]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/netinet/tcp_timer.c
xnu-792.24.17.tar.gz
[apple/xnu.git] / bsd / netinet / tcp_timer.c
index 3819a60cfb762913c35f76162573e267ebbf40bb..effeffde5292a482eae0a71a37509d71d20f4307 100644 (file)
  * SUCH DAMAGE.
  *
  *     @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
+ * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.11 2001/08/22 00:59:12 silby Exp $
  */
 
-#if ISFB31
-#include "opt_compat.h"
-#include "opt_tcpdebug.h"
-#endif
 
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/kernel.h>
+#include <sys/mbuf.h>
 #include <sys/sysctl.h>
 #include <sys/socket.h>
 #include <sys/socketvar.h>
 #include <sys/protosw.h>
+#include <kern/locks.h>
 
 #include <kern/cpu_number.h>   /* before tcp_seq.h, for tcp_random18() */
 
 
 #include <netinet/in.h>
 #include <netinet/in_systm.h>
-#include <netinet/ip.h>
-#include <netinet/ip6.h>
 #include <netinet/in_pcb.h>
+#if INET6
+#include <netinet6/in6_pcb.h>
+#endif
 #include <netinet/ip_var.h>
 #include <netinet/tcp.h>
 #include <netinet/tcp_fsm.h>
 #define DBG_FNC_TCP_FAST       NETDBG_CODE(DBG_NETTCP, (5 << 8))
 #define DBG_FNC_TCP_SLOW       NETDBG_CODE(DBG_NETTCP, (5 << 8) | 1)
 
+/*
+ * NOTE - WARNING
+ *
+ *
+ * 
+ *
+ */
+static int
+sysctl_msec_to_ticks SYSCTL_HANDLER_ARGS
+{
+       int error, s, tt;
+
+       tt = *(int *)oidp->oid_arg1;
+       s = tt * 1000 / hz;
+
+       error = sysctl_handle_int(oidp, &s, 0, req);
+       if (error || !req->newptr)
+               return (error);
+
+       tt = s * hz / 1000;
+       if (tt < 1)
+               return (EINVAL);
+
+       *(int *)oidp->oid_arg1 = tt;
+        return (0);
+}
+
+int    tcp_keepinit;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
+    &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
 
-int    tcp_keepinit = TCPTV_KEEP_INIT;
-SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit,
-       CTLFLAG_RW, &tcp_keepinit , 0, "");
+int    tcp_keepidle;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
+    &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
 
-int    tcp_keepidle = TCPTV_KEEP_IDLE;
-SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
-       CTLFLAG_RW, &tcp_keepidle , 0, "");
+int    tcp_keepintvl;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
+    &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
 
-static int     tcp_keepintvl = TCPTV_KEEPINTVL;
-SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
-       CTLFLAG_RW, &tcp_keepintvl , 0, "");
+int    tcp_delacktime;
+SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
+    CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
+    "Time before a delayed ACK is sent");
+int    tcp_msl;
+SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
+    &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
 
 static int     always_keepalive = 0;
-SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive,
-       CTLFLAG_RW, &always_keepalive , 0, "");
+SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW, 
+    &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
 
 static int     tcp_keepcnt = TCPTV_KEEPCNT;
        /* max idle probes */
-static int     tcp_maxpersistidle = TCPTV_KEEP_IDLE;
+int    tcp_maxpersistidle;
        /* max idle time in persist */
 int    tcp_maxidle;
 
-
-
 struct inpcbhead       time_wait_slots[N_TIME_WAIT_SLOTS];
 int            cur_tw_slot = 0;
 
 u_long         *delack_bitmask;
-u_long         current_active_connections = 0;
-u_long         last_active_conn_count = 0;
 
 
-void   add_to_time_wait(tp) 
+void   add_to_time_wait_locked(tp) 
        struct tcpcb    *tp;
 {
        int             tw_slot;
 
+       /* pcb list should be locked when we get here */        
+#if 0
+       lck_mtx_assert(tp->t_inpcb->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
+#endif
+
        LIST_REMOVE(tp->t_inpcb, inp_list);
 
        if (tp->t_timer[TCPT_2MSL] == 0) 
            tp->t_timer[TCPT_2MSL] = 1;
 
