2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_output.c 8.4 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
37 #include "opt_ipsec.h"
38 #include "opt_tcpdebug.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/domain.h>
43 #include <sys/hhook.h>
44 #include <sys/kernel.h>
47 #include <sys/mutex.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
54 #include <net/route.h>
57 #include <netinet/cc.h>
58 #include <netinet/in.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/ip.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/ip_options.h>
65 #include <netinet6/in6_pcb.h>
66 #include <netinet/ip6.h>
67 #include <netinet6/ip6_var.h>
70 #include <netinet/tcp_fsm.h>
71 #include <netinet/tcp_seq.h>
72 #include <netinet/tcp_timer.h>
73 #include <netinet/tcp_var.h>
74 #include <netinet/tcpip.h>
76 #include <netinet/tcp_debug.h>
79 #include <netinet/tcp_offload.h>
83 #include <netipsec/ipsec.h>
86 #include <machine/in_cksum.h>
88 #include <security/mac/mac_framework.h>
90 VNET_DEFINE(int, path_mtu_discovery) = 1;
91 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
92 &VNET_NAME(path_mtu_discovery), 1,
93 "Enable Path MTU Discovery");
95 VNET_DEFINE(int, tcp_do_tso) = 1;
96 #define V_tcp_do_tso VNET(tcp_do_tso)
97 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_RW,
98 &VNET_NAME(tcp_do_tso), 0,
99 "Enable TCP Segmentation Offload");
101 VNET_DEFINE(int, tcp_sendspace) = 1024*32;
102 #define V_tcp_sendspace VNET(tcp_sendspace)
103 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
104 &VNET_NAME(tcp_sendspace), 0, "Initial send socket buffer size");
106 VNET_DEFINE(int, tcp_do_autosndbuf) = 1;
107 #define V_tcp_do_autosndbuf VNET(tcp_do_autosndbuf)
108 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_RW,
109 &VNET_NAME(tcp_do_autosndbuf), 0,
110 "Enable automatic send buffer sizing");
112 VNET_DEFINE(int, tcp_autosndbuf_inc) = 8*1024;
113 #define V_tcp_autosndbuf_inc VNET(tcp_autosndbuf_inc)
114 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_RW,
115 &VNET_NAME(tcp_autosndbuf_inc), 0,
116 "Incrementor step size of automatic send buffer");
118 VNET_DEFINE(int, tcp_autosndbuf_max) = 2*1024*1024;
119 #define V_tcp_autosndbuf_max VNET(tcp_autosndbuf_max)
120 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_RW,
121 &VNET_NAME(tcp_autosndbuf_max), 0,
122 "Max size of automatic send buffer");
124 static void inline hhook_run_tcp_est_out(struct tcpcb *tp,
125 struct tcphdr *th, struct tcpopt *to,
127 static void inline cc_after_idle(struct tcpcb *tp);
130 * Wrapper for the TCP established ouput helper hook.
133 hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th,
134 struct tcpopt *to, long len, int tso)
136 struct tcp_hhook_data hhook_data;
138 if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) {
142 hhook_data.len = len;
143 hhook_data.tso = tso;
145 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data,
151 * CC wrapper hook functions
154 cc_after_idle(struct tcpcb *tp)
156 INP_WLOCK_ASSERT(tp->t_inpcb);
158 if (CC_ALGO(tp)->after_idle != NULL)
159 CC_ALGO(tp)->after_idle(tp->ccv);
163 * Tcp output routine: figure out what should be sent and send it.
166 tcp_output(struct tcpcb *tp)
168 struct socket *so = tp->t_inpcb->inp_socket;
169 long len, recwin, sendwin;
170 int off, flags, error = 0; /* Keep compiler happy */
172 struct ip *ip = NULL;
173 struct ipovly *ipov = NULL;
175 u_char opt[TCP_MAXOLEN];
176 unsigned ipoptlen, optlen, hdrlen;
178 unsigned ipsec_optlen = 0;
181 int sack_rxmit, sack_bytes_rxmt;
186 int maxburst = TCP_MAXBURST;
189 struct ip6_hdr *ip6 = NULL;
192 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
195 INP_WLOCK_ASSERT(tp->t_inpcb);
198 if (tp->t_flags & TF_TOE)
199 return (tcp_offload_output(tp));
203 * Determine length of data that should be transmitted,
204 * and flags that will be used.
205 * If there is some data or critical controls (SYN, RST)
206 * to send, then transmit; otherwise, investigate further.
208 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
209 if (idle && ticks - tp->t_rcvtime >= tp->t_rxtcur)
211 tp->t_flags &= ~TF_LASTIDLE;
213 if (tp->t_flags & TF_MORETOCOME) {
214 tp->t_flags |= TF_LASTIDLE;
220 * If we've recently taken a timeout, snd_max will be greater than
221 * snd_nxt. There may be SACK information that allows us to avoid
222 * resending already delivered data. Adjust snd_nxt accordingly.
224 if ((tp->t_flags & TF_SACK_PERMIT) &&
225 SEQ_LT(tp->snd_nxt, tp->snd_max))
230 off = tp->snd_nxt - tp->snd_una;
231 sendwin = min(tp->snd_wnd, tp->snd_cwnd);
233 flags = tcp_outflags[tp->t_state];
235 * Send any SACK-generated retransmissions. If we're explicitly trying
236 * to send out new data (when sendalot is 1), bypass this function.
237 * If we retransmit in fast recovery mode, decrement snd_cwnd, since
238 * we're replacing a (future) new transmission with a retransmission
239 * now, and we previously incremented snd_cwnd in tcp_input().
