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("$FreeBSD$");
   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 = ((int32_t)ulmin(cwin,
  302                                     SEQ_SUB(tp->snd_recover, p->rxmit)));
  303                         }
  304                 } else {
  305                         len = ((int32_t)ulmin(cwin,
  306                             SEQ_SUB(p->end, p->rxmit)));
  307                 }
  308                 off = SEQ_SUB(p->rxmit, tp->snd_una);
  309                 KASSERT(off >= 0,("%s: sack block to the left of una : %d",
  310                     __func__, off));
  311                 if (len > 0) {
  312                         sack_rxmit = 1;
  313                         sendalot = 1;
  314                         TCPSTAT_INC(tcps_sack_rexmits);
  315                         TCPSTAT_ADD(tcps_sack_rexmit_bytes,
  316                             min(len, tp->t_maxseg));
  317                 }
  318         }
  319 after_sack_rexmit:
  320         /*
  321          * Get standard flags, and add SYN or FIN if requested by 'hidden'
  322          * state flags.
  323          */
  324         if (tp->t_flags & TF_NEEDFIN)
  325                 flags |= TH_FIN;
  326         if (tp->t_flags & TF_NEEDSYN)
  327                 flags |= TH_SYN;
  328 
  329         SOCKBUF_LOCK(&so->so_snd);
  330         /*
  331          * If in persist timeout with window of 0, send 1 byte.
  332          * Otherwise, if window is small but nonzero
  333          * and timer expired, we will send what we can
  334          * and go to transmit state.
  335          */
  336         if (tp->t_flags & TF_FORCEDATA) {
  337                 if (sendwin == 0) {
  338                         /*
  339                          * If we still have some data to send, then
  340                          * clear the FIN bit.  Usually this would
  341                          * happen below when it realizes that we
  342                          * aren't sending all the data.  However,
  343                          * if we have exactly 1 byte of unsent data,
  344                          * then it won't clear the FIN bit below,
  345                          * and if we are in persist state, we wind
  346                          * up sending the packet without recording
  347                          * that we sent the FIN bit.
  348                          *
  349                          * We can't just blindly clear the FIN bit,
  350                          * because if we don't have any more data
  351                          * to send then the probe will be the FIN
  352                          * itself.
  353                          */
  354                         if (off < sbused(&so->so_snd))
  355                                 flags &= ~TH_FIN;
  356                         sendwin = 1;
  357                 } else {
  358                         tcp_timer_activate(tp, TT_PERSIST, 0);
  359                         tp->t_rxtshift = 0;
  360                 }
  361         }
  362 
  363         /*
  364          * If snd_nxt == snd_max and we have transmitted a FIN, the
  365          * offset will be > 0 even if so_snd.sb_cc is 0, resulting in
  366          * a negative length.  This can also occur when TCP opens up
  367          * its congestion window while receiving additional duplicate
  368          * acks after fast-retransmit because TCP will reset snd_nxt
  369          * to snd_max after the fast-retransmit.
  370          *
  371          * In the normal retransmit-FIN-only case, however, snd_nxt will
  372          * be set to snd_una, the offset will be 0, and the length may
  373          * wind up 0.
  374          *
  375          * If sack_rxmit is true we are retransmitting from the scoreboard
  376          * in which case len is already set.
  377          */
  378         if (sack_rxmit == 0) {
  379                 if (sack_bytes_rxmt == 0)
  380                         len = ((long)ulmin(sbavail(&so->so_snd), sendwin) -
  381                             off);
  382                 else {
  383                         long cwin;
  384 
  385                         /*
  386                          * We are inside of a SACK recovery episode and are
  387                          * sending new data, having retransmitted all the
  388                          * data possible in the scoreboard.
  389                          */
  390                         len = ((long)ulmin(sbavail(&so->so_snd), tp->snd_wnd) -
  391                             off);
  392                         /*
  393                          * Don't remove this (len > 0) check !
  394                          * We explicitly check for len > 0 here (although it 
  395                          * isn't really necessary), to work around a gcc 
  396                          * optimization issue - to force gcc to compute
  397                          * len above. Without this check, the computation
  398                          * of len is bungled by the optimizer.
  399                          */
  400                         if (len > 0) {
  401                                 cwin = tp->snd_cwnd - 
  402                                         (tp->snd_nxt - tp->sack_newdata) -
  403                                         sack_bytes_rxmt;
  404                                 if (cwin < 0)
  405                                         cwin = 0;
  406                                 len = lmin(len, cwin);
  407                         }
  408                 }
  409         }
  410 
  411         /*
  412          * Lop off SYN bit if it has already been sent.  However, if this
  413          * is SYN-SENT state and if segment contains data and if we don't
  414          * know that foreign host supports TAO, suppress sending segment.
  415          */
  416         if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
  417                 if (tp->t_state != TCPS_SYN_RECEIVED)
  418                         flags &= ~TH_SYN;
  419 #ifdef TCP_RFC7413
  420                 /*
  421                  * When sending additional segments following a TFO SYN|ACK,
  422                  * do not include the SYN bit.
  423                  */
  424                 if ((tp->t_flags & TF_FASTOPEN) &&
  425                     (tp->t_state == TCPS_SYN_RECEIVED))
  426                         flags &= ~TH_SYN;
  427 #endif
  428                 off--, len++;
  429         }
  430 
  431         /*
  432          * Be careful not to send data and/or FIN on SYN segments.
  433          * This measure is needed to prevent interoperability problems
  434          * with not fully conformant TCP implementations.
  435          */
  436         if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) {
  437                 len = 0;
  438                 flags &= ~TH_FIN;
  439         }
  440 
  441 #ifdef TCP_RFC7413
  442         /*
  443          * When retransmitting SYN|ACK on a passively-created TFO socket,
  444          * don't include data, as the presence of data may have caused the
  445          * original SYN|ACK to have been dropped by a middlebox.
  446          */
  447         if ((tp->t_flags & TF_FASTOPEN) &&
  448             (((tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_rxtshift > 0)) ||
  449              (flags & TH_RST)))
  450                 len = 0;
  451 #endif
  452         if (len <= 0) {
  453                 /*
  454                  * If FIN has been sent but not acked,
  455                  * but we haven't been called to retransmit,
  456                  * len will be < 0.  Otherwise, window shrank
  457                  * after we sent into it.  If window shrank to 0,
  458                  * cancel pending retransmit, pull snd_nxt back
  459                  * to (closed) window, and set the persist timer
  460                  * if it isn't already going.  If the window didn't
  461                  * close completely, just wait for an ACK.
  462                  *
  463                  * We also do a general check here to ensure that
  464                  * we will set the persist timer when we have data
  465                  * to send, but a 0-byte window. This makes sure
  466                  * the persist timer is set even if the packet
  467                  * hits one of the "goto send" lines below.
  468                  */
  469                 len = 0;
  470                 if ((sendwin == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) &&
  471                         (off < (int) sbavail(&so->so_snd))) {
  472                         tcp_timer_activate(tp, TT_REXMT, 0);
  473                         tp->t_rxtshift = 0;
  474                         tp->snd_nxt = tp->snd_una;
  475                         if (!tcp_timer_active(tp, TT_PERSIST))
  476                                 tcp_setpersist(tp);
  477                 }
  478         }
  479 
  480         /* len will be >= 0 after this point. */
  481         KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
  482 
  483         /*
  484          * Automatic sizing of send socket buffer.  Often the send buffer
  485          * size is not optimally adjusted to the actual network conditions
  486          * at hand (delay bandwidth product).  Setting the buffer size too
  487          * small limits throughput on links with high bandwidth and high
  488          * delay (eg. trans-continental/oceanic links).  Setting the
  489          * buffer size too big consumes too much real kernel memory,
  490          * especially with many connections on busy servers.
  491          *
  492          * The criteria to step up the send buffer one notch are:
  493          *  1. receive window of remote host is larger than send buffer
  494          *     (with a fudge factor of 5/4th);
  495          *  2. send buffer is filled to 7/8th with data (so we actually
  496          *     have data to make use of it);
  497          *  3. send buffer fill has not hit maximal automatic size;
  498          *  4. our send window (slow start and cogestion controlled) is
  499          *     larger than sent but unacknowledged data in send buffer.
