The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/netinet/tcp_output.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice, this list of conditions and the following disclaimer.
   10  * 2. Redistributions in binary form must reproduce the above copyright
   11  *    notice, this list of conditions and the following disclaimer in the
   12  *    documentation and/or other materials provided with the distribution.
   13  * 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.
   16  *
   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
   27  * SUCH DAMAGE.
   28  *
   29  *      @(#)tcp_output.c        8.4 (Berkeley) 5/24/95
   30  */
   31 
   32 #include <sys/cdefs.h>
   33 __FBSDID("$FreeBSD: releng/11.1/sys/netinet/tcp_output.c 319397 2017-06-01 08:19:45Z tuexen $");
   34 
   35 #include "opt_inet.h"
   36 #include "opt_inet6.h"
   37 #include "opt_ipsec.h"
   38 #include "opt_tcpdebug.h"
   39 
   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>
   45 #include <sys/lock.h>
   46 #include <sys/mbuf.h>
   47 #include <sys/mutex.h>
   48 #include <sys/protosw.h>
   49 #include <sys/sdt.h>
   50 #include <sys/socket.h>
   51 #include <sys/socketvar.h>
   52 #include <sys/sysctl.h>
   53 
   54 #include <net/if.h>
   55 #include <net/route.h>
   56 #include <net/vnet.h>
   57 
   58 #include <netinet/in.h>
   59 #include <netinet/in_kdtrace.h>
   60 #include <netinet/in_systm.h>
   61 #include <netinet/ip.h>
   62 #include <netinet/in_pcb.h>
   63 #include <netinet/ip_var.h>
   64 #include <netinet/ip_options.h>
   65 #ifdef INET6
   66 #include <netinet6/in6_pcb.h>
   67 #include <netinet/ip6.h>
   68 #include <netinet6/ip6_var.h>
   69 #endif
   70 #ifdef TCP_RFC7413
   71 #include <netinet/tcp_fastopen.h>
   72 #endif
   73 #include <netinet/tcp.h>
   74 #define TCPOUTFLAGS
   75 #include <netinet/tcp_fsm.h>
   76 #include <netinet/tcp_seq.h>
   77 #include <netinet/tcp_timer.h>
   78 #include <netinet/tcp_var.h>
   79 #include <netinet/tcpip.h>
   80 #include <netinet/cc/cc.h>
   81 #ifdef TCPPCAP
   82 #include <netinet/tcp_pcap.h>
   83 #endif
   84 #ifdef TCPDEBUG
   85 #include <netinet/tcp_debug.h>
   86 #endif
   87 #ifdef TCP_OFFLOAD
   88 #include <netinet/tcp_offload.h>
   89 #endif
   90 
   91 #include <netipsec/ipsec_support.h>
   92 
   93 #include <machine/in_cksum.h>
   94 
   95 #include <security/mac/mac_framework.h>
   96 
   97 VNET_DEFINE(int, path_mtu_discovery) = 1;
   98 SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_VNET | CTLFLAG_RW,
   99         &VNET_NAME(path_mtu_discovery), 1,
  100         "Enable Path MTU Discovery");
  101 
  102 VNET_DEFINE(int, tcp_do_tso) = 1;
  103 #define V_tcp_do_tso            VNET(tcp_do_tso)
  104 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_VNET | CTLFLAG_RW,
  105         &VNET_NAME(tcp_do_tso), 0,
  106         "Enable TCP Segmentation Offload");
  107 
  108 VNET_DEFINE(int, tcp_sendspace) = 1024*32;
  109 #define V_tcp_sendspace VNET(tcp_sendspace)
  110 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_VNET | CTLFLAG_RW,
  111         &VNET_NAME(tcp_sendspace), 0, "Initial send socket buffer size");
  112 
  113 VNET_DEFINE(int, tcp_do_autosndbuf) = 1;
  114 #define V_tcp_do_autosndbuf     VNET(tcp_do_autosndbuf)
  115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
  116         &VNET_NAME(tcp_do_autosndbuf), 0,
  117         "Enable automatic send buffer sizing");
  118 
  119 VNET_DEFINE(int, tcp_autosndbuf_inc) = 8*1024;
  120 #define V_tcp_autosndbuf_inc    VNET(tcp_autosndbuf_inc)
  121 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
  122         &VNET_NAME(tcp_autosndbuf_inc), 0,
  123         "Incrementor step size of automatic send buffer");
  124 
  125 VNET_DEFINE(int, tcp_autosndbuf_max) = 2*1024*1024;
  126 #define V_tcp_autosndbuf_max    VNET(tcp_autosndbuf_max)
  127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
  128         &VNET_NAME(tcp_autosndbuf_max), 0,
  129         "Max size of automatic send buffer");
  130 
  131 /*
  132  * Make sure that either retransmit or persist timer is set for SYN, FIN and
  133  * non-ACK.
  134  */
  135 #define TCP_XMIT_TIMER_ASSERT(tp, len, th_flags)                        \
  136         KASSERT(((len) == 0 && ((th_flags) & (TH_SYN | TH_FIN)) == 0) ||\
  137             tcp_timer_active((tp), TT_REXMT) ||                         \
  138             tcp_timer_active((tp), TT_PERSIST),                         \
  139             ("neither rexmt nor persist timer is set"))
  140 
  141 static void inline      hhook_run_tcp_est_out(struct tcpcb *tp,
  142                             struct tcphdr *th, struct tcpopt *to,
  143                             long len, int tso);
  144 static void inline      cc_after_idle(struct tcpcb *tp);
  145 
  146 /*
  147  * Wrapper for the TCP established output helper hook.
  148  */
  149 static void inline
  150 hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th,
  151     struct tcpopt *to, long len, int tso)
  152 {
  153         struct tcp_hhook_data hhook_data;
  154 
  155         if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) {
  156                 hhook_data.tp = tp;
  157                 hhook_data.th = th;
  158                 hhook_data.to = to;
  159                 hhook_data.len = len;
  160                 hhook_data.tso = tso;
  161 
  162                 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data,
  163                     tp->osd);
  164         }
  165 }
  166 
  167 /*
  168  * CC wrapper hook functions
  169  */
  170 static void inline
  171 cc_after_idle(struct tcpcb *tp)
  172 {
  173         INP_WLOCK_ASSERT(tp->t_inpcb);
  174 
  175         if (CC_ALGO(tp)->after_idle != NULL)
  176                 CC_ALGO(tp)->after_idle(tp->ccv);
  177 }
  178 
  179 /*
  180  * Tcp output routine: figure out what should be sent and send it.
  181  */
  182 int
  183 tcp_output(struct tcpcb *tp)
  184 {
  185         struct socket *so = tp->t_inpcb->inp_socket;
  186         long len, recwin, sendwin;
  187         int off, flags, error = 0;      /* Keep compiler happy */
  188         struct mbuf *m;
  189         struct ip *ip = NULL;
  190         struct ipovly *ipov = NULL;
  191         struct tcphdr *th;
  192         u_char opt[TCP_MAXOLEN];
  193         unsigned ipoptlen, optlen, hdrlen;
  194 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
  195         unsigned ipsec_optlen = 0;
  196 #endif
  197         int idle, sendalot;
  198         int sack_rxmit, sack_bytes_rxmt;
  199         struct sackhole *p;
  200         int tso, mtu;
  201         struct tcpopt to;
  202 #if 0
  203         int maxburst = TCP_MAXBURST;
  204 #endif
  205 #ifdef INET6
  206         struct ip6_hdr *ip6 = NULL;
  207         int isipv6;
  208 
  209         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
  210 #endif
  211 
  212         INP_WLOCK_ASSERT(tp->t_inpcb);
  213 
  214 #ifdef TCP_OFFLOAD
  215         if (tp->t_flags & TF_TOE)
  216                 return (tcp_offload_output(tp));
  217 #endif
  218 
  219 #ifdef TCP_RFC7413
  220         /*
  221          * For TFO connections in SYN_RECEIVED, only allow the initial
  222          * SYN|ACK and those sent by the retransmit timer.
  223          */
  224         if ((tp->t_flags & TF_FASTOPEN) &&
  225             (tp->t_state == TCPS_SYN_RECEIVED) &&
  226             SEQ_GT(tp->snd_max, tp->snd_una) &&    /* initial SYN|ACK sent */
  227             (tp->snd_nxt != tp->snd_una))          /* not a retransmit */
  228                 return (0);
  229 #endif
  230         /*
  231          * Determine length of data that should be transmitted,
  232          * and flags that will be used.
  233          * If there is some data or critical controls (SYN, RST)
  234          * to send, then transmit; otherwise, investigate further.
  235          */
  236         idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
  237         if (idle && ticks - tp->t_rcvtime >= tp->t_rxtcur)
  238                 cc_after_idle(tp);
  239         tp->t_flags &= ~TF_LASTIDLE;
  240         if (idle) {
  241                 if (tp->t_flags & TF_MORETOCOME) {
  242                         tp->t_flags |= TF_LASTIDLE;
  243                         idle = 0;
  244                 }
  245         }
  246 again:
  247         /*
  248          * If we've recently taken a timeout, snd_max will be greater than
  249          * snd_nxt.  There may be SACK information that allows us to avoid
  250          * resending already delivered data.  Adjust snd_nxt accordingly.
