Register keyword removal
[dragonfly.git] / sys / netinet / tcp_output.c
CommitLineData
984263bc
MD
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: src/sys/netinet/tcp_output.c,v 1.39.2.20 2003/01/29 22:45:36 hsu Exp $
2256ba69 35 * $DragonFly: src/sys/netinet/tcp_output.c,v 1.3 2003/07/26 21:00:04 rob Exp $
984263bc
MD
36 */
37
38#include "opt_inet6.h"
39#include "opt_ipsec.h"
40#include "opt_tcpdebug.h"
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/kernel.h>
45#include <sys/sysctl.h>
46#include <sys/mbuf.h>
47#include <sys/domain.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#include <netinet/in_pcb.h>
58#include <netinet/ip_var.h>
59#ifdef INET6
60#include <netinet6/in6_pcb.h>
61#include <netinet/ip6.h>
62#include <netinet6/ip6_var.h>
63#endif
64#include <netinet/tcp.h>
65#define TCPOUTFLAGS
66#include <netinet/tcp_fsm.h>
67#include <netinet/tcp_seq.h>
68#include <netinet/tcp_timer.h>
69#include <netinet/tcp_var.h>
70#include <netinet/tcpip.h>
71#ifdef TCPDEBUG
72#include <netinet/tcp_debug.h>
73#endif
74
75#ifdef IPSEC
76#include <netinet6/ipsec.h>
77#endif /*IPSEC*/
78
79#ifdef FAST_IPSEC
80#include <netipsec/ipsec.h>
81#define IPSEC
82#endif /*FAST_IPSEC*/
83
84#include <machine/in_cksum.h>
85
86#ifdef notyet
87extern struct mbuf *m_copypack();
88#endif
89
90int path_mtu_discovery = 1;
91SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
92 &path_mtu_discovery, 1, "Enable Path MTU Discovery");
93
94int ss_fltsz = 1;
95SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
96 &ss_fltsz, 1, "Slow start flight size");
97
98int ss_fltsz_local = 4;
99SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
100 &ss_fltsz_local, 1, "Slow start flight size for local networks");
101
102int tcp_do_newreno = 1;
103SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno,
104 0, "Enable NewReno Algorithms");
105
106/*
107 * Tcp output routine: figure out what should be sent and send it.
108 */
109int
110tcp_output(tp)
2256ba69 111 struct tcpcb *tp;
984263bc 112{
2256ba69
RG
113 struct socket *so = tp->t_inpcb->inp_socket;
114 long len, win;
984263bc 115 int off, flags, error;
2256ba69 116 struct mbuf *m;
984263bc 117 struct ip *ip = NULL;
2256ba69 118 struct ipovly *ipov = NULL;
984263bc
MD
119#ifdef INET6
120 struct ip6_hdr *ip6 = NULL;
121#endif /* INET6 */
2256ba69 122 struct tcphdr *th;
984263bc
MD
123 u_char opt[TCP_MAXOLEN];
124 unsigned ipoptlen, optlen, hdrlen;
125 int idle, sendalot;
126#if 0
127 int maxburst = TCP_MAXBURST;
128#endif
129 struct rmxp_tao *taop;
130#ifdef INET6
131 int isipv6;
132#endif
133
134#ifdef INET6
135 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
136#endif
137
138 /*
139 * Determine length of data that should be transmitted,
140 * and flags that will be used.
141 * If there is some data or critical controls (SYN, RST)
142 * to send, then transmit; otherwise, investigate further.
143 */
144 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
145 if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
146 /*
147 * We have been idle for "a while" and no acks are
148 * expected to clock out any data we send --
149 * slow start to get ack "clock" running again.
150 *
151 * Set the slow-start flight size depending on whether
152 * this is a local network or not.
