idnits 2.17.1 draft-mrw-nat66-08.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 abstract seems to contain references ([RFC2119]), which it shouldn't. Please replace those with straight textual mentions of the documents in question. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year == Line 1281 has weird spacing: '...d short inner...' == Line 1283 has weird spacing: '...d short outer...' == Line 1285 has weird spacing: '...d short inner...' == Line 1286 has weird spacing: '...d short packe...' == Line 1287 has weird spacing: '...ed char chec...' == (12 more instances...) -- The document date (February 28, 2011) is 4799 days in the past. Is this intentional? -- Found something which looks like a code comment -- if you have code sections in the document, please surround them with '' and '' lines. Checking references for intended status: Experimental ---------------------------------------------------------------------------- -- Looks like a reference, but probably isn't: '8' on line 1288 -- Looks like a reference, but probably isn't: '65536' on line 1287 -- Looks like a reference, but probably isn't: '3' on line 1515 -- Looks like a reference, but probably isn't: '1' on line 1515 -- Looks like a reference, but probably isn't: '0' on line 1515 -- Looks like a reference, but probably isn't: '2' on line 1515 -- Looks like a reference, but probably isn't: '4' on line 1516 -- Looks like a reference, but probably isn't: '5' on line 1516 -- Looks like a reference, but probably isn't: '6' on line 1516 -- Looks like a reference, but probably isn't: '7' on line 1516 -- Obsolete informational reference (is this intentional?): RFC 2629 (Obsoleted by RFC 7749) -- Obsolete informational reference (is this intentional?): RFC 3484 (Obsoleted by RFC 6724) Summary: 1 error (**), 0 flaws (~~), 7 warnings (==), 14 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group M. Wasserman 3 Internet-Draft Painless Security 4 Intended status: Experimental F. Baker 5 Expires: September 1, 2011 Cisco Systems 6 February 28, 2011 8 IPv6-to-IPv6 Network Prefix Translation 9 draft-mrw-nat66-08 11 Abstract 13 This document describes a stateless, transport-agnostic IPv6-to-IPv6 14 Network Prefix Translation (NPTv6) function that provides the address 15 independence benefit associated with IPv4-to-IPv4 NAT (NAPT44), and 16 in addition provides a 1:1 relationship between addresses in the 17 "inside" and "outside" prefixes, preserving end to end reachability 18 at the network layer. 20 Requirements Terminology 22 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 23 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 24 document are to be interpreted as described in RFC 2119 [RFC2119]. 26 Status of this Memo 28 This Internet-Draft is submitted in full conformance with the 29 provisions of BCP 78 and BCP 79. 31 Internet-Drafts are working documents of the Internet Engineering 32 Task Force (IETF). Note that other groups may also distribute 33 working documents as Internet-Drafts. The list of current Internet- 34 Drafts is at http://datatracker.ietf.org/drafts/current/. 36 Internet-Drafts are draft documents valid for a maximum of six months 37 and may be updated, replaced, or obsoleted by other documents at any 38 time. It is inappropriate to use Internet-Drafts as reference 39 material or to cite them other than as "work in progress." 41 This Internet-Draft will expire on September 1, 2011. 43 Copyright Notice 45 Copyright (c) 2011 IETF Trust and the persons identified as the 46 document authors. All rights reserved. 48 This document is subject to BCP 78 and the IETF Trust's Legal 49 Provisions Relating to IETF Documents 50 (http://trustee.ietf.org/license-info) in effect on the date of 51 publication of this document. Please review these documents 52 carefully, as they describe your rights and restrictions with respect 53 to this document. Code Components extracted from this document must 54 include Simplified BSD License text as described in Section 4.e of 55 the Trust Legal Provisions and are provided without warranty as 56 described in the Simplified BSD License. 58 Table of Contents 60 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4 61 1.1. What is Address Independence? . . . . . . . . . . . . . . 4 62 1.2. NPTv6 Applicability . . . . . . . . . . . . . . . . . . . 6 63 2. NPTv6 Overview . . . . . . . . . . . . . . . . . . . . . . . . 7 64 2.1. NPTv6: the simplest case . . . . . . . . . . . . . . . . 7 65 2.2. NPTv6 between peer networks . . . . . . . . . . . . . . . 8 66 2.3. NPTv6 redundnacy and load-sharing . . . . . . . . . . . . 9 67 2.4. NPTv6 multihoming . . . . . . . . . . . . . . . . . . . . 10 68 2.5. Mapping with No Per-Flow State . . . . . . . . . . . . . 10 69 2.6. Checksum-Neutral Mapping . . . . . . . . . . . . . . . . 11 70 3. NPTv6 Algorithmic Specification . . . . . . . . . . . . . . . 11 71 3.1. NPTv6 configuration calculations . . . . . . . . . . . . 11 72 3.2. NPTv6 translation, internal network to external 73 network . . . . . . . . . . . . . . . . . . . . . . . . . 12 74 3.3. NPTv6 translation, external network to internal 75 network . . . . . . . . . . . . . . . . . . . . . . . . . 12 76 3.4. NPTv6 with a /48 or shorter prefix . . . . . . . . . . . 13 77 3.5. NPTv6 with a /49 or longer prefix . . . . . . . . . . . . 13 78 3.6. /48 Prefix Mapping Example . . . . . . . . . . . . . . . 13 79 3.7. Address Mapping for Longer Prefixes . . . . . . . . . . . 14 80 4. Implications of Network Address Translator Behavioral 81 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . 15 82 4.1. Prefix configuration and generation . . . . . . . . . . . 15 83 4.2. Subnet numbering . . . . . . . . . . . . . . . . . . . . 15 84 4.3. NAT Behavioral Requirements . . . . . . . . . . . . . . . 15 85 5. Implications for Applications . . . . . . . . . . . . . . . . 16 86 5.1. Recommendation for network planners considering use 87 of NPTv6 Translator . . . . . . . . . . . . . . . . . . . 18 88 5.2. Recommendations for application writers . . . . . . . . . 18 89 5.3. Recommendation for future work . . . . . . . . . . . . . 18 90 6. A Note on Port Mapping . . . . . . . . . . . . . . . . . . . . 18 91 7. Security Considerations . . . . . . . . . . . . . . . . . . . 19 92 8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 20 93 9. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 20 94 10. Change Log . . . . . . . . . . . . . . . . . . . . . . . . . . 20 95 10.1. Changes Between draft-mrw-behave-nat66-00 and -01 . . . . 20 96 10.2. Changes between *behave-nat66-01 and -02 . . . . . . . . 20 97 10.3. Changes between *nat66-00 and *nat66-01 . . . . . . . . . 21 98 10.4. Changes between *nat66-01 and *nat66-02 . . . . . . . . . 21 99 10.5. Changes between *nat66-02 and *nat66-03 . . . . . . . . . 22 100 10.6. Changes between *nat66-03 and *nat66-04 . . . . . . . . . 22 101 10.7. Changes between *nat66-04 and *nat66-05 . . . . . . . . . 22 102 10.8. Changes between *nat66-05 and *nat66-06 . . . . . . . . . 22 103 10.9. Changes between *nat66-06 and *nat66-07 . . . . . . . . . 22 104 10.10. Changes between *nat66-07 and *nat66-08 . . . . . . . . . 22 105 11. References . . . . . . . . . . . . . . . . . . . . . . . . . . 23 106 11.1. Normative References . . . . . . . . . . . . . . . . . . 23 107 11.2. Informative References . . . . . . . . . . . . . . . . . 23 108 Appendix A. Why GSE? . . . . . . . . . . . . . . . . . . . . . . 25 109 Appendix B. Verification code . . . . . . . . . . . . . . . . . . 27 110 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 34 112 1. Introduction 114 This document describes a stateless IPv6-to-IPv6 Network Prefix 115 Translation (NPTv6) function, designed to provide address 116 independence to the edge network. It is transport-agnostic with 117 respect to transports that don't checksum the IP header, such as SCTP 118 or DCCP, and to transports that use the TCP/UDP pseudo-header and 119 checksum [RFC1071]. 121 This has several ramifications: 123 o Any security benefit that NAPT44 might offer is not present in 124 NPTv6, necessitating the use of a firewall to obtain those 125 benefits if desired. An example of such a firewall is described 126 in [RFC6092]. 128 o End to end reachability is preserved, although the address used 129 "inside" the edge network differs from the address used "outside" 130 the edge network. This has implications for application referrals 131 and other uses of Internet layer addresses. 133 o If there are multiple identically-configured prefix translators 134 between two networks, there is no need for them to exchange 135 dynamic state, as there is no dynamic state - the algorithmic 136 translation will be identical across each of them. The network 137 can therefore asymmetrically route, load-share, and fail-over 138 among them without issue. 