-       tp->t_idle += tp->t_timer[TCPT_2MSL] & (N_TIME_WAIT_SLOTS - 1);
+       tp->t_rcvtime += tp->t_timer[TCPT_2MSL] & (N_TIME_WAIT_SLOTS - 1);
        tw_slot = (tp->t_timer[TCPT_2MSL] & (N_TIME_WAIT_SLOTS - 1)) + cur_tw_slot; 
        if (tw_slot >= N_TIME_WAIT_SLOTS)
            tw_slot -= N_TIME_WAIT_SLOTS;
@@ -142,6 +178,19 @@ void       add_to_time_wait(tp)
        LIST_INSERT_HEAD(&time_wait_slots[tw_slot], tp->t_inpcb, inp_list);
 }
 
+void   add_to_time_wait(tp) 
+       struct tcpcb    *tp;
+{
+       struct inpcbinfo *pcbinfo               = &tcbinfo;
+       
+       if (!lck_rw_try_lock_exclusive(pcbinfo->mtx)) {
+               tcp_unlock(tp->t_inpcb->inp_socket, 0, 0);
+               lck_rw_lock_exclusive(pcbinfo->mtx);
+               tcp_lock(tp->t_inpcb->inp_socket, 0, 0);
+       }
+       add_to_time_wait_locked(tp);
+       lck_rw_done(pcbinfo->mtx);
+}
 
 
 
@@ -152,66 +201,46 @@ void      add_to_time_wait(tp)
 void
 tcp_fasttimo()
 {
-    register struct inpcb *inp;
+    struct inpcb *inp, *inpnxt;
     register struct tcpcb *tp;
 
 
-    register u_long                    i,j;
-    register u_long                    temp_mask;
-    register u_long                    elem_base = 0;
-    struct inpcbhead                   *head;
-    int s = splnet();
+    struct inpcbinfo *pcbinfo  = &tcbinfo;
 
-    static
-    int delack_checked = 0;
+    int delack_checked = 0, delack_done = 0;
 
     KERNEL_DEBUG(DBG_FNC_TCP_FAST | DBG_FUNC_START, 0,0,0,0,0);
 
-    if (!tcp_delack_enabled
+    if (tcp_delack_enabled == 0
        return;
 
-    if ((current_active_connections > DELACK_BITMASK_THRESH) &&
-       (last_active_conn_count > DELACK_BITMASK_THRESH)) {
-       for (i=0; i < (tcbinfo.hashsize / 32); i++) {
-           if (delack_bitmask[i]) {
-               temp_mask = 1;
-               for (j=0; j < 32; j++) {
-                   if (temp_mask & delack_bitmask[i]) {
-                       head = &tcbinfo.hashbase[elem_base + j];
-                       for (inp=head->lh_first; inp != 0; inp = inp->inp_hash.le_next) {
-                           delack_checked++;
-                           if ((tp = (struct tcpcb *)inp->inp_ppcb) && (tp->t_flags & TF_DELACK)) {
-                               tp->t_flags &= ~TF_DELACK;
-                               tp->t_flags |= TF_ACKNOW;
-                               tcpstat.tcps_delack++;
-                               (void) tcp_output(tp);
-                           }
-                       }
-                   }
-                   temp_mask <<=  1;
-               }
-               delack_bitmask[i] = 0;
-           }
-           elem_base += 32;
-       }
-    }
-    else 
-    {
-       for (inp = tcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
-           if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
-               (tp->t_flags & TF_DELACK)) {
-               tp->t_flags &= ~TF_DELACK;
-               tp->t_flags |= TF_ACKNOW;
-               tcpstat.tcps_delack++;
-               (void) tcp_output(tp);
-           }
-       }
-    }
+    lck_rw_lock_shared(pcbinfo->mtx);
 
-    last_active_conn_count = current_active_connections;
-    KERNEL_DEBUG(DBG_FNC_TCP_FAST | DBG_FUNC_END, delack_checked,tcpstat.tcps_delack,0,0,0);
-    splx(s);
+    /* Walk the list of valid tcpcbs and send ACKS on the ones with DELACK bit set */
 
