Re: FC5 NM/Madwifi "roaming" loses IP Address

[Derek Atkins <warlord MIT EDU>]

Quoting Dan Williams <dcbw redhat com>:

> On Wed, 2006-07-12 at 09:09 -0400, Derek Atkins wrote:
> > Quoting Robert Love <rml novell com>:
> >
> > > On Tue, 2006-07-11 at 15:05 -0400, Derek Atkins wrote:
> > >
> > >> I'm sitting here in Montréal at the IETF and noticed a really
> > >> annoying problem.  When I wander from one room to another
> > >> I have to roam from one AP to another, and NM seems to disconnect
> > >> the network.  When I arrive at the other room I have to manually
> > >> re-select the network and when it comes back up it's forgotten my
> > >> DHCP lease and gets me a new IP Address.
> > >
> > > Actually, I think Madwifi has a bug where it cannot roam.  Instead of
> > > silently roaming to a new AP (a process that does not involve NM), the
> > > driver loses association.
> >
> > Well, it would be nice if that bug were fixed...  But it doesn't answer
> > the question of why, when I come back up, do I get a new IP Address from
> > DHCP?  It's almost as if NM/dhclient don't know that it can just make
> > a DHCP_REQUEST and instead always falls back to a DHCP_DISCOVER.  This
> > is true even if the Network Name (ESSID) hasn't changed!  *THIS* I consider
> > a bug in NM.
>
> Well, I think that what should be done here is to try to get dhclient to
> do a DHCP renew when we notice an AP change, as long as that AP is in
> the same cached BSSID list as the old AP, rather than doing a request
> from the start.  That's a bit complicated I guess, but it would be more
> ideal than the current situation.

Or even if it's the same ESSID (regardless of whether the BSSID is cached
or not)?  If I walk across the IETF floor quickly it's possible that my
ultimate destination might not be a BSSID that was in range at the beginning
of the session, so the BSSID might not be cached.  Or I might never have
been in that room before.  Indeed, as a /user/ I should never have to
worry about the BSSID sitting under the ESSID (except, perhaps, in the
case where there's an (a) vs. (b/g) network difference).

Thanks!

> Dan

-derek

--
      Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
      Member, MIT Student Information Processing Board  (SIPB)
      URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
      warlord MIT EDU                        PGP key available