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>
80 #include <netipsec/ipsec.h>
83 #include <machine/in_cksum.h>
85 #include <security/mac/mac_framework.h>
87 VNET_DEFINE(int, path_mtu_discovery) = 1;
88 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_RW,
89 &VNET_NAME(path_mtu_discovery), 1,
90 "Enable Path MTU Discovery");
92 VNET_DEFINE(int, ss_fltsz) = 1;
93 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, slowstart_flightsize, CTLFLAG_RW,
94 &VNET_NAME(ss_fltsz), 1,
95 "Slow start flight size");
97 VNET_DEFINE(int, ss_fltsz_local) = 4;
98 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, local_slowstart_flightsize,
99 CTLFLAG_RW, &VNET_NAME(ss_fltsz_local), 1,
100 "Slow start flight size for local networks");
102 VNET_DEFINE(int, tcp_do_tso) = 1;
103 #define V_tcp_do_tso VNET(tcp_do_tso)
104 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_RW,
105 &VNET_NAME(tcp_do_tso), 0,
106 "Enable TCP Segmentation Offload");
108 VNET_DEFINE(int, tcp_do_autosndbuf) = 1;
109 #define V_tcp_do_autosndbuf VNET(tcp_do_autosndbuf)
110 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_RW,
111 &VNET_NAME(tcp_do_autosndbuf), 0,
112 "Enable automatic send buffer sizing");
114 VNET_DEFINE(int, tcp_autosndbuf_inc) = 8*1024;
115 #define V_tcp_autosndbuf_inc VNET(tcp_autosndbuf_inc)
116 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_RW,
117 &VNET_NAME(tcp_autosndbuf_inc), 0,
118 "Incrementor step size of automatic send buffer");
120 VNET_DEFINE(int, tcp_autosndbuf_max) = 2*1024*1024;
121 #define V_tcp_autosndbuf_max VNET(tcp_autosndbuf_max)
122 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_RW,
123 &VNET_NAME(tcp_autosndbuf_max), 0,
124 "Max size of automatic send buffer");
126 static void inline hhook_run_tcp_est_out(struct tcpcb *tp,
127 struct tcphdr *th, struct tcpopt *to,
129 static void inline cc_after_idle(struct tcpcb *tp);
132 * Wrapper for the TCP established ouput helper hook.
135 hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th,
136 struct tcpopt *to, long len, int tso)
138 struct tcp_hhook_data hhook_data;
140 if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) {
144 hhook_data.len = len;
145 hhook_data.tso = tso;
147 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data,
153 * CC wrapper hook functions
156 cc_after_idle(struct tcpcb *tp)
158 INP_WLOCK_ASSERT(tp->t_inpcb);
160 if (CC_ALGO(tp)->after_idle != NULL)
161 CC_ALGO(tp)->after_idle(tp->ccv);
165 * Tcp output routine: figure out what should be sent and send it.
168 tcp_output(struct tcpcb *tp)
170 struct socket *so = tp->t_inpcb->inp_socket;
171 long len, recwin, sendwin;
172 int off, flags, error = 0; /* Keep compiler happy */
174 struct ip *ip = NULL;
175 struct ipovly *ipov = NULL;
177 u_char opt[TCP_MAXOLEN];
178 unsigned ipoptlen, optlen, hdrlen;
180 unsigned ipsec_optlen = 0;
183 int sack_rxmit, sack_bytes_rxmt;
188 int maxburst = TCP_MAXBURST;
191 struct ip6_hdr *ip6 = NULL;
194 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
197 INP_WLOCK_ASSERT(tp->t_inpcb);
200 * Determine length of data that should be transmitted,
201 * and flags that will be used.
202 * If there is some data or critical controls (SYN, RST)
203 * to send, then transmit; otherwise, investigate further.
205 idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
206 if (idle && ticks - tp->t_rcvtime >= tp->t_rxtcur)
208 tp->t_flags &= ~TF_LASTIDLE;
210 if (tp->t_flags & TF_MORETOCOME) {
211 tp->t_flags |= TF_LASTIDLE;
217 * If we've recently taken a timeout, snd_max will be greater than
218 * snd_nxt. There may be SACK information that allows us to avoid
219 * resending already delivered data. Adjust snd_nxt accordingly.
221 if ((tp->t_flags & TF_SACK_PERMIT) &&
222 SEQ_LT(tp->snd_nxt, tp->snd_max))
226 off = tp->snd_nxt - tp->snd_una;
227 sendwin = min(tp->snd_wnd, tp->snd_cwnd);
229 flags = tcp_outflags[tp->t_state];
231 * Send any SACK-generated retransmissions. If we're explicitly trying
232 * to send out new data (when sendalot is 1), bypass this function.
233 * If we retransmit in fast recovery mode, decrement snd_cwnd, since
234 * we're replacing a (future) new transmission with a retransmission
235 * now, and we previously incremented snd_cwnd in tcp_input().
238 * Still in sack recovery , reset rxmit flag to zero.
