From: Herbert Poetzl (herbert_at_13thfloor.at)
Date: Tue 06 Jan 2004 - 23:33:02 GMT
On Tue, Jan 06, 2004 at 06:43:43PM +0100, Christian Mayrhuber wrote:
> Christian Mayrhuber wrote:
> Does the latest vserver 1.22 still posess the SMP bug?
hmm, what is 'the SMP bug'?
- the uts_sem issue present since ctx-2 (in words two)
- the dynamic allocation deadlock?
- the dynamic wraparound lockup?
those have benn fixed in 1.22 and should
be still there in 1.00 ;)
> I think I hit it on a dual xeon machine, but had no physical
> access, so somebody else did a reboot back to vserver 1.00.
> The non SMP Athlon test machine is still up and running with
> vserver 1.22 ;-)
currently we are tracking some hard to trigger SMP
races with or within the procfs (or the way current
development versions do use it), but that should not
hit you, except if you spawn 100 contexts per minute
while banging at the procfs entries ...
> The IPV4 bug is in vserver 1.00, but fixed in vserver 1.22, right?
> Recently, I had problems to send mail to a machine behind a netfilter
> firewall from a machine with a vserver 1.00 kernel.
> The firewall did not complain about corrupted packets, but the smtp
> server behind the firewall did. This happened with a ctx17 kernel, too.
> Things worked fine with a standard kernel.
> I'll answer this myself.
> Both questions YES.
> Following patch should fix it for vserver 1.00:
yeah, this was a bug I introduced ;) it isn't present
in ctx17 and it was removed in 1.21, if there is interest
in updating some parts of vs1.00, please let me know
> I'll use that for my servers.
> Seems to be the only stable release that will work reliable on SMP
> systems and not do strange things to IPV4 packets.
> Please, correct me if I'm wrong.
hmm, I would say 1.22 should do better, but I tell you
I don't know ... although feedback is always welcome ...
if you are interested in hunting down and/or improving
any IPV4/6 issues, just let me know, I'm all ears ...
Vserver mailing list