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_inet.h"
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
87 extern struct mbuf *m_copypack();
88 #endif
89
90 int path_mtu_discovery = 1;
91 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
92 &path_mtu_discovery, 1, "Enable Path MTU Discovery");
93
94 int ss_fltsz = 1;
95 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
96 &ss_fltsz, 1, "Slow start flight size");
97
98 int ss_fltsz_local = 4;
99 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
100 &ss_fltsz_local, 1, "Slow start flight size for local networks");
101
102 int tcp_do_newreno = 1;
103 SYSCTL_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 */
109 int
110 tcp_output(tp)
111 register struct tcpcb *tp;
112 {
113 register struct socket *so = tp->t_inpcb->inp_socket;
114 register long len, win;
115 int off, flags, error;
116 #ifdef TCP_SIGNATURE
117 int sigoff = 0;
118 #endif
119 register struct mbuf *m;
120 struct ip *ip = NULL;
121 register struct ipovly *ipov = NULL;
122 #ifdef INET6
123 struct ip6_hdr *ip6 = NULL;
124 #endif /* INET6 */
125 register struct tcphdr *th;
126 u_char opt[TCP_MAXOLEN];
127 unsigned ipoptlen, optlen, hdrlen;
128 int idle, sendalot;
129 #if 0
130 int maxburst = TCP_MAXBURST;
131 #endif
132 struct rmxp_tao *taop;
133 #ifdef INET6
134 int isipv6;
135 #endif
136
137 #ifdef INET6
138 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
139 #endif
140
141 /*
142 * Determine length of data that should be transmitted,
143 * and flags that will be used.
144 * If there is some data or critical controls (SYN, RST)
145 * to send, then transmit; otherwise, investigate further.
146 */
147 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
148 if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
149 /*
150 * We have been idle for "a while" and no acks are
151 * expected to clock out any data we send --
152 * slow start to get ack "clock" running again.
153 *
154 * Set the slow-start flight size depending on whether
155 * this is a local network or not.
156 */
157 if (
158 #ifdef INET6
159 (isipv6 && in6_localaddr(&tp->t_inpcb->in6p_faddr)) ||
160 (!isipv6 &&
161 #endif
162 in_localaddr(tp->t_inpcb->inp_faddr)
163 #ifdef INET6
164 )
165 #endif
166 )
167 tp->snd_cwnd = tp->t_maxseg * ss_fltsz_local;
168 else
169 tp->snd_cwnd = tp->t_maxseg * ss_fltsz;
170 }
171 tp->t_flags &= ~TF_LASTIDLE;
172 if (idle) {
173 if (tp->t_flags & TF_MORETOCOME) {
174 tp->t_flags |= TF_LASTIDLE;
175 idle = 0;
176 }
177 }
178 again:
179 sendalot = 0;
180 off = tp->snd_nxt - tp->snd_una;
181 win = min(tp->snd_wnd, tp->snd_cwnd);
182 win = min(win, tp->snd_bwnd);
183
184 flags = tcp_outflags[tp->t_state];
185 /*
186 * Get standard flags, and add SYN or FIN if requested by 'hidden'
187 * state flags.
188 */
189 if (tp->t_flags & TF_NEEDFIN)
190 flags |= TH_FIN;
191 if (tp->t_flags & TF_NEEDSYN)
192 flags |= TH_SYN;
193
194 /*
195 * If in persist timeout with window of 0, send 1 byte.
196 * Otherwise, if window is small but nonzero
197 * and timer expired, we will send what we can
198 * and go to transmit state.
199 */
200 if (tp->t_force) {
201 if (win == 0) {
202 /*
203 * If we still have some data to send, then
204 * clear the FIN bit. Usually this would
205 * happen below when it realizes that we
206 * aren't sending all the data. However,
207 * if we have exactly 1 byte of unsent data,
208 * then it won't clear the FIN bit below,
209 * and if we are in persist state, we wind
210 * up sending the packet without recording
211 * that we sent the FIN bit.
212 *
213 * We can't just blindly clear the FIN bit,
214 * because if we don't have any more data
215 * to send then the probe will be the FIN
216 * itself.