153 */
154 if (
155#ifdef INET6
156 (isipv6 && in6_localaddr(&tp->t_inpcb->in6p_faddr)) ||
157 (!isipv6 &&
158#endif
159 in_localaddr(tp->t_inpcb->inp_faddr)
160#ifdef INET6
161 )
162#endif
163 )
164 tp->snd_cwnd = tp->t_maxseg * ss_fltsz_local;
165 else
166 tp->snd_cwnd = tp->t_maxseg * ss_fltsz;
167 }
168 tp->t_flags &= ~TF_LASTIDLE;
169 if (idle) {
170 if (tp->t_flags & TF_MORETOCOME) {
171 tp->t_flags |= TF_LASTIDLE;
172 idle = 0;
173 }
174 }
175again:
176 sendalot = 0;
177 off = tp->snd_nxt - tp->snd_una;
178 win = min(tp->snd_wnd, tp->snd_cwnd);
179 win = min(win, tp->snd_bwnd);
180
181 flags = tcp_outflags[tp->t_state];
182 /*
183 * Get standard flags, and add SYN or FIN if requested by 'hidden'
184 * state flags.
185 */
186 if (tp->t_flags & TF_NEEDFIN)
187 flags |= TH_FIN;
188 if (tp->t_flags & TF_NEEDSYN)
189 flags |= TH_SYN;
190
191 /*
192 * If in persist timeout with window of 0, send 1 byte.
193 * Otherwise, if window is small but nonzero
194 * and timer expired, we will send what we can
195 * and go to transmit state.
196 */
197 if (tp->t_force) {
198 if (win == 0) {
199 /*
200 * If we still have some data to send, then
201 * clear the FIN bit. Usually this would
202 * happen below when it realizes that we
203 * aren't sending all the data. However,
204 * if we have exactly 1 byte of unsent data,
205 * then it won't clear the FIN bit below,
206 * and if we are in persist state, we wind
207 * up sending the packet without recording
208 * that we sent the FIN bit.
209 *
210 * We can't just blindly clear the FIN bit,
211 * because if we don't have any more data
212 * to send then the probe will be the FIN
213 * itself.
214 */
215 if (off < so->so_snd.sb_cc)
216 flags &= ~TH_FIN;
217 win = 1;
218 } else {
219 callout_stop(tp->tt_persist);
220 tp->t_rxtshift = 0;
221 }
222 }
223
224 /*
225 * If snd_nxt == snd_max and we have transmitted a FIN, the
226 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
227 * a negative length. This can also occur when tcp opens up
228 * its congestion window while receiving additional duplicate
229 * acks after fast-retransmit because TCP will reset snd_nxt
230 * to snd_max after the fast-retransmit.
231 *
232 * In the normal retransmit-FIN-only case, however, snd_nxt will
233 * be set to snd_una, the offset will be 0, and the length may
234 * wind up 0.
235 */
236 len = (long)ulmin(so->so_snd.sb_cc, win) - off;
237
238 taop = tcp_gettaocache(&tp->t_inpcb->inp_inc);
239
240 /*
241 * Lop off SYN bit if it has already been sent. However, if this
242 * is SYN-SENT state and if segment contains data and if we don't
243 * know that foreign host supports TAO, suppress sending segment.
244 */
245 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
246 flags &= ~TH_SYN;
247 off--, len++;
248 if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
249 (taop == NULL || taop->tao_ccsent == 0))
250 return 0;
251 }
252
253 /*
254 * Be careful not to send data and/or FIN on SYN segments
255 * in cases when no CC option will be sent.
256 * This measure is needed to prevent interoperability problems
257 * with not fully conformant TCP implementations.
258 */
259 if ((flags & TH_SYN) &&
260 ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) ||
261 ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) {
262 len = 0;
263 flags &= ~TH_FIN;
264 }
265
266 if (len < 0) {
267 /*
268 * If FIN has been sent but not acked,
269 * but we haven't been called to retransmit,
270 * len will be < 0. Otherwise, window shrank
271 * after we sent into it. If window shrank to 0,
272 * cancel pending retransmit, pull snd_nxt back
273 * to (closed) window, and set the persist timer
274 * if it isn't already going. If the window didn't
275 * close completely, just wait for an ACK.
