Mysterious jail lockups

[D Hill]

On Tue, 9 Oct 2007 at 10:16 -0400, wmoran at collaborativefusion.com confabulated:

> Has anyone else seen this?
>
> The symptoms are a jail that has no processes in it, and thus can not
> be stopped/killed/whatever.  Only solution is to reboot the host system.
> Trying to jexec into the jail results in an error, so new processes can't
> be started therein.
>
> It doesn't happen very often, and I've been unable to reproduce it on
> demand.  What I'm looking for at this point are whether or not anyone
> else has seen this, and advice on how to track it down/reproduce it, with
> the eventual goal of fixing the problem.
>
> It would be nice if there were a command, let's say "jkill" that killed
> the _jail_.  There is a port called jkill that (allegedly) does this, but
> looking at the perl code, all it does it loop through a ps listing
> killing off processes.  In the event of a jail with no processes, this
> doesn't help any.
>
> Theoretically, this would be some sort of kernel bug, whereby the
> reference counter to the jail is not properly decremented when processes
> die and thus the jail never shuts down.  Given the infrequency of the
> occurrence and my inability to produce a reproducible case, I expect
> it to be challenging to track down.
>
> Any advice?

Same thing seen here running:

FreeBSD ns1jail.example.com 6.2-RELEASE-p7 FreeBSD 6.2-RELEASE-p7 #1:
   Thu Aug  9 18:59:52 UTC 2007 d.hill at ns1jail.example.com:/usr/obj/usr/src/sys/NS1JAIL  i386

After I '/etc/rc.d/jail stop ns1_ynp' it will linger and finally stop. 
Other times it will still show a day or two later. Displaying the process 
list on the host does not show any process running with the 'J' jail 
indicator. Like you, the host has to be rebooted.

-----
  _|_
(_| |