217 */
218 if (off < so->so_snd.sb_cc)
219 flags &= ~TH_FIN;
220 win = 1;
221 } else {
222 callout_stop(tp->tt_persist);
223 tp->t_rxtshift = 0;
224 }
225 }
226
227 /*
228 * If snd_nxt == snd_max and we have transmitted a FIN, the
229 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
230 * a negative length. This can also occur when tcp opens up
231 * its congestion window while receiving additional duplicate
232 * acks after fast-retransmit because TCP will reset snd_nxt
233 * to snd_max after the fast-retransmit.
234 *
235 * In the normal retransmit-FIN-only case, however, snd_nxt will
236 * be set to snd_una, the offset will be 0, and the length may
237 * wind up 0.
238 */
239 len = (long)ulmin(so->so_snd.sb_cc, win) - off;
240
241 taop = tcp_gettaocache(&tp->t_inpcb->inp_inc);
242
243 /*
244 * Lop off SYN bit if it has already been sent. However, if this
245 * is SYN-SENT state and if segment contains data and if we don't
246 * know that foreign host supports TAO, suppress sending segment.
247 */
248 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
249 flags &= ~TH_SYN;
250 off--, len++;
251 if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
252 (taop == NULL || taop->tao_ccsent == 0))
253 return 0;
254 }
255
256 /*
257 * Be careful not to send data and/or FIN on SYN segments
258 * in cases when no CC option will be sent.
259 * This measure is needed to prevent interoperability problems
260 * with not fully conformant TCP implementations.
261 */
262 if ((flags & TH_SYN) &&
263 ((tp->t_flags & TF_NOOPT) || !(tp->t_flags & TF_REQ_CC) ||
264 ((flags & TH_ACK) && !(tp->t_flags & TF_RCVD_CC)))) {
265 len = 0;
266 flags &= ~TH_FIN;
267 }
268
269 if (len < 0) {
270 /*
271 * If FIN has been sent but not acked,
272 * but we haven't been called to retransmit,
273 * len will be < 0. Otherwise, window shrank
274 * after we sent into it. If window shrank to 0,
275 * cancel pending retransmit, pull snd_nxt back
276 * to (closed) window, and set the persist timer
277 * if it isn't already going. If the window didn't
278 * close completely, just wait for an ACK.
279 */
280 len = 0;
281 if (win == 0) {
282 callout_stop(tp->tt_rexmt);
283 tp->t_rxtshift = 0;
284 tp->snd_nxt = tp->snd_una;
285 if (!callout_active(tp->tt_persist))
286 tcp_setpersist(tp);
287 }
288 }
289
290 /*
291 * len will be >= 0 after this point. Truncate to the maximum
292 * segment length and ensure that FIN is removed if the length
293 * no longer contains the last data byte.
294 */
295 if (len > tp->t_maxseg) {
296 len = tp->t_maxseg;
297 sendalot = 1;
298 }
299 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
300 flags &= ~TH_FIN;
301
302 win = sbspace(&so->so_rcv);
303
304 /*
305 * Sender silly window avoidance. We transmit under the following
306 * conditions when len is non-zero:
307 *
308 * - We have a full segment
309 * - This is the last buffer in a write()/send() and we are
310 * either idle or running NODELAY
311 * - we've timed out (e.g. persist timer)
312 * - we have more then 1/2 the maximum send window's worth of
313 * data (receiver may be limited the window size)
314 * - we need to retransmit
315 */
316 if (len) {
317 if (len == tp->t_maxseg)
318 goto send;
319 /*
320 * NOTE! on localhost connections an 'ack' from the remote
321 * end may occur synchronously with the output and cause
322 * us to flush a buffer queued with moretocome. XXX
323 *
324 * note: the len + off check is almost certainly unnecessary.
325 */
326 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
327 (idle || (tp->t_flags & TF_NODELAY)) &&
328 len + off >= so->so_snd.sb_cc &&
329 (tp->t_flags & TF_NOPUSH) == 0) {
330 goto send;
331 }
332 if (tp->t_force) /* typ. timeout case */
333 goto send;
334 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
335 goto send;
336 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
337 goto send;
338 }
339
340 /*
341 * Compare available window to amount of window
342 * known to peer (as advertised window less
343 * next expected input). If the difference is at least two
344 * max size segments, or at least 50% of the maximum possible
345 * window, then want to send a window update to peer.
