Re: [PATCH linux-wpan 0/3] Fix handling of skbs of type PACKET_OTHERHOST

[Alexander Aring <alex.aring@gmail.com>]

Hi Jukka,

On Thu, Oct 02, 2014 at 01:12:44PM +0300, Jukka Rissanen wrote:
> Hi Alex,
> 
> On to, 2014-10-02 at 11:58 +0200, Alexander Aring wrote:
> > On Thu, Oct 02, 2014 at 11:54:30AM +0200, Alexander Aring wrote:
> > > Jukka,
> > > 
> > > On Thu, Oct 02, 2014 at 12:43:05PM +0300, Jukka Rissanen wrote:
> > > > Hi Simon,
> > > > 
> > > > On to, 2014-10-02 at 10:36 +0100, Simon Vincent wrote:
> > > > > I will add this patch when I rebase on bluetooth-next.
> > > > 
> > > > and thanks for doing that.
> > > > 
> > > 
> > > I see now, that the bluetooth layer never sets this value. I don't know
> > > how bluetooth deal with that. But you need to set PACKET_HOST if packet
> > > belongs to you, OTHERHOST if packets belongs not to you, or
> > > PACKET_BROADCAST if packet belongs to you, but is a broadcast.
> > > 
> > > Also for PACKET_MULTICAST, if bluetooth support MULTICAST frames.
> > > 
> > > 
> > > I think you need handling for this in bluetooth 6lowpan layer.
> > > 
> > 
> > and this should be fixed before Simon's patches. I add Marcel here in
> > cc, maybe he can bring some new information if bluetooth ever set's the
> > pkt_type in sk_buff.
> > 
> > Point is IPv6 needs this information and we should set it to a valid value.
> > ... and this is not always set PACKET_HOST, IPv6 also check on PACKET_BROADCAST.
> 
> Hmm, bluetooth 6lowpan is a bit different from ieee802154 as the
> bluetooth link is a point-to-point one and typically only uses link
> local addresses so normally every packet is really meant for that host.
> 
> I will investigate this a bit more, so perhaps it is best if Simon does
> not change the pkt_type field after all. Everything seems to work just
> fine with unicast and multicast packets at the moment (when pkt_type ==
> PACKET_HOST).
> 

Okay, then maybe bluetooth 6lowpan need to set it to PACKET_HOST always.

Another idea would be:
We could also set PACKET_BROADCAST if the IPv6 address is a
multicast/broadcast addresss.

But it looks wrong, the PACKET_FOO values should be set by mac layer only.


What I deteced is at [0]. There it checks if the "mac frame" was
broadcast or multicast. I don't know when this occurs, I need to check
this but it seems to be broken, when we always set the PACKET_HOST.



Simon had issues with routing when IPv6 Layer gets a skb with PACKET_HOST and
it was PACKET_OTHERHOSTS before.


Simon:

Maybe only send patch:

[PATCH linux-wpan 3/3] ieee802154: 6lowpan: Drop PACKET_OTHERHOST skbs in 6lowpan

this should fix your routing problems, but issue with [0] still exists.

- Alex

[0] http://lxr.free-electrons.com/source/net/ipv6/mcast.c#L1413