2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
37 * License terms: all terms for the DragonFly license above plus the following:
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
53 * The Regents of the University of California. All rights reserved.
55 * Redistribution and use in source and binary forms, with or without
56 * modification, are permitted provided that the following conditions
58 * 1. Redistributions of source code must retain the above copyright
59 * notice, this list of conditions and the following disclaimer.
60 * 2. Redistributions in binary form must reproduce the above copyright
61 * notice, this list of conditions and the following disclaimer in the
62 * documentation and/or other materials provided with the distribution.
63 * 3. All advertising materials mentioning features or use of this software
64 * must display the following acknowledgement:
65 * This product includes software developed by the University of
66 * California, Berkeley and its contributors.
67 * 4. Neither the name of the University nor the names of its contributors
68 * may be used to endorse or promote products derived from this software
69 * without specific prior written permission.
71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
83 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95
84 * $FreeBSD: src/sys/netinet/tcp_output.c,v 1.39.2.20 2003/01/29 22:45:36 hsu Exp $
85 * $DragonFly: src/sys/netinet/tcp_output.c,v 1.17 2004/08/03 00:25:54 dillon Exp $
88 #include "opt_inet6.h"
89 #include "opt_ipsec.h"
90 #include "opt_tcpdebug.h"
92 #include <sys/param.h>
93 #include <sys/systm.h>
94 #include <sys/kernel.h>
95 #include <sys/sysctl.h>
97 #include <sys/domain.h>
98 #include <sys/protosw.h>
99 #include <sys/socket.h>
100 #include <sys/socketvar.h>
101 #include <sys/in_cksum.h>
102 #include <sys/thread.h>
103 #include <sys/globaldata.h>
105 #include <net/route.h>
107 #include <netinet/in.h>
108 #include <netinet/in_systm.h>
109 #include <netinet/ip.h>
110 #include <netinet/in_pcb.h>
111 #include <netinet/ip_var.h>
112 #include <netinet6/in6_pcb.h>
113 #include <netinet/ip6.h>
114 #include <netinet6/ip6_var.h>
115 #include <netinet/tcp.h>
117 #include <netinet/tcp_fsm.h>
118 #include <netinet/tcp_seq.h>
119 #include <netinet/tcp_timer.h>
120 #include <netinet/tcp_var.h>
121 #include <netinet/tcpip.h>
123 #include <netinet/tcp_debug.h>
127 #include <netinet6/ipsec.h>
131 #include <netipsec/ipsec.h>
133 #endif /*FAST_IPSEC*/
136 extern struct mbuf *m_copypack();
139 int path_mtu_discovery = 1;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
141 &path_mtu_discovery, 1, "Enable Path MTU Discovery");
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
145 &ss_fltsz, 1, "Slow start flight size");
147 int ss_fltsz_local = 4;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
149 &ss_fltsz_local, 1, "Slow start flight size for local networks");
151 static int avoid_pure_win_update = 1;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, avoid_pure_win_update, CTLFLAG_RW,
153 &avoid_pure_win_update, 1, "Avoid pure window updates when possible");
156 * Tcp output routine: figure out what should be sent and send it.
162 struct inpcb * const inp = tp->t_inpcb;
163 struct socket *so = inp->inp_socket;
164 long len, recvwin, sendwin;
165 int off, flags, error;
167 struct ip *ip = NULL;
168 struct ipovly *ipov = NULL;
170 u_char opt[TCP_MAXOLEN];
171 unsigned ipoptlen, optlen, hdrlen;
173 struct ip6_hdr *ip6 = NULL;
175 const boolean_t isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
177 const boolean_t isipv6 = FALSE;
179 struct rmxp_tao *taop;
182 * Determine length of data that should be transmitted,
183 * and flags that will be used.
184 * If there is some data or critical controls (SYN, RST)
185 * to send, then transmit; otherwise, investigate further.