140 o Since translation is 1:1 at the network layer, there is no need to 141 modify port numbers or other transport parameters. 143 1.1. What is Address Independence? 145 For the purposes of this document, IPv6 Address Independence consists 146 of the following set of properties: 148 From the perspective of the edge network: 150 * The IPv6 addresses used inside the local network (for 151 interfaces, access lists, and logs) do not need to be 152 renumbered if the global prefix(es) assigned for use by the 153 edge network are changed. 155 * The IPv6 addresses used inside the edge network (for 156 interfaces, access lists, and logs) or within other upstream 157 networks (such as when multihoming) do not need to be 158 renumbered when a site adds, drops, or changes upstream 159 networks. 161 * It is not necessary for an administration to convince an 162 upstream network to route its internal IPv6 prefixes, or for it 163 to advertise prefixes derived from other upstream networks into 164 it. 166 * Unless it wants to optimize routing between multiple upstream 167 networks in the process of multihoming, there is therefore no 168 need for a BGP exchange with the upstream network. 170 From the perspective of the upstream network: 172 * IPv6 addresses used by the edge network are guaranteed to have 173 a provider-allocated prefix, eliminating the need and concern 174 for BCP 38 [RFC2827] ingress filtering and the advertisement of 175 customer-specific prefixes. 177 Thus, address independence has ramifications for the edge network, 178 networks it directly connects with (especially its upstream 179 networks), and for the Internet as a whole. The desire for address 180 independence has been a primary driver for IPv4 NAT deployment in 181 medium to large-sized enterprise networks, including NAT deployments 182 in enterprises that have plenty of IPv4 provider-independent address 183 space (from IPv4 "swamp space"). It has also been a driver for edge 184 networks to become members of RIR communities, seeking to obtain BGP 185 Autonomous System Numbers and provider-independent prefixes, and as a 186 result has been one of the drivers of the explosion of the IPv4 route 187 table. Service providers have stated that the lack of address 188 independence from their customers has been a negative incentive to 189 deployment, due to the impact of customer routing expected in their 190 networks. 192 The Local Network Protection [RFC4864] document discusses a related 193 concept called "Address Autonomy" as a benefit of NAPT44. [RFC4864] 194 indicates that address autonomy can be achieved by the simultaneous 195 use of global addresses on all nodes within a site that need external 196 connectivity, and Unique Local Addresses (ULAs) [RFC4193] for all 197 internal communication. However, this solution fails to meet the 198 requirement for address independence, because if an ISP renumbering 199 event occurs, all of the hosts, routers, DHCP servers, ACLs, 200 firewalls and other internal systems that are configured with global 201 addresses from the ISP will need to be renumbered before global 202 connectivity is fully restored. 204 The use of IPv6 Provider Independent (PI) addresses has also been 205 suggested as a means to fulfill the address independence requirement. 206 However, this solution requires that an enterprise qualify to receive 207 a PI assignment and persuade their ISP to install specific routes for 208 the enterprise's PI addresses. There are a number of practical 209 issues with this approach, especially if there is a desire to route 210 to a number of geographically and topologically diverse set of sites, 211 which can sometimes involve coordinating with several ISPs to route 212 portions of a single PI prefix. These problems have caused numerous 213 enterprises with plenty of IPv4 swamp space to choose to use IPv4 NAT 214 for part, or substantially all, of their internal network instead of 215 using their provider-independent address space. 217 1.2. NPTv6 Applicability 219 NPTv6 provides a simple and compelling solution to meet the Address 220 Independence requirement in IPv6. The address independence benefit 221 stems directly from the translation function of the network prefix 222 translator. To avoid as many of the issues associated with NAPT44 as 223 possible, NPTv6 is defined to include a two-way, checksum-neutral, 224 algorithmic translation function, and nothing else. 226 NPTv6 does not include a port mapping function, and the defined 227 address mapping mechanism is checksum-neutral. This avoids the need 228 for a NPTv6 Translator to re-write transport layer headers, making it 229 feasible to deploy new or improved transport layer protocols without 230 upgrading NPTv6 Translators. Because NPTv6 does not involve re- 231 writing transport-layer headers, NPTv6 will not interfere with 232 encryption of the full IP payload in many cases. 234 The default NPTv6 address mapping mechanism is purely algorithmic, so 235 NPTv6 translators do not need to maintain per-node or per-connection 236 state, allowing deployment of more robust and adaptive networks than 237 can be deployed using NAPT44. Since the default NPTv6 mapping can be 238 performed in either direction, it does not interfere with inbound 239 connection establishment, thus allowing internal nodes to participate 240 in direct Peer-to-Peer applications without the application layer 241 overhead one finds in many IPv4 Peer-to-Peer applications. 243 Although NPTv6 compares favorably to NAPT44 in several ways, it does 244 not eliminate all of the architectural problems associated with IPv4 245 NAT, as described in [RFC2993]. NPTv6 involves modifying IP headers 246 in transit, so it is not compatible with security mechanisms, such as 247 the IPsec Authentication Header, that provide integrity protection 248 for the IP header. NPTv6 may interfere with the use of application 249 protocols that transmit IP addresses in the application-specific 250 portion of the IP packet. These applications currently require 251 application layer gateways (ALGs) to work correctly through NAPT44 252 devices, and similar ALGs may be required for these applications to 253 work through NPTv6 Translators. The use of separate internal and 254 external prefixes creates complexity for DNS deployment, due the 255 desire for internal nodes to communicate with other internal nodes 256 using internal addresses, while external nodes need to obtain 257 external addresses to communicate with the same nodes. This 258 frequently results in the deployment of "split DNS", which may add 259 complexity to network configuration. 261 The choice of address within the edge network bears consideration. 262 One could use a ULA, which maximizes address independence. That 263 could also be considered a misuse of the ULA; if the expectation is 264 that a ULA prevents access to a system from outside the range of the 265 ULA, NPTv6 overrides that. On the other hand, the administration is 266 aware that it has made that choice, and could if it desired deploy a 267 second ULA for the purpose of privacy; the only prefix that will be 268 translated is one that has a NPTv6 Translator configured to translate 269 to or from it. Also, using any other global scope address format 270 makes one either obtain a PI prefix or be at the mercy of the agency 271 from which it was allocated. 273 There are significant technical impacts associated with the 274 deployment of any prefix translation mechanism, including NPTv6, and 275 we strongly encourage anyone who is considering the implementation or 276 deployment of NPTv6 to read [RFC4864], and to carefully consider the 277 alternatives described in that document, some of which may cause 278 fewer problems than NPTv6. 280 2. NPTv6 Overview 282 NPTv6 may be implemented in an IPv6 router to map one IPv6 address 283 prefix to another IPv6 prefix as each IPv6 packet transits the 284 router. A router that implements a NPTv6 prefix translation function 285 is referred to as an NPTv6 Translator. 287 2.1. NPTv6: the simplest case 289 In its simplest form, a NPTv6 Translator interconnects two network 290 links, one of which is an "internal" network link attached to a leaf 291 network within a single administrative domain, and the other of which 292 is an "external" network with connectivity to the global Internet. 