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 * Compare available window to amount of window
549 * known to peer (as advertised window less
550 * next expected input). If the difference is at least two
551 * max size segments, or at least 50% of the maximum possible
552 * window, then want to send a window update to peer.
553 * Skip this if the connection is in T/TCP half-open state.
554 * Don't send pure window updates when the peer has closed
555 * the connection and won't ever send more data.
557 if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) &&
558 !TCPS_HAVERCVDFIN(tp->t_state)) {
560 * "adv" is the amount we can increase the window,
561 * taking into account that we are limited by
562 * TCP_MAXWIN << tp->rcv_scale.
567 adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale);
568 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) {
569 oldwin = (tp->rcv_adv - tp->rcv_nxt);
575 * If the new window size ends up being the same as the old
576 * size when it is scaled, then don't force a window update.
578 if (oldwin >> tp->rcv_scale == (adv + oldwin) >> tp->rcv_scale)
580 if (adv >= (long) (2 * tp->t_maxseg))
582 if (2 * adv >= (long) so->so_rcv.sb_hiwat)
588 * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
589 * is also a catch-all for the retransmit timer timeout case.
591 if (tp->t_flags & TF_ACKNOW)
593 if ((flags & TH_RST) ||
594 ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
596 if (SEQ_GT(tp->snd_up, tp->snd_una))
599 * If our state indicates that FIN should be sent
600 * and we have not yet done so, then we need to send.
602 if (flags & TH_FIN &&
603 ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
606 * In SACK, it is possible for tcp_output to fail to send a segment
607 * after the retransmission timer has been turned off. Make sure
608 * that the retransmission timer is set.
610 if ((tp->t_flags & TF_SACK_PERMIT) &&
611 SEQ_GT(tp->snd_max, tp->snd_una) &&
612 !tcp_timer_active(tp, TT_REXMT) &&
613 !tcp_timer_active(tp, TT_PERSIST)) {
614 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
618 * TCP window updates are not reliable, rather a polling protocol
619 * using ``persist'' packets is used to insure receipt of window
620 * updates. The three ``states'' for the output side are:
621 * idle not doing retransmits or persists
622 * persisting to move a small or zero window
623 * (re)transmitting and thereby not persisting
625 * tcp_timer_active(tp, TT_PERSIST)
626 * is true when we are in persist state.
627 * (tp->t_flags & TF_FORCEDATA)
628 * is set when we are called to send a persist packet.
629 * tcp_timer_active(tp, TT_REXMT)
630 * is set when we are retransmitting
631 * The output side is idle when both timers are zero.
633 * If send window is too small, there is data to transmit, and no
634 * retransmit or persist is pending, then go to persist state.
635 * If nothing happens soon, send when timer expires:
636 * if window is nonzero, transmit what we can,
637 * otherwise force out a byte.
639 if (so->so_snd.sb_cc && !tcp_timer_active(tp, TT_REXMT) &&
640 !tcp_timer_active(tp, TT_PERSIST)) {
646 * No reason to send a segment, just return.
649 SOCKBUF_UNLOCK(&so->so_snd);
653 SOCKBUF_LOCK_ASSERT(&so->so_snd);
655 * Before ESTABLISHED, force sending of initial options
656 * unless TCP set not to do any options.
657 * NOTE: we assume that the IP/TCP header plus TCP options
658 * always fit in a single mbuf, leaving room for a maximum
660 * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
665 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
668 hdrlen = sizeof (struct tcpiphdr);
671 * Compute options for segment.