346 */
347 if (win > 0) {
348 /*
349 * "adv" is the amount we can increase the window,
350 * taking into account that we are limited by
351 * TCP_MAXWIN << tp->rcv_scale.
352 */
353 long adv = min(win, (long)TCP_MAXWIN << tp->rcv_scale) -
354 (tp->rcv_adv - tp->rcv_nxt);
355
356 if (adv >= (long) (2 * tp->t_maxseg))
357 goto send;
358 if (2 * adv >= (long) so->so_rcv.sb_hiwat)
359 goto send;
360 }
361
362 /*
363 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
364 * is also a catch-all for the retransmit timer timeout case.
365 */
366 if (tp->t_flags & TF_ACKNOW)
367 goto send;
368 if ((flags & TH_RST) ||
369 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
370 goto send;
371 if (SEQ_GT(tp->snd_up, tp->snd_una))
372 goto send;
373 /*
374 * If our state indicates that FIN should be sent
375 * and we have not yet done so, then we need to send.
376 */
377 if (flags & TH_FIN &&
378 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
379 goto send;
380
381 /*
382 * TCP window updates are not reliable, rather a polling protocol
383 * using ``persist'' packets is used to insure receipt of window
384 * updates. The three ``states'' for the output side are:
385 * idle not doing retransmits or persists
386 * persisting to move a small or zero window
387 * (re)transmitting and thereby not persisting
388 *
389 * callout_active(tp->tt_persist)
390 * is true when we are in persist state.
391 * tp->t_force
392 * is set when we are called to send a persist packet.
393 * callout_active(tp->tt_rexmt)
394 * is set when we are retransmitting
395 * The output side is idle when both timers are zero.
396 *
397 * If send window is too small, there is data to transmit, and no
398 * retransmit or persist is pending, then go to persist state.
399 * If nothing happens soon, send when timer expires:
400 * if window is nonzero, transmit what we can,
401 * otherwise force out a byte.
402 */
403 if (so->so_snd.sb_cc && !callout_active(tp->tt_rexmt) &&
404 !callout_active(tp->tt_persist)) {
405 tp->t_rxtshift = 0;
406 tcp_setpersist(tp);
407 }
408
409 /*
410 * No reason to send a segment, just return.
411 */
412 return (0);
413
414 send:
415 /*
416 * Before ESTABLISHED, force sending of initial options
417 * unless TCP set not to do any options.
418 * NOTE: we assume that the IP/TCP header plus TCP options
419 * always fit in a single mbuf, leaving room for a maximum
420 * link header, i.e.
421 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
422 */
423 optlen = 0;
424 #ifdef INET6
425 if (isipv6)
426 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
427 else
428 #endif
429 hdrlen = sizeof (struct tcpiphdr);
430 if (flags & TH_SYN) {
431 tp->snd_nxt = tp->iss;
432 if ((tp->t_flags & TF_NOOPT) == 0) {
433 u_short mss;
434
435 opt[0] = TCPOPT_MAXSEG;
436 opt[1] = TCPOLEN_MAXSEG;
437 mss = htons((u_short) tcp_mssopt(tp));
438 (void)memcpy(opt + 2, &mss, sizeof(mss));
439 optlen = TCPOLEN_MAXSEG;
440
441 if ((tp->t_flags & TF_REQ_SCALE) &&
442 ((flags & TH_ACK) == 0 ||
443 (tp->t_flags & TF_RCVD_SCALE))) {
444 *((u_int32_t *)(opt + optlen)) = htonl(
445 TCPOPT_NOP << 24 |
446 TCPOPT_WINDOW << 16 |
447 TCPOLEN_WINDOW << 8 |
448 tp->request_r_scale);
449 optlen += 4;
450 }
451 }
452 }
453
454 /*
455 * Send a timestamp and echo-reply if this is a SYN and our side
456 * wants to use timestamps (TF_REQ_TSTMP is set) or both our side
457 * and our peer have sent timestamps in our SYN's.