244 if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) &&
245 (p = tcp_sack_output(tp, &sack_bytes_rxmt))) {
248 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt;
251 /* Do not retransmit SACK segments beyond snd_recover */
252 if (SEQ_GT(p->end, tp->snd_recover)) {
254 * (At least) part of sack hole extends beyond
255 * snd_recover. Check to see if we can rexmit data
258 if (SEQ_GEQ(p->rxmit, tp->snd_recover)) {
260 * Can't rexmit any more data for this hole.
261 * That data will be rexmitted in the next
262 * sack recovery episode, when snd_recover
263 * moves past p->rxmit.
266 goto after_sack_rexmit;
268 /* Can rexmit part of the current hole */
269 len = ((long)ulmin(cwin,
270 tp->snd_recover - p->rxmit));
272 len = ((long)ulmin(cwin, p->end - p->rxmit));
273 off = p->rxmit - tp->snd_una;
274 KASSERT(off >= 0,("%s: sack block to the left of una : %d",
279 TCPSTAT_INC(tcps_sack_rexmits);
280 TCPSTAT_ADD(tcps_sack_rexmit_bytes,
281 min(len, tp->t_maxseg));
286 * Get standard flags, and add SYN or FIN if requested by 'hidden'
289 if (tp->t_flags & TF_NEEDFIN)
291 if (tp->t_flags & TF_NEEDSYN)
294 SOCKBUF_LOCK(&so->so_snd);
296 * If in persist timeout with window of 0, send 1 byte.
297 * Otherwise, if window is small but nonzero
298 * and timer expired, we will send what we can
299 * and go to transmit state.
301 if (tp->t_flags & TF_FORCEDATA) {
304 * If we still have some data to send, then
305 * clear the FIN bit. Usually this would
306 * happen below when it realizes that we
307 * aren't sending all the data. However,
308 * if we have exactly 1 byte of unsent data,
309 * then it won't clear the FIN bit below,
310 * and if we are in persist state, we wind
311 * up sending the packet without recording
312 * that we sent the FIN bit.
314 * We can't just blindly clear the FIN bit,
315 * because if we don't have any more data
316 * to send then the probe will be the FIN
319 if (off < so->so_snd.sb_cc)
323 tcp_timer_activate(tp, TT_PERSIST, 0);
329 * If snd_nxt == snd_max and we have transmitted a FIN, the
330 * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
331 * a negative length. This can also occur when TCP opens up
332 * its congestion window while receiving additional duplicate
333 * acks after fast-retransmit because TCP will reset snd_nxt
334 * to snd_max after the fast-retransmit.
336 * In the normal retransmit-FIN-only case, however, snd_nxt will
337 * be set to snd_una, the offset will be 0, and the length may
340 * If sack_rxmit is true we are retransmitting from the scoreboard
341 * in which case len is already set.
343 if (sack_rxmit == 0) {
344 if (sack_bytes_rxmt == 0)
345 len = ((long)ulmin(so->so_snd.sb_cc, sendwin) - off);
350 * We are inside of a SACK recovery episode and are
351 * sending new data, having retransmitted all the
352 * data possible in the scoreboard.
354 len = ((long)ulmin(so->so_snd.sb_cc, tp->snd_wnd)
357 * Don't remove this (len > 0) check !
358 * We explicitly check for len > 0 here (although it
359 * isn't really necessary), to work around a gcc
360 * optimization issue - to force gcc to compute
361 * len above. Without this check, the computation
362 * of len is bungled by the optimizer.
365 cwin = tp->snd_cwnd -
366 (tp->snd_nxt - tp->sack_newdata) -
370 len = lmin(len, cwin);
376 * Lop off SYN bit if it has already been sent. However, if this
377 * is SYN-SENT state and if segment contains data and if we don't
378 * know that foreign host supports TAO, suppress sending segment.
380 if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
381 if (tp->t_state != TCPS_SYN_RECEIVED)
387 * Be careful not to send data and/or FIN on SYN segments.
388 * This measure is needed to prevent interoperability problems
389 * with not fully conformant TCP implementations.
391 if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) {
398 * If FIN has been sent but not acked,
399 * but we haven't been called to retransmit,
400 * len will be < 0. Otherwise, window shrank
401 * after we sent into it. If window shrank to 0,
402 * cancel pending retransmit, pull snd_nxt back
403 * to (closed) window, and set the persist timer
404 * if it isn't already going. If the window didn't
405 * close completely, just wait for an ACK.
409 tcp_timer_activate(tp, TT_REXMT, 0);
411 tp->snd_nxt = tp->snd_una;
412 if (!tcp_timer_active(tp, TT_PERSIST))
417 /* len will be >= 0 after this point. */
418 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
421 * Automatic sizing of send socket buffer. Often the send buffer
422 * size is not optimally adjusted to the actual network conditions
423 * at hand (delay bandwidth product). Setting the buffer size too
424 * small limits throughput on links with high bandwidth and high
425 * delay (eg. trans-continental/oceanic links). Setting the
426 * buffer size too big consumes too much real kernel memory,
427 * especially with many connections on busy servers.
429 * The criteria to step up the send buffer one notch are:
430 * 1. receive window of remote host is larger than send buffer
431 * (with a fudge factor of 5/4th);
432 * 2. send buffer is filled to 7/8th with data (so we actually
433 * have data to make use of it);
434 * 3. send buffer fill has not hit maximal automatic size;
435 * 4. our send window (slow start and cogestion controlled) is
436 * larger than sent but unacknowledged data in send buffer.