  500          *
  501          * The remote host receive window scaling factor may limit the
  502          * growing of the send buffer before it reaches its allowed
  503          * maximum.
  504          *
  505          * It scales directly with slow start or congestion window
  506          * and does at most one step per received ACK.  This fast
  507          * scaling has the drawback of growing the send buffer beyond
  508          * what is strictly necessary to make full use of a given
  509          * delay*bandwidth product.  However testing has shown this not
  510          * to be much of an problem.  At worst we are trading wasting
  511          * of available bandwidth (the non-use of it) for wasting some
  512          * socket buffer memory.
  513          *
  514          * TODO: Shrink send buffer during idle periods together
  515          * with congestion window.  Requires another timer.  Has to
  516          * wait for upcoming tcp timer rewrite.
  517          *
  518          * XXXGL: should there be used sbused() or sbavail()?
  519          */
  520         if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) {
  521                 if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat &&
  522                     sbused(&so->so_snd) >= (so->so_snd.sb_hiwat / 8 * 7) &&
  523                     sbused(&so->so_snd) < V_tcp_autosndbuf_max &&
  524                     sendwin >= (sbused(&so->so_snd) -
  525                     (tp->snd_nxt - tp->snd_una))) {
  526                         if (!sbreserve_locked(&so->so_snd,
  527                             min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc,
  528                              V_tcp_autosndbuf_max), so, curthread))
  529                                 so->so_snd.sb_flags &= ~SB_AUTOSIZE;
  530                 }
  531         }
  532 
  533         /*
  534          * Decide if we can use TCP Segmentation Offloading (if supported by
  535          * hardware).
  536          *
  537          * TSO may only be used if we are in a pure bulk sending state.  The
  538          * presence of TCP-MD5, SACK retransmits, SACK advertizements and
  539          * IP options prevent using TSO.  With TSO the TCP header is the same
  540          * (except for the sequence number) for all generated packets.  This
  541          * makes it impossible to transmit any options which vary per generated
  542          * segment or packet.
  543          *
  544          * IPv4 handling has a clear separation of ip options and ip header
  545          * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() does
  546          * the right thing below to provide length of just ip options and thus
  547          * checking for ipoptlen is enough to decide if ip options are present.
  548          */
  549 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
  550         /*
  551          * Pre-calculate here as we save another lookup into the darknesses
  552          * of IPsec that way and can actually decide if TSO is ok.
  553          */
  554 #ifdef INET6
  555         if (isipv6 && IPSEC_ENABLED(ipv6))
  556                 ipsec_optlen = IPSEC_HDRSIZE(ipv6, tp->t_inpcb);
  557 #ifdef INET
  558         else
  559 #endif
  560 #endif /* INET6 */
  561 #ifdef INET
  562         if (IPSEC_ENABLED(ipv4))
  563                 ipsec_optlen = IPSEC_HDRSIZE(ipv4, tp->t_inpcb);
  564 #endif /* INET */
  565 #endif /* IPSEC */
  566 #ifdef INET6
  567         if (isipv6)
  568                 ipoptlen = ip6_optlen(tp->t_inpcb);
  569         else
  570 #endif
  571         if (tp->t_inpcb->inp_options)
  572                 ipoptlen = tp->t_inpcb->inp_options->m_len -
  573                                 offsetof(struct ipoption, ipopt_list);
  574         else
  575                 ipoptlen = 0;
  576 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
  577         ipoptlen += ipsec_optlen;
  578 #endif
  579 
  580         if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg &&
  581             ((tp->t_flags & TF_SIGNATURE) == 0) &&
  582             tp->rcv_numsacks == 0 && sack_rxmit == 0 &&
  583             ipoptlen == 0)
  584                 tso = 1;
  585 
  586         if (sack_rxmit) {
  587                 if (SEQ_LT(p->rxmit + len, tp->snd_una + sbused(&so->so_snd)))
  588                         flags &= ~TH_FIN;
  589         } else {
  590                 if (SEQ_LT(tp->snd_nxt + len, tp->snd_una +
  591                     sbused(&so->so_snd)))
  592                         flags &= ~TH_FIN;
  593         }
  594 
  595         recwin = sbspace(&so->so_rcv);
  596 
  597         /*
  598          * Sender silly window avoidance.   We transmit under the following
  599          * conditions when len is non-zero:
  600          *
  601          *      - We have a full segment (or more with TSO)
  602          *      - This is the last buffer in a write()/send() and we are
  603          *        either idle or running NODELAY
  604          *      - we've timed out (e.g. persist timer)
  605          *      - we have more then 1/2 the maximum send window's worth of
  606          *        data (receiver may be limited the window size)
  607          *      - we need to retransmit
  608          */
  609         if (len) {
  610                 if (len >= tp->t_maxseg)
  611                         goto send;
  612                 /*
  613                  * NOTE! on localhost connections an 'ack' from the remote
  614                  * end may occur synchronously with the output and cause
  615                  * us to flush a buffer queued with moretocome.  XXX
  616                  *
  617                  * note: the len + off check is almost certainly unnecessary.
  618                  */
  619                 if (!(tp->t_flags & TF_MORETOCOME) &&   /* normal case */
  620                     (idle || (tp->t_flags & TF_NODELAY)) &&
  621                     len + off >= sbavail(&so->so_snd) &&
  622                     (tp->t_flags & TF_NOPUSH) == 0) {
  623                         goto send;
  624                 }
  625                 if (tp->t_flags & TF_FORCEDATA)         /* typ. timeout case */
  626                         goto send;
  627                 if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
  628                         goto send;
  629                 if (SEQ_LT(tp->snd_nxt, tp->snd_max))   /* retransmit case */
  630                         goto send;
  631                 if (sack_rxmit)
  632                         goto send;
  633         }
  634 
  635         /*
  636          * Sending of standalone window updates.
  637          *
  638          * Window updates are important when we close our window due to a
  639          * full socket buffer and are opening it again after the application
  640          * reads data from it.  Once the window has opened again and the
  641          * remote end starts to send again the ACK clock takes over and
  642          * provides the most current window information.
  643          *
  644          * We must avoid the silly window syndrome whereas every read
  645          * from the receive buffer, no matter how small, causes a window
  646          * update to be sent.  We also should avoid sending a flurry of
  647          * window updates when the socket buffer had queued a lot of data
  648          * and the application is doing small reads.
  649          *
  650          * Prevent a flurry of pointless window updates by only sending
  651          * an update when we can increase the advertized window by more
  652          * than 1/4th of the socket buffer capacity.  When the buffer is
  653          * getting full or is very small be more aggressive and send an
  654          * update whenever we can increase by two mss sized segments.
  655          * In all other situations the ACK's to new incoming data will
  656          * carry further window increases.
  657          *
  658          * Don't send an independent window update if a delayed
  659          * ACK is pending (it will get piggy-backed on it) or the
  660          * remote side already has done a half-close and won't send
  661          * more data.  Skip this if the connection is in T/TCP
  662          * half-open state.
  663          */
  664         if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) &&
  665             !(tp->t_flags & TF_DELACK) &&
  666             !TCPS_HAVERCVDFIN(tp->t_state)) {
  667                 /*
  668                  * "adv" is the amount we could increase the window,
  669                  * taking into account that we are limited by
  670                  * TCP_MAXWIN << tp->rcv_scale.
  671                  */
  672                 long adv;
  673                 int oldwin;
  674 
  675                 adv = min(recwin, (long)TCP_MAXWIN << tp->rcv_scale);
  676                 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) {
  677                         oldwin = (tp->rcv_adv - tp->rcv_nxt);
  678                         adv -= oldwin;
  679                 } else
  680                         oldwin = 0;
  681 
  682                 /* 
  683                  * If the new window size ends up being the same as or less
  684                  * than the old size when it is scaled, then don't force
  685                  * a window update.