458 */
459 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
460 (flags & TH_RST) == 0 &&
461 ((flags & TH_ACK) == 0 ||
462 (tp->t_flags & TF_RCVD_TSTMP))) {
463 u_int32_t *lp = (u_int32_t *)(opt + optlen);
464
465 /* Form timestamp option as shown in appendix A of RFC 1323. */
466 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
467 *lp++ = htonl(ticks);
468 *lp = htonl(tp->ts_recent);
469 optlen += TCPOLEN_TSTAMP_APPA;
470 }
471
472 /*
473 * Send `CC-family' options if our side wants to use them (TF_REQ_CC),
474 * options are allowed (!TF_NOOPT) and it's not a RST.
475 */
476 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
477 (flags & TH_RST) == 0) {
478 switch (flags & (TH_SYN|TH_ACK)) {
479 /*
480 * This is a normal ACK, send CC if we received CC before
481 * from our peer.
482 */
483 case TH_ACK:
484 if (!(tp->t_flags & TF_RCVD_CC))
485 break;
486 /*FALLTHROUGH*/
487
488 /*
489 * We can only get here in T/TCP's SYN_SENT* state, when
490 * we're a sending a non-SYN segment without waiting for
491 * the ACK of our SYN. A check above assures that we only
492 * do this if our peer understands T/TCP.
493 */
494 case 0:
495 opt[optlen++] = TCPOPT_NOP;
496 opt[optlen++] = TCPOPT_NOP;
497 opt[optlen++] = TCPOPT_CC;
498 opt[optlen++] = TCPOLEN_CC;
499 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
500
501 optlen += 4;
502 break;
503
504 /*
505 * This is our initial SYN, check whether we have to use
506 * CC or CC.new.
507 */
508 case TH_SYN:
509 opt[optlen++] = TCPOPT_NOP;
510 opt[optlen++] = TCPOPT_NOP;
511 opt[optlen++] = tp->t_flags & TF_SENDCCNEW ?
512 TCPOPT_CCNEW : TCPOPT_CC;
513 opt[optlen++] = TCPOLEN_CC;
514 *(u_int32_t *)&opt[optlen] = htonl(tp->cc_send);
515 optlen += 4;
516 break;
517
518 /*
519 * This is a SYN,ACK; send CC and CC.echo if we received
520 * CC from our peer.
521 */
522 case (TH_SYN|TH_ACK):
523 if (tp->t_flags & TF_RCVD_CC) {
524 opt[optlen++] = TCPOPT_NOP;
525 opt[optlen++] = TCPOPT_NOP;
526 opt[optlen++] = TCPOPT_CC;
527 opt[optlen++] = TCPOLEN_CC;
528 *(u_int32_t *)&opt[optlen] =
529 htonl(tp->cc_send);
530 optlen += 4;
531 opt[optlen++] = TCPOPT_NOP;
532 opt[optlen++] = TCPOPT_NOP;
533 opt[optlen++] = TCPOPT_CCECHO;
534 opt[optlen++] = TCPOLEN_CC;
535 *(u_int32_t *)&opt[optlen] =
536 htonl(tp->cc_recv);
537 optlen += 4;
538 }
539 break;
540 }
541 }
542
543 #ifdef TCP_SIGNATURE
544 #ifdef INET6
545 if (!isipv6)
546 #endif
547 if (tp->t_flags & TF_SIGNATURE) {
548 int i;
549 u_char *bp;
550 /*
551 * Initialize TCP-MD5 option (RFC2385)
552 */
553 bp = (u_char *)opt + optlen;
554 *bp++ = TCPOPT_SIGNATURE;
555 *bp++ = TCPOLEN_SIGNATURE;
556 sigoff = optlen + 2;
557 for (i = 0; i < TCP_SIGLEN; i++)
558 *bp++ = 0;
559 optlen += TCPOLEN_SIGNATURE;
560 /*
561 * Terminate options list and maintain 32-bit alignment.
