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: releng/5.2/sys/netinet/tcp_output.c 122922 2003-11-20 20:07:39Z andre $
35 */
36
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 #include "opt_tcpdebug.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/domain.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mac.h>
48 #include <sys/mbuf.h>
49 #include <sys/mutex.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/sysctl.h>
54
55 #include <net/route.h>
56
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
60 #include <netinet/in_pcb.h>
61 #include <netinet/ip_var.h>
62 #ifdef INET6
63 #include <netinet6/in6_pcb.h>
64 #include <netinet/ip6.h>
65 #include <netinet6/ip6_var.h>
66 #endif
67 #include <netinet/tcp.h>
68 #define TCPOUTFLAGS
69 #include <netinet/tcp_fsm.h>
70 #include <netinet/tcp_seq.h>
71 #include <netinet/tcp_timer.h>
72 #include <netinet/tcp_var.h>
73 #include <netinet/tcpip.h>
74 #ifdef TCPDEBUG
75 #include <netinet/tcp_debug.h>
76 #endif
77
78 #ifdef IPSEC
79 #include <netinet6/ipsec.h>
80 #endif /*IPSEC*/
81
82 #ifdef FAST_IPSEC
83 #include <netipsec/ipsec.h>
84 #define IPSEC
85 #endif /*FAST_IPSEC*/
86
87 #include <machine/in_cksum.h>
88
89 #ifdef notyet
90 extern struct mbuf *m_copypack();
91 #endif
92
93 int path_mtu_discovery = 1;
94 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
95 &path_mtu_discovery, 1, "Enable Path MTU Discovery");
96
97 int ss_fltsz = 1;
98 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
99 &ss_fltsz, 1, "Slow start flight size");
100
101 int ss_fltsz_local = 4;
102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize, CTLFLAG_RW,
103 &ss_fltsz_local, 1, "Slow start flight size for local networks");
104
105 int tcp_do_newreno = 1;
106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno, CTLFLAG_RW, &tcp_do_newreno,
107 0, "Enable NewReno Algorithms");
108
109 /*
110 * Tcp output routine: figure out what should be sent and send it.
111 */
112 int
113 tcp_output(struct tcpcb *tp)
114 {
115 struct socket *so = tp->t_inpcb->inp_socket;
116 long len, win;
117 int off, flags, error;
118 struct mbuf *m;
119 struct ip *ip = NULL;
120 struct ipovly *ipov = NULL;
121 struct tcphdr *th;
122 u_char opt[TCP_MAXOLEN];
123 unsigned ipoptlen, optlen, hdrlen;
124 int idle, sendalot;
125 #if 0
126 int maxburst = TCP_MAXBURST;
127 #endif
128 struct rmxp_tao tao;
129 #ifdef INET6
130 struct ip6_hdr *ip6 = NULL;
131 int isipv6;
132
133 bzero(&tao, sizeof(tao));
134 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
135 #endif
136
137 INP_LOCK_ASSERT(tp->t_inpcb);
138
139 /*
140 * Determine length of data that should be transmitted,
141 * and flags that will be used.
142 * If there is some data or critical controls (SYN, RST)
143 * to send, then transmit; otherwise, investigate further.
144 */
145 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
146 if (idle && (ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
147 /*
148 * We have been idle for "a while" and no acks are
149 * expected to clock out any data we send --
150 * slow start to get ack "clock" running again.
151 *
152 * Set the slow-start flight size depending on whether
153 * this is a local network or not.
