2.2.1a on Solaris; runaway fcntl bugfix !

[David Lee]

On Wed, 5 Sep 2001, Jeremy Allison wrote:

> Ok, due to some *amazingly* helpful debugging work by
> Gerry, I'm pretty sure I've found the fcntl spinning
> problem and I've fixed it in CVS.

Thanks, Gerry.

> I'm also including the patch that applies cleanly to
> 2.2.1a to this message.

Thanks, Jeremy, including for any extra work of retrofitting to 2.2.1a .

> The problem was in the code that was added to abstract
> the kernel oplock interface between IRIX and Linux.
> 
> What it did was it changed the logic of the oplock
> break processing so that incoming SMB messages from
> the client were being given precedence over incoming
> oplock break message from other smbds. The original
> 2.0.x code gave precedence to oplock break messages
> over client smb messages.
> 
> This would cause the messages to get lost when the receiving
> smbd was under heavy client load - just the kind of hard
> to reproduce bug that we were seeing.

Sounds plausible.

> Please try the attached patch against 2.2.1a installations
> and let me know if it fixes the problem (it should :-).
> Fingers crossed......

I put it on my test server earlier this afternoon: no adverse effects
apparent yet.  So I have just put it onto our smallest production server
(currently 5-10 active sessions).  Again, no adverse effects apparent yet. 

Both those servers are "light", so have not seen the original problem. 
But I have to play safe, especially with this change which looks so
fundamental to the logic of receiving each and every SMB packet! 

It is now too close to the weekend (we are a seven-day-a-week shop) to be
any more adventurous (near 15:00, Friday afternoon). 

But early next week (perhaps even at the end of the weekend) I might try
it on two larger servers (~50 simultaneous connections).  After a further
day or so, I would hope then to try it on our main server, which does
suffer the problem: it has ~170 connections now which rises to 800++
when term starts in a few weeks' time.

Thanks again.

-- 

:  David Lee                                I.T. Service          :
:  Systems Programmer                       Computer Centre       :
:                                           University of Durham  :
:  http://www.dur.ac.uk/t.d.lee/            South Road            :
:                                           Durham                :
:  Phone: +44 191 374 2882                  U.K.                  :