Re: [PATCH net-next] net/vxlan: Avoid unaligned access in vxlan_build_skb()

On (09/20/16 18:43), Jiri Benc wrote:

IPv6 header is certainly not 20 bytes.

vxlan encapsulates an IPv6/ethernet frame in a UDP/IPv4/ethernet packet.
the ipv4 header is 20 bytes (without options).
mld_newpack is dealing with the vxlan net_device in this case,
which has ->hard_header_len of 14, and ->needed_headroom of 64,
so hlen (used in mld_newpack) comes out to be 80.

If you see unaligned access, something is wrong. But I very much doubt
that the problem is at the place you're trying to fix. Could you share
the traces with us?

Exactly what traces do you want? I think you can check this easily
by using printk's predicated on IS_ALIGNED checks in vxlan_build_skb().
IPv6 MLD triggers this quite easily.

I will try out Hannes' solution (which makes sense) in a moment, 
and report back.

--Sowmini