+    for (inp = tcb.lh_first; inp != NULL; inp = inpnxt) {
+       inpnxt = inp->inp_list.le_next;
+       /* NOTE: it's OK to check the tp because the pcb can't be removed while we hold pcbinfo->mtx) */
+       if ((tp = (struct tcpcb *)inp->inp_ppcb) && (tp->t_flags & TF_DELACK)) {
+               if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) 
+                       continue;
+               tcp_lock(inp->inp_socket, 1, 0);
+               if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
+                       tcp_unlock(inp->inp_socket, 1, 0);
+                       continue;
+               }
+               if (tp->t_flags & TF_DELACK) {
+                       delack_done++;
+                       tp->t_flags &= ~TF_DELACK;
+                       tp->t_flags |= TF_ACKNOW;
+                       tcpstat.tcps_delack++;
+                       (void) tcp_output(tp);
+               }
+               tcp_unlock(inp->inp_socket, 1, 0);
+       }
+    }
+    KERNEL_DEBUG(DBG_FNC_TCP_FAST | DBG_FUNC_END, delack_checked, delack_done, tcpstat.tcps_delack,0,0);
+    lck_rw_done(pcbinfo->mtx);
 }
 
 /*
@@ -222,35 +251,50 @@ tcp_fasttimo()
 void
 tcp_slowtimo()
 {
-       register struct inpcb *ip, *ipnxt;
-       register struct tcpcb *tp;
-       register int i;
-       int s;
+       struct inpcb *inp, *inpnxt;
+       struct tcpcb *tp;
+       struct socket *so;
+       int i;
 #if TCPDEBUG
        int ostate;
 #endif
 #if KDEBUG
        static int tws_checked;
 #endif
+       struct inpcbinfo *pcbinfo               = &tcbinfo;
 
        KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_START, 0,0,0,0,0);
-       s = splnet();
 
        tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
 
-       ip = tcb.lh_first;
-       if (ip == NULL) {
-               splx(s);
-               return;
-       }
+       lck_rw_lock_shared(pcbinfo->mtx);
+
        /*
         * Search through tcb's and update active timers.
         */
-       for (; ip != NULL; ip = ipnxt) {
-               ipnxt = ip->inp_list.le_next;
-               tp = intotcpcb(ip);
-               if (tp == 0 || tp->t_state == TCPS_LISTEN)
+       for (inp = tcb.lh_first; inp != NULL; inp = inpnxt) {
+               inpnxt = inp->inp_list.le_next;
+
+               so = inp->inp_socket;
+
+               if (so == &tcbinfo.nat_dummy_socket)
+                               continue;
+
+               if (in_pcb_checkstate(inp, WNT_ACQUIRE,0) == WNT_STOPUSING) 
                        continue;
+
+               tcp_lock(so, 1, 0);
+
+               if ((in_pcb_checkstate(inp, WNT_RELEASE,1) == WNT_STOPUSING)  && so->so_usecount == 1) {
+                       tcp_unlock(so, 1, 0);
+                       continue;
+               }
+               tp = intotcpcb(inp);
+               if (tp == 0 || tp->t_state == TCPS_LISTEN) {
+                       tcp_unlock(so, 1, 0);
+                       continue; 
+               }
+
                for (i = 0; i < TCPT_NTIMERS; i++) {
                        if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
 #if TCPDEBUG
@@ -269,12 +313,12 @@ tcp_slowtimo()
 #endif
                        }
                }
-               tp->t_idle++;
-               tp->t_duration++;
-               if (tp->t_rtt)
-                       tp->t_rtt++;
+               tp->t_rcvtime++;
+               tp->t_starttime++;
+               if (tp->t_rtttime)
+                       tp->t_rtttime++;        
 tpgone:
-               ;
+               tcp_unlock(so, 1, 0);
        }
 
 #if KDEBUG
@@ -286,34 +330,106 @@ tpgone:
         * Process the items in the current time-wait slot
         */
 
