4over6 Transit Solution using IP Encapsulation and MP-BGP Extensions

Due to the lack of IPv4 address space, more and more IPv6 networks have been deployed not only on edge networks, but also on backbone networks. However, there are still a large number of legacy IPv4 hosts and applications in the coming years. The emerging and growing deployment of IPv6 networks, in particular IPv6 backbone networks, will introduce cases where connectivity with IPv4 networks is desired. Some IPv6 backbones will need to offer transit services to attached IPv4 access networks. The ideal method to achieve this would be to encapsulate and then transport the IPv4 payloads inside IPv6 tunnels spanning the backbone. There are some IPv6/IPv4-related tunneling protocols and mechanisms defined in the literature, but most of these existing techniques focus on the problem of IPv6 over IPv4, rather than the case of IPv4 over IPv6. Encapsulation methods, (e.g. ) specified for generic packet tunneling in IPv6 do exist and has been implemented. However they do not offer an easy means to provision such tunnels which can place a manual configuration burden on the operator, in particular if the number of required tunnels grows large. Thus, new techniques are needed to automatically create tunnels across an IPv6 backbone network containing IPv4 payloads. The mechanisms defined in this document are referred to as 4over6. The 4over6 mechanism concerns two aspects: the control plane and the data plane. The control plane needs to address the problem of how to setup an IPv4 over IPv6 tunnel in an automatic and scalable fashion between a large number of edge routers. This document defines extensions to MP-BGP employed to communicate tunnel end-point information and establish 4over6 tunnels between dual-stack Provider Edge (PE) routers positioned at the edge of the IPv6 backbone network. Once the 4over6 tunnel is in place, the data plane focuses on the packet forwarding processes of encapsulation and decapsulation.

In the topology shown in figure 1, a number of IPv6-only P routers compose a native IPv6 backbone. The PE routers are dual-stack and referred to as 4over6 PE routers. The IPv6 backbone acts as a transit core to transport IPv4 packets across the IPv6 backbone. This enables each of IPv4 access islands to communicate with each other via 4over6 tunnels spanning the IPv6 transit core.

Figure 1: IPv4 over IPv6 network topology As shown in figure 1, there are multiple dual-stack PE routers connected to the IPv6 transit core. In order for the ingress 4over6 PE router to forward an IPv4 packet across the IPv6 backbone to the correct egress 4over6 PE router, the ingress 4over6 PE router must learn which IPv4 destination prefixes are reachable through each egress 4over6 PE router. MP-BGP will be extended to distribute the destination IPv4 prefix information between peering dual-stack PE routers. Section 4 of this document presents the definition of the 4over6 protocol field in MP-BGP and section 5 describes MP-BGP's extended behavior in support of this capability. After the ingress 4over6 PE router learns the correct egress 4over6 PE router via MP-BGP, it will forward the packet across the IPv6 backbone using IP encapsulation. The egress 4over6 PE router will receive the encapsulated packet, remove the IPv6 header and then forward the original IPv4 packet to its final IPv4 destination. Section 6 describes the procedure of packet forwarding.

An implementatoin of the 4over6 mechanisms described in this document was developed, tested and deployed on Linux with kernel version 2.4. The prototype system is composed of three components: packet forwarding, the encapsulation table and MP-BGP extensions. The packet forwarding and encapsulation table are Linux kernel modules and the MP-BGP extension was developed by extending Zebra routing software. The following sections will discuss these parts in detail.

Forwarding an IPv4 packet through the IPv6 transit core includes 3 parts: encapsulation of the incoming IPv4 packet with the IPv6 tunnel header; transmission of the encapsulated packet over the IPv6 transit backbone; and decapsulation of the IPv6 header and forwarding of the original IPv4 packet. Native IPv6 routing and forwarding are employed in the backbone network since the P routers take the 4over6 tunneled packets as just native IPv6 packets. Therefore, 4over6 packet forwarding involves only the encapsulation process and the decapsulation process, both of which are peformed on the 4over6 PE routers.

Tunnel Tunnel Entry-Point Exit-Point Node Node +-+ IPv4 +--+ IPv6 Transit Core +--+ IPv4 +-+ |S|-->--//-->--|PE|=====>=====//=====>=====|PE|-->--//-->--|D| +-+ +--+ +--+ +-+ Original Ingress PE Egress PE Original Packet (Encapsulation) (Decapsulation) Packet Source Destination Node Node ]]> Figure 2: Packet forwarding along 4over6 Tunnel As shown in Figure 2, packet encapsulation and decapsulaion are both on the dual-stack 4over6 PE routers. Figure 3 shows the format of packet encapsulation and decapsulation.

The encapsulation format to apply is IPv4 encapsulated in IPv6 as outlined in .