  686                  */
  687                 if (oldwin >> tp->rcv_scale >= (adv + oldwin) >> tp->rcv_scale)
  688                         goto dontupdate;
  689 
  690                 if (adv >= (long)(2 * tp->t_maxseg) &&
  691                     (adv >= (long)(so->so_rcv.sb_hiwat / 4) ||
  692                      recwin <= (long)(so->so_rcv.sb_hiwat / 8) ||
  693                      so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg))
  694                         goto send;
  695                 if (2 * adv >= (int32_t)so->so_rcv.sb_hiwat)
  696                         goto send;
  697         }
  698 dontupdate:
  699 
  700         /*
  701          * Send if we owe the peer an ACK, RST, SYN, or urgent data.  ACKNOW
  702          * is also a catch-all for the retransmit timer timeout case.
  703          */
  704         if (tp->t_flags & TF_ACKNOW)
  705                 goto send;
  706         if ((flags & TH_RST) ||
  707             ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0))
  708                 goto send;
  709         if (SEQ_GT(tp->snd_up, tp->snd_una))
  710                 goto send;
  711         /*
  712          * If our state indicates that FIN should be sent
  713          * and we have not yet done so, then we need to send.
  714          */
  715         if (flags & TH_FIN &&
  716             ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
  717                 goto send;
  718         /*
  719          * In SACK, it is possible for tcp_output to fail to send a segment
  720          * after the retransmission timer has been turned off.  Make sure
  721          * that the retransmission timer is set.
  722          */
  723         if ((tp->t_flags & TF_SACK_PERMIT) &&
  724             SEQ_GT(tp->snd_max, tp->snd_una) &&
  725             !tcp_timer_active(tp, TT_REXMT) &&
  726             !tcp_timer_active(tp, TT_PERSIST)) {
  727                 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
  728                 goto just_return;
  729         } 
  730         /*
  731          * TCP window updates are not reliable, rather a polling protocol
  732          * using ``persist'' packets is used to insure receipt of window
  733          * updates.  The three ``states'' for the output side are:
  734          *      idle                    not doing retransmits or persists
  735          *      persisting              to move a small or zero window
  736          *      (re)transmitting        and thereby not persisting
  737          *
  738          * tcp_timer_active(tp, TT_PERSIST)
  739          *      is true when we are in persist state.
  740          * (tp->t_flags & TF_FORCEDATA)
  741          *      is set when we are called to send a persist packet.
  742          * tcp_timer_active(tp, TT_REXMT)
  743          *      is set when we are retransmitting
  744          * The output side is idle when both timers are zero.
  745          *
  746          * If send window is too small, there is data to transmit, and no
  747          * retransmit or persist is pending, then go to persist state.
  748          * If nothing happens soon, send when timer expires:
  749          * if window is nonzero, transmit what we can,
  750          * otherwise force out a byte.
  751          */
  752         if (sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) &&
  753             !tcp_timer_active(tp, TT_PERSIST)) {
  754                 tp->t_rxtshift = 0;
  755                 tcp_setpersist(tp);
  756         }
  757 
  758         /*
  759          * No reason to send a segment, just return.
  760          */
  761 just_return:
  762         SOCKBUF_UNLOCK(&so->so_snd);
  763         return (0);
  764 
  765 send:
  766         SOCKBUF_LOCK_ASSERT(&so->so_snd);
  767         if (len > 0) {
  768                 if (len >= tp->t_maxseg)
  769                         tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT;
  770                 else
  771                         tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT;
  772         }
  773         /*
  774          * Before ESTABLISHED, force sending of initial options
  775          * unless TCP set not to do any options.
  776          * NOTE: we assume that the IP/TCP header plus TCP options
  777          * always fit in a single mbuf, leaving room for a maximum
  778          * link header, i.e.
  779          *      max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES
  780          */
  781         optlen = 0;
  782 #ifdef INET6
  783         if (isipv6)
  784                 hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
  785         else
  786 #endif
  787                 hdrlen = sizeof (struct tcpiphdr);
  788 
  789         /*
  790          * Compute options for segment.
  791          * We only have to care about SYN and established connection
  792          * segments.  Options for SYN-ACK segments are handled in TCP
  793          * syncache.
  794          */
  795         to.to_flags = 0;
  796         if ((tp->t_flags & TF_NOOPT) == 0) {
  797                 /* Maximum segment size. */
  798                 if (flags & TH_SYN) {
  799                         tp->snd_nxt = tp->iss;
  800                         to.to_mss = tcp_mssopt(&tp->t_inpcb->inp_inc);
  801                         to.to_flags |= TOF_MSS;
  802 #ifdef TCP_RFC7413
  803                         /*
  804                          * Only include the TFO option on the first
  805                          * transmission of the SYN|ACK on a
  806                          * passively-created TFO socket, as the presence of
  807                          * the TFO option may have caused the original
  808                          * SYN|ACK to have been dropped by a middlebox.
  809                          */
  810                         if ((tp->t_flags & TF_FASTOPEN) &&
  811                             (tp->t_state == TCPS_SYN_RECEIVED) &&
  812                             (tp->t_rxtshift == 0)) {
  813                                 to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN;
  814                                 to.to_tfo_cookie = (u_char *)&tp->t_tfo_cookie;
  815                                 to.to_flags |= TOF_FASTOPEN;
  816                         }
  817 #endif
  818                 }
  819                 /* Window scaling. */
  820                 if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) {
  821                         to.to_wscale = tp->request_r_scale;
  822                         to.to_flags |= TOF_SCALE;
  823                 }
  824                 /* Timestamps. */
  825                 if ((tp->t_flags & TF_RCVD_TSTMP) ||
  826                     ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) {
  827                         to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
  828                         to.to_tsecr = tp->ts_recent;
  829                         to.to_flags |= TOF_TS;
  830                 }
  831 
  832                 /* Set receive buffer autosizing timestamp. */
  833                 if (tp->rfbuf_ts == 0 &&
  834                     (so->so_rcv.sb_flags & SB_AUTOSIZE))
  835                         tp->rfbuf_ts = tcp_ts_getticks();
  836 
  837                 /* Selective ACK's. */
  838                 if (tp->t_flags & TF_SACK_PERMIT) {
  839                         if (flags & TH_SYN)
  840                                 to.to_flags |= TOF_SACKPERM;
  841                         else if (TCPS_HAVEESTABLISHED(tp->t_state) &&
  842                             (tp->t_flags & TF_SACK_PERMIT) &&
  843                             tp->rcv_numsacks > 0) {
  844                                 to.to_flags |= TOF_SACK;
  845                                 to.to_nsacks = tp->rcv_numsacks;
  846                                 to.to_sacks = (u_char *)tp->sackblks;
  847                         }
  848                 }
  849 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
  850                 /* TCP-MD5 (RFC2385). */
  851                 /*
  852                  * Check that TCP_MD5SIG is enabled in tcpcb to
  853                  * account the size needed to set this TCP option.
  854                  */
  855                 if (tp->t_flags & TF_SIGNATURE)
  856                         to.to_flags |= TOF_SIGNATURE;
  857 #endif /* TCP_SIGNATURE */
  858 
  859                 /* Processing the options. */
  860                 hdrlen += optlen = tcp_addoptions(&to, opt);
  861         }
  862 
  863         /*
  864          * Adjust data length if insertion of options will
  865          * bump the packet length beyond the t_maxseg length.
  866          * Clear the FIN bit because we cut off the tail of
  867          * the segment.
