2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
33 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95
34 * $FreeBSD: src/sys/netinet/tcp_output.c,v 1.39.2.20 2003/01/29 22:45:36 hsu Exp $
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
46 #include <sys/domain.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
51 #include <net/route.h>
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>
59 #include <netinet6/in6_pcb.h>
60 #include <netinet/ip6.h>
61 #include <netinet6/ip6_var.h>
63 #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>
71 #include <netinet/tcp_debug.h>
75 #include <netinet6/ipsec.h>
79 #include <netipsec/ipsec.h>
83 #include <machine/in_cksum.h>
86 extern struct mbuf *m_copypack();
89 int path_mtu_discovery = 1;
90 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
91 &path_mtu_discovery, 1, "Enable Path MTU Discovery");
94 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
95 &ss_fltsz, 1, "Slow start flight size");
97 int ss_fltsz_local = 4;
98 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
99 &ss_fltsz_local, 1, "Slow start flight size for local networks");
101 int tcp_do_newreno = 1;
102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno,
103 0, "Enable NewReno Algorithms");
106 * Tcp output routine: figure out what should be sent and send it.
110 register struct tcpcb *tp;
112 register struct socket *so = tp->t_inpcb->inp_socket;
113 register long len, win;
114 int off, flags, error;
115 register struct mbuf *m;
116 struct ip *ip = NULL;
117 register struct ipovly *ipov = NULL;
119 struct ip6_hdr *ip6 = NULL;
121 register struct tcphdr *th;
122 u_char opt[TCP_MAXOLEN];
123 unsigned ipoptlen, optlen, hdrlen;
126 int maxburst = TCP_MAXBURST;
128 struct rmxp_tao *taop;
134 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
138 * Determine length of data that should be transmitted,
139 * and flags that will be used.
140 * If there is some data or critical controls (SYN, RST)
141 * to send, then transmit; otherwise, investigate further.
143 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
144 if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
146 * We have been idle for "a while" and no acks are
147 * expected to clock out any data we send --
148 * slow start to get ack "clock" running again.
150 * Set the slow-start flight size depending on whether
151 * this is a local network or not.
155 (isipv6 && in6_localaddr(&tp->t_inpcb->in6p_faddr)) ||
158 in_localaddr(tp->t_inpcb->inp_faddr)
163 tp->snd_cwnd = tp->t_maxseg * ss_fltsz_local;
165 tp->snd_cwnd = tp->t_maxseg * ss_fltsz;
167 tp->t_flags &= ~TF_LASTIDLE;
169 if (tp->t_flags & TF_MORETOCOME) {
170 tp->t_flags |= TF_LASTIDLE;
176 off = tp->snd_nxt - tp->snd_una;
177 win = min(tp->snd_wnd, tp->snd_cwnd);
178 win = min(win, tp->snd_bwnd);
180 flags = tcp_outflags[tp->t_state];
182 * Get standard flags, and add SYN or FIN if requested by 'hidden'
185 if (tp->t_flags & TF_NEEDFIN)
187 if (tp->t_flags & TF_NEEDSYN)
191 * If in persist timeout with window of 0, send 1 byte.
192 * Otherwise, if window is small but nonzero
193 * and timer expired, we will send what we can
194 * and go to transmit state.
199 * If we still have some data to send, then
200 * clear the FIN bit. Usually this would
201 * happen below when it realizes that we
202 * aren't sending all the data. However,
203 * if we have exactly 1 byte of unsent data,
204 * then it won't clear the FIN bit below,
205 * and if we are in persist state, we wind
206 * up sending the packet without recording
207 * that we sent the FIN bit.
209 * We can't just blindly clear the FIN bit,
210 * because if we don't have any more data
211 * to send then the probe will be the FIN
214 if (off < so->so_snd.sb_cc)
218 callout_stop(tp->tt_persist);
224 * If snd_nxt == snd_max and we have transmitted a FIN, the
225 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
226 * a negative length. This can also occur when tcp opens up
227 * its congestion window while receiving additional duplicate
228 * acks after fast-retransmit because TCP will reset snd_nxt
229 * to snd_max after the fast-retransmit.