154 */
155 int ss = ss_fltsz;
156 #ifdef INET6
157 if (isipv6) {
158 if (in6_localaddr(&tp->t_inpcb->in6p_faddr))
159 ss = ss_fltsz_local;
160 } else
161 #endif /* INET6 */
162 if (in_localaddr(tp->t_inpcb->inp_faddr))
163 ss = ss_fltsz_local;
164 tp->snd_cwnd = tp->t_maxseg * ss;
165 }
166 tp->t_flags &= ~TF_LASTIDLE;
167 if (idle) {
168 if (tp->t_flags & TF_MORETOCOME) {
169 tp->t_flags |= TF_LASTIDLE;
170 idle = 0;
171 }
172 }
173 again:
174 sendalot = 0;
175 off = tp->snd_nxt - tp->snd_una;
176 win = min(tp->snd_wnd, tp->snd_cwnd);
177 win = min(win, tp->snd_bwnd);
178
179 flags = tcp_outflags[tp->t_state];
180 /*
181 * Get standard flags, and add SYN or FIN if requested by 'hidden'
182 * state flags.
183 */
184 if (tp->t_flags & TF_NEEDFIN)
185 flags |= TH_FIN;
186 if (tp->t_flags & TF_NEEDSYN)
187 flags |= TH_SYN;
188
189 /*
190 * If in persist timeout with window of 0, send 1 byte.
191 * Otherwise, if window is small but nonzero
192 * and timer expired, we will send what we can
193 * and go to transmit state.
194 */
195 if (tp->t_force) {
196 if (win == 0) {
197 /*
198 * If we still have some data to send, then
199 * clear the FIN bit. Usually this would
200 * happen below when it realizes that we
201 * aren't sending all the data. However,
202 * if we have exactly 1 byte of unsent data,
203 * then it won't clear the FIN bit below,
204 * and if we are in persist state, we wind
205 * up sending the packet without recording
206 * that we sent the FIN bit.
207 *
208 * We can't just blindly clear the FIN bit,
209 * because if we don't have any more data
210 * to send then the probe will be the FIN
211 * itself.
212 */
213 if (off < so->so_snd.sb_cc)
214 flags &= ~TH_FIN;
215 win = 1;
216 } else {
217 callout_stop(tp->tt_persist);
218 tp->t_rxtshift = 0;
219 }
220 }
221
222 /*
223 * If snd_nxt == snd_max and we have transmitted a FIN, the
224 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
225 * a negative length. This can also occur when tcp opens up
226 * its congestion window while receiving additional duplicate
227 * acks after fast-retransmit because TCP will reset snd_nxt
228 * to snd_max after the fast-retransmit.
229 *
230 * In the normal retransmit-FIN-only case, however, snd_nxt will
231 * be set to snd_una, the offset will be 0, and the length may
232 * wind up 0.
233 */
234 len = (long)ulmin(so->so_snd.sb_cc, win) - off;
235
236
237 /*
238 * Lop off SYN bit if it has already been sent. However, if this
239 * is SYN-SENT state and if segment contains data and if we don't
240 * know that foreign host supports TAO, suppress sending segment.
241 */
242 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
243 flags &= ~TH_SYN;
244 off--, len++;
245 if (tcp_do_rfc1644)
246 tcp_hc_gettao(&tp->t_inpcb->inp_inc, &tao);
247 if (len > 0 && tp->t_state == TCPS_SYN_SENT &&
248 tao.tao_ccsent == 0)
249 return 0;
250 }
251
252 /*
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.
257 */
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)))) {
261 len = 0;
262 flags &= ~TH_FIN;
263 }
264
265 if (len < 0) {
266 /*
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.
275 */
276 len = 0;
277 if (win == 0) {
278 callout_stop(tp->tt_rexmt);
279 tp->t_rxtshift = 0;
280 tp->snd_nxt = tp->snd_una;
281 if (!callout_active(tp->tt_persist))
282 tcp_setpersist(tp);
283 }
284 }
285
286 /*
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.
290 */
291 if (len > tp->t_maxseg) {
292 len = tp->t_maxseg;
293 sendalot = 1;
294 }
295 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
296 flags &= ~TH_FIN;
297
298 win = sbspace(&so->so_rcv);
299
300 /*
301 * Sender silly window avoidance. We transmit under the following
302 * conditions when len is non-zero:
303 *
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
311 */
312 if (len) {
313 if (len == tp->t_maxseg)
314 goto send;
315 /*
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
319 *
320 * note: the len + off check is almost certainly unnecessary.
