new aio engine

[Jeremy Allison]

On Thu, Jul 19, 2012 at 02:00:13PM +0200, Christian Ambach wrote:
> 
> I like the idea behind this (and I think it is the correct thing to do
> from an engineering perspective): instead of cluttering AIO code all
> over the code, have a single place in the code that deals with aio and
> the various approaches to it on the different platforms.
> 
> However, I am not sure about the impact it will have on the
> implementation and stackability of VFS modules, especially ones that
> are not shipped as part of the Samba source.
> e.g. Antivirus modules might want to perform their own open() of a file
> before letting smbd do the actual open(). Thus this would have to be
> done asynchronously as well to not introduce synchronous IO again and so
> libasys must sit at the very low end of the chain.
> 
> Maybe this will require some guidance for external VFS module authors
> how to properly implement secondary IO.

Yes, we will need to update Richard Sharpe's excellent guide to
writing Samba VFS modules to cope with hooking into the async
engine.

That's going to be hard - but plenty of example code should
help.

Jeremy.