void
tcp_timer_rexmt(void *arg)
{
struct tcpcb *tp = arg;
uint32_t rto;
#ifdef TCP_DEBUG
struct socket *so = NULL;
short ostate;
#endif
mutex_enter(softnet_lock);
if ((tp->t_flags & TF_DEAD) != 0) {
mutex_exit(softnet_lock);
return;
}
if (!callout_expired(&tp->t_timer[TCPT_REXMT])) {
mutex_exit(softnet_lock);
return;
}
KERNEL_LOCK(1, NULL);
if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_ourmss))) {
extern struct sockaddr_in icmpsrc;
struct icmp icmp;
tp->t_flags &= ~TF_PMTUD_PEND;
/* XXX create fake icmp message with relevant entries */
icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
icmpsrc.sin_addr = tp->t_inpcb->inp_faddr;
icmp_mtudisc(&icmp, icmpsrc.sin_addr);
/*
* Notify all connections to the same peer about
* new mss and trigger retransmit.
*/
in_pcbnotifyall(&tcbtable, icmpsrc.sin_addr, EMSGSIZE,
tcp_mtudisc);
KERNEL_UNLOCK_ONE(NULL);
mutex_exit(softnet_lock);
return;
}
#ifdef TCP_DEBUG
#ifdef INET
if (tp->t_inpcb)
so = tp->t_inpcb->inp_socket;
#endif
#ifdef INET6
if (tp->t_in6pcb)
so = tp->t_in6pcb->in6p_socket;
#endif
ostate = tp->t_state;
#endif /* TCP_DEBUG */
/*
* Clear the SACK scoreboard, reset FACK estimate.
*/
tcp_free_sackholes(tp);
tp->snd_fack = tp->snd_una;
/*
* Retransmission timer went off. Message has not
* been acked within retransmit interval. Back off
* to a longer retransmit interval and retransmit one segment.
*/
if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
tp->t_rxtshift = TCP_MAXRXTSHIFT;
TCP_STATINC(TCP_STAT_TIMEOUTDROP);
tp = tcp_drop(tp, tp->t_softerror ?
tp->t_softerror : ETIMEDOUT);
goto out;
}
TCP_STATINC(TCP_STAT_REXMTTIMEO);
rto = TCP_REXMTVAL(tp);
if (rto < tp->t_rttmin)
rto = tp->t_rttmin;
TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
tp->t_rttmin, TCPTV_REXMTMAX);
TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
/*
* If we are losing and we are trying path MTU discovery,
* try turning it off. This will avoid black holes in
* the network which suppress or fail to send "packet
* too big" ICMP messages. We should ideally do
* lots more sophisticated searching to find the right
* value here...
*/
if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
TCP_STATINC(TCP_STAT_PMTUBLACKHOLE);
#ifdef INET
/* try turning PMTUD off */
if (tp->t_inpcb)
tp->t_mtudisc = 0;
#endif
#ifdef INET6
/* try using IPv6 minimum MTU */
if (tp->t_in6pcb)
tp->t_mtudisc = 0;
#endif
/* XXX: more sophisticated Black hole recovery code? */
}
/*
* 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) {
#ifdef INET
if (tp->t_inpcb)
in_losing(tp->t_inpcb);
#endif
#ifdef INET6
if (tp->t_in6pcb)
in6_losing(tp->t_in6pcb);
#endif
/*
* This operation is not described in RFC2988. The
* point is to keep srtt+4*rttvar constant, so we
* should shift right 2 bits to divide by 4, and then
* shift right one bit because the storage
* representation of rttvar is 1/16s vs 1/32s for
* srtt.
*/
tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
tp->t_srtt = 0;
}
tp->snd_nxt = tp->snd_una;
tp->snd_high = tp->snd_max;
/*
* If timing a segment in this window, stop the timer.
*/
tp->t_rtttime = 0;
/*
* Remember if we are retransmitting a SYN, because if
* we do, set the initial congestion window must be set
* to 1 segment.
*/
if (tp->t_state == TCPS_SYN_SENT)
tp->t_flags |= TF_SYN_REXMT;
/*
* Adjust congestion control parameters.
