idnits 2.17.1 draft-marques-l3vpn-end-system-07.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) ** There are 2 instances of too long lines in the document, the longest one being 2 characters in excess of 72. ** The abstract seems to contain references ([RFC4364]), which it shouldn't. Please replace those with straight textual mentions of the documents in question. == There are 11 instances of lines with private range IPv4 addresses in the document. If these are generic example addresses, they should be changed to use any of the ranges defined in RFC 6890 (or successor): 192.0.2.x, 198.51.100.x or 203.0.113.x. ** The document seems to lack a both a reference to RFC 2119 and the recommended RFC 2119 boilerplate, even if it appears to use RFC 2119 keywords. RFC 2119 keyword, line 307: '... VPN Forwarder MAY be co-located in ...' RFC 2119 keyword, line 337: '...irtual MAC address which SHOULD be the...' RFC 2119 keyword, line 339: '... virtual MAC address SHALL default to the VRRP [RFC5798] virtual...' RFC 2119 keyword, line 400: '... MUST implement the following functi...' RFC 2119 keyword, line 426: '...he VPN Forwarder MAY support the abili...' (18 more instances...) Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (August 2012) is 4272 days in the past. Is this intentional? Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Unused Reference: 'RFC4271' is defined on line 883, but no explicit reference was found in the text == Unused Reference: 'I-D.marques-sdnp-flow-spec' is defined on line 916, but no explicit reference was found in the text ** Obsolete normative reference: RFC 5575 (Obsoleted by RFC 8955) Summary: 5 errors (**), 0 flaws (~~), 4 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group P. Marques 3 Internet-Draft Contrail Systems 4 Intended status: Standards Track L. Fang 5 Expires: January 31, 2013 Cisco Systems 6 P. Pan 7 Infinera Corp 8 A. Shukla 9 Juniper Networks 10 M. Napierala 11 AT&T Labs 12 N. Bitar 13 Verizon 14 August 2012 16 BGP-signaled end-system IP/VPNs. 17 draft-marques-l3vpn-end-system-07 19 Abstract 21 This document describes a solution in which the control plane 22 protocol specified in BGP/MPLS IP VPNs [RFC4364] is used to provide a 23 Virtual Network service to end-systems. These end-systems may be 24 used to provide network services or may directly host end-to-end 25 applications. 27 Status of this Memo 29 This Internet-Draft is submitted in full conformance with the 30 provisions of BCP 78 and BCP 79. 32 Internet-Drafts are working documents of the Internet Engineering 33 Task Force (IETF). Note that other groups may also distribute 34 working documents as Internet-Drafts. The list of current Internet- 35 Drafts is at http://datatracker.ietf.org/drafts/current/. 37 Internet-Drafts are draft documents valid for a maximum of six months 38 and may be updated, replaced, or obsoleted by other documents at any 39 time. It is inappropriate to use Internet-Drafts as reference 40 material or to cite them other than as "work in progress." 42 This Internet-Draft will expire on January 31, 2013. 44 Copyright Notice 46 Copyright (c) 2012 IETF Trust and the persons identified as the 47 document authors. All rights reserved. 49 This document is subject to BCP 78 and the IETF Trust's Legal 50 Provisions Relating to IETF Documents (http://trustee.ietf.org/ 51 license-info) in effect on the date of publication of this document. 52 Please review these documents carefully, as they describe your rights 53 and restrictions with respect to this document. Code Components 54 extracted from this document must include Simplified BSD License text 55 as described in Section 4.e of the Trust Legal Provisions and are 56 provided without warranty as described in the Simplified BSD License. 58 Table of Contents 60 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 2 61 1.1. Terminology . . . . . . . . . . . . . . . . . . . . . . . 2 62 2. Requirements . . . . . . . . . . . . . . . . . . . . . . . . . 3 63 3. Applicability of BGP IP VPNs . . . . . . . . . . . . . . . . . 4 64 4. Virtual network end-points . . . . . . . . . . . . . . . . . . 6 65 5. VPN Forwarder . . . . . . . . . . . . . . . . . . . . . . . . 8 66 6. XMPP signaling protocol . . . . . . . . . . . . . . . . . . . 10 67 7. End-System Route Server behavior . . . . . . . . . . . . . . . 14 68 8. Operational Model . . . . . . . . . . . . . . . . . . . . . . 14 69 9. Security Considerations . . . . . . . . . . . . . . . . . . . 17 70 10. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 18 71 11. References . . . . . . . . . . . . . . . . . . . . . . . . . . 18 72 11.1. Normative References . . . . . . . . . . . . . . . . . . 18 73 11.2. Informational References . . . . . . . . . . . . . . . . 18 74 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 19 76 1. Introduction 78 This document describes the requirements for a network virtualization 79 solution that provides an IP service to end-system virtual 80 interfaces. It then discusses how the BGP IP VPNs [RFC4364] control 81 plane can be used to provide a solution that meets these 82 requirements. Subsequent sections provide a detailed discussion of 83 the control and forwarding plane components. 