-       for (ip = time_wait_slots[cur_tw_slot].lh_first; ip; ip = ipnxt)
+       for (inp = time_wait_slots[cur_tw_slot].lh_first; inp; inp = inpnxt)
        {
+               inpnxt = inp->inp_list.le_next;
 #if KDEBUG
                tws_checked++;
 #endif
-               ipnxt = ip->inp_list.le_next;
-               tp = intotcpcb(ip);
+
+               if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) 
+                       continue;
+
+               tcp_lock(inp->inp_socket, 1, 0);
+
+               if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) 
+                       goto twunlock;
+
+               tp = intotcpcb(inp);
+               if (tp == NULL) { /* tp already closed, remove from list */
+#if TEMPDEBUG
+                       printf("tcp_slowtimo: tp is null in time-wait slot!\n");
+#endif
+                       goto twunlock;
+               }
                if (tp->t_timer[TCPT_2MSL] >= N_TIME_WAIT_SLOTS) {
                    tp->t_timer[TCPT_2MSL] -= N_TIME_WAIT_SLOTS;
-                   tp->t_idle += N_TIME_WAIT_SLOTS;
+                   tp->t_rcvtime += N_TIME_WAIT_SLOTS;
                }
                else
                    tp->t_timer[TCPT_2MSL] = 0;
 
-               if (tp->t_timer[TCPT_2MSL] == 0) 
-                   tp = tcp_timers(tp, TCPT_2MSL);
+               if (tp->t_timer[TCPT_2MSL] == 0)  
+                   tp = tcp_timers(tp, TCPT_2MSL);     /* tp can be returned null if tcp_close is called */
+twunlock:
+               tcp_unlock(inp->inp_socket, 1, 0);
        }
 
-       if (++cur_tw_slot >= N_TIME_WAIT_SLOTS)
-               cur_tw_slot = 0;
+       if (lck_rw_lock_shared_to_exclusive(pcbinfo->mtx) != 0)
+               lck_rw_lock_exclusive(pcbinfo->mtx);    /* Upgrade failed, lost lock no take it again exclusive */
+
 
-       tcp_iss += TCP_ISSINCR/PR_SLOWHZ;               /* increment iss */
-#if TCP_COMPAT_42
-       if ((int)tcp_iss < 0)
-               tcp_iss = TCP_ISSINCR;                  /* XXX */
+       for (inp = tcb.lh_first; inp != NULL; inp = inpnxt) {
+               inpnxt = inp->inp_list.le_next;
+               /* Ignore nat/SharedIP dummy pcbs */
+               if (inp->inp_socket == &tcbinfo.nat_dummy_socket)
+                               continue;
+
+               if (inp->inp_wantcnt != WNT_STOPUSING) 
+                       continue;
+
+               so = inp->inp_socket;
+               if (!lck_mtx_try_lock(inp->inpcb_mtx)) {/* skip if in use */
+#if TEMPDEBUG
+                       printf("tcp_slowtimo so=%x STOPUSING but locked...\n", so);
 #endif
-       tcp_now++;                                      /* for timestamps */
-       splx(s);
+                       continue;
+               }
+
+               if (so->so_usecount == 0) 
+                       in_pcbdispose(inp);
+               else {
+                       tp = intotcpcb(inp);
+                       /* Check for embryonic socket stuck on listener queue (4023660) */
+                       if ((so->so_usecount == 1) && (tp->t_state == TCPS_CLOSED) &&
+                           (so->so_head != NULL) && (so->so_state & SS_INCOMP)) {
+                               so->so_usecount--; 
+                               in_pcbdispose(inp);
+                       } else
+                               lck_mtx_unlock(inp->inpcb_mtx);
+               }
+       }
+
+       /* Now cleanup the time wait ones */
+       for (inp = time_wait_slots[cur_tw_slot].lh_first; inp; inp = inpnxt)
+       {
+               inpnxt = inp->inp_list.le_next;
+
+               if (inp->inp_wantcnt != WNT_STOPUSING) 
+                       continue;
+
+               so = inp->inp_socket;
+               if (!lck_mtx_try_lock(inp->inpcb_mtx)) /* skip if in use */
+                       continue;
+               if (so->so_usecount == 0)  
+                       in_pcbdispose(inp);
+               else  {
+                       tp = intotcpcb(inp);
+                       /* Check for embryonic socket stuck on listener queue (4023660) */
+                       if ((so->so_usecount == 1) && (tp->t_state == TCPS_CLOSED) &&
+                           (so->so_head != NULL) && (so->so_state & SS_INCOMP)) {
+                               so->so_usecount--; 
+                               in_pcbdispose(inp);
+                       } else
+                               lck_mtx_unlock(inp->inpcb_mtx);
+               }
+       }
+
+       tcp_now++;
+       if (++cur_tw_slot >= N_TIME_WAIT_SLOTS)
+               cur_tw_slot = 0;
+       
+       lck_rw_done(pcbinfo->mtx);
        KERNEL_DEBUG(DBG_FNC_TCP_SLOW | DBG_FUNC_END, tws_checked, cur_tw_slot,0,0,0);
 }
 