231 * In the normal retransmit-FIN-only case, however, snd_nxt will
232 * be set to snd_una, the offset will be 0, and the length may
235 len = (long)ulmin(so->so_snd.sb_cc, win) - off;
237 taop = tcp_gettaocache(&tp->t_inpcb->inp_inc);
240 * Lop off SYN bit if it has already been sent. However, if this
241 * is SYN-SENT state and if segment contains data and if we don't
242 * know that foreign host supports TAO, suppress sending segment.
244 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
247 if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
248 (taop == NULL || taop->tao_ccsent == 0))
253 * Be careful not to send data and/or FIN on SYN segments
254 * in cases when no CC option will be sent.
255 * This measure is needed to prevent interoperability problems
256 * with not fully conformant TCP implementations.
258 if ((flags & TH_SYN) &&
259 ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) ||
260 ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) {
267 * If FIN has been sent but not acked,
268 * but we haven't been called to retransmit,
269 * len will be < 0. Otherwise, window shrank
270 * after we sent into it. If window shrank to 0,
271 * cancel pending retransmit, pull snd_nxt back
272 * to (closed) window, and set the persist timer
273 * if it isn't already going. If the window didn't
274 * close completely, just wait for an ACK.
278 callout_stop(tp->tt_rexmt);
280 tp->snd_nxt = tp->snd_una;
281 if (!callout_active(tp->tt_persist))
287 * len will be >= 0 after this point. Truncate to the maximum
288 * segment length and ensure that FIN is removed if the length
289 * no longer contains the last data byte.
291 if (len > tp->t_maxseg) {
295 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
298 win = sbspace(&so->so_rcv);
301 * Sender silly window avoidance. We transmit under the following
302 * conditions when len is non-zero:
304 * - We have a full segment
305 * - This is the last buffer in a write()/send() and we are
306 * either idle or running NODELAY
307 * - we've timed out (e.g. persist timer)
308 * - we have more then 1/2 the maximum send window's worth of
309 * data (receiver may be limited the window size)
310 * - we need to retransmit
313 if (len == tp->t_maxseg)
316 * NOTE! on localhost connections an 'ack' from the remote
317 * end may occur synchronously with the output and cause
318 * us to flush a buffer queued with moretocome. XXX
320 * note: the len + off check is almost certainly unnecessary.
322 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
323 (idle || (tp->t_flags & TF_NODELAY)) &&
324 len + off >= so->so_snd.sb_cc &&
325 (tp->t_flags & TF_NOPUSH) == 0) {
328 if (tp->t_force) /* typ. timeout case */
330 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
332 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
337 * Compare available window to amount of window
338 * known to peer (as advertised window less
339 * next expected input). If the difference is at least two
340 * max size segments, or at least 50% of the maximum possible
341 * window, then want to send a window update to peer.
345 * "adv" is the amount we can increase the window,
346 * taking into account that we are limited by
347 * TCP_MAXWIN << tp->rcv_scale.
349 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
350 (tp->rcv_adv - tp->rcv_nxt);
352 if (adv >= (long) (2 * tp->t_maxseg))
354 if (2 * adv >= (long) so->so_rcv.sb_hiwat)
359 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
360 * is also a catch-all for the retransmit timer timeout case.
362 if (tp->t_flags & TF_ACKNOW)
364 if ((flags & TH_RST) ||
365 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
367 if (SEQ_GT(tp->snd_up, tp->snd_una))
370 * If our state indicates that FIN should be sent
371 * and we have not yet done so, then we need to send.
373 if (flags & TH_FIN &&
374 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
378 * TCP window updates are not reliable, rather a polling protocol
379 * using ``persist'' packets is used to insure receipt of window
380 * updates. The three ``states'' for the output side are:
381 * idle not doing retransmits or persists
382 * persisting to move a small or zero window
383 * (re)transmitting and thereby not persisting
385 * callout_active(tp->tt_persist)
386 * is true when we are in persist state.