672 * We only have to care about SYN and established connection
673 * segments. Options for SYN-ACK segments are handled in TCP
676 if ((tp->t_flags & TF_NOOPT) == 0) {
678 /* Maximum segment size. */
679 if (flags & TH_SYN) {
680 tp->snd_nxt = tp->iss;
681 to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc);
682 to.to_flags |= TOF_MSS;
684 /* Window scaling. */
685 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) {
686 to.to_wscale = tp->request_r_scale;
687 to.to_flags |= TOF_SCALE;
690 if ((tp->t_flags & TF_RCVD_TSTMP) ||
691 ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) {
692 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
693 to.to_tsecr = tp->ts_recent;
694 to.to_flags |= TOF_TS;
695 /* Set receive buffer autosizing timestamp. */
696 if (tp->rfbuf_ts == 0 &&
697 (so->so_rcv.sb_flags & SB_AUTOSIZE))
698 tp->rfbuf_ts = tcp_ts_getticks();
700 /* Selective ACK's. */
701 if (tp->t_flags & TF_SACK_PERMIT) {
703 to.to_flags |= TOF_SACKPERM;
704 else if (TCPS_HAVEESTABLISHED(tp->t_state) &&
705 (tp->t_flags & TF_SACK_PERMIT) &&
706 tp->rcv_numsacks > 0) {
707 to.to_flags |= TOF_SACK;
708 to.to_nsacks = tp->rcv_numsacks;
709 to.to_sacks = (u_char *)tp->sackblks;
713 /* TCP-MD5 (RFC2385). */
714 if (tp->t_flags & TF_SIGNATURE)
715 to.to_flags |= TOF_SIGNATURE;
716 #endif /* TCP_SIGNATURE */
718 /* Processing the options. */
719 hdrlen += optlen = tcp_addoptions(&to, opt);
724 ipoptlen = ip6_optlen(tp->t_inpcb);
727 if (tp->t_inpcb->inp_options)
728 ipoptlen = tp->t_inpcb->inp_options->m_len -
729 offsetof(struct ipoption, ipopt_list);
733 ipoptlen += ipsec_optlen;
737 * Adjust data length if insertion of options will
738 * bump the packet length beyond the t_maxopd length.
739 * Clear the FIN bit because we cut off the tail of
742 if (len + optlen + ipoptlen > tp->t_maxopd) {
746 KASSERT(ipoptlen == 0,
747 ("%s: TSO can't do IP options", __func__));
750 * Limit a burst to IP_MAXPACKET minus IP,
751 * TCP and options length to keep ip->ip_len
754 if (len > IP_MAXPACKET - hdrlen) {
755 len = IP_MAXPACKET - hdrlen;
760 * Prevent the last segment from being
761 * fractional unless the send sockbuf can
764 if (sendalot && off + len < so->so_snd.sb_cc) {
765 len -= len % (tp->t_maxopd - optlen);
770 * Send the FIN in a separate segment
771 * after the bulk sending is done.
772 * We don't trust the TSO implementations
773 * to clear the FIN flag on all but the
776 if (tp->t_flags & TF_NEEDFIN)
780 len = tp->t_maxopd - optlen - ipoptlen;
786 KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET,
787 ("%s: len > IP_MAXPACKET", __func__));
789 /*#ifdef DIAGNOSTIC*/
791 if (max_linkhdr + hdrlen > MCLBYTES)
793 if (max_linkhdr + hdrlen > MHLEN)
795 panic("tcphdr too big");
799 * This KASSERT is here to catch edge cases at a well defined place.
800 * Before, those had triggered (random) panic conditions further down.
802 KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
805 * Grab a header mbuf, attaching a copy of data to
806 * be transmitted, and initialize the header from
807 * the template for sends on this connection.
813 if ((tp->t_flags & TF_FORCEDATA) && len == 1)
814 TCPSTAT_INC(tcps_sndprobe);
815 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) {
816 tp->t_sndrexmitpack++;
817 TCPSTAT_INC(tcps_sndrexmitpack);
818 TCPSTAT_ADD(tcps_sndrexmitbyte, len);
820 TCPSTAT_INC(tcps_sndpack);
821 TCPSTAT_ADD(tcps_sndbyte, len);
823 MGETHDR(m, M_DONTWAIT, MT_DATA);
825 SOCKBUF_UNLOCK(&so->so_snd);
830 if (MHLEN < hdrlen + max_linkhdr) {
831 MCLGET(m, M_DONTWAIT);
832 if ((m->m_flags & M_EXT) == 0) {
833 SOCKBUF_UNLOCK(&so->so_snd);
840 m->m_data += max_linkhdr;
844 * Start the m_copy functions from the closest mbuf
845 * to the offset in the socket buffer chain.
