstat.tcps_sndrexmitpack++;
552 			tcpstat.tcps_sndrexmitbyte += len;
553 		} else {
554 			tcpstat.tcps_sndpack++;
555 			tcpstat.tcps_sndbyte += len;
556 		}
557 #ifdef notyet
558 		if ((m = m_copypack(so->so_snd.sb_mb, off,
559 		    (int)len, max_linkhdr + hdrlen)) == 0) {
560 			error = ENOBUFS;
561 			goto out;
562 		}
563 		/*
564 		 * m_copypack left space for our hdr; use it.
565 		 */
566 		m->m_len += hdrlen;
567 		m->m_data -= hdrlen;
568 #else
569 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
570 		if (m == NULL) {
571 			error = ENOBUFS;
572 			goto out;
573 		}
574 #ifdef INET6
575 		if (MHLEN < hdrlen + max_linkhdr) {
576 			MCLGET(m, M_DONTWAIT);
577 			if ((m->m_flags & M_EXT) == 0) {
578 				m_freem(m);
579 				error = ENOBUFS;
580 				goto out;
581 			}
582 		}
583 #endif
584 		m->m_data += max_linkhdr;
585 		m->m_len = hdrlen;
586 		if (len <= MHLEN - hdrlen - max_linkhdr) {
587 			m_copydata(so->so_snd.sb_mb, off, (int) len,
588 			    mtod(m, caddr_t) + hdrlen);
589 			m->m_len += len;
590 		} else {
591 			m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
592 			if (m->m_next == 0) {
593 				(void) m_free(m);
594 				error = ENOBUFS;
595 				goto out;
596 			}
597 		}
598 #endif
599 		/*
600 		 * If we're sending everything we've got, set PUSH.
601 		 * (This will keep happy those implementations which only
602 		 * give data to the user when a buffer fills or
603 		 * a PUSH comes in.)
604 		 */
605 		if (off + len == so->so_snd.sb_cc)
606 			flags |= TH_PUSH;
607 	} else {
608 		if (tp->t_flags & TF_ACKNOW)
609 			tcpstat.tcps_sndacks++;
610 		else if (flags & (TH_SYN|TH_FIN|TH_RST))
611 			tcpstat.tcps_sndctrl++;
612 		else if (SEQ_GT(tp->snd_up, tp->snd_una))
613 			tcpstat.tcps_sndurg++;
614 		else
615 			tcpstat.tcps_sndwinup++;
616 
617 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
618 		if (m == NULL) {
619 			error = ENOBUFS;
620 			goto out;
621 		}
622 #ifdef INET6
623 		if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
624 		    MHLEN >= hdrlen) {
625 			MH_ALIGN(m, hdrlen);
626 		} else
627 #endif
628 		m->m_data += max_linkhdr;
629 		m->m_len = hdrlen;
630 	}
631 	m->m_pkthdr.rcvif = (struct ifnet *)0;
632 	if (tp->t_template == 0)
633 		panic("tcp_output");
634 #ifdef INET6
635 	if (isipv6) {
636 		ip6 = mtod(m, struct ip6_hdr *);
637 		th = (struct tcphdr *)(ip6 + 1);
638 		bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip6,
639 		      sizeof(struct ip6_hdr));
640 		bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
641 		      sizeof(struct tcphdr));
642 	} else
643 #endif /* INET6 */
644       {
645 	ip = mtod(m, struct ip *);
646 	ipov = (struct ipovly *)ip;
647 	th = (struct tcphdr *)(ip + 1);
648 	bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip,
649 	      sizeof(struct ip));
650 	bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
651 	      sizeof(struct tcphdr));
652       }
653 
654 	/*
655 	 * Fill in fields, remembering maximum advertised
656 	 * window for use in delaying messages about window sizes.
657 	 * If resending a FIN, be sure not to use a new sequence number.
658 	 */
659 	if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
660 	    tp->snd_nxt == tp->snd_max)
661 		tp->snd_nxt--;
662 	/*
663 	 * If we are doing retransmissions, then snd_nxt will
664 	 * not reflect the first unsent octet.  For ACK only
665 	 * packets, we do not want the sequence number of the
666 	 * retransmitted packet, we want the sequence number
667 	 * of the next unsent octet.  So, if there is no data
668 	 * (and no SYN or FIN), use snd_max instead of snd_nxt
669 	 * when filling in ti_seq.  But if we are in persist
670 	 * state, snd_max might reflect one byte beyond the
671 	 * right edge of the window, so use snd_nxt in that
672 	 * case, since we know we aren't doing a retransmission.