  868          */
  869         if (len + optlen + ipoptlen > tp->t_maxseg) {
  870                 flags &= ~TH_FIN;
  871 
  872                 if (tso) {
  873                         u_int if_hw_tsomax;
  874                         u_int if_hw_tsomaxsegcount;
  875                         u_int if_hw_tsomaxsegsize;
  876                         struct mbuf *mb;
  877                         u_int moff;
  878                         int max_len;
  879 
  880                         /* extract TSO information */
  881                         if_hw_tsomax = tp->t_tsomax;
  882                         if_hw_tsomaxsegcount = tp->t_tsomaxsegcount;
  883                         if_hw_tsomaxsegsize = tp->t_tsomaxsegsize;
  884 
  885                         /*
  886                          * Limit a TSO burst to prevent it from
  887                          * overflowing or exceeding the maximum length
  888                          * allowed by the network interface:
  889                          */
  890                         KASSERT(ipoptlen == 0,
  891                             ("%s: TSO can't do IP options", __func__));
  892 
  893                         /*
  894                          * Check if we should limit by maximum payload
  895                          * length:
  896                          */
  897                         if (if_hw_tsomax != 0) {
  898                                 /* compute maximum TSO length */
  899                                 max_len = (if_hw_tsomax - hdrlen -
  900                                     max_linkhdr);
  901                                 if (max_len <= 0) {
  902                                         len = 0;
  903                                 } else if (len > max_len) {
  904                                         sendalot = 1;
  905                                         len = max_len;
  906                                 }
  907                         }
  908 
  909                         /*
  910                          * Check if we should limit by maximum segment
  911                          * size and count:
  912                          */
  913                         if (if_hw_tsomaxsegcount != 0 &&
  914                             if_hw_tsomaxsegsize != 0) {
  915                                 /*
  916                                  * Subtract one segment for the LINK
  917                                  * and TCP/IP headers mbuf that will
  918                                  * be prepended to this mbuf chain
  919                                  * after the code in this section
  920                                  * limits the number of mbufs in the
  921                                  * chain to if_hw_tsomaxsegcount.
  922                                  */
  923                                 if_hw_tsomaxsegcount -= 1;
  924                                 max_len = 0;
  925                                 mb = sbsndmbuf(&so->so_snd, off, &moff);
  926 
  927                                 while (mb != NULL && max_len < len) {
  928                                         u_int mlen;
  929                                         u_int frags;
  930 
  931                                         /*
  932                                          * Get length of mbuf fragment
  933                                          * and how many hardware frags,
  934                                          * rounded up, it would use:
  935                                          */
  936                                         mlen = (mb->m_len - moff);
  937                                         frags = howmany(mlen,
  938                                             if_hw_tsomaxsegsize);
  939 
  940                                         /* Handle special case: Zero Length Mbuf */
  941                                         if (frags == 0)
  942                                                 frags = 1;
  943 
  944                                         /*
  945                                          * Check if the fragment limit
  946                                          * will be reached or exceeded:
  947                                          */
  948                                         if (frags >= if_hw_tsomaxsegcount) {
  949                                                 max_len += min(mlen,
  950                                                     if_hw_tsomaxsegcount *
  951                                                     if_hw_tsomaxsegsize);
  952                                                 break;
  953                                         }
  954                                         max_len += mlen;
  955                                         if_hw_tsomaxsegcount -= frags;
  956                                         moff = 0;
  957                                         mb = mb->m_next;
  958                                 }
  959                                 if (max_len <= 0) {
  960                                         len = 0;
  961                                 } else if (len > max_len) {
  962                                         sendalot = 1;
  963                                         len = max_len;
  964                                 }
  965                         }
  966 
  967                         /*
  968                          * Prevent the last segment from being
  969                          * fractional unless the send sockbuf can be
  970                          * emptied:
  971                          */
  972                         max_len = (tp->t_maxseg - optlen);
  973                         if ((off + len) < sbavail(&so->so_snd)) {
  974                                 moff = len % max_len;
  975                                 if (moff != 0) {
  976                                         len -= moff;
  977                                         sendalot = 1;
  978                                 }
  979                         }
  980 
  981                         /*
  982                          * In case there are too many small fragments
  983                          * don't use TSO:
  984                          */
  985                         if (len <= max_len) {
  986                                 len = max_len;
  987                                 sendalot = 1;
  988                                 tso = 0;
  989                         }
  990 
  991                         /*
  992                          * Send the FIN in a separate segment
  993                          * after the bulk sending is done.
  994                          * We don't trust the TSO implementations
  995                          * to clear the FIN flag on all but the
  996                          * last segment.
  997                          */
  998                         if (tp->t_flags & TF_NEEDFIN)
  999                                 sendalot = 1;
 1000 
 1001                 } else {
 1002                         len = tp->t_maxseg - optlen - ipoptlen;
 1003                         sendalot = 1;
 1004                 }
 1005         } else
 1006                 tso = 0;
 1007 
 1008         KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET,
 1009             ("%s: len > IP_MAXPACKET", __func__));
 1010 
 1011 /*#ifdef DIAGNOSTIC*/
 1012 #ifdef INET6
 1013         if (max_linkhdr + hdrlen > MCLBYTES)
 1014 #else
 1015         if (max_linkhdr + hdrlen > MHLEN)
 1016 #endif
 1017                 panic("tcphdr too big");
 1018 /*#endif*/
 1019 
 1020         /*
 1021          * This KASSERT is here to catch edge cases at a well defined place.
 1022          * Before, those had triggered (random) panic conditions further down.
 1023          */
 1024         KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__));
 1025 
 1026         /*
 1027          * Grab a header mbuf, attaching a copy of data to
 1028          * be transmitted, and initialize the header from
 1029          * the template for sends on this connection.
 1030          */
 1031         if (len) {
 1032                 struct mbuf *mb;
 1033                 u_int moff;
 1034 
 1035                 if ((tp->t_flags & TF_FORCEDATA) && len == 1)
 1036                         TCPSTAT_INC(tcps_sndprobe);
 1037                 else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) {
 1038                         tp->t_sndrexmitpack++;
 1039                         TCPSTAT_INC(tcps_sndrexmitpack);
 1040                         TCPSTAT_ADD(tcps_sndrexmitbyte, len);
 1041                 } else {
 1042                         TCPSTAT_INC(tcps_sndpack);
 1043                         TCPSTAT_ADD(tcps_sndbyte, len);
 1044                 }
 1045 #ifdef INET6
 1046                 if (MHLEN < hdrlen + max_linkhdr)
 1047                         m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
 1048                 else
 1049 #endif
 1050                         m = m_gethdr(M_NOWAIT, MT_DATA);
 1051 
 1052                 if (m == NULL) {
 1053                         SOCKBUF_UNLOCK(&so->so_snd);
 1054                         error = ENOBUFS;
 1055                         sack_rxmit = 0;
 1056                         goto out;
 1057                 }
 1058 
 1059                 m->m_data += max_linkhdr;
 1060                 m->m_len = hdrlen;
 1061 
 1062                 /*
 1063                  * Start the m_copy functions from the closest mbuf
 1064                  * to the offset in the socket buffer chain.
 1065                  */
 1066                 mb = sbsndptr(&so->so_snd, off, len, &moff);
 1067 
 1068                 if (len <= MHLEN - hdrlen - max_linkhdr) {
 1069                         m_copydata(mb, moff, (int)len,
 1070                             mtod(m, caddr_t) + hdrlen);
 1071                         m->m_len += len;
 1072                 } else {
 1073                         m->m_next = m_copy(mb, moff, (int)len);
 1074                         if (m->m_next == NULL) {
 1075                                 SOCKBUF_UNLOCK(&so->so_snd);
 1076                                 (void) m_free(m);
 1077                                 error = ENOBUFS;
 1078                                 sack_rxmit = 0;
 1079                                 goto out;
 1080                         }
 1081                 }
 1082 
 1083                 /*
 1084                  * If we're sending everything we've got, set PUSH.
 1085                  * (This will keep happy those implementations which only
 1086                  * give data to the user when a buffer fills or
 1087                  * a PUSH comes in.)
