misc/95180: Unexplained Hard Crash on 6.0 and 6.1BETA's. Reproducible

[Peter Thoenen]

>Number:         95180
>Category:       misc
>Synopsis:       Unexplained Hard Crash on 6.0 and 6.1BETA's.  Reproducible
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Sat Apr 01 00:00:32 GMT 2006
>Closed-Date:
>Last-Modified:
>Originator:     Peter Thoenen
>Release:        6.0 and 6.1BETA3
>Organization:
N/A
>Environment:
FreeBSD herz.nan-elmoth.net 6.1-PRERELEASE FreeBSD 6.1-PRERELEASE #3: Sat Feb 18 21:50:32 EST 2006     root at herz.nan-elmoth.net:/usr/obj/usr/src/sys/BITCH  amd64

NOTE: This is not a arch problem.  Have see on i386 also.
>Description:
System, after 2 to 4 hours of running, hard crashes and powers off after starting certain software.  If said software is not started, sytem remains up.  No panic or dump is generated.  Problem is reproducible.  Problem did not exist on FBSD 5.x.  As the tor-devel maintainer, I have had multiple people on a variety of different platforms report this identical issue to me.  The only common thread is we are all running FBSD 6.0 and 6.1BETA.  It is NOT an application issue as problem is reproducible with other applications.  Nor do I believe it to be a faulty hardware issue (as a variety of poeple have reborted) and works just fine on FBSD 5.x.  Not a system limit issue either as far as I can tell as I have spent the last month playing with tunables and sysctrl.  I can provide additional information if needed, just tell me exactly what you want.  As it stands now, my colo (along with a couple other FBSD folk's colo's) are completely unusable.
>How-To-Repeat:
1.  Start either: tor-devel (server mode), freenet, i2p.
2.  Wait 2 to 4 hours.
3.  System hard powers off.
4.  Repeat.

NOTE: Not everybody gets this so assuming something obscure.  I have had a dozen or so folk email me though with identical issue.  I believe it has something to do with the new network code introduced in FBSD6.  Only common thread I can see in the above applications are all open a large number of concurrent connections and high constant bandwidth.  In my case I am usually running 4Mbs constant with ~8000 open connections and even more pf states.
>Fix:
Unknown and at a lose on how to further troubleshoot.
>Release-Note:
>Audit-Trail:
>Unformatted: