Existing netlink kernel <--> kernel communication

Existing netlink kernel <--> kernel communication

[Luis R. Rodriguez]

Was curious if there are existing use cases of netlink for kernel <-->
kernel communication rather than the typical kernel <--> userspace.
The original RFC 3549 [1] does specify it could have been used for
"intra-kernel messaging", but am just not aware of existing use cases.
A pointer could help.

[1] http://www.faqs.org/rfcs/rfc3549.html

  Luis

Re: Existing netlink kernel <--> kernel communication

[Marcel Holtmann]

Hi Luis,

> Was curious if there are existing use cases of netlink for kernel <-->
> kernel communication rather than the typical kernel <--> userspace.
> The original RFC 3549 [1] does specify it could have been used for
> "intra-kernel messaging", but am just not aware of existing use cases.
> A pointer could help.

I would use direct function calls for kernel to kernel communication. No
need to put any messaging system in between.

Regards

Marcel

Re: Existing netlink kernel <--> kernel communication

[Luis R. Rodriguez]

On Mon, Jul 19, 2010 at 9:07 AM, Patrick McManus <mcmanus@ducksong.com> wrote:
> Hi Luis,
>
> I'm following up to a linux kernel post of yours back in December:
>
> http://marc.info/?l=linux-kernel&m=125053428528207&w=3
>
> Basically, I have the same interest - an internal netlink client to the
> qdisc interfaces.. I was wondering if you made any progress on the topic
> you could share? It seems maddening that netlink is promoted as an
> intra-kernel mechanism but it is non-obvious how to write a client..

The reason we at Atheros were considering it was a possible way to
communicate between two different subsystems, in our case Bluetooth
and 802.11. While it is technically possible to use netlink for such
purposes it introduces a layer of complexity which can be avoided by
just using direct exported routines, and by coordinating between the
two subsystems you need communication with.

So, the real question I think is why do you need a rigid API ? If
there is a good need, perhaps its a good idea, but so far I guess no
one simply has come up with one.

  Luis