  251          */
  252         if ((tp->t_flags & TF_SACK_PERMIT) &&
  253             SEQ_LT(tp->snd_nxt, tp->snd_max))
  254                 tcp_sack_adjust(tp);
  255         sendalot = 0;
  256         tso = 0;
  257         mtu = 0;
  258         off = tp->snd_nxt - tp->snd_una;
  259         sendwin = min(tp->snd_wnd, tp->snd_cwnd);
  260 
  261         flags = tcp_outflags[tp->t_state];
  262         /*
  263          * Send any SACK-generated retransmissions.  If we're explicitly trying
  264          * to send out new data (when sendalot is 1), bypass this function.
  265          * If we retransmit in fast recovery mode, decrement snd_cwnd, since
  266          * we're replacing a (future) new transmission with a retransmission
  267          * now, and we previously incremented snd_cwnd in tcp_input().
  268          */
  269         /*
  270          * Still in sack recovery , reset rxmit flag to zero.
  271          */
  272         sack_rxmit = 0;
  273         sack_bytes_rxmt = 0;
  274         len = 0;
  275         p = NULL;
  276         if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) &&
  277             (p = tcp_sack_output(tp, &sack_bytes_rxmt))) {
  278                 long cwin;
  279                 
  280                 cwin = min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt;
  281                 if (cwin < 0)
  282                         cwin = 0;
  283                 /* Do not retransmit SACK segments beyond snd_recover */
  284                 if (SEQ_GT(p->end, tp->snd_recover)) {
  285                         /*
  286                          * (At least) part of sack hole extends beyond
  287                          * snd_recover. Check to see if we can rexmit data
  288                          * for this hole.
  289                          */
  290                         if (SEQ_GEQ(p->rxmit, tp->snd_recover)) {
  291                                 /*
  292                                  * Can't rexmit any more data for this hole.
  293                                  * That data will be rexmitted in the next
  294                                  * sack recovery episode, when snd_recover
  295                                  * moves past p->rxmit.
  296                                  */
  297                                 p = NULL;
  298                                 goto after_sack_rexmit;
  299                         } else
  300                                 /* Can rexmit part of the current hole */
  301                                 len = ((long)ulmin(cwin,
  302                                                    tp->snd_recover - p->rxmit));
  303                 } else
  304                         len = ((long)ulmin(cwin, p->end - p->rxmit));
  305                 off = p->rxmit - tp->snd_una;
  306                 KASSERT(off >= 0,("%s: sack block to the left of una : %d",
  307                     __func__, off));
  308                 if (len > 0) {
  309                         sack_rxmit = 1;
  310                         sendalot = 1;
  311                         TCPSTAT_INC(tcps_sack_rexmits);
  312                         TCPSTAT_ADD(tcps_sack_rexmit_bytes,
  313                             min(len, tp->t_maxseg));
  314                 }
  315         }
  316 after_sack_rexmit:
  317         /*
  318          * Get standard flags, and add SYN or FIN if requested by 'hidden'
  319          * state flags.
  320          */
  321         if (tp->t_flags & TF_NEEDFIN)
  322                 flags |= TH_FIN;
  323         if (tp->t_flags & TF_NEEDSYN)
  324                 flags |= TH_SYN;
  325 
  326         SOCKBUF_LOCK(&so->so_snd);
  327         /*
  328          * If in persist timeout with window of 0, send 1 byte.
  329          * Otherwise, if window is small but nonzero
  330          * and timer expired, we will send what we can
  331          * and go to transmit state.
  332          */
  333         if (tp->t_flags & TF_FORCEDATA) {
  334                 if (sendwin == 0) {
  335                         /*
  336                          * If we still have some data to send, then
  337                          * clear the FIN bit.  Usually this would
  338                          * happen below when it realizes that we
  339                          * aren't sending all the data.  However,
  340                          * if we have exactly 1 byte of unsent data,
  341                          * then it won't clear the FIN bit below,
  342                          * and if we are in persist state, we wind
  343                          * up sending the packet without recording
  344                          * that we sent the FIN bit.
  345                          *
  346                          * We can't just blindly clear the FIN bit,
  347                          * because if we don't have any more data
  348                          * to send then the probe will be the FIN
  349                          * itself.
  350                          */
  351                         if (off < sbused(&so->so_snd))
  352                                 flags &= ~TH_FIN;
  353                         sendwin = 1;
  354                 } else {
  355                         tcp_timer_activate(tp, TT_PERSIST, 0);
  356                         tp->t_rxtshift = 0;
  357                 }
  358         }
  359 
  360         /*
  361          * If snd_nxt == snd_max and we have transmitted a FIN, the
  362          * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
  363          * a negative length.  This can also occur when TCP opens up
  364          * its congestion window while receiving additional duplicate
  365          * acks after fast-retransmit because TCP will reset snd_nxt
  366          * to snd_max after the fast-retransmit.
  367          *
  368          * In the normal retransmit-FIN-only case, however, snd_nxt will
  369          * be set to snd_una, the offset will be 0, and the length may
  370          * wind up 0.
  371          *
  372          * If sack_rxmit is true we are retransmitting from the scoreboard
  373          * in which case len is already set.
  374          */
  375         if (sack_rxmit == 0) {
  376                 if (sack_bytes_rxmt == 0)
  377                         len = ((long)ulmin(sbavail(&so->so_snd), sendwin) -
  378                             off);
  379                 else {
  380                         long cwin;
  381 
  382                         /*
  383                          * We are inside of a SACK recovery episode and are
  384                          * sending new data, having retransmitted all the
  385                          * data possible in the scoreboard.
  386                          */
  387                         len = ((long)ulmin(sbavail(&so->so_snd), tp->snd_wnd) -
  388                             off);
  389                         /*
  390                          * Don't remove this (len > 0) check !
  391                          * We explicitly check for len > 0 here (although it 
  392                          * isn't really necessary), to work around a gcc 
  393                          * optimization issue - to force gcc to compute
  394                          * len above. Without this check, the computation
  395                          * of len is bungled by the optimizer.
  396                          */
  397                         if (len > 0) {
  398                                 cwin = tp->snd_cwnd - 
  399                                         (tp->snd_nxt - tp->sack_newdata) -
  400                                         sack_bytes_rxmt;
  401                                 if (cwin < 0)
  402                                         cwin = 0;
  403                                 len = lmin(len, cwin);
  404                         }
  405                 }
  406         }
  407 
  408         /*
  409          * Lop off SYN bit if it has already been sent.  However, if this
  410          * is SYN-SENT state and if segment contains data and if we don't
  411          * know that foreign host supports TAO, suppress sending segment.
  412          */
  413         if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
  414                 if (tp->t_state != TCPS_SYN_RECEIVED)
  415                         flags &= ~TH_SYN;
  416 #ifdef TCP_RFC7413
  417                 /*
  418                  * When sending additional segments following a TFO SYN|ACK,
  419                  * do not include the SYN bit.
  420                  */
  421                 if ((tp->t_flags & TF_FASTOPEN) &&
  422                     (tp->t_state == TCPS_SYN_RECEIVED))
  423                         flags &= ~TH_SYN;
  424 #endif
  425                 off--, len++;
  426         }
  427 
  428         /*
  429          * Be careful not to send data and/or FIN on SYN segments.
  430          * This measure is needed to prevent interoperability problems
  431          * with not fully conformant TCP implementations.
  432          */
  433         if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) {
  434                 len = 0;
  435                 flags &= ~TH_FIN;
  436         }
  437 
  438 #ifdef TCP_RFC7413
  439         /*
  440          * When retransmitting SYN|ACK on a passively-created TFO socket,
  441          * don't include data, as the presence of data may have caused the
  442          * original SYN|ACK to have been dropped by a middlebox.
  443          */
  444         if ((tp->t_flags & TF_FASTOPEN) &&
  445             (((tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_rxtshift > 0)) ||
  446              (flags & TH_RST)))
  447                 len = 0;
  448 #endif
  449         if (len <= 0) {
  450                 /*
  451                  * If FIN has been sent but not acked,
  452                  * but we haven't been called to retransmit,
  453                  * len will be < 0.  Otherwise, window shrank
  454                  * after we sent into it.  If window shrank to 0,
  455                  * cancel pending retransmit, pull snd_nxt back
  456                  * to (closed) window, and set the persist timer
  457                  * if it isn't already going.  If the window didn't
  458                  * close completely, just wait for an ACK.
  459                  *
  460                  * We also do a general check here to ensure that
  461                  * we will set the persist timer when we have data
  462                  * to send, but a 0-byte window. This makes sure
  463                  * the persist timer is set even if the packet
  464                  * hits one of the "goto send" lines below.
  465                  */
  466                 len = 0;
  467                 if ((sendwin == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) &&
  468                         (off < (int) sbavail(&so->so_snd))) {
  469                         tcp_timer_activate(tp, TT_REXMT, 0);
  470                         tp->t_rxtshift = 0;
  471                         tp->snd_nxt = tp->snd_una;
  472                         if (!tcp_timer_active(tp, TT_PERSIST))
  473                                 tcp_setpersist(tp);
  474                 }
  475         }
  476 
  477         /* len will be >= 0 after this point. */
  478         KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
  479 
  480         /*
  481          * Automatic sizing of send socket buffer.  Often the send buffer
  482          * size is not optimally adjusted to the actual network conditions
  483          * at hand (delay bandwidth product).  Setting the buffer size too
  484          * small limits throughput on links with high bandwidth and high
  485          * delay (eg. trans-continental/oceanic links).  Setting the
  486          * buffer size too big consumes too much real kernel memory,
  487          * especially with many connections on busy servers.