562 */
563 *bp++ = TCPOPT_NOP;
564 *bp++ = TCPOPT_EOL;
565 optlen += 2;
566 }
567 #endif /* TCP_SIGNATURE */
568
569 hdrlen += optlen;
570
571 #ifdef INET6
572 if (isipv6)
573 ipoptlen = ip6_optlen(tp->t_inpcb);
574 else
575 #endif
576 {
577 if (tp->t_inpcb->inp_options) {
578 ipoptlen = tp->t_inpcb->inp_options->m_len -
579 offsetof(struct ipoption, ipopt_list);
580 } else {
581 ipoptlen = 0;
582 }
583 }
584 #ifdef IPSEC
585 ipoptlen += ipsec_hdrsiz_tcp(tp);
586 #endif
587
588 /*
589 * Adjust data length if insertion of options will
590 * bump the packet length beyond the t_maxopd length.
591 * Clear the FIN bit because we cut off the tail of
592 * the segment.
593 */
594 if (len + optlen + ipoptlen > tp->t_maxopd) {
595 /*
596 * If there is still more to send, don't close the connection.
597 */
598 flags &= ~TH_FIN;
599 len = tp->t_maxopd - optlen - ipoptlen;
600 sendalot = 1;
601 }
602
603 /*#ifdef DIAGNOSTIC*/
604 #ifdef INET6
605 if (max_linkhdr + hdrlen > MCLBYTES)
606 panic("tcphdr too big");
607 #else
608 if (max_linkhdr + hdrlen > MHLEN)
609 panic("tcphdr too big");
610 #endif
611 /*#endif*/
612
613 /*
614 * Grab a header mbuf, attaching a copy of data to
615 * be transmitted, and initialize the header from
616 * the template for sends on this connection.
617 */
618 if (len) {
619 if (tp->t_force && len == 1)
620 tcpstat.tcps_sndprobe++;
621 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
622 tcpstat.tcps_sndrexmitpack++;
623 tcpstat.tcps_sndrexmitbyte += len;
624 } else {
625 tcpstat.tcps_sndpack++;
626 tcpstat.tcps_sndbyte += len;
627 }
628 #ifdef notyet
629 if ((m = m_copypack(so->so_snd.sb_mb, off,
630 (int)len, max_linkhdr + hdrlen)) == 0) {
631 error = ENOBUFS;
632 goto out;
633 }
634 /*
635 * m_copypack left space for our hdr; use it.
636 */
637 m->m_len += hdrlen;
638 m->m_data -= hdrlen;
639 #else
640 MGETHDR(m, M_DONTWAIT, MT_HEADER);
641 if (m == NULL) {
642 error = ENOBUFS;
643 goto out;
644 }
645 #ifdef INET6
646 if (MHLEN < hdrlen + max_linkhdr) {
647 MCLGET(m, M_DONTWAIT);
648 if ((m->m_flags & M_EXT) == 0) {
649 m_freem(m);
650 error = ENOBUFS;
651 goto out;
652 }
653 }
654 #endif
655 m->m_data += max_linkhdr;
656 m->m_len = hdrlen;
657 if (len <= MHLEN - hdrlen - max_linkhdr) {
658 m_copydata(so->so_snd.sb_mb, off, (int) len,
659 mtod(m, caddr_t) + hdrlen);
660 m->m_len += len;
661 } else {
662 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
663 if (m->m_next == 0) {
664 (void) m_free(m);
665 error = ENOBUFS;
666 goto out;
667 }
668 }
669 #endif
670 /*
671 * If we're sending everything we've got, set PUSH.
672 * (This will keep happy those implementations which only
673 * give data to the user when a buffer fills or
674 * a PUSH comes in.)