242 * Still in sack recovery , reset rxmit flag to zero.
248 if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) &&
249 (p = tcp_sack_output(tp, &sack_bytes_rxmt))) {
252 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt;
255 /* Do not retransmit SACK segments beyond snd_recover */
256 if (SEQ_GT(p->end, tp->snd_recover)) {
258 * (At least) part of sack hole extends beyond
259 * snd_recover. Check to see if we can rexmit data
262 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) {
264 * Can't rexmit any more data for this hole.
265 * That data will be rexmitted in the next
266 * sack recovery episode, when snd_recover
267 * moves past p->rxmit.
270 goto after_sack_rexmit;
272 /* Can rexmit part of the current hole */
273 len = ((long)ulmin(cwin,
274 tp->snd_recover - p->rxmit));
276 len = ((long)ulmin(cwin, p->end - p->rxmit));
277 off = p->rxmit - tp->snd_una;
278 KASSERT(off >= 0,("%s: sack block to the left of una : %d",
283 TCPSTAT_INC(tcps_sack_rexmits);
284 TCPSTAT_ADD(tcps_sack_rexmit_bytes,
285 min(len, tp->t_maxseg));
290 * Get standard flags, and add SYN or FIN if requested by 'hidden'
293 if (tp->t_flags & TF_NEEDFIN)
295 if (tp->t_flags & TF_NEEDSYN)
298 SOCKBUF_LOCK(&so->so_snd);
300 * If in persist timeout with window of 0, send 1 byte.
301 * Otherwise, if window is small but nonzero
302 * and timer expired, we will send what we can
303 * and go to transmit state.
305 if (tp->t_flags & TF_FORCEDATA) {
308 * If we still have some data to send, then
309 * clear the FIN bit. Usually this would
310 * happen below when it realizes that we
311 * aren't sending all the data. However,
312 * if we have exactly 1 byte of unsent data,
313 * then it won't clear the FIN bit below,
314 * and if we are in persist state, we wind
315 * up sending the packet without recording
316 * that we sent the FIN bit.
318 * We can't just blindly clear the FIN bit,
319 * because if we don't have any more data
320 * to send then the probe will be the FIN
323 if (off < so->so_snd.sb_cc)
327 tcp_timer_activate(tp, TT_PERSIST, 0);
333 * If snd_nxt == snd_max and we have transmitted a FIN, the
334 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
335 * a negative length. This can also occur when TCP opens up
336 * its congestion window while receiving additional duplicate
337 * acks after fast-retransmit because TCP will reset snd_nxt
338 * to snd_max after the fast-retransmit.
340 * In the normal retransmit-FIN-only case, however, snd_nxt will
341 * be set to snd_una, the offset will be 0, and the length may
344 * If sack_rxmit is true we are retransmitting from the scoreboard
345 * in which case len is already set.
347 if (sack_rxmit == 0) {
348 if (sack_bytes_rxmt == 0)
349 len = ((long)ulmin(so->so_snd.sb_cc, sendwin) - off);
354 * We are inside of a SACK recovery episode and are
355 * sending new data, having retransmitted all the
356 * data possible in the scoreboard.
358 len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd)
361 * Don't remove this (len > 0) check !
362 * We explicitly check for len > 0 here (although it
363 * isn't really necessary), to work around a gcc
364 * optimization issue - to force gcc to compute
365 * len above. Without this check, the computation
366 * of len is bungled by the optimizer.
369 cwin = tp->snd_cwnd -
370 (tp->snd_nxt - tp->sack_newdata) -
374 len = lmin(len, cwin);
380 * Lop off SYN bit if it has already been sent. However, if this
381 * is SYN-SENT state and if segment contains data and if we don't
382 * know that foreign host supports TAO, suppress sending segment.
384 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
385 if (tp->t_state != TCPS_SYN_RECEIVED)
391 * Be careful not to send data and/or FIN on SYN segments.
392 * This measure is needed to prevent interoperability problems
393 * with not fully conformant TCP implementations.
395 if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) {
402 * If FIN has been sent but not acked,
403 * but we haven't been called to retransmit,
404 * len will be < 0. Otherwise, window shrank
405 * after we sent into it. If window shrank to 0,
406 * cancel pending retransmit, pull snd_nxt back
407 * to (closed) window, and set the persist timer
408 * if it isn't already going. If the window didn't
409 * close completely, just wait for an ACK.
413 tcp_timer_activate(tp, TT_REXMT, 0);
415 tp->snd_nxt = tp->snd_una;
416 if (!tcp_timer_active(tp, TT_PERSIST))
421 /* len will be >= 0 after this point. */
422 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
425 * Automatic sizing of send socket buffer. Often the send buffer
426 * size is not optimally adjusted to the actual network conditions
427 * at hand (delay bandwidth product). Setting the buffer size too
428 * small limits throughput on links with high bandwidth and high
429 * delay (eg. trans-continental/oceanic links). Setting the
430 * buffer size too big consumes too much real kernel memory,
431 * especially with many connections on busy servers.
433 * The criteria to step up the send buffer one notch are:
434 * 1. receive window of remote host is larger than send buffer
435 * (with a fudge factor of 5/4th);
436 * 2. send buffer is filled to 7/8th with data (so we actually
437 * have data to make use of it);
438 * 3. send buffer fill has not hit maximal automatic size;
439 * 4. our send window (slow start and cogestion controlled) is
440 * larger than sent but unacknowledged data in send buffer.
