Re: [PATCH net] geneve: allow changing DF behavior after creation

[Stefano Brivio <sbrivio@redhat.com>]

On Thu, 18 Jun 2020 15:03:47 +0200
Sabrina Dubroca <sd@queasysnail.net> wrote:

> 2020-06-18, 12:26:29 +0200, Stefano Brivio wrote:
> > On Thu, 18 Jun 2020 12:13:22 +0200
> > Sabrina Dubroca <sd@queasysnail.net> wrote:
> >   
> > > Currently, trying to change the DF parameter of a geneve device does
> > > nothing:
> > > 
> > >     # ip -d link show geneve1
> > >     14: geneve1: <snip>
> > >         link/ether <snip>
> > >         geneve id 1 remote 10.0.0.1 ttl auto df set dstport 6081 <snip>
> > >     # ip link set geneve1 type geneve id 1 df unset
> > >     # ip -d link show geneve1
> > >     14: geneve1: <snip>
> > >         link/ether <snip>
> > >         geneve id 1 remote 10.0.0.1 ttl auto df set dstport 6081 <snip>
> > > 
> > > We just need to update the value in geneve_changelink.
> > > 
> > > Fixes: a025fb5f49ad ("geneve: Allow configuration of DF behaviour")
> > > Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
> > > ---
> > >  drivers/net/geneve.c | 1 +
> > >  1 file changed, 1 insertion(+)
> > > 
> > > diff --git a/drivers/net/geneve.c b/drivers/net/geneve.c
> > > index 75266580b586..4661ef865807 100644
> > > --- a/drivers/net/geneve.c
> > > +++ b/drivers/net/geneve.c
> > > @@ -1649,6 +1649,7 @@ static int geneve_changelink(struct net_device *dev, struct nlattr *tb[],
> > >  	geneve->collect_md = metadata;
> > >  	geneve->use_udp6_rx_checksums = use_udp6_rx_checksums;
> > >  	geneve->ttl_inherit = ttl_inherit;
> > > +	geneve->df = df;  
> > 
> > I introduced this bug as I didn't notice the asymmetry with VXLAN,
> > where vxlan_nl2conf() takes care of this for both new links and link
> > changes.  
> 
> Yeah, I didn't notice either :/
> 
> > Here, this block is duplicated in geneve_configure(), which,
> > somewhat surprisingly given the name, is not called from
> > geneve_changelink(). Did you consider factoring out (at least) this
> > block to have it shared?  
> 
> Then I'd have to introduce another lovely function with an absurdly
> long argument list. I'd rather clean that up in all of geneve and
> introduce something like struct vxlan_config, but it's a bit much for
> net. I'll do that once this fix finds its way into net-next.

Yeah, sure, I didn't mean you should simply copy and paste that
somewhere. Either something like struct vxlan_config used around the
current logic, or perhaps even better, something like vxlan_nl2conf()
does.

I didn't check whether something in GENEVE is so special compared to
VXLAN as to prevent this, though.

-- 
Stefano