  488          *
  489          * The criteria to step up the send buffer one notch are:
  490          *  1. receive window of remote host is larger than send buffer
  491          *     (with a fudge factor of 5/4th);
  492          *  2. send buffer is filled to 7/8th with data (so we actually
  493          *     have data to make use of it);
  494          *  3. send buffer fill has not hit maximal automatic size;
  495          *  4. our send window (slow start and cogestion controlled) is
  496          *     larger than sent but unacknowledged data in send buffer.
  497          *
  498          * The remote host receive window scaling factor may limit the
  499          * growing of the send buffer before it reaches its allowed
  500          * maximum.
  501          *
  502          * It scales directly with slow start or congestion window
  503          * and does at most one step per received ACK.  This fast
  504          * scaling has the drawback of growing the send buffer beyond
  505          * what is strictly necessary to make full use of a given
  506          * delay*bandwidth product.  However testing has shown this not
  507          * to be much of an problem.  At worst we are trading wasting
  508          * of available bandwidth (the non-use of it) for wasting some
  509          * socket buffer memory.
  510          *
  511          * TODO: Shrink send buffer during idle periods together
  512          * with congestion window.  Requires another timer.  Has to
  513          * wait for upcoming tcp timer rewrite.
  514          *
  515          * XXXGL: should there be used sbused() or sbavail()?
  516          */
  517         if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) {
  518                 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat &&
  519                     sbused(&so->so_snd) >= (so->so_snd.sb_hiwat / 8 * 7) &&
  520                     sbused(&so->so_snd) < V_tcp_autosndbuf_max &&
  521                     sendwin >= (sbused(&so->so_snd) -
  522                     (tp->snd_nxt - tp->snd_una))) {
  523                         if (!sbreserve_locked(&so->so_snd,
  524                             min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc,
  525                              V_tcp_autosndbuf_max), so, curthread))
  526                                 so->so_snd.sb_flags &= ~SB_AUTOSIZE;
  527                 }
  528         }
  529 
  530         /*
  531          * Decide if we can use TCP Segmentation Offloading (if supported by
  532          * hardware).
  533          *
  534          * TSO may only be used if we are in a pure bulk sending state.  The
  535          * presence of TCP-MD5, SACK retransmits, SACK advertizements and
  536          * IP options prevent using TSO.  With TSO the TCP header is the same
  537          * (except for the sequence number) for all generated packets.  This
  538          * makes it impossible to transmit any options which vary per generated
  539          * segment or packet.
  540          *
  541          * IPv4 handling has a clear separation of ip options and ip header
  542          * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() does
  543          * the right thing below to provide length of just ip options and thus
  544          * checking for ipoptlen is enough to decide if ip options are present.
  545          */
  546 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
  547         /*
  548          * Pre-calculate here as we save another lookup into the darknesses
  549          * of IPsec that way and can actually decide if TSO is ok.
  550          */
  551 #ifdef INET6
  552         if (isipv6 && IPSEC_ENABLED(ipv6))
  553                 ipsec_optlen = IPSEC_HDRSIZE(ipv6, tp->t_inpcb);
  554 #ifdef INET
  555         else
  556 #endif
  557 #endif /* INET6 */
  558 #ifdef INET
  559         if (IPSEC_ENABLED(ipv4))
  560                 ipsec_optlen = IPSEC_HDRSIZE(ipv4, tp->t_inpcb);
  561 #endif /* INET */
  562 #endif /* IPSEC */
  563 #ifdef INET6
  564         if (isipv6)
  565                 ipoptlen = ip6_optlen(tp->t_inpcb);
  566         else
  567 #endif
  568         if (tp->t_inpcb->inp_options)
  569                 ipoptlen = tp->t_inpcb->inp_options->m_len -
  570                                 offsetof(struct ipoption, ipopt_list);
  571         else
  572                 ipoptlen = 0;
  573 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
  574         ipoptlen += ipsec_optlen;
  575 #endif
  576 
  577         if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg &&
  578             ((tp->t_flags & TF_SIGNATURE) == 0) &&
  579             tp->rcv_numsacks == 0 && sack_rxmit == 0 &&
  580             ipoptlen == 0)
  581                 tso = 1;
  582 
  583         if (sack_rxmit) {
  584                 if (SEQ_LT(p->rxmit + len, tp->snd_una + sbused(&so->so_snd)))
  585                         flags &= ~TH_FIN;
  586         } else {
  587                 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una +
  588                     sbused(&so->so_snd)))
  589                         flags &= ~TH_FIN;
  590         }
  591 
  592         recwin = sbspace(&so->so_rcv);
  593 
  594         /*
  595          * Sender silly window avoidance.   We transmit under the following
  596          * conditions when len is non-zero:
  597          *
  598          *      - We have a full segment (or more with TSO)
  599          *      - This is the last buffer in a write()/send() and we are
  600          *        either idle or running NODELAY
  601          *      - we've timed out (e.g. persist timer)
  602          *      - we have more then 1/2 the maximum send window's worth of
  603          *        data (receiver may be limited the window size)
  604          *      - we need to retransmit
  605          */
  606         if (len) {
  607                 if (len >= tp->t_maxseg)
  608                         goto send;
  609                 /*
  610                  * NOTE! on localhost connections an 'ack' from the remote
  611                  * end may occur synchronously with the output and cause
  612                  * us to flush a buffer queued with moretocome.  XXX
  613                  *
  614                  * note: the len + off check is almost certainly unnecessary.
  615                  */
  616                 if (!(tp->t_flags & TF_MORETOCOME) &&   /* normal case */
  617                     (idle || (tp->t_flags & TF_NODELAY)) &&
  618                     len + off >= sbavail(&so->so_snd) &&
  619                     (tp->t_flags & TF_NOPUSH) == 0) {
  620                         goto send;
  621                 }
  622                 if (tp->t_flags & TF_FORCEDATA)         /* typ. timeout case */
  623                         goto send;
  624                 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
  625                         goto send;
  626                 if (SEQ_LT(tp->snd_nxt, tp->snd_max))   /* retransmit case */
  627                         goto send;
  628                 if (sack_rxmit)
  629                         goto send;
  630         }
  631 
  632         /*
  633          * Sending of standalone window updates.
  634          *
  635          * Window updates are important when we close our window due to a
  636          * full socket buffer and are opening it again after the application
  637          * reads data from it.  Once the window has opened again and the
  638          * remote end starts to send again the ACK clock takes over and
  639          * provides the most current window information.
  640          *
  641          * We must avoid the silly window syndrome whereas every read
  642          * from the receive buffer, no matter how small, causes a window
  643          * update to be sent.  We also should avoid sending a flurry of
  644          * window updates when the socket buffer had queued a lot of data
  645          * and the application is doing small reads.
  646          *
  647          * Prevent a flurry of pointless window updates by only sending
  648          * an update when we can increase the advertized window by more
  649          * than 1/4th of the socket buffer capacity.  When the buffer is
  650          * getting full or is very small be more aggressive and send an
  651          * update whenever we can increase by two mss sized segments.
  652          * In all other situations the ACK's to new incoming data will
  653          * carry further window increases.
  654          *
  655          * Don't send an independent window update if a delayed
  656          * ACK is pending (it will get piggy-backed on it) or the
  657          * remote side already has done a half-close and won't send
  658          * more data.  Skip this if the connection is in T/TCP
  659          * half-open state.
  660          */
  661         if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) &&
  662             !(tp->t_flags & TF_DELACK) &&
  663             !TCPS_HAVERCVDFIN(tp->t_state)) {
  664                 /*
  665                  * "adv" is the amount we could increase the window,
  666                  * taking into account that we are limited by
  667                  * TCP_MAXWIN << tp->rcv_scale.
  668                  */
  669                 long adv;
  670                 int oldwin;
  671 
  672                 adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale);
  673                 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) {
  674                         oldwin = (tp->rcv_adv - tp->rcv_nxt);
  675                         adv -= oldwin;
  676                 } else
  677                         oldwin = 0;
  678 
  679                 /* 
  680                  * If the new window size ends up being the same as the old
  681                  * size when it is scaled, then don't force a window update.
  682                  */
  683                 if (oldwin >> tp->rcv_scale == (adv + oldwin) >> tp->rcv_scale)
  684                         goto dontupdate;
  685 
  686                 if (adv >= (long)(2 * tp->t_maxseg) &&
  687                     (adv >= (long)(so->so_rcv.sb_hiwat / 4) ||
  688                      recwin <= (long)(so->so_rcv.sb_hiwat / 8) ||
  689                      so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg))
  690                         goto send;
  691                 if (2 * adv >= (int32_t)so->so_rcv.sb_hiwat)
  692                         goto send;
  693         }
  694 dontupdate:
  695 
  696         /*
  697          * Send if we owe the peer an ACK, RST, SYN, or urgent data.  ACKNOW
  698          * is also a catch-all for the retransmit timer timeout case.
