Re: NetBSD-5 Problems with Samba and AIO

[christos%astron.com@localhost (Christos Zoulas)]

In article <20091027231242.GB1304%web.de@localhost>,
Stefan Krüger  <stadtkind2%gmx.de@localhost> wrote:
>Hello,
>
>with asynchronous I/O as a major new feature in the 5.0 NetBSD release, I
>thought I could give it a try and the first application with AIO support that
>came to my mind was... Samba
>
>So, I recompiled Samba with the following line
>
>CONFIGURE_ARGS+=       --with-aio-support
>
>added to pkgsrc/net/samba/Makefile, configure output shows everything's
>detected etc. and no compile-time (or run-time later) error occurred
>
>next step was adding
>
>aio read size = 4096
>aio write size = 4096
>
>to smb.conf
>
>But... I does not work correctly, the Windows XP client <-> NetBSD 5.0.1
>Samba server connection is dropped after some time, nothing in the Samba
>logs, and I have no idea why (this does *not* happen when AIO is disabled or
>commented out in smb.conf)
>
>If anyone here wants to reproduce this, I found Bart's Stuff Test 5
>( http://www.nu2.nu/bst/ ) can be used to trigger the disconnect/failure
>easily, just run the test(s) on the samba share, it takes only a few seconds
>here to stall+error the program (so, the test never even comes close to
>completion)
>
>I've used the same smb.conf without problems (and with AIO enabled ofc) on
>both FreeBSD and Solaris10, and never had any problems with it, so I guess
>it's neither a configuration issue nor faulty AIO code inside Samba)
>
>To draw a conclusion, I'd be glad for pointers how to further analyze (and,
>hopefully, fix) this strange problem

Please file a PR. You can use ktrace to see what's going on with the AIO
calls.

christos