293 All of the hosts on the internal network will use addresses from a 294 single, locally-routed prefix, and those addresses will be translated 295 to/from addresses in a globally-routable prefix as IP packets transit 296 the NPTv6 Translator. The lengths of these two prefixes will be 297 functionally the same; if they differ, the longer of the two will 298 limit the ability to use subnets in the shorter. 300 External Network: Prefix = 2001:0DB8:0001:/48 301 -------------------------------------- 302 | 303 | 304 +-------------+ 305 | NPTv6 | 306 | Translator | 307 +-------------+ 308 | 309 | 310 -------------------------------------- 311 Internal Network: Prefix = FD01:0203:0405:/48 313 Figure 1: A simple translator 315 Figure 1 shows a NPTv6 Translator attached to two networks. In this 316 example, the internal network uses IPv6 Unique Local Addresses (ULAs) 317 [RFC4193] to represent the internal IPv6 nodes, and the external 318 network uses globally routable IPv6 addresses to represent the same 319 nodes. 321 When a NPTv6 Translator forwards packets in the "outbound" direction, 322 from the internal network to the external network, NPTv6 overwrites 323 the IPv6 source prefix (in the IPv6 header) with a corresponding 324 external prefix. When packets are forwarded in the "inbound" 325 direction, from the external network to the internal network, the 326 IPv6 destination prefix is overwritten with a corresponding internal 327 prefix. Using the prefixes shown in the diagram above, as an IP 328 packet passes through the NPTv6 Translator in the outbound direction, 329 the source prefix (FD01:0203:0405:/48) will be overwritten with the 330 external prefix (2001:0DB8:0001:/48). In an inbound packet, the 331 destination prefix (2001:0DB8:0001:/48) will be overwritten with the 332 internal prefix (FD01:0203:0405:/48). In both cases, it is the local 333 IPv6 prefix that is overwritten; the remote IPv6 prefix remains 334 unchanged. Nodes on the internal network are said to be "behind" the 335 NPTv6 Translator. 337 2.2. NPTv6 between peer networks 339 NPTv6 can also be used between two private networks. In these cases, 340 both networks may use ULA prefixes, with each subnet in one network 341 mapped into a corresponding subnet in the other network, and vice 342 versa. Or, each network may use ULA prefixes for internal 343 addressing, and global unicast addresses on the other network. 345 Internal Prefix = FD01:4444:5555:/48 346 -------------------------------------- 347 V | External Prefix 348 V | 2001:0DB8:6666:/48 349 V +---------+ ^ 350 V | NPTv6 | ^ 351 V | Device | ^ 352 V +---------+ ^ 353 External Prefix | ^ 354 2001:0DB8:0001:/48 | ^ 355 -------------------------------------- 356 Internal Prefix = FD01:0203:0405:/48 358 Figure 2: Flow of Information in Translation 360 2.3. NPTv6 redundnacy and load-sharing 362 In some cases, more than one NPTv6 Translator may be attached to a 363 network, as show in Figure 3. In such cases, NPTv6 Translators are 364 configured with the same internal and external prefixes. Since there 365 is only one translation, even though there are multiple translators, 366 they map only one external address (prefix and IID) to the internal 367 address. 369 External Network: Prefix = 2001:0DB8:0001:/48 370 -------------------------------------- 371 | | 372 | | 373 +-------------+ +-------------+ 374 | NPTv6 | | NPTv6 | 375 | Translator | | Translator | 376 | #1 | | #2 | 377 +-------------+ +-------------+ 378 | | 379 | | 380 -------------------------------------- 381 Internal Network: Prefix = FD01:0203:0405:/48 383 Figure 3: Parallel Translators 385 2.4. NPTv6 multihoming 387 External Network #1: External Network #2: 388 Prefix = 2001:0DB8:0001:/48 Prefix = 2001:0DB8:5555:/48 389 --------------------------- -------------------------- 390 | | 391 | | 392 +-------------+ +-------------+ 393 | NPTv6 | | NPTv6 | 394 | Translator | | Translator | 395 | #1 | | #2 | 396 +-------------+ +-------------+ 397 | | 398 | | 399 -------------------------------------- 400 Internal Network: Prefix = FD01:0203:0405:/48 402 Figure 4: Parallel Translators with different upstream networks 404 When multihoming, NPTv6 Translators are attached to an internal 405 network, as show in Figure 4, but connected to different external 406 networks. In such cases, NPTv6 Translators are configured with the 407 same internal prefix, but different external prefixes. Since there 408 are multiple translations, they map multiple external addresses 409 (prefix and IID) to the common internal address. A system within the 410 edge network is unable to determine which external address it is 411 using apart from services such as STUN. 413 Multihoming in this sense has one negative feature as compared with 414 multihoming with a provider-independent address; when routes change 415 between NPTv6 Translators, since the upstream network changes, the 416 translated prefix can change. This would case sessions and referrals 417 dependent on it to fail as well. This is not expected to be a major 418 real issue, however, in networks where routing is generally stable. 420 2.5. Mapping with No Per-Flow State 422 When NPTv6 is used as described in this document, no per-node or per- 423 flow state is maintained in the NPTv6 Translator. Both inbound and 424 outbound packets are translated algorithmically, using only 425 information found in the IPv6 header. Due to this property, NPTv6's 426 two-way, algorithmic address mapping can support both outbound and 427 inbound connection establishment without the need for state-priming 428 or rendezvous mechanisms, or the maintenance of mapping state. This 429 is a significant improvement over NAPT44 devices, but it also has 430 significant security implications which are described in the Security 431 Considerations section. 433 2.6. Checksum-Neutral Mapping 435 When a change is made to one of the IP header fields in the IPv6 436 pseudo-header checksum (such as one of the IP addresses), the 437 checksum field in the transport layer header may become invalid. 438 Fortunately, an incremental change in the area covered by the 439 Internet standard checksum [RFC1071] will result in a well-defined 440 change to the checksum value [RFC1624]. So, a checksum change caused 441 by modifying part of the area covered by the checksum can be 442 corrected by making a complementary change to a different 16-bit 443 field covered by the same checksum. 445 The NPTv6 mapping mechanisms described in this document are checksum- 446 neutral, which means that they result in IP headers that will 447 generate the same IPv6 pseudo-header checksum when the checksum is 448 calculated using the standard Internet checksum algorithm [RFC1071]. 449 Any changes that are made during translation of the IPv6 prefix are 450 offset by changes to other parts of the IPv6 address. This results 451 in transport layers that use the Internet checksum (such as TCP and 452 UDP) calculating the same IPv6 pseudo header checksum for both the 453 internal and external forms of the same packet, which avoids the need 454 for the NPTv6 Translator to modify those transport layer headers to 455 correct the checksum value. 457 As noted in Section 4.2, this mapping results in an edge network 458 using a /48 external prefix to be unable to use subnet 0xFFFF. 460 3. NPTv6 Algorithmic Specification 462 The [RFC4291] IPv6 Address is reproduced for clarity in Figure 5. 464 0 15 16 31 32 47 48 63 64 79 80 95 96 111 112 127 465 +-------+-------+-------+-------+-------+-------+-------+-------+ 466 | Routing Prefix | Subnet| Interface Identifier (IID) | 467 +-------+-------+-------+-------+-------+-------+-------+-------+ 469 Figure 5: Enumeration of the IPv6 Address [RFC4291] 471 3.1. NPTv6 configuration calculations 473 When an NPTv6 Translation function is configured, it is configured 474 with 476 o one or more "internal" interfaces with their "internal" routing 477 domain prefixes, and 479 o one or more "external" interfaces with their "external" routing 480 domain prefixes. 482 In the simple case, there is one of each. If a single router 483 provides NPTv6 translation services between a multiplicity of domains 484 (as might be true when multihoming), each internal/external pair must 485 be thought of as a separate NPTv6 Translator from the perspective of 486 this specification. 488 When an NPTv6 Translator is configured, the translation function 489 first ensures that the internal and external prefixes are the same 490 length, if necessary by extending the shorter of the two with zeroes. 491 These two prefixes will be used in the prefix translation function 492 described in Section 3.2 and Section 3.3. 