388 * is set when we are called to send a persist packet.
389 * callout_active(tp->tt_rexmt)
390 * is set when we are retransmitting
391 * The output side is idle when both timers are zero.
393 * If send window is too small, there is data to transmit, and no
394 * retransmit or persist is pending, then go to persist state.
395 * If nothing happens soon, send when timer expires:
396 * if window is nonzero, transmit what we can,
397 * otherwise force out a byte.
399 if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) &&
400 !callout_active(tp->tt_persist)) {
406 * No reason to send a segment, just return.
412 * Before ESTABLISHED, force sending of initial options
413 * unless TCP set not to do any options.
414 * NOTE: we assume that the IP/TCP header plus TCP options
415 * always fit in a single mbuf, leaving room for a maximum
417 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
422 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
425 hdrlen = sizeof (struct tcpiphdr);
426 if (flags & TH_SYN) {
427 tp->snd_nxt = tp->iss;
428 if ((tp->t_flags & TF_NOOPT) == 0) {
431 opt[0] = TCPOPT_MAXSEG;
432 opt[1] = TCPOLEN_MAXSEG;
433 mss = htons((u_short) tcp_mssopt(tp));
434 (void)memcpy(opt + 2, &mss, sizeof(mss));
435 optlen = TCPOLEN_MAXSEG;
437 if ((tp->t_flags & TF_REQ_SCALE) &&
438 ((flags & TH_ACK) == 0 ||
439 (tp->t_flags & TF_RCVD_SCALE))) {
440 *((u_int32_t *)(opt + optlen)) = htonl(
442 TCPOPT_WINDOW << 16 |
443 TCPOLEN_WINDOW << 8 |
444 tp->request_r_scale);
451 * Send a timestamp and echo-reply if this is a SYN and our side
452 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
453 * and our peer have sent timestamps in our SYN's.
455 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
456 (flags & TH_RST) == 0 &&
457 ((flags & TH_ACK) == 0 ||
458 (tp->t_flags & TF_RCVD_TSTMP))) {
459 u_int32_t *lp = (u_int32_t *)(opt + optlen);
461 /* Form timestamp option as shown in appendix A of RFC 1323. */
462 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
463 *lp++ = htonl(ticks);
464 *lp = htonl(tp->ts_recent);
465 optlen += TCPOLEN_TSTAMP_APPA;
469 * Send `CC-family' options if our side wants to use them (TF_REQ_CC),
470 * options are allowed (!TF_NOOPT) and it's not a RST.
472 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
473 (flags & TH_RST) == 0) {
474 switch (flags & (TH_SYN|TH_ACK)) {
476 * This is a normal ACK, send CC if we received CC before
480 if (!(tp->t_flags & TF_RCVD_CC))
485 * We can only get here in T/TCP's SYN_SENT* state, when
486 * we're a sending a non-SYN segment without waiting for
487 * the ACK of our SYN. A check above assures that we only
488 * do this if our peer understands T/TCP.
491 opt[optlen++] = TCPOPT_NOP;
492 opt[optlen++] = TCPOPT_NOP;
493 opt[optlen++] = TCPOPT_CC;
494 opt[optlen++] = TCPOLEN_CC;
495 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
501 * This is our initial SYN, check whether we have to use
505 opt[optlen++] = TCPOPT_NOP;
506 opt[optlen++] = TCPOPT_NOP;
507 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ?
