Re: [PATCH RFC can-next 3/3] can: add can diag interface

[Davide Caratti <dcaratti@redhat.com>]

On Thu, Apr 02, 2026 at 03:07:53PM +0200, Oliver Hartkopp wrote:
> Hello Filippo,
> 
> many thanks for your patches.
> 
> I'm not yet convinced we would need this kind of interface as many features
> in ss(8) are only relevant for IP and not really for CAN.
> 
> Btw. having an overview over PIDs and open sockets might be a nice
> informational feature.


hi, 

thanks for reading!
(an unsolicited answer to the above sentence, just because of the
"Suggested-by:" tag in patch 3/3 :) )

Enumerating open sockets with {uid, pid, interface, protocol} looked to me
a good-enough reason for adding support for ss(8): this is done also by
other non-IP socket families, such as AF_XDP [1] and AF_PACKET for the same
purpose. An alternative would be a tool like tcpstates [2], but it needs to
add tracepoints as well; the diag module (at the expense of serializing
'can_create()' with 'proto.release()' with a mutex) looked a cheaper
solution.

[1] speaking of this, I wonder whether we should pack some of the info we
have now in 'rep', e.g. the result of 'getname()', into a dedicated
attribute. That would save from returning zeros for modules that have stub
getname(), like can_bcm. WDYT?
[2] https://github.com/iovisor/bcc/blob/master/tools/tcpstates.py

-- 
davide