494 They are then zero-extended to /64, for the purposes of a 495 calculation. The translation function calculates the ones-complement 496 sum of the 16 bit words of the /64 external prefix and the /64 497 internal prefix. It then calculates the difference between these 498 values: internal minus external. This value, called the 499 "adjustment", is effectively constant for the lifetime of the NPTv6 500 Translator configuration, and used in per-packet processing. 502 3.2. NPTv6 translation, internal network to external network 504 When a datagram passes through the NPTv6 Translator from an internal 505 to an external network, its IPv6 Source Address is changed in two 506 ways: 508 o If the internal subnet number has no mapping, such as being 0xFFFF 509 or simply not mapped, discard the datagram. This SHOULD result in 510 an ICMP Destination Unreachable. 512 o The internal prefix is overwritten with the external prefix, in 513 effect subtracting the difference between the two checksums (the 514 adjustment) from the pseudo-header's checksum, and 516 o A 16-bit word of the address has the adjustment added to it using 517 one's complement arithmetic. If the result is 0xFFFF, it is 518 overwritten as zero. The choice of word is as specified in 519 Section 3.4 or Section 3.5 as appropriate. 521 3.3. NPTv6 translation, external network to internal network 523 When a datagram passes through the NPTv6 Translator from an external 524 to an internal network, its IPv6 Destination Address is changed in 525 two ways: 527 o The external prefix is overwritten with the internal prefix, in 528 effect adding the difference between the two checksums (the 529 adjustment) to the pseudoheader's checksum, and 531 o A 16-bit word of the address has the adjustment subtracted from it 532 (bitwise inverted and added to it) it using one's complement 533 arithmetic. If the result is 0xFFFF, it is overwritten as zero. 534 The choice of word is as specified in Section 3.4 or Section 3.5 535 as appropriate. 537 3.4. NPTv6 with a /48 or shorter prefix 539 When a NPTv6 Translator is configured with internal and external 540 prefixes that are 48 bits in length (a /48) or shorter, the 541 adjustment MUST be added to or subtracted from bits 48..63 of the 542 address. 544 This mapping results in no modification of the Interface Identifier 545 (IID), which is held in the lower half of the IPv6 address, so it 546 will not interfere with future protocols that may use unique IIDs for 547 node identification. 549 NPTv6 Translator implementations MUST implement the /48 mapping. 551 3.5. NPTv6 with a /49 or longer prefix 553 When a NPTv6 Translator is configured with internal and external 554 prefixes that are longer than 48 bits in length (such as a /52, /56, 555 or /60), the adjustment must be added to or subtracted from one of 556 the words in bits 64..79, 80..95, 96..111, or 112..127 of the 557 address. While the choice of word is immaterial as long as it is 558 consistent, for consistency's sake, these words MUST be inspected in 559 that sequence, and the first that is not initially 0xFFFF chosen. 561 NPTv6 Translator implementations SHOULD implement the mapping for 562 longer prefixes. 564 3.6. /48 Prefix Mapping Example 566 For the network shown in Figure 1, the Internal Prefix is FD01:0203: 567 0405:/48, and the External Prefix is 2001:0DB8:0001:/48 569 If a node with internal address FD01:0203:0405:0001::1234 sends an 570 outbound packet through the NPTv6 Translator, the resulting external 571 address will be 2001:0DB8:0001:D550::1234. The resulting address is 572 obtained by calculating the checksum of both the internal and 573 external 48-bit prefixes, subtracting the internal prefix from the 574 external prefix using one's complement arithmetic to calculate the 575 "adjustment", and adding the adjustment to the 16-bit subnet field 576 (in this case 0x0001). 578 To show the work: 580 The one's complement checksum of FD01:0203:0405 is 0xFCF5. The one's 581 complement checksum of 2001:0DB8:0001 is 0xD245. Using one's 582 complement arithmetic, 0xD245 - 0xFCF5 = 0xD54F. The subnet in the 583 original packet is 0x0001. Using one's complement arithmetic, 0x0001 584 + 0xD54F = 0xD550. Since 0xD550 != 0xFFFF, it is not changed to 585 0x0000. 587 So, the value 0xD550 is written in the 16-bit subnet area, resulting 588 in a mapped external address of 2001:0DB8:0001:D550::1234. 590 When a response packet is received, it will contain the destination 591 address 2001:0DB8:0001:D550::0001, which will be mapped using the 592 inverse mapping algorithm, back to FD01:0203:0405:0001::1234. 594 In this case, the difference between the two prefixes will be 595 calculated as follows: 597 Using one's complement arithmetic, 0xFCF5 - 0xD245 = 0x2AB0. The 598 subnet in the original packet = 0xD550. Using one's complement 599 arithmetic, 0xD550 + 0x2AB0 = 0x0001. Since 0x0001 != 0xFFFF, it is 600 not changed to 0x0000. 602 So the value 0x0001 is written into the subnet field, and the 603 internal value of the subnet field is properly restored. 605 3.7. Address Mapping for Longer Prefixes 607 If the prefix being mapped is longer than 48 bits, the algorithm is 608 slightly more complex. A common case will be that the internal and 609 external prefixes are of different length. In such a case, the 610 shorter prefix is zero-extended to the length of the longer as 611 described in Section 3.1 for the purposes of overwriting the prefix. 612 Then, they are both zero-extended to 64 bits to facilitate one's 613 complement arithmetic. The "adjustment" is calculated using those 64 614 bit prefixes. 616 For example if the internal prefix is a /48 ULA and the external 617 prefix is a /56 provider-allocated prefix, the ULA becomes a /56 with 618 zeros in bits 48..55. For purposes of one's complement arithmetic, 619 they are then both zero-extended to 64 bits. A side-effect of this 620 is that a subset of the subnets possible in the shorter prefix are 621 untranslatable. While the security value of this is debatable, the 622 administration may choose to use them for subnets that it knows need 623 no external accessibility. 625 We then find the first word in the IID that does not have the value 626 0xFFFF, trying bits 64..79, and then 80..95, 96..111, and finally 627 112..127. We perform the same calculation (with the same proof of 628 correctness) as in Section 3.6, but applying it to that word. 630 Although any 16-bit portion of an IPv6 IID could contain 0xFFFF, an 631 IID of all-ones is a reserved anycast identifier that should not be 632 used on the network [RFC2526]. If a NPTv6 Translator discovers a 633 packet with an IID of all-zeros while performing address mapping, 634 that packet MUST be dropped, and an ICMPv6 Parameter Problem error 635 SHOULD be generated [RFC4443]. 637 Note: this mechanism does involve modification of the IID; it may not 638 be compatible with future mechanisms that use unique IIDs for node 639 identification. 641 4. Implications of Network Address Translator Behavioral Requirements 643 4.1. Prefix configuration and generation 645 NPTv6 Translators MUST support manual configuration of internal and 646 external prefixes, and MUST NOT place any restrictions on those 647 prefixes except that they be valid IPv6 unicast prefixes as described 648 in [RFC4291]. They MAY also support random generation of ULA 649 addresses on command. Since the most common place anticipated for 650 the implementation of an NPTv6 Translator is a CPE router, the reader 651 is urged to consider the requirements of 652 [I-D.ietf-v6ops-ipv6-cpe-router]. 654 4.2. Subnet numbering 656 For reasons detailed in Appendix B, a network using NPTv6 Translation 657 and a /48 external prefix MUST NOT use the value 0xFFFF to designate 658 a subnet that it expects to be translated. 660 4.3. NAT Behavioral Requirements 662 NPTv6 Translators MUST support hairpinning behavior, as defined in 663 the NAT Behavioral Requirements for UDP document [RFC4787]. This 664 means that when a NPTv6 Translator receives a packet on the internal 665 interface that has a destination address that matches the site's 666 external prefix, it will translate the packet and forward it 667 internally. This allows internal nodes to reach other internal nodes 668 using their external, global addresses when necessary. 670 Conceptually, the datagram leaves the domain (is translated as 671 described in Section 3.2), and returns (is again translated as 672 described in Section 3.3). As a result, the datagram exchange will 673 be through the NPTv6 Translator in both directions for the lifetime 674 of the session. The alternative would be to require the NPTv6 675 Translator to drop the datagram, forcing the sender to use the 676 correct internal prefix for its peer. Performing only the external- 677 to-internal translation results in the datagram being sent from the 678 untranslated internal address of the source to the translated and 679 therefore internal address of its peer, which would enable the 680 session to bypass the NPTv6 Translator for future datagrams. It 681 would also mean that the original sender would be unlikely to 682 recognize the response when it arrived. 