508 TCPOPT_CCNEW : TCPOPT_CC;
509 opt[optlen++] = TCPOLEN_CC;
510 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
515 * This is a SYN,ACK; send CC and CC.echo if we received
518 case (TH_SYN|TH_ACK):
519 if (tp->t_flags & TF_RCVD_CC) {
520 opt[optlen++] = TCPOPT_NOP;
521 opt[optlen++] = TCPOPT_NOP;
522 opt[optlen++] = TCPOPT_CC;
523 opt[optlen++] = TCPOLEN_CC;
524 *(u_int32_t *)&opt[optlen] =
527 opt[optlen++] = TCPOPT_NOP;
528 opt[optlen++] = TCPOPT_NOP;
529 opt[optlen++] = TCPOPT_CCECHO;
530 opt[optlen++] = TCPOLEN_CC;
531 *(u_int32_t *)&opt[optlen] =
543 ipoptlen = ip6_optlen(tp->t_inpcb);
547 if (tp->t_inpcb->inp_options) {
548 ipoptlen = tp->t_inpcb->inp_options->m_len -
549 offsetof(struct ipoption, ipopt_list);
555 ipoptlen += ipsec_hdrsiz_tcp(tp);
559 * Adjust data length if insertion of options will
560 * bump the packet length beyond the t_maxopd length.
561 * Clear the FIN bit because we cut off the tail of
564 if (len + optlen + ipoptlen > tp->t_maxopd) {
566 * If there is still more to send, don't close the connection.
569 len = tp->t_maxopd - optlen - ipoptlen;
573 /*#ifdef DIAGNOSTIC*/
575 if (max_linkhdr + hdrlen > MCLBYTES)
576 panic("tcphdr too big");
578 if (max_linkhdr + hdrlen > MHLEN)
579 panic("tcphdr too big");
584 * Grab a header mbuf, attaching a copy of data to
585 * be transmitted, and initialize the header from
586 * the template for sends on this connection.
589 if (tp->t_force && len == 1)
590 tcpstat.tcps_sndprobe++;
591 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
592 tcpstat.tcps_sndrexmitpack++;
593 tcpstat.tcps_sndrexmitbyte += len;
595 tcpstat.tcps_sndpack++;
596 tcpstat.tcps_sndbyte += len;
599 if ((m = m_copypack(so->so_snd.sb_mb, off,
600 (int)len, max_linkhdr + hdrlen)) == 0) {
605 * m_copypack left space for our hdr; use it.
610 MGETHDR(m, M_DONTWAIT, MT_HEADER);
616 if (MHLEN < hdrlen + max_linkhdr) {
617 MCLGET(m, M_DONTWAIT);
618 if ((m->m_flags & M_EXT) == 0) {
625 m->m_data += max_linkhdr;
627 if (len <= MHLEN - hdrlen - max_linkhdr) {
628 m_copydata(so->so_snd.sb_mb, off, (int) len,
629 mtod(m, caddr_t) + hdrlen);
632 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
633 if (m->m_next == 0) {
641 * If we're sending everything we've got, set PUSH.
642 * (This will keep happy those implementations which only
643 * give data to the user when a buffer fills or
646 if (off + len == so->so_snd.sb_cc)
649 if (tp->t_flags & TF_ACKNOW)
650 tcpstat.tcps_sndacks++;
651 else if (flags & (TH_SYN|TH_FIN|TH_RST))
652 tcpstat.tcps_sndctrl++;
653 else if (SEQ_GT(tp->snd_up, tp->snd_una))
654 tcpstat.tcps_sndurg++;
656 tcpstat.tcps_sndwinup++;
658 MGETHDR(m, M_DONTWAIT, MT_HEADER);
664 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
669 m->m_data += max_linkhdr;
672 m->m_pkthdr.rcvif = (struct ifnet *)0;
675 ip6 = mtod(m, struct ip6_hdr *);
676 th = (struct tcphdr *)(ip6 + 1);
677 tcp_fillheaders(tp, ip6, th);
681 ip = mtod(m, struct ip *);
682 ipov = (struct ipovly *)ip;
683 th = (struct tcphdr *)(ip + 1);
684 /* this picks up the pseudo header (w/o the length) */
685 tcp_fillheaders(tp, ip, th);
689 * Fill in fields, remembering maximum advertised
690 * window for use in delaying messages about window sizes.
691 * If resending a FIN, be sure not to use a new sequence number.