321 */
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) {
326 goto send;
327 }
328 if (tp->t_force) /* typ. timeout case */
329 goto send;
330 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
331 goto send;
332 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
333 goto send;
334 }
335
336 /*
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.
342 * Skip this if the connection is in T/TCP half-open state.
343 */
344 if (win > 0 && !(tp->t_flags & TF_NEEDSYN)) {
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
411 send:
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->t_inpcb->inp_inc));
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 if (tp->t_inpcb->inp_options)
548 ipoptlen = tp->t_inpcb->inp_options->m_len -
549 offsetof(struct ipoption, ipopt_list);
550 else
551 ipoptlen = 0;
552 #ifdef IPSEC
553 ipoptlen += ipsec_hdrsiz_tcp(tp);
554 #endif
555
556 /*
557 * Adjust data length if insertion of options will
558 * bump the packet length beyond the t_maxopd length.
559 * Clear the FIN bit because we cut off the tail of
560 * the segment.
561 */
562 if (len + optlen + ipoptlen > tp->t_maxopd) {
563 /*
564 * If there is still more to send, don't close the connection.
565 */
566 flags &= ~TH_FIN;
567 len = tp->t_maxopd - optlen - ipoptlen;
568 sendalot = 1;
569 }
570
571 /*#ifdef DIAGNOSTIC*/
572 #ifdef INET6
573 if (max_linkhdr + hdrlen > MCLBYTES)
574 #else
575 if (max_linkhdr + hdrlen > MHLEN)
576 #endif
577 panic("tcphdr too big");
578 /*#endif*/
579
580 /*
581 * Grab a header mbuf, attaching a copy of data to
582 * be transmitted, and initialize the header from
583 * the template for sends on this connection.
584 */
585 if (len) {
586 if (tp->t_force && len == 1)
587 tcpstat.tcps_sndprobe++;
588 else if (SEQ_LT(tp->snd_nxt, tp->snd_max)) {
589 tcpstat.tcps_sndrexmitpack++;
590 tcpstat.tcps_sndrexmitbyte += len;
591 } else {
592 tcpstat.tcps_sndpack++;
593 tcpstat.tcps_sndbyte += len;
594 }
595 #ifdef notyet
596 if ((m = m_copypack(so->so_snd.sb_mb, off,
597 (int)len, max_linkhdr + hdrlen)) == 0) {
598 error = ENOBUFS;
599 goto out;
600 }
601 /*
602 * m_copypack left space for our hdr; use it.
603 */
604 m->m_len += hdrlen;
605 m->m_data -= hdrlen;
606 #else
607 MGETHDR(m, M_DONTWAIT, MT_HEADER);
608 if (m == NULL) {
609 error = ENOBUFS;
610 goto out;
611 }
612 #ifdef INET6
613 if (MHLEN < hdrlen + max_linkhdr) {
614 MCLGET(m, M_DONTWAIT);
615 if ((m->m_flags & M_EXT) == 0) {
616 m_freem(m);
617 error = ENOBUFS;
618 goto out;
619 }
620 }
621 #endif
622 m->m_data += max_linkhdr;
623 m->m_len = hdrlen;
624 if (len <= MHLEN - hdrlen - max_linkhdr) {
625 m_copydata(so->so_snd.sb_mb, off, (int) len,
626 mtod(m, caddr_t) + hdrlen);
627 m->m_len += len;
628 } else {
629 m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
630 if (m->m_next == 0) {
631 (void) m_free(m);
632 error = ENOBUFS;
633 goto out;
634 }
635 }
636 #endif
637 /*
638 * If we're sending everything we've got, set PUSH.
639 * (This will keep happy those implementations which only
640 * give data to the user when a buffer fills or
641 * a PUSH comes in.)
