Add qemu 2.4.0
[kvmfornfv.git] / qemu / slirp / tcp_timer.c
diff --git a/qemu/slirp/tcp_timer.c b/qemu/slirp/tcp_timer.c
new file mode 100644 (file)
index 0000000..6c5bb11
--- /dev/null
@@ -0,0 +1,292 @@
+/*
+ * Copyright (c) 1982, 1986, 1988, 1990, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)tcp_timer.c 8.1 (Berkeley) 6/10/93
+ * tcp_timer.c,v 1.2 1994/08/02 07:49:10 davidg Exp
+ */
+
+#include <slirp.h>
+
+static struct tcpcb *tcp_timers(register struct tcpcb *tp, int timer);
+
+/*
+ * Fast timeout routine for processing delayed acks
+ */
+void
+tcp_fasttimo(Slirp *slirp)
+{
+       register struct socket *so;
+       register struct tcpcb *tp;
+
+       DEBUG_CALL("tcp_fasttimo");
+
+       so = slirp->tcb.so_next;
+       if (so)
+       for (; so != &slirp->tcb; so = so->so_next)
+               if ((tp = (struct tcpcb *)so->so_tcpcb) &&
+                   (tp->t_flags & TF_DELACK)) {
+                       tp->t_flags &= ~TF_DELACK;
+                       tp->t_flags |= TF_ACKNOW;
+                       (void) tcp_output(tp);
+               }
+}
+
+/*
+ * Tcp protocol timeout routine called every 500 ms.
+ * Updates the timers in all active tcb's and
+ * causes finite state machine actions if timers expire.
+ */
+void
+tcp_slowtimo(Slirp *slirp)
+{
+       register struct socket *ip, *ipnxt;
+       register struct tcpcb *tp;
+       register int i;
+
+       DEBUG_CALL("tcp_slowtimo");
+
+       /*
+        * Search through tcb's and update active timers.
+        */
+       ip = slirp->tcb.so_next;
+        if (ip == NULL) {
+            return;
+        }
+       for (; ip != &slirp->tcb; ip = ipnxt) {
+               ipnxt = ip->so_next;
+               tp = sototcpcb(ip);
+                if (tp == NULL) {
+                        continue;
+                }
+               for (i = 0; i < TCPT_NTIMERS; i++) {
+                       if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
+                               tcp_timers(tp,i);
+                               if (ipnxt->so_prev != ip)
+                                       goto tpgone;
+                       }
+               }
+               tp->t_idle++;
+               if (tp->t_rtt)
+                  tp->t_rtt++;
+tpgone:
+               ;
+       }
+       slirp->tcp_iss += TCP_ISSINCR/PR_SLOWHZ;        /* increment iss */
+       slirp->tcp_now++;                               /* for timestamps */
+}
+
+/*
+ * Cancel all timers for TCP tp.
+ */
+void
+tcp_canceltimers(struct tcpcb *tp)
+{
+       register int i;
+
+       for (i = 0; i < TCPT_NTIMERS; i++)
+               tp->t_timer[i] = 0;
+}
+
+const int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
+   { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
+
+/*
+ * TCP timer processing.
+ */
+static struct tcpcb *
+tcp_timers(register struct tcpcb *tp, int timer)
+{
+       register int rexmt;
+
+       DEBUG_CALL("tcp_timers");
+
+       switch (timer) {
+
+       /*
+        * 2 MSL timeout in shutdown went off.  If we're closed but
+        * still waiting for peer to close and connection has been idle
+        * too long, or if 2MSL time is up from TIME_WAIT, delete connection
+        * control block.  Otherwise, check again in a bit.
+        */
+       case TCPT_2MSL:
+               if (tp->t_state != TCPS_TIME_WAIT &&
+                   tp->t_idle <= TCP_MAXIDLE)
+                       tp->t_timer[TCPT_2MSL] = TCPTV_KEEPINTVL;
+               else
+                       tp = tcp_close(tp);
+               break;
+
+       /*
+        * Retransmission timer went off.  Message has not
+        * been acked within retransmit interval.  Back off
+        * to a longer retransmit interval and retransmit one segment.
+        */
+       case TCPT_REXMT:
+
+               /*
+                * XXXXX If a packet has timed out, then remove all the queued
+                * packets for that session.
+                */
+
+               if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
+                       /*
+                        * This is a hack to suit our terminal server here at the uni of canberra
+                        * since they have trouble with zeroes... It usually lets them through
+                        * unharmed, but under some conditions, it'll eat the zeros.  If we
+                        * keep retransmitting it, it'll keep eating the zeroes, so we keep
+                        * retransmitting, and eventually the connection dies...
