Ufs dead-locks on freebsd 6.2

Eric Anderson anderson at freebsd.org
Wed May 16 16:32:09 UTC 2007

On 05/16/07 11:08, Andrew Edwards wrote:
> I have a system running a dual intel zeon 2.8Ghz with 4G of ram and
> using an intel raid controller model SRCU42X which uses the amr driver.
> I have had this server running 5.4 upgraded to 6.2 and was running fine
> for several months and then after a normal reboot I've started having
> all sorts of problems with what appears to be dead-locks in the
> filesystem.  This server is my backup server and I rsync files from
> various servers onto this one fairly non-stop.  If I stop the rsync's
> the system appears to be stable although I did have a kernel core just
> last night.
> When I have been able to observe the problem I ususally see one
> filesystem become inaccessible, perhaps var but I'm not sure, and then
> in a short period of time the whole system is inaccessible.  Usually if
> I startup just one of the rsync's within a couple of hours the system
> will be un-usable.
> I did find this thread which seems to describe similar issues but this
> is a different driver.
> http://lists.freebsd.org/pipermail/freebsd-questions/2006-August/127835.
> html
> Currently I'm running with debug.mpsafevfs=0, debug.mpsafenet=1and
> debug.mpsafevm=0 but this doesn't seem to help.
> On perahps a related issue I have two other nearly identical systems
> which were going to be upgrading to 6.2 as on 5.4 I am experiencing
> deadlocks and when I hit ctrl-t I see the system is either stuck in ufs
> or zoneinfo and I have not found very much information about zoneinfo.

What's ctrl-t?

> Does anyone have any suggestions on what I can look for or other tuning
> options or had similar experiences?

When you say deadlock, you mean no access at all, but pingable?  Does it 
panic?  Can you still run things like 'ps'?

You should (if you haven't already) add debugging to your kernel, and 
make sure you are running 6-STABLE.

A ps -auxl of a system would be great if you can get it.  Otherwise, 
you'll need to get a crash dump.  You said you have a kernel core, can 
you send the output of a full back trace to the list?

Also - I've cc'ed freebsd-fs@, since it's more appropriate there.


More information about the freebsd-performance mailing list