187 if ((tp->snd_max == tp->snd_una) &&
188 (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
190 * We have been idle for "a while" and no acks are
191 * expected to clock out any data we send --
192 * slow start to get ack "clock" running again.
194 * Set the slow-start flight size depending on whether
195 * this is a local network or not.
197 if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
198 (!isipv6 && in_localaddr(inp->inp_faddr)))
199 tp->snd_cwnd = tp->t_maxseg * ss_fltsz_local;
201 tp->snd_cwnd = tp->t_maxseg * ss_fltsz;
203 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
204 if (idle && (tp->t_flags & TF_MORETOCOME))
205 tp->t_flags |= TF_LASTIDLE;
207 tp->t_flags &= ~TF_LASTIDLE;
211 off = tp->snd_nxt - tp->snd_una;
212 sendwin = min(tp->snd_wnd, tp->snd_cwnd);
213 sendwin = min(sendwin, tp->snd_bwnd);
215 flags = tcp_outflags[tp->t_state];
217 * Get standard flags, and add SYN or FIN if requested by 'hidden'
220 if (tp->t_flags & TF_NEEDFIN)
222 if (tp->t_flags & TF_NEEDSYN)
226 * If in persist timeout with window of 0, send 1 byte.
227 * Otherwise, if window is small but nonzero
228 * and timer expired, we will send what we can
229 * and go to transmit state.
231 if (tp->t_flags & TF_FORCE) {
234 * If we still have some data to send, then
235 * clear the FIN bit. Usually this would
236 * happen below when it realizes that we
237 * aren't sending all the data. However,
238 * if we have exactly 1 byte of unsent data,
239 * then it won't clear the FIN bit below,
240 * and if we are in persist state, we wind
241 * up sending the packet without recording
242 * that we sent the FIN bit.
244 * We can't just blindly clear the FIN bit,
245 * because if we don't have any more data
246 * to send then the probe will be the FIN
249 if (off < so->so_snd.sb_cc)
253 callout_stop(tp->tt_persist);
259 * If snd_nxt == snd_max and we have transmitted a FIN, the
260 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
261 * a negative length. This can also occur when TCP opens up
262 * its congestion window while receiving additional duplicate
263 * acks after fast-retransmit because TCP will reset snd_nxt
264 * to snd_max after the fast-retransmit.
266 * In the normal retransmit-FIN-only case, however, snd_nxt will
267 * be set to snd_una, the offset will be 0, and the length may
270 len = (long)ulmin(so->so_snd.sb_cc, sendwin) - off;
273 * Lop off SYN bit if it has already been sent. However, if this
274 * is SYN-SENT state and if segment contains data and if we don't
275 * know that foreign host supports TAO, suppress sending segment.
277 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
280 if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
281 ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL ||
282 taop->tao_ccsent == 0))
287 * Be careful not to send data and/or FIN on SYN segments
288 * in cases when no CC option will be sent.
289 * This measure is needed to prevent interoperability problems
290 * with not fully conformant TCP implementations.
292 if ((flags & TH_SYN) &&
293 ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) ||
294 ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) {
301 * If FIN has been sent but not acked,
302 * but we haven't been called to retransmit,
303 * len will be < 0. Otherwise, window shrank
304 * after we sent into it. If window shrank to 0,
305 * cancel pending retransmit, pull snd_nxt back
306 * to (closed) window, and set the persist timer
307 * if it isn't already going. If the window didn't
308 * close completely, just wait for an ACK.
312 callout_stop(tp->tt_rexmt);
314 tp->snd_nxt = tp->snd_una;
315 if (!callout_active(tp->tt_persist))
321 * len will be >= 0 after this point. Truncate to the maximum
322 * segment length and ensure that FIN is removed if the length
323 * no longer contains the last data byte.
