read/write multiplexed support

[Jeremy Allison]

On Tue, May 26, 2009 at 09:38:09PM +0200, Marcel Müller wrote:
> Volker Lendecke wrote:
>> On Sat, May 16, 2009 at 06:17:45PM +0200, Marcel Müller wrote:
>>> Done. See http://home.arcor.de/maazl/temp/tlogs.zip
>>
>> Thanks for those. If possible, can you catch the WriteMPX
>> against Windows? Maybe we're not sending the right error
>> code?
>
> Unfortunately I have no Windows at home.
>
> But further analysis showed that OS/2 chooses sometimes Read/Write RAW
> and sometimes Read/Write MPX at the connection setup.
> The issue might be quite old without much noticeable effect, since Samba
> 3.0.x supported at least Write MPX by default and obviously no windows
> server will ever accept the MPX modes over TCPBEUI. Samba 3.2.x only
> made the problem to show up.
>
> Most likely there is some kind of issue during the negotiation. Maybe
> samba returns some wrong error code that made the client to choose the
> MPX modes, maybe something else.
> However, there seems to be a much easier solution. Since I disabled the
> MPX modes in the requester configuration, I was no longer be able to
> reproduce the problem. OS/2 now always uses the Read/Write RAW modes,
> that provide similar performance than the multiplexed mode.
> There should be no significant drawback for the same reasons why the
> multiplexed modes are not essential to samba anymore. Ok the multiplexed
> modes might be slightly more responsive, because other commands can be
> sent while a package is transferred.
>
>
> Solution:
> =========
>
> Disable the MPX modes in the requester configuration:
>
> In bootdrive:\IBMLAN\IBMLAN.INI reset bit 14 and 15 in the wrkheuristics
> setting (requires restart of the requester)
>
> ;                           1         2         3         4
> ;                 012345678901234567890123456789012345678901
> ; wrkheuristics = 1111111121311111110001011120111221001111100
> ;                               ^^  (old)
>   wrkheuristics = 1111111121311100110001011120111221001111100
> ;                               ^^  (new)

As there is a workaround on the client side I'd be loathe
to put that code back into the server (even though it was
originally my code :-). The MPX calls really shouldn't be
made over TCP (IMHO).

Jeremy.