85 In BGP IP VPNs, Customer Edge (CE) interfaces connect to a Provider 86 Edge (PE) device which provides both the control plane and VPN 87 encapsulation functions required to implement a Virtual Network 88 service. This document decouples the control plane and forwarding 89 functionality of the PE device in order to enable the forwarding 90 functionality to be implemented in multiple devices. For instance, 91 the forwarding function can be implemented directly on the operating 92 system of application servers or network appliances. 94 1.1. Terminology 96 This document makes use of the following terms: 98 End-System Route Server A software application that implements the 99 control plane functionality of a BGP IP VPN PE device and a XMPP 100 server that interacts with VPN Forwarders. 102 Virtual Interface An interface in an end-system that is used by a 103 virtual machine or by applications. It performs the role of a CE 104 interface in a BGP IP VPN network. 106 VPN Forwarder The forwarding component of a BGP IP VPN PE device. 107 This functionality may be co-located with the virtual interface or 108 implemented by an external device. 110 2. Requirements 112 Network Virtualization is used in both service provider as well as 113 enterprise networks to support multi-tenancy, network-based access 114 control. It may also be used to facilitate end-system mobility. 116 Multi-tenancy allows a physical network to provide services to 117 multiple "customers" or "tenants", whether these are external 118 entities in the case of a Service Provider providing managed VPN 119 services or internal departments sharing an IT facility. Multi- 120 tenancy requires isolation of traffic and routing information between 121 tenants. 123 Within a tenant, it is often required to create multiple distinct 124 virtual networks, in order to be able to provide network-based access 125 control. In this service model, each virtual network behaves as a 126 "Closed User Group" (CUG) of end-systems that are allowed to exchange 127 traffic freely, while traffic between virtual networks is subject to 128 access controls. This scenario can be found in both enterprise 129 campus networks, branch offices and data-centers. 131 It is often the case when network access control is used, that the 132 traffic patterns are such that there is significantly more traffic 133 crossing a CUG boundary than staying within such boundary. As an 134 example, in campus networks it is common to segregate users into CUGs 135 based on some classification such as the user's department. Campus 136 networks often see traffic patterns in which almost all the traffic 137 flows northbound to the data-center or internet boundaries. Similar 138 traffic patterns can be found in multi-tier applications in IT data- 139 centers. 141 End-systems are often configured to expect the concept of IP subnet 142 to match its closed user group. A network virtualization solution 143 should be able to provide this concept of IP subnet regardless of 144 whether the underlying implementation uses a multi-access network or 145 not. 147 End-system virtual interfaces should be able to directly access 148 multiple closed user groups without needing to traverse a gateway. 149 Network access policy should allow this access whether the source and 150 destination CUGs for a particular traffic flow belong to the same 151 tenant or different tenants. It is often the case that 152 infrastructure services are provided to multiple tenants. One such 153 example is voice-over-IP gateway services for branch offices. 155 Independently, but often associated with the previous two functions, 156 IP mobility is another network function that can be implemented using 157 network virtualization. By abstracting the externally visible 158 network address from the underlying infrastructure address, mobility 159 can be implemented without having to recur to home agents or large L2 160 broadcast domains. Alternative techniques that are used in both 161 Service Provider as well as enterprise networks. 163 IP Mobility requires the ability to "move" a device without 164 disrupting its TCP (or UDP) transport sessions. These sessions often 165 deploy second or sub-second keepalives to detect application failure. 166 Experience with failure restoration in Service Provider networks 167 shows that fast-failure restoration often requires the pre- 168 provisioning of a restoration path. 170 IP Mobility can be a result of devices physically moving (e.g., a 171 WiFi enabled laptop) or workload being diverted between physical 172 systems such as network appliances or application servers. 174 3. Applicability of BGP IP VPNs 176 BGP IP VPNs [RFC4364] is the industry de-facto standard for providing 177 "closed user group" functionality in WAN environments. It is used by 178 service providers in environments where several millions of routes 179 are present. It supports both isolated VPNs as well as overlapping 180 VPNs (often referred to as "extranets"). 182 In its traditional usage in Service Provider networks, BGP IP VPN 183 functionality is implemented in a Provider Edge (PE) device that 184 combines both BGP signaling as well as VRF-based forwarding 185 functions. In practice, most PE devices in current use are multi- 186 component systems with the signaling and forwarding functionality 187 actually implemented in different processors attached to an internal 188 network. 