438 * The remote host receive window scaling factor may limit the
439 * growing of the send buffer before it reaches its allowed
442 * It scales directly with slow start or congestion window
443 * and does at most one step per received ACK. This fast
444 * scaling has the drawback of growing the send buffer beyond
445 * what is strictly necessary to make full use of a given
446 * delay*bandwith product. However testing has shown this not
447 * to be much of an problem. At worst we are trading wasting
448 * of available bandwith (the non-use of it) for wasting some
449 * socket buffer memory.
451 * TODO: Shrink send buffer during idle periods together
452 * with congestion window. Requires another timer. Has to
453 * wait for upcoming tcp timer rewrite.
455 if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) {
456 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat &&
457 so->so_snd.sb_cc >= (so->so_snd.sb_hiwat / 8 * 7) &&
458 so->so_snd.sb_cc < V_tcp_autosndbuf_max &&
459 sendwin >= (so->so_snd.sb_cc - (tp->snd_nxt - tp->snd_una))) {
460 if (!sbreserve_locked(&so->so_snd,
461 min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc,
462 V_tcp_autosndbuf_max), so, curthread))
463 so->so_snd.sb_flags &= ~SB_AUTOSIZE;
468 * Decide if we can use TCP Segmentation Offloading (if supported by
471 * TSO may only be used if we are in a pure bulk sending state. The
472 * presence of TCP-MD5, SACK retransmits, SACK advertizements and
473 * IP options prevent using TSO. With TSO the TCP header is the same
474 * (except for the sequence number) for all generated packets. This
475 * makes it impossible to transmit any options which vary per generated
480 * Pre-calculate here as we save another lookup into the darknesses
481 * of IPsec that way and can actually decide if TSO is ok.
483 ipsec_optlen = ipsec_hdrsiz_tcp(tp);
485 if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg &&
486 ((tp->t_flags & TF_SIGNATURE) == 0) &&
487 tp->rcv_numsacks == 0 && sack_rxmit == 0 &&
491 tp->t_inpcb->inp_options == NULL &&
492 tp->t_inpcb->in6p_options == NULL)
496 if (SEQ_LT(p->rxmit + len, tp->snd_una + so->so_snd.sb_cc))
499 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
503 recwin = sbspace(&so->so_rcv);
506 * Sender silly window avoidance. We transmit under the following
507 * conditions when len is non-zero:
509 * - We have a full segment (or more with TSO)
510 * - This is the last buffer in a write()/send() and we are
511 * either idle or running NODELAY
512 * - we've timed out (e.g. persist timer)
513 * - we have more then 1/2 the maximum send window's worth of
514 * data (receiver may be limited the window size)
515 * - we need to retransmit
518 if (len >= tp->t_maxseg)
521 * NOTE! on localhost connections an 'ack' from the remote
522 * end may occur synchronously with the output and cause
523 * us to flush a buffer queued with moretocome. XXX
525 * note: the len + off check is almost certainly unnecessary.
527 if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
528 (idle || (tp->t_flags & TF_NODELAY)) &&
529 len + off >= so->so_snd.sb_cc &&
530 (tp->t_flags & TF_NOPUSH) == 0) {
533 if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */
535 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
537 if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
544 * Compare available window to amount of window
545 * known to peer (as advertised window less
546 * next expected input). If the difference is at least two
547 * max size segments, or at least 50% of the maximum possible
548 * window, then want to send a window update to peer.
549 * Skip this if the connection is in T/TCP half-open state.
550 * Don't send pure window updates when the peer has closed
551 * the connection and won't ever send more data.
553 if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) &&
554 !TCPS_HAVERCVDFIN(tp->t_state)) {
556 * "adv" is the amount we can increase the window,
557 * taking into account that we are limited by
558 * TCP_MAXWIN << tp->rcv_scale.
563 adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale);
564 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) {
565 oldwin = (tp->rcv_adv - tp->rcv_nxt);
571 * If the new window size ends up being the same as the old
572 * size when it is scaled, then don't force a window update.
574 if (oldwin >> tp->rcv_scale == (adv + oldwin) >> tp->rcv_scale)
576 if (adv >= (long) (2 * tp->t_maxseg))
578 if (2 * adv >= (long) so->so_rcv.sb_hiwat)
584 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
585 * is also a catch-all for the retransmit timer timeout case.
587 if (tp->t_flags & TF_ACKNOW)
589 if ((flags & TH_RST) ||
590 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
592 if (SEQ_GT(tp->snd_up, tp->snd_una))
595 * If our state indicates that FIN should be sent
596 * and we have not yet done so, then we need to send.
598 if (flags & TH_FIN &&
599 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
602 * In SACK, it is possible for tcp_output to fail to send a segment
603 * after the retransmission timer has been turned off. Make sure
604 * that the retransmission timer is set.