 1088                  */
 1089                 if ((off + len == sbused(&so->so_snd)) && !(flags & TH_SYN))
 1090                         flags |= TH_PUSH;
 1091                 SOCKBUF_UNLOCK(&so->so_snd);
 1092         } else {
 1093                 SOCKBUF_UNLOCK(&so->so_snd);
 1094                 if (tp->t_flags & TF_ACKNOW)
 1095                         TCPSTAT_INC(tcps_sndacks);
 1096                 else if (flags & (TH_SYN|TH_FIN|TH_RST))
 1097                         TCPSTAT_INC(tcps_sndctrl);
 1098                 else if (SEQ_GT(tp->snd_up, tp->snd_una))
 1099                         TCPSTAT_INC(tcps_sndurg);
 1100                 else
 1101                         TCPSTAT_INC(tcps_sndwinup);
 1102 
 1103                 m = m_gethdr(M_NOWAIT, MT_DATA);
 1104                 if (m == NULL) {
 1105                         error = ENOBUFS;
 1106                         sack_rxmit = 0;
 1107                         goto out;
 1108                 }
 1109 #ifdef INET6
 1110                 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) &&
 1111                     MHLEN >= hdrlen) {
 1112                         M_ALIGN(m, hdrlen);
 1113                 } else
 1114 #endif
 1115                 m->m_data += max_linkhdr;
 1116                 m->m_len = hdrlen;
 1117         }
 1118         SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
 1119         m->m_pkthdr.rcvif = (struct ifnet *)0;
 1120 #ifdef MAC
 1121         mac_inpcb_create_mbuf(tp->t_inpcb, m);
 1122 #endif
 1123 #ifdef INET6
 1124         if (isipv6) {
 1125                 ip6 = mtod(m, struct ip6_hdr *);
 1126                 th = (struct tcphdr *)(ip6 + 1);
 1127                 tcpip_fillheaders(tp->t_inpcb, ip6, th);
 1128         } else
 1129 #endif /* INET6 */
 1130         {
 1131                 ip = mtod(m, struct ip *);
 1132                 ipov = (struct ipovly *)ip;
 1133                 th = (struct tcphdr *)(ip + 1);
 1134                 tcpip_fillheaders(tp->t_inpcb, ip, th);
 1135         }
 1136 
 1137         /*
 1138          * Fill in fields, remembering maximum advertised
 1139          * window for use in delaying messages about window sizes.
 1140          * If resending a FIN, be sure not to use a new sequence number.
 1141          */
 1142         if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
 1143             tp->snd_nxt == tp->snd_max)
 1144                 tp->snd_nxt--;
 1145         /*
 1146          * If we are starting a connection, send ECN setup
 1147          * SYN packet. If we are on a retransmit, we may
 1148          * resend those bits a number of times as per
 1149          * RFC 3168.
 1150          */
 1151         if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn == 1) {
 1152                 if (tp->t_rxtshift >= 1) {
 1153                         if (tp->t_rxtshift <= V_tcp_ecn_maxretries)
 1154                                 flags |= TH_ECE|TH_CWR;
 1155                 } else
 1156                         flags |= TH_ECE|TH_CWR;
 1157         }
 1158 
 1159         if (TCPS_HAVEESTABLISHED(tp->t_state) &&
 1160             (tp->t_flags & TF_ECN_PERMIT)) {
 1161                 /*
 1162                  * If the peer has ECN, mark data packets with
 1163                  * ECN capable transmission (ECT).
 1164                  * Ignore pure ack packets, retransmissions and window probes.
 1165                  */
 1166                 if (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) &&
 1167                     !((tp->t_flags & TF_FORCEDATA) && len == 1 &&
 1168                     SEQ_LT(tp->snd_una, tp->snd_max))) {
 1169 #ifdef INET6
 1170                         if (isipv6)
 1171                                 ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
 1172                         else
 1173 #endif
 1174                                 ip->ip_tos |= IPTOS_ECN_ECT0;
 1175                         TCPSTAT_INC(tcps_ecn_ect0);
 1176                         /*
 1177                          * Reply with proper ECN notifications.
 1178                          * Only set CWR on new data segments.
 1179                          */
 1180                         if (tp->t_flags & TF_ECN_SND_CWR) {
 1181                                 flags |= TH_CWR;
 1182                                 tp->t_flags &= ~TF_ECN_SND_CWR;
 1183                         }
 1184                 }
 1185                 if (tp->t_flags & TF_ECN_SND_ECE)
 1186                         flags |= TH_ECE;
 1187         }
 1188         
 1189         /*
 1190          * If we are doing retransmissions, then snd_nxt will
 1191          * not reflect the first unsent octet.  For ACK only
 1192          * packets, we do not want the sequence number of the
 1193          * retransmitted packet, we want the sequence number
 1194          * of the next unsent octet.  So, if there is no data
 1195          * (and no SYN or FIN), use snd_max instead of snd_nxt
 1196          * when filling in ti_seq.  But if we are in persist
 1197          * state, snd_max might reflect one byte beyond the
 1198          * right edge of the window, so use snd_nxt in that
 1199          * case, since we know we aren't doing a retransmission.
 1200          * (retransmit and persist are mutually exclusive...)
 1201          */
 1202         if (sack_rxmit == 0) {
 1203                 if (len || (flags & (TH_SYN|TH_FIN)) ||
 1204                     tcp_timer_active(tp, TT_PERSIST))
 1205                         th->th_seq = htonl(tp->snd_nxt);
 1206                 else
 1207                         th->th_seq = htonl(tp->snd_max);
 1208         } else {
 1209                 th->th_seq = htonl(p->rxmit);
 1210                 p->rxmit += len;
 1211                 tp->sackhint.sack_bytes_rexmit += len;
 1212         }
 1213         th->th_ack = htonl(tp->rcv_nxt);
 1214         if (optlen) {
 1215                 bcopy(opt, th + 1, optlen);
 1216                 th->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
 1217         }
 1218         th->th_flags = flags;
 1219         /*
 1220          * Calculate receive window.  Don't shrink window,
 1221          * but avoid silly window syndrome.
 1222          */
 1223         if (recwin < (long)(so->so_rcv.sb_hiwat / 4) &&
 1224             recwin < (long)tp->t_maxseg)
 1225                 recwin = 0;
 1226         if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) &&
 1227             recwin < (long)(tp->rcv_adv - tp->rcv_nxt))
 1228                 recwin = (long)(tp->rcv_adv - tp->rcv_nxt);
 1229         if (recwin > (long)TCP_MAXWIN << tp->rcv_scale)
 1230                 recwin = (long)TCP_MAXWIN << tp->rcv_scale;
 1231 
 1232         /*
 1233          * According to RFC1323 the window field in a SYN (i.e., a <SYN>
 1234          * or <SYN,ACK>) segment itself is never scaled.  The <SYN,ACK>
 1235          * case is handled in syncache.
 1236          */
 1237         if (flags & TH_SYN)
 1238                 th->th_win = htons((u_short)
 1239                                 (min(sbspace(&so->so_rcv), TCP_MAXWIN)));
 1240         else
 1241                 th->th_win = htons((u_short)(recwin >> tp->rcv_scale));
 1242 
 1243         /*
 1244          * Adjust the RXWIN0SENT flag - indicate that we have advertised
 1245          * a 0 window.  This may cause the remote transmitter to stall.  This
 1246          * flag tells soreceive() to disable delayed acknowledgements when
 1247          * draining the buffer.  This can occur if the receiver is attempting
 1248          * to read more data than can be buffered prior to transmitting on
 1249          * the connection.
 1250          */
 1251         if (th->th_win == 0) {
 1252                 tp->t_sndzerowin++;
 1253                 tp->t_flags |= TF_RXWIN0SENT;
 1254         } else
 1255                 tp->t_flags &= ~TF_RXWIN0SENT;
 1256         if (SEQ_GT(tp->snd_up, tp->snd_nxt)) {
 1257                 th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt));
 1258                 th->th_flags |= TH_URG;
 1259         } else
 1260                 /*
 1261                  * If no urgent pointer to send, then we pull
 1262                  * the urgent pointer to the left edge of the send window
 1263                  * so that it doesn't drift into the send window on sequence
 1264                  * number wraparound.
 1265                  */
 1266                 tp->snd_up = tp->snd_una;               /* drag it along */
 1267 
 1268         /*
 1269          * Put TCP length in extended header, and then
 1270          * checksum extended header and data.
 1271          */
 1272         m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */
 1273         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
 1274 
 1275 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
 1276         if (to.to_flags & TOF_SIGNATURE) {
 1277                 /*
 1278                  * Calculate MD5 signature and put it into the place
 1279                  * determined before.
 1280                  * NOTE: since TCP options buffer doesn't point into
 1281                  * mbuf's data, calculate offset and use it.
