idnits 2.17.1 draft-ietf-tsvwg-transport-encrypt-10.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 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 761: '... IPv6 "source nodes SHOULD assign each...' Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (January 9, 2020) is 1570 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Outdated reference: A later version (-17) exists of draft-ietf-ippm-ioam-data-06 == Outdated reference: A later version (-34) exists of draft-ietf-quic-transport-22 == Outdated reference: A later version (-12) exists of draft-ietf-taps-transport-security-08 -- Obsolete informational reference (is this intentional?): RFC 4995 (Obsoleted by RFC 5795) -- Obsolete informational reference (is this intentional?): RFC 6347 (Obsoleted by RFC 9147) Summary: 1 error (**), 0 flaws (~~), 4 warnings (==), 3 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 TSVWG G. Fairhurst 3 Internet-Draft University of Aberdeen 4 Intended status: Informational C. Perkins 5 Expires: July 12, 2020 University of Glasgow 6 January 9, 2020 8 Considerations around Transport Header Confidentiality, Network 9 Operations, and the Evolution of Internet Transport Protocols 10 draft-ietf-tsvwg-transport-encrypt-10 12 Abstract 14 To protect user data and privacy, Internet transport protocols have 15 supported payload encryption and authentication for some time. Such 16 encryption and authentication is now also starting to be applied to 17 the transport protocol headers. This helps avoid transport protocol 18 ossification by middleboxes, while also protecting metadata about the 19 communication. Current operational practice in some networks inspect 20 transport header information within the network, but this is no 21 longer possible when those transport headers are encrypted. This 22 document discusses the possible impact when network traffic uses a 23 protocol with an encrypted transport header. It suggests issues to 24 consider when designing new transport protocols, to account for 25 network operations, prevent network ossification, and enable 26 transport evolution, while still respecting user privacy. 28 Status of This Memo 30 This Internet-Draft is submitted in full conformance with the 31 provisions of BCP 78 and BCP 79. 33 Internet-Drafts are working documents of the Internet Engineering 34 Task Force (IETF). Note that other groups may also distribute 35 working documents as Internet-Drafts. The list of current Internet- 36 Drafts is at https://datatracker.ietf.org/drafts/current/. 38 Internet-Drafts are draft documents valid for a maximum of six months 39 and may be updated, replaced, or obsoleted by other documents at any 40 time. It is inappropriate to use Internet-Drafts as reference 41 material or to cite them other than as "work in progress." 43 This Internet-Draft will expire on July 12, 2020. 45 Copyright Notice 47 Copyright (c) 2020 IETF Trust and the persons identified as the 48 document authors. All rights reserved. 50 This document is subject to BCP 78 and the IETF Trust's Legal 51 Provisions Relating to IETF Documents 52 (https://trustee.ietf.org/license-info) in effect on the date of 53 publication of this document. Please review these documents 54 carefully, as they describe your rights and restrictions with respect 55 to this document. Code Components extracted from this document must 56 include Simplified BSD License text as described in Section 4.e of 57 the Trust Legal Provisions and are provided without warranty as 58 described in the Simplified BSD License. 60 Table of Contents 62 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 63 2. Context and Rationale . . . . . . . . . . . . . . . . . . . . 4 64 2.1. Use of Transport Header Information in the Network . . . 5 65 2.2. Authentication of Transport Header Information . . . . . 7 66 2.3. Observable Transport Header Fields . . . . . . . . . . . 7 67 3. Current uses of Transport Headers within the Network . . . . 10 68 3.1. Observing Transport Information in the Network . . . . . 11 69 3.2. Transport Measurement . . . . . . . . . . . . . . . . . . 18 70 3.3. Use for Network Diagnostics and Troubleshooting . . . . . 21 71 3.4. Header Compression . . . . . . . . . . . . . . . . . . . 23 72 4. Encryption and Authentication of Transport Headers . . . . . 23 73 4.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 23 74 4.2. Approaches to Transport Header Protection . . . . . . . . 24 75 5. Addition of Transport Information to Network-Layer Headers . 