847 mb = sbsndptr(&so->so_snd, off, len, &moff);
849 if (len <= MHLEN - hdrlen - max_linkhdr) {
850 m_copydata(mb, moff, (int)len,
851 mtod(m, caddr_t) + hdrlen);
854 m->m_next = m_copy(mb, moff, (int)len);
855 if (m->m_next == NULL) {
856 SOCKBUF_UNLOCK(&so->so_snd);
864 * If we're sending everything we've got, set PUSH.
865 * (This will keep happy those implementations which only
866 * give data to the user when a buffer fills or
869 if (off + len == so->so_snd.sb_cc)
871 SOCKBUF_UNLOCK(&so->so_snd);
873 SOCKBUF_UNLOCK(&so->so_snd);
874 if (tp->t_flags & TF_ACKNOW)
875 TCPSTAT_INC(tcps_sndacks);
876 else if (flags & (TH_SYN|TH_FIN|TH_RST))
877 TCPSTAT_INC(tcps_sndctrl);
878 else if (SEQ_GT(tp->snd_up, tp->snd_una))
879 TCPSTAT_INC(tcps_sndurg);
881 TCPSTAT_INC(tcps_sndwinup);
883 MGETHDR(m, M_DONTWAIT, MT_DATA);
889 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
894 m->m_data += max_linkhdr;
897 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
898 m->m_pkthdr.rcvif = (struct ifnet *)0;
900 mac_inpcb_create_mbuf(tp->t_inpcb, m);
904 ip6 = mtod(m, struct ip6_hdr *);
905 th = (struct tcphdr *)(ip6 + 1);
906 tcpip_fillheaders(tp->t_inpcb, ip6, th);
910 ip = mtod(m, struct ip *);
911 ipov = (struct ipovly *)ip;
912 th = (struct tcphdr *)(ip + 1);
913 tcpip_fillheaders(tp->t_inpcb, ip, th);
917 * Fill in fields, remembering maximum advertised
918 * window for use in delaying messages about window sizes.
919 * If resending a FIN, be sure not to use a new sequence number.
921 if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
922 tp->snd_nxt == tp->snd_max)
925 * If we are starting a connection, send ECN setup
926 * SYN packet. If we are on a retransmit, we may
927 * resend those bits a number of times as per
930 if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn) {
931 if (tp->t_rxtshift >= 1) {
932 if (tp->t_rxtshift <= V_tcp_ecn_maxretries)
933 flags |= TH_ECE|TH_CWR;
935 flags |= TH_ECE|TH_CWR;
938 if (tp->t_state == TCPS_ESTABLISHED &&
939 (tp->t_flags & TF_ECN_PERMIT)) {
941 * If the peer has ECN, mark data packets with
942 * ECN capable transmission (ECT).
943 * Ignore pure ack packets, retransmissions and window probes.
945 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) &&
946 !((tp->t_flags & TF_FORCEDATA) && len == 1)) {
949 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
952 ip->ip_tos |= IPTOS_ECN_ECT0;
953 TCPSTAT_INC(tcps_ecn_ect0);
957 * Reply with proper ECN notifications.
959 if (tp->t_flags & TF_ECN_SND_CWR) {
961 tp->t_flags &= ~TF_ECN_SND_CWR;
963 if (tp->t_flags & TF_ECN_SND_ECE)
968 * If we are doing retransmissions, then snd_nxt will
969 * not reflect the first unsent octet. For ACK only
970 * packets, we do not want the sequence number of the
971 * retransmitted packet, we want the sequence number
972 * of the next unsent octet. So, if there is no data
973 * (and no SYN or FIN), use snd_max instead of snd_nxt
974 * when filling in ti_seq. But if we are in persist
975 * state, snd_max might reflect one byte beyond the
976 * right edge of the window, so use snd_nxt in that
977 * case, since we know we aren't doing a retransmission.
