Re: [vserver] netstat in vserver kills machine

About this list Date view Thread view Subject view Author view Attachment view

From: Jacques Gelinas (
Date: Sat Mar 23 2002 - 12:11:54 EST

On Sat, 23 Mar 2002 17:19:57 -0500, Martin Josefsson wrote
> On Sat, 23 Mar 2002, James MacLean wrote:
> > On Sat, 23 Mar 2002, Martin Josefsson wrote:
> >
> > > On Sat, 23 Mar 2002, James MacLean wrote:
> > > > I seem to remember it was an Ieee then the machine completely froze...
> > > > Except for sysrq combos which confirmed the info reported :).
> > > Ok thanks, one more thing, do you remember if numlock worked?
> > > It sounds like an infinite loop in tcp_get_info :(
> >
> > That I can not remember, but since the first time it happened, another
> > staff member rebooted it, I believe he would have tried numlock.
> >
> > I know I tried ctrl-alt-delete and it didn't do anything :).
> >
> > Also, we blew it up a few more times through the day. Once with netstat at
> > ctx 0 (normal root user) and once we were not sure what caused it although
> > we had started an IceCast server on its own vserver.
> ctx8 which is the first version to include my netstat stuff has been very
> stable here on my machine, nothing has crashed or anything, I've tried
> running netstat in a loop and it's been working fine.
> and another user said that ctx8 worked fine on his machine but ctx9
> doesn't and ctx9 doesn't touch this stuff at all so I'd actually expect
> ctx9 to work as well as ctx8 does.

I have replicated this problem with ctx-9 on two machines, using netstat. Very
odd. On one machine, I have hammered on netstat for a while without any problems
then few hours later (uptime hours later), I crashed the machine twice in a row
using netstat.

The trace shows that the kernel is dying in __get_free_page(), called from
proc_file_read, just before calling tcp_get_info. My idea was that the new
stuff introduced in ctx-8 (private vserver netstat) was doing some corruption.
But ctx-8 has never crashed on me.

I have reviewed and reviewed tcp_get_info() and it seems all fine to me.

Very strange. I wonder if for some reason the kernel ctx-9 has been
improperly compiled. In ctx-9, I introduced a new field in "struct sock" for
the broadcast address.

Anyone has withness a crash with ctx-9 with a kernel they compiled themselves
(not using my binary) ?

Jacques Gelinas <>
vserver: run general purpose virtual servers on one box, full speed!

About this list Date view Thread view Subject view Author view Attachment view

This archive was generated by hypermail 2.1.4 : Mon Aug 19 2002 - 12:01:01 EDT