606 if ((tp->t_flags & TF_SACK_PERMIT) &&
607 SEQ_GT(tp->snd_max, tp->snd_una) &&
608 !tcp_timer_active(tp, TT_REXMT) &&
609 !tcp_timer_active(tp, TT_PERSIST)) {
610 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
614 * TCP window updates are not reliable, rather a polling protocol
615 * using ``persist'' packets is used to insure receipt of window
616 * updates. The three ``states'' for the output side are:
617 * idle not doing retransmits or persists
618 * persisting to move a small or zero window
619 * (re)transmitting and thereby not persisting
621 * tcp_timer_active(tp, TT_PERSIST)
622 * is true when we are in persist state.
623 * (tp->t_flags & TF_FORCEDATA)
624 * is set when we are called to send a persist packet.
625 * tcp_timer_active(tp, TT_REXMT)
626 * is set when we are retransmitting
627 * The output side is idle when both timers are zero.
629 * If send window is too small, there is data to transmit, and no
630 * retransmit or persist is pending, then go to persist state.
631 * If nothing happens soon, send when timer expires:
632 * if window is nonzero, transmit what we can,
633 * otherwise force out a byte.
635 if (so->so_snd.sb_cc && !tcp_timer_active(tp, TT_REXMT) &&
636 !tcp_timer_active(tp, TT_PERSIST)) {
642 * No reason to send a segment, just return.
645 SOCKBUF_UNLOCK(&so->so_snd);
649 SOCKBUF_LOCK_ASSERT(&so->so_snd);
651 * Before ESTABLISHED, force sending of initial options
652 * unless TCP set not to do any options.
653 * NOTE: we assume that the IP/TCP header plus TCP options
654 * always fit in a single mbuf, leaving room for a maximum
656 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
661 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
664 hdrlen = sizeof (struct tcpiphdr);
667 * Compute options for segment.
668 * We only have to care about SYN and established connection
669 * segments. Options for SYN-ACK segments are handled in TCP
672 if ((tp->t_flags & TF_NOOPT) == 0) {
674 /* Maximum segment size. */
675 if (flags & TH_SYN) {
676 tp->snd_nxt = tp->iss;
677 to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc);
678 to.to_flags |= TOF_MSS;
680 /* Window scaling. */
681 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) {
682 to.to_wscale = tp->request_r_scale;
683 to.to_flags |= TOF_SCALE;
686 if ((tp->t_flags & TF_RCVD_TSTMP) ||
687 ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) {
688 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
689 to.to_tsecr = tp->ts_recent;
690 to.to_flags |= TOF_TS;
691 /* Set receive buffer autosizing timestamp. */
692 if (tp->rfbuf_ts == 0 &&
693 (so->so_rcv.sb_flags & SB_AUTOSIZE))
694 tp->rfbuf_ts = tcp_ts_getticks();
696 /* Selective ACK's. */
697 if (tp->t_flags & TF_SACK_PERMIT) {
699 to.to_flags |= TOF_SACKPERM;
700 else if (TCPS_HAVEESTABLISHED(tp->t_state) &&
701 (tp->t_flags & TF_SACK_PERMIT) &&
702 tp->rcv_numsacks > 0) {
703 to.to_flags |= TOF_SACK;
704 to.to_nsacks = tp->rcv_numsacks;
705 to.to_sacks = (u_char *)tp->sackblks;
709 /* TCP-MD5 (RFC2385). */
710 if (tp->t_flags & TF_SIGNATURE)
711 to.to_flags |= TOF_SIGNATURE;
712 #endif /* TCP_SIGNATURE */
714 /* Processing the options. */
715 hdrlen += optlen = tcp_addoptions(&to, opt);
720 ipoptlen = ip6_optlen(tp->t_inpcb);
723 if (tp->t_inpcb->inp_options)
724 ipoptlen = tp->t_inpcb->inp_options->m_len -
725 offsetof(struct ipoption, ipopt_list);
729 ipoptlen += ipsec_optlen;
733 * Adjust data length if insertion of options will
734 * bump the packet length beyond the t_maxopd length.
735 * Clear the FIN bit because we cut off the tail of
738 if (len + optlen + ipoptlen > tp->t_maxopd) {
742 KASSERT(ipoptlen == 0,
743 ("%s: TSO can't do IP options", __func__));
746 * Limit a burst to IP_MAXPACKET minus IP,
747 * TCP and options length to keep ip->ip_len
750 if (len > IP_MAXPACKET - hdrlen) {
751 len = IP_MAXPACKET - hdrlen;
756 * Prevent the last segment from being
757 * fractional unless the send sockbuf can
760 if (sendalot && off + len < so->so_snd.sb_cc) {
761 len -= len % (tp->t_maxopd - optlen);
766 * Send the FIN in a separate segment
767 * after the bulk sending is done.
768 * We don't trust the TSO implementations
769 * to clear the FIN flag on all but the
772 if (tp->t_flags & TF_NEEDFIN)
776 len = tp->t_maxopd - optlen - ipoptlen;
782 KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET,
783 ("%s: len > IP_MAXPACKET", __func__));
785 /*#ifdef DIAGNOSTIC*/
787 if (max_linkhdr + hdrlen > MCLBYTES)
789 if (max_linkhdr + hdrlen > MHLEN)
791 panic("tcphdr too big");
795 * This KASSERT is here to catch edge cases at a well defined place.