325 if (len > tp->t_maxseg) {
329 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
332 recvwin = sbspace(&so->so_rcv);
335 * Sender silly window avoidance. We transmit under the following
336 * conditions when len is non-zero:
338 * - We have a full segment
339 * - This is the last buffer in a write()/send() and we are
340 * either idle or running NODELAY
341 * - we've timed out (e.g. persist timer)
342 * - we have more then 1/2 the maximum send window's worth of
343 * data (receiver may be limited the window size)
344 * - we need to retransmit
347 if (len == tp->t_maxseg)
350 * NOTE! on localhost connections an 'ack' from the remote
351 * end may occur synchronously with the output and cause
352 * us to flush a buffer queued with moretocome. XXX
354 * note: the len + off check is almost certainly unnecessary.
356 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
357 (idle || (tp->t_flags & TF_NODELAY)) &&
358 len + off >= so->so_snd.sb_cc &&
359 !(tp->t_flags & TF_NOPUSH)) {
362 if (tp->t_flags & TF_FORCE) /* typ. timeout case */
364 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
366 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
371 * Compare available window to amount of window
372 * known to peer (as advertised window less
373 * next expected input). If the difference is at least two
374 * max size segments, or at least 50% of the maximum possible
375 * window, then want to send a window update to peer.
379 * "adv" is the amount we can increase the window,
380 * taking into account that we are limited by
381 * TCP_MAXWIN << tp->rcv_scale.
383 long adv = min(recvwin, (long)TCP_MAXWIN << tp->rcv_scale) -
384 (tp->rcv_adv - tp->rcv_nxt);
387 * This ack case typically occurs when the user has drained
388 * the TCP socket buffer sufficiently to warrent an ack
389 * containing a 'pure window update'... that is, an ack that
390 * ONLY updates the tcp window.
392 * It is unclear why we would need to do a pure window update
393 * past 2 segments if we are going to do one at 1/2 the high
394 * water mark anyway, especially since under normal conditions
395 * the user program will drain the socket buffer quickly.
396 * The 2-segment pure window update will often add a large
397 * number of extra, unnecessary acks to the stream.
399 * avoid_pure_win_update now defaults to 1.
401 if (avoid_pure_win_update == 0) {
402 if (adv >= (long) (2 * tp->t_maxseg))
405 if (2 * adv >= (long) so->so_rcv.sb_hiwat)
410 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
411 * is also a catch-all for the retransmit timer timeout case.
413 if (tp->t_flags & TF_ACKNOW)
415 if ((flags & TH_RST) ||
416 ((flags & TH_SYN) && !(tp->t_flags & TF_NEEDSYN)))
418 if (SEQ_GT(tp->snd_up, tp->snd_una))
421 * If our state indicates that FIN should be sent
422 * and we have not yet done so, then we need to send.
424 if (flags & TH_FIN &&
425 (!(tp->t_flags & TF_SENTFIN) || tp->snd_nxt == tp->snd_una))
429 * TCP window updates are not reliable, rather a polling protocol
430 * using ``persist'' packets is used to insure receipt of window
431 * updates. The three ``states'' for the output side are:
432 * idle not doing retransmits or persists
433 * persisting to move a small or zero window
434 * (re)transmitting and thereby not persisting
436 * callout_active(tp->tt_persist)
437 * is true when we are in persist state.
438 * The TF_FORCE flag in tp->t_flags
439 * is set when we are called to send a persist packet.
440 * callout_active(tp->tt_rexmt)
441 * is set when we are retransmitting
442 * The output side is idle when both timers are zero.
444 * If send window is too small, there is data to transmit, and no
445 * retransmit or persist is pending, then go to persist state.
446 * If nothing happens soon, send when timer expires:
447 * if window is nonzero, transmit what we can,
448 * otherwise force out a byte.
450 if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) &&
451 !callout_active(tp->tt_persist)) {
457 * No reason to send a segment, just return.
463 * Before ESTABLISHED, force sending of initial options
464 * unless TCP set not to do any options.