673 	 * (retransmit and persist are mutually exclusive...)
674 	 */
675 	if (len || (flags & (TH_SYN|TH_FIN))
676 	    || callout_active(tp->tt_persist))
677 		th->th_seq = htonl(tp->snd_nxt);
678 	else
679 		th->th_seq = htonl(tp->snd_max);
680 	th->th_ack = htonl(tp->rcv_nxt);
681 	if (optlen) {
682 		bcopy(opt, th + 1, optlen);
683 		th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
684 	}
685 	th->th_flags = flags;
686 	/*
687 	 * Calculate receive window.  Don't shrink window,
688 	 * but avoid silly window syndrome.
689 	 */
690 	if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
691 		win = 0;
692 	if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
693 		win = (long)(tp->rcv_adv - tp->rcv_nxt);
694 	if (win > (long)TCP_MAXWIN << tp->rcv_scale)
695 		win = (long)TCP_MAXWIN << tp->rcv_scale;
696 	th->th_win = htons((u_short) (win>>tp->rcv_scale));
697 	if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
698 		th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
699 		th->th_flags |= TH_URG;
700 	} else
701 		/*
702 		 * If no urgent pointer to send, then we pull
703 		 * the urgent pointer to the left edge of the send window
704 		 * so that it doesn't drift into the send window on sequence
705 		 * number wraparound.
706 		 */
707 		tp->snd_up = tp->snd_una;		/* drag it along */
708 
709 	/*
710 	 * Put TCP length in extended header, and then
711 	 * checksum extended header and data.
712 	 */
713 	m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
714 #ifdef INET6
715 	if (isipv6)
716 		/*
717 		 * ip6_plen is not need to be filled now, and will be filled
718 		 * in ip6_output.
719 		 */
720 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
721 				       sizeof(struct tcphdr) + optlen + len);
722 	else
723 #endif /* INET6 */
724       {
725 	if (len + optlen)
726 		ipov->ih_len = htons((u_short)(sizeof (struct tcphdr) +
727 		    optlen + len));
728 	th->th_sum = in_cksum(m, (int)(hdrlen + len));
729 #ifdef INET6
730 	/* Re-initialization for later version check */
731 	ip->ip_v = IPVERSION;
732 
733 #endif /* INET6 */
734       }
735 
736 	/*
737 	 * In transmit state, time the transmission and arrange for
738 	 * the retransmit.  In persist state, just set snd_max.
739 	 */
740 	if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
741 		tcp_seq startseq = tp->snd_nxt;
742 
743 		/*
744 		 * Advance snd_nxt over sequence space of this segment.
745 		 */
746 		if (flags & (TH_SYN|TH_FIN)) {
747 			if (flags & TH_SYN)
748 				tp->snd_nxt++;
749 			if (flags & TH_FIN) {
750 				tp->snd_nxt++;
751 				tp->t_flags |= TF_SENTFIN;
752 			}
753 		}
754 		tp->snd_nxt += len;
755 		if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
756 			tp->snd_max = tp->snd_nxt;
757 			/*
758 			 * Time this transmission if not a retransmission and
759 			 * not currently timing anything.
760 			 */
761 			if (tp->t_rtttime == 0) {
762 				tp->t_rtttime = ticks;
763 				tp->t_rtseq = startseq;
764 				tcpstat.tcps_segstimed++;
765 			}
766 		}
767 
768 		/*
769 		 * Set retransmit timer if not currently set,
770 		 * and not doing an ack or a keep-alive probe.
771 		 * Initial value for retransmit timer is smoothed
772 		 * round-trip time + 2 * round-trip time variance.
773 		 * Initialize shift counter which is used for backoff
774 		 * of retransmit time.
775 		 */
776 		if (!callout_active(tp->tt_rexmt) &&
777 		    tp->snd_nxt != tp->snd_una) {
778 			callout_reset(tp->tt_rexmt, tp->t_rxtcur,
779 				      tcp_timer_rexmt, tp);
780 			if (callout_active(tp->tt_persist)) {
781 				callout_stop(tp->tt_persist);
782 				tp->t_rxtshift = 0;
783 			}
784 		}
785 	} else
786 		if (SEQ_GT(tp->snd_nxt + len, tp->snd_max))
787 			tp->snd_max = tp->snd_nxt + len;
788 
789 #ifdef TCPDEBUG
790 	/*
791 	 * Trace.