693 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
694 tp->snd_nxt == tp->snd_max)
697 * If we are doing retransmissions, then snd_nxt will
698 * not reflect the first unsent octet. For ACK only
699 * packets, we do not want the sequence number of the
700 * retransmitted packet, we want the sequence number
701 * of the next unsent octet. So, if there is no data
702 * (and no SYN or FIN), use snd_max instead of snd_nxt
703 * when filling in ti_seq. But if we are in persist
704 * state, snd_max might reflect one byte beyond the
705 * right edge of the window, so use snd_nxt in that
706 * case, since we know we aren't doing a retransmission.
707 * (retransmit and persist are mutually exclusive...)
709 if (len || (flags & (TH_SYN|TH_FIN))
710 || callout_active(tp->tt_persist))
711 th->th_seq = htonl(tp->snd_nxt);
713 th->th_seq = htonl(tp->snd_max);
714 th->th_ack = htonl(tp->rcv_nxt);
716 bcopy(opt, th + 1, optlen);
717 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
719 th->th_flags = flags;
721 * Calculate receive window. Don't shrink window,
722 * but avoid silly window syndrome.
724 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
726 if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
727 win = (long)(tp->rcv_adv - tp->rcv_nxt);
728 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
729 win = (long)TCP_MAXWIN << tp->rcv_scale;
730 th->th_win = htons((u_short) (win>>tp->rcv_scale));
733 * Adjust the RXWIN0SENT flag - indicate that we have advertised
734 * a 0 window. This may cause the remote transmitter to stall. This
735 * flag tells soreceive() to disable delayed acknowledgements when
736 * draining the buffer. This can occur if the receiver is attempting
737 * to read more data then can be buffered prior to transmitting on
741 tp->t_flags |= TF_RXWIN0SENT;
743 tp->t_flags &= ~TF_RXWIN0SENT;
745 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
746 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
747 th->th_flags |= TH_URG;
750 * If no urgent pointer to send, then we pull
751 * the urgent pointer to the left edge of the send window
752 * so that it doesn't drift into the send window on sequence
755 tp->snd_up = tp->snd_una; /* drag it along */
758 * Put TCP length in extended header, and then
759 * checksum extended header and data.
761 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
765 * ip6_plen is not need to be filled now, and will be filled
768 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
769 sizeof(struct tcphdr) + optlen + len);
773 m->m_pkthdr.csum_flags = CSUM_TCP;
774 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
776 th->th_sum = in_addword(th->th_sum,
777 htons((u_short)(optlen + len)));
779 /* IP version must be set here for ipv4/ipv6 checking later */
780 KASSERT(ip->ip_v == IPVERSION,
781 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
785 * In transmit state, time the transmission and arrange for
786 * the retransmit. In persist state, just set snd_max.
788 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
789 tcp_seq startseq = tp->snd_nxt;
792 * Advance snd_nxt over sequence space of this segment.
794 if (flags & (TH_SYN|TH_FIN)) {
797 if (flags & TH_FIN) {
799 tp->t_flags |= TF_SENTFIN;
803 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
804 tp->snd_max = tp->snd_nxt;
806 * Time this transmission if not a retransmission and
807 * not currently timing anything.
809 if (tp->t_rtttime == 0) {
810 tp->t_rtttime = ticks;
811 tp->t_rtseq = startseq;
812 tcpstat.tcps_segstimed++;
817 * Set retransmit timer if not currently set,
818 * and not doing a pure ack or a keep-alive probe.
819 * Initial value for retransmit timer is smoothed
820 * round-trip time + 2 * round-trip time variance.
821 * Initialize shift counter which is used for backoff
822 * of retransmit time.
824 if (!callout_active(tp->tt_rexmt) &&
825 tp->snd_nxt != tp->snd_una) {
826 if (callout_active(tp->tt_persist)) {
827 callout_stop(tp->tt_persist);
830 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
831 tcp_timer_rexmt, tp);
835 * Persist case, update snd_max but since we are in
836 * persist mode (no window) we do not update snd_nxt.
