[ switching mail address, as I believe the list is subscriber-only ]

On Wed, 23 Mar 2005 11:54:31 -0500, Timothy Miller wrote:
> We should put this back on-list, since what you say is informative.
[...]
> My objection to using something other than ioctl is that I was under
> the impression that it was the best thing designed for the use we were
> discussing.  You point out, however, that Linux kernel developers have
> specifically designed another mechanism for that purpose.  If so, and
> it's an efficient mechanism, that is what we should use.

I do _not_ claim netlink is the thing to use. I am merely pointing out
that Linux does have a socket-based userland-kernel interface (and so
does BSD, if memory serves): netlink. That interface is used for async
communication and was meant to replace ioctls at least for some cases.

Also, there's some consensus on LKML that ioctls are evil and only a
last resort.

For those who don't know netlink: IIRC it was initially designed
for setting up routing tables and friends. It is now also used for
other purposes, for instance as a transport for event signalling
(e.g. auditing messages, file change notification). I have written a
machine-readable, scalable interface that offers the same information
as procfs via netlink. The problem with netlink is hardly overhead but
that the interface is somewhat documentation-challenged.

Again, I make no claims about the suitability of netlink for this
purpose. However, the argument that sockets must only ever be used for
inter-user communication holds no water.

Roger
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to