The same condition that led us to discover the “double deference” in ISC-Bugs #18097, also led us to find the “bad parameter to execute ping”.  We are still not able to find the cause of why our box gets into this condition and causes dhclient to get into the TIMEOUT state.  If you can provide any thoughts or suggestions, it would be helpful.

When this condition occurs, dhclient gets a valid offer from the server in DHCPOFFER and DHCPACK, but gets into the TIMEOUT state.  Because $1 is not initialized in the script below, “ping” in line 209 cannot execute, and exits with error.  The script then goes to line 229 and exits with error code 1.  This causes dhclient to execute line 959 in dhclient.c and send a DHCPDECLINE.

We have about one in a hundred boxes that occasionally get into this condition, and stays in this condition at least for a few minutes.  When performing lease renewal in this condition, dhclient declines an offer executing the code path described above, so the server offers the next IP address.  The dhclient declines this next IP address executing the same code path, so the server again offers the next IP address, until the server runs out of IP address in its address pool.

I do not know if the intent of “ping” in line 209 is to find out if another device is already using $new_ip_address.  If so, the “ping” should be executed before “ifconfig”.  But it does not hurt to “ping –q –c 1 $new_ip_address”, as the IP address is already configured in $interface and is always pingable.  This is our short term work around.

From the file client/scripts/linux

202 if [ x$reason = xTIMEOUT ]; then

203   if [ x$alias_ip_address != x ]; then

204     ifconfig $interface:0- inet 0

205   fi

206   ifconfig $interface inet $new_ip_address $new_subnet_arg \

207                     $new_broadcast_arg

208   set $new_routers

209   if ping -q -c 1 $1; then

210     if [ x$new_ip_address != x$alias_ip_address ] && \

211             [ x$alias_ip_address != x ]; then

212       ifconfig $interface:0 inet $alias_ip_address $alias_subnet_arg

213       route add -host $alias_ip_address dev $interface:0

214     fi

215     if [ $relmajor -lt 2 ] || \

216         ( [ $relmajor -eq 2 ] && [ $relminor -eq 0 ] ); then

217       route add -net $new_network_number

218     fi

219     for router in $new_routers; do

220       if [ "x$new_subnet_mask" = "x255.255.255.255" ] ; then

221     route add -host $router dev $interface

222       fi

223       route add default gw $router

224     done

225     make_resolv_conf

226     exit_with_hooks 0

227   fi

228   ifconfig $interface inet 0 down

229   exit_with_hooks 1

230 fi

From file client/dhclient.c

 955     /* If the BOUND/RENEW code detects another machine using the

 956        offered address, it exits nonzero.  We need to send a

 957        DHCPDECLINE and toss the lease. */

 958     if (script_go (client)) {

 959         make_decline (client, client -> new);

 960         send_decline (client);

 961         destroy_client_lease (client -> new);

 962         client -> new = (struct client_lease *)0;

 963         state_init (client);

 964         return;

 965     }

Pauline Yeung

Cisco Systems, Inc.