  699          */
  700         if (tp->t_flags & TF_ACKNOW)
  701                 goto send;
  702         if ((flags & TH_RST) ||
  703             ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
  704                 goto send;
  705         if (SEQ_GT(tp->snd_up, tp->snd_una))
  706                 goto send;
  707         /*
  708          * If our state indicates that FIN should be sent
  709          * and we have not yet done so, then we need to send.
  710          */
  711         if (flags & TH_FIN &&
  712             ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
  713                 goto send;
  714         /*
  715          * In SACK, it is possible for tcp_output to fail to send a segment
  716          * after the retransmission timer has been turned off.  Make sure
  717          * that the retransmission timer is set.
  718          */
  719         if ((tp->t_flags & TF_SACK_PERMIT) &&
  720             SEQ_GT(tp->snd_max, tp->snd_una) &&
  721             !tcp_timer_active(tp, TT_REXMT) &&
  722             !tcp_timer_active(tp, TT_PERSIST)) {
  723                 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
  724                 goto just_return;
  725         } 
  726         /*
  727          * TCP window updates are not reliable, rather a polling protocol
  728          * using ``persist'' packets is used to insure receipt of window
  729          * updates.  The three ``states'' for the output side are:
  730          *      idle                    not doing retransmits or persists
  731          *      persisting              to move a small or zero window
  732          *      (re)transmitting        and thereby not persisting
  733          *
  734          * tcp_timer_active(tp, TT_PERSIST)
  735          *      is true when we are in persist state.
  736          * (tp->t_flags & TF_FORCEDATA)
  737          *      is set when we are called to send a persist packet.
  738          * tcp_timer_active(tp, TT_REXMT)
  739          *      is set when we are retransmitting
  740          * The output side is idle when both timers are zero.
  741          *
  742          * If send window is too small, there is data to transmit, and no
  743          * retransmit or persist is pending, then go to persist state.
  744          * If nothing happens soon, send when timer expires:
  745          * if window is nonzero, transmit what we can,
  746          * otherwise force out a byte.
  747          */
  748         if (sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) &&
  749             !tcp_timer_active(tp, TT_PERSIST)) {
  750                 tp->t_rxtshift = 0;
  751                 tcp_setpersist(tp);
  752         }
  753 
  754         /*
  755          * No reason to send a segment, just return.
  756          */
  757 just_return:
  758         SOCKBUF_UNLOCK(&so->so_snd);
  759         return (0);
  760 
  761 send:
  762         SOCKBUF_LOCK_ASSERT(&so->so_snd);
  763         if (len > 0) {
  764                 if (len >= tp->t_maxseg)
  765                         tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT;
  766                 else
  767                         tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT;
  768         }
  769         /*
  770          * Before ESTABLISHED, force sending of initial options
  771          * unless TCP set not to do any options.
  772          * NOTE: we assume that the IP/TCP header plus TCP options
  773          * always fit in a single mbuf, leaving room for a maximum
  774          * link header, i.e.
  775          *      max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
  776          */
  777         optlen = 0;
  778 #ifdef INET6
  779         if (isipv6)
  780                 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
  781         else
  782 #endif
  783                 hdrlen = sizeof (struct tcpiphdr);
  784 
  785         /*
  786          * Compute options for segment.
  787          * We only have to care about SYN and established connection
  788          * segments.  Options for SYN-ACK segments are handled in TCP
  789          * syncache.
  790          */
  791         to.to_flags = 0;
  792         if ((tp->t_flags & TF_NOOPT) == 0) {
  793                 /* Maximum segment size. */
  794                 if (flags & TH_SYN) {
  795                         tp->snd_nxt = tp->iss;
  796                         to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc);
  797                         to.to_flags |= TOF_MSS;
  798 #ifdef TCP_RFC7413
  799                         /*
  800                          * Only include the TFO option on the first
  801                          * transmission of the SYN|ACK on a
  802                          * passively-created TFO socket, as the presence of
  803                          * the TFO option may have caused the original
  804                          * SYN|ACK to have been dropped by a middlebox.
  805                          */
  806                         if ((tp->t_flags & TF_FASTOPEN) &&
  807                             (tp->t_state == TCPS_SYN_RECEIVED) &&
  808                             (tp->t_rxtshift == 0)) {
  809                                 to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN;
  810                                 to.to_tfo_cookie = (u_char *)&tp->t_tfo_cookie;
  811                                 to.to_flags |= TOF_FASTOPEN;
  812                         }
  813 #endif
  814                 }
  815                 /* Window scaling. */
  816                 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) {
  817                         to.to_wscale = tp->request_r_scale;
  818                         to.to_flags |= TOF_SCALE;
  819                 }
  820                 /* Timestamps. */
  821                 if ((tp->t_flags & TF_RCVD_TSTMP) ||
  822                     ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) {
  823                         to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
  824                         to.to_tsecr = tp->ts_recent;
  825                         to.to_flags |= TOF_TS;
  826                 }
  827 
  828                 /* Set receive buffer autosizing timestamp. */
  829                 if (tp->rfbuf_ts == 0 &&
  830                     (so->so_rcv.sb_flags & SB_AUTOSIZE))
  831                         tp->rfbuf_ts = tcp_ts_getticks();
  832 
  833                 /* Selective ACK's. */
  834                 if (tp->t_flags & TF_SACK_PERMIT) {
  835                         if (flags & TH_SYN)
  836                                 to.to_flags |= TOF_SACKPERM;
  837                         else if (TCPS_HAVEESTABLISHED(tp->t_state) &&
  838                             (tp->t_flags & TF_SACK_PERMIT) &&
  839                             tp->rcv_numsacks > 0) {
  840                                 to.to_flags |= TOF_SACK;
  841                                 to.to_nsacks = tp->rcv_numsacks;
  842                                 to.to_sacks = (u_char *)tp->sackblks;
  843                         }
  844                 }
  845 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
  846                 /* TCP-MD5 (RFC2385). */
  847                 /*
  848                  * Check that TCP_MD5SIG is enabled in tcpcb to
  849                  * account the size needed to set this TCP option.
  850                  */
  851                 if (tp->t_flags & TF_SIGNATURE)
  852                         to.to_flags |= TOF_SIGNATURE;
  853 #endif /* TCP_SIGNATURE */
  854 
  855                 /* Processing the options. */
  856                 hdrlen += optlen = tcp_addoptions(&to, opt);
  857         }
  858 
  859         /*
  860          * Adjust data length if insertion of options will
  861          * bump the packet length beyond the t_maxseg length.
  862          * Clear the FIN bit because we cut off the tail of
  863          * the segment.
  864          */
  865         if (len + optlen + ipoptlen > tp->t_maxseg) {
  866                 flags &= ~TH_FIN;
  867 
  868                 if (tso) {
  869                         u_int if_hw_tsomax;
  870                         u_int if_hw_tsomaxsegcount;
  871                         u_int if_hw_tsomaxsegsize;
  872                         struct mbuf *mb;
  873                         u_int moff;
  874                         int max_len;
  875 
  876                         /* extract TSO information */
  877                         if_hw_tsomax = tp->t_tsomax;
  878                         if_hw_tsomaxsegcount = tp->t_tsomaxsegcount;
  879                         if_hw_tsomaxsegsize = tp->t_tsomaxsegsize;
  880 
  881                         /*
  882                          * Limit a TSO burst to prevent it from
  883                          * overflowing or exceeding the maximum length
  884                          * allowed by the network interface:
  885                          */
  886                         KASSERT(ipoptlen == 0,
  887                             ("%s: TSO can't do IP options", __func__));
  888 
  889                         /*
  890                          * Check if we should limit by maximum payload
  891                          * length:
  892                          */
  893                         if (if_hw_tsomax != 0) {
  894                                 /* compute maximum TSO length */
  895                                 max_len = (if_hw_tsomax - hdrlen -
  896                                     max_linkhdr);
  897                                 if (max_len <= 0) {
  898                                         len = 0;
  899                                 } else if (len > max_len) {
  900                                         sendalot = 1;
  901                                         len = max_len;
  902                                 }
  903                         }
  904 
  905                         /*
  906                          * Check if we should limit by maximum segment
  907                          * size and count:
  908                          */
  909                         if (if_hw_tsomaxsegcount != 0 &&
  910                             if_hw_tsomaxsegsize != 0) {
  911                                 /*
  912                                  * Subtract one segment for the LINK
  913                                  * and TCP/IP headers mbuf that will
  914                                  * be prepended to this mbuf chain
  915                                  * after the code in this section
  916                                  * limits the number of mbufs in the
  917                                  * chain to if_hw_tsomaxsegcount.