978 * (retransmit and persist are mutually exclusive...)
980 if (sack_rxmit == 0) {
981 if (len || (flags & (TH_SYN|TH_FIN)) ||
982 tcp_timer_active(tp, TT_PERSIST))
983 th->th_seq = htonl(tp->snd_nxt);
985 th->th_seq = htonl(tp->snd_max);
987 th->th_seq = htonl(p->rxmit);
989 tp->sackhint.sack_bytes_rexmit += len;
991 th->th_ack = htonl(tp->rcv_nxt);
993 bcopy(opt, th + 1, optlen);
994 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
996 th->th_flags = flags;
998 * Calculate receive window. Don't shrink window,
999 * but avoid silly window syndrome.
1001 if (recwin < (long)(so->so_rcv.sb_hiwat / 4) &&
1002 recwin < (long)tp->t_maxseg)
1004 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
1005 recwin < (long)(tp->rcv_adv - tp->rcv_nxt))
1006 recwin = (long)(tp->rcv_adv - tp->rcv_nxt);
1007 if (recwin > (long)TCP_MAXWIN << tp->rcv_scale)
1008 recwin = (long)TCP_MAXWIN << tp->rcv_scale;
1011 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1012 * or <SYN,ACK>) segment itself is never scaled. The <SYN,ACK>
1013 * case is handled in syncache.
1016 th->th_win = htons((u_short)
1017 (min(sbspace(&so->so_rcv), TCP_MAXWIN)));
1019 th->th_win = htons((u_short)(recwin >> tp->rcv_scale));
1022 * Adjust the RXWIN0SENT flag - indicate that we have advertised
1023 * a 0 window. This may cause the remote transmitter to stall. This
1024 * flag tells soreceive() to disable delayed acknowledgements when
1025 * draining the buffer. This can occur if the receiver is attempting
1026 * to read more data than can be buffered prior to transmitting on
1029 if (th->th_win == 0) {
1031 tp->t_flags |= TF_RXWIN0SENT;
1033 tp->t_flags &= ~TF_RXWIN0SENT;
1034 if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
1035 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
1036 th->th_flags |= TH_URG;
1039 * If no urgent pointer to send, then we pull
1040 * the urgent pointer to the left edge of the send window
1041 * so that it doesn't drift into the send window on sequence
1042 * number wraparound.
1044 tp->snd_up = tp->snd_una; /* drag it along */
1046 #ifdef TCP_SIGNATURE
1047 if (tp->t_flags & TF_SIGNATURE) {
1048 int sigoff = to.to_signature - opt;
1049 tcp_signature_compute(m, 0, len, optlen,
1050 (u_char *)(th + 1) + sigoff, IPSEC_DIR_OUTBOUND);
1055 * Put TCP length in extended header, and then
1056 * checksum extended header and data.
1058 m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
1059 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1063 * ip6_plen is not need to be filled now, and will be filled
1066 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1067 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) +
1068 optlen + len, IPPROTO_TCP, 0);
1071 #if defined(INET6) && defined(INET)
1076 m->m_pkthdr.csum_flags = CSUM_TCP;
1077 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1078 htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen));
1080 /* IP version must be set here for ipv4/ipv6 checking later */
1081 KASSERT(ip->ip_v == IPVERSION,
1082 ("%s: IP version incorrect: %d", __func__, ip->ip_v));
1087 * Enable TSO and specify the size of the segments.
1088 * The TCP pseudo header checksum is always provided.
1089 * XXX: Fixme: This is currently not the case for IPv6.
1092 KASSERT(len > tp->t_maxopd - optlen,
1093 ("%s: len <= tso_segsz", __func__));
1094 m->m_pkthdr.csum_flags |= CSUM_TSO;
1095 m->m_pkthdr.tso_segsz = tp->t_maxopd - optlen;
1099 KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL),
1100 ("%s: mbuf chain shorter than expected: %ld + %u + %u - %u != %u",
1101 __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL)));
1103 KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL),
1104 ("%s: mbuf chain shorter than expected: %ld + %u + %u != %u",
1105 __func__, len, hdrlen, ipoptlen, m_length(m, NULL)));
1109 * In transmit state, time the transmission and arrange for
1110 * the retransmit. In persist state, just set snd_max.