276 */
277 len = 0;
278 if (win == 0) {
279 callout_stop(tp->tt_rexmt);
280 tp->t_rxtshift = 0;
281 tp->snd_nxt = tp->snd_una;
282 if (!callout_active(tp->tt_persist))
283 tcp_setpersist(tp);
284 }
285 }
286
287 /*
288 * len will be >= 0 after this point. Truncate to the maximum
289 * segment length and ensure that FIN is removed if the length
290 * no longer contains the last data byte.
291 */
292 if (len > tp->t_maxseg) {
293 len = tp->t_maxseg;
294 sendalot = 1;
295 }
296 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
297 flags &= ~TH_FIN;
298
299 win = sbspace(&so->so_rcv);
300
301 /*
302 * Sender silly window avoidance. We transmit under the following
303 * conditions when len is non-zero:
304 *
305 * - We have a full segment
306 * - This is the last buffer in a write()/send() and we are
307 * either idle or running NODELAY
308 * - we've timed out (e.g. persist timer)
309 * - we have more then 1/2 the maximum send window's worth of
310 * data (receiver may be limited the window size)
311 * - we need to retransmit
312 */
313 if (len) {
314 if (len == tp->t_maxseg)
315 goto send;
316 /*
317 * NOTE! on localhost connections an 'ack' from the remote
318 * end may occur synchronously with the output and cause
319 * us to flush a buffer queued with moretocome. XXX
320 *
321 * note: the len + off check is almost certainly unnecessary.
322 */
323 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
324 (idle || (tp->t_flags & TF_NODELAY)) &&
325 len + off >= so->so_snd.sb_cc &&
326 (tp->t_flags & TF_NOPUSH) == 0) {
327 goto send;
328 }
329 if (tp->t_force) /* typ. timeout case */
330 goto send;
331 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
332 goto send;
333 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
334 goto send;
335 }
336
337 /*
338 * Compare available window to amount of window
339 * known to peer (as advertised window less
340 * next expected input). If the difference is at least two
341 * max size segments, or at least 50% of the maximum possible
342 * window, then want to send a window update to peer.
343 */
344 if (win > 0) {
345 /*
346 * "adv" is the amount we can increase the window,
347 * taking into account that we are limited by
348 * TCP_MAXWIN << tp->rcv_scale.
349 */
350 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
351 (tp->rcv_adv - tp->rcv_nxt);
352
353 if (adv >= (long) (2 * tp->t_maxseg))
354 goto send;
355 if (2 * adv >= (long) so->so_rcv.sb_hiwat)
356 goto send;
357 }
358
359 /*
360 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
361 * is also a catch-all for the retransmit timer timeout case.
362 */
363 if (tp->t_flags & TF_ACKNOW)
364 goto send;
365 if ((flags & TH_RST) ||
366 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
367 goto send;
368 if (SEQ_GT(tp->snd_up, tp->snd_una))
369 goto send;
370 /*
371 * If our state indicates that FIN should be sent
372 * and we have not yet done so, then we need to send.
373 */
374 if (flags & TH_FIN &&
375 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
376 goto send;
377
378 /*
379 * TCP window updates are not reliable, rather a polling protocol
380 * using ``persist'' packets is used to insure receipt of window
381 * updates. The three ``states'' for the output side are:
382 * idle not doing retransmits or persists
383 * persisting to move a small or zero window
384 * (re)transmitting and thereby not persisting
385 *
386 * callout_active(tp->tt_persist)
387 * is true when we are in persist state.
388 * tp->t_force
389 * is set when we are called to send a persist packet.
390 * callout_active(tp->tt_rexmt)
391 * is set when we are retransmitting
392 * The output side is idle when both timers are zero.
393 *
394 * If send window is too small, there is data to transmit, and no
395 * retransmit or persist is pending, then go to persist state.
396 * If nothing happens soon, send when timer expires:
397 * if window is nonzero, transmit what we can,
398 * otherwise force out a byte.
399 */
400 if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) &&
401 !callout_active(tp->tt_persist)) {
402 tp->t_rxtshift = 0;
403 tcp_setpersist(tp);
404 }
405
406 /*
407 * No reason to send a segment, just return.