Each 4over6 PE router maintains an encapsulation table as depicted in Figure 4. Each entry in the encapsulation table consists of an IPv4 prefix and its corresponding IPv6 address. The IPv4 prefix is a particular network located in an IPv4 access island network. The IPv6 address is the 4over6 virtual interface (VIF) address of the 4over6 PE router that the IPv4 prefix is reachable through. The encapsulation table is built and maintained using local configuration information and MP-BGP advertisements received from remote 4over6 PE routers. The 4over6 VIF is an IPv6 /128 address that is locally configured on each 4over6 router. This address, as an ordinary global IPv6 address, must also be injected into the IPv6 IGP so that it is reachable across the IPv6 backbone.

Figure 4: Encapsulation Table When an IPv4 packet arrives at the ingress 4over6 PE router, a lookup in the local IPv4 routing table will result in a pointer to the local encapsulation table entry with the matching destination IPv4 prefix. There is a corresponding IPv6 address in the encapsulation table. The IPv4 packet is encapsulated in an IPv6 header. The source address in the IPv6 header is the IPv6 VIF address of the local 4over6 PE router and the destination address is the IPv6 VIF address of the remote 4over6 PE router contained in the local encapsulation table. The packet is then subjected to normal IPv6 forwarding for transport across the IPv6 backbone. When the encapsulated packet arrives at the egress 4over6 PE router, the IPv6 header is removed and the original IPv4 packet is forwarded to the destination IPv4 network based on the outcome of the lookup in the IPv4 routing table contained in the egress 4over6 PE router.

Each 4over6 PE router possesses an IPv4 interface connected to an IPv4 access network(s). It can peer with other IPv4 routers using IGP or BGP routing protocols to exchange local IPv4 routing information. Routing information can also be installed on the 4over6 PE router using static configuration methods. Each 4over6 PE also possesses at least one IPv6 interface to connect it into the IPv6 transit backbone. The 4over6 PE typically uses IGP routing protocols to exchange IPv6 backbone routing information with other IPv6 P routers. The 4over6 PE router will also form an MP-iBGP peering relationship with other 4over6 PE routers connected to the IPv6 backbone network. The use of MP-iBGP suggests that the participating 4over6 PE routers that share a route reflector or form a full mesh of TCP connections are contained in the same autonomous system (AS). This implementation is in fact only deployed over a single AS. This was not an intentional design constraint but rather reflected the single AS topology of the CNGI-CERNET2 national IPv6 backbone used in the testing and deployment of this solution.

When a 4over6 PE router learns routing information from the locally attached IPv4 access networks, the 4over6 MP-iBGP entity should process the information as follows: Install and maintain local IPv4 routing information in the IPv4 routing database. Install and maintain new entries in encapsulation table. Each entry should consist of the IPv4 prefix and the local IPv6 VIF address. Advertise the new contents of the local encapsulation table in the form of MP_REACH_NLRI update information to remote 4over6 PE routers. The format of these updates is as follows: AFI = 1 (IPv4) SAFI = 67 (4OVER6) NLRI = IPv4 network prefix Network Address of Next Hop = IPv6 address of its 4over6 VIF A new SAFI for this solution was obtained as SAFI_4OVER6 (67) from IANA. We call BGP update with SAFI being 67 as 4over6 routing information.

A local 4over6 PE router will receive MP_REACH_NLRI updates from remote 4over6 routers and use that information to populate the local encapsulation table and the BGP routing database. After validating correctness of the received attribute, the following procedures are used to update the local encapsulation table and redistribute to local IPv4 routing table: Validate the received BGP update packet as 4over6 routing information by AFI = 1 (IPv4) and SAFI = 67 (4OVER6) Extract the IPv4 network address from the NLRI field and install as the IPv4 network prefix Extract the IPv6 address from the Network Address of the Next Hop field and place that as an associated entry next to the IPv4 network index. (Note this describes the update of local encap table.) Install and maintain a new entry in encapsulation table with the extracted IPv4 prefix and its corresponding IPv6 address. Redistribute the new 4over6 routing information to local IPv4 routing table. Set the destination network prefix as the extracted IPv4 prefix, set the Next Hop as Null, and Set the OUTPUT Interface as the 4over6 VIF on the local 4over6 PE router. Therefore, when an ingress 4over6 PE router receives an IPv4 packet, the lookup in its IPv4 routing table will have a result of the output interface as the local 4over6 VIF, where the incoming IPv4 packet will be encapsulated with a new IPv6 header as indicated in the encapsulation table.