1112 if ((tp->t_flags & TF_FORCEDATA) == 0 ||
1113 !tcp_timer_active(tp, TT_PERSIST)) {
1114 tcp_seq startseq = tp->snd_nxt;
1117 * Advance snd_nxt over sequence space of this segment.
1119 if (flags & (TH_SYN|TH_FIN)) {
1122 if (flags & TH_FIN) {
1124 tp->t_flags |= TF_SENTFIN;
1130 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
1131 tp->snd_max = tp->snd_nxt;
1133 * Time this transmission if not a retransmission and
1134 * not currently timing anything.
1136 if (tp->t_rtttime == 0) {
1137 tp->t_rtttime = ticks;
1138 tp->t_rtseq = startseq;
1139 TCPSTAT_INC(tcps_segstimed);
1144 * Set retransmit timer if not currently set,
1145 * and not doing a pure ack or a keep-alive probe.
1146 * Initial value for retransmit timer is smoothed
1147 * round-trip time + 2 * round-trip time variance.
1148 * Initialize shift counter which is used for backoff
1149 * of retransmit time.
1152 if (!tcp_timer_active(tp, TT_REXMT) &&
1153 ((sack_rxmit && tp->snd_nxt != tp->snd_max) ||
1154 (tp->snd_nxt != tp->snd_una))) {
1155 if (tcp_timer_active(tp, TT_PERSIST)) {
1156 tcp_timer_activate(tp, TT_PERSIST, 0);
1159 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1163 * Persist case, update snd_max but since we are in
1164 * persist mode (no window) we do not update snd_nxt.
1169 if (flags & TH_FIN) {
1171 tp->t_flags |= TF_SENTFIN;
1173 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
1174 tp->snd_max = tp->snd_nxt + len;
1177 /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */
1178 hhook_run_tcp_est_out(tp, th, &to, len, tso);
1184 if (so->so_options & SO_DEBUG) {
1190 save = ipov->ih_len;
1191 ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */);
1193 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
1197 ipov->ih_len = save;
1199 #endif /* TCPDEBUG */
1202 * Fill in IP length and desired time to live and
1203 * send to IP level. There should be a better way
1204 * to handle ttl and tos; we could keep them in
1205 * the template, but need a way to checksum without them.
1208 * m->m_pkthdr.len should have been set before cksum calcuration,
1209 * because in6_cksum() need it.
1213 struct route_in6 ro;
1215 bzero(&ro, sizeof(ro));
1217 * we separately set hoplimit for every segment, since the
1218 * user might want to change the value via setsockopt.
1219 * Also, desired default hop limit might be changed via
1220 * Neighbor Discovery.
1222 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL);
1224 /* TODO: IPv6 IP6TOS_ECT bit on */
1225 error = ip6_output(m, tp->t_inpcb->in6p_outputopts, &ro,
1226 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
1227 NULL, NULL, tp->t_inpcb);
1229 if (error == EMSGSIZE && ro.ro_rt != NULL)
1230 mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1234 #if defined(INET) && defined(INET6)
1241 bzero(&ro, sizeof(ro));
1242 ip->ip_len = m->m_pkthdr.len;
1244 if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO)
1245 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL);
1248 * If we do path MTU discovery, then we set DF on every packet.
1249 * This might not be the best thing to do according to RFC3390
1250 * Section 2. However the tcp hostcache migitates the problem
1251 * so it affects only the first tcp connection with a host.
1253 * NB: Don't set DF on small MTU/MSS to have a safe fallback.