408 */
409 return (0);
410
411send:
412 /*
413 * Before ESTABLISHED, force sending of initial options
414 * unless TCP set not to do any options.
415 * NOTE: we assume that the IP/TCP header plus TCP options
416 * always fit in a single mbuf, leaving room for a maximum
417 * link header, i.e.
418 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
419 */
420 optlen = 0;
421#ifdef INET6
422 if (isipv6)
423 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
424 else
425#endif
426 hdrlen = sizeof (struct tcpiphdr);
427 if (flags & TH_SYN) {
428 tp->snd_nxt = tp->iss;
429 if ((tp->t_flags & TF_NOOPT) == 0) {
430 u_short mss;
431
432 opt[0] = TCPOPT_MAXSEG;
433 opt[1] = TCPOLEN_MAXSEG;
434 mss = htons((u_short) tcp_mssopt(tp));
435 (void)memcpy(opt + 2, &mss, sizeof(mss));
436 optlen = TCPOLEN_MAXSEG;
437
438 if ((tp->t_flags & TF_REQ_SCALE) &&
439 ((flags & TH_ACK) == 0 ||
440 (tp->t_flags & TF_RCVD_SCALE))) {
441 *((u_int32_t *)(opt + optlen)) = htonl(
442 TCPOPT_NOP << 24 |
443 TCPOPT_WINDOW << 16 |
444 TCPOLEN_WINDOW << 8 |
445 tp->request_r_scale);
446 optlen += 4;
447 }
448 }
449 }
450
451 /*
452 * Send a timestamp and echo-reply if this is a SYN and our side
453 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
454 * and our peer have sent timestamps in our SYN's.
455 */
456 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
457 (flags & TH_RST) == 0 &&
458 ((flags & TH_ACK) == 0 ||
459 (tp->t_flags & TF_RCVD_TSTMP))) {
460 u_int32_t *lp = (u_int32_t *)(opt + optlen);
461
462 /* Form timestamp option as shown in appendix A of RFC 1323. */
463 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
464 *lp++ = htonl(ticks);
465 *lp = htonl(tp->ts_recent);
466 optlen += TCPOLEN_TSTAMP_APPA;
467 }
468
469 /*
470 * Send `CC-family' options if our side wants to use them (TF_REQ_CC),
471 * options are allowed (!TF_NOOPT) and it's not a RST.
472 */
473 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
474 (flags & TH_RST) == 0) {
475 switch (flags & (TH_SYN|TH_ACK)) {
476 /*
477 * This is a normal ACK, send CC if we received CC before
478 * from our peer.
479 */
480 case TH_ACK:
481 if (!(tp->t_flags & TF_RCVD_CC))
482 break;
483 /*FALLTHROUGH*/
484
485 /*
486 * We can only get here in T/TCP's SYN_SENT* state, when
487 * we're a sending a non-SYN segment without waiting for
488 * the ACK of our SYN. A check above assures that we only
489 * do this if our peer understands T/TCP.
490 */
491 case 0:
492 opt[optlen++] = TCPOPT_NOP;
493 opt[optlen++] = TCPOPT_NOP;
494 opt[optlen++] = TCPOPT_CC;
495 opt[optlen++] = TCPOLEN_CC;
496 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
497
498 optlen += 4;
499 break;
500
501 /*
502 * This is our initial SYN, check whether we have to use
503 * CC or CC.new.
504 */
505 case TH_SYN:
506 opt[optlen++] = TCPOPT_NOP;
507 opt[optlen++] = TCPOPT_NOP;
508 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ?
509 TCPOPT_CCNEW : TCPOPT_CC;
510 opt[optlen++] = TCPOLEN_CC;
511 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
512 optlen += 4;
513 break;
514
515 /*
516 * This is a SYN,ACK; send CC and CC.echo if we received
517 * CC from our peer.