190 This document assumes a similar separation of functionality in which 191 software appliances, the End-System Route Servers, implement the 192 control plane functionality of a PE device and a VPN Forwarder 193 implements the forwarding function usually found in a PE device 194 "line-card". The VPN Forwarder functionality may be co-located with 195 the end-system virtual interface (e.g., implemented in the hypervisor 196 switch or host OS network drivers). It may also be external to the 197 end-system residing in a data-center switch or specialized appliance. 199 Operationally, BGP IP VPN technology has several important 200 characteristics: 202 It has a high-level of aggregation between customer interfaces and 203 managed entities (Provider Edge devices). 205 It defines VPNs as policies, allowing an interface to directly 206 exchange traffic with multiple VPNs and allowing for the topology 207 of the virtual network to be modified by modifying the policy 208 configuration. 210 It scales horizontally in terms of event propagation. By 211 increasing the number of signaling devices implementing the PE 212 control plane, it is possible to decrease the load on each 213 signaling device when it comes to propagating events that 214 originate in a specific location and must be propagated across the 215 network. 217 The last point is particularly relevant to the convergence 218 characteristics required for large scale deployments. BGP's 219 hierarchical route distribution capabilities allow a deployment to 220 divide the workload by increasing the number of End-System Route 221 Servers. 223 As an example consider a topology in which 100 End-System Route 224 Servers are deployed in a network each serving a subset of the VPN 225 forwarding elements. The Route Servers inter-connect to two top- 226 level BGP Route Reflectors [RFC4456]. 228 If an event (i.e. a VPN route change) needs to be propagated from a 229 specific end-system to 10.000 clients randomly distributed across the 230 network, each of the End-System Route Servers must generate 100 231 updates to its respective downstream clients. 233 By modifying this topology such that another 100 End-System Route 234 Servers are added, then each Route Server is now responsible to 235 generate 50 client updates. This example illustrates the linear 236 scaling properties of BGP: doubling the number of Route Servers (i.e. 237 the processing capacity) reduces in half the number of updates 238 generated by each (i.e. load at each processing node). 240 The same horizontal scaling techniques can be applied to the Route 241 Reflector layer in the example above by subsetting the VPN Route 242 space according to some pre-defined criteria (for instance VPN route 243 target) and using a pair of Route Reflectors per subset. 245 In the previous example we assumed a dense membership in which all 246 Route Servers have local clients that are interested in a particular 247 event. BGP also optimizes the route distribution for sparse events. 249 The Route Target Constraint [RFC4684] extension, builds an optimal 250 distribution tree for message propagation based on VPN membership. 251 It ensures that only the PEs with local receivers for a particular 252 event do receive it also decreasing the total load on the upstream 253 BGP speaker. 255 In the WAN environment, BGP IP VPN control plane scaling is focused 256 not primarily on route convergence times but on memory footprint of 257 embedded devices. While memory footprint does not have a similar 258 linear scaling behavior, memory technology available to software 259 appliances is often at 10x the scale of what is commonly found in WAN 260 environments. 262 The functionality present in the BGP IP VPN control plane addresses 263 the requirements specified in the previous section. Specifically, it 264 supports multiple potentially overlapping "groups", regular or "hub 265 and spoke" topologies and the scaling characteristics necessary. 267 The BGP IP VPN control plane supports not only the definition of 268 "closed user-groups" (VPNs in its terminology) but also the 269 propagation of inter-VPN traffic policies [RFC5575]. An application 270 of that mechanism to "end-system" VPNs is presented in [I-D.marques- 271 sdnp-flow-spec]. 273 Note that the signaling protocol itself is rather agnostic of the 274 encapsulation used on the wire as long as this encapsulation has the 275 ability to carry a 20 bit label. 277 Several network environments use a network infrastructure that is 278 only capable of providing an IP unicast service. In order to support 279 them, implementations of this document should support the MPLS in GRE 280 [RFC4023] encapsulation. Other encapsulations are possible, 281 including UDP based encapsulations. 283 4. Virtual network end-points 285 This document assumes that end-systems support one or more virtual 286 network interfaces in addition to a physical interface that is 287 associated with the underlying network infrastructure. Virtual 288 network interfaces can be associated with a restricted list of 289 applications via OS-dependent mechanisms, a Virtual Machine (VM), or 290 they can be used to provide network connectivity to all user 291 applications in the same way that a "VPN tunnel" interface is used to 292 provide access between an end-system (e.g., a laptop) and a remote 293 corporate network. 295 From an IP address assignment point of view, a virtual network 296 interface is addressed out of the virtual IP topology and associated 297 with a "closed user group" or VPN, while the physical interface of 298 the machine is addressed in the network infrastructure topology. As 299 a security measure, it is recommended that virtual and infrastructure 300 topologies never be allowed to exchange traffic directly. 