1255 if (V_path_mtu_discovery && tp->t_maxopd > V_tcp_minmss)
1256 ip->ip_off |= IP_DF;
1258 error = ip_output(m, tp->t_inpcb->inp_options, &ro,
1259 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0,
1262 if (error == EMSGSIZE && ro.ro_rt != NULL)
1263 mtu = ro.ro_rt->rt_rmx.rmx_mtu;
1270 * We know that the packet was lost, so back out the
1271 * sequence number advance, if any.
1273 * If the error is EPERM the packet got blocked by the
1274 * local firewall. Normally we should terminate the
1275 * connection but the blocking may have been spurious
1276 * due to a firewall reconfiguration cycle. So we treat
1277 * it like a packet loss and let the retransmit timer and
1278 * timeouts do their work over time.
1279 * XXX: It is a POLA question whether calling tcp_drop right
1280 * away would be the really correct behavior instead.
1282 if (((tp->t_flags & TF_FORCEDATA) == 0 ||
1283 !tcp_timer_active(tp, TT_PERSIST)) &&
1284 ((flags & TH_SYN) == 0) &&
1288 tp->sackhint.sack_bytes_rexmit -= len;
1289 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
1290 ("sackhint bytes rtx >= 0"));
1295 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */
1298 tp->t_softerror = error;
1301 if (!tcp_timer_active(tp, TT_REXMT) &&
1302 !tcp_timer_active(tp, TT_PERSIST))
1303 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1304 tp->snd_cwnd = tp->t_maxseg;
1308 * For some reason the interface we used initially
1309 * to send segments changed to another or lowered
1311 * If TSO was active we either got an interface
1312 * without TSO capabilits or TSO was turned off.
1313 * If we obtained mtu from ip_output() then update
1317 tp->t_flags &= ~TF_TSO;
1319 tcp_mss_update(tp, -1, mtu, NULL, NULL);
1327 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1328 tp->t_softerror = error;
1336 TCPSTAT_INC(tcps_sndtotal);
1339 * Data sent (as far as we can tell).
1340 * If this advertises a larger window than any other segment,
1341 * then remember the size of the advertised window.
1342 * Any pending ACK has now been sent.
1344 if (recwin >= 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv))
1345 tp->rcv_adv = tp->rcv_nxt + recwin;
1346 tp->last_ack_sent = tp->rcv_nxt;
1347 tp->t_flags &= ~(TF_ACKNOW | TF_DELACK);
1348 if (tcp_timer_active(tp, TT_DELACK))
1349 tcp_timer_activate(tp, TT_DELACK, 0);
1352 * This completely breaks TCP if newreno is turned on. What happens
1353 * is that if delayed-acks are turned on on the receiver, this code
1354 * on the transmitter effectively destroys the TCP window, forcing
1355 * it to four packets (1.5Kx4 = 6K window).
1357 if (sendalot && --maxburst)
1366 tcp_setpersist(struct tcpcb *tp)
1368 int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
1371 tp->t_flags &= ~TF_PREVVALID;
1372 if (tcp_timer_active(tp, TT_REXMT))
1373 panic("tcp_setpersist: retransmit pending");
1375 * Start/restart persistance timer.
1377 TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
1378 TCPTV_PERSMIN, TCPTV_PERSMAX);
1379 tcp_timer_activate(tp, TT_PERSIST, tt);
1380 if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
1385 * Insert TCP options according to the supplied parameters to the place
1386 * optp in a consistent way. Can handle unaligned destinations.
1388 * The order of the option processing is crucial for optimal packing and
1389 * alignment for the scarce option space.
1391 * The optimal order for a SYN/SYN-ACK segment is:
1392 * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) +
1393 * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40.
1395 * The SACK options should be last. SACK blocks consume 8*n+2 bytes.
1396 * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks).
1397 * At minimum we need 10 bytes (to generate 1 SACK block). If both
1398 * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present,
1399 * we only have 10 bytes for SACK options (40 - (12 + 18)).