442 * The remote host receive window scaling factor may limit the
443 * growing of the send buffer before it reaches its allowed
446 * It scales directly with slow start or congestion window
447 * and does at most one step per received ACK. This fast
448 * scaling has the drawback of growing the send buffer beyond
449 * what is strictly necessary to make full use of a given
450 * delay*bandwith product. However testing has shown this not
451 * to be much of an problem. At worst we are trading wasting
452 * of available bandwith (the non-use of it) for wasting some
453 * socket buffer memory.
455 * TODO: Shrink send buffer during idle periods together
456 * with congestion window. Requires another timer. Has to
457 * wait for upcoming tcp timer rewrite.
459 if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) {
460 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat &&
461 so->so_snd.sb_cc >= (so->so_snd.sb_hiwat / 8 * 7) &&
462 so->so_snd.sb_cc < V_tcp_autosndbuf_max &&
463 sendwin >= (so->so_snd.sb_cc - (tp->snd_nxt - tp->snd_una))) {
464 if (!sbreserve_locked(&so->so_snd,
465 min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc,
466 V_tcp_autosndbuf_max), so, curthread))
467 so->so_snd.sb_flags &= ~SB_AUTOSIZE;
472 * Decide if we can use TCP Segmentation Offloading (if supported by
475 * TSO may only be used if we are in a pure bulk sending state. The
476 * presence of TCP-MD5, SACK retransmits, SACK advertizements and
477 * IP options prevent using TSO. With TSO the TCP header is the same
478 * (except for the sequence number) for all generated packets. This
479 * makes it impossible to transmit any options which vary per generated
484 * Pre-calculate here as we save another lookup into the darknesses
485 * of IPsec that way and can actually decide if TSO is ok.
487 ipsec_optlen = ipsec_hdrsiz_tcp(tp);
489 if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg &&
490 ((tp->t_flags & TF_SIGNATURE) == 0) &&
491 tp->rcv_numsacks == 0 && sack_rxmit == 0 &&
495 tp->t_inpcb->inp_options == NULL &&
496 tp->t_inpcb->in6p_options == NULL)
500 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc))
503 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
507 recwin = sbspace(&so->so_rcv);
510 * Sender silly window avoidance. We transmit under the following
511 * conditions when len is non-zero:
513 * - We have a full segment (or more with TSO)
514 * - This is the last buffer in a write()/send() and we are
515 * either idle or running NODELAY
516 * - we've timed out (e.g. persist timer)
517 * - we have more then 1/2 the maximum send window's worth of
518 * data (receiver may be limited the window size)
519 * - we need to retransmit
522 if (len >= tp->t_maxseg)
525 * NOTE! on localhost connections an 'ack' from the remote
526 * end may occur synchronously with the output and cause
527 * us to flush a buffer queued with moretocome. XXX
529 * note: the len + off check is almost certainly unnecessary.
531 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
532 (idle || (tp->t_flags & TF_NODELAY)) &&
533 len + off >= so->so_snd.sb_cc &&
534 (tp->t_flags & TF_NOPUSH) == 0) {
537 if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */
539 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
541 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
548 * Sending of standalone window updates.
550 * Window updates are important when we close our window due to a
551 * full socket buffer and are opening it again after the application
552 * reads data from it. Once the window has opened again and the
553 * remote end starts to send again the ACK clock takes over and
554 * provides the most current window information.
556 * We must avoid the silly window syndrome whereas every read
557 * from the receive buffer, no matter how small, causes a window
558 * update to be sent. We also should avoid sending a flurry of
559 * window updates when the socket buffer had queued a lot of data
560 * and the application is doing small reads.
562 * Prevent a flurry of pointless window updates by only sending
563 * an update when we can increase the advertized window by more
564 * than 1/4th of the socket buffer capacity. When the buffer is
565 * getting full or is very small be more aggressive and send an
566 * update whenever we can increase by two mss sized segments.
567 * In all other situations the ACK's to new incoming data will
568 * carry further window increases.
570 * Don't send an independent window update if a delayed
571 * ACK is pending (it will get piggy-backed on it) or the
572 * remote side already has done a half-close and won't send
573 * more data. Skip this if the connection is in T/TCP
576 if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) &&
577 !(tp->t_flags & TF_DELACK) &&
578 !TCPS_HAVERCVDFIN(tp->t_state)) {
580 * "adv" is the amount we could increase the window,
581 * taking into account that we are limited by
582 * TCP_MAXWIN << tp->rcv_scale.
587 adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale);
588 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) {
589 oldwin = (tp->rcv_adv - tp->rcv_nxt);
595 * If the new window size ends up being the same as the old
596 * size when it is scaled, then don't force a window update.
598 if (oldwin >> tp->rcv_scale == (adv + oldwin) >> tp->rcv_scale)
601 if (adv >= (long)(2 * tp->t_maxseg) &&
602 (adv >= (long)(so->so_rcv.sb_hiwat / 4) ||
603 recwin <= (long)(so->so_rcv.sb_hiwat / 8) ||
604 so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg))
610 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
611 * is also a catch-all for the retransmit timer timeout case.
613 if (tp->t_flags & TF_ACKNOW)
615 if ((flags & TH_RST) ||
616 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
618 if (SEQ_GT(tp->snd_up, tp->snd_una))
621 * If our state indicates that FIN should be sent
622 * and we have not yet done so, then we need to send.
624 if (flags & TH_FIN &&
625 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
628 * In SACK, it is possible for tcp_output to fail to send a segment
629 * after the retransmission timer has been turned off. Make sure
630 * that the retransmission timer is set.
