Patrick McHardy wrote:

Oliver Hartkopp wrote:
  

quoted

Patrick McHardy wrote:
    

Yes, its working, but only in certain combinations and you're breaking
the rules for skb->cb, making it impossible for other layers to use.
skb->sk is "stable" at the output path, the regular loopback device
orphans the skb in hard_start_xmit. So you can at least use it there.

  

quoted

Would therefore skb->cb left unchanged in my skb's? Or is there any flag
that can be set in the skb to keep the packet scheduler's hands off?
    

No, and I don't think we want a flag to signal that something is
violating the rules for skb->cb, there are other users of this
besides qdiscs.
  

Hm - regarding Patricks and Urs' last mails i just had the idea to put
the sk-reference that's needed for this special
CAN-only-loopback-functionality into the data section of the skb, e.g.
by introducing a new struct can_skb_data:

struct can_skb_data {
    struct can_frame cf;
    sock *txsk;
};

So instead of allocating the space of struct can_frame the alloc_skb()
would allocate the size of struct can_skb_data.
The needed txsk would be stable in any case and could be used like the
currently missused skb->cb. This would also lead to a type proof(!)
implementation.

In raw_rcv() in raw.c there could be a check for the size of struct
can_skb_data first before checking the txsk - this would also guarantee
the backward compatibility for current CAN drivers that allocate only
the size of struct can_frame. For me this looks like a safe and
compatible (Kernel & CAN) solution.

Any objections/comments for this approach?

Best regards,
Oliver