----- On Feb 23, 2017, at 2:49 PM, pablo pa...@netfilter.org wrote: > On Wed, Feb 22, 2017 at 01:47:17PM -0800, Tom Herbert wrote: >> On Wed, Feb 22, 2017 at 1:29 PM, Or Gerlitz <gerlitz...@gmail.com> wrote: >> > On Thu, Feb 16, 2017 at 11:58 PM, Andreas Schultz <aschu...@tpip.net> >> > wrote: >> >> Hi Or, >> >> ----- On Feb 16, 2017, at 3:59 PM, Or Gerlitz ogerl...@mellanox.com wrote: >> >> >> >>> Generate the source udp header according to the flow represented by >> >>> the packet we are encapsulating, as done for other udp tunnels. This >> >>> helps on the receiver side to apply RSS spreading. >> >> >> >> This might work for GTPv0-U, However, for GTPv1-U this could interfere >> >> with error handling in the user space control process when the UDP port >> >> extension header is used in error indications. >> > >> > >> > in the document you posted there's this quote "The source IP and port >> > have no meaning and can change at any time" -- I assume it refers to >> > v0? can we identify in the kernel code that we're on v0 and have the >> > patch come into play? >> > >> >> 3GPP TS 29.281 Rel 13, section 5.2.2.1 defines the UDP port extension and >> >> section 7.3.1 says that the UDP source port extension can be used to >> >> mitigate DOS attacks. This would IMHO imply that the user space control >> >> process needs to know the TEID to UDP source port mapping. >> > >> >> The other question is, on what is this actually hashing. When I understand >> >> the code correctly, this will hash on the source/destination of the >> >> orignal >> >> flow. I would expect that a SGSN/SGW/eNodeB would like the keep flow >> >> processing on a per TEID base, so the port hashing should be base on the >> >> TEID. >> > >> > is it possible for packets belonging to the same TCP session or UDP >> > "pseudo session" (given pair of src/dst ip/port) to be encapsulated >> > using different TEID? >> > >> > hashing on the TEID imposes a harder requirement on the NIC HW vs. >> > just UDP based RSS. >> >> This shouldn't be taken as a HW requirement and it's unlikely we'd add >> explicit GTP support in flow_dissector. If we can't get entropy in the >> UDP source port then IPv6 flow label is a potential alternative (so >> that should be supported in NICs for RSS). > > According to specs, section 4.4.2.3 Encapsulated T-PDU, TS 29.281. > > "The UDP Source Port is a locally allocated port number at the sending > GTP-U entity." > > Older specs that refer to GTP-U such as TS 09.60 and TS 29.060 also > state the same.
It is absolutely valid the choose any sending port you want. I only think you should know which port you did send on. TS 29.281, Sect. 5.2.2.1 defines the UDP port extension to be used in error indications. It provides the UDP source port of a G-PDU that triggered an error. If the send side does not know which port it uses to send, how can it use this indication to correlate an error? That's the reason I thought it would be better to add the UDP source port to the PDP context and allow the control path to assign the source port on context creation. Of course, this header is optional and the receiving side is not required to use it. About the RSS spreading in the receive side, I would think that a receiver would prefer to process all packets that belong to a give TEID with the same receive instance. So keeping the UDP source port for a given TEID stable would be beneficial. As far as I understand it, the hash used in the patch uses the source and destination values from the original flow. This would mean that GTP packets that belong to the same TEID would end up with different UDP source ports. So what about this as a compromise, we dd a UDP source port field to the PDP context, it defaults to 0 (zero), the control instance can optionally initialize this field, when we hit the xmit code and the port is non zero, use that value, if it is zero use the hash? Regards Andreas > > So Or patch looks fine to me.