632 if ((tp->t_flags & TF_SACK_PERMIT) &&
633 SEQ_GT(tp->snd_max, tp->snd_una) &&
634 !tcp_timer_active(tp, TT_REXMT) &&
635 !tcp_timer_active(tp, TT_PERSIST)) {
636 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
640 * TCP window updates are not reliable, rather a polling protocol
641 * using ``persist'' packets is used to insure receipt of window
642 * updates. The three ``states'' for the output side are:
643 * idle not doing retransmits or persists
644 * persisting to move a small or zero window
645 * (re)transmitting and thereby not persisting
647 * tcp_timer_active(tp, TT_PERSIST)
648 * is true when we are in persist state.
649 * (tp->t_flags & TF_FORCEDATA)
650 * is set when we are called to send a persist packet.
651 * tcp_timer_active(tp, TT_REXMT)
652 * is set when we are retransmitting
653 * The output side is idle when both timers are zero.
655 * If send window is too small, there is data to transmit, and no
656 * retransmit or persist is pending, then go to persist state.
657 * If nothing happens soon, send when timer expires:
658 * if window is nonzero, transmit what we can,
659 * otherwise force out a byte.
661 if (so->so_snd.sb_cc && !tcp_timer_active(tp, TT_REXMT) &&
662 !tcp_timer_active(tp, TT_PERSIST)) {
668 * No reason to send a segment, just return.
671 SOCKBUF_UNLOCK(&so->so_snd);
675 SOCKBUF_LOCK_ASSERT(&so->so_snd);
677 * Before ESTABLISHED, force sending of initial options
678 * unless TCP set not to do any options.
679 * NOTE: we assume that the IP/TCP header plus TCP options
680 * always fit in a single mbuf, leaving room for a maximum
682 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
687 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
690 hdrlen = sizeof (struct tcpiphdr);
693 * Compute options for segment.
694 * We only have to care about SYN and established connection
695 * segments. Options for SYN-ACK segments are handled in TCP
698 if ((tp->t_flags & TF_NOOPT) == 0) {
700 /* Maximum segment size. */
701 if (flags & TH_SYN) {
702 tp->snd_nxt = tp->iss;
703 to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc);
704 to.to_flags |= TOF_MSS;
706 /* Window scaling. */
707 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) {
708 to.to_wscale = tp->request_r_scale;
709 to.to_flags |= TOF_SCALE;
712 if ((tp->t_flags & TF_RCVD_TSTMP) ||
713 ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) {
714 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
715 to.to_tsecr = tp->ts_recent;
716 to.to_flags |= TOF_TS;
717 /* Set receive buffer autosizing timestamp. */
718 if (tp->rfbuf_ts == 0 &&
719 (so->so_rcv.sb_flags & SB_AUTOSIZE))
720 tp->rfbuf_ts = tcp_ts_getticks();
722 /* Selective ACK's. */
723 if (tp->t_flags & TF_SACK_PERMIT) {
725 to.to_flags |= TOF_SACKPERM;
726 else if (TCPS_HAVEESTABLISHED(tp->t_state) &&
727 (tp->t_flags & TF_SACK_PERMIT) &&
728 tp->rcv_numsacks > 0) {
729 to.to_flags |= TOF_SACK;
730 to.to_nsacks = tp->rcv_numsacks;
731 to.to_sacks = (u_char *)tp->sackblks;
735 /* TCP-MD5 (RFC2385). */
736 if (tp->t_flags & TF_SIGNATURE)
737 to.to_flags |= TOF_SIGNATURE;
738 #endif /* TCP_SIGNATURE */
740 /* Processing the options. */
741 hdrlen += optlen = tcp_addoptions(&to, opt);
746 ipoptlen = ip6_optlen(tp->t_inpcb);
749 if (tp->t_inpcb->inp_options)
750 ipoptlen = tp->t_inpcb->inp_options->m_len -
751 offsetof(struct ipoption, ipopt_list);
755 ipoptlen += ipsec_optlen;
759 * Adjust data length if insertion of options will
760 * bump the packet length beyond the t_maxopd length.
761 * Clear the FIN bit because we cut off the tail of
764 if (len + optlen + ipoptlen > tp->t_maxopd) {
769 u_int if_hw_tsomaxsegcount;
770 u_int if_hw_tsomaxsegsize;
775 /* extract TSO information */
776 if_hw_tsomax = tp->t_tsomax;
777 if_hw_tsomaxsegcount = tp->t_tsomaxsegcount;
778 if_hw_tsomaxsegsize = tp->t_tsomaxsegsize;
781 * Limit a TSO burst to prevent it from
782 * overflowing or exceeding the maximum length
783 * allowed by the network interface:
785 KASSERT(ipoptlen == 0,
786 ("%s: TSO can't do IP options", __func__));
789 * Check if we should limit by maximum payload
792 if (if_hw_tsomax != 0) {
793 /* compute maximum TSO length */
794 max_len = (if_hw_tsomax - hdrlen);
797 } else if (len > max_len) {
804 * Check if we should limit by maximum segment
807 if (if_hw_tsomaxsegcount != 0 &&
808 if_hw_tsomaxsegsize != 0) {
810 mb = sbsndmbuf(&so->so_snd, off, &moff);
812 while (mb != NULL && max_len < len) {
817 * Get length of mbuf fragment
818 * and how many hardware frags,
819 * rounded up, it would use:
821 mlen = (mb->m_len - moff);
822 frags = howmany(mlen,
823 if_hw_tsomaxsegsize);
825 /* Handle special case: Zero Length Mbuf */
830 * Check if the fragment limit
831 * will be reached or exceeded:
833 if (frags >= if_hw_tsomaxsegcount) {
835 if_hw_tsomaxsegcount *
836 if_hw_tsomaxsegsize);
840 if_hw_tsomaxsegcount -= frags;
846 } else if (len > max_len) {
853 * Prevent the last segment from being
854 * fractional unless the send sockbuf can be
857 max_len = (tp->t_maxopd - optlen);
858 if ((off + len) < so->so_snd.sb_cc) {
859 moff = len % max_len;
867 * In case there are too many small fragments
870 if (len <= max_len) {
877 * Send the FIN in a separate segment
878 * after the bulk sending is done.
