On Apr 24, 2005, at 00:41, christos@netbsd.org wrote:
> Synopsis: "release of unlocked lock" panic with null fs
>
> State-Changed-From-To: open->feedback
> State-Changed-By: christos@netbsd.org
> State-Changed-When: Sun, 24 Apr 2005 00:41:34 -0400
> State-Changed-Why:
> Can you please try this patch?
> It will look in the hold queue, if nothing in the free queue was 
> successful.

(BTW, the mail I received included no patch, but I pulled it out of the 
problem report.)

I finally got this installed Friday night, tweaked to log a message 
when it switches over from the free list to the hold list.  The daily 
cron job (standard, plus daily.local includes 
download-vulnerability-list and a propagation of my KDC database) seems 
to trigger it:

May 29 03:17:08 raeburn /netbsd: getcleanvnode: switching from free 
list to hold list
May 29 03:17:39 raeburn last message repeated 5895 times

... but at least Friday and Saturday nights' test builds of Kerberos 
(the cron job that frequently triggered the problem) didn't cause a 
crash or hang, even if they do continue to trigger this code path:

May 29 05:21:04 raeburn /netbsd: getcleanvnode: switching from free 
list to hold list
May 29 05:21:35 raeburn last message repeated 2872 times

Though even if it does fix my problem (and it'd be a few more days 
before I can really be confident of that, but it looks good so far), 
unless it's now impossible to fail to allocate a vnode, there's still 
another bug, that failure to allocate, at least when dealing with a 
null file system, can leave a node locked.

Ken