+                        * (this only happens on incoming data)
+                        *
+                        * So, if we were gonna drop the connection from too many retransmits,
+                        * don't... instead halve the t_maxseg, which might break up the NULLs and
+                        * let them through
+                        *
+                        * *sigh*
+                        */
+
+                       tp->t_maxseg >>= 1;
+                       if (tp->t_maxseg < 32) {
+                               /*
+                                * We tried our best, now the connection must die!
+                                */
+                               tp->t_rxtshift = TCP_MAXRXTSHIFT;
+                               tp = tcp_drop(tp, tp->t_softerror);
+                               /* tp->t_softerror : ETIMEDOUT); */ /* XXX */
+                               return (tp); /* XXX */
+                       }
+
+                       /*
+                        * Set rxtshift to 6, which is still at the maximum
+                        * backoff time
+                        */
+                       tp->t_rxtshift = 6;
+               }
+               rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
+               TCPT_RANGESET(tp->t_rxtcur, rexmt,
+                   (short)tp->t_rttmin, TCPTV_REXMTMAX); /* XXX */
+               tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
+               /*
+                * If losing, let the lower level know and try for
+                * a better route.  Also, if we backed off this far,
+                * our srtt estimate is probably bogus.  Clobber it
+                * so we'll take the next rtt measurement as our srtt;
+                * move the current srtt into rttvar to keep the current
+                * retransmit times until then.
+                */
+               if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
+                       tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
+                       tp->t_srtt = 0;
+               }
+               tp->snd_nxt = tp->snd_una;
+               /*
+                * If timing a segment in this window, stop the timer.
+                */
+               tp->t_rtt = 0;
+               /*
+                * Close the congestion window down to one segment
+                * (we'll open it by one segment for each ack we get).
+                * Since we probably have a window's worth of unacked
+                * data accumulated, this "slow start" keeps us from
+                * dumping all that data as back-to-back packets (which
+                * might overwhelm an intermediate gateway).
+                *
+                * There are two phases to the opening: Initially we
+                * open by one mss on each ack.  This makes the window
+                * size increase exponentially with time.  If the
+                * window is larger than the path can handle, this
+                * exponential growth results in dropped packet(s)
+                * almost immediately.  To get more time between
+                * drops but still "push" the network to take advantage
+                * of improving conditions, we switch from exponential
+                * to linear window opening at some threshold size.
+                * For a threshold, we use half the current window
+                * size, truncated to a multiple of the mss.
+                *
+                * (the minimum cwnd that will give us exponential
+                * growth is 2 mss.  We don't allow the threshold
+                * to go below this.)
+                */
+               {
+               u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
+               if (win < 2)
+                       win = 2;
+               tp->snd_cwnd = tp->t_maxseg;
+               tp->snd_ssthresh = win * tp->t_maxseg;
+               tp->t_dupacks = 0;
+               }
+               (void) tcp_output(tp);
+               break;
+
+       /*
+        * Persistence timer into zero window.
+        * Force a byte to be output, if possible.
+        */
+       case TCPT_PERSIST:
+               tcp_setpersist(tp);
+               tp->t_force = 1;
+               (void) tcp_output(tp);
+               tp->t_force = 0;
+               break;
+
+       /*
+        * Keep-alive timer went off; send something
+        * or drop connection if idle for too long.
+        */
+       case TCPT_KEEP:
+               if (tp->t_state < TCPS_ESTABLISHED)
+                       goto dropit;
+
+               if ((SO_OPTIONS) && tp->t_state <= TCPS_CLOSE_WAIT) {
+                       if (tp->t_idle >= TCPTV_KEEP_IDLE + TCP_MAXIDLE)
+                               goto dropit;
+                       /*
+                        * Send a packet designed to force a response
+                        * if the peer is up and reachable:
+                        * either an ACK if the connection is still alive,
+                        * or an RST if the peer has closed the connection
+                        * due to timeout or reboot.
+                        * Using sequence number tp->snd_una-1
+                        * causes the transmitted zero-length segment
+                        * to lie outside the receive window;
+                        * by the protocol spec, this requires the
+                        * correspondent TCP to respond.
+                        */
+                       tcp_respond(tp, &tp->t_template, (struct mbuf *)NULL,
+                           tp->rcv_nxt, tp->snd_una - 1, 0);
+                       tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
+               } else
+                       tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
+               break;
+
+       dropit:
+               tp = tcp_drop(tp, 0);
+               break;
+       }
+
+       return (tp);
+}