879 * We don't trust the TSO implementations
880 * to clear the FIN flag on all but the
883 if (tp->t_flags & TF_NEEDFIN)
887 len = tp->t_maxopd - optlen - ipoptlen;
893 KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET,
894 ("%s: len > IP_MAXPACKET", __func__));
896 /*#ifdef DIAGNOSTIC*/
898 if (max_linkhdr + hdrlen > MCLBYTES)
900 if (max_linkhdr + hdrlen > MHLEN)
902 panic("tcphdr too big");
906 * This KASSERT is here to catch edge cases at a well defined place.
907 * Before, those had triggered (random) panic conditions further down.
909 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
912 * Grab a header mbuf, attaching a copy of data to
913 * be transmitted, and initialize the header from
914 * the template for sends on this connection.
920 if ((tp->t_flags & TF_FORCEDATA) && len == 1)
921 TCPSTAT_INC(tcps_sndprobe);
922 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) {
923 tp->t_sndrexmitpack++;
924 TCPSTAT_INC(tcps_sndrexmitpack);
925 TCPSTAT_ADD(tcps_sndrexmitbyte, len);
927 TCPSTAT_INC(tcps_sndpack);
928 TCPSTAT_ADD(tcps_sndbyte, len);
930 MGETHDR(m, M_DONTWAIT, MT_DATA);
932 SOCKBUF_UNLOCK(&so->so_snd);
937 if (MHLEN < hdrlen + max_linkhdr) {
938 MCLGET(m, M_DONTWAIT);
939 if ((m->m_flags & M_EXT) == 0) {
940 SOCKBUF_UNLOCK(&so->so_snd);
947 m->m_data += max_linkhdr;
951 * Start the m_copy functions from the closest mbuf
952 * to the offset in the socket buffer chain.
954 mb = sbsndptr(&so->so_snd, off, len, &moff);
956 if (len <= MHLEN - hdrlen - max_linkhdr) {
957 m_copydata(mb, moff, (int)len,
958 mtod(m, caddr_t) + hdrlen);
961 m->m_next = m_copy(mb, moff, (int)len);
962 if (m->m_next == NULL) {
963 SOCKBUF_UNLOCK(&so->so_snd);
971 * If we're sending everything we've got, set PUSH.
972 * (This will keep happy those implementations which only
973 * give data to the user when a buffer fills or
976 if (off + len == so->so_snd.sb_cc)
978 SOCKBUF_UNLOCK(&so->so_snd);
980 SOCKBUF_UNLOCK(&so->so_snd);
981 if (tp->t_flags & TF_ACKNOW)
982 TCPSTAT_INC(tcps_sndacks);
983 else if (flags & (TH_SYN|TH_FIN|TH_RST))
984 TCPSTAT_INC(tcps_sndctrl);
985 else if (SEQ_GT(tp->snd_up, tp->snd_una))
986 TCPSTAT_INC(tcps_sndurg);
988 TCPSTAT_INC(tcps_sndwinup);
990 MGETHDR(m, M_DONTWAIT, MT_DATA);
996 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
1001 m->m_data += max_linkhdr;
1004 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
1005 m->m_pkthdr.rcvif = (struct ifnet *)0;
1007 mac_inpcb_create_mbuf(tp->t_inpcb, m);
1011 ip6 = mtod(m, struct ip6_hdr *);
1012 th = (struct tcphdr *)(ip6 + 1);
1013 tcpip_fillheaders(tp->t_inpcb, ip6, th);
1017 ip = mtod(m, struct ip *);
1018 ipov = (struct ipovly *)ip;
1019 th = (struct tcphdr *)(ip + 1);
1020 tcpip_fillheaders(tp->t_inpcb, ip, th);
1024 * Fill in fields, remembering maximum advertised
1025 * window for use in delaying messages about window sizes.
1026 * If resending a FIN, be sure not to use a new sequence number.
1028 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
1029 tp->snd_nxt == tp->snd_max)
1032 * If we are starting a connection, send ECN setup
1033 * SYN packet. If we are on a retransmit, we may
1034 * resend those bits a number of times as per
1037 if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn) {
1038 if (tp->t_rxtshift >= 1) {
1039 if (tp->t_rxtshift <= V_tcp_ecn_maxretries)
1040 flags |= TH_ECE|TH_CWR;
1042 flags |= TH_ECE|TH_CWR;
1045 if (tp->t_state == TCPS_ESTABLISHED &&
1046 (tp->t_flags & TF_ECN_PERMIT)) {
1048 * If the peer has ECN, mark data packets with
1049 * ECN capable transmission (ECT).
1050 * Ignore pure ack packets, retransmissions and window probes.