796 * Before, those had triggered (random) panic conditions further down.
798 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
801 * Grab a header mbuf, attaching a copy of data to
802 * be transmitted, and initialize the header from
803 * the template for sends on this connection.
809 if ((tp->t_flags & TF_FORCEDATA) && len == 1)
810 TCPSTAT_INC(tcps_sndprobe);
811 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) {
812 tp->t_sndrexmitpack++;
813 TCPSTAT_INC(tcps_sndrexmitpack);
814 TCPSTAT_ADD(tcps_sndrexmitbyte, len);
816 TCPSTAT_INC(tcps_sndpack);
817 TCPSTAT_ADD(tcps_sndbyte, len);
819 MGETHDR(m, M_DONTWAIT, MT_DATA);
821 SOCKBUF_UNLOCK(&so->so_snd);
826 if (MHLEN < hdrlen + max_linkhdr) {
827 MCLGET(m, M_DONTWAIT);
828 if ((m->m_flags & M_EXT) == 0) {
829 SOCKBUF_UNLOCK(&so->so_snd);
836 m->m_data += max_linkhdr;
840 * Start the m_copy functions from the closest mbuf
841 * to the offset in the socket buffer chain.
843 mb = sbsndptr(&so->so_snd, off, len, &moff);
845 if (len <= MHLEN - hdrlen - max_linkhdr) {
846 m_copydata(mb, moff, (int)len,
847 mtod(m, caddr_t) + hdrlen);
850 m->m_next = m_copy(mb, moff, (int)len);
851 if (m->m_next == NULL) {
852 SOCKBUF_UNLOCK(&so->so_snd);
860 * If we're sending everything we've got, set PUSH.
861 * (This will keep happy those implementations which only
862 * give data to the user when a buffer fills or
865 if (off + len == so->so_snd.sb_cc)
867 SOCKBUF_UNLOCK(&so->so_snd);
869 SOCKBUF_UNLOCK(&so->so_snd);
870 if (tp->t_flags & TF_ACKNOW)
871 TCPSTAT_INC(tcps_sndacks);
872 else if (flags & (TH_SYN|TH_FIN|TH_RST))
873 TCPSTAT_INC(tcps_sndctrl);
874 else if (SEQ_GT(tp->snd_up, tp->snd_una))
875 TCPSTAT_INC(tcps_sndurg);
877 TCPSTAT_INC(tcps_sndwinup);
879 MGETHDR(m, M_DONTWAIT, MT_DATA);
885 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
890 m->m_data += max_linkhdr;
893 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
894 m->m_pkthdr.rcvif = (struct ifnet *)0;
896 mac_inpcb_create_mbuf(tp->t_inpcb, m);
900 ip6 = mtod(m, struct ip6_hdr *);
901 th = (struct tcphdr *)(ip6 + 1);
902 tcpip_fillheaders(tp->t_inpcb, ip6, th);
906 ip = mtod(m, struct ip *);
907 ipov = (struct ipovly *)ip;
908 th = (struct tcphdr *)(ip + 1);
909 tcpip_fillheaders(tp->t_inpcb, ip, th);
913 * Fill in fields, remembering maximum advertised
914 * window for use in delaying messages about window sizes.
915 * If resending a FIN, be sure not to use a new sequence number.
917 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
918 tp->snd_nxt == tp->snd_max)
921 * If we are starting a connection, send ECN setup
922 * SYN packet. If we are on a retransmit, we may
923 * resend those bits a number of times as per
926 if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn) {
927 if (tp->t_rxtshift >= 1) {
928 if (tp->t_rxtshift <= V_tcp_ecn_maxretries)
929 flags |= TH_ECE|TH_CWR;
931 flags |= TH_ECE|TH_CWR;
934 if (tp->t_state == TCPS_ESTABLISHED &&
935 (tp->t_flags & TF_ECN_PERMIT)) {
937 * If the peer has ECN, mark data packets with
938 * ECN capable transmission (ECT).
939 * Ignore pure ack packets, retransmissions and window probes.
941 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) &&
942 !((tp->t_flags & TF_FORCEDATA) && len == 1)) {
945 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
948 ip->ip_tos |= IPTOS_ECN_ECT0;
949 TCPSTAT_INC(tcps_ecn_ect0);
953 * Reply with proper ECN notifications.
955 if (tp->t_flags & TF_ECN_SND_CWR) {
957 tp->t_flags &= ~TF_ECN_SND_CWR;
959 if (tp->t_flags & TF_ECN_SND_ECE)
964 * If we are doing retransmissions, then snd_nxt will
965 * not reflect the first unsent octet. For ACK only
966 * packets, we do not want the sequence number of the
967 * retransmitted packet, we want the sequence number
968 * of the next unsent octet. So, if there is no data
969 * (and no SYN or FIN), use snd_max instead of snd_nxt
970 * when filling in ti_seq. But if we are in persist
971 * state, snd_max might reflect one byte beyond the
972 * right edge of the window, so use snd_nxt in that
973 * case, since we know we aren't doing a retransmission.
974 * (retransmit and persist are mutually exclusive...)
