Re: [PATCH net-next 1/2] udp: Do not require sock in udp_tunnel_xmit_skb

[Tom Herbert <therbert@google.com>]

On Tue, Jan 20, 2015 at 1:47 PM, Or Gerlitz <gerlitz.or@gmail.com> wrote:
> On Tue, Jan 20, 2015 at 7:36 PM, Tom Herbert <therbert@google.com> wrote:
>> On Sun, Jan 18, 2015 at 2:43 PM, Or Gerlitz <gerlitz.or@gmail.com> wrote:
>>> On Sat, Jan 17, 2015 at 8:18 PM, Tom Herbert <therbert@google.com> wrote:
>>>> The UDP tunnel transmit functions udp_tunnel_xmit_skb and
>>>> udp_tunnel6_xmit_skb include a socket argument. The socket being
>>>> passed to the functions (from VXLAN) is a UDP created for receive
>>>> side. The only thing that the socket is used for in the transmit
>>>> functions is to get the setting for checksum (enabled or zero).
>>>
>>> Tom, just to clarify - re the sockets usage in the transmit side,
>>> somewhere bind or alike is done on them such that we have multiple
>>> source UDP ports for given host VXLAN traffic. Here for example the
>>> sender host is 192.168.31.17 and two ports are seen here 54206 and
>>> 50795.
>>>
>>> Just wanted to make sure this series doesn't change that, since if
>>> this is the case, we introduce here a regression w.r.t RSS hash
>>> spreading from the outer UDP header data at the receiver side (which
>>> is the right thing to do, per your LKS session...)
>>>
>> Hi Or,
>>
>> Using or not using a socket on transmit should have no bearing to the
>> receive side. RSS works based on the hash of the UDP 5-tuple which
>> should include a source port set to a value for the inner flow. Since
>> the UDP socket is unconnected it should have no bearing on RFS or XPS
>> either.
>
> Hi Tom,
>
> You say "which should include a source port set to a value for the
> inner flow", well this series doesn't add such logic, nor I am fully
> clear what piece exactly is responsible for the fact that I see
> multiple source udp ports used from vxlan traffic flowing out of a
> certain host. I just wanted to make sure that these patches don't
> introduce a regression w.r.t to the **current** (not future) state of
> things, can you confirm this?

vxlan_xmit_one calls udp_flow_src_port to get a source port value
based on the encapsulated flow.