Hi Xiaohu, On 5/23/2016 7:59 PM, Xuxiaohu wrote: > > Hi Joe, > > > > Thanks for your response. Please see my response with [Xiaohu] inline. > > > > *From:*Joe Touch [mailto:[email protected]] > *Sent:* Tuesday, May 24, 2016 12:31 AM > *To:* Xuxiaohu; Fred Baker (fred); Wassim Haddad > *Cc:* [email protected] > *Subject:* Re: [Int-area] Call for adoption of > draft-xu-intarea-ip-in-udp-03 > > > > > > On 5/22/2016 7:44 PM, Xuxiaohu wrote: > > Hi Joe,, > > > > As for the four things that you have pointed out (see below), > > > > We know of at least four things that tunnels need that IP-in-UDP > ignores: > > > > - stronger checksums > > > > [Xiaohu] it said in the draft that > > > > “ For IPv4 UDP encapsulation, this field is RECOMMENDED to be set > > to zero for performance or implementation reasons because the > > IPv4 header includes a checksum and use of the UDP > checksum is > > optional with IPv4. For IPv6 UDP encapsulation, the IPv6 > > header does not include a checksum, so this field MUST > contain > > a UDP checksum that MUST be used as specified in [RFC0768 > <https://tools.ietf.org/html/rfc0768>] and > > [RFC2460 <https://tools.ietf.org/html/rfc2460>] unless > one of the exceptions that allows use of UDP > > zero-checksum mode (as specified in [RFC6935 > <https://tools.ietf.org/html/rfc6935>]) applies.” > > > > In other words, the stronger checksum (i.e., non-zero checksum of > UDP) is strongly recommended especially for IPv6 UDP case. > Furthermore, as stated in RFC6935, “Tunnel protocols that > encapsulate IP will generally be safe for > > deployment, because all IPv4 and IPv6 packets include at least one > > checksum at either the network or transport layer.” This > statement is just applied to the IP-in-UDP case. >
The problem is that this advice overlooks the need to account for errors of fragmentation and reassembly (see below). Something stronger than the Internet checksum is typically useful, e.g., as was developed for SEAL (e.g., RFC5320). > > > > > - fragmentation support > > > > [Xiaohu] It said in the draft that > > “ Ingress AFBRs MUST NOT fragment I-IP packets > (i.e., UDP encapsulated > > IP packets), and when the outer IP header is IPv4, ingress > AFBRs MUST > > set the DF bit in the outer IPv4 header. It is strongly > RECOMMENDED > > that I-IP transit core be configured to carry an MTU at least large > > enough to accommodate the added encapsulation headers. > Meanwhile, it > > is strongly RECOMMENDED that Path MTU Discovery [RFC1191 > <https://tools.ietf.org/html/rfc1191>] [RFC1981 > <https://tools.ietf.org/html/rfc1981>] > > is used to prevent or minimize fragmentation. Once an ingress AFBR > > needs to perform fragmentation on an E-IP packet before > > encapsulating, it MUST use the same source UDP port for all > > fragmented packets so as to ensures these fragmented packets are > > > > > > > > Xu, et al. Expires August 3, 2016 > [Page 5] > > ------------------------------------------------------------------------ > > <https://tools.ietf.org/html/draft-xu-intarea-ip-in-udp-03#page-6> > > Internet-Draft Encapsulating IP in UDP > January 2016 > > > > > > always forwarded on the same path. In a word, IP-in-UDP is just > > applicable in those Softwire network environments where > fragmentation > > on the tunnel layer is not needed.” > > > > In other words, fragmentation on I-IP packets (i.e., UDP > encapsulated IP packets) must be avoided. fragmentation on E-IP > packets should be avoided by some means (e.g., I-IP transit core > is configured to carry an MTU at least large enough to accommodate > the added encapsulation headers or Path MTU discovery is enabled > on the UDP tunnels). Fragmentation on E-IP packets is possible > only when the E-IP packets are IPv4 packets and the DF bit is not > set. > While this is a nice goal, it's not viable. In order to provide a viable IPv6 tunnel, the payload MUST transit IPv6 payloads of at least 1280 bytes.However, there's no reason to assume that the UDP/IPv6 tunnel can support 1280 + UDP + IPv6. So you have only one choice - you can deploy these tunnels only where you can guarantee paths that support 1280+8+40 byte payloads over the tunnel path. However, there's no way to ensure that solely by assuming IPv6 along that path. I.e., you MUST support source fragmentation at the ingress at the outer IPv6 layer (because UDP doesn't have support for fragmentation and reassembly). If you make this requirement, you can handle IPv6 over the tunnel. IPv4 will work over a UDP/IPv6 tunnel fairly easily, even assuming the common typical IPv4 MTU (576, even though that's really the reassembled MTU, not the path MTU). IPv4 over IPv4 is much more difficult, and not only requires fragmentation at the outer packet but also additional reassembly requirements at the egress above and beyond the IPv4 minimums. > > > - signalling support (e.g., to test whether a tunnel is up or > > to measure MTUs) > > > > [Xiaohu] It said in the draft that > > “it is strongly RECOMMENDED that Path MTU Discovery [RFC1191 > <https://tools.ietf.org/html/rfc1191>] [RFC1981 > <https://tools.ietf.org/html/rfc1981>] > > is used to prevent or minimize fragmentation.” > > > > > > - support for robust ID fields (related to fragmentation, > > e.g., to overcome the limits of IPv4 ID as per RFC 6864) > > in other words, the PMTUD signaling is recommended to be supported. > PMTUD is a recipe for black-holing. PLPMTUD is preferred for many reasons. However, PLPMTUD only ensures proper discovery of the tunnel MTU. It does not describe how the ICMPs inside the tunnel relate to those that are generated by the ingress. Again, for all of these, see draft-ietf-intarea-tunnels Joe > > > [Xiaohu] Is it enough to add a sentence like “to overcome the > limits of IPv4 ID as per RFC 6864 when performing fragmentation on > E-IP IPv4 packets, the specifications as described in RFC6864 MUST > be followed.”? > > > > Best regards, > > Xiaohu > > > > > > I’m wondering whether those issues are specific to IP-in-UDP or > not. In other words, are those issue also applicable to other > X-in-UDP approaches (where X could be LISP, TRILL, VXLAN, > VXLAN-GPE, GENEVE, GUE, NSH or MPLS)? > > > That set of issues applies to all IP-in-(xlist), where xlist includes > IP again. The issue is what IP expects and what the tunnel exports. > > The other items I pointed out are specific to transport-in-UDP or this > specific approach. > > Joe >
_______________________________________________ Int-area mailing list [email protected] https://www.ietf.org/mailman/listinfo/int-area