302 Both static and dynamic IP address allocation can be supported. The 303 later assumes that the VPN Forwarder implements a DHCP relay or DHCP 304 proxy functionality. 306 A virtual network interface is connected to a VPN Forwarder. This 307 VPN Forwarder MAY be co-located in the end-system or external. 309 Traffic that ingresses or egresses through a virtual network 310 interface is routed at the VPN Forwarder which acts as the first-hop 311 router (in the virtual topology). The IP configuration on the client 312 side of this virtual network interface (e.g., in the guest OS) can 313 follow one of two models: 315 point-to-point interface model. 317 multipoint interface model. 319 In a point-to-point interface model, the VPN client routing table 320 (e.g., on the guest OS) contains the following routing entires: a 321 host route to the local IP address, a host route to the first-hop 322 router via the virtual interface and a default route to the first-hop 323 router. This is the model typically used in "VPN tunnel" 324 configurations or other access technologies such as cable deployments 325 or DSL. When this model is used, the first-hop router IP address is a 326 link-local address that is the same on all first-hop routers across a 327 specific deployment. This first-hop IP address should not change 328 when a virtual interface moves between different machines. 330 In a multi-point interface model, the VPN client routing table (e.g., 331 on the guest OS) contains the following routing entires: a host route 332 to the local IP address, a subnet route to the local interface and 333 optionally a default route to a specific router address within that 334 subnet. In this model, the VPN client IP stack will issue address 335 resolution requests for any IP addresses it considers to be directly 336 attached to the subnet. The VPN Forwarder shall answer all address 337 resolution requests with a virtual MAC address which SHOULD be the 338 same across all VPN Forwarders in a specific deployment. This 339 virtual MAC address SHALL default to the VRRP [RFC5798] virtual 340 router MAC address for Virtual Router Identifier (VRID) 1. 342 When the virtual topology first-hop router resides on the same 343 physical machine, the host OS is responsible to map the virtual 344 interface with a VPN specific routing table (without taking L2 345 addresses into consideration). In this case the mac-addresses known 346 to the guest OS are not used on the wire. 348 When the virtual topology first-hop router resides in an external 349 system (e.g., the first hop-switch) the virtual interface shall be 350 identified by the combination of the mac-address assigned to physical 351 interface of the end-system and a 802.1Q VLAN tag. The first-hop 352 switch should use a virtual router MAC address to answer any address 353 resolution queries. 355 Whenever an external VPN Forwarder is used and resiliency is desired, 356 the external VPN Forwarder should be redundant. It is desirable to 357 use VRRP as a mechanism to control the flow of traffic between the 358 end-system and the external VPN Forwarder. VRRP already defines the 359 necessary procedures to elect a single forwarder for a LAN. 361 This specification uses the VRRP virtual router MAC address as the 362 default L2 address for the VPN Forwarder as a client virtual 363 interface may move between locations where redundancy may not be 364 present. 366 While the VRRP Virtual Router MAC will be used to answer any address 367 resolution request made by the virtual interface client (e.g., the 368 guest VM) this does not imply that a single default router is elected 369 per virtual IP subnet. The ingress VPN Forwarder will perform an IP 370 forwarding decision based on the destination IP address of the 371 (payload) traffic. 373 VRRP router election is only relevant in selecting the VPN Forwarder 374 associated with a specific machine, when external forwarders are in 375 use. 377 5. VPN Forwarder 379 In this solution, the Host OS/Hypervisor in the end-system must 380 participate in the virtual network service. Given an end-system with 381 multiple virtual interfaces, these virtual interfaces must be mapped 382 onto the network by the guest OS such that applications on one 383 virtual interface are not allowed to impersonate another virtual 384 interface. 386 When VPN forwarder functionality is implemented by the Host OS/ 387 Hypervisor, intermediate systems in the network do not require any 388 knowledge of the virtual network topology. This can simplify the 389 design and operation of the physical network. 391 When it is not possible or desirable to add the VPN forwarding 392 functionality to the end-system, it may be implemented by an external 393 system, typically located as close as possible to the end-system 394 itself. 396 Both models, co-located and external VPN Forwarder can co-exist in a 397 deployment. 399 In order to implement the BGP IP VPN Forwarder functionality a device 400 MUST implement the following functionality: 402 Support for multiple "Virtual Routing and Forwarding" (VRF) 403 tables; 404 VRF route entries map prefixes in the virtual network topology 405 to a next-hop containing a infrastructure IP address and a 406 20-bit label allocated by the destination Forwarder. The VRF 407 table lookup follows the standard IP lookup (best-match) 408 algorithm. 