 1282                  */
 1283                 if (!TCPMD5_ENABLED() || (error = TCPMD5_OUTPUT(m, th,
 1284                     (u_char *)(th + 1) + (to.to_signature - opt))) != 0) {
 1285                         /*
 1286                          * Do not send segment if the calculation of MD5
 1287                          * digest has failed.
 1288                          */
 1289                         m_freem(m);
 1290                         goto out;
 1291                 }
 1292         }
 1293 #endif
 1294 #ifdef INET6
 1295         if (isipv6) {
 1296                 /*
 1297                  * There is no need to fill in ip6_plen right now.
 1298                  * It will be filled later by ip6_output.
 1299                  */
 1300                 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
 1301                 th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) +
 1302                     optlen + len, IPPROTO_TCP, 0);
 1303         }
 1304 #endif
 1305 #if defined(INET6) && defined(INET)
 1306         else
 1307 #endif
 1308 #ifdef INET
 1309         {
 1310                 m->m_pkthdr.csum_flags = CSUM_TCP;
 1311                 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 1312                     htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen));
 1313 
 1314                 /* IP version must be set here for ipv4/ipv6 checking later */
 1315                 KASSERT(ip->ip_v == IPVERSION,
 1316                     ("%s: IP version incorrect: %d", __func__, ip->ip_v));
 1317         }
 1318 #endif
 1319 
 1320         /*
 1321          * Enable TSO and specify the size of the segments.
 1322          * The TCP pseudo header checksum is always provided.
 1323          */
 1324         if (tso) {
 1325                 KASSERT(len > tp->t_maxseg - optlen,
 1326                     ("%s: len <= tso_segsz", __func__));
 1327                 m->m_pkthdr.csum_flags |= CSUM_TSO;
 1328                 m->m_pkthdr.tso_segsz = tp->t_maxseg - optlen;
 1329         }
 1330 
 1331 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 1332         KASSERT(len + hdrlen + ipoptlen - ipsec_optlen == m_length(m, NULL),
 1333             ("%s: mbuf chain shorter than expected: %ld + %u + %u - %u != %u",
 1334             __func__, len, hdrlen, ipoptlen, ipsec_optlen, m_length(m, NULL)));
 1335 #else
 1336         KASSERT(len + hdrlen + ipoptlen == m_length(m, NULL),
 1337             ("%s: mbuf chain shorter than expected: %ld + %u + %u != %u",
 1338             __func__, len, hdrlen, ipoptlen, m_length(m, NULL)));
 1339 #endif
 1340 
 1341         /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */
 1342         hhook_run_tcp_est_out(tp, th, &to, len, tso);
 1343 
 1344 #ifdef TCPDEBUG
 1345         /*
 1346          * Trace.
 1347          */
 1348         if (so->so_options & SO_DEBUG) {
 1349                 u_short save = 0;
 1350 #ifdef INET6
 1351                 if (!isipv6)
 1352 #endif
 1353                 {
 1354                         save = ipov->ih_len;
 1355                         ipov->ih_len = htons(m->m_pkthdr.len /* - hdrlen + (th->th_off << 2) */);
 1356                 }
 1357                 tcp_trace(TA_OUTPUT, tp->t_state, tp, mtod(m, void *), th, 0);
 1358 #ifdef INET6
 1359                 if (!isipv6)
 1360 #endif
 1361                 ipov->ih_len = save;
 1362         }
 1363 #endif /* TCPDEBUG */
 1364         TCP_PROBE3(debug__output, tp, th, m);
 1365 
 1366         /*
 1367          * Fill in IP length and desired time to live and
 1368          * send to IP level.  There should be a better way
 1369          * to handle ttl and tos; we could keep them in
 1370          * the template, but need a way to checksum without them.
 1371          */
 1372         /*
 1373          * m->m_pkthdr.len should have been set before checksum calculation,
 1374          * because in6_cksum() need it.
 1375          */
 1376 #ifdef INET6
 1377         if (isipv6) {
 1378                 /*
 1379                  * we separately set hoplimit for every segment, since the
 1380                  * user might want to change the value via setsockopt.
 1381                  * Also, desired default hop limit might be changed via
 1382                  * Neighbor Discovery.
 1383                  */
 1384                 ip6->ip6_hlim = in6_selecthlim(tp->t_inpcb, NULL);
 1385 
 1386                 /*
 1387                  * Set the packet size here for the benefit of DTrace probes.
 1388                  * ip6_output() will set it properly; it's supposed to include
 1389                  * the option header lengths as well.
 1390                  */
 1391                 ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
 1392 
 1393                 if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss)
 1394                         tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 1395                 else
 1396                         tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 1397 
 1398                 if (tp->t_state == TCPS_SYN_SENT)
 1399                         TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th);
 1400 
 1401                 TCP_PROBE5(send, NULL, tp, ip6, tp, th);
 1402 
 1403 #ifdef TCPPCAP
 1404                 /* Save packet, if requested. */
 1405                 tcp_pcap_add(th, m, &(tp->t_outpkts));
 1406 #endif
 1407 
 1408                 /* TODO: IPv6 IP6TOS_ECT bit on */
 1409                 error = ip6_output(m, tp->t_inpcb->in6p_outputopts,
 1410                     &tp->t_inpcb->inp_route6,
 1411                     ((so->so_options & SO_DONTROUTE) ?  IP_ROUTETOIF : 0),
 1412                     NULL, NULL, tp->t_inpcb);
 1413 
 1414                 if (error == EMSGSIZE && tp->t_inpcb->inp_route6.ro_rt != NULL)
 1415                         mtu = tp->t_inpcb->inp_route6.ro_rt->rt_mtu;
 1416         }
 1417 #endif /* INET6 */
 1418 #if defined(INET) && defined(INET6)
 1419         else
 1420 #endif
 1421 #ifdef INET
 1422     {
 1423         ip->ip_len = htons(m->m_pkthdr.len);
 1424 #ifdef INET6
 1425         if (tp->t_inpcb->inp_vflag & INP_IPV6PROTO)
 1426                 ip->ip_ttl = in6_selecthlim(tp->t_inpcb, NULL);
 1427 #endif /* INET6 */
 1428         /*
 1429          * If we do path MTU discovery, then we set DF on every packet.
 1430          * This might not be the best thing to do according to RFC3390
 1431          * Section 2. However the tcp hostcache migitates the problem
 1432          * so it affects only the first tcp connection with a host.
 1433          *
 1434          * NB: Don't set DF on small MTU/MSS to have a safe fallback.
 1435          */
 1436         if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) {
 1437                 ip->ip_off |= htons(IP_DF);
 1438                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
 1439         } else {
 1440                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
 1441         }
 1442 
 1443         if (tp->t_state == TCPS_SYN_SENT)
 1444                 TCP_PROBE5(connect__request, NULL, tp, ip, tp, th);
 1445 
 1446         TCP_PROBE5(send, NULL, tp, ip, tp, th);
 1447 
 1448 #ifdef TCPPCAP
 1449         /* Save packet, if requested. */
 1450         tcp_pcap_add(th, m, &(tp->t_outpkts));
 1451 #endif
 1452 
 1453         error = ip_output(m, tp->t_inpcb->inp_options, &tp->t_inpcb->inp_route,
 1454             ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0,
 1455             tp->t_inpcb);
 1456 
 1457         if (error == EMSGSIZE && tp->t_inpcb->inp_route.ro_rt != NULL)
 1458                 mtu = tp->t_inpcb->inp_route.ro_rt->rt_mtu;
 1459     }
 1460 #endif /* INET */
 1461 
 1462 out:
 1463         /*
 1464          * In transmit state, time the transmission and arrange for
 1465          * the retransmit.  In persist state, just set snd_max.
 1466          */
 1467         if ((tp->t_flags & TF_FORCEDATA) == 0 || 
 1468             !tcp_timer_active(tp, TT_PERSIST)) {
 1469                 tcp_seq startseq = tp->snd_nxt;
 1470 
 1471                 /*
 1472                  * Advance snd_nxt over sequence space of this segment.