684 Because NPTv6 does not perform port mapping and uses a one-to-one, 685 reversible mapping algorithm, none of the other NAT behavioral 686 requirements apply to NPTv6. 688 5. Implications for Applications 690 NPTv6 Translation does not create several of the problems known to 691 exist with other kinds of NATs and discussed in [RFC2993]. In 692 particular: NPTv6 Translation is stateless, so a "reset" or brief 693 outage of an NPTv6 Translator does not break connections that 694 traverse the translation function, and if multiple NPTv6 Translators 695 exist between the same two networks, load can shift or be dynamically 696 loaded-shared among them. Also, an NPTv6 Translator does not 697 aggregate traffic for several hosts/interfaces behind a lesser number 698 of external addresses, so there is no inherent expectation for an 699 NPTv6 Translator to block new inbound flows from external hosts, and 700 no issue with a filter or blacklist associated with one prefix within 701 the domain affecting another. A firewall can of course be used in 702 conjunction with NPTv6 Translator; this would allow the network 703 administrator more flexibility to specify security policy than would 704 be possible with a traditional NAT. 706 However, NPTv6 Translation does create difficulties for some kinds of 707 applications. e.g.: 709 o An application instance "behind" an NPTv6 Translator will see a 710 different address for its connections than its peers "outside" the 711 NPTv6 Translator. 713 o An application instance "outside" an NPTv6 Translator will see a 714 different address for its connections than any peers which are 715 "behind" an NPTv6 Translator. 717 o An application instance wishing to establish communication with a 718 peer which is "behind" an NPTv6 Translator, may need to use a 719 different address to reach that peer depending on whether the 720 instance is behind the same NPTv6 Translator or external to it. 721 If the NPTv6 Translator implements hairpinning (Section 4.3), it 722 suffices for applications to always use their external addresses. 723 However, this creates inefficiencies in the local network and may 724 also complicate implementation of the NPTv6 Translator. [RFC3484] 725 also would prefer the private address in such a case in order to 726 reduce those inefficiencies. 728 o An application instance which moves from a realm "behind" an NPTv6 729 Translator to a realm that is "outside" the network, or vice 730 versa, may find that it is no longer able to reach its peers at 731 the same addresses it was previously able to use. 733 o An application instance which is intermittently communicating with 734 a peer that moves from behind an NPTv6 Translator, to "outside" 735 the of it, or vice versa, may find that it is no longer able to 736 reach that peer at the same address that it had previously used. 738 Many, but not all, of the applications which are adversely affected 739 by NPTv6 Translation are those that do "referrals" - where an 740 application instance passes its own addresses, and/or addresses of 741 its peers, to other peers. (Some believe referrals are inherently 742 undesirable; others believe that they are necessary in some 743 circumstances. A discussion of the merits of referrals, or lack 744 thereof, is beyond the scope of this document.) 746 To some extent, the incidence of these difficulties can be reduced by 747 DNS hacks that attempt to expose addresses "behind" an NPTv6 748 Translator only to hosts which are also behind the same NPTv6 749 Translator; and perhaps also, to expose only the "internal" addresses 750 of hosts behind the NPTv6 Translator to other hosts behind the same 751 NPTv6 Translator. However, this cannot be a complete solution. A 752 full discussion of these issues is out of scope for this document, 753 but briefly: (a) reliance on DNS to solve this problem depends on 754 hosts always making queries from DNS servers in the same realm as 755 they are (or on DNS interception proxies, which create their own 756 problems), and on mobile hosts/applications not caching those 757 results; (b) reliance on DNS to solve this problem depends on network 758 administrators on all networks using such applications to reliably 759 and accurately maintain current DNS entries for every host using 760 those applications; and (c) reliance on DNS to solve this problem 761 depends on applications always using DNS names, even though they 762 often must run in environments where DNS names are not reliably 763 maintained for every host. Other issues are that there is often no 764 single distinguished name for a host, no reliable way for a host to 765 determine what DNS names are associated with it, and which names are 766 appropriate to use in which contexts. 768 5.1. Recommendation for network planners considering use of NPTv6 769 Translator 771 In light of the above, network planners considering the use of NPTv6 772 translation should carefully consider the kinds of applications that 773 they will need to run in the future, and determine whether the 774 address stability and provider independence benefits are consistent 775 with their application requirements. 777 5.2. Recommendations for application writers 779 Several mechanisms (e.g. STUN, TURN, ICE) have been used with 780 traditional IPv4 NAT to circumvent some of the limitations of such 781 devices. Similar mechanisms could also be applied to circumvent some 782 of the issues with NPTv6 Translator. However, all of these require 783 the assistance of an external server or a function co-located with 784 the translator that can tell an "internal" host what its "external" 785 addresses are. 787 5.3. Recommendation for future work 789 It might be desirable to define a general mechanism which would allow 790 hosts within a translation domain to determine their external 791 addresses and/or request that inbound traffic be permitted. If such 792 a mechanism were to be defined, it would ideally be general enough to 793 also accommodate other types of NAT likely to be encountered by IPV6 794 applications - in particular, IPv4/IPv6 Translation 795 [I-D.ietf-behave-v6v4-framework] [I-D.ietf-behave-dns64] 796 [I-D.ietf-behave-v6v4-xlate] [I-D.ietf-behave-v6v4-xlate-stateful] 797 [RFC6052]. For this and other reasons, such a mechanism is beyond 798 the scope of this document. 800 6. A Note on Port Mapping 802 In addition to overwriting IP addresses when packets are forwarded, 803 NAPT44 devices overwrite the source port number in outbound traffic, 804 and the destination port number in inbound traffic. This mechanism 805 is called "port mapping". 807 The major benefit of port mapping is that it allows multiple 808 computers to share a single IPv4 address. A large number of internal 809 IPv4 addresses (typically from one of the [RFC1918] private address 810 spaces) can be mapped into a single external, globally routable IPv4 811 address, with the local port number used to identify which internal 812 node should receive each inbound packet. This address amplification 813 feature is not generally foreseen as a necessity at this time. 815 Since port mapping requires re-writing a portion of the transport 816 layer header, it requires NAPT44 devices to be aware of all of the 817 transport protocols that they forward, thus stifling the development 818 of new and improved transport protocols and preventing the use of 819 IPsec encryption. Modifying the transport layer header is 820 incompatible with security mechanisms that encrypt the full IP 821 payload, and restricts the NAPT44 to forwarding transport layers that 822 use weak checksum algorithms that are easily recalculated in routers. 824 Since there is significant detriment caused by modifying transport 825 layer headers and very little, if any, benefit to the use of port 826 mapping in IPv6, NPTv6 Translators that comply with this 827 specification MUST NOT perform port mapping. 829 7. Security Considerations 831 When NPTv6 is deployed using either of the two-way, algorithmic 832 mappings defined in the document, it allows direct inbound 833 connections to internal nodes. While this can be viewed as a benefit 834 of NPTv6 vs. NAPT44, it does open internal nodes to attacks that 835 would be more difficult in a NAPT44 network. Although this situation 836 is not substantially worse, from a security standpoint, than running 837 IPv6 with no NAT, some enterprises may assume that a NPTv6 Translator 838 will offer similar protection to a NAPT44 device. 