Re: [patch 5/7] CAN: Add virtual CAN netdevice driver

[Oliver Hartkopp <oliver@hartkopp.net>]

Patrick McHardy wrote:
> I have a set of patches coming up that introduce a rtnetlink API
> for adding/modifying/deleting software network devices. I would
> prefer if you could switch this driver over instead of doing the
> "create N devices during loading" that many current drivers do,
> leaving you with 20 unused devices after boot.
>   

Hi Patrick,

next Friday? Too late ;-)

Ok, your approach is indeed an interesting idea and we would look on it,
as we also not interested in creating unused devices and do any double
work.

> And it allows you have both loopback and non-loopback devices
> in case that would be useful.
>
> (..)
> Qdiscs might change skb->cb. Maybe use skb->sk?
>
>
>   

The loopback functionality in CAN is a bit tricky (maybe you can take a
look into the Documentation patch [7/7] at chapter 3.2 and 4.1.4).

The problem is, that we need a per socket(!) option that enables the
loopback for the sent CAN-Frames or not, so the information about
loopback a skb content or not is contained inside the skb sent down to
the CAN netdevice. If the CAN networkdevice is not capable to perform a
loopback itself (see dev->flags, chapter 6) the CAN core has to do the
loopback as a fallback.

The use of skb->cb is also needed in the receive path to recognize and
trash loopback'ed CAN-frames inside the *originating* socket (see
chapter 4.1.4).

Please consider the referenced documentation and give me a feedback, if
you have a better idea how to deal with variable loopbacks on driver
level (with is vital for CAN).

Thanks for your feedback,
Oliver