*/
tp->t_congctl->slow_retransmit(tp);
(void) tcp_output(tp);
out:
#ifdef TCP_DEBUG
if (tp && so->so_options & SO_DEBUG)
tcp_trace(TA_USER, ostate, tp, NULL,
PRU_SLOWTIMO | (TCPT_REXMT << 8));
#endif
KERNEL_UNLOCK_ONE(NULL);
mutex_exit(softnet_lock);
}
//-------------------------------------------------------------------------//
// TCP timer processing.
//-------------------------------------------------------------------------//
static
struct tcpcb * tcp_timers(
Node *node,
struct tcpcb *tp,
int timer,
UInt32 tcp_now,
struct tcpstat *tcp_stat)
{
int rexmt;
TransportDataTcp *tcpLayer = (TransportDataTcp *)
node->transportData.tcp;
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 <= TCPTV_MAXIDLE)
tp->t_timer[TCPT_2MSL] = TCPTV_KEEPINTVL;
else {
// printf("TCP: Connection closed by timer\n");
tp = tcp_close(node, tp, tcp_stat);
}
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:
if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
tp->t_rxtshift = TCP_MAXRXTSHIFT;
//if (tcp_stat)
//tcp_stat->tcps_timeoutdrop++;
printf("TCP: Retransmission timer went off\n");
tp = tcp_drop(node, tp, tcp_now, tcp_stat);
break;
}
//if (tcp_stat)
//tcp_stat->tcps_rexmttimeo++;
rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
TCPT_RANGESET(tp->t_rxtcur, rexmt,
tp->t_rttmin, TCPTV_REXMTMAX);
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
//
// 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 - TCP_DELTA_SHIFT));
tp->t_srtt = 0;
}
tp->snd_nxt = tp->snd_una;
if (TCP_VARIANT_IS_SACK(tp) && tp->isSackFastRextOn) {
TransportTcpSackRextTimeoutInit(tp);
TransportTcpTrace(node, 0, 0, "Faxt: timeout");
}
// Force a segment to be sent.
tp->t_flags |= TF_ACKNOW;
// If timing a segment in this window, stop the timer.
// The retransmitted segment shouldn't be timed.
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 threshhold size.
// For a threshhold, 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 threshhold
// to go below this.)
//
{
unsigned int win;
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_partialacks = -1;
tp->t_dupacks = 0;
}
tp->t_ecnFlags |= TF_CWND_REDUCED;
TransportTcpTrace(node, 0, 0, "Rext: timeout");
//
// To eliminates the problem of multiple Fast Retransmits we uses this
// new variable "send_high", whose initial value is the initial send
// sequence number. After each retransmit timeout, the highest sequence
// numbers transmitted so far is recorded in the variable "send_high".
//
if (TCP_VARIANT_IS_NEWRENO(tp)) {
tp->send_high = tp->snd_max;
}
tcp_output(node, tp, tcp_now, tcp_stat);
break;
//
// Persistance timer into zero window.
// Force a byte to be output, if possible.
//
case TCPT_PERSIST:
//if (tcp_stat)
//tcp_stat->tcps_persisttimeo++;
//
// Hack: if the peer is dead/unreachable, we do not
// time out if the window is closed. After a full
// backoff, drop the connection if the idle time
// (no responses to probes) reaches the maximum
// backoff that we would use if retransmitting.
//
if (tp->t_rxtshift == TCP_MAXRXTSHIFT) {
UInt32 maxidle = TCP_REXMTVAL(tp);
if (maxidle < tp->t_rttmin)
maxidle = tp->t_rttmin;
maxidle *= tcp_totbackoff;
if (tp->t_idle >= TCPTV_KEEP_IDLE ||
tp->t_idle >= maxidle) {
//if (tcp_stat)
//tcp_stat->tcps_persistdrop++;
printf("TCP: Idle timer went off\n");
tp = tcp_drop(node, tp, tcp_now, tcp_stat);
break;
}
}
tcp_setpersist(tp);
tp->t_force = 1;
tcp_output(node, tp, tcp_now, tcp_stat);
tp->t_force = 0;
break;
//
// Keep-alive timer went off; send something
// or drop connection if idle for too long.