1052 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) &&
1053 !((tp->t_flags & TF_FORCEDATA) && len == 1)) {
1056 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
1059 ip->ip_tos |= IPTOS_ECN_ECT0;
1060 TCPSTAT_INC(tcps_ecn_ect0);
1064 * Reply with proper ECN notifications.
1066 if (tp->t_flags & TF_ECN_SND_CWR) {
1068 tp->t_flags &= ~TF_ECN_SND_CWR;
1070 if (tp->t_flags & TF_ECN_SND_ECE)
1075 * If we are doing retransmissions, then snd_nxt will
1076 * not reflect the first unsent octet. For ACK only
1077 * packets, we do not want the sequence number of the
1078 * retransmitted packet, we want the sequence number
1079 * of the next unsent octet. So, if there is no data
1080 * (and no SYN or FIN), use snd_max instead of snd_nxt
1081 * when filling in ti_seq. But if we are in persist
1082 * state, snd_max might reflect one byte beyond the
1083 * right edge of the window, so use snd_nxt in that
1084 * case, since we know we aren't doing a retransmission.
1085 * (retransmit and persist are mutually exclusive...)
1087 if (sack_rxmit == 0) {
1088 if (len || (flags & (TH_SYN|TH_FIN)) ||
1089 tcp_timer_active(tp, TT_PERSIST))
1090 th->th_seq = htonl(tp->snd_nxt);
1092 th->th_seq = htonl(tp->snd_max);
1094 th->th_seq = htonl(p->rxmit);
1096 tp->sackhint.sack_bytes_rexmit += len;
1098 th->th_ack = htonl(tp->rcv_nxt);
1100 bcopy(opt, th + 1, optlen);
1101 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1103 th->th_flags = flags;
1105 * Calculate receive window. Don't shrink window,
1106 * but avoid silly window syndrome.
1108 if (recwin < (long)(so->so_rcv.sb_hiwat / 4) &&
1109 recwin < (long)tp->t_maxseg)
1111 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
1112 recwin < (long)(tp->rcv_adv - tp->rcv_nxt))
1113 recwin = (long)(tp->rcv_adv - tp->rcv_nxt);
1114 if (recwin > (long)TCP_MAXWIN << tp->rcv_scale)
1115 recwin = (long)TCP_MAXWIN << tp->rcv_scale;
1118 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1119 * or <SYN,ACK>) segment itself is never scaled. The <SYN,ACK>
1120 * case is handled in syncache.
1123 th->th_win = htons((u_short)
1124 (min(sbspace(&so->so_rcv), TCP_MAXWIN)));
1126 th->th_win = htons((u_short)(recwin >> tp->rcv_scale));
1129 * Adjust the RXWIN0SENT flag - indicate that we have advertised
1130 * a 0 window. This may cause the remote transmitter to stall. This
1131 * flag tells soreceive() to disable delayed acknowledgements when
1132 * draining the buffer. This can occur if the receiver is attempting
1133 * to read more data than can be buffered prior to transmitting on
1136 if (th->th_win == 0) {
1138 tp->t_flags |= TF_RXWIN0SENT;
1140 tp->t_flags &= ~TF_RXWIN0SENT;
1141 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
1142 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
1143 th->th_flags |= TH_URG;
1146 * If no urgent pointer to send, then we pull
1147 * the urgent pointer to the left edge of the send window
1148 * so that it doesn't drift into the send window on sequence
1149 * number wraparound.
1151 tp->snd_up = tp->snd_una; /* drag it along */
1153 #ifdef TCP_SIGNATURE
1154 if (tp->t_flags & TF_SIGNATURE) {
1155 int sigoff = to.to_signature - opt;
1156 tcp_signature_compute(m, 0, len, optlen,
1157 (u_char *)(th + 1) + sigoff, IPSEC_DIR_OUTBOUND);
1162 * Put TCP length in extended header, and then
1163 * checksum extended header and data.
1165 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
1166 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1170 * ip6_plen is not need to be filled now, and will be filled
1173 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1174 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) +
1175 optlen + len, IPPROTO_TCP, 0);
1178 #if defined(INET6) && defined(INET)
1183 m->m_pkthdr.csum_flags = CSUM_TCP;
1184 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1185 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen));
1187 /* IP version must be set here for ipv4/ipv6 checking later */
1188 KASSERT(ip->ip_v == IPVERSION,
1189 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
1194 * Enable TSO and specify the size of the segments.
1195 * The TCP pseudo header checksum is always provided.
1196 * XXX: Fixme: This is currently not the case for IPv6.
1199 KASSERT(len > tp->t_maxopd - optlen,
1200 ("%s: len <= tso_segsz", __func__));
1201 m->m_pkthdr.csum_flags |= CSUM_TSO;
1202 m->m_pkthdr.tso_segsz = tp->t_maxopd - optlen;
1206 KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL),
1207 ("%s: mbuf chain shorter than expected: %ld + %u + %u - %u != %u",
1208 __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL)));
1210 KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL),
1211 ("%s: mbuf chain shorter than expected: %ld + %u + %u != %u",
1212 __func__, len, hdrlen, ipoptlen, m_length(m, NULL)));
1216 * In transmit state, time the transmission and arrange for
1217 * the retransmit. In persist state, just set snd_max.
1219 if ((tp->t_flags & TF_FORCEDATA) == 0 ||
1220 !tcp_timer_active(tp, TT_PERSIST)) {
1221 tcp_seq startseq = tp->snd_nxt;
1224 * Advance snd_nxt over sequence space of this segment.