976 if (sack_rxmit == 0) {
977 if (len || (flags & (TH_SYN|TH_FIN)) ||
978 tcp_timer_active(tp, TT_PERSIST))
979 th->th_seq = htonl(tp->snd_nxt);
981 th->th_seq = htonl(tp->snd_max);
983 th->th_seq = htonl(p->rxmit);
985 tp->sackhint.sack_bytes_rexmit += len;
987 th->th_ack = htonl(tp->rcv_nxt);
989 bcopy(opt, th + 1, optlen);
990 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
992 th->th_flags = flags;
994 * Calculate receive window. Don't shrink window,
995 * but avoid silly window syndrome.
997 if (recwin < (long)(so->so_rcv.sb_hiwat / 4) &&
998 recwin < (long)tp->t_maxseg)
1000 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
1001 recwin < (long)(tp->rcv_adv - tp->rcv_nxt))
1002 recwin = (long)(tp->rcv_adv - tp->rcv_nxt);
1003 if (recwin > (long)TCP_MAXWIN << tp->rcv_scale)
1004 recwin = (long)TCP_MAXWIN << tp->rcv_scale;
1007 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1008 * or <SYN,ACK>) segment itself is never scaled. The <SYN,ACK>
1009 * case is handled in syncache.
1012 th->th_win = htons((u_short)
1013 (min(sbspace(&so->so_rcv), TCP_MAXWIN)));
1015 th->th_win = htons((u_short)(recwin >> tp->rcv_scale));
1018 * Adjust the RXWIN0SENT flag - indicate that we have advertised
1019 * a 0 window. This may cause the remote transmitter to stall. This
1020 * flag tells soreceive() to disable delayed acknowledgements when
1021 * draining the buffer. This can occur if the receiver is attempting
1022 * to read more data than can be buffered prior to transmitting on
1025 if (th->th_win == 0) {
1027 tp->t_flags |= TF_RXWIN0SENT;
1029 tp->t_flags &= ~TF_RXWIN0SENT;
1030 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
1031 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
1032 th->th_flags |= TH_URG;
1035 * If no urgent pointer to send, then we pull
1036 * the urgent pointer to the left edge of the send window
1037 * so that it doesn't drift into the send window on sequence
1038 * number wraparound.
1040 tp->snd_up = tp->snd_una; /* drag it along */
1042 #ifdef TCP_SIGNATURE
1043 if (tp->t_flags & TF_SIGNATURE) {
1044 int sigoff = to.to_signature - opt;
1045 tcp_signature_compute(m, 0, len, optlen,
1046 (u_char *)(th + 1) + sigoff, IPSEC_DIR_OUTBOUND);
1051 * Put TCP length in extended header, and then
1052 * checksum extended header and data.
1054 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
1055 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1059 * ip6_plen is not need to be filled now, and will be filled
1062 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1063 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) +
1064 optlen + len, IPPROTO_TCP, 0);
1067 #if defined(INET6) && defined(INET)
1072 m->m_pkthdr.csum_flags = CSUM_TCP;
1073 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1074 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen));
1076 /* IP version must be set here for ipv4/ipv6 checking later */
1077 KASSERT(ip->ip_v == IPVERSION,
1078 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
1083 * Enable TSO and specify the size of the segments.
1084 * The TCP pseudo header checksum is always provided.
1085 * XXX: Fixme: This is currently not the case for IPv6.
1088 KASSERT(len > tp->t_maxopd - optlen,
1089 ("%s: len <= tso_segsz", __func__));
1090 m->m_pkthdr.csum_flags |= CSUM_TSO;
1091 m->m_pkthdr.tso_segsz = tp->t_maxopd - optlen;
1095 KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL),
1096 ("%s: mbuf chain shorter than expected: %ld + %u + %u - %u != %u",
1097 __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL)));
1099 KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL),
1100 ("%s: mbuf chain shorter than expected: %ld + %u + %u != %u",
1101 __func__, len, hdrlen, ipoptlen, m_length(m, NULL)));
1105 * In transmit state, time the transmission and arrange for
1106 * the retransmit. In persist state, just set snd_max.
1108 if ((tp->t_flags & TF_FORCEDATA) == 0 ||
1109 !tcp_timer_active(tp, TT_PERSIST)) {
1110 tcp_seq startseq = tp->snd_nxt;
1113 * Advance snd_nxt over sequence space of this segment.
1115 if (flags & (TH_SYN|TH_FIN)) {
1118 if (flags & TH_FIN) {
1120 tp->t_flags |= TF_SENTFIN;
1126 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
1127 tp->snd_max = tp->snd_nxt;
1129 * Time this transmission if not a retransmission and
1130 * not currently timing anything.
1132 if (tp->t_rtttime == 0) {
1133 tp->t_rtttime = ticks;
1134 tp->t_rtseq = startseq;
1135 TCPSTAT_INC(tcps_segstimed);
1140 * Set retransmit timer if not currently set,
1141 * and not doing a pure ack or a keep-alive probe.
1142 * Initial value for retransmit timer is smoothed
1143 * round-trip time + 2 * round-trip time variance.
1144 * Initialize shift counter which is used for backoff
1145 * of retransmit time.