 1473                  */
 1474                 if (flags & (TH_SYN|TH_FIN)) {
 1475                         if (flags & TH_SYN)
 1476                                 tp->snd_nxt++;
 1477                         if (flags & TH_FIN) {
 1478                                 tp->snd_nxt++;
 1479                                 tp->t_flags |= TF_SENTFIN;
 1480                         }
 1481                 }
 1482                 if (sack_rxmit)
 1483                         goto timer;
 1484                 tp->snd_nxt += len;
 1485                 if (SEQ_GT(tp->snd_nxt, tp->snd_max)) {
 1486                         tp->snd_max = tp->snd_nxt;
 1487                         /*
 1488                          * Time this transmission if not a retransmission and
 1489                          * not currently timing anything.
 1490                          */
 1491                         if (tp->t_rtttime == 0) {
 1492                                 tp->t_rtttime = ticks;
 1493                                 tp->t_rtseq = startseq;
 1494                                 TCPSTAT_INC(tcps_segstimed);
 1495                         }
 1496                 }
 1497 
 1498                 /*
 1499                  * Set retransmit timer if not currently set,
 1500                  * and not doing a pure ack or a keep-alive probe.
 1501                  * Initial value for retransmit timer is smoothed
 1502                  * round-trip time + 2 * round-trip time variance.
 1503                  * Initialize shift counter which is used for backoff
 1504                  * of retransmit time.
 1505                  */
 1506 timer:
 1507                 if (!tcp_timer_active(tp, TT_REXMT) &&
 1508                     ((sack_rxmit && tp->snd_nxt != tp->snd_max) ||
 1509                      (tp->snd_nxt != tp->snd_una))) {
 1510                         if (tcp_timer_active(tp, TT_PERSIST)) {
 1511                                 tcp_timer_activate(tp, TT_PERSIST, 0);
 1512                                 tp->t_rxtshift = 0;
 1513                         }
 1514                         tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
 1515                 } else if (len == 0 && sbavail(&so->so_snd) &&
 1516                     !tcp_timer_active(tp, TT_REXMT) &&
 1517                     !tcp_timer_active(tp, TT_PERSIST)) {
 1518                         /*
 1519                          * Avoid a situation where we do not set persist timer
 1520                          * after a zero window condition. For example:
 1521                          * 1) A -> B: packet with enough data to fill the window
 1522                          * 2) B -> A: ACK for #1 + new data (0 window
 1523                          *    advertisement)
 1524                          * 3) A -> B: ACK for #2, 0 len packet
 1525                          *
 1526                          * In this case, A will not activate the persist timer,
 1527                          * because it chose to send a packet. Unless tcp_output
 1528                          * is called for some other reason (delayed ack timer,
 1529                          * another input packet from B, socket syscall), A will
 1530                          * not send zero window probes.
 1531                          *
 1532                          * So, if you send a 0-length packet, but there is data
 1533                          * in the socket buffer, and neither the rexmt or
 1534                          * persist timer is already set, then activate the
 1535                          * persist timer.
 1536                          */
 1537                         tp->t_rxtshift = 0;
 1538                         tcp_setpersist(tp);
 1539                 }
 1540         } else {
 1541                 /*
 1542                  * Persist case, update snd_max but since we are in
 1543                  * persist mode (no window) we do not update snd_nxt.
 1544                  */
 1545                 int xlen = len;
 1546                 if (flags & TH_SYN)
 1547                         ++xlen;
 1548                 if (flags & TH_FIN) {
 1549                         ++xlen;
 1550                         tp->t_flags |= TF_SENTFIN;
 1551                 }
 1552                 if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max))
 1553                         tp->snd_max = tp->snd_nxt + xlen;
 1554         }
 1555         if ((error == 0) &&
 1556             (TCPS_HAVEESTABLISHED(tp->t_state) &&
 1557              (tp->t_flags & TF_SACK_PERMIT) &&
 1558              tp->rcv_numsacks > 0)) {
 1559                     /* Clean up any DSACK's sent */
 1560                     tcp_clean_dsack_blocks(tp);
 1561         }
 1562         if (error) {
 1563 
 1564                 /*
 1565                  * We know that the packet was lost, so back out the
 1566                  * sequence number advance, if any.
 1567                  *
 1568                  * If the error is EPERM the packet got blocked by the
 1569                  * local firewall.  Normally we should terminate the
 1570                  * connection but the blocking may have been spurious
 1571                  * due to a firewall reconfiguration cycle.  So we treat
 1572                  * it like a packet loss and let the retransmit timer and
 1573                  * timeouts do their work over time.
 1574                  * XXX: It is a POLA question whether calling tcp_drop right
 1575                  * away would be the really correct behavior instead.
 1576                  */
 1577                 if (((tp->t_flags & TF_FORCEDATA) == 0 ||
 1578                     !tcp_timer_active(tp, TT_PERSIST)) &&
 1579                     ((flags & TH_SYN) == 0) &&
 1580                     (error != EPERM)) {
 1581                         if (sack_rxmit) {
 1582                                 p->rxmit -= len;
 1583                                 tp->sackhint.sack_bytes_rexmit -= len;
 1584                                 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
 1585                                     ("sackhint bytes rtx >= 0"));
 1586                         } else
 1587                                 tp->snd_nxt -= len;
 1588                 }
 1589                 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);     /* Check gotos. */
 1590                 switch (error) {
 1591                 case EACCES:
 1592                 case EPERM:
 1593                         tp->t_softerror = error;
 1594                         return (error);
 1595                 case ENOBUFS:
 1596                         TCP_XMIT_TIMER_ASSERT(tp, len, flags);
 1597                         tp->snd_cwnd = tp->t_maxseg;
 1598                         return (0);
 1599                 case EMSGSIZE:
 1600                         /*
 1601                          * For some reason the interface we used initially
 1602                          * to send segments changed to another or lowered
 1603                          * its MTU.
 1604                          * If TSO was active we either got an interface
 1605                          * without TSO capabilits or TSO was turned off.
 1606                          * If we obtained mtu from ip_output() then update
 1607                          * it and try again.
 1608                          */
 1609                         if (tso)
 1610                                 tp->t_flags &= ~TF_TSO;
 1611                         if (mtu != 0) {
 1612                                 tcp_mss_update(tp, -1, mtu, NULL, NULL);
 1613                                 goto again;
 1614                         }
 1615                         return (error);
 1616                 case EHOSTDOWN:
 1617                 case EHOSTUNREACH:
 1618                 case ENETDOWN:
 1619                 case ENETUNREACH:
 1620                         if (TCPS_HAVERCVDSYN(tp->t_state)) {
 1621                                 tp->t_softerror = error;
 1622                                 return (0);
 1623                         }
 1624                         /* FALLTHROUGH */
 1625                 default:
 1626                         return (error);
 1627                 }
 1628         }
 1629         TCPSTAT_INC(tcps_sndtotal);
 1630 
 1631         /*
 1632          * Data sent (as far as we can tell).
 1633          * If this advertises a larger window than any other segment,
 1634          * then remember the size of the advertised window.
 1635          * Any pending ACK has now been sent.
 1636          */
 1637         if (recwin >= 0 && SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv))
 1638                 tp->rcv_adv = tp->rcv_nxt + recwin;
 1639         tp->last_ack_sent = tp->rcv_nxt;
 1640         tp->t_flags &= ~(TF_ACKNOW | TF_DELACK);
 1641         if (tcp_timer_active(tp, TT_DELACK))
 1642                 tcp_timer_activate(tp, TT_DELACK, 0);
 1643 #if 0
 1644         /*
 1645          * This completely breaks TCP if newreno is turned on.  What happens
 1646          * is that if delayed-acks are turned on on the receiver, this code
 1647          * on the transmitter effectively destroys the TCP window, forcing
 1648          * it to four packets (1.5Kx4 = 6K window).