518 */
519 case (TH_SYN|TH_ACK):
520 if (tp->t_flags & TF_RCVD_CC) {
521 opt[optlen++] = TCPOPT_NOP;
522 opt[optlen++] = TCPOPT_NOP;
523 opt[optlen++] = TCPOPT_CC;
524 opt[optlen++] = TCPOLEN_CC;
525 *(u_int32_t *)&opt[optlen] =
526 htonl(tp->cc_send);
527 optlen += 4;
528 opt[optlen++] = TCPOPT_NOP;
529 opt[optlen++] = TCPOPT_NOP;
530 opt[optlen++] = TCPOPT_CCECHO;
531 opt[optlen++] = TCPOLEN_CC;
532 *(u_int32_t *)&opt[optlen] =
533 htonl(tp->cc_recv);
534 optlen += 4;
535 }
536 break;
537 }
538 }
539
540 hdrlen += optlen;
541
542#ifdef INET6
543 if (isipv6)
544 ipoptlen = ip6_optlen(tp->t_inpcb);
545 else
546#endif
547 {
548 if (tp->t_inpcb->inp_options) {
549 ipoptlen = tp->t_inpcb->inp_options->m_len -
550 offsetof(struct ipoption, ipopt_list);
551 } else {
552 ipoptlen = 0;
553 }
554 }
555#ifdef IPSEC
556 ipoptlen += ipsec_hdrsiz_tcp(tp);
557#endif
558
559 /*
560 * Adjust data length if insertion of options will
561 * bump the packet length beyond the t_maxopd length.
562 * Clear the FIN bit because we cut off the tail of
563 * the segment.
564 */
565 if (len + optlen + ipoptlen > tp->t_maxopd) {
566 /*
567 * If there is still more to send, don't close the connection.
568 */
569 flags &= ~TH_FIN;
570 len = tp->t_maxopd - optlen - ipoptlen;
571 sendalot = 1;
572 }
573
574/*#ifdef DIAGNOSTIC*/
575#ifdef INET6
576 if (max_linkhdr + hdrlen > MCLBYTES)
577 panic("tcphdr too big");
578#else
579 if (max_linkhdr + hdrlen > MHLEN)
580 panic("tcphdr too big");
581#endif
582/*#endif*/
583
584 /*
585 * Grab a header mbuf, attaching a copy of data to
586 * be transmitted, and initialize the header from
587 * the template for sends on this connection.
588 */
589 if (len) {
590 if (tp->t_force && len == 1)
591 tcpstat.tcps_sndprobe++;
592 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
593 tcpstat.tcps_sndrexmitpack++;
594 tcpstat.tcps_sndrexmitbyte += len;
595 } else {
596 tcpstat.tcps_sndpack++;
597 tcpstat.tcps_sndbyte += len;
598 }
599#ifdef notyet
600 if ((m = m_copypack(so->so_snd.sb_mb, off,
601 (int)len, max_linkhdr + hdrlen)) == 0) {
602 error = ENOBUFS;
603 goto out;
604 }
605 /*
606 * m_copypack left space for our hdr; use it.
607 */
608 m->m_len += hdrlen;
609 m->m_data -= hdrlen;
610#else
611 MGETHDR(m, M_DONTWAIT, MT_HEADER);
612 if (m == NULL) {
613 error = ENOBUFS;
614 goto out;
615 }
616#ifdef INET6
617 if (MHLEN < hdrlen + max_linkhdr) {
618 MCLGET(m, M_DONTWAIT);
619 if ((m->m_flags & M_EXT) == 0) {
620 m_freem(m);
621 error = ENOBUFS;
622 goto out;
623 }
624 }
625#endif
626 m->m_data += max_linkhdr;
627 m->m_len = hdrlen;
628 if (len <= MHLEN - hdrlen - max_linkhdr) {
629 m_copydata(so->so_snd.sb_mb, off, (int) len,
630 mtod(m, caddr_t) + hdrlen);
631 m->m_len += len;
632 } else {
633 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
634 if (m->m_next == 0) {
635 (void) m_free(m);
636 error = ENOBUFS;
637 goto out;
638 }
639 }
640#endif
641 /*
642 * If we're sending everything we've got, set PUSH.
643 * (This will keep happy those implementations which only
644 * give data to the user when a buffer fills or
645 * a PUSH comes in.)