@@ -330,6 +446,9 @@ tcp_canceltimers(tp)
                tp->t_timer[i] = 0;
 }
 
+int    tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
+    { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
+
 int    tcp_backoff[TCP_MAXRXTSHIFT + 1] =
     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
 
@@ -345,10 +464,18 @@ tcp_timers(tp, timer)
 {
        register int rexmt;
        struct socket *so_tmp;
+       struct tcptemp *t_template;
+
+#if TCPDEBUG
+       int ostate;
+#endif
+
 #if INET6
        int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV4) == 0;
 #endif /* INET6 */
 
+       so_tmp = tp->t_inpcb->inp_socket;
+
        switch (timer) {
 
        /*
@@ -358,13 +485,16 @@ tcp_timers(tp, timer)
         * control block.  Otherwise, check again in a bit.
         */
        case TCPT_2MSL:
+               tcp_free_sackholes(tp);
                if (tp->t_state != TCPS_TIME_WAIT &&
-                   tp->t_idle <= tcp_maxidle) {
-                       tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
-                       add_to_time_wait(tp);
+                   tp->t_rcvtime <= tcp_maxidle) {
+                       tp->t_timer[TCPT_2MSL] = (unsigned long)tcp_keepintvl;
+                       add_to_time_wait_locked(tp);
                }
-               else
+               else {
                        tp = tcp_close(tp);
+                       return(tp);
+               }
                break;
 
        /*
@@ -373,20 +503,53 @@ tcp_timers(tp, timer)
         * to a longer retransmit interval and retransmit one segment.
         */
        case TCPT_REXMT:
+               tcp_free_sackholes(tp);
                if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
                        tp->t_rxtshift = TCP_MAXRXTSHIFT;
                        tcpstat.tcps_timeoutdrop++;
-                       so_tmp = tp->t_inpcb->inp_socket;
                        tp = tcp_drop(tp, tp->t_softerror ?
                            tp->t_softerror : ETIMEDOUT);
                        postevent(so_tmp, 0, EV_TIMEOUT);                       
                        break;
                }
+
+               if (tp->t_rxtshift == 1) {
+                       /*
+                        * first retransmit; record ssthresh and cwnd so they can
+                        * be recovered if this turns out to be a "bad" retransmit.
+                        * A retransmit is considered "bad" if an ACK for this 
+                        * segment is received within RTT/2 interval; the assumption
+                        * here is that the ACK was already in flight.  See 
+                        * "On Estimating End-to-End Network Path Properties" by
+                        * Allman and Paxson for more details.
+                        */
+                       tp->snd_cwnd_prev = tp->snd_cwnd;
+                       tp->snd_ssthresh_prev = tp->snd_ssthresh;
+                       tp->snd_recover_prev = tp->snd_recover;
+                       if (IN_FASTRECOVERY(tp))
+                                 tp->t_flags |= TF_WASFRECOVERY;
+                       else
+                                 tp->t_flags &= ~TF_WASFRECOVERY;
+                       tp->t_badrxtwin = tcp_now + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
+               }
                tcpstat.tcps_rexmttimeo++;
-               rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
+               if (tp->t_state == TCPS_SYN_SENT)
+                       rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
+               else
+                       rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
                TCPT_RANGESET(tp->t_rxtcur, rexmt,
-                   tp->t_rttmin, TCPTV_REXMTMAX);
+                       tp->t_rttmin, TCPTV_REXMTMAX);
                tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
+
+               /*
+                * Disable rfc1323 and rfc1644 if we havn't got any response to
+                * our third SYN to work-around some broken terminal servers 
+                * (most of which have hopefully been retired) that have bad VJ 
+                * header compression code which trashes TCP segments containing 
+                * unknown-to-them TCP options.
+                */
+               if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
+                               tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
                /*
                 * If losing, let the lower level know and try for
                 * a better route.  Also, if we backed off this far,
@@ -406,6 +569,11 @@ tcp_timers(tp, timer)
                        tp->t_srtt = 0;
                }
                tp->snd_nxt = tp->snd_una;
+               /*
+                * Note:  We overload snd_recover to function also as the
+                * snd_last variable described in RFC 2582
+                */
+               tp->snd_recover = tp->snd_max;
                /*
                 * Force a segment to be sent.
                 */
@@ -413,7 +581,7 @@ tcp_timers(tp, timer)
                /*
                 * If timing a segment in this window, stop the timer.
                 */
-               tp->t_rtt = 0;
+               tp->t_rtttime = 0;
                /*
                 * Close the congestion window down to one segment
                 * (we'll open it by one segment for each ack we get).
@@ -446,6 +614,7 @@ tcp_timers(tp, timer)
                tp->snd_ssthresh = win * tp->t_maxseg;
                tp->t_dupacks = 0;
                }
+               EXIT_FASTRECOVERY(tp);
                (void) tcp_output(tp);
                break;
 