  918                                  */
  919                                 if_hw_tsomaxsegcount -= 1;
  920                                 max_len = 0;
  921                                 mb = sbsndmbuf(&so->so_snd, off, &moff);
  922 
  923                                 while (mb != NULL && max_len < len) {
  924                                         u_int mlen;
  925                                         u_int frags;
  926 
  927                                         /*
  928                                          * Get length of mbuf fragment
  929                                          * and how many hardware frags,
  930                                          * rounded up, it would use:
  931                                          */
  932                                         mlen = (mb->m_len - moff);
  933                                         frags = howmany(mlen,
  934                                             if_hw_tsomaxsegsize);
  935 
  936                                         /* Handle special case: Zero Length Mbuf */
  937                                         if (frags == 0)
  938                                                 frags = 1;
  939 
  940                                         /*
  941                                          * Check if the fragment limit
  942                                          * will be reached or exceeded:
  943                                          */
  944                                         if (frags >= if_hw_tsomaxsegcount) {
  945                                                 max_len += min(mlen,
  946                                                     if_hw_tsomaxsegcount *
  947                                                     if_hw_tsomaxsegsize);
  948                                                 break;
  949                                         }
  950                                         max_len += mlen;
  951                                         if_hw_tsomaxsegcount -= frags;
  952                                         moff = 0;
  953                                         mb = mb->m_next;
  954                                 }
  955                                 if (max_len <= 0) {
  956                                         len = 0;
  957                                 } else if (len > max_len) {
  958                                         sendalot = 1;
  959                                         len = max_len;
  960                                 }
  961                         }
  962 
  963                         /*
  964                          * Prevent the last segment from being
  965                          * fractional unless the send sockbuf can be
  966                          * emptied:
  967                          */
  968                         max_len = (tp->t_maxseg - optlen);
  969                         if ((off + len) < sbavail(&so->so_snd)) {
  970                                 moff = len % max_len;
  971                                 if (moff != 0) {
  972                                         len -= moff;
  973                                         sendalot = 1;
  974                                 }
  975                         }
  976 
  977                         /*
  978                          * In case there are too many small fragments
  979                          * don't use TSO:
  980                          */
  981                         if (len <= max_len) {
  982                                 len = max_len;
  983                                 sendalot = 1;
  984                                 tso = 0;
  985                         }
  986 
  987                         /*
  988                          * Send the FIN in a separate segment
  989                          * after the bulk sending is done.
  990                          * We don't trust the TSO implementations
  991                          * to clear the FIN flag on all but the
  992                          * last segment.
  993                          */
  994                         if (tp->t_flags & TF_NEEDFIN)
  995                                 sendalot = 1;
  996 
  997                 } else {
  998                         len = tp->t_maxseg - optlen - ipoptlen;
  999                         sendalot = 1;
 1000                 }
 1001         } else
 1002                 tso = 0;
 1003 
 1004         KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET,
 1005             ("%s: len > IP_MAXPACKET", __func__));
 1006 
 1007 /*#ifdef DIAGNOSTIC*/
 1008 #ifdef INET6
 1009         if (max_linkhdr + hdrlen > MCLBYTES)
 1010 #else
 1011         if (max_linkhdr + hdrlen > MHLEN)
 1012 #endif
 1013                 panic("tcphdr too big");
 1014 /*#endif*/
 1015 
 1016         /*
 1017          * This KASSERT is here to catch edge cases at a well defined place.
 1018          * Before, those had triggered (random) panic conditions further down.
 1019          */
 1020         KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
 1021 
 1022         /*
 1023          * Grab a header mbuf, attaching a copy of data to
 1024          * be transmitted, and initialize the header from
 1025          * the template for sends on this connection.
 1026          */
 1027         if (len) {
 1028                 struct mbuf *mb;
 1029                 u_int moff;
 1030 
 1031                 if ((tp->t_flags & TF_FORCEDATA) && len == 1)
 1032                         TCPSTAT_INC(tcps_sndprobe);
 1033                 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) {
 1034                         tp->t_sndrexmitpack++;
 1035                         TCPSTAT_INC(tcps_sndrexmitpack);
 1036                         TCPSTAT_ADD(tcps_sndrexmitbyte, len);
 1037                 } else {
 1038                         TCPSTAT_INC(tcps_sndpack);
 1039                         TCPSTAT_ADD(tcps_sndbyte, len);
 1040                 }
 1041 #ifdef INET6
 1042                 if (MHLEN < hdrlen + max_linkhdr)
 1043                         m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 1044                 else
 1045 #endif
 1046                         m = m_gethdr(M_NOWAIT, MT_DATA);
 1047 
 1048                 if (m == NULL) {
 1049                         SOCKBUF_UNLOCK(&so->so_snd);
 1050                         error = ENOBUFS;
 1051                         sack_rxmit = 0;
 1052                         goto out;
 1053                 }
 1054 
 1055                 m->m_data += max_linkhdr;
 1056                 m->m_len = hdrlen;
 1057 
 1058                 /*
 1059                  * Start the m_copy functions from the closest mbuf
 1060                  * to the offset in the socket buffer chain.
 1061                  */
 1062                 mb = sbsndptr(&so->so_snd, off, len, &moff);
 1063 
 1064                 if (len <= MHLEN - hdrlen - max_linkhdr) {
 1065                         m_copydata(mb, moff, (int)len,
 1066                             mtod(m, caddr_t) + hdrlen);
 1067                         m->m_len += len;
 1068                 } else {
 1069                         m->m_next = m_copy(mb, moff, (int)len);
 1070                         if (m->m_next == NULL) {
 1071                                 SOCKBUF_UNLOCK(&so->so_snd);
 1072                                 (void) m_free(m);
 1073                                 error = ENOBUFS;
 1074                                 sack_rxmit = 0;
 1075                                 goto out;
 1076                         }
 1077                 }
 1078 
 1079                 /*
 1080                  * If we're sending everything we've got, set PUSH.
 1081                  * (This will keep happy those implementations which only
 1082                  * give data to the user when a buffer fills or
 1083                  * a PUSH comes in.)
 1084                  */
 1085                 if ((off + len == sbused(&so->so_snd)) && !(flags & TH_SYN))
 1086                         flags |= TH_PUSH;
 1087                 SOCKBUF_UNLOCK(&so->so_snd);
 1088         } else {
 1089                 SOCKBUF_UNLOCK(&so->so_snd);
 1090                 if (tp->t_flags & TF_ACKNOW)
 1091                         TCPSTAT_INC(tcps_sndacks);
 1092                 else if (flags & (TH_SYN|TH_FIN|TH_RST))
 1093                         TCPSTAT_INC(tcps_sndctrl);
 1094                 else if (SEQ_GT(tp->snd_up, tp->snd_una))
 1095                         TCPSTAT_INC(tcps_sndurg);
 1096                 else
 1097                         TCPSTAT_INC(tcps_sndwinup);
 1098 
 1099                 m = m_gethdr(M_NOWAIT, MT_DATA);
 1100                 if (m == NULL) {
 1101                         error = ENOBUFS;
 1102                         sack_rxmit = 0;
 1103                         goto out;
 1104                 }
 1105 #ifdef INET6
 1106                 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
 1107                     MHLEN >= hdrlen) {
 1108                         M_ALIGN(m, hdrlen);
 1109                 } else
 1110 #endif
 1111                 m->m_data += max_linkhdr;
 1112                 m->m_len = hdrlen;
 1113         }
 1114         SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
 1115         m->m_pkthdr.rcvif = (struct ifnet *)0;
 1116 #ifdef MAC
 1117         mac_inpcb_create_mbuf(tp->t_inpcb, m);
 1118 #endif
 1119 #ifdef INET6
 1120         if (isipv6) {
 1121                 ip6 = mtod(m, struct ip6_hdr *);
 1122                 th = (struct tcphdr *)(ip6 + 1);
 1123                 tcpip_fillheaders(tp->t_inpcb, ip6, th);
 1124         } else
 1125 #endif /* INET6 */
 1126         {
 1127                 ip = mtod(m, struct ip *);
 1128                 ipov = (struct ipovly *)ip;
 1129                 th = (struct tcphdr *)(ip + 1);
 1130                 tcpip_fillheaders(tp->t_inpcb, ip, th);
 1131         }
 1132 
 1133         /*
 1134          * Fill in fields, remembering maximum advertised
 1135          * window for use in delaying messages about window sizes.
 1136          * If resending a FIN, be sure not to use a new sequence number.
 1137          */
 1138         if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
 1139             tp->snd_nxt == tp->snd_max)
 1140                 tp->snd_nxt--;
 1141         /*
 1142          * If we are starting a connection, send ECN setup
 1143          * SYN packet. If we are on a retransmit, we may
 1144          * resend those bits a number of times as per
 1145          * RFC 3168.
 1146          */
 1147         if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn == 1) {
 1148                 if (tp->t_rxtshift >= 1) {
 1149                         if (tp->t_rxtshift <= V_tcp_ecn_maxretries)
 1150                                 flags |= TH_ECE|TH_CWR;
 1151                 } else
 1152                         flags |= TH_ECE|TH_CWR;
 1153         }
 1154         
 1155         if (tp->t_state == TCPS_ESTABLISHED &&
 1156             (tp->t_flags & TF_ECN_PERMIT)) {
 1157                 /*
 1158                  * If the peer has ECN, mark data packets with
 1159                  * ECN capable transmission (ECT).
 1160                  * Ignore pure ack packets, retransmissions and window probes.