410 Associate an end-system virtual interface with a specific VRF 411 table; 413 When the the Forwarder is co-located with the end-system, this 414 association is implemented by an internal mechanism. When the 415 Forwarder is external the association is performed using the 416 mac-address of the end-system and a IEEE 802.1Q tag that 417 identifies the virtual interface within the end-system. 419 Encapsulate outgoing traffic (end-system to network) according to 420 the result of the VRF lookup; 422 Associate incoming packets (network to end-system) to a VRF 423 according to the 20-bit label contained immediately after the GRE 424 header; 426 The VPN Forwarder MAY support the ability to associate multiple 427 virtual interfaces with the same VRF. When that is the case, locally 428 originated routes, that is IP routes to the local virtual interfaces 429 SHALL NOT be used to forward outbound traffic (from the virtual 430 interfaces to the outside) unless a route advertisement has been 431 received that matches that specific IP prefix and next-hop 432 information. 434 As an example, if a given VRF contains two virtual interfaces, 435 "veth0" and "veth1", with the addresses 10.0.1.1/32 and 10.0.1.2/32 436 respectively, the initial forwarding state must be initialized such 437 that traffic from either of these interfaces does not match the 438 other's routing table entry. It may for instance match a default 439 route advertised by a remote system. Traffic received from other VPN 440 Forwarders, however, must be delivered to the correct local 441 interface. If at a subsequent stage a route is received from the 442 Route Server such that 10.0.1.2/32 has a next-hop with the IP address 443 of the local host and the correct label, the system may subsequently 444 install a local routing table entry that delivers traffic directly to 445 the "veth1" interface. 447 The 20-bit label which is associated with a virtual-interface is of 448 local significance only and SHOULD be allocated by the VPN Forwarder. 450 When an external VPN Forwarder is used the end-system MUST associate 451 each virtual interface with a VLAN [IEEE.802-1Q] that is unique on 452 the end-system. The switching infrastructure MUST be configured such 453 that multi-destination frames sourced from an end-system are only 454 delivered to VPN Forwarders used by this end-system and not to other 455 end-systems. 457 6. XMPP signaling protocol 459 End-System Route Servers must be aware of VPN membership on each 460 Forwarder as well as what IP addresses are currently associated with 461 each virtual interface. 463 VPN Forwarders must receive VPN route information from which to 464 populate their forwarding tables. External VPN Forwarders also need 465 to receive the virtual interface and IP address events from the end- 466 system for which they are VPN forwarders. In this case the end- 467 system assigns an 802.1Q VLAN tag to each virtual interface and 468 communicates that information to the Forwarder. 470 In order to exchange this information this specification uses the 471 XMPP [RFC6120] protocol along with the PubSub Collection Nodes 472 [pubsub] extension. 474 When an external VPN Forwarder is used, end-systems establish XMPP 475 sessions with VPN Forwarders. VPN forwarders (both co-located and 476 external) establish XMPP sessions with End-System Route Servers. VPN 477 Forwarders act as an XMPP clients of a End-System Route Server. 478 External VPN Forwarders act as XMPP servers for end-systems which are 479 associated with them. These sessions are persistent and MUST use the 480 XMPP Ping [xmpp-ping] extension in order to detect end-system 481 failures. 483 A VPN Forwarder MAY connect to multiple End-System Route Servers for 484 reliability. In this case it SHOULD publish its information to each 485 of the Route Servers. It MAY choose to subscribe to VPN routing 486 information once only from one of the available gateways. 488 The information advertised by an XMPP client SHOULD be deleted after 489 a configurable timeout, when the session closes. This timeout should 490 default to 60 seconds. 492 +---------+ +--------+ 493 | RS | ----------- | BGP | 494 +---------+ +--------+ 495 // \ / 496 XMPP \ / 497 // \ / 498 +------------+ \ / 499 | end-system | \ / 500 +------------+ \/ 501 \\ /\ 502 XMPP / \ 503 \\ / \ 504 +---------+ / \ +--------+ 505 | RS | ----------- | BGP | 506 +---------+ +--------+ 508 The figure above represents a typical configuration in which an end- 509 system with a co-located VPN Forwarder is directly connected to two 510 End-System Routes Servers, which are in turn connected to multiple 511 BGP speakers which may be other L3VPN PEs or BGP route reflectors. 513 In deployment the number of End-System Route Servers used will depend 514 on the desired Route Server to VPN Forwarder ratio which affects the 515 convergence time of event propagation. 517 The XMPP "jid" used by the client shall be a 6-byte value that 518 uniquely identifies it in its administrative domain. This 519 specification recommends the use of the MAC address of one of the 520 physical ethernet interfaces. 522 Each VPN shall be identified by a 128 octet ASCII character string. 