646 */
647 if (off + len == so->so_snd.sb_cc)
648 flags |= TH_PUSH;
649 } else {
650 if (tp->t_flags & TF_ACKNOW)
651 tcpstat.tcps_sndacks++;
652 else if (flags & (TH_SYN|TH_FIN|TH_RST))
653 tcpstat.tcps_sndctrl++;
654 else if (SEQ_GT(tp->snd_up, tp->snd_una))
655 tcpstat.tcps_sndurg++;
656 else
657 tcpstat.tcps_sndwinup++;
658
659 MGETHDR(m, M_DONTWAIT, MT_HEADER);
660 if (m == NULL) {
661 error = ENOBUFS;
662 goto out;
663 }
664#ifdef INET6
665 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
666 MHLEN >= hdrlen) {
667 MH_ALIGN(m, hdrlen);
668 } else
669#endif
670 m->m_data += max_linkhdr;
671 m->m_len = hdrlen;
672 }
673 m->m_pkthdr.rcvif = (struct ifnet *)0;
674#ifdef INET6
675 if (isipv6) {
676 ip6 = mtod(m, struct ip6_hdr *);
677 th = (struct tcphdr *)(ip6 + 1);
678 tcp_fillheaders(tp, ip6, th);
679 } else
680#endif /* INET6 */
681 {
682 ip = mtod(m, struct ip *);
683 ipov = (struct ipovly *)ip;
684 th = (struct tcphdr *)(ip + 1);
685 /* this picks up the pseudo header (w/o the length) */
686 tcp_fillheaders(tp, ip, th);
687 }
688
689 /*
690 * Fill in fields, remembering maximum advertised
691 * window for use in delaying messages about window sizes.
692 * If resending a FIN, be sure not to use a new sequence number.
693 */
694 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
695 tp->snd_nxt == tp->snd_max)
696 tp->snd_nxt--;
697 /*
698 * If we are doing retransmissions, then snd_nxt will
699 * not reflect the first unsent octet. For ACK only
700 * packets, we do not want the sequence number of the
701 * retransmitted packet, we want the sequence number
702 * of the next unsent octet. So, if there is no data
703 * (and no SYN or FIN), use snd_max instead of snd_nxt
704 * when filling in ti_seq. But if we are in persist
705 * state, snd_max might reflect one byte beyond the
706 * right edge of the window, so use snd_nxt in that
707 * case, since we know we aren't doing a retransmission.
708 * (retransmit and persist are mutually exclusive...)
709 */
710 if (len || (flags & (TH_SYN|TH_FIN))
711 || callout_active(tp->tt_persist))
712 th->th_seq = htonl(tp->snd_nxt);
713 else
714 th->th_seq = htonl(tp->snd_max);
715 th->th_ack = htonl(tp->rcv_nxt);
716 if (optlen) {
717 bcopy(opt, th + 1, optlen);
718 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
719 }
720 th->th_flags = flags;
721 /*
722 * Calculate receive window. Don't shrink window,
723 * but avoid silly window syndrome.
724 */
725 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
726 win = 0;
727 if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
728 win = (long)(tp->rcv_adv - tp->rcv_nxt);
729 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
730 win = (long)TCP_MAXWIN << tp->rcv_scale;
731 th->th_win = htons((u_short) (win>>tp->rcv_scale));
732
733 /*
734 * Adjust the RXWIN0SENT flag - indicate that we have advertised
735 * a 0 window. This may cause the remote transmitter to stall. This
736 * flag tells soreceive() to disable delayed acknowledgements when
737 * draining the buffer. This can occur if the receiver is attempting
738 * to read more data then can be buffered prior to transmitting on
739 * the connection.
740 */
741 if (win == 0)
742 tp->t_flags |= TF_RXWIN0SENT;
743 else
744 tp->t_flags &= ~TF_RXWIN0SENT;
745
746 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
747 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
748 th->th_flags |= TH_URG;
749 } else
750 /*
751 * If no urgent pointer to send, then we pull
752 * the urgent pointer to the left edge of the send window
753 * so that it doesn't drift into the send window on sequence
754 * number wraparound.
