On 08/03/2017 07:03 AM, Oliver Hartkopp wrote:
>> Patch 9 removed the need for struct raw_sock::ifindex from the raw sock, by
>> using struct sock::sk_bound_dev_if from the generic socket structure.
> 
> Have a bad feeling on that - will comment later.

>> Patch 11: Cleans up the newly integrated CAN net namespace support.
> 
> need to review

>> Patches 13-14: Where to put the per device protocol specific memory? af_can
>> allocated it's memory during a netdev_notifier call, life cycle proves to be
>> rather complicated (see remove_on_zero_entries, etc...), adding the j1939
>> memory makes it even more compilcated. So I decided to allocate the memory
>> during the allocation if net_device. And this seems to work. More details in
>> the individual patches.
> 
> 'Seems to work' sounds frightening. There was a racy reason to have that 
> implementation as-is. Although this approach sounds interesting. Need to 
> review too.
> 
> I'm currently pretty busy at work.
> Please to not push these things without my ACK (as we had it with the 
> namespace support where I crashed my easter holiday to fix/implement all 
> the missing stuff to fit the merge window).

Have you found some time to look at these issues?

regards,
Marc

-- 
Pengutronix e.K.                  | Marc Kleine-Budde           |
Industrial Linux Solutions        | Phone: +49-231-2826-924     |
Vertretung West/Dortmund          | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686  | http://www.pengutronix.de   |