642 */
643 if (off + len == so->so_snd.sb_cc)
644 flags |= TH_PUSH;
645 } else {
646 if (tp->t_flags & TF_ACKNOW)
647 tcpstat.tcps_sndacks++;
648 else if (flags & (TH_SYN|TH_FIN|TH_RST))
649 tcpstat.tcps_sndctrl++;
650 else if (SEQ_GT(tp->snd_up, tp->snd_una))
651 tcpstat.tcps_sndurg++;
652 else
653 tcpstat.tcps_sndwinup++;
654
655 MGETHDR(m, M_DONTWAIT, MT_HEADER);
656 if (m == NULL) {
657 error = ENOBUFS;
658 goto out;
659 }
660 #ifdef INET6
661 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
662 MHLEN >= hdrlen) {
663 MH_ALIGN(m, hdrlen);
664 } else
665 #endif
666 m->m_data += max_linkhdr;
667 m->m_len = hdrlen;
668 }
669 m->m_pkthdr.rcvif = (struct ifnet *)0;
670 #ifdef MAC
671 mac_create_mbuf_from_socket(so, m);
672 #endif
673 #ifdef INET6
674 if (isipv6) {
675 ip6 = mtod(m, struct ip6_hdr *);
676 th = (struct tcphdr *)(ip6 + 1);
677 tcpip_fillheaders(tp->t_inpcb, ip6, th);
678 } else
679 #endif /* INET6 */
680 {
681 ip = mtod(m, struct ip *);
682 ipov = (struct ipovly *)ip;
683 th = (struct tcphdr *)(ip + 1);
684 tcpip_fillheaders(tp->t_inpcb, ip, th);
685 }
686
687 /*
688 * Fill in fields, remembering maximum advertised
689 * window for use in delaying messages about window sizes.
690 * If resending a FIN, be sure not to use a new sequence number.
691 */
692 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
693 tp->snd_nxt == tp->snd_max)
694 tp->snd_nxt--;
695 /*
696 * If we are doing retransmissions, then snd_nxt will
697 * not reflect the first unsent octet. For ACK only
698 * packets, we do not want the sequence number of the
699 * retransmitted packet, we want the sequence number
700 * of the next unsent octet. So, if there is no data
701 * (and no SYN or FIN), use snd_max instead of snd_nxt
702 * when filling in ti_seq. But if we are in persist
703 * state, snd_max might reflect one byte beyond the
704 * right edge of the window, so use snd_nxt in that
705 * case, since we know we aren't doing a retransmission.
706 * (retransmit and persist are mutually exclusive...)
707 */
708 if (len || (flags & (TH_SYN|TH_FIN))
709 || callout_active(tp->tt_persist))
710 th->th_seq = htonl(tp->snd_nxt);
711 else
712 th->th_seq = htonl(tp->snd_max);
713 th->th_ack = htonl(tp->rcv_nxt);
714 if (optlen) {
715 bcopy(opt, th + 1, optlen);
716 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
717 }
718 th->th_flags = flags;
719 /*
720 * Calculate receive window. Don't shrink window,
721 * but avoid silly window syndrome.
722 */
723 if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
724 win = 0;
725 if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
726 win = (long)(tp->rcv_adv - tp->rcv_nxt);
727 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
728 win = (long)TCP_MAXWIN << tp->rcv_scale;
729 th->th_win = htons((u_short) (win>>tp->rcv_scale));
730
731
732 /*
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
738 * the connection.
739 */
740 if (win == 0)
741 tp->t_flags |= TF_RXWIN0SENT;
742 else
743 tp->t_flags &= ~TF_RXWIN0SENT;
744 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
745 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
746 th->th_flags |= TH_URG;
747 } else
748 /*
749 * If no urgent pointer to send, then we pull
750 * the urgent pointer to the left edge of the send window
751 * so that it doesn't drift into the send window on sequence
752 * number wraparound.