755 */
756 tp->snd_up = tp->snd_una; /* drag it along */
757
758 /*
759 * Put TCP length in extended header, and then
760 * checksum extended header and data.
761 */
762 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
763#ifdef INET6
764 if (isipv6)
765 /*
766 * ip6_plen is not need to be filled now, and will be filled
767 * in ip6_output.
768 */
769 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
770 sizeof(struct tcphdr) + optlen + len);
771 else
772#endif /* INET6 */
773 {
774 m->m_pkthdr.csum_flags = CSUM_TCP;
775 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
776 if (len + optlen)
777 th->th_sum = in_addword(th->th_sum,
778 htons((u_short)(optlen + len)));
779
780 /* IP version must be set here for ipv4/ipv6 checking later */
781 KASSERT(ip->ip_v == IPVERSION,
782 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
783 }
784
785 /*
786 * In transmit state, time the transmission and arrange for
787 * the retransmit. In persist state, just set snd_max.
788 */
789 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
790 tcp_seq startseq = tp->snd_nxt;
791
792 /*
793 * Advance snd_nxt over sequence space of this segment.
794 */
795 if (flags & (TH_SYN|TH_FIN)) {
796 if (flags & TH_SYN)
797 tp->snd_nxt++;
798 if (flags & TH_FIN) {
799 tp->snd_nxt++;
800 tp->t_flags |= TF_SENTFIN;
801 }
802 }
803 tp->snd_nxt += len;
804 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
805 tp->snd_max = tp->snd_nxt;
806 /*
807 * Time this transmission if not a retransmission and
808 * not currently timing anything.
809 */
810 if (tp->t_rtttime == 0) {
811 tp->t_rtttime = ticks;
812 tp->t_rtseq = startseq;
813 tcpstat.tcps_segstimed++;
814 }
815 }
816
817 /*
818 * Set retransmit timer if not currently set,
819 * and not doing a pure ack or a keep-alive probe.
820 * Initial value for retransmit timer is smoothed
821 * round-trip time + 2 * round-trip time variance.
822 * Initialize shift counter which is used for backoff
823 * of retransmit time.
824 */
825 if (!callout_active(tp->tt_rexmt) &&
826 tp->snd_nxt != tp->snd_una) {
827 if (callout_active(tp->tt_persist)) {
828 callout_stop(tp->tt_persist);
829 tp->t_rxtshift = 0;
830 }
831 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
832 tcp_timer_rexmt, tp);
833 }
834 } else {
835 /*
836 * Persist case, update snd_max but since we are in
837 * persist mode (no window) we do not update snd_nxt.
838 */
839 int xlen = len;
840 if (flags & TH_SYN)
841 ++xlen;
842 if (flags & TH_FIN) {
843 ++xlen;
844 tp->t_flags |= TF_SENTFIN;
845 }
846 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
847 tp->snd_max = tp->snd_nxt + xlen;
848 }
849
850#ifdef TCPDEBUG
851 /*
852 * Trace.
853 */
854 if (so->so_options & SO_DEBUG)
855 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
856#endif
857
858 /*
859 * Fill in IP length and desired time to live and
860 * send to IP level. There should be a better way
861 * to handle ttl and tos; we could keep them in
862 * the template, but need a way to checksum without them.
863 */
864 /*
865 * m->m_pkthdr.len should have been set before cksum calcuration,
866 * because in6_cksum() need it.
867 */
868#ifdef INET6
869 if (isipv6) {
870 /*
871 * we separately set hoplimit for every segment, since the
872 * user might want to change the value via setsockopt.
873 * Also, desired default hop limit might be changed via
874 * Neighbor Discovery.
875 */
876 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb,
877 tp->t_inpcb->in6p_route.ro_rt ?
