Re: Win95 won't forget old IP address

[Mark Sirota <[e-mail]>]

Ted Lemon wrote:
> A trace of the output of DHCP would be very helpful here - it should
> be telling you if the packets are coming and going through a BOOTP
> gateway or not.   If the DHCPREQUEST is coming through a BOOTP
> gateway, it'll be printed like this:

Quite right!  Many thanks to both Ted Lemon and Ken Key for helping me to
diagnose this.

To summarize, the situation is that we have a Win95 DHCP client, the ISC
DHCP server Beta release 4 patchlevel 4 running under Digital UNIX 3.2D,
and a Cisco router in the middle running IOS 11.1 with a helper address
configured to point to the server.

We've just moved the client to a new subnet and attempted to boot.
The sequence of events is:
(1) Client sends REQUEST via all-1's broadcast
(2) Router forwards to server via unicast
(3) Server sends NAK to router via unicast
(4) Router forwards to client via unicast
(5) Client drops packet because it's not an all-1's broadcast

Conclusion?  The Win95 DHCP client is broken.

Unfortunately, I can't prove the broadcast/unicast statements above, because
the router broke sometime between 17:42:41 last night and this morning when
I went back to try it again.  Sigh.  I probably won't be able to try again
until tomorrow at the earliest.

I see that RFC 1541 provides for the broadcast bit, which says that in step
4 above, the router should use a broadcast rather than a unicast if the bit
is set.  I cannot tell at the moment if the bit was set right now either;
that too can be tested when I get the replacement router.

In any case, we are somewhat confused about why we would need to do this.
Using a broadcast address at both the link layer and the IP layer at this
step seems very expensive and unnecessary...  Why can't we just use the
client's MAC address at the link layer, regardless of what's at the IP layer?

Our supposition si that the router is (was) indeed doing just this, but
the Win95 DHCP client dropped the NAK packet because the broadcast address
was not all-1's.  Does anyone else's experience back this up or refute it?

If this is true, it may be impossible for us to support Windows 95 DHCP
clients, until Microsoft fixes this.  This would be something of a disaster;
people have been clamoring for support for Windows 95 DHCP clients, but it
appears that without modifying either the Cisco code or the Microsoft code,
it may be impossible.

Several people have said that I need to be running Cisco IOS 9.21(4) or
later, or 10.0(3) or later.  Anyone know if 11.1 qualifies as "later"?
-- 
Mark Sirota, Network Systems Engineer
University of Pennsylvania, Information Systems and Computing
[e-mail], 215/573-7214