465 * NOTE: we assume that the IP/TCP header plus TCP options
466 * always fit in a single mbuf, leaving room for a maximum
468 * max_linkhdr + sizeof(struct tcpiphdr) + optlen <= MCLBYTES
472 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
474 hdrlen = sizeof(struct tcpiphdr);
475 if (flags & TH_SYN) {
476 tp->snd_nxt = tp->iss;
477 if (!(tp->t_flags & TF_NOOPT)) {
480 opt[0] = TCPOPT_MAXSEG;
481 opt[1] = TCPOLEN_MAXSEG;
482 mss = htons((u_short) tcp_mssopt(tp));
483 (void)memcpy(opt + 2, &mss, sizeof(mss));
484 optlen = TCPOLEN_MAXSEG;
486 if ((tp->t_flags & TF_REQ_SCALE) &&
487 (!(flags & TH_ACK) ||
488 (tp->t_flags & TF_RCVD_SCALE))) {
489 *((u_int32_t *)(opt + optlen)) = htonl(
491 TCPOPT_WINDOW << 16 |
492 TCPOLEN_WINDOW << 8 |
493 tp->request_r_scale);
500 * Send a timestamp and echo-reply if this is a SYN and our side
501 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
502 * and our peer have sent timestamps in our SYN's.
504 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
506 (!(flags & TH_ACK) || (tp->t_flags & TF_RCVD_TSTMP))) {
507 u_int32_t *lp = (u_int32_t *)(opt + optlen);
509 /* Form timestamp option as shown in appendix A of RFC 1323. */
510 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
511 *lp++ = htonl(ticks);
512 *lp = htonl(tp->ts_recent);
513 optlen += TCPOLEN_TSTAMP_APPA;
517 * Send `CC-family' options if our side wants to use them (TF_REQ_CC),
518 * options are allowed (!TF_NOOPT) and it's not a RST.
520 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
522 switch (flags & (TH_SYN|TH_ACK)) {
524 * This is a normal ACK, send CC if we received CC before
528 if (!(tp->t_flags & TF_RCVD_CC))
533 * We can only get here in T/TCP's SYN_SENT* state, when
534 * we're a sending a non-SYN segment without waiting for
535 * the ACK of our SYN. A check above assures that we only
536 * do this if our peer understands T/TCP.
539 opt[optlen++] = TCPOPT_NOP;
540 opt[optlen++] = TCPOPT_NOP;
541 opt[optlen++] = TCPOPT_CC;
542 opt[optlen++] = TCPOLEN_CC;
543 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
549 * This is our initial SYN, check whether we have to use
553 opt[optlen++] = TCPOPT_NOP;
554 opt[optlen++] = TCPOPT_NOP;
555 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ?
556 TCPOPT_CCNEW : TCPOPT_CC;
557 opt[optlen++] = TCPOLEN_CC;
558 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
563 * This is a SYN,ACK; send CC and CC.echo if we received
566 case (TH_SYN|TH_ACK):
567 if (tp->t_flags & TF_RCVD_CC) {
568 opt[optlen++] = TCPOPT_NOP;
569 opt[optlen++] = TCPOPT_NOP;
570 opt[optlen++] = TCPOPT_CC;
571 opt[optlen++] = TCPOLEN_CC;
572 *(u_int32_t *)&opt[optlen] =
575 opt[optlen++] = TCPOPT_NOP;
576 opt[optlen++] = TCPOPT_NOP;
577 opt[optlen++] = TCPOPT_CCECHO;
578 opt[optlen++] = TCPOLEN_CC;
579 *(u_int32_t *)&opt[optlen] =
590 ipoptlen = ip6_optlen(inp);
592 if (inp->inp_options) {
593 ipoptlen = inp->inp_options->m_len -
594 offsetof(struct ipoption, ipopt_list);
600 ipoptlen += ipsec_hdrsiz_tcp(tp);
604 * Adjust data length if insertion of options will
605 * bump the packet length beyond the t_maxopd length.