 1649          */
 1650         if (sendalot && --maxburst)
 1651                 goto again;
 1652 #endif
 1653         if (sendalot)
 1654                 goto again;
 1655         return (0);
 1656 }
 1657 
 1658 void
 1659 tcp_setpersist(struct tcpcb *tp)
 1660 {
 1661         int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
 1662         int tt;
 1663 
 1664         tp->t_flags &= ~TF_PREVVALID;
 1665         if (tcp_timer_active(tp, TT_REXMT))
 1666                 panic("tcp_setpersist: retransmit pending");
 1667         /*
 1668          * Start/restart persistence timer.
 1669          */
 1670         TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift],
 1671                       tcp_persmin, tcp_persmax);
 1672         tcp_timer_activate(tp, TT_PERSIST, tt);
 1673         if (tp->t_rxtshift < TCP_MAXRXTSHIFT)
 1674                 tp->t_rxtshift++;
 1675 }
 1676 
 1677 /*
 1678  * Insert TCP options according to the supplied parameters to the place
 1679  * optp in a consistent way.  Can handle unaligned destinations.
 1680  *
 1681  * The order of the option processing is crucial for optimal packing and
 1682  * alignment for the scarce option space.
 1683  *
 1684  * The optimal order for a SYN/SYN-ACK segment is:
 1685  *   MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) +
 1686  *   Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40.
 1687  *
 1688  * The SACK options should be last.  SACK blocks consume 8*n+2 bytes.
 1689  * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks).
 1690  * At minimum we need 10 bytes (to generate 1 SACK block).  If both
 1691  * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present,
 1692  * we only have 10 bytes for SACK options (40 - (12 + 18)).
 1693  */
 1694 int
 1695 tcp_addoptions(struct tcpopt *to, u_char *optp)
 1696 {
 1697         u_int32_t mask, optlen = 0;
 1698 
 1699         for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) {
 1700                 if ((to->to_flags & mask) != mask)
 1701                         continue;
 1702                 if (optlen == TCP_MAXOLEN)
 1703                         break;
 1704                 switch (to->to_flags & mask) {
 1705                 case TOF_MSS:
 1706                         while (optlen % 4) {
 1707                                 optlen += TCPOLEN_NOP;
 1708                                 *optp++ = TCPOPT_NOP;
 1709                         }
 1710                         if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG)
 1711                                 continue;
 1712                         optlen += TCPOLEN_MAXSEG;
 1713                         *optp++ = TCPOPT_MAXSEG;
 1714                         *optp++ = TCPOLEN_MAXSEG;
 1715                         to->to_mss = htons(to->to_mss);
 1716                         bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss));
 1717                         optp += sizeof(to->to_mss);
 1718                         break;
 1719                 case TOF_SCALE:
 1720                         while (!optlen || optlen % 2 != 1) {
 1721                                 optlen += TCPOLEN_NOP;
 1722                                 *optp++ = TCPOPT_NOP;
 1723                         }
 1724                         if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW)
 1725                                 continue;
 1726                         optlen += TCPOLEN_WINDOW;
 1727                         *optp++ = TCPOPT_WINDOW;
 1728                         *optp++ = TCPOLEN_WINDOW;
 1729                         *optp++ = to->to_wscale;
 1730                         break;
 1731                 case TOF_SACKPERM:
 1732                         while (optlen % 2) {
 1733                                 optlen += TCPOLEN_NOP;
 1734                                 *optp++ = TCPOPT_NOP;
 1735                         }
 1736                         if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED)
 1737                                 continue;
 1738                         optlen += TCPOLEN_SACK_PERMITTED;
 1739                         *optp++ = TCPOPT_SACK_PERMITTED;
 1740                         *optp++ = TCPOLEN_SACK_PERMITTED;
 1741                         break;
 1742                 case TOF_TS:
 1743                         while (!optlen || optlen % 4 != 2) {
 1744                                 optlen += TCPOLEN_NOP;
 1745                                 *optp++ = TCPOPT_NOP;
 1746                         }
 1747                         if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP)
 1748                                 continue;
 1749                         optlen += TCPOLEN_TIMESTAMP;
 1750                         *optp++ = TCPOPT_TIMESTAMP;
 1751                         *optp++ = TCPOLEN_TIMESTAMP;
 1752                         to->to_tsval = htonl(to->to_tsval);
 1753                         to->to_tsecr = htonl(to->to_tsecr);
 1754                         bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval));
 1755                         optp += sizeof(to->to_tsval);
 1756                         bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr));
 1757                         optp += sizeof(to->to_tsecr);
 1758                         break;
 1759                 case TOF_SIGNATURE:
 1760                         {
 1761                         int siglen = TCPOLEN_SIGNATURE - 2;
 1762 
 1763                         while (!optlen || optlen % 4 != 2) {
 1764                                 optlen += TCPOLEN_NOP;
 1765                                 *optp++ = TCPOPT_NOP;
 1766                         }
 1767                         if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE) {
 1768                                 to->to_flags &= ~TOF_SIGNATURE;
 1769                                 continue;
 1770                         }
 1771                         optlen += TCPOLEN_SIGNATURE;
 1772                         *optp++ = TCPOPT_SIGNATURE;
 1773                         *optp++ = TCPOLEN_SIGNATURE;
 1774                         to->to_signature = optp;
 1775                         while (siglen--)
 1776                                  *optp++ = 0;
 1777                         break;
 1778                         }
 1779                 case TOF_SACK:
 1780                         {
 1781                         int sackblks = 0;
 1782                         struct sackblk *sack = (struct sackblk *)to->to_sacks;
 1783                         tcp_seq sack_seq;
 1784 
 1785                         while (!optlen || optlen % 4 != 2) {
 1786                                 optlen += TCPOLEN_NOP;
 1787                                 *optp++ = TCPOPT_NOP;
 1788                         }
 1789                         if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK)
 1790                                 continue;
 1791                         optlen += TCPOLEN_SACKHDR;
 1792                         *optp++ = TCPOPT_SACK;
 1793                         sackblks = min(to->to_nsacks,
 1794                                         (TCP_MAXOLEN - optlen) / TCPOLEN_SACK);
 1795                         *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK;
 1796                         while (sackblks--) {
 1797                                 sack_seq = htonl(sack->start);
 1798                                 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
 1799                                 optp += sizeof(sack_seq);
 1800                                 sack_seq = htonl(sack->end);
 1801                                 bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq));
 1802                                 optp += sizeof(sack_seq);
 1803                                 optlen += TCPOLEN_SACK;
 1804                                 sack++;
 1805                         }
 1806                         TCPSTAT_INC(tcps_sack_send_blocks);
 1807                         break;
 1808                         }
 1809 #ifdef TCP_RFC7413
 1810                 case TOF_FASTOPEN:
 1811                         {
 1812                         int total_len;
 1813 
 1814                         /* XXX is there any point to aligning this option? */
 1815                         total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len;
 1816                         if (TCP_MAXOLEN - optlen < total_len)
 1817                                 continue;
 1818                         *optp++ = TCPOPT_FAST_OPEN;
 1819                         *optp++ = total_len;
 1820                         if (to->to_tfo_len > 0) {
 1821                                 bcopy(to->to_tfo_cookie, optp, to->to_tfo_len);
 1822                                 optp += to->to_tfo_len;
 1823                         }
 1824                         optlen += total_len;
 1825                         break;
 1826                         }
 1827 #endif
 1828                 default:
 1829                         panic("%s: unknown TCP option type", __func__);
 1830                         break;
 1831                 }
 1832         }
 1833 
 1834         /* Terminate and pad TCP options to a 4 byte boundary. */
 1835         if (optlen % 4) {
 1836                 optlen += TCPOLEN_EOL;
 1837                 *optp++ = TCPOPT_EOL;
 1838         }
 1839         /*
 1840          * According to RFC 793 (STD0007):
 1841          *   "The content of the header beyond the End-of-Option option
 1842          *    must be header padding (i.e., zero)."
 1843          *   and later: "The padding is composed of zeros."
 1844          */
 1845         while (optlen % 4) {
 1846                 optlen += TCPOLEN_PAD;
 1847                 *optp++ = TCPOPT_PAD;
 1848         }
 1849 
 1850         KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__));
 1851         return (optlen);
 1852 }

Cache object: cb84b7248d82f4b1bdfd660070b4ae29


[ 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.