1148 if (!tcp_timer_active(tp, TT_REXMT) &&
1149 ((sack_rxmit && tp->snd_nxt != tp->snd_max) ||
1150 (tp->snd_nxt != tp->snd_una))) {
1151 if (tcp_timer_active(tp, TT_PERSIST)) {
1152 tcp_timer_activate(tp, TT_PERSIST, 0);
1155 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1159 * Persist case, update snd_max but since we are in
1160 * persist mode (no window) we do not update snd_nxt.
1165 if (flags & TH_FIN) {
1167 tp->t_flags |= TF_SENTFIN;
1169 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
1170 tp->snd_max = tp->snd_nxt + len;
1173 /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */
1174 hhook_run_tcp_est_out(tp, th, &to, len, tso);
1180 if (so->so_options & SO_DEBUG) {
1186 save = ipov->ih_len;
1187 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */);
1189 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
1193 ipov->ih_len = save;
1195 #endif /* TCPDEBUG */
1198 * Fill in IP length and desired time to live and
1199 * send to IP level. There should be a better way
1200 * to handle ttl and tos; we could keep them in
1201 * the template, but need a way to checksum without them.
1204 * m->m_pkthdr.len should have been set before cksum calcuration,
1205 * because in6_cksum() need it.
1210 * we separately set hoplimit for every segment, since the
1211 * user might want to change the value via setsockopt.
1212 * Also, desired default hop limit might be changed via
1213 * Neighbor Discovery.
1215 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL);
1217 /* TODO: IPv6 IP6TOS_ECT bit on */
1218 error = ip6_output(m,
1219 tp->t_inpcb->in6p_outputopts, NULL,
1220 ((so->so_options & SO_DONTROUTE) ?
1221 IP_ROUTETOIF : 0), NULL, NULL, tp->t_inpcb);
1224 #if defined(INET) && defined(INET6)
1229 ip->ip_len = m->m_pkthdr.len;
1231 if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO)
1232 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL);
1235 * If we do path MTU discovery, then we set DF on every packet.
1236 * This might not be the best thing to do according to RFC3390
1237 * Section 2. However the tcp hostcache migitates the problem
1238 * so it affects only the first tcp connection with a host.
1240 * NB: Don't set DF on small MTU/MSS to have a safe fallback.
1242 if (V_path_mtu_discovery && tp->t_maxopd > V_tcp_minmss)
1243 ip->ip_off |= IP_DF;
1245 error = ip_output(m, tp->t_inpcb->inp_options, NULL,
1246 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0,
1253 * We know that the packet was lost, so back out the
1254 * sequence number advance, if any.
1256 * If the error is EPERM the packet got blocked by the
1257 * local firewall. Normally we should terminate the
1258 * connection but the blocking may have been spurious
1259 * due to a firewall reconfiguration cycle. So we treat
1260 * it like a packet loss and let the retransmit timer and
1261 * timeouts do their work over time.
1262 * XXX: It is a POLA question whether calling tcp_drop right
1263 * away would be the really correct behavior instead.
1265 if (((tp->t_flags & TF_FORCEDATA) == 0 ||
1266 !tcp_timer_active(tp, TT_PERSIST)) &&
1267 ((flags & TH_SYN) == 0) &&
1271 tp->sackhint.sack_bytes_rexmit -= len;
1272 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
1273 ("sackhint bytes rtx >= 0"));
1278 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */
1281 tp->t_softerror = error;
1284 if (!tcp_timer_active(tp, TT_REXMT) &&
1285 !tcp_timer_active(tp, TT_PERSIST))
1286 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1287 tp->snd_cwnd = tp->t_maxseg;
1291 * For some reason the interface we used initially
1292 * to send segments changed to another or lowered
1295 * tcp_mtudisc() will find out the new MTU and as
1296 * its last action, initiate retransmission, so it
1297 * is important to not do so here.
1299 * If TSO was active we either got an interface
1300 * without TSO capabilits or TSO was turned off.
1301 * Disable it for this connection as too and
1302 * immediatly retry with MSS sized segments generated
1306 tp->t_flags &= ~TF_TSO;
1307 tcp_mtudisc(tp->t_inpcb, -1);
1313 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1314 tp->t_softerror = error;
1322 TCPSTAT_INC(tcps_sndtotal);
1325 * Data sent (as far as we can tell).
1326 * If this advertises a larger window than any other segment,
1327 * then remember the size of the advertised window.
1328 * Any pending ACK has now been sent.
1330 if (recwin >= 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv))
1331 tp->rcv_adv = tp->rcv_nxt + recwin;
1332 tp->last_ack_sent = tp->rcv_nxt;
1333 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK);
1334 if (tcp_timer_active(tp, TT_DELACK))
1335 tcp_timer_activate(tp, TT_DELACK, 0);
1338 * This completely breaks TCP if newreno is turned on. What happens
1339 * is that if delayed-acks are turned on on the receiver, this code
1340 * on the transmitter effectively destroys the TCP window, forcing
1341 * it to four packets (1.5Kx4 = 6K window).