753 */
754 tp->snd_up = tp->snd_una; /* drag it along */
755
756 /*
757 * Put TCP length in extended header, and then
758 * checksum extended header and data.
759 */
760 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
761 #ifdef INET6
762 if (isipv6)
763 /*
764 * ip6_plen is not need to be filled now, and will be filled
765 * in ip6_output.
766 */
767 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
768 sizeof(struct tcphdr) + optlen + len);
769 else
770 #endif /* INET6 */
771 {
772 m->m_pkthdr.csum_flags = CSUM_TCP;
773 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
774 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
775 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen));
776
777 /* IP version must be set here for ipv4/ipv6 checking later */
778 KASSERT(ip->ip_v == IPVERSION,
779 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
780 }
781
782 /*
783 * In transmit state, time the transmission and arrange for
784 * the retransmit. In persist state, just set snd_max.
785 */
786 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
787 tcp_seq startseq = tp->snd_nxt;
788
789 /*
790 * Advance snd_nxt over sequence space of this segment.
791 */
792 if (flags & (TH_SYN|TH_FIN)) {
793 if (flags & TH_SYN)
794 tp->snd_nxt++;
795 if (flags & TH_FIN) {
796 tp->snd_nxt++;
797 tp->t_flags |= TF_SENTFIN;
798 }
799 }
800 tp->snd_nxt += len;
801 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
802 tp->snd_max = tp->snd_nxt;
803 /*
804 * Time this transmission if not a retransmission and
805 * not currently timing anything.
806 */
807 if (tp->t_rtttime == 0) {
808 tp->t_rtttime = ticks;
809 tp->t_rtseq = startseq;
810 tcpstat.tcps_segstimed++;
811 }
812 }
813
814 /*
815 * Set retransmit timer if not currently set,
816 * and not doing a pure ack or a keep-alive probe.
817 * Initial value for retransmit timer is smoothed
818 * round-trip time + 2 * round-trip time variance.
819 * Initialize shift counter which is used for backoff
820 * of retransmit time.
821 */
822 if (!callout_active(tp->tt_rexmt) &&
823 tp->snd_nxt != tp->snd_una) {
824 if (callout_active(tp->tt_persist)) {
825 callout_stop(tp->tt_persist);
826 tp->t_rxtshift = 0;
827 }
828 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
829 tcp_timer_rexmt, tp);
830 }
831 } else {
832 /*
833 * Persist case, update snd_max but since we are in
834 * persist mode (no window) we do not update snd_nxt.
835 */
836 int xlen = len;
837 if (flags & TH_SYN)
838 ++xlen;
839 if (flags & TH_FIN) {
840 ++xlen;
841 tp->t_flags |= TF_SENTFIN;
842 }
843 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
844 tp->snd_max = tp->snd_nxt + len;
845 }
846
847 #ifdef TCPDEBUG
848 /*
849 * Trace.
850 */
851 if (so->so_options & SO_DEBUG) {
852 u_short save = ipov->ih_len;
853 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */);
854 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
855 ipov->ih_len = save;
856 }
857 #endif
858
859 /*
860 * Fill in IP length and desired time to live and
861 * send to IP level. There should be a better way
862 * to handle ttl and tos; we could keep them in
863 * the template, but need a way to checksum without them.
864 */
865 /*
866 * m->m_pkthdr.len should have been set before cksum calcuration,
867 * because in6_cksum() need it.
868 */
869 #ifdef INET6
870 if (isipv6) {
871 /*
872 * we separately set hoplimit for every segment, since the
873 * user might want to change the value via setsockopt.
874 * Also, desired default hop limit might be changed via
875 * Neighbor Discovery.