840 The port mapping mechanism in NAPT44 implementations require that 841 state be created in both directions. This has lead to an industry- 842 wide perception that NAT functionality is the same as a stateful 843 firewall. It is not. The translation function of the NAT only 844 creates dynamic state in one direction and has no policy. For this 845 reason, it is RECOMMENDED that NPTv6 Translators also implement 846 firewall functionality such as described in [RFC6092], with 847 appropriate configuration options including turning it on or off. 849 When [RFC4864] talks about randomizing the subnet identifier, the 850 idea is to make it harder for worms to guess a valid subnet 851 identifier at an advertised network prefix. This should not be 852 interpreted as endorsing concealing the subnet identifier behind the 853 obfuscating function of a translator such as NPTv6. [RFC4864] 854 specifically talks about how to obtain the desired properties of 855 concealment without using a translator. Topology hiding when using 856 NAT is often ineffective in environments where the topology is 857 visible in application layer messaging protocols such as DNS, SIP, 858 SMTP, etc. If the information were not available through the 859 application layer, [RFC2993] would not be valid. 861 8. IANA Considerations 863 This document has no IANA considerations. 865 9. Acknowledgements 867 The checksum-neutral algorithmic address mapping described in this 868 document is based on e-mail written by Iljtsch Van Beijnum. 870 The following people provided advice or review comments that 871 substantially improved this document: Christian Huitema, Dave Thaler, 872 Ed Jankiewicz, Eric Kline, Iljtsch Van Beijnum, Jari Arkko, Keith 873 Moore, Mark Townsley, Merike Kaeo, Ralph Droms, Remi Depres, Steve 874 Blake, and Tony Hain. 876 This document was written using the xml2rfc tool described in RFC 877 2629 [RFC2629]. 879 10. Change Log 881 This section should be removed by the RFC Editor. 883 10.1. Changes Between draft-mrw-behave-nat66-00 and -01 885 There were several minor changes made between the *behave-nat66-00 886 and -01 versions of this draft: 888 o Added Fred Baker as a co-author. 890 o Minor arithmetic corrections. 892 o Added AH to paragraph on NAT security issues. 894 o Added additional NAT topologies to overview (diagrams TBD). 896 10.2. Changes between *behave-nat66-01 and -02 898 There were further changes made between *behave-nat66-01 and -02: 900 o Removed topology hiding mechanism. 902 o Added diagrams. 904 o Made minor updates based on mailing list feedback. 906 o Added discussion of IPv6 SAF document. 908 o Added applicability section. 910 o Added discussion of Address Independence requirement. 912 o Added hairpinning requirement and discussion of applicability of 913 other NAT behavioral requirements. 915 10.3. Changes between *nat66-00 and *nat66-01 917 There were further changes made between nat66-01 and nat66-02: 919 o Added mapping for prefixes longer than /48. 921 o Change draft name to remove reference to the behave WG. 923 o Resolved various open issues and fixed typos. 925 10.4. Changes between *nat66-01 and *nat66-02 927 o Change the acronym "NAT66" to "NPTv6", so people don't read "NAT" 928 and MEGO. 930 o Change the term used to refer to the function from "NAT66 device" 931 to "NPTv6 Translator". It's not a "device" function, it's a 932 function that is applied between two interfaces. Consider a 933 router with two upstreams and two legs in the local network; it 934 will not translate between the local legs, but will translate to 935 and from each upstream, and be configured differently for each of 936 the two ISPs. 938 o Comment specifically on the security aspects. 940 o Comment specifically on the application issues raised on this 941 list. 943 o Comment specifically on multihoming, load-sharing, and asymmetric 944 routing. 946 o Spell out the hairpinning requirement and its implications. 948 o Spell out the service provider side of Address Independence. 950 o 00 focuses on the edge's view 951 o Detail the algorithm in a manner clearer to the implementor (I 952 think) 954 o Spell out the case for GSE-style DMZs between the edge and the 955 transit network, which is about the implications for the global 956 routing table. 958 o Refer to [RFC6092] as a CPE firewall description. 960 10.5. Changes between *nat66-02 and *nat66-03 962 o Added an appendix on Verification code 964 o Various minor markups in response to Ralph Droms 966 10.6. Changes between *nat66-03 and *nat66-04 968 o Markups in response to Christian Huitema, mostly surrounding the 969 issue of subnet 0xFFFF. 971 o Refer to [I-D.ietf-v6ops-ipv6-cpe-router] for CPE router 972 requirements. 974 10.7. Changes between *nat66-04 and *nat66-05 976 o Update statistics in appendix A per BGP report of 17 December 2010 978 o Update security considerations using text supplied by Merike Kaeo. 980 10.8. Changes between *nat66-05 and *nat66-06 982 o restore a code snippet inadvertently removed in version -05 984 10.9. Changes between *nat66-06 and *nat66-07 986 o Changed requested status to experimental 988 o Incorporated comments from Eric Kline 990 10.10. Changes between *nat66-07 and *nat66-08 992 The section on Application Considerations was expanded after 993 discussion with Keith Moore. 995 11. References 996 11.1. Normative References 998 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 999 Requirement Levels", BCP 14, RFC 2119, March 1997. 1001 [RFC2526] Johnson, D. and S. Deering, "Reserved IPv6 Subnet Anycast 1002 Addresses", RFC 2526, March 1999. 1004 [RFC4193] Hinden, R. and B. Haberman, "Unique Local IPv6 Unicast 1005 Addresses", RFC 4193, October 2005. 1007 [RFC4291] Hinden, R. and S. Deering, "IP Version 6 Addressing 1008 Architecture", RFC 4291, February 2006. 1010 [RFC4443] Conta, A., Deering, S., and M. Gupta, "Internet Control 1011 Message Protocol (ICMPv6) for the Internet Protocol 1012 Version 6 (IPv6) Specification", RFC 4443, March 2006. 1014 [RFC4787] Audet, F. and C. Jennings, "Network Address Translation 1015 (NAT) Behavioral Requirements for Unicast UDP", BCP 127, 1016 RFC 4787, January 2007. 1018 11.2. Informative References 1020 [GSE] O'Dell, M., "GSE - An Alternate Addressing Architecture 1021 for IPv6", February 1997, 1022 . 1024 [I-D.ietf-behave-dns64] 1025 Bagnulo, M., Sullivan, A., Matthews, P., and I. Beijnum, 1026 "DNS64: DNS extensions for Network Address Translation 1027 from IPv6 Clients to IPv4 Servers", 1028 draft-ietf-behave-dns64-11 (work in progress), 1029 October 2010. 1031 [I-D.ietf-behave-v6v4-framework] 1032 Baker, F., Li, X., Bao, C., and K. Yin, "Framework for 1033 IPv4/IPv6 Translation", 1034 draft-ietf-behave-v6v4-framework-10 (work in progress), 1035 August 2010. 1037 [I-D.ietf-behave-v6v4-xlate] 1038 Li, X., Bao, C., and F. Baker, "IP/ICMP Translation 1039 Algorithm", draft-ietf-behave-v6v4-xlate-23 (work in 1040 progress), September 2010. 1042 [I-D.ietf-behave-v6v4-xlate-stateful] 1043 Bagnulo, M., Matthews, P., and I. Beijnum, "Stateful 1044 NAT64: Network Address and Protocol Translation from IPv6 1045 Clients to IPv4 Servers", 1046 draft-ietf-behave-v6v4-xlate-stateful-12 (work in 1047 progress), July 2010. 1049 [I-D.ietf-v6ops-ipv6-cpe-router] 1050 Singh, H., Beebee, W., Donley, C., Stark, B., and O. 1051 Troan, "Basic Requirements for IPv6 Customer Edge 1052 Routers", draft-ietf-v6ops-ipv6-cpe-router-09 (work in 1053 progress), December 2010. 1055 [RFC1071] Braden, R., Borman, D., Partridge, C., and W. Plummer, 1056 "Computing the Internet checksum", RFC 1071, 1057 September 1988. 1059 [RFC1624] Rijsinghani, A., "Computation of the Internet Checksum via 1060 Incremental Update", RFC 1624, May 1994. 1062 [RFC1918] Rekhter, Y., Moskowitz, R., Karrenberg, D., Groot, G., and 1063 E. Lear, "Address Allocation for Private Internets", 1064 BCP 5, RFC 1918, February 1996. 1066 [RFC2629] Rose, M., "Writing I-Ds and RFCs using XML", RFC 2629, 1067 June 1999. 1069 [RFC2827] Ferguson, P. and D. Senie, "Network Ingress Filtering: 1070 Defeating Denial of Service Attacks which employ IP Source 1071 Address Spoofing", BCP 38, RFC 2827, May 2000. 1073 [RFC2993] Hain, T., "Architectural Implications of NAT", RFC 2993, 1074 November 2000. 1076 [RFC3484] Draves, R., "Default Address Selection for Internet 1077 Protocol version 6 (IPv6)", RFC 3484, February 2003. 