@@ -463,8 +632,8 @@ tcp_timers(tp, timer)
                 * backoff that we would use if retransmitting.
                 */
                if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
-                   (tp->t_idle >= tcp_maxpersistidle ||
-                   tp->t_idle >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
+                   (tp->t_rcvtime >= tcp_maxpersistidle ||
+                   tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
                        tcpstat.tcps_persistdrop++;
                        so_tmp = tp->t_inpcb->inp_socket;
                        tp = tcp_drop(tp, ETIMEDOUT);
@@ -487,8 +656,8 @@ tcp_timers(tp, timer)
                        goto dropit;
                if ((always_keepalive ||
                    tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
-                   tp->t_state <= TCPS_CLOSING) {
-                       if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
+                   tp->t_state <= TCPS_CLOSING || tp->t_state == TCPS_FIN_WAIT_2) {
+                       if (tp->t_rcvtime >= TCP_KEEPIDLE(tp) + (unsigned long)tcp_maxidle)
                                goto dropit;
                        /*
                         * Send a packet designed to force a response
@@ -503,44 +672,25 @@ tcp_timers(tp, timer)
                         * correspondent TCP to respond.
                         */
                        tcpstat.tcps_keepprobe++;
-#if TCP_COMPAT_42
-                       /*
-                        * The keepalive packet must have nonzero length
-                        * to get a 4.2 host to respond.
-                        */
-#if INET6
-                       if (isipv6)
-                               tcp_respond(tp, (void *)&tp->t_template->tt_i6,
-                                           &tp->t_template->tt_t,
-                                           (struct mbuf *)NULL,
-                                           tp->rcv_nxt - 1, tp->snd_una - 1,
-                                           0, isipv6);
-                       else
-#endif /* INET6 */
-                       tcp_respond(tp, (void *)&tp->t_template->tt_i,
-                                   &tp->t_template->tt_t, (struct mbuf *)NULL,
-                                   tp->rcv_nxt - 1, tp->snd_una - 1, 0,
-                                   isipv6);
-#else
-#if INET6
-                       if (isipv6)
-                               tcp_respond(tp, (void *)&tp->t_template->tt_i6,
-                                           &tp->t_template->tt_t,
-                                           (struct mbuf *)NULL, tp->rcv_nxt,
-                                           tp->snd_una - 1, 0, isipv6);
-                       else
-#endif /* INET6 */
-                       tcp_respond(tp, (void *)&tp->t_template->tt_i,
-                                   &tp->t_template->tt_t, (struct mbuf *)NULL,
-                                   tp->rcv_nxt, tp->snd_una - 1, 0, isipv6);
-#endif
+                       t_template = tcp_maketemplate(tp);
+                       if (t_template) {
+                               tcp_respond(tp, t_template->tt_ipgen,
+                                   &t_template->tt_t, (struct mbuf *)NULL,
+                                   tp->rcv_nxt, tp->snd_una - 1, 0);
+                               (void) m_free(dtom(t_template));
+                       }
                        tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
                } else
-                       tp->t_timer[TCPT_KEEP] = tcp_keepidle;
+                       tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp);
                break;
+
+#if TCPDEBUG
+       if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
+               tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
+                         PRU_SLOWTIMO);
+#endif
        dropit:
                tcpstat.tcps_keepdrops++;
-               so_tmp = tp->t_inpcb->inp_socket;
                tp = tcp_drop(tp, ETIMEDOUT);
                postevent(so_tmp, 0, EV_TIMEOUT);
                break;