524 When external Forwarders are used, its control software operates as a 525 XMPP server processing requests from end-systems and as a client of 526 one or more End-System Route Servers. The control software relays to 527 the End-System Route Server(s) VPN membership messages it receives 528 from the end-system. VPN routing information received from the Route 529 Server(s) SHOULD NOT be propagated to the end-system. 531 When a virtual interface is created on a end-system, the host 532 operating-system software shall generate an XMPP Subscribe message to 533 its server (the End-System Route Server or external VPN Forwarder). 535 Subscription request from co-located VPN Forwarder to Route Server: 537 541 542 543 544 546 The request above, instructs the End-System Route Server to start 547 populating the client's VRF table with any routing information that 548 is available for this VPN. The XMPP node 'vpn-customer-name' is 549 assumed to be a collection which is implicitly created by the End- 550 System Route Server. Creation of a virtual interface may precede any 551 IP address becoming active on the interface, as it is the case with 552 VM instantiation. 554 Subscription request from end-system to external VPN Forwarder: 556 560 561 562 563 564 vlan-id 565 566 567 568 570 When an external VPN Forwarder is used the end-system should include 571 the VLAN identifier it assigned to the virtual interface as a 572 subscription option. 574 When a IP address is added to a virtual interface, the end-system 575 will generate an XMPP Publish request. 577 Publish request from VPN Forwarder to End-System Route Server: 579 581 to='network-control.domain.org' 582 id='request1'> 583 584 585 586 587 'vpn-ip-address/32' 588 'infrastructure-ip-address' 589 590 591 592 593 594 595 597 601 602 603 604 605 606 607 The End-System Route Server will convert the information received in 608 a the 'publish' request into the corresponding BGP route information 609 such that:. 611 It associates the specific request with a local VRF which it 612 resolves by using a combination of the originator system-id and 613 the collection 'node' attribute. 615 It creates a BGP VPN route with a 'Route Distinguisher' (RD) which 616 contains the the end-system's 'system-id' value and the specified 617 IP prefix and 'label' received from the VPN Forwarder as the 618 Network Layer Reachability Information (NLRI). 620 The BGP next-hop address is set to the address of the VPN 621 Forwarder. 623 It optionally associates the route with an extended community TDB 624 containing a version number of the virtual-interface. 626 Update notification from Route Server to VPN Forwarder: 628 629 630 631 632 633 'vpn-ip-address>/32' 634 'infrastructure-ip-address' 635 636 637 638 639 640 ... 641 642 643 644 646 Notifications should be generated whenever a VPN route is added, 647 modified or deleted. 649 Note that the Update from the Route Server to the VPN Forwarder does 650 not contain the system-id of the destination end-system. The "from" 651 attribute in the 'message' element contains a "jid" associated with 652 the Route Servers in the domain. The XMPP messages are point-to- 653 point in nature, between a Forwarder and Route Server. Even in the 654 case when one XMPP publish request from a Forwarder may cause the 655 Route Server to generate one or more event notifications. 657 When multiple possible routes exist for a given VPN IP address within 658 a VRF it is the responsibility of the Route Server to select the best 659 path to advertise to the Forwarder. 661 When routes are withdrawn, the End-System Route Server generates both 662 a "collection disassociate" request as well as a node "delete" 663 request. 665 In situations where an automated system is controlling the 666 instantiation of virtual interfaces it may be possible to have that 667 system assign a non-decreasing version number for each instantiation 668 of that particular interface. In that case, this number, carried in 669 the 'version' field may be used to help gateways select the most 670 recent instantiation of an interface during the interval of time 671 where multiple routes are present. 673 7. End-System Route Server behavior 675 End-System Route Servers SHALL support the BGP address families: VPN- 676 IPv4 (1, 128), VPN-IPv6 (2, 128) and RT-Constraint (1, 132) 677 [RFC4684]. 679 When an End-System Route Server receives a request to create or 680 modify a VPN route it SHALL generate a BGP VPN route advertisement 681 with the corresponding information. 683 It is assumed that the End-System Route Servers have information 684 regarding the mapping between end-system tuple ('system-id', 'vpn- 685 customer-names') and BGP Route Targets used to import and export 686 information from the associated VRFs. This mapping is known via an 687 out-of-band mechanism not specified in this document. 689 Whenever the End-System Route Server receives an XMPP subscription 690 request, it SHALL consult its RT-Constraint Routing Information Base 691 (RIB). If the Route Server does not already have locally originated 692 route for the route target the corresponds to the vpn-name present in 693 the request, it SHALL create one and generate the corresponding BGP 694 route advertisement. This route advertisement should only be 695 withdrawn when there are no more downstream XMPP clients subscribed 696 to the VPN. 698 The 32bit route version number defined in the XML schema is 699 advertised into BGP as an Extended community with type TBD. 701 End-System Route Servers SHOULD automatically assign a BGP route 702 distinguisher per VPN routing table. 