[cifs-protocol] clarification of MaxMpx value in MS-CIFS

[Christopher R. Hertel]

On 03/01/2012 11:19 AM, Jeff Layton wrote:
> Dearest Dochelp...
>
> We in the Linux CIFS client community are attempting to clean up our
> code to make it better respect the MaxMpxCount value that the server
> advertises in the NEGOTIATE request.
>
> MS-CIFS currently says this:
>
>> MaxMpxCount (2 bytes): The maximum number of outstanding SMB operations
>> that the server supports. This value includes existing OpLocks, the
>> NT_TRANSACT_NOTIFY_CHANGE subcommand, and any other commands that are
>> pending on the server.
>
> It is rumored however that servers are not supposed to count certain
> command types against this limit. For instance, SMB echoes and blocking
> locks. My questions are:
>
> 1/ what do windows servers do in this regard? Do they count all
> outstanding requests against the maxmpx value, or are certain commands
> exempted?

Section 3.3.5.33, which covers the server's handling of an echo request, 
does not list the echo as exempt.  It does, however, indicate how a cancel 
request would be handled.  As I recall (the truth is in the code), you 
cannot cancel a request that is *not* in the PendingRequestTable.

That would suggest that SMB_COM_ECHO is not exempt from MaxMpxCount.

...but it would be good to check.  My memory is that SMB_COM_ECHO may be 
handled separately since it can be executed without a SessionSetup or 
TreeConnect.  The thing is, an SMB_COM_ECHO must follow (at minimum) a 
NegotiateProtocol, which would provide the server's MaxMpxCount.

...so, SMB_COM_ECHO *may* be counted against the MaxMpxCount.

> 2/ do windows clients ever send more than MaxMpxCount requests at a
> time? For instance, suppose I have a MaxMpxCount of 50 and I spawn 100
> threads that attempt to acquire blocking locks. What happens at that
> point? Will the client end up using all of the slots for blocking locks
> in such a way that nothing else can get through? Or does it allow you
> to exceed that value somehow?
>
> Thanks for the info!

There is code specific to this in both Windows 98SE and NT4 Server (and all 
subsequent clients).  I do not recall what the code says.  A shame, as 
that's where the answer lies.

Note to DocHelp:  The client code will go through and cancel outstanding 
requests that have timed out.  That code is rarely exercised, I'd guess.  I 
just recall that it's there.

Chris -)-----

-- 
"Implementing CIFS - the Common Internet FileSystem" ISBN: 013047116X
Samba Team -- http://www.samba.org/     -)-----   Christopher R. Hertel
jCIFS Team -- http://jcifs.samba.org/   -)-----   ubiqx development, uninq.
ubiqx Team -- http://www.ubiqx.org/     -)-----   crh at ubiqx.mn.org
OnLineBook -- http://ubiqx.org/cifs/    -)-----   crh at ubiqx.org