876 */
877 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL);
878
879 /* TODO: IPv6 IP6TOS_ECT bit on */
880 #if defined(IPSEC) && !defined(FAST_IPSEC)
881 if (ipsec_setsocket(m, so) != 0) {
882 m_freem(m);
883 error = ENOBUFS;
884 goto out;
885 }
886 #endif /*IPSEC*/
887 error = ip6_output(m,
888 tp->t_inpcb->in6p_outputopts, NULL,
889 (so->so_options & SO_DONTROUTE), NULL, NULL,
890 tp->t_inpcb);
891 } else
892 #endif /* INET6 */
893 {
894 ip->ip_len = m->m_pkthdr.len;
895 #ifdef INET6
896 if (INP_CHECK_SOCKAF(so, AF_INET6))
897 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL);
898 #endif /* INET6 */
899 /*
900 * If we do path MTU discovery, then we set DF on every packet.
901 * This might not be the best thing to do according to RFC3390
902 * Section 2. However the tcp hostcache migitates the problem
903 * so it affects only the first tcp connection with a host.
904 */
905 if (path_mtu_discovery)
906 ip->ip_off |= IP_DF;
907
908 error = ip_output(m, tp->t_inpcb->inp_options, NULL,
909 (so->so_options & SO_DONTROUTE), 0, tp->t_inpcb);
910 }
911 if (error) {
912
913 /*
914 * We know that the packet was lost, so back out the
915 * sequence number advance, if any.
916 */
917 if (tp->t_force == 0 || !callout_active(tp->tt_persist)) {
918 /*
919 * No need to check for TH_FIN here because
920 * the TF_SENTFIN flag handles that case.
921 */
922 if ((flags & TH_SYN) == 0)
923 tp->snd_nxt -= len;
924 }
925
926 out:
927 if (error == ENOBUFS) {
928 if (!callout_active(tp->tt_rexmt) &&
929 !callout_active(tp->tt_persist))
930 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
931 tcp_timer_rexmt, tp);
932 tcp_quench(tp->t_inpcb, 0);
933 return (0);
934 }
935 if (error == EMSGSIZE) {
936 /*
937 * ip_output() will have already fixed the route
938 * for us. tcp_mtudisc() will, as its last action,
939 * initiate retransmission, so it is important to
940 * not do so here.
941 */
942 tcp_mtudisc(tp->t_inpcb, 0);
943 return 0;
944 }
945 if ((error == EHOSTUNREACH || error == ENETDOWN)
946 && TCPS_HAVERCVDSYN(tp->t_state)) {
947 tp->t_softerror = error;
948 return (0);
949 }
950 return (error);
951 }
952 tcpstat.tcps_sndtotal++;
953
954 /*
955 * Data sent (as far as we can tell).
956 * If this advertises a larger window than any other segment,
957 * then remember the size of the advertised window.
958 * Any pending ACK has now been sent.
959 */
960 if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
961 tp->rcv_adv = tp->rcv_nxt + win;
962 tp->last_ack_sent = tp->rcv_nxt;
963 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK);
964 if (callout_active(tp->tt_delack))
965 callout_stop(tp->tt_delack);
966 #if 0
967 /*
968 * This completely breaks TCP if newreno is turned on. What happens
969 * is that if delayed-acks are turned on on the receiver, this code
970 * on the transmitter effectively destroys the TCP window, forcing
971 * it to four packets (1.5Kx4 = 6K window).
972 */
973 if (sendalot && (!tcp_do_newreno || --maxburst))
974 goto again;
975 #endif
976 if (sendalot)
977 goto again;
978 return (0);
979 }
980
981 void
982 tcp_setpersist(tp)
983 register struct tcpcb *tp;
984 {
985 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
986 int tt;
987
988 if (callout_active(tp->tt_rexmt))
989 panic("tcp_setpersist: retransmit pending");
990 /*
991 * Start/restart persistance timer.
992 */
993 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
994 TCPTV_PERSMIN, TCPTV_PERSMAX);
995 callout_reset(tp->tt_persist, tt, tcp_timer_persist, tp);
996 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
997 tp->t_rxtshift++;
998 }
Cache object: 3d385f65d2b2b9c070847c7fa48e4da6
|