704 8. Operational Model 706 In the simplest case, a VPN is a collection of systems that are 707 allowed to exchange traffic with each other and only with each other. 708 Since all the forwarding tables in this VPN have the same routing 709 entries they are often referred to as symmetrical VPNs. 711 In order to better illustrate the operation of the protocol we 712 consider a simple example in which "host 1" and "host 2" both contain 713 a virtual interface that is a member of the same VPN. 715 Each of these hosts has an XMPP session with an End-System Route 716 Server, RS1 and RS2 our example, and these Route Servers are part of 717 the same BGP mesh. 719 When a virtual interface is created on "host 1", the local XMPP 720 client generates a XMPP subscription message to its respective Route 721 Server. This message contains a VPN identifier that has been 722 assigned by the provisioning system. The Route Server maps that 723 identifier to a BGP IP VPN configuration which contains the list of 724 import and export route targets to be used for that particular VRF. 726 Once the interface is operational, "host 1" will publish any IP 727 addresses that are configured on the respective virtual interface. 728 This will in turn cause the End-System Route Server to advertise 729 these (directly or indirectly) to any other BGP speaker on the 730 network which is connected to an attachment point of that VPN. 732 +--------+ +------------+ +----------+ 733 | host 1 | <===> | End-System | <===> | BGP mesh | 734 +--------+ |Route Server| +----------+ 735 +------------+ 737 +----------------+-------------+-------+-----------+ 738 | VPN IP address | NEXT-HOP | label | Known via | 739 +----------------+-------------+-------+-----------+ 740 | 10.1.1.1/32 | 192.168.1.1 | 10000 | XMPP | 741 | 10.1.1.2/32 | 192.168.2.1 | 20000 | BGP | 742 +----------------+-------------+-------+-----------+ 744 VPN Routing table on Route Server 746 The figure above represents the contents of the VRF routing table on 747 RS1 after the IPv4 address 10.1.1.1 has been added to the virtual 748 interface on host 1. It assumes that there is another attachement 749 point for this VPN with the IPv4 address of 10.1.1.2. Host 1 has an 750 infrastructure IP address of 192.168.1.1 configured on its physical 751 interface while host 2 has IP address 192.168.2.1. 753 The contents of the VRF routing table in the End-System Route Servers 754 are advertised via XMPP Update notifications sent to host 1. This 755 information is the used by the host to populate the forwarding table 756 associated with that VPN. 758 +--------+ +--------+ 759 app -- veth0 --| host 1 |=== [network] ===| host 2 |-- veth0 -- app 760 +--------+ +--------+ 762 IP pkt ===> GRE encap ===> [IP net] ===> GRE decap ===> IP pkt 763 [192.168.2.1, 20] map 20 to veth0 765 +----------------+--------------+-------+ 766 | VPN IP address | Host address | label | 767 +----------------+--------------+-------+ 768 | 10.1.1.1/32 | localhost | 10000 | 769 | 10.1.1.2/32 | 192.168.2.1 | 20000 | 770 +----------------+--------------+-------+ 772 VRF table on host1 774 When an application that uses the virtual interface on host 1 775 generates packets with a destination IP address of 10.1.1.2 these are 776 routed by the VPN Forwarder implemented in the Host OS. The packets 777 are encapsulated with a GRE header that contains a 20-bit label 778 assigned by host 2. 780 In the case the virtual interface on host is associated with a guest 781 OS, this guest OS has had its address resolution queries answered 782 with the Virtual Router MAC address. As a result, that is the 783 address it uses as the destination MAC address in packets it 784 originates. This MAC address is not present on the GRE encapsulated 785 packet. 787 End-System Route Servers are software applications the implement both 788 the BGP IP VPN PE control plane as well as XMPP server functionality. 789 These application are not in the forwarding plane and do not need to 790 be co-located with a network device. 792 Network devices MAY have direct BGP sessions to the End-System Route 793 Servers. For instance, a router or security appliance that supports 794 BGP/MPLS IP VPNs over GRE may use its existing functionality to 795 inter-operate directly with a collection of Virtual Machines or other 796 network appliances that support this specification. 798 End-System Route Servers implement the VRF import policy and export 799 policy functionality that is associated with PE routers in standard 800 BGP IP/VPN deployments. VPN Forwarders receive forwarding 801 information after policy and route selection is applied. These are 802 unqualified routes in a specific VRF rather than VPN routing 803 information qualified by a Route Distinguisher and with a set of 804 Route Targets. 