1079 [RFC4864] Van de Velde, G., Hain, T., Droms, R., Carpenter, B., and 1080 E. Klein, "Local Network Protection for IPv6", RFC 4864, 1081 May 2007. 1083 [RFC6052] Bao, C., Huitema, C., Bagnulo, M., Boucadair, M., and X. 1084 Li, "IPv6 Addressing of IPv4/IPv6 Translators", RFC 6052, 1085 October 2010. 1087 [RFC6092] Woodyatt, J., "Recommended Simple Security Capabilities in 1088 Customer Premises Equipment (CPE) for Providing 1089 Residential IPv6 Internet Service", RFC 6092, 1090 January 2011. 1092 Appendix A. Why GSE? 1094 For the purpose of this discussion, let us over-simplify the 1095 Internet's structure by distinguishing between two broad classes of 1096 networks: transit and edge. A "transit network", in this context, is 1097 a network that provides connectivity services to other networks. Its 1098 AS number may show up in a non-final position in BGP AS paths, or in 1099 the case of mobile and residential broadband networks, it may offer 1100 network services to smaller networks that can't justify RIR 1101 membership. An "edge network", in contrast, is any network that is 1102 not a transit network; it is the ultimate customer, and while it 1103 provides internal connectivity for its own use, it is in other 1104 respects is a consumer of transit services. In terms of routing, a 1105 network in the transit domain generally needs some way to make 1106 choices about how it routes to other networks; an edge network is 1107 generally quite satisfied with a simple default route. 1109 The [GSE] proposal, and as a result this proposal (which is similar 1110 to GSE in most respects and inspired by it), responds directly to 1111 current concerns in the RIR communities. Edge networks are used to 1112 an environment in IPv4 in which their addressing is disjoint from 1113 that of their upstream transit networks; it is either provider 1114 independent, or a network prefix translator makes their external 1115 address distinct from their internal address, and they like the 1116 distinction. In IPv6, there is a mantra that edge network addresses 1117 should be derived from their upstream, and if they have multiple 1118 upstreams, edge networks are expected to design their networks to use 1119 all of those prefixes equivalently. They see this as unnecessary and 1120 unwanted operational complexity, and are as a result pushing very 1121 hard in the RIR communities for provider independent addressing. 1123 Widespread use of provider independent addressing has a natural and 1124 perhaps unavoidable side-effect that is likely to be very expensive 1125 in the long term. It means that the routing table will enumerate the 1126 networks at the edge of the transit domain, the edge networks, rather 1127 than enumerating the transit domain. Per the BGP Update Report of 17 1128 December 2010, there are currently over 36,000 Autonomous Systems 1129 being advertised in BGP, of which over 15,000 advertise only one 1130 prefix. There are in the neighborhood of 5000 AS's that show up in a 1131 non-final position in AS paths, and perhaps another 5000 networks 1132 whose AS numbers are terminal in more than one AS path. In other 1133 words, we have prefixes for some 36,000 transit and edge networks in 1134 the route table now, many of which arguably need an Autonomous System 1135 number only for multihoming. Current estimates suggest that we could 1136 easily see that be on the order of 10,000,000 within fifteen years. 1137 Tens of thousands of entries in the 36,264 Autonomous Systems being 1138 advertised in BGP, of which 31,137 provide no visible transit service 1139 to another AS, and 23,595 of those are visible in only one AS path 1140 (have only one upstream network). In addition, of the 36,264 AS's in 1141 the world, 15,439 advertise only a single prefix. In other words, we 1142 have prefixes for some 36,000 transit and edge networks in the route 1143 table now, many of which arguably need an Autonomous System number 1144 only for multihoming. However, the vast majority of networks (2/3) 1145 having the tools necessary to multihome are not visibly doing so, and 1146 would be well served by any solution that gives them address 1147 independence without the overhead of RIR membership and BGP routing. 1149 Current growth estimates suggest that we could easily see that be on 1150 the order of 10,000,000 within fifteen years. Tens of thousands of 1151 entries in the route table is very survivable; while our protocols 1152 and computers will likely do quite well with tens of millions of 1153 routes, the heat produced and power consumed by those routers, and 1154 the inevitable impact on the cost of those routers, is not a good 1155 outcome. To avoid having a massive and unscalable route table, we 1156 need to find a way that is politically acceptable and returns us to 1157 enumerating the transit domain, not the edge. 1159 There have been a number of proposals. As described, shim6 moves the 1160 complexity to the edge, and the edge is rebelling. Geographic 1161 addressing in essence forces ISPs to "own" geographic territory from 1162 a routing perspective, as otherwise there is no clue in the address 1163 as to what network a datagram should be delivered to in order to 1164 reach it. Metropolitan Addressing can imply regulatory authority, 1165 and even if it is implemented using internet exchange consortia, 1166 visits a great deal of complexity on the transit networks that 1167 directly serve the edge. The one that is likely to be most 1168 acceptable is any proposal that enables an edge network to be 1169 operationally independent of its upstreams, with no obligation to 1170 renumber when it adds, drops, or changes ISPs, and with no additional 1171 burden placed either on the ISP or the edge network as a result. 1172 From an application perspective, an additional operational 1173 requirement in the words of US NIST's Roadmap for the Smart Grid, is 1174 that 1176 "...the Network should enable an application in a particular 1177 domain to communicate with an application in any other domain in 1178 the information network, with proper management control over who 1179 and where applications can be interconnected." 1181 In other words, the structure of the network should allow for and 1182 enable appropriate access control, but the structure of the network 1183 should not inherently limit access. 1185 The GSE model, by statelessly translating the prefix between an edge 1186 network and its upstream transit network, accomplishes that with a 1187 minimum of fuss and bother. Stated in the simplest terms, it enables 1188 the edge network to behave as if it has a provider-independent prefix 1189 from a multihoming and renumbering perspective without the overhead 1190 of RIR membership or maintaining BGP connectivity, and it enables the 1191 transit networks to aggressively aggregate what are from their 1192 perspective provider-allocated customer prefixes, to maintain a 1193 rational-sized routing table. 1195 Appendix B. Verification code 1197 This non-normative appendix is presented as a proof of concept. It 1198 is in no sense optimized; for example, one's complement arithmetic is 1199 implemented in portable subroutines, where operational 1200 implementations might use one's complement arithmetic instructions 1201 through a pragma; such implementations probably need to explicitly 1202 force 0xFFFF to 0x0000, as the instruction will not. The original 1203 purpose of the code was to verify whether or not it was necessary to 1204 suppress 0xFFFF by overwriting with zero, and whether predicted 1205 issues with subnet numbering were real. 1207 The point is to 1209 o demonstrate that if one or the other representation of zero is not 1210 used in the word the checksum is updated in, the program maps 1211 inner and outer addresses in a manner that is, mathematically, 1:1 1212 and onto (each inner address maps to a unique outer address, and 1213 that outer address maps back to exactly the same inner address), 1214 and 1216 o give guidance on the suppression of 0xFFFF checksums. 1218 In short, in one's complement arithmetic, x-x=0, but will take the 1219 negative representation of zero. If 0xFFFF results are forced to the 1220 value 0x0000, as is recommended in [RFC1071], the word the checksum 1221 is adjusted in cannot be initially 0xFFFF, as on the return it will 1222 be forced to 0. If 0xFFFF results are not forced to the value 0x0000 1223 as is recommended in [RFC1071], the word the checksum is adjusted in 1224 cannot be initially 0, as on the return it will be calculated as 1225 0+(~0) = 0xFFFF. We chose to follow [RFC1071]'s recommendations, 1226 which implies a requirement to not use 0xFFFF as a subnet number in 1227 networks with a /48 external prefix. 