1343 if (sendalot && --maxburst)
1352 tcp_setpersist(struct tcpcb *tp)
1354 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
1357 tp->t_flags &= ~TF_PREVVALID;
1358 if (tcp_timer_active(tp, TT_REXMT))
1359 panic("tcp_setpersist: retransmit pending");
1361 * Start/restart persistance timer.
1363 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
1364 TCPTV_PERSMIN, TCPTV_PERSMAX);
1365 tcp_timer_activate(tp, TT_PERSIST, tt);
1366 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
1371 * Insert TCP options according to the supplied parameters to the place
1372 * optp in a consistent way. Can handle unaligned destinations.
1374 * The order of the option processing is crucial for optimal packing and
1375 * alignment for the scarce option space.
1377 * The optimal order for a SYN/SYN-ACK segment is:
1378 * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) +
1379 * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40.
1381 * The SACK options should be last. SACK blocks consume 8*n+2 bytes.
1382 * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks).
1383 * At minimum we need 10 bytes (to generate 1 SACK block). If both
1384 * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present,
1385 * we only have 10 bytes for SACK options (40 - (12 + 18)).
1388 tcp_addoptions(struct tcpopt *to, u_char *optp)
1390 u_int mask, optlen = 0;
1392 for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) {
1393 if ((to->to_flags & mask) != mask)
1395 if (optlen == TCP_MAXOLEN)
1397 switch (to->to_flags & mask) {
1399 while (optlen % 4) {
1400 optlen += TCPOLEN_NOP;
1401 *optp++ = TCPOPT_NOP;
1403 if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG)
1405 optlen += TCPOLEN_MAXSEG;
1406 *optp++ = TCPOPT_MAXSEG;
1407 *optp++ = TCPOLEN_MAXSEG;
1408 to->to_mss = htons(to->to_mss);
1409 bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss));
1410 optp += sizeof(to->to_mss);
1413 while (!optlen || optlen % 2 != 1) {
1414 optlen += TCPOLEN_NOP;
1415 *optp++ = TCPOPT_NOP;
1417 if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW)
1419 optlen += TCPOLEN_WINDOW;
1420 *optp++ = TCPOPT_WINDOW;
1421 *optp++ = TCPOLEN_WINDOW;
1422 *optp++ = to->to_wscale;
1425 while (optlen % 2) {
1426 optlen += TCPOLEN_NOP;
1427 *optp++ = TCPOPT_NOP;
1429 if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED)
1431 optlen += TCPOLEN_SACK_PERMITTED;
1432 *optp++ = TCPOPT_SACK_PERMITTED;
1433 *optp++ = TCPOLEN_SACK_PERMITTED;
1436 while (!optlen || optlen % 4 != 2) {
1437 optlen += TCPOLEN_NOP;
1438 *optp++ = TCPOPT_NOP;
1440 if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP)
1442 optlen += TCPOLEN_TIMESTAMP;
1443 *optp++ = TCPOPT_TIMESTAMP;
1444 *optp++ = TCPOLEN_TIMESTAMP;
1445 to->to_tsval = htonl(to->to_tsval);
1446 to->to_tsecr = htonl(to->to_tsecr);
1447 bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval));
1448 optp += sizeof(to->to_tsval);
1449 bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr));
1450 optp += sizeof(to->to_tsecr);
1454 int siglen = TCPOLEN_SIGNATURE - 2;
1456 while (!optlen || optlen % 4 != 2) {
1457 optlen += TCPOLEN_NOP;
1458 *optp++ = TCPOPT_NOP;
1460 if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE)
1462 optlen += TCPOLEN_SIGNATURE;
1463 *optp++ = TCPOPT_SIGNATURE;
1464 *optp++ = TCPOLEN_SIGNATURE;
1465 to->to_signature = optp;
1473 struct sackblk *sack = (struct sackblk *)to->to_sacks;
1476 while (!optlen || optlen % 4 != 2) {
1477 optlen += TCPOLEN_NOP;
1478 *optp++ = TCPOPT_NOP;
1480 if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK)
1482 optlen += TCPOLEN_SACKHDR;
1483 *optp++ = TCPOPT_SACK;
1484 sackblks = min(to->to_nsacks,
1485 (TCP_MAXOLEN - optlen) / TCPOLEN_SACK);
1486 *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK;
1487 while (sackblks--) {
1488 sack_seq = htonl(sack->start);
1489 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
1490 optp += sizeof(sack_seq);
1491 sack_seq = htonl(sack->end);
1492 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
1493 optp += sizeof(sack_seq);
1494 optlen += TCPOLEN_SACK;
1497 TCPSTAT_INC(tcps_sack_send_blocks);
1501 panic("%s: unknown TCP option type", __func__);
1506 /* Terminate and pad TCP options to a 4 byte boundary. */
1508 optlen += TCPOLEN_EOL;
1509 *optp++ = TCPOPT_EOL;
1512 * According to RFC 793 (STD0007):
1513 * "The content of the header beyond the End-of-Option option
1514 * must be header padding (i.e., zero)."
1515 * and later: "The padding is composed of zeros."
1517 while (optlen % 4) {
1518 optlen += TCPOLEN_PAD;
1519 *optp++ = TCPOPT_PAD;
1522 KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__));