606 * Clear the FIN bit because we cut off the tail of
609 if (len + optlen + ipoptlen > tp->t_maxopd) {
611 * If there is still more to send, don't close the connection.
614 len = tp->t_maxopd - optlen - ipoptlen;
619 KASSERT(max_linkhdr + hdrlen <= MCLBYTES, ("tcphdr too big"));
621 KASSERT(max_linkhdr + hdrlen <= MHLEN, ("tcphdr too big"));
625 * Grab a header mbuf, attaching a copy of data to
626 * be transmitted, and initialize the header from
627 * the template for sends on this connection.
630 if ((tp->t_flags & TF_FORCE) && len == 1)
631 tcpstat.tcps_sndprobe++;
632 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
633 tcpstat.tcps_sndrexmitpack++;
634 tcpstat.tcps_sndrexmitbyte += len;
636 tcpstat.tcps_sndpack++;
637 tcpstat.tcps_sndbyte += len;
640 if ((m = m_copypack(so->so_snd.sb_mb, off,
641 (int)len, max_linkhdr + hdrlen)) == 0) {
646 * m_copypack left space for our hdr; use it.
651 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
657 if (MHLEN < hdrlen + max_linkhdr) {
658 MCLGET(m, MB_DONTWAIT);
659 if (!(m->m_flags & M_EXT)) {
666 m->m_data += max_linkhdr;
668 if (len <= MHLEN - hdrlen - max_linkhdr) {
669 m_copydata(so->so_snd.sb_mb, off, (int) len,
670 mtod(m, caddr_t) + hdrlen);
673 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
674 if (m->m_next == 0) {
682 * If we're sending everything we've got, set PUSH.
683 * (This will keep happy those implementations which only
684 * give data to the user when a buffer fills or
687 if (off + len == so->so_snd.sb_cc)
690 if (tp->t_flags & TF_ACKNOW)
691 tcpstat.tcps_sndacks++;
692 else if (flags & (TH_SYN|TH_FIN|TH_RST))
693 tcpstat.tcps_sndctrl++;
694 else if (SEQ_GT(tp->snd_up, tp->snd_una))
695 tcpstat.tcps_sndurg++;
697 tcpstat.tcps_sndwinup++;
699 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
705 (hdrlen + max_linkhdr > MHLEN) && hdrlen <= MHLEN)
708 m->m_data += max_linkhdr;
711 m->m_pkthdr.rcvif = (struct ifnet *)0;
713 ip6 = mtod(m, struct ip6_hdr *);
714 th = (struct tcphdr *)(ip6 + 1);
715 tcp_fillheaders(tp, ip6, th);
717 ip = mtod(m, struct ip *);
718 ipov = (struct ipovly *)ip;
719 th = (struct tcphdr *)(ip + 1);
720 /* this picks up the pseudo header (w/o the length) */
721 tcp_fillheaders(tp, ip, th);
725 * Fill in fields, remembering maximum advertised
726 * window for use in delaying messages about window sizes.
727 * If resending a FIN, be sure not to use a new sequence number.
729 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
730 tp->snd_nxt == tp->snd_max)
733 * If we are doing retransmissions, then snd_nxt will
734 * not reflect the first unsent octet. For ACK only
735 * packets, we do not want the sequence number of the
736 * retransmitted packet, we want the sequence number
737 * of the next unsent octet. So, if there is no data
738 * (and no SYN or FIN), use snd_max instead of snd_nxt
739 * when filling in ti_seq. But if we are in persist
740 * state, snd_max might reflect one byte beyond the
741 * right edge of the window, so use snd_nxt in that
742 * case, since we know we aren't doing a retransmission.
743 * (retransmit and persist are mutually exclusive...)