806 A symmetrical VPN uses a vrf import and vrf export polices that 807 contain a single route target, where the route target used for both 808 import and export is the same. 810 Different VPN topologies can be created by manipulating the vrf 811 import and export configuration including "hub-and-spoke" topologies 812 or overlapping VPNs. 814 An example of a hub-and-spoke VPN configuration is one where all the 815 traffic from the VPN clients must be redirected though a middle-box 816 for inspection. Assuming that the virtual interfaces of a particular 817 user are configured to be in the VPN "tenant1". At an initial stage 818 this "tenant1" VPN is symmetrical and uses a single Route Target in 819 both its import and export policies. The middle-box functionality 820 can be incrementally deployed by defining a new VPN, "tenant1-hub", 821 and an associated Route Target. Accompanied with a change in the 822 End-System Route Server configuration such that VPN "tenant1" only 823 imports routes with the Route Target associated with the hub. The 824 "hub" VPN is assumed to advertise a prefix that covers all the VPN 825 clients IP addresses. The "hub" VPN imports the VPN routes in order 826 for it to be able to generate the XMPP updates to the "hub" end- 827 system. This information is required for the return traffic from the 828 hub to the spokes (the VPN clients). In such a scenario a single 829 physical interface can connect the middle-box to the clients in a 830 given VPN which appear logically as downstream from it. Such a 831 middle-box would often require connectivity to multiple VPNs, such as 832 for instance an "outside" VPN which provides external connectivity to 833 one or more "inside" VPNs. 835 The functionality defined in this document in which the BGP IP VPN PE 836 functionality is split into its control (End-System Route Servers) 837 and forwarding (VPN Forwarder) components is fully interoperable with 838 existing BGP IP VPN PEs. 840 This makes it possible to reuse existing systems. For example, at 841 the edge of a data-center facility it may be desirable to use an 842 existing router or appliance that aggregates IP VPN routing 843 information and/or provides IP based services such as stateful packet 844 inspection. 846 Such a system can be configured, based on existing functionality, to 847 suppress more specific routes than a specified aggregate while 848 advertising the aggregate with a BGP NEXT_HOP containing the PE's IP 849 address and a locally assigned label corresponding to a VRF where the 850 more specific routes are present. 852 9. Security Considerations 854 The signaling protocol defines the access control policies for each 855 virtual interface and any guest application associated with it. It 856 is important to secure the end-system access to End-System Route 857 Servers and the BGP infrastructure itself. 859 The XMPP session between end-systems and the Route Servers MUST use 860 mutual authentication. One possible strategy is to distribute pre- 861 signed certificates to end-systems which are presented as proof of 862 authorization to the Route Server. 864 BGP sessions MUST be authenticated. This document recommends that 865 BGP speaking systems filter traffic on port 179 such that only IP 866 addresses which are known to participate in the BGP signaling 867 protocol are allowed. 869 10. Acknowledgements 871 Yakov Rekhter has contributed to this document by providing detailed 872 feedback and suggestions. The authors would also like to thank 873 Thomas Morin for his comments. 875 11. References 877 11.1. Normative References 879 [RFC4023] Worster, T., Rekhter, Y. and E. Rosen, "Encapsulating MPLS 880 in IP or Generic Routing Encapsulation (GRE)", RFC 4023, 881 March 2005. 883 [RFC4271] Rekhter, Y., Li, T. and S. Hares, "A Border Gateway 884 Protocol 4 (BGP-4)", RFC 4271, January 2006. 886 [RFC4364] Rosen, E. and Y. Rekhter, "BGP/MPLS IP Virtual Private 887 Networks (VPNs)", RFC 4364, February 2006. 889 [RFC4456] Bates, T., Chen, E. and R. Chandra, "BGP Route Reflection: 890 An Alternative to Full Mesh Internal BGP (IBGP)", RFC 891 4456, April 2006. 893 [RFC4684] Marques, P., Bonica, R., Fang, L., Martini, L., Raszuk, 894 R., Patel, K. and J. Guichard, "Constrained Route 895 Distribution for Border Gateway Protocol/MultiProtocol 896 Label Switching (BGP/MPLS) Internet Protocol (IP) Virtual 897 Private Networks (VPNs)", RFC 4684, November 2006. 899 [RFC5575] Marques, P., Sheth, N., Raszuk, R., Greene, B., Mauch, J. 900 and D. McPherson, "Dissemination of Flow Specification 901 Rules", RFC 5575, August 2009. 903 [RFC5798] Nadas, S., "Virtual Router Redundancy Protocol (VRRP) 904 Version 3 for IPv4 and IPv6", RFC 5798, March 2010. 906 [RFC6120] Saint-Andre, P., "Extensible Messaging and Presence 907 Protocol (XMPP): Core", RFC 6120, March 2011. 909 [xmpp-ping] 910 "XMPP Ping", XEP 0199, June 2009. 912 [pubsub] "PubSub Collection Nodes", XEP 0248, September 2010. 914 11.2. Informational References 916 [I-D.marques-sdnp-flow-spec] 917 Marques, P., Fang, L., Pan, P., Shukla, A. and M. 918 Napierala, "Traffic classification in end-system IP 919 VPNs.", Internet-Draft draft-marques-sdnp-flow-spec-01, 920 April 2012. 922 [IEEE.802-1Q] 923 Institute of Electrical and Electronics Engineers, "Local 924 and Metropolitan Area Networks: Virtual Bridged Local Area 925 Networks", IEEE Std 802.1Q-2005, May 2006. 927 Authors' Addresses 929 Pedro Marques 930 Contrail Systems 931 2350 Mission College Blvd. 932 Santa Clara, CA 95054 934 Email: roque@contrailsystems.com 936 Luyuan Fang 937 Cisco Systems 938 111 Wood Avenue South 939 Iselin, NJ 08830 941 Email: lufang@cisco.com 943 Ping Pan 944 Infinera Corp 945 140 Caspian Ct. 946 Sunnyvale, CA 94089 948 Email: ppan@infinera.com 950 Amit Shukla 951 Juniper Networks 952 1194 N. Mathilda Av. 953 Sunnyvale, CA 94089 955 Email: amit@juniper.net 957 Maria Napierala 958 AT&T Labs 959 200 Laurel Avenue 960 Middletown, NJ 07748 962 Email: mnapierala@att.com 963 Nabil Bitar 964 Verizon 965 40 Sylvan Rd. 966 Waltham, MA 02145 968 Email: nabil.bitar@verizon.com