Re: [PATCH net-next 0/11] pull-request: can-next 2026-10-22

[Oliver Hartkopp <socketcan@hartkopp.net>]

On 22.10.26 14:41, Marc Kleine-Budde wrote:
> On 22.10.2026 13:19:08, Oliver Hartkopp wrote:
>> Unfortunately sashiko-bot complained about something new in my patch 11/11
>> and suggests to move some release_sock() calls 4 lines down behind
>> sk_error_report(sk) :-/
> 
> Is this review already available to the public?

Yes. It was some delayed v5 (PR 2026-10-21) review:

https://lore.kernel.org/linux-can/179266767682.3135.6448103178144570295@kernel.org/

- [Medium] After this patch, CAN_RAW serializes bind/setsockopt/release
   against NETDEV_UNREGISTER only with lock_sock().

This is the issue that sockopt() uses lock_sock() which might block the 
rtnl_lock in the notifier. This has been partly addressed in patch 11/11 
so that we are better than today's mainline status. On the TODO list.

- [Medium] can: raw/isotp: race in bind() overwrites ENODEV with 
ENETDOWN on unbind

This is a new topic of v6 which wasn't remarked by sashiko in the v5 
review before. Fixed in v7.

- [Low] isotp_bind() now publishes so->dev with WRITE_ONCE(so->dev, dev)
   inside the RCU lookup (net/can/isotp.c:1623).

Fixed in v6.

Pre-existing issues:
- [Low] isotp_bind() calls can_rx_register() twice: for rx_id when
   isotp_register_rxid(so) is true (net/can/isotp.c:1648-1650), and for…

Groundhog day. Not critical. On the TODO list.
> 
>> Would you like to have that re-spin PR to be named as "net" PR ?
>> And when would it be best to be sent?
> 
> I'll rebase the changes to net, and usually "net" is always open.
> (Correct me, if this is outdated information.)

Ok. As you suggested to work on the two TODO topics, I'll post a new 
series on CAN ML and wait for the cross-review.

This should be fine for a 'net' PR then, after the current net-next hit 
Linus' tree.

Best regards,
Oliver