 1161                  */
 1162                 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) &&
 1163                     !((tp->t_flags & TF_FORCEDATA) && len == 1)) {
 1164 #ifdef INET6
 1165                         if (isipv6)
 1166                                 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
 1167                         else
 1168 #endif
 1169                                 ip->ip_tos |= IPTOS_ECN_ECT0;
 1170                         TCPSTAT_INC(tcps_ecn_ect0);
 1171                 }
 1172                 
 1173                 /*
 1174                  * Reply with proper ECN notifications.
 1175                  */
 1176                 if (tp->t_flags & TF_ECN_SND_CWR) {
 1177                         flags |= TH_CWR;
 1178                         tp->t_flags &= ~TF_ECN_SND_CWR;
 1179                 } 
 1180                 if (tp->t_flags & TF_ECN_SND_ECE)
 1181                         flags |= TH_ECE;
 1182         }
 1183         
 1184         /*
 1185          * If we are doing retransmissions, then snd_nxt will
 1186          * not reflect the first unsent octet.  For ACK only
 1187          * packets, we do not want the sequence number of the
 1188          * retransmitted packet, we want the sequence number
 1189          * of the next unsent octet.  So, if there is no data
 1190          * (and no SYN or FIN), use snd_max instead of snd_nxt
 1191          * when filling in ti_seq.  But if we are in persist
 1192          * state, snd_max might reflect one byte beyond the
 1193          * right edge of the window, so use snd_nxt in that
 1194          * case, since we know we aren't doing a retransmission.
 1195          * (retransmit and persist are mutually exclusive...)
 1196          */
 1197         if (sack_rxmit == 0) {
 1198                 if (len || (flags & (TH_SYN|TH_FIN)) ||
 1199                     tcp_timer_active(tp, TT_PERSIST))
 1200                         th->th_seq = htonl(tp->snd_nxt);
 1201                 else
 1202                         th->th_seq = htonl(tp->snd_max);
 1203         } else {
 1204                 th->th_seq = htonl(p->rxmit);
 1205                 p->rxmit += len;
 1206                 tp->sackhint.sack_bytes_rexmit += len;
 1207         }
 1208         th->th_ack = htonl(tp->rcv_nxt);
 1209         if (optlen) {
 1210                 bcopy(opt, th + 1, optlen);
 1211                 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
 1212         }
 1213         th->th_flags = flags;
 1214         /*
 1215          * Calculate receive window.  Don't shrink window,
 1216          * but avoid silly window syndrome.
 1217          */
 1218         if (recwin < (long)(so->so_rcv.sb_hiwat / 4) &&
 1219             recwin < (long)tp->t_maxseg)
 1220                 recwin = 0;
 1221         if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
 1222             recwin < (long)(tp->rcv_adv - tp->rcv_nxt))
 1223                 recwin = (long)(tp->rcv_adv - tp->rcv_nxt);
 1224         if (recwin > (long)TCP_MAXWIN << tp->rcv_scale)
 1225                 recwin = (long)TCP_MAXWIN << tp->rcv_scale;
 1226 
 1227         /*
 1228          * According to RFC1323 the window field in a SYN (i.e., a <SYN>
 1229          * or <SYN,ACK>) segment itself is never scaled.  The <SYN,ACK>
 1230          * case is handled in syncache.
 1231          */
 1232         if (flags & TH_SYN)
 1233                 th->th_win = htons((u_short)
 1234                                 (min(sbspace(&so->so_rcv), TCP_MAXWIN)));
 1235         else
 1236                 th->th_win = htons((u_short)(recwin >> tp->rcv_scale));
 1237 
 1238         /*
 1239          * Adjust the RXWIN0SENT flag - indicate that we have advertised
 1240          * a 0 window.  This may cause the remote transmitter to stall.  This
 1241          * flag tells soreceive() to disable delayed acknowledgements when
 1242          * draining the buffer.  This can occur if the receiver is attempting
 1243          * to read more data than can be buffered prior to transmitting on
 1244          * the connection.
 1245          */
 1246         if (th->th_win == 0) {
 1247                 tp->t_sndzerowin++;
 1248                 tp->t_flags |= TF_RXWIN0SENT;
 1249         } else
 1250                 tp->t_flags &= ~TF_RXWIN0SENT;
 1251         if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
 1252                 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
 1253                 th->th_flags |= TH_URG;
 1254         } else
 1255                 /*
 1256                  * If no urgent pointer to send, then we pull
 1257                  * the urgent pointer to the left edge of the send window
 1258                  * so that it doesn't drift into the send window on sequence
 1259                  * number wraparound.
 1260                  */
 1261                 tp->snd_up = tp->snd_una;               /* drag it along */
 1262 
 1263         /*
 1264          * Put TCP length in extended header, and then
 1265          * checksum extended header and data.
 1266          */
 1267         m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
 1268         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
 1269 
 1270 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
 1271         if (to.to_flags & TOF_SIGNATURE) {
 1272                 /*
 1273                  * Calculate MD5 signature and put it into the place
 1274                  * determined before.
 1275                  * NOTE: since TCP options buffer doesn't point into
 1276                  * mbuf's data, calculate offset and use it.
 1277                  */
 1278                 if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, th,
 1279                     (u_char *)(th + 1) + (to.to_signature - opt)) != 0) {
 1280                         /*
 1281                          * Do not send segment if the calculation of MD5
 1282                          * digest has failed.
 1283                          */
 1284                         goto out;
 1285                 }
 1286         }
 1287 #endif
 1288 #ifdef INET6
 1289         if (isipv6) {
 1290                 /*
 1291                  * There is no need to fill in ip6_plen right now.
 1292                  * It will be filled later by ip6_output.
 1293                  */
 1294                 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
 1295                 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) +
 1296                     optlen + len, IPPROTO_TCP, 0);
 1297         }
 1298 #endif
 1299 #if defined(INET6) && defined(INET)
 1300         else
 1301 #endif
 1302 #ifdef INET
 1303         {
 1304                 m->m_pkthdr.csum_flags = CSUM_TCP;
 1305                 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 1306                     htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen));
 1307 
 1308                 /* IP version must be set here for ipv4/ipv6 checking later */
 1309                 KASSERT(ip->ip_v == IPVERSION,
 1310                     ("%s: IP version incorrect: %d", __func__, ip->ip_v));
 1311         }
 1312 #endif
 1313 
 1314         /*
 1315          * Enable TSO and specify the size of the segments.
 1316          * The TCP pseudo header checksum is always provided.
 1317          */
 1318         if (tso) {
 1319                 KASSERT(len > tp->t_maxseg - optlen,
 1320                     ("%s: len <= tso_segsz", __func__));
 1321                 m->m_pkthdr.csum_flags |= CSUM_TSO;
 1322                 m->m_pkthdr.tso_segsz = tp->t_maxseg - optlen;
 1323         }
 1324 
 1325 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 1326         KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL),
 1327             ("%s: mbuf chain shorter than expected: %ld + %u + %u - %u != %u",
 1328             __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL)));
 1329 #else
 1330         KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL),
 1331             ("%s: mbuf chain shorter than expected: %ld + %u + %u != %u",
 1332             __func__, len, hdrlen, ipoptlen, m_length(m, NULL)));
 1333 #endif
 1334 
 1335         /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */
 1336         hhook_run_tcp_est_out(tp, th, &to, len, tso);
 1337 
 1338 #ifdef TCPDEBUG
 1339         /*
 1340          * Trace.
 1341          */
 1342         if (so->so_options & SO_DEBUG) {
 1343                 u_short save = 0;
 1344 #ifdef INET6
 1345                 if (!isipv6)
 1346 #endif
 1347                 {
 1348                         save = ipov->ih_len;
 1349                         ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */);
 1350                 }
 1351                 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
 1352 #ifdef INET6
 1353                 if (!isipv6)
 1354 #endif
 1355                 ipov->ih_len = save;
 1356         }
 1357 #endif /* TCPDEBUG */
 1358         TCP_PROBE3(debug__output, tp, th, m);
 1359 
 1360         /*
 1361          * Fill in IP length and desired time to live and
 1362          * send to IP level.  There should be a better way
 1363          * to handle ttl and tos; we could keep them in
 1364          * the template, but need a way to checksum without them.
 1365          */
 1366         /*
 1367          * m->m_pkthdr.len should have been set before checksum calculation,
 1368          * because in6_cksum() need it.
 1369          */
 1370 #ifdef INET6
 1371         if (isipv6) {
 1372                 /*
 1373                  * we separately set hoplimit for every segment, since the
 1374                  * user might want to change the value via setsockopt.
 1375                  * Also, desired default hop limit might be changed via
 1376                  * Neighbor Discovery.
 1377                  */
 1378                 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL);
 1379 
 1380                 /*
 1381                  * Set the packet size here for the benefit of DTrace probes.
 1382                  * ip6_output() will set it properly; it's supposed to include
 1383                  * the option header lengths as well.