841 if (flags & TH_FIN) {
843 tp->t_flags |= TF_SENTFIN;
845 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
846 tp->snd_max = tp->snd_nxt + xlen;
853 if (so->so_options & SO_DEBUG)
854 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
858 * Fill in IP length and desired time to live and
859 * send to IP level. There should be a better way
860 * to handle ttl and tos; we could keep them in
861 * the template, but need a way to checksum without them.
864 * m->m_pkthdr.len should have been set before cksum calcuration,
865 * because in6_cksum() need it.
870 * we separately set hoplimit for every segment, since the
871 * user might want to change the value via setsockopt.
872 * Also, desired default hop limit might be changed via
873 * Neighbor Discovery.
875 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb,
876 tp->t_inpcb->in6p_route.ro_rt ?
877 tp->t_inpcb->in6p_route.ro_rt->rt_ifp
880 /* TODO: IPv6 IP6TOS_ECT bit on */
881 error = ip6_output(m,
882 tp->t_inpcb->in6p_outputopts,
883 &tp->t_inpcb->in6p_route,
884 (so->so_options & SO_DONTROUTE), NULL, NULL,
890 ip->ip_len = m->m_pkthdr.len;
892 if (INP_CHECK_SOCKAF(so, AF_INET6))
893 ip->ip_ttl = in6_selecthlim(tp->t_inpcb,
894 tp->t_inpcb->in6p_route.ro_rt ?
895 tp->t_inpcb->in6p_route.ro_rt->rt_ifp
899 ip->ip_ttl = tp->t_inpcb->inp_ip_ttl; /* XXX */
900 ip->ip_tos = tp->t_inpcb->inp_ip_tos; /* XXX */
902 * See if we should do MTU discovery. We do it only if the following
904 * 1) we have a valid route to the destination
905 * 2) the MTU is not locked (if it is, then discovery has been
908 if (path_mtu_discovery
909 && (rt = tp->t_inpcb->inp_route.ro_rt)
910 && rt->rt_flags & RTF_UP
911 && !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
914 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
915 (so->so_options & SO_DONTROUTE), 0, tp->t_inpcb);
920 * We know that the packet was lost, so back out the
921 * sequence number advance, if any.
923 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
925 * No need to check for TH_FIN here because
926 * the TF_SENTFIN flag handles that case.
928 if ((flags & TH_SYN) == 0)
933 if (error == ENOBUFS) {
934 if (!callout_active(tp->tt_rexmt) &&
935 !callout_active(tp->tt_persist))
936 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
937 tcp_timer_rexmt, tp);
938 tcp_quench(tp->t_inpcb, 0);
941 if (error == EMSGSIZE) {
943 * ip_output() will have already fixed the route
944 * for us. tcp_mtudisc() will, as its last action,
945 * initiate retransmission, so it is important to
948 tcp_mtudisc(tp->t_inpcb, 0);
951 if ((error == EHOSTUNREACH || error == ENETDOWN)
952 && TCPS_HAVERCVDSYN(tp->t_state)) {
953 tp->t_softerror = error;
958 tcpstat.tcps_sndtotal++;
961 * Data sent (as far as we can tell).
962 * If this advertises a larger window than any other segment,
963 * then remember the size of the advertised window.
964 * Any pending ACK has now been sent.
966 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
967 tp->rcv_adv = tp->rcv_nxt + win;
968 tp->last_ack_sent = tp->rcv_nxt;
969 tp->t_flags &= ~TF_ACKNOW;
970 if (tcp_delack_enabled)
971 callout_stop(tp->tt_delack);
974 * This completely breaks TCP if newreno is turned on. What happens
975 * is that if delayed-acks are turned on on the receiver, this code
976 * on the transmitter effectively destroys the TCP window, forcing
977 * it to four packets (1.5Kx4 = 6K window).
979 if (sendalot && (!tcp_do_newreno || --maxburst))
989 register struct tcpcb *tp;
991 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
994 if (callout_active(tp->tt_rexmt))
995 panic("tcp_setpersist: retransmit pending");
997 * Start/restart persistance timer.
999 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
1000 TCPTV_PERSMIN, TCPTV_PERSMAX);
1001 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp);
1002 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)