//
case TCPT_KEEP:
//if (tcp_stat)
//tcp_stat->tcps_keeptimeo++;
if (tp->t_state < TCPS_ESTABLISHED)
printf("TCP: Keep-alive timer went off before established\n");
goto dropit;
if (tcpLayer->tcpUseKeepAliveProbes && tp->t_state <= TCPS_CLOSING) {
//
// If the connection has been idle for more than the sum of
// TCPTV_KEEP_IDLE (set to 2 hours) and TCPTV_MAXIDLE
// (set to the total time taken to send all the probes),
// it's time to drop the connection.
//
if (tp->t_idle >= TCPTV_KEEP_IDLE + TCPTV_MAXIDLE)
printf("TCP: Keep-alive timer went off\n");
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.
//
//if (tcp_stat)
//tcp_stat->tcps_keepprobe++;
tcp_respond(node, tp, tp->t_template,
0, tp->rcv_nxt, tp->snd_una - 1,
0, tcp_stat);
tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
} else {
//
// If the tcpUseKeepAliveProbes is FALSE
// or the connection state is greater than TCPS_CLOSING,
// reset the keepalive timer to TCPTV_KEEP_IDLE.
//
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
}
break;
dropit:
//if (tcp_stat) {
//
// Note that this counter counts connection drops due to
// failure in connection establishment and the keepalive
// timer timeouts
//
//tcp_stat->tcps_keepdrops++;
//}
// printf("TCP: Unknown timer went off\n");
tp = tcp_drop(node, tp, tcp_now, tcp_stat);
break;
}
/*
* 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;
}
/*
* TCP timer processing.
*/
struct tcpcb*
tcp_timers(struct tcpcb *tp, int timer)
{
int rexmt;
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.
* If TIME_WAIT is not set, this is FIN_WAIT_2 timer.
*/
case TCPT_2MSL:
if (tp->t_state != TCPS_TIME_WAIT &&
tp->t_idle <= g_tcp_maxidle)
tp->t_timer[TCPT_2MSL] = g_tcp_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:
if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
tp->t_rxtshift = TCP_MAXRXTSHIFT;
g_tcpstat.tcps_timeoutdrop++;
tp = tcp_drop(tp, tp->t_softerror ?
tp->t_softerror : ETIMEDOUT);
break;
}
g_tcpstat.tcps_rexmttimeo++;
rexmt = TCP_REXMTVAL(tp) * g_tcp_backoff[tp->t_rxtshift];
TCPT_RANGESET(tp->t_rxtcur, rexmt,
tp->t_rttmin, TCPTV_REXMTMAX);
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) {
in_losing(tp->t_inpcb);
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 threshhold size.
* For a threshhold, 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 threshhold
* 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;
/*
* Persistance timer into zero window.
* Force a byte to be output, if possible.
*/
case TCPT_PERSIST:
g_tcpstat.tcps_persisttimeo++;
/*
* Hack: if the peer is dead/unreachable, we do not
* time out if the window is closed. After a full
* backoff, drop the connection if the idle time
* (no responses to probes) reaches the maximum
* backoff that we would use if retransmitting.
*/
if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
(tp->t_idle >= g_tcp_maxpersistidle ||
tp->t_idle >= TCP_REXMTVAL(tp) * g_tcp_totbackoff)) {
g_tcpstat.tcps_persistdrop++;
tp = tcp_drop(tp, ETIMEDOUT);
break;
}
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:
g_tcpstat.tcps_keeptimeo++;
if (tp->t_state < TCPS_ESTABLISHED) // connection-establishment timer.
goto dropit;
if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
tp->t_state <= TCPS_CLOSE_WAIT) { // keepalive timer.
if (tp->t_idle >= g_tcp_keepidle + g_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.
*/
g_tcpstat.tcps_keepprobe++;
tcp_respond(tp, tp->t_template, (usn_mbuf_t *)NULL,
tp->rcv_nxt, tp->snd_una - 1, 0);
tp->t_timer[TCPT_KEEP] = g_tcp_keepintvl;
} else
tp->t_timer[TCPT_KEEP] = g_tcp_keepidle;
break;
dropit:
g_tcpstat.tcps_keepdrops++;
tp = tcp_drop(tp, ETIMEDOUT);
break;
}