1226 if (flags & (TH_SYN|TH_FIN)) {
1229 if (flags & TH_FIN) {
1231 tp->t_flags |= TF_SENTFIN;
1237 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
1238 tp->snd_max = tp->snd_nxt;
1240 * Time this transmission if not a retransmission and
1241 * not currently timing anything.
1243 if (tp->t_rtttime == 0) {
1244 tp->t_rtttime = ticks;
1245 tp->t_rtseq = startseq;
1246 TCPSTAT_INC(tcps_segstimed);
1251 * Set retransmit timer if not currently set,
1252 * and not doing a pure ack or a keep-alive probe.
1253 * Initial value for retransmit timer is smoothed
1254 * round-trip time + 2 * round-trip time variance.
1255 * Initialize shift counter which is used for backoff
1256 * of retransmit time.
1259 if (!tcp_timer_active(tp, TT_REXMT) &&
1260 ((sack_rxmit && tp->snd_nxt != tp->snd_max) ||
1261 (tp->snd_nxt != tp->snd_una))) {
1262 if (tcp_timer_active(tp, TT_PERSIST)) {
1263 tcp_timer_activate(tp, TT_PERSIST, 0);
1266 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1270 * Persist case, update snd_max but since we are in
1271 * persist mode (no window) we do not update snd_nxt.
1276 if (flags & TH_FIN) {
1278 tp->t_flags |= TF_SENTFIN;
1280 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
1281 tp->snd_max = tp->snd_nxt + len;
1284 /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */
1285 hhook_run_tcp_est_out(tp, th, &to, len, tso);
1291 if (so->so_options & SO_DEBUG) {
1297 save = ipov->ih_len;
1298 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */);
1300 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
1304 ipov->ih_len = save;
1306 #endif /* TCPDEBUG */
1309 * Fill in IP length and desired time to live and
1310 * send to IP level. There should be a better way
1311 * to handle ttl and tos; we could keep them in
1312 * the template, but need a way to checksum without them.
1315 * m->m_pkthdr.len should have been set before cksum calcuration,
1316 * because in6_cksum() need it.
1320 struct route_in6 ro;
1322 bzero(&ro, sizeof(ro));
1324 * we separately set hoplimit for every segment, since the
1325 * user might want to change the value via setsockopt.
1326 * Also, desired default hop limit might be changed via
1327 * Neighbor Discovery.
1329 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL);
1331 /* TODO: IPv6 IP6TOS_ECT bit on */
1332 error = ip6_output(m, tp->t_inpcb->in6p_outputopts, &ro,
1333 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1334 NULL, NULL, tp->t_inpcb);
1336 if (error == EMSGSIZE && ro.ro_rt != NULL)
1337 mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1341 #if defined(INET) && defined(INET6)
1348 bzero(&ro, sizeof(ro));
1349 ip->ip_len = m->m_pkthdr.len;
1351 if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO)
1352 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL);
1355 * If we do path MTU discovery, then we set DF on every packet.
1356 * This might not be the best thing to do according to RFC3390
1357 * Section 2. However the tcp hostcache migitates the problem
1358 * so it affects only the first tcp connection with a host.
1360 * NB: Don't set DF on small MTU/MSS to have a safe fallback.
1362 if (V_path_mtu_discovery && tp->t_maxopd > V_tcp_minmss)
1363 ip->ip_off |= IP_DF;
1365 error = ip_output(m, tp->t_inpcb->inp_options, &ro,
1366 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0,
1369 if (error == EMSGSIZE && ro.ro_rt != NULL)
1370 mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1377 * We know that the packet was lost, so back out the
1378 * sequence number advance, if any.
1380 * If the error is EPERM the packet got blocked by the
1381 * local firewall. Normally we should terminate the
1382 * connection but the blocking may have been spurious
1383 * due to a firewall reconfiguration cycle. So we treat
1384 * it like a packet loss and let the retransmit timer and
1385 * timeouts do their work over time.
1386 * XXX: It is a POLA question whether calling tcp_drop right
1387 * away would be the really correct behavior instead.
1389 if (((tp->t_flags & TF_FORCEDATA) == 0 ||
1390 !tcp_timer_active(tp, TT_PERSIST)) &&
1391 ((flags & TH_SYN) == 0) &&
1395 tp->sackhint.sack_bytes_rexmit -= len;
1396 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
1397 ("sackhint bytes rtx >= 0"));
1402 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */
1405 tp->t_softerror = error;
1408 if (!tcp_timer_active(tp, TT_REXMT) &&
1409 !tcp_timer_active(tp, TT_PERSIST))
1410 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1411 tp->snd_cwnd = tp->t_maxseg;
1415 * For some reason the interface we used initially
1416 * to send segments changed to another or lowered
1418 * If TSO was active we either got an interface
1419 * without TSO capabilits or TSO was turned off.
1420 * If we obtained mtu from ip_output() then update
1424 tp->t_flags &= ~TF_TSO;
1426 tcp_mss_update(tp, -1, mtu, NULL, NULL);
1434 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1435 tp->t_softerror = error;
1443 TCPSTAT_INC(tcps_sndtotal);
1446 * Data sent (as far as we can tell).
1447 * If this advertises a larger window than any other segment,
1448 * then remember the size of the advertised window.
1449 * Any pending ACK has now been sent.
1451 if (recwin >= 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv))
1452 tp->rcv_adv = tp->rcv_nxt + recwin;
1453 tp->last_ack_sent = tp->rcv_nxt;
1454 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK);
1455 if (tcp_timer_active(tp, TT_DELACK))
1456 tcp_timer_activate(tp, TT_DELACK, 0);
1459 * This completely breaks TCP if newreno is turned on. What happens
1460 * is that if delayed-acks are turned on on the receiver, this code
1461 * on the transmitter effectively destroys the TCP window, forcing
1462 * it to four packets (1.5Kx4 = 6K window).