745 if (len || (flags & (TH_SYN|TH_FIN)) || callout_active(tp->tt_persist))
746 th->th_seq = htonl(tp->snd_nxt);
748 th->th_seq = htonl(tp->snd_max);
749 th->th_ack = htonl(tp->rcv_nxt);
751 bcopy(opt, th + 1, optlen);
752 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
754 th->th_flags = flags;
756 * Calculate receive window. Don't shrink window,
757 * but avoid silly window syndrome.
759 if (recvwin < (long)(so->so_rcv.sb_hiwat / 4) &&
760 recvwin < (long)tp->t_maxseg)
762 if (recvwin < (long)(tp->rcv_adv - tp->rcv_nxt))
763 recvwin = (long)(tp->rcv_adv - tp->rcv_nxt);
764 if (recvwin > (long)TCP_MAXWIN << tp->rcv_scale)
765 recvwin = (long)TCP_MAXWIN << tp->rcv_scale;
766 th->th_win = htons((u_short) (recvwin>>tp->rcv_scale));
769 * Adjust the RXWIN0SENT flag - indicate that we have advertised
770 * a 0 window. This may cause the remote transmitter to stall. This
771 * flag tells soreceive() to disable delayed acknowledgements when
772 * draining the buffer. This can occur if the receiver is attempting
773 * to read more data then can be buffered prior to transmitting on
777 tp->t_flags |= TF_RXWIN0SENT;
779 tp->t_flags &= ~TF_RXWIN0SENT;
781 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
782 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
783 th->th_flags |= TH_URG;
786 * If no urgent pointer to send, then we pull
787 * the urgent pointer to the left edge of the send window
788 * so that it doesn't drift into the send window on sequence
791 tp->snd_up = tp->snd_una; /* drag it along */
795 * Put TCP length in extended header, and then
796 * checksum extended header and data.
798 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
801 * ip6_plen is not need to be filled now, and will be filled
804 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
805 sizeof(struct tcphdr) + optlen + len);
807 m->m_pkthdr.csum_flags = CSUM_TCP;
808 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
810 th->th_sum = in_addword(th->th_sum,
811 htons((u_short)(optlen + len)));
813 /* IP version must be set here for ipv4/ipv6 checking later */
814 KASSERT(ip->ip_v == IPVERSION,
815 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
819 * In transmit state, time the transmission and arrange for
820 * the retransmit. In persist state, just set snd_max.
822 if (!(tp->t_flags & TF_FORCE) || !callout_active(tp->tt_persist)) {
823 tcp_seq startseq = tp->snd_nxt;
826 * Advance snd_nxt over sequence space of this segment.
828 if (flags & (TH_SYN|TH_FIN)) {
831 if (flags & TH_FIN) {
833 tp->t_flags |= TF_SENTFIN;
837 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
838 tp->snd_max = tp->snd_nxt;
840 * Time this transmission if not a retransmission and
841 * not currently timing anything.
843 if (tp->t_rtttime == 0) {
844 tp->t_rtttime = ticks;
845 tp->t_rtseq = startseq;
846 tcpstat.tcps_segstimed++;
851 * Set retransmit timer if not currently set,
852 * and not doing a pure ack or a keep-alive probe.
853 * Initial value for retransmit timer is smoothed
854 * round-trip time + 2 * round-trip time variance.
855 * Initialize shift counter which is used for backoff
856 * of retransmit time.
858 if (!callout_active(tp->tt_rexmt) &&
859 tp->snd_nxt != tp->snd_una) {
860 if (callout_active(tp->tt_persist)) {
861 callout_stop(tp->tt_persist);
864 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
865 tcp_timer_rexmt, tp);
869 * Persist case, update snd_max but since we are in
870 * persist mode (no window) we do not update snd_nxt.
