Re: [PATCH net] team: replace team lock with rtnl lock

[Stanislav Fomichev <stfomichev@gmail.com>]

On 05/22, Tetsuo Handa wrote:
> On 2025/05/22 10:00, Tetsuo Handa wrote:
> > On 2025/05/22 3:00, Jakub Kicinski wrote:
> >> And as he pointed out this patch promptly generates all sort 
> >> of locking warnings, please test this properly.
> > 
> > I didn't get any compile-time warnings, and
> > https://lkml.kernel.org/r/682e6b1f.a00a0220.2a3337.0007.GAE@google.com didn't
> > get any run-time locking warnings.
> > 
> > What locking warnings did you get? Is there an automated testing environment
> > (like https://lkml.kernel.org/r/66a4b1a7.050a0220.12c792.8f9e@mx.google.com )
> > which I can use for testing this patch?
> > 
> 
> Ah, I got which posts you are referring to. I was failing to receive Jiri's mails
> because my spam filter setting was sending mails from .us domain to trash.
> Now I removed the .us entry.
> 
> 
> 
> Jiri Pirko wrote:
> > Sat, May 17, 2025 at 09:32:20AM +0200, penguin-kernel@I-love.SAKURA.ne.jp wrote:
> > 
> > [..]
> > 
> > >@@ -2319,13 +2301,12 @@ static struct team *team_nl_team_get(struct genl_info *info)
> > > 	}
> > > 
> > > 	team = netdev_priv(dev);
> > >-	mutex_lock(&team->lock);
> > > 	return team;
> > > }
> > 
> > 
> > Why do you think this is safe?
> > 
> > Rtnl is held only for set doit.
> 
> I assumed that the caller already held rtnl lock.
> 
> > 
> > 
> > > 
> > > static void team_nl_team_put(struct team *team)
> > > {
> > >-	mutex_unlock(&team->lock);
> > >+	ASSERT_RTNL();
> > 
> > Did you test this? How? Howcome you didn't hit this assertion?
> 
> Tests using syzbot's reproducer did not hit this assertion.
> 
> > 
> > 
> > > 	dev_put(team->dev);
> > > }
> > > 

[..]

> Anyway, we can't remove team lock. Too bad.

I was hoping to see another revision, but just noticed this part. Can
you share more on why we can't remove the team lock? I can try to
give it a stab if you're not planning to send a follow up...