> > 3- you have SMBFS enabled, as a module, but you have long filename
> > support compiled into the kernel. Isnt this lethal if the NT server
> > broadcasts some silly SMB stuff and your idle machine has the module
> > unloaded?
> >
> I had SMBFS and VFAT (is this the long file name stuff you meant?), and
> anything else having to do with WinXX, *available* as modules only. I was
> running kerneld, so it should automagically load the needed modules if
> something like that happened. Shouldn't it?
it would be a good test to compile them into the kernel not as modules,
and test 2.0.27 again? I'm not saying this is the bug, it's just a
feeling.
> But why in the world would it do that? I'm not mounting any drives from
> the NT boxes, nor exporting anything, so why should Linux care what the NT
> box does?
well SMB (and its zillion subversions like NT LM 0.12) has crazy stuff
like networkwide enforced file locks. If memory serves, as an SMB client,
you might end up listening to broadcasts about changes in the locking
space.
> Both times it happened the NT machines were basically idle (they are servers
> for the company tech support, management and Win* people - who had all gone
> home), but the Linux box was pretty active. (Running the WWW proxy, sendmail,
> named and more for a small ISP). The reboots both occurred during the
> evening, when traffic is near its peak.
they are never idle, they regularly announce silly services.
so i'm not sure, but there is some chance that the problems are in this
area. The recompile should sort out this kind of stuff.
-- mingo