878 tp->t_inpcb->in6p_route.ro_rt->rt_ifp
879 : NULL);
880
881 /* TODO: IPv6 IP6TOS_ECT bit on */
882 error = ip6_output(m,
883 tp->t_inpcb->in6p_outputopts,
884 &tp->t_inpcb->in6p_route,
885 (so->so_options & SO_DONTROUTE), NULL, NULL,
886 tp->t_inpcb);
887 } else
888#endif /* INET6 */
889 {
890 struct rtentry *rt;
891 ip->ip_len = m->m_pkthdr.len;
892#ifdef INET6
893 if (INP_CHECK_SOCKAF(so, AF_INET6))
894 ip->ip_ttl = in6_selecthlim(tp->t_inpcb,
895 tp->t_inpcb->in6p_route.ro_rt ?
896 tp->t_inpcb->in6p_route.ro_rt->rt_ifp
897 : NULL);
898 else
899#endif /* INET6 */
900 ip->ip_ttl = tp->t_inpcb->inp_ip_ttl; /* XXX */
901 ip->ip_tos = tp->t_inpcb->inp_ip_tos; /* XXX */
902 /*
903 * See if we should do MTU discovery. We do it only if the following
904 * are true:
905 * 1) we have a valid route to the destination
906 * 2) the MTU is not locked (if it is, then discovery has been
907 * disabled)
908 */
909 if (path_mtu_discovery
910 && (rt = tp->t_inpcb->inp_route.ro_rt)
911 && rt->rt_flags & RTF_UP
912 && !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
913 ip->ip_off |= IP_DF;
914 }
915 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
916 (so->so_options & SO_DONTROUTE), 0, tp->t_inpcb);
917 }
918 if (error) {
919
920 /*
921 * We know that the packet was lost, so back out the
922 * sequence number advance, if any.
923 */
924 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
925 /*
926 * No need to check for TH_FIN here because
927 * the TF_SENTFIN flag handles that case.
928 */
929 if ((flags & TH_SYN) == 0)
930 tp->snd_nxt -= len;
931 }
932
933out:
934 if (error == ENOBUFS) {
935 if (!callout_active(tp->tt_rexmt) &&
936 !callout_active(tp->tt_persist))
937 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
938 tcp_timer_rexmt, tp);
939 tcp_quench(tp->t_inpcb, 0);
940 return (0);
941 }
942 if (error == EMSGSIZE) {
943 /*
944 * ip_output() will have already fixed the route
945 * for us. tcp_mtudisc() will, as its last action,
946 * initiate retransmission, so it is important to
947 * not do so here.
948 */
949 tcp_mtudisc(tp->t_inpcb, 0);
950 return 0;
951 }
952 if ((error == EHOSTUNREACH || error == ENETDOWN)
953 && TCPS_HAVERCVDSYN(tp->t_state)) {
954 tp->t_softerror = error;
955 return (0);
956 }
957 return (error);
958 }
959 tcpstat.tcps_sndtotal++;
960
961 /*
962 * Data sent (as far as we can tell).
963 * If this advertises a larger window than any other segment,
964 * then remember the size of the advertised window.
965 * Any pending ACK has now been sent.
966 */
967 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
968 tp->rcv_adv = tp->rcv_nxt + win;
969 tp->last_ack_sent = tp->rcv_nxt;
970 tp->t_flags &= ~TF_ACKNOW;
971 if (tcp_delack_enabled)
972 callout_stop(tp->tt_delack);
973#if 0
974 /*
975 * This completely breaks TCP if newreno is turned on. What happens
976 * is that if delayed-acks are turned on on the receiver, this code
977 * on the transmitter effectively destroys the TCP window, forcing
978 * it to four packets (1.5Kx4 = 6K window).
979 */
980 if (sendalot && (!tcp_do_newreno || --maxburst))
981 goto again;
982#endif
983 if (sendalot)
984 goto again;
985 return (0);
986}
987
988void
989tcp_setpersist(tp)
2256ba69 990 struct tcpcb *tp;
984263bc
MD
991{
992 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
993 int tt;
994
995 if (callout_active(tp->tt_rexmt))
996 panic("tcp_setpersist: retransmit pending");
997 /*
998 * Start/restart persistance timer.
999 */
1000 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
1001 TCPTV_PERSMIN, TCPTV_PERSMAX);
1002 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp);
1003 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
1004 tp->t_rxtshift++;
1005}