Hard crash on 6.x -- reproducible, multiple people affected

[Kris Kennaway]

On Sun, Apr 02, 2006 at 08:22:56PM +0100, Chris wrote:
> On 31/03/06, Peter Thoenen <eol1 at yahoo.com> wrote:
> > Hallo everybody.  Will prob end up PR'ing this but want to post to the
> > list first and get some feelers.  At first I thought it was just me but
> > as the port maintainer for tor-devel I have had a couple other folk
> > email me with the exact same problem (believing tor to be the cause).
> > Its not.
> >
> > Problem: When tor (running in server mode), i2p, or freenet is ran (all
> > high bandwidth, high number of concurrent tcp / udp / ip connections,
> > maybe 2000+ sim connects @ a constant 4 mb) between the 2 and 4 hour
> > mark a system running FBSD 6 will hard crash and power off.  This is
> > reproducible (though not on demand, just on arbitrary time between 2
> > and 4 hours).
> >
> > Troubleshooting so far:
> >
> > - It is NOT an issue on 5.x.  It is an issue on both 6.0 and 6.1BETA4.
> >
> >
> > - It is not easy to reproduce.  I assume a couple hundred people use
> > these ports and I have only had a dozen or so report this problem to
> > me.  All of the people suffering from this though CAN reproduce this
> > crash on demand.  I have to assume though the other 99% of users though
> > do not see this error for unknown reasons.
> >
> > - It is not a port issue (as tor/i2p/freenet all can cause it).  It is
> > not a java issue (was another early worry of mine but tor is written in
> > C).
> >
> > - The hard crash does NOT generate a dump or panic for anybody I have
> > spoken with.  /var/crash comes up empty every time.
> >
> > - It is not platform or hardware specific (this was my initial guess).
> > I have seen this on both the amd64 and i386 archs.  I have also seen it
> > on various motherboards and nic's (bge, nv, em).  To reiterate, running
> > 5.x on identical hardware does not cause this crash.
> >
> > - It is not a tunable or sysctrl issue as far as I can tell.  Spent the
> > last month tuning and watching my system limits and all appears to be
> > good.
> >
> > - My gut feeling is this has something to do with the new network stack
> > introduced in 6.0.
> >
> > - When not running one of the aforementioned ports, system does not
> > crash.  I have a three week uptime currently on one of my testbeds
> > experiencing this problem.  If I start tor / i2p / or freenet I will
> > crash within hours.
> >
> > Help would be appreciated in resolving this or troubleshooting further.
> >  I am at a loss here and me (along with a couple other folk that have
> > emailed me) have some nice expensive paperweights now as they crash on
> > use.  If you are developer and want to look into this, I can provide
> > all the info you need, just let me know exactly what it is.
> >
> > Thanks,
> >
> > -Peter
> > _______________________________________________
> > freebsd-questions at freebsd.org mailing list
> > http://lists.freebsd.org/mailman/listinfo/freebsd-questions
> > To unsubscribe, send any mail to "freebsd-questions-unsubscribe at freebsd.org"
> >
> 
> I am interested in the development of this issue, as everyone appears
> disinterested have you submitted a PR?

It's not that I'm disinterested, but it's really hard to imagine what
kind of kernel bug could cause your system to power off in this way.

Are any of these applications Linux binaries?  The Linux halt syscall
maps to some other common syscall in FreeBSD, so if a Linux binary is
incorrectly branded as a FreeBSD binary it can shut down the system
when run.

That's the only thing I can think of.

Kris
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-questions/attachments/20060402/b245e8d9/attachment.pgp