 1384                  */
 1385                 ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
 1386 
 1387                 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss)
 1388                         tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 1389                 else
 1390                         tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 1391 
 1392                 if (tp->t_state == TCPS_SYN_SENT)
 1393                         TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th);
 1394 
 1395                 TCP_PROBE5(send, NULL, tp, ip6, tp, th);
 1396 
 1397 #ifdef TCPPCAP
 1398                 /* Save packet, if requested. */
 1399                 tcp_pcap_add(th, m, &(tp->t_outpkts));
 1400 #endif
 1401 
 1402                 /* TODO: IPv6 IP6TOS_ECT bit on */
 1403                 error = ip6_output(m, tp->t_inpcb->in6p_outputopts,
 1404                     &tp->t_inpcb->inp_route6,
 1405                     ((so->so_options & SO_DONTROUTE) ?  IP_ROUTETOIF : 0),
 1406                     NULL, NULL, tp->t_inpcb);
 1407 
 1408                 if (error == EMSGSIZE && tp->t_inpcb->inp_route6.ro_rt != NULL)
 1409                         mtu = tp->t_inpcb->inp_route6.ro_rt->rt_mtu;
 1410         }
 1411 #endif /* INET6 */
 1412 #if defined(INET) && defined(INET6)
 1413         else
 1414 #endif
 1415 #ifdef INET
 1416     {
 1417         ip->ip_len = htons(m->m_pkthdr.len);
 1418 #ifdef INET6
 1419         if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO)
 1420                 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL);
 1421 #endif /* INET6 */
 1422         /*
 1423          * If we do path MTU discovery, then we set DF on every packet.
 1424          * This might not be the best thing to do according to RFC3390
 1425          * Section 2. However the tcp hostcache migitates the problem
 1426          * so it affects only the first tcp connection with a host.
 1427          *
 1428          * NB: Don't set DF on small MTU/MSS to have a safe fallback.
 1429          */
 1430         if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) {
 1431                 ip->ip_off |= htons(IP_DF);
 1432                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 1433         } else {
 1434                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 1435         }
 1436 
 1437         if (tp->t_state == TCPS_SYN_SENT)
 1438                 TCP_PROBE5(connect__request, NULL, tp, ip, tp, th);
 1439 
 1440         TCP_PROBE5(send, NULL, tp, ip, tp, th);
 1441 
 1442 #ifdef TCPPCAP
 1443         /* Save packet, if requested. */
 1444         tcp_pcap_add(th, m, &(tp->t_outpkts));
 1445 #endif
 1446 
 1447         error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
 1448             ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0,
 1449             tp->t_inpcb);
 1450 
 1451         if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_rt != NULL)
 1452                 mtu = tp->t_inpcb->inp_route.ro_rt->rt_mtu;
 1453     }
 1454 #endif /* INET */
 1455 
 1456 out:
 1457         /*
 1458          * In transmit state, time the transmission and arrange for
 1459          * the retransmit.  In persist state, just set snd_max.
 1460          */
 1461         if ((tp->t_flags & TF_FORCEDATA) == 0 || 
 1462             !tcp_timer_active(tp, TT_PERSIST)) {
 1463                 tcp_seq startseq = tp->snd_nxt;
 1464 
 1465                 /*
 1466                  * Advance snd_nxt over sequence space of this segment.
 1467                  */
 1468                 if (flags & (TH_SYN|TH_FIN)) {
 1469                         if (flags & TH_SYN)
 1470                                 tp->snd_nxt++;
 1471                         if (flags & TH_FIN) {
 1472                                 tp->snd_nxt++;
 1473                                 tp->t_flags |= TF_SENTFIN;
 1474                         }
 1475                 }
 1476                 if (sack_rxmit)
 1477                         goto timer;
 1478                 tp->snd_nxt += len;
 1479                 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
 1480                         tp->snd_max = tp->snd_nxt;
 1481                         /*
 1482                          * Time this transmission if not a retransmission and
 1483                          * not currently timing anything.
 1484                          */
 1485                         if (tp->t_rtttime == 0) {
 1486                                 tp->t_rtttime = ticks;
 1487                                 tp->t_rtseq = startseq;
 1488                                 TCPSTAT_INC(tcps_segstimed);
 1489                         }
 1490                 }
 1491 
 1492                 /*
 1493                  * Set retransmit timer if not currently set,
 1494                  * and not doing a pure ack or a keep-alive probe.
 1495                  * Initial value for retransmit timer is smoothed
 1496                  * round-trip time + 2 * round-trip time variance.
 1497                  * Initialize shift counter which is used for backoff
 1498                  * of retransmit time.
 1499                  */
 1500 timer:
 1501                 if (!tcp_timer_active(tp, TT_REXMT) &&
 1502                     ((sack_rxmit && tp->snd_nxt != tp->snd_max) ||
 1503                      (tp->snd_nxt != tp->snd_una))) {
 1504                         if (tcp_timer_active(tp, TT_PERSIST)) {
 1505                                 tcp_timer_activate(tp, TT_PERSIST, 0);
 1506                                 tp->t_rxtshift = 0;
 1507                         }
 1508                         tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
 1509                 } else if (len == 0 && sbavail(&so->so_snd) &&
 1510                     !tcp_timer_active(tp, TT_REXMT) &&
 1511                     !tcp_timer_active(tp, TT_PERSIST)) {
 1512                         /*
 1513                          * Avoid a situation where we do not set persist timer
 1514                          * after a zero window condition. For example:
 1515                          * 1) A -> B: packet with enough data to fill the window
 1516                          * 2) B -> A: ACK for #1 + new data (0 window
 1517                          *    advertisement)
 1518                          * 3) A -> B: ACK for #2, 0 len packet
 1519                          *
 1520                          * In this case, A will not activate the persist timer,
 1521                          * because it chose to send a packet. Unless tcp_output
 1522                          * is called for some other reason (delayed ack timer,
 1523                          * another input packet from B, socket syscall), A will
 1524                          * not send zero window probes.
 1525                          *
 1526                          * So, if you send a 0-length packet, but there is data
 1527                          * in the socket buffer, and neither the rexmt or
 1528                          * persist timer is already set, then activate the
 1529                          * persist timer.
 1530                          */
 1531                         tp->t_rxtshift = 0;
 1532                         tcp_setpersist(tp);
 1533                 }
 1534         } else {
 1535                 /*
 1536                  * Persist case, update snd_max but since we are in
 1537                  * persist mode (no window) we do not update snd_nxt.
 1538                  */
 1539                 int xlen = len;
 1540                 if (flags & TH_SYN)
 1541                         ++xlen;
 1542                 if (flags & TH_FIN) {
 1543                         ++xlen;
 1544                         tp->t_flags |= TF_SENTFIN;
 1545                 }
 1546                 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
 1547                         tp->snd_max = tp->snd_nxt + len;
 1548         }
 1549 
 1550         if (error) {
 1551 
 1552                 /*
 1553                  * We know that the packet was lost, so back out the
 1554                  * sequence number advance, if any.
 1555                  *
 1556                  * If the error is EPERM the packet got blocked by the
 1557                  * local firewall.  Normally we should terminate the
 1558                  * connection but the blocking may have been spurious
 1559                  * due to a firewall reconfiguration cycle.  So we treat
 1560                  * it like a packet loss and let the retransmit timer and
 1561                  * timeouts do their work over time.
 1562                  * XXX: It is a POLA question whether calling tcp_drop right
 1563                  * away would be the really correct behavior instead.
 1564                  */
 1565                 if (((tp->t_flags & TF_FORCEDATA) == 0 ||
 1566                     !tcp_timer_active(tp, TT_PERSIST)) &&
 1567                     ((flags & TH_SYN) == 0) &&
 1568                     (error != EPERM)) {
 1569                         if (sack_rxmit) {
 1570                                 p->rxmit -= len;
 1571                                 tp->sackhint.sack_bytes_rexmit -= len;
 1572                                 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
 1573                                     ("sackhint bytes rtx >= 0"));
 1574                         } else
 1575                                 tp->snd_nxt -= len;
 1576                 }
 1577                 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);     /* Check gotos. */
 1578                 switch (error) {
 1579                 case EACCES:
 1580                         tp->t_softerror = error;
 1581                         return (0);
 1582                 case EPERM:
 1583                         tp->t_softerror = error;
 1584                         return (error);
 1585                 case ENOBUFS:
 1586                         TCP_XMIT_TIMER_ASSERT(tp, len, flags);
 1587                         tp->snd_cwnd = tp->t_maxseg;
 1588                         return (0);
 1589                 case EMSGSIZE:
 1590                         /*
 1591                          * For some reason the interface we used initially
 1592                          * to send segments changed to another or lowered
 1593                          * its MTU.
 1594                          * If TSO was active we either got an interface
 1595                          * without TSO capabilits or TSO was turned off.
 1596                          * If we obtained mtu from ip_output() then update
 1597                          * it and try again.
 1598                          */
 1599                         if (tso)
 1600                                 tp->t_flags &= ~TF_TSO;
 1601                         if (mtu != 0) {
 1602                                 tcp_mss_update(tp, -1, mtu, NULL, NULL);
 1603                                 goto again;
 1604                         }
 1605                         return (error);
 1606                 case EHOSTDOWN:
 1607                 case EHOSTUNREACH:
 1608                 case ENETDOWN:
 1609                 case ENETUNREACH:
 1610                         if (TCPS_HAVERCVDSYN(tp->t_state)) {
 1611                                 tp->t_softerror = error;
 1612                                 return (0);
 1613                         }
 1614                         /* FALLTHROUGH */
 1615                 default:
 1616                         return (error);
 1617                 }
 1618         }
 1619         TCPSTAT_INC(tcps_sndtotal);
 1620 
 1621         /*
 1622          * Data sent (as far as we can tell).