26 76 5.1. Use of OAM within a Maintenance Domain . . . . . . . . . 26 77 5.2. Use of OAM across Multiple Maintenance Domains . . . . . 26 78 6. Implications of Protecting the Transport Headers . . . . . . 27 79 6.1. Independent Measurement . . . . . . . . . . . . . . . . . 28 80 6.2. Characterising "Unknown" Network Traffic . . . . . . . . 30 81 6.3. Accountability and Internet Transport Protocols . . . . . 30 82 6.4. Impact on Operational Cost . . . . . . . . . . . . . . . 31 83 6.5. Impact on Research, Development and Deployment . . . . . 31 84 7. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 32 85 8. Security Considerations . . . . . . . . . . . . . . . . . . . 35 86 9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 37 87 10. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 38 88 11. Informative References . . . . . . . . . . . . . . . . . . . 38 89 Appendix A. Revision information . . . . . . . . . . . . . . . . 45 90 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 47 92 1. Introduction 94 Transport protocols have supported end-to-end encryption of payload 95 data for many years. Examples include Transport Layer Security (TLS) 96 over TCP [RFC8446], Datagram TLS (DTLS) over UDP [RFC6347], Secure 97 RTP [RFC3711], and TCPcrypt [RFC8548] which permits opportunistic 98 encryption of the TCP transport payload. Some of these also provide 99 integrity protection of all or part of the transport header. 101 This end-to-end transport payload encryption brings many benefits in 102 terms of providing confidentiality and protecting user privacy. The 103 benefits have been widely discussed, for example in [RFC7258] and 104 [RFC7624]. This document strongly supports and encourages increased 105 use of end-to-end payload encryption in transport protocols. The 106 implications of protecting the transport payload data are therefore 107 not further discussed in this document. 109 A further level of protection can be achieved by encrypting the 110 entire network layer payload, including both the transport headers 111 and the payload. This does not expose any transport information to 112 devices in the network, and therefore also prevents modification 113 along a network path. An example of encryption at the network layer 114 is the IPsec Encapsulating Security Payload (ESP) [RFC4303] in tunnel 115 mode. Virtual Private Networks (VPNs) typically also operate in this 116 way. This form of encryption is not further discussed in this 117 document. 119 There is also a middle ground, comprising transport protocols that 120 encrypt some, or all, of the transport layer header information, in 121 addition to encrypting the payload. An example of such a protocol, 122 that is now seeing widespread interest and deployment, is the QUIC 123 transport protocol [I-D.ietf-quic-transport]. The encryption and 124 authentication of transport header information can prevent unwanted 125 modification of transport headers by middleboxes, reducing the risk 126 of protocol ossification. It also reduces the amount of metadata 127 about the progress of the transport connection that is visible to the 128 network. 130 As discussed in [RFC7258], Pervasive Monitoring (PM) is a technical 131 attack that needs to be mitigated in the design of IETF protocols. 132 This document supports that conclusion and the use of transport 133 header encryption to protect against pervasive monitoring. RFC 7258 134 also notes, though, that "Making networks unmanageable to mitigate PM 135 is not an acceptable outcome, but ignoring PM would go against the 136 consensus documented here. An appropriate balance will emerge over 137 time as real instances of this tension are considered". 139 The transport protocols developed for the Internet are used across a 140 wide range of paths across network segments with many different 141 regulatory, commercial, and engineering considerations. This 142 document considers some of the costs and changes to network 143 management and research that are implied by widespread use of 144 transport protocols that encrypt their transport header information. 145 It reviews the implications of developing transport protocols that 146 use end-to-end encryption to provide confidentiality of their 147 transport layer headers, and considers the effect of such changes on 148 transport protocol design and network operations. It also considers 149 some anticipated implications on transport and application evolution. 150 This provides considerations relating to the design of transport