Re: DHCP transaction timing out, even though working

[Dan Williams <dcbw redhat com>]

On Mon, 2006-09-11 at 14:19 -0700, Doug Bunting wrote:
> [If this gets through, please reply-all since I am not subscribed to this list.]
> 
> A late reply based on recent experience with OpenSuSE 10.1 -- parts
> also posted at Beyond Abstraction
> <http://beyondabstraction.net/2005/12/14/networkmanager-kde/>.  I
> don't know if Simon's issue has been resolved.
> ----
> 
> I recently worked though a similar problem with my SuSE 10.1 / Gnome setup.
> 
> The main issue in my case was the flaky DHCP clients — at least in
> SuSE.  I long ago set up my router to provide static leases, allowing
> DNS settings to automatically update without losing easy access to my
> other machines. Some routers turn static leases into infinite leases
> and Linux DHCP clients start complaining!
> 
> You might be able to find a /var/log/localmessages line such as
> "dhcpcd[7824]: infinite IP address lease time. Exiting" but
> "dhcpcd[7824]: dhcpIPaddrLeaseTime=4294967295 in DHCP server
> response." was all I could find at first. (0xFFFFFFF or a 32 bit -1
> should probably have set off alarm bells right away.)  In any case,
> NetworkManager was waiting for a DBUS message from a process that had
> died.
> 
> Reconfigured my DHCP server and I was on my way…
> 
> BTW: Other routers, such as my older default DHCP server, honour the
> DHCP lease request a client sends — providing an easier work-around in
> /etc/sysconfig/network/dhcp.
> ----
> 
> From a correction and improvement pov, the NetworkManager handling of
> a dying dhcpcd service could be improved.  The timeout message could
> cover many ills.  Does NetworkManager not monitor the status of
> services it requires?

Yes, NM will know when dhcdbd exits; but the logic to deal with an exit
during a connection attempt may not work correctly.

> I'm not sure where to complain about dhcpcd's poor handling of valid
> but infinite leases.  The problems are not limited to dhcpcd actually:
> dhclient may spit out "dhclient: send_packet: please consult README
> file regarding broadcast address." in the same situation -- after
> receiving the SIGTERM from NetworkManager, providing an irrelevant
> solution on recent Linux kernels.  Any ideas?

Best is to probably file a bug report in Red Hat bugzilla against the
dhcdbd package.  If dhcdbd doesn't like any of the values that dhclient
returns, it should certainly send an error rather than exit.

Dan