1229 /* 1230 * Copyright (c) 2010 IETF Trust and the persons identified as 1231 * authors of the code. All rights reserved. Redistribution 1232 * and use in source and binary forms, with or without 1233 * modification, are permitted provided that the following 1234 * conditions are met: 1236 * 1237 * o Redistributions of source code must retain the above 1238 * copyright notice, this list of conditions and the 1239 * following disclaimer. 1240 * 1241 * o Redistributions in binary form must reproduce the above 1242 * copyright notice, this list of conditions and the 1243 * following disclaimer in the documentation and/or other 1244 * materials provided with the distribution. 1245 * 1246 * o Neither the name of Internet Society, IETF or IETF Trust, 1247 * nor the names of specific contributors, may be used to 1248 * endorse or promote products derived from this software 1249 * without specific prior written permission. 1250 * 1251 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND 1252 * CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, 1253 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 1254 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 1255 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR 1256 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 1257 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 1258 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 1259 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 1260 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 1261 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 1262 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 1263 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 1264 */ 1265 #include "stdio.h" 1266 #include "assert.h" 1267 /* 1268 * program to verify the NPTv6 algorithm 1269 * 1270 * argument: 1271 * perform negative zero suppression: boolean 1272 * 1273 * method: 1274 * We specify an internal and an external prefix. The prefix 1275 * length is presumed to be the common length of both, and for 1276 * this is a /48. We perform the three algorithms specified. 1277 * the "packet" address is in effect the source address 1278 * internal->external and the destination address 1279 * external->internal. 1280 */ 1281 unsigned short inner_init[] = { 1282 0xFD01, 0x0203, 0x0405, 1, 2, 3, 4, 5}; 1283 unsigned short outer_init[] = { 1284 0x2001, 0x0db8, 0x0001, 1, 2, 3, 4, 5}; 1285 unsigned short inner[8]; 1286 unsigned short packet[8]; 1287 unsigned char checksum[65536] = {0}; 1288 unsigned short outer[8]; 1289 unsigned short adjustment; 1290 unsigned short suppress; 1291 /* 1292 * One's complement sum. 1293 * return number1 + number2 1294 */ 1295 unsigned short 1296 add1(number1, number2) 1297 unsigned short number1; 1298 unsigned short number2; 1299 { 1300 unsigned int result; 1302 result = number1; 1303 result += number2; 1304 if (suppress) { 1305 while (0xFFFF <= result) { 1306 result = result + 1 - 0x10000; 1307 } 1308 } else { 1309 while (0xFFFF < result) { 1310 result = result + 1 - 0x10000; 1311 } 1312 } 1313 return result; 1314 } 1316 /* 1317 * One's complement difference 1318 * return number1 - number2 1319 */ 1320 unsigned short 1321 sub1(number1, number2) 1322 unsigned short number1; 1323 unsigned short number2; 1324 { 1325 return add1(number1, ~number2); 1326 } 1328 /* 1329 * return one's complement sum of an array of numbers 1330 */ 1331 unsigned short 1332 sum1(numbers, count) 1333 unsigned short *numbers; 1334 int count; 1335 { 1336 unsigned int result; 1338 result = *numbers++; 1339 while (--count > 0) { 1340 result += *numbers++; 1341 } 1343 if (suppress) { 1344 while (0xFFFF <= result) { 1345 result = result + 1 - 0x10000; 1346 } 1347 } else { 1348 while (0xFFFF < result) { 1349 result = result + 1 - 0x10000; 1350 } 1351 } 1352 return result; 1353 } 1355 /* 1356 * NPTv6 initialization: section 3.1 assuming section 3.4 1357 * 1358 * create the /48, a source address in internal format, and a 1359 * source address in external format. calculate the adjustment 1360 * if one /48 is overwritten with the other. 1361 */ 1362 void 1363 nptv6_initialization(subnet) 1364 unsigned short subnet; 1365 { 1366 int i; 1367 unsigned short inner48; 1368 unsigned short outer48; 1370 /* initialize the internal and external prefixes. */ 1371 for (i = 0; i < 8; i++) { 1372 inner[i] = inner_init[i]; 1373 outer[i] = outer_init[i]; 1374 } 1375 inner[3] = subnet; 1376 outer[3] = subnet; 1377 /* calculate the checksum adjustment */ 1378 inner48 = sum1(inner, 3); 1379 outer48 = sum1(outer, 3); 1380 adjustment = sub1(inner48, outer48); 1381 } 1383 /* 1384 * NPTv6 packet from edge to transit: section 3.2 assuming 1385 * section 3.4 1386 * 1387 * overwrite the prefix in the source address with the outer 1388 * prefix, and adjust the checksum 1389 */ 1390 void 1391 nptv6_inner_to_outer() 1392 { 1393 int i; 1395 /* let's get the source address into the packet */ 1396 for (i = 0; i < 8; i++) { 1397 packet[i] = inner[i]; 1398 } 1400 /* overwrite the prefix with the outer prefix */ 1401 for (i = 0; i < 3; i++) { 1402 packet[i] = outer[i]; 1403 } 1405 /* adjust the checksum */ 1406 packet[3] = add1(packet[3], adjustment); 1407 } 1409 /* 1410 * NPTv6 packet from transit to edge:: section 3.3 assuming 1411 * section 3.4 1412 * 1413 * overwrite the prefix in the destination address with the 1414 * inner prefix, and adjust the checksum 1415 */ 1416 void 1417 nptv6_outer_to_inner() 1418 { 1419 int i; 1421 /* overwrite the prefix with the outer prefix */ 1422 for (i = 0; i < 3; i++) { 1423 packet[i] = inner[i]; 1424 } 1426 /* adjust the checksum */ 1427 packet[3] = sub1(packet[3], adjustment); 1429 } 1431 /* 1432 * main program 1433 */ 1434 main(argc, argv) 1435 int argc; 1436 char **argv; 1437 { 1438 unsigned subnet; 1439 int i; 1441 if (argc < 2) { 1442 fprintf(stderr, "usage: nptv6 supression\n"); 1443 assert(0); 1444 } 1445 suppress = atoi(argv[1]); 1446 assert(suppress <= 1); 1448 for (subnet = 0; subnet < 0x10000; subnet++) { 1449 /* section 3.1: initialize the system */ 1450 nptv6_initialization(subnet); 1452 /* section 3.2: take a packet from inside to outside */ 1453 nptv6_inner_to_outer(); 1455 /* the resulting checksum value should be unique */ 1456 if (checksum[subnet]) { 1457 printf("inner->outer duplicated checksum: " 1458 "inner: %x:%x:%x:%x:%x:%x:%x:%x(%x) " 1459 "calculated: %x:%x:%x:%x:%x:%x:%x:%x(%x)\n", 1460 inner[0], inner[1], inner[2], inner[3], 1461 inner[4], inner[5], inner[6], inner[7], 1462 sum1(inner, 8), 1463 packet[0], packet[1], packet[2], packet[3], 1464 packet[4], packet[5], packet[6], packet[7], 1465 sum1(packet, 8)); 1466 } 1468 checksum[subnet] = 1; 1470 /* 1471 * the resulting checksum should be the same as the inner 1472 * address's checksum 1473 */ 1474 if (sum1(packet, 8) != sum1(inner, 8)) { 1475 printf("inner->outer incorrect: " 1476 "inner: %x:%x:%x:%x:%x:%x:%x:%x(%x) " 1477 "calculated: %x:%x:%x:%x:%x:%x:%x:%x(%x)\n", 1478 inner[0], inner[1], inner[2], inner[3], 1479 inner[4], inner[5], inner[6], inner[7], 1480 sum1(inner, 8), 1481 packet[0], packet[1], packet[2], packet[3], 1482 packet[4], packet[5], packet[6], packet[7], 1483 sum1(packet, 8)); 1484 } 1486 /* section 3.3: take a packet from outside to inside */ 1487 nptv6_outer_to_inner(); 1489 /* 1490 * the returning packet should have the same checksum it 1491 * left with 1492 */ 1493 if (sum1(packet, 8) != sum1(inner, 8)) { 1494 printf("outer->inner checksum incorrect: " 1495 "calculated: %x:%x:%x:%x:%x:%x:%x:%x(%x) " 1496 "inner: %x:%x:%x:%x:%x:%x:%x:%x(%x)\n", 1497 packet[0], packet[1], packet[2], packet[3], 1498 packet[4], packet[5], packet[6], packet[7], 1499 sum1(packet, 8), inner[0], inner[1], inner[2], 1500 inner[3], inner[4], inner[5], inner[6], 1501 inner[7], sum1(inner, 8)); 1502 } 1504 /* 1505 * and every octet should calculate back to the same inner 1506 * value 1507 */ 1508 for (i = 0; i < 8; i++) { 1509 if (inner[i] != packet[i]) { 1510 printf("outer->inner different: " 1511 "calculated: %x:%x:%x:%x:%x:%x:%x:%x " 1512 "inner: %x:%x:%x:%x:%x:%x:%x:%x\n", 1513 packet[0], packet[1], packet[2], packet[3], 1514 packet[4], packet[5], packet[6], packet[7], 1515 inner[0], inner[1], inner[2], inner[3], 1516 inner[4], inner[5], inner[6], inner[7]); 1517 break; 1518 } 1519 } 1520 } 1521 } 1523 Authors' Addresses 1525 Margaret Wasserman 1526 Painless Security 1527 North Andover, MA 01845 1528 USA 1530 Phone: +1 781 405 7464 1531 Email: mrw@painless-security.com 1532 URI: http://www.painless-secuirty.com 1534 Fred Baker 1535 Cisco Systems 1536 Santa Barbara, California 93117 1537 USA 1539 Phone: +1-408-526-4257 1540 Email: fred@cisco.com