792 	 */
793 	if (so->so_options & SO_DEBUG)
794 		tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
795 #endif
796 
797 	/*
798 	 * Fill in IP length and desired time to live and
799 	 * send to IP level.  There should be a better way
800 	 * to handle ttl and tos; we could keep them in
801 	 * the template, but need a way to checksum without them.
802 	 */
803 	/*
804 	 * m->m_pkthdr.len should have been set before cksum calcuration,
805 	 * because in6_cksum() need it.
806 	 */
807 #ifdef INET6
808 	if (isipv6) {
809 		/*
810 		 * we separately set hoplimit for every segment, since the
811 		 * user might want to change the value via setsockopt.
812 		 * Also, desired default hop limit might be changed via
813 		 * Neighbor Discovery.
814 		 */
815 		ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb,
816 					       tp->t_inpcb->in6p_route.ro_rt ?
817 					       tp->t_inpcb->in6p_route.ro_rt->rt_ifp
818 					       : NULL);
819 
820 		/* TODO: IPv6 IP6TOS_ECT bit on */
821 #ifdef IPSEC
822 		m->m_pkthdr.rcvif = (struct ifnet *)so;
823 #endif /*IPSEC*/
824 		error = ip6_output(m,
825 			    tp->t_inpcb->in6p_outputopts,
826 			    &tp->t_inpcb->in6p_route,
827 			    (so->so_options & SO_DONTROUTE)|IPV6_SOCKINMRCVIF,
828 			    NULL, NULL);
829 	} else
830 #endif /* INET6 */
831     {
832 	struct rtentry *rt;
833 	ip->ip_len = m->m_pkthdr.len;
834 	ip->ip_ttl = tp->t_inpcb->inp_ip_ttl;	/* XXX */
835 	ip->ip_tos = tp->t_inpcb->inp_ip_tos;	/* XXX */
836 	/*
837 	 * See if we should do MTU discovery.  We do it only if the following
838 	 * are true:
839 	 *	1) we have a valid route to the destination
840 	 *	2) the MTU is not locked (if it is, then discovery has been
841 	 *	   disabled)
842 	 */
843 	if (path_mtu_discovery
844 	    && (rt = tp->t_inpcb->inp_route.ro_rt)
845 	    && rt->rt_flags & RTF_UP
846 	    && !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
847 		ip->ip_off |= IP_DF;
848 	}
849 	error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
850 	    (so->so_options & SO_DONTROUTE)|IP_SOCKINMRCVIF, 0);
851     }
852 	if (error) {
853 out:
854 		if (error == ENOBUFS) {
855 			tcp_quench(tp->t_inpcb, 0);
856 			return (0);
857 		}
858 		if (error == EMSGSIZE) {
859 			/*
860 			 * ip_output() will have already fixed the route
861 			 * for us.  tcp_mtudisc() will, as its last action,
862 			 * initiate retransmission, so it is important to
863 			 * not do so here.
864 			 */
865 			tcp_mtudisc(tp->t_inpcb, 0);
866 			return 0;
867 		}
868 		if ((error == EHOSTUNREACH || error == ENETDOWN)
869 		    && TCPS_HAVERCVDSYN(tp->t_state)) {
870 			tp->t_softerror = error;
871 			return (0);
872 		}
873 		return (error);
874 	}
875 	tcpstat.tcps_sndtotal++;
876 
877 	/*
878 	 * Data sent (as far as we can tell).
879 	 * If this advertises a larger window than any other segment,
880 	 * then remember the size of the advertised window.
881 	 * Any pending ACK has now been sent.
882 	 */
883 	if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
884 		tp->rcv_adv = tp->rcv_nxt + win;
885 	tp->last_ack_sent = tp->rcv_nxt;
886 	tp->t_flags &= ~TF_ACKNOW;
887 	if (tcp_delack_enabled)
888 		callout_stop(tp->tt_delack);
889 	if (sendalot)
890 		goto again;
891 	return (0);
892 }
893 
894 void
895 tcp_setpersist(tp)
896 	register struct tcpcb *tp;
897 {
898 	int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
899 	int tt;
900 
901 	if (callout_active(tp->tt_rexmt))
902 		panic("tcp_setpersist: retransmit pending");
903 	/*
904 	 * Start/restart persistance timer.
905 	 */
906 	TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
907 		      TCPTV_PERSMIN, TCPTV_PERSMAX);
908 	callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp);
909 	if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
910 		tp->t_rxtshift++;
911 }
912