1464 if (sendalot && --maxburst)
1473 tcp_setpersist(struct tcpcb *tp)
1475 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
1478 tp->t_flags &= ~TF_PREVVALID;
1479 if (tcp_timer_active(tp, TT_REXMT))
1480 panic("tcp_setpersist: retransmit pending");
1482 * Start/restart persistance timer.
1484 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
1485 TCPTV_PERSMIN, TCPTV_PERSMAX);
1486 tcp_timer_activate(tp, TT_PERSIST, tt);
1487 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
1492 * Insert TCP options according to the supplied parameters to the place
1493 * optp in a consistent way. Can handle unaligned destinations.
1495 * The order of the option processing is crucial for optimal packing and
1496 * alignment for the scarce option space.
1498 * The optimal order for a SYN/SYN-ACK segment is:
1499 * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) +
1500 * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40.
1502 * The SACK options should be last. SACK blocks consume 8*n+2 bytes.
1503 * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks).
1504 * At minimum we need 10 bytes (to generate 1 SACK block). If both
1505 * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present,
1506 * we only have 10 bytes for SACK options (40 - (12 + 18)).
1509 tcp_addoptions(struct tcpopt *to, u_char *optp)
1511 u_int mask, optlen = 0;
1513 for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) {
1514 if ((to->to_flags & mask) != mask)
1516 if (optlen == TCP_MAXOLEN)
1518 switch (to->to_flags & mask) {
1520 while (optlen % 4) {
1521 optlen += TCPOLEN_NOP;
1522 *optp++ = TCPOPT_NOP;
1524 if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG)
1526 optlen += TCPOLEN_MAXSEG;
1527 *optp++ = TCPOPT_MAXSEG;
1528 *optp++ = TCPOLEN_MAXSEG;
1529 to->to_mss = htons(to->to_mss);
1530 bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss));
1531 optp += sizeof(to->to_mss);
1534 while (!optlen || optlen % 2 != 1) {
1535 optlen += TCPOLEN_NOP;
1536 *optp++ = TCPOPT_NOP;
1538 if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW)
1540 optlen += TCPOLEN_WINDOW;
1541 *optp++ = TCPOPT_WINDOW;
1542 *optp++ = TCPOLEN_WINDOW;
1543 *optp++ = to->to_wscale;
1546 while (optlen % 2) {
1547 optlen += TCPOLEN_NOP;
1548 *optp++ = TCPOPT_NOP;
1550 if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED)
1552 optlen += TCPOLEN_SACK_PERMITTED;
1553 *optp++ = TCPOPT_SACK_PERMITTED;
1554 *optp++ = TCPOLEN_SACK_PERMITTED;
1557 while (!optlen || optlen % 4 != 2) {
1558 optlen += TCPOLEN_NOP;
1559 *optp++ = TCPOPT_NOP;
1561 if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP)
1563 optlen += TCPOLEN_TIMESTAMP;
1564 *optp++ = TCPOPT_TIMESTAMP;
1565 *optp++ = TCPOLEN_TIMESTAMP;
1566 to->to_tsval = htonl(to->to_tsval);
1567 to->to_tsecr = htonl(to->to_tsecr);
1568 bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval));
1569 optp += sizeof(to->to_tsval);
1570 bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr));
1571 optp += sizeof(to->to_tsecr);
1575 int siglen = TCPOLEN_SIGNATURE - 2;
1577 while (!optlen || optlen % 4 != 2) {
1578 optlen += TCPOLEN_NOP;
1579 *optp++ = TCPOPT_NOP;
1581 if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE)
1583 optlen += TCPOLEN_SIGNATURE;
1584 *optp++ = TCPOPT_SIGNATURE;
1585 *optp++ = TCPOLEN_SIGNATURE;
1586 to->to_signature = optp;
1594 struct sackblk *sack = (struct sackblk *)to->to_sacks;
1597 while (!optlen || optlen % 4 != 2) {
1598 optlen += TCPOLEN_NOP;
1599 *optp++ = TCPOPT_NOP;
1601 if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK)
1603 optlen += TCPOLEN_SACKHDR;
1604 *optp++ = TCPOPT_SACK;
1605 sackblks = min(to->to_nsacks,
1606 (TCP_MAXOLEN - optlen) / TCPOLEN_SACK);
1607 *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK;
1608 while (sackblks--) {
1609 sack_seq = htonl(sack->start);
1610 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
1611 optp += sizeof(sack_seq);
1612 sack_seq = htonl(sack->end);
1613 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
1614 optp += sizeof(sack_seq);
1615 optlen += TCPOLEN_SACK;
1618 TCPSTAT_INC(tcps_sack_send_blocks);
1622 panic("%s: unknown TCP option type", __func__);
1627 /* Terminate and pad TCP options to a 4 byte boundary. */
1629 optlen += TCPOLEN_EOL;
1630 *optp++ = TCPOPT_EOL;
1633 * According to RFC 793 (STD0007):
1634 * "The content of the header beyond the End-of-Option option
1635 * must be header padding (i.e., zero)."
1636 * and later: "The padding is composed of zeros."
1638 while (optlen % 4) {
1639 optlen += TCPOLEN_PAD;
1640 *optp++ = TCPOPT_PAD;
1643 KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__));