Re: [Xen-devel] Re: Oopses with xen 1.2

[Ian Pratt <Ian.Pratt@xxxxxxxxxxxx>]

> > I'm seeing oops messages on xen 1.2 on a system we have here. I
> > currently can't keep the box up for more than a couple of hours.
> >
> > The box runs: apache, vsftpd, postfix, and ecartis (a mailing
> > list manager).
> > 
> > Hardware: Pentium III 500mhz, 192mb ram. At the moment it's only
> > running domain 0 (with 32mb ram) and then one other domain with
> > the remaining ram. I've run memtest86 on the box for multiple
> > passes, and the ram returns perfect results. The system runs
> > on IDE. It has a eepro1000 network card.
> 
> I want to provide some feedback related to the below.
> 
> Firstly, I'm still running 1.2 on that box, successfully,
> given that it used to crash every few hours. It's now been
> up just less than 2 days without problems - and seems to give
> every indication that it's not having problems. I'll send more
> details if this isn't the case.

Interesting. Which version of 1.2?

If you still get problems, please can you try moving to
xeno-unstable. I know of at least one bug that's fixed in
xeno-unstable that is a pain to back port to 1.2 so hasn't been
done yet (and may never be done with the forthcoming 2.0 release)
 
> I don't have any need to run the vsftp daemon on the machine,
> and recently disabled it, as it seemed to be mentioned in a
> lot of the oops messages. Since then, the machine has been
> stable. One thought I have is that if someone was trying an
> exploit on the machine, it might never have done anything on
> the normal kernel, but does on xen.

That's certainly a possibility. There'll certainly be some
exploits that will fail due to differences in the kernel.

> I'm running the standard debian woody vsftpd. The details are
> below. As mentioned in the description, it uses sendfile, which
> I thought might have something to do with this, as it's not
> a common function call, and might have less testing than most
> other calls.

We've exercised Apache very heavily, but I'm not sure whether it
uses sendfile.

The oops message is interesting. It's deep in the bowels of mm
code, so its the sort of thing you might expect to see if there
was a problem with arch xen. 

> The other option is it might have something to do
> with the ftp-firewalling modules, which I run on that machine
> to ensure my packet filters can handle established/related
> tcp filters.

Unlikely that this would lead to a xen-specific crash.
 
> I will spend some time when I can trying to replicate this
> exactly.

If you can find a way to reproduce it reliably on xeno-unstable
we'd be very interested in fixing it.

Cheers,
Ian


-------------------------------------------------------
This SF.Net email is sponsored by the new InstallShield X.
>From Windows to Linux, servers to mobile, InstallShield X is the one
installation-authoring solution that does it all. Learn more and
evaluate today! http://www.installshield.com/Dev2Dev/0504
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/xen-devel