1402 tcp_addoptions(struct tcpopt *to, u_char *optp)
1404 u_int mask, optlen = 0;
1406 for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) {
1407 if ((to->to_flags & mask) != mask)
1409 if (optlen == TCP_MAXOLEN)
1411 switch (to->to_flags & mask) {
1413 while (optlen % 4) {
1414 optlen += TCPOLEN_NOP;
1415 *optp++ = TCPOPT_NOP;
1417 if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG)
1419 optlen += TCPOLEN_MAXSEG;
1420 *optp++ = TCPOPT_MAXSEG;
1421 *optp++ = TCPOLEN_MAXSEG;
1422 to->to_mss = htons(to->to_mss);
1423 bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss));
1424 optp += sizeof(to->to_mss);
1427 while (!optlen || optlen % 2 != 1) {
1428 optlen += TCPOLEN_NOP;
1429 *optp++ = TCPOPT_NOP;
1431 if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW)
1433 optlen += TCPOLEN_WINDOW;
1434 *optp++ = TCPOPT_WINDOW;
1435 *optp++ = TCPOLEN_WINDOW;
1436 *optp++ = to->to_wscale;
1439 while (optlen % 2) {
1440 optlen += TCPOLEN_NOP;
1441 *optp++ = TCPOPT_NOP;
1443 if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED)
1445 optlen += TCPOLEN_SACK_PERMITTED;
1446 *optp++ = TCPOPT_SACK_PERMITTED;
1447 *optp++ = TCPOLEN_SACK_PERMITTED;
1450 while (!optlen || optlen % 4 != 2) {
1451 optlen += TCPOLEN_NOP;
1452 *optp++ = TCPOPT_NOP;
1454 if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP)
1456 optlen += TCPOLEN_TIMESTAMP;
1457 *optp++ = TCPOPT_TIMESTAMP;
1458 *optp++ = TCPOLEN_TIMESTAMP;
1459 to->to_tsval = htonl(to->to_tsval);
1460 to->to_tsecr = htonl(to->to_tsecr);
1461 bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval));
1462 optp += sizeof(to->to_tsval);
1463 bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr));
1464 optp += sizeof(to->to_tsecr);
1468 int siglen = TCPOLEN_SIGNATURE - 2;
1470 while (!optlen || optlen % 4 != 2) {
1471 optlen += TCPOLEN_NOP;
1472 *optp++ = TCPOPT_NOP;
1474 if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE)
1476 optlen += TCPOLEN_SIGNATURE;
1477 *optp++ = TCPOPT_SIGNATURE;
1478 *optp++ = TCPOLEN_SIGNATURE;
1479 to->to_signature = optp;
1487 struct sackblk *sack = (struct sackblk *)to->to_sacks;
1490 while (!optlen || optlen % 4 != 2) {
1491 optlen += TCPOLEN_NOP;
1492 *optp++ = TCPOPT_NOP;
1494 if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK)
1496 optlen += TCPOLEN_SACKHDR;
1497 *optp++ = TCPOPT_SACK;
1498 sackblks = min(to->to_nsacks,
1499 (TCP_MAXOLEN - optlen) / TCPOLEN_SACK);
1500 *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK;
1501 while (sackblks--) {
1502 sack_seq = htonl(sack->start);
1503 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
1504 optp += sizeof(sack_seq);
1505 sack_seq = htonl(sack->end);
1506 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
1507 optp += sizeof(sack_seq);
1508 optlen += TCPOLEN_SACK;
1511 TCPSTAT_INC(tcps_sack_send_blocks);
1515 panic("%s: unknown TCP option type", __func__);
1520 /* Terminate and pad TCP options to a 4 byte boundary. */
1522 optlen += TCPOLEN_EOL;
1523 *optp++ = TCPOPT_EOL;
1526 * According to RFC 793 (STD0007):
1527 * "The content of the header beyond the End-of-Option option
1528 * must be header padding (i.e., zero)."
1529 * and later: "The padding is composed of zeros."
1531 while (optlen % 4) {
1532 optlen += TCPOLEN_PAD;
1533 *optp++ = TCPOPT_PAD;
1536 KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__));