Re: [PATCH can-next v3 0/3] Introduce diag support for CAN

[Filippo Storniolo <fstornio@redhat.com>]

On Wed, Oct 21, 2026 at 10:35:31AM +0200, Oliver Hartkopp wrote:
>
>The example does not solve the question about the port number having 
>only 16 bit.
>

Davide says:
Actually in ss, the port values (lport and rport in struct sockstat
in ss.c) are int.
Moreover, the port field in ss is been used even for things that
do not represent a "port". 
For example:
ss --packet -n
Netid Recv-Q Send-Q Local Address:Port      Peer Address:Port
p_dgr 0      0            [35085]:wlp0s20f3             *    
p_dgr 0      0            [34958]:*                     *    
p_dgr 0      0             [2054]:wlp0s20f3             *

when requesting AF_PACKET diag, the Address columnn contains the
protocol passed through socket(AF_PACKET, ..., protocol):
- 2054 -> 0x806 -> arp
- 34958 -> 0x888e -> pae
while port column contains the ifnameindex.

>If someone uses 'ss' as you propose will there be a specific support 
>for CAN?
>

At the moment it does not exist but in the future we might have.
Normally, when somebody adds a diag module they follow up with
the a patch for iproute2 (therefore: me).

>And what happens to older 'ss' binaries when used together 
>with the CAN diag support?
>

Nothing. Old binaries won't support it because they won't require
AF_CAN inside sdiag_family in the netlink request message.


The use of the couple Address/Port in ss as far as I see is arbitrary
and for this reason I think it can still offer a bit of support
when asked to show open CAN sockets in the kernel.

And in worse case scenario, it would still be possible to write
from scratch a userspace program that uses CAN diag and show
open CAN socket stats in a better fashion.

Best regards,
Filippo