675 */
676 if (off + len == so->so_snd.sb_cc)
677 flags |= TH_PUSH;
678 } else {
679 if (tp->t_flags & TF_ACKNOW)
680 tcpstat.tcps_sndacks++;
681 else if (flags & (TH_SYN|TH_FIN|TH_RST))
682 tcpstat.tcps_sndctrl++;
683 else if (SEQ_GT(tp->snd_up, tp->snd_una))
684 tcpstat.tcps_sndurg++;
685 else
686 tcpstat.tcps_sndwinup++;
687
688 MGETHDR(m, M_DONTWAIT, MT_HEADER);
689 if (m == NULL) {
690 error = ENOBUFS;
691 goto out;
692 }
693 #ifdef INET6
694 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
695 MHLEN >= hdrlen) {
696 MH_ALIGN(m, hdrlen);
697 } else
698 #endif
699 m->m_data += max_linkhdr;
700 m->m_len = hdrlen;
701 }
702 m->m_pkthdr.rcvif = (struct ifnet *)0;
703 #ifdef INET6
704 if (isipv6) {
705 ip6 = mtod(m, struct ip6_hdr *);
706 th = (struct tcphdr *)(ip6 + 1);
707 tcp_fillheaders(tp, ip6, th);
708 } else
709 #endif /* INET6 */
710 {
711 ip = mtod(m, struct ip *);
712 ipov = (struct ipovly *)ip;
713 th = (struct tcphdr *)(ip + 1);
714 /* this picks up the pseudo header (w/o the length) */
715 tcp_fillheaders(tp, ip, th);
716 }
717
718 /*
719 * Fill in fields, remembering maximum advertised
720 * window for use in delaying messages about window sizes.
721 * If resending a FIN, be sure not to use a new sequence number.
722 */
723 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
724 tp->snd_nxt == tp->snd_max)
725 tp->snd_nxt--;
726 /*
727 * If we are doing retransmissions, then snd_nxt will
728 * not reflect the first unsent octet. For ACK only
729 * packets, we do not want the sequence number of the
730 * retransmitted packet, we want the sequence number
731 * of the next unsent octet. So, if there is no data
732 * (and no SYN or FIN), use snd_max instead of snd_nxt
733 * when filling in ti_seq. But if we are in persist
734 * state, snd_max might reflect one byte beyond the
735 * right edge of the window, so use snd_nxt in that
736 * case, since we know we aren't doing a retransmission.
737 * (retransmit and persist are mutually exclusive...)
738 */
739 if (len || (flags & (TH_SYN|TH_FIN))
740 || callout_active(tp->tt_persist))
741 th->th_seq = htonl(tp->snd_nxt);
742 else
743 th->th_seq = htonl(tp->snd_max);
744 th->th_ack = htonl(tp->rcv_nxt);
745 if (optlen) {
746 bcopy(opt, th + 1, optlen);
747 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
748 }
749 th->th_flags = flags;
750 /*
751 * Calculate receive window. Don't shrink window,
752 * but avoid silly window syndrome.
753 */
754 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
755 win = 0;
756 if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
757 win = (long)(tp->rcv_adv - tp->rcv_nxt);
758 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
759 win = (long)TCP_MAXWIN << tp->rcv_scale;
760 th->th_win = htons((u_short) (win>>tp->rcv_scale));
761
762 /*
763 * Adjust the RXWIN0SENT flag - indicate that we have advertised
764 * a 0 window. This may cause the remote transmitter to stall. This
765 * flag tells soreceive() to disable delayed acknowledgements when
766 * draining the buffer. This can occur if the receiver is attempting
767 * to read more data then can be buffered prior to transmitting on
768 * the connection.
769 */
770 if (win == 0)
771 tp->t_flags |= TF_RXWIN0SENT;
772 else
773 tp->t_flags &= ~TF_RXWIN0SENT;
774
775 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
776 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
777 th->th_flags |= TH_URG;
778 } else
779 /*
780 * If no urgent pointer to send, then we pull
781 * the urgent pointer to the left edge of the send window
782 * so that it doesn't drift into the send window on sequence
783 * number wraparound.
784 */
785 tp->snd_up = tp->snd_una; /* drag it along */
786
787 #ifdef TCP_SIGNATURE
788 #ifdef INET6
789 if (!isipv6)
790 #endif
791 if (tp->t_flags & TF_SIGNATURE)
792 tcp_signature_compute(m, sizeof(struct ip), len, optlen,
793 (u_char *)(th + 1) + sigoff, IPSEC_DIR_OUTBOUND);
794 #endif
795
796 /*
797 * Put TCP length in extended header, and then
798 * checksum extended header and data.