A prototype of the 4over6 solution is implemented and deployed on CNGI-CERNET2. CNGI-CERNET2 is one of the China Next Generation Internet (CNGI) backbones, operated by the China Education and Research Network (CERNET). CNGI-CERNET2 connects approximately 25 core nodes distributed in 20 cities in China at speeds of 2.5-10 Gb/s. The CNGI-CERNET2 backbone is IPv6-only with some attached customer premise networks (CPN) being dual-stack. The CNGI-CERNET2 backbone, attached CNGI-CERNET2 CPNs, and CNGI-6IX Exchange all have globally unique AS numbers. This IPv6 backbone is used to provide transit IPv4 services for customer IPv4 networks connected via 4over6 PE routers to the backbone.

Figure 5 shows 4over6 deployment network topology.

Figure 5: 4over6 deployment network topology The IPv4-only access networks are equipped with servers and clients running different applications. The 4over6 PE routers are deployed at 8 x IPv6 nodes of CNGI-CERNET2, located in 7 universities and 5 cities across China. As suggested in figure 5 some of the IPv4 access networks are connected to both IPv6 and IPv4 networks and others are only connected to the IPv6 backbone. In the deployment, users in different IPv4 networks can communicate with each other through 4over6 tunnels.

A number of 4over6 PE routers were deployed and configured to support the 4over6 transit solution. MP-BGP peerings were established, and successful distribution of 4over6 SAFI information occured. Inspection of the BGP routing and encapsulation tables confirmed that the correct entries were sent and received. ICMP ping traffic indicated that IPv4 packets were successfully transiting the IPv6 backbone. In addition other application protocols were successfully tested per the following: HTTP. A client running Internet Explorer in one IPv4 client network was able to access and download multiple objects from an HTTP server located in another IPv4 client network. P2P. BitComet software running on several PCs placed in different IPv4 client networks were able to find each other and share files. Other protocols, including FTP, SSH, IM(MSN, GTalk) and Multimedia Streaming, all functioned correctly.

The 4over6 solution was presented at the IETF Softwires Working Group Interim meeting in Honk Kong in January 2006. The existence of this large-scale implementation and deployment clearly showed that MP-BGP could be employed to support tunnel setup in a scalable fashion across an IPv6 backbone. Perhaps most important was the use-case presented, that being an IPv6 backbone offering transit to attached client IPv4 networks. The 4over6 solution can be viewed as a precursor to softwires mesh framework. However there are several differences with this solution and the effort that emerged from the softwires working group called softwires mesh framework. MP-BGP Extensions. 4over6 employs a new SAFI (4OVER6) to convey client IPv4 prefixes between 4over6 PE routers. Softwires Mesh retains original AFI-SAFI designations but uses a modified MP_REACH_NLRI format to convey IPv4 NLRI prefix information with an IPv6 next_hop address. Encapsulation. 4over6 assumes IP-in-IP or it is possible to configure GRE. Softwires uses those two scenarios configured locally or for IP headers that require dynamic updating. As a result, the BGP encap safi is introduced in Softwires. Multicast. The basic 4over6 solution only implemented unicast communications. The multicast communications is specified in the Softwire Mesh Framework, and also supported by the multicast extension of 4over6. Use-Cases. The 4over6 solution in this document specifies the 4over6 use-case, which is also pretty easy to extend for the use-case of 6over4. The softwires mesh framework supports both 4over6 and 6over4.

A new SAFI value (67) was assigned by IANA for 4over6 BGP extension: SAFI_4OVER6.

Tunneling mechanisms, especially automatic ones, often have potential problems of DDoS attacks on the tunnel-entry point or tunnel-end point. However, since 4over6 BGP extension don't allocate resources to each flow or maintain the state of each flow, the 4over6 PE routers will have a capacity of enduring DDoS attacks as a common router. I-BGP peering relationship may be maintained over IPSec or other secure communications.

The emerging and growing deployment of IPv6 networks, in particular IPv6 backbone networks, will introduce cases where connectivity with IPv4 networks is desired. Some IPv6 backbones will need to offer transit services to attached IPv4 access networks. The 4over6 solution outlined in this document supports such a capability through an extension to MP-BGP to convey IPv4 routing information along with an associated IPv6 address. Basic IP encapsulation is used in the dataplane as IPv4 packets are tunneled through the IPv6 backbone. An actual implemention has been developed and deployed on the CNGI-CERNET2 IPv6 backbone.

During the design procedure of the 4over6 framework and definition of BGP-MP 4over6 extension, Professor Ke Xu gave the authors many valuable comments. The support of IETF softwire WG is also gratefully acknowledged with special thanks to David Ward and Mark Townsley for their rich experience and knowledge in this field. Many thanks to Yakov Rekhter for his helpful comments and advice. The deployment and test for the prototype system was conducted among 7 universities -- namely, Tsinghua University, Peking University, Beijing University of Post and Telecommunications, Shanghai Jiaotong University, Huazhong University of Science and Technology, Southeast University, South China University of Technology. The authors would like to thank everyone involved in this effort in these universities.