 1623          * If this advertises a larger window than any other segment,
 1624          * then remember the size of the advertised window.
 1625          * Any pending ACK has now been sent.
 1626          */
 1627         if (recwin >= 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv))
 1628                 tp->rcv_adv = tp->rcv_nxt + recwin;
 1629         tp->last_ack_sent = tp->rcv_nxt;
 1630         tp->t_flags &= ~(TF_ACKNOW | TF_DELACK);
 1631         if (tcp_timer_active(tp, TT_DELACK))
 1632                 tcp_timer_activate(tp, TT_DELACK, 0);
 1633 #if 0
 1634         /*
 1635          * This completely breaks TCP if newreno is turned on.  What happens
 1636          * is that if delayed-acks are turned on on the receiver, this code
 1637          * on the transmitter effectively destroys the TCP window, forcing
 1638          * it to four packets (1.5Kx4 = 6K window).
 1639          */
 1640         if (sendalot && --maxburst)
 1641                 goto again;
 1642 #endif
 1643         if (sendalot)
 1644                 goto again;
 1645         return (0);
 1646 }
 1647 
 1648 void
 1649 tcp_setpersist(struct tcpcb *tp)
 1650 {
 1651         int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
 1652         int tt;
 1653 
 1654         tp->t_flags &= ~TF_PREVVALID;
 1655         if (tcp_timer_active(tp, TT_REXMT))
 1656                 panic("tcp_setpersist: retransmit pending");
 1657         /*
 1658          * Start/restart persistence timer.
 1659          */
 1660         TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
 1661                       tcp_persmin, tcp_persmax);
 1662         tcp_timer_activate(tp, TT_PERSIST, tt);
 1663         if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
 1664                 tp->t_rxtshift++;
 1665 }
 1666 
 1667 /*
 1668  * Insert TCP options according to the supplied parameters to the place
 1669  * optp in a consistent way.  Can handle unaligned destinations.
 1670  *
 1671  * The order of the option processing is crucial for optimal packing and
 1672  * alignment for the scarce option space.
 1673  *
 1674  * The optimal order for a SYN/SYN-ACK segment is:
 1675  *   MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) +
 1676  *   Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40.
 1677  *
 1678  * The SACK options should be last.  SACK blocks consume 8*n+2 bytes.
 1679  * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks).
 1680  * At minimum we need 10 bytes (to generate 1 SACK block).  If both
 1681  * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present,
 1682  * we only have 10 bytes for SACK options (40 - (12 + 18)).
 1683  */
 1684 int
 1685 tcp_addoptions(struct tcpopt *to, u_char *optp)
 1686 {
 1687         u_int32_t mask, optlen = 0;
 1688 
 1689         for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) {
 1690                 if ((to->to_flags & mask) != mask)
 1691                         continue;
 1692                 if (optlen == TCP_MAXOLEN)
 1693                         break;
 1694                 switch (to->to_flags & mask) {
 1695                 case TOF_MSS:
 1696                         while (optlen % 4) {
 1697                                 optlen += TCPOLEN_NOP;
 1698                                 *optp++ = TCPOPT_NOP;
 1699                         }
 1700                         if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG)
 1701                                 continue;
 1702                         optlen += TCPOLEN_MAXSEG;
 1703                         *optp++ = TCPOPT_MAXSEG;
 1704                         *optp++ = TCPOLEN_MAXSEG;
 1705                         to->to_mss = htons(to->to_mss);
 1706                         bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss));
 1707                         optp += sizeof(to->to_mss);
 1708                         break;
 1709                 case TOF_SCALE:
 1710                         while (!optlen || optlen % 2 != 1) {
 1711                                 optlen += TCPOLEN_NOP;
 1712                                 *optp++ = TCPOPT_NOP;
 1713                         }
 1714                         if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW)
 1715                                 continue;
 1716                         optlen += TCPOLEN_WINDOW;
 1717                         *optp++ = TCPOPT_WINDOW;
 1718                         *optp++ = TCPOLEN_WINDOW;
 1719                         *optp++ = to->to_wscale;
 1720                         break;
 1721                 case TOF_SACKPERM:
 1722                         while (optlen % 2) {
 1723                                 optlen += TCPOLEN_NOP;
 1724                                 *optp++ = TCPOPT_NOP;
 1725                         }
 1726                         if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED)
 1727                                 continue;
 1728                         optlen += TCPOLEN_SACK_PERMITTED;
 1729                         *optp++ = TCPOPT_SACK_PERMITTED;
 1730                         *optp++ = TCPOLEN_SACK_PERMITTED;
 1731                         break;
 1732                 case TOF_TS:
 1733                         while (!optlen || optlen % 4 != 2) {
 1734                                 optlen += TCPOLEN_NOP;
 1735                                 *optp++ = TCPOPT_NOP;
 1736                         }
 1737                         if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP)
 1738                                 continue;
 1739                         optlen += TCPOLEN_TIMESTAMP;
 1740                         *optp++ = TCPOPT_TIMESTAMP;
 1741                         *optp++ = TCPOLEN_TIMESTAMP;
 1742                         to->to_tsval = htonl(to->to_tsval);
 1743                         to->to_tsecr = htonl(to->to_tsecr);
 1744                         bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval));
 1745                         optp += sizeof(to->to_tsval);
 1746                         bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr));
 1747                         optp += sizeof(to->to_tsecr);
 1748                         break;
 1749                 case TOF_SIGNATURE:
 1750                         {
 1751                         int siglen = TCPOLEN_SIGNATURE - 2;
 1752 
 1753                         while (!optlen || optlen % 4 != 2) {
 1754                                 optlen += TCPOLEN_NOP;
 1755                                 *optp++ = TCPOPT_NOP;
 1756                         }
 1757                         if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE) {
 1758                                 to->to_flags &= ~TOF_SIGNATURE;
 1759                                 continue;
 1760                         }
 1761                         optlen += TCPOLEN_SIGNATURE;
 1762                         *optp++ = TCPOPT_SIGNATURE;
 1763                         *optp++ = TCPOLEN_SIGNATURE;
 1764                         to->to_signature = optp;
 1765                         while (siglen--)
 1766                                  *optp++ = 0;
 1767                         break;
 1768                         }
 1769                 case TOF_SACK:
 1770                         {
 1771                         int sackblks = 0;
 1772                         struct sackblk *sack = (struct sackblk *)to->to_sacks;
 1773                         tcp_seq sack_seq;
 1774 
 1775                         while (!optlen || optlen % 4 != 2) {
 1776                                 optlen += TCPOLEN_NOP;
 1777                                 *optp++ = TCPOPT_NOP;
 1778                         }
 1779                         if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK)
 1780                                 continue;
 1781                         optlen += TCPOLEN_SACKHDR;
 1782                         *optp++ = TCPOPT_SACK;
 1783                         sackblks = min(to->to_nsacks,
 1784                                         (TCP_MAXOLEN - optlen) / TCPOLEN_SACK);
 1785                         *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK;
 1786                         while (sackblks--) {
 1787                                 sack_seq = htonl(sack->start);
 1788                                 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
 1789                                 optp += sizeof(sack_seq);
 1790                                 sack_seq = htonl(sack->end);
 1791                                 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
 1792                                 optp += sizeof(sack_seq);
 1793                                 optlen += TCPOLEN_SACK;
 1794                                 sack++;
 1795                         }
 1796                         TCPSTAT_INC(tcps_sack_send_blocks);
 1797                         break;
 1798                         }
 1799 #ifdef TCP_RFC7413
 1800                 case TOF_FASTOPEN:
 1801                         {
 1802                         int total_len;
 1803 
 1804                         /* XXX is there any point to aligning this option? */
 1805                         total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len;
 1806                         if (TCP_MAXOLEN - optlen < total_len)
 1807                                 continue;
 1808                         *optp++ = TCPOPT_FAST_OPEN;
 1809                         *optp++ = total_len;
 1810                         if (to->to_tfo_len > 0) {
 1811                                 bcopy(to->to_tfo_cookie, optp, to->to_tfo_len);
 1812                                 optp += to->to_tfo_len;
 1813                         }
 1814                         optlen += total_len;
 1815                         break;
 1816                         }
 1817 #endif
 1818                 default:
 1819                         panic("%s: unknown TCP option type", __func__);
 1820                         break;
 1821                 }
 1822         }
 1823 
 1824         /* Terminate and pad TCP options to a 4 byte boundary. */
 1825         if (optlen % 4) {
 1826                 optlen += TCPOLEN_EOL;
 1827                 *optp++ = TCPOPT_EOL;
 1828         }
 1829         /*
 1830          * According to RFC 793 (STD0007):
 1831          *   "The content of the header beyond the End-of-Option option
 1832          *    must be header padding (i.e., zero)."
 1833          *   and later: "The padding is composed of zeros."
 1834          */
 1835         while (optlen % 4) {
 1836                 optlen += TCPOLEN_PAD;
 1837                 *optp++ = TCPOPT_PAD;
 1838         }
 1839 
 1840         KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__));
 1841         return (optlen);
 1842 }

Cache object: 1abee23dc24bdb839e537e852365afc2


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.