799 */
800 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
801 #ifdef INET6
802 if (isipv6)
803 /*
804 * ip6_plen is not need to be filled now, and will be filled
805 * in ip6_output.
806 */
807 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
808 sizeof(struct tcphdr) + optlen + len);
809 else
810 #endif /* INET6 */
811 {
812 m->m_pkthdr.csum_flags = CSUM_TCP;
813 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
814 if (len + optlen)
815 th->th_sum = in_addword(th->th_sum,
816 htons((u_short)(optlen + len)));
817
818 /* IP version must be set here for ipv4/ipv6 checking later */
819 KASSERT(ip->ip_v == IPVERSION,
820 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
821 }
822
823 /*
824 * In transmit state, time the transmission and arrange for
825 * the retransmit. In persist state, just set snd_max.
826 */
827 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
828 tcp_seq startseq = tp->snd_nxt;
829
830 /*
831 * Advance snd_nxt over sequence space of this segment.
832 */
833 if (flags & (TH_SYN|TH_FIN)) {
834 if (flags & TH_SYN)
835 tp->snd_nxt++;
836 if (flags & TH_FIN) {
837 tp->snd_nxt++;
838 tp->t_flags |= TF_SENTFIN;
839 }
840 }
841 tp->snd_nxt += len;
842 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
843 tp->snd_max = tp->snd_nxt;
844 /*
845 * Time this transmission if not a retransmission and
846 * not currently timing anything.
847 */
848 if (tp->t_rtttime == 0) {
849 tp->t_rtttime = ticks;
850 tp->t_rtseq = startseq;
851 tcpstat.tcps_segstimed++;
852 }
853 }
854
855 /*
856 * Set retransmit timer if not currently set,
857 * and not doing a pure ack or a keep-alive probe.
858 * Initial value for retransmit timer is smoothed
859 * round-trip time + 2 * round-trip time variance.
860 * Initialize shift counter which is used for backoff
861 * of retransmit time.
862 */
863 if (!callout_active(tp->tt_rexmt) &&
864 tp->snd_nxt != tp->snd_una) {
865 if (callout_active(tp->tt_persist)) {
866 callout_stop(tp->tt_persist);
867 tp->t_rxtshift = 0;
868 }
869 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
870 tcp_timer_rexmt, tp);
871 }
872 } else {
873 /*
874 * Persist case, update snd_max but since we are in
875 * persist mode (no window) we do not update snd_nxt.
876 */
877 int xlen = len;
878 if (flags & TH_SYN)
879 ++xlen;
880 if (flags & TH_FIN) {
881 ++xlen;
882 tp->t_flags |= TF_SENTFIN;
883 }
884 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
885 tp->snd_max = tp->snd_nxt + xlen;
886 }
887
888 #ifdef TCPDEBUG
889 /*
890 * Trace.
891 */
892 if (so->so_options & SO_DEBUG)
893 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
894 #endif
895
896 /*
897 * Fill in IP length and desired time to live and
898 * send to IP level. There should be a better way
899 * to handle ttl and tos; we could keep them in
900 * the template, but need a way to checksum without them.
901 */
902 /*
903 * m->m_pkthdr.len should have been set before cksum calcuration,
904 * because in6_cksum() need it.
905 */
906 #ifdef INET6
907 if (isipv6) {
908 /*
909 * we separately set hoplimit for every segment, since the
910 * user might want to change the value via setsockopt.
911 * Also, desired default hop limit might be changed via
912 * Neighbor Discovery.
913 */
914 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb,
915 tp->t_inpcb->in6p_route.ro_rt ?
916 tp->t_inpcb->in6p_route.ro_rt->rt_ifp
917 : NULL);
918
919 /* TODO: IPv6 IP6TOS_ECT bit on */
920 error = ip6_output(m,
921 tp->t_inpcb->in6p_outputopts,
922 &tp->t_inpcb->in6p_route,
923 (so->so_options & SO_DONTROUTE), NULL, NULL,
924 tp->t_inpcb);
925 } else
926 #endif /* INET6 */
927 {
928 struct rtentry *rt;
929 ip->ip_len = m->m_pkthdr.len;
930 #ifdef INET6
931 if (INP_CHECK_SOCKAF(so, AF_INET6))
932 ip->ip_ttl = in6_selecthlim(tp->t_inpcb,
933 tp->t_inpcb->in6p_route.ro_rt ?
