PRIVATE: smbd spins on f_setlkw (PR#21673)

[Jeremy Allison]

tagoldth at uhnres.utoronto.ca wrote:

> I find I have to kill about 20-30 smbds each day, they loop forever on F_SETLKW
> for bytes in a tdb.  I saw a message regarding this in the mailing lists, but
> no resolutions, and the links I clicked on here didn't mention anything, so
> I'll
> submit a report.  It wouldn't be so bad if the locks get spun-on if you would
> at
> least put a usleep() in there to cause them not to consume the machine!  They
> do
> die off eventually when the client box goes away, but that can be a long long
> time.  After much tracing, I did note that the same client has the same byte
> locked
> through another smbd, but nothing is checking for this condition.  They all are
> typically (from a live core image):

This seems to be a problem particular to Solaris (although
some people have seen it also on Linux, not 100% sure).

I have committed several fixes for this in the current 2.2
CVS tree, if you are able to test this code to see if it
fixes the problem (or at least gives more information)
I'd appreciate it.

Thanks,

	Jeremy Allison,
	Samba Team.