875 if (flags & TH_FIN) {
877 tp->t_flags |= TF_SENTFIN;
879 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
880 tp->snd_max = tp->snd_nxt + xlen;
887 if (so->so_options & SO_DEBUG)
888 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
892 * Fill in IP length and desired time to live and
893 * send to IP level. There should be a better way
894 * to handle ttl and tos; we could keep them in
895 * the template, but need a way to checksum without them.
898 * m->m_pkthdr.len should have been set before cksum calcuration,
899 * because in6_cksum() need it.
903 * we separately set hoplimit for every segment, since the
904 * user might want to change the value via setsockopt.
905 * Also, desired default hop limit might be changed via
906 * Neighbor Discovery.
908 ip6->ip6_hlim = in6_selecthlim(inp,
909 (inp->in6p_route.ro_rt ?
910 inp->in6p_route.ro_rt->rt_ifp : NULL));
912 /* TODO: IPv6 IP6TOS_ECT bit on */
913 error = ip6_output(m, inp->in6p_outputopts, &inp->in6p_route,
914 (so->so_options & SO_DONTROUTE), NULL, NULL, inp);
917 ip->ip_len = m->m_pkthdr.len;
919 if (INP_CHECK_SOCKAF(so, AF_INET6))
920 ip->ip_ttl = in6_selecthlim(inp,
921 (inp->in6p_route.ro_rt ?
922 inp->in6p_route.ro_rt->rt_ifp : NULL));
925 ip->ip_ttl = inp->inp_ip_ttl; /* XXX */
927 ip->ip_tos = inp->inp_ip_tos; /* XXX */
929 * See if we should do MTU discovery.
930 * We do it only if the following are true:
931 * 1) we have a valid route to the destination
932 * 2) the MTU is not locked (if it is,
933 * then discovery has been disabled)
935 if (path_mtu_discovery &&
936 (rt = inp->inp_route.ro_rt) && (rt->rt_flags & RTF_UP) &&
937 !(rt->rt_rmx.rmx_locks & RTV_MTU))
940 error = ip_output(m, inp->inp_options, &inp->inp_route,
941 (so->so_options & SO_DONTROUTE), NULL, inp);
946 * We know that the packet was lost, so back out the
947 * sequence number advance, if any.
949 if (!(tp->t_flags & TF_FORCE) ||
950 !callout_active(tp->tt_persist)) {
952 * No need to check for TH_FIN here because
953 * the TF_SENTFIN flag handles that case.
955 if (!(flags & TH_SYN))
960 if (error == ENOBUFS) {
962 * If we can't send, make sure there is something
963 * to get us going again later. Persist state
964 * is not necessarily right, but it is close enough.
966 if (!callout_active(tp->tt_rexmt) &&
967 !callout_active(tp->tt_persist)) {
974 if (error == EMSGSIZE) {
976 * ip_output() will have already fixed the route
977 * for us. tcp_mtudisc() will, as its last action,
978 * initiate retransmission, so it is important to
984 if ((error == EHOSTUNREACH || error == ENETDOWN) &&
985 TCPS_HAVERCVDSYN(tp->t_state)) {
986 tp->t_softerror = error;
991 tcpstat.tcps_sndtotal++;
994 * Data sent (as far as we can tell).
995 * If this advertises a larger window than any other segment,
996 * then remember the size of the advertised window.
997 * Any pending ACK has now been sent.
999 if (recvwin > 0 && SEQ_GT(tp->rcv_nxt + recvwin, tp->rcv_adv))
1000 tp->rcv_adv = tp->rcv_nxt + recvwin;
1001 tp->last_ack_sent = tp->rcv_nxt;
1002 tp->t_flags &= ~TF_ACKNOW;
1003 if (tcp_delack_enabled)
1004 callout_stop(tp->tt_delack);
1014 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
1017 if (callout_active(tp->tt_rexmt))
1018 panic("tcp_setpersist: retransmit pending");
1020 * Start/restart persistance timer.
1022 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
1023 TCPTV_PERSMIN, TCPTV_PERSMAX);
1024 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp);
1025 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)