934 tp->t_inpcb->in6p_route.ro_rt->rt_ifp
935 : NULL);
936 else
937 #endif /* INET6 */
938 ip->ip_ttl = tp->t_inpcb->inp_ip_ttl; /* XXX */
939 ip->ip_tos = tp->t_inpcb->inp_ip_tos; /* XXX */
940 /*
941 * See if we should do MTU discovery. We do it only if the following
942 * are true:
943 * 1) we have a valid route to the destination
944 * 2) the MTU is not locked (if it is, then discovery has been
945 * disabled)
946 */
947 if (path_mtu_discovery
948 && (rt = tp->t_inpcb->inp_route.ro_rt)
949 && rt->rt_flags & RTF_UP
950 && !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
951 ip->ip_off |= IP_DF;
952 }
953 error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
954 (so->so_options & SO_DONTROUTE), 0, tp->t_inpcb);
955 }
956 if (error) {
957
958 /*
959 * We know that the packet was lost, so back out the
960 * sequence number advance, if any.
961 */
962 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
963 /*
964 * No need to check for TH_FIN here because
965 * the TF_SENTFIN flag handles that case.
966 */
967 if ((flags & TH_SYN) == 0)
968 tp->snd_nxt -= len;
969 }
970
971 out:
972 if (error == ENOBUFS) {
973 if (!callout_active(tp->tt_rexmt) &&
974 !callout_active(tp->tt_persist))
975 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
976 tcp_timer_rexmt, tp);
977 tcp_quench(tp->t_inpcb, 0);
978 return (0);
979 }
980 if (error == EMSGSIZE) {
981 /*
982 * ip_output() will have already fixed the route
983 * for us. tcp_mtudisc() will, as its last action,
984 * initiate retransmission, so it is important to
985 * not do so here.
986 */
987 tcp_mtudisc(tp->t_inpcb, 0);
988 return 0;
989 }
990 if ((error == EHOSTUNREACH || error == ENETDOWN)
991 && TCPS_HAVERCVDSYN(tp->t_state)) {
992 tp->t_softerror = error;
993 return (0);
994 }
995 return (error);
996 }
997 tcpstat.tcps_sndtotal++;
998
999 /*
1000 * Data sent (as far as we can tell).
1001 * If this advertises a larger window than any other segment,
1002 * then remember the size of the advertised window.
1003 * Any pending ACK has now been sent.
1004 */
1005 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
1006 tp->rcv_adv = tp->rcv_nxt + win;
1007 tp->last_ack_sent = tp->rcv_nxt;
1008 tp->t_flags &= ~TF_ACKNOW;
1009 if (tcp_delack_enabled)
1010 callout_stop(tp->tt_delack);
1011 #if 0
1012 /*
1013 * This completely breaks TCP if newreno is turned on. What happens
1014 * is that if delayed-acks are turned on on the receiver, this code
1015 * on the transmitter effectively destroys the TCP window, forcing
1016 * it to four packets (1.5Kx4 = 6K window).
1017 */
1018 if (sendalot && (!tcp_do_newreno || --maxburst))
1019 goto again;
1020 #endif
1021 if (sendalot)
1022 goto again;
1023 return (0);
1024 }
1025
1026 void
1027 tcp_setpersist(tp)
1028 register struct tcpcb *tp;
1029 {
1030 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
1031 int tt;
1032
1033 if (callout_active(tp->tt_rexmt))
1034 panic("tcp_setpersist: retransmit pending");
1035 /*
1036 * Start/restart persistance timer.
1037 */
1038 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
1039 TCPTV_PERSMIN, TCPTV_PERSMAX);
1040 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp);
1041 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
1042 tp->t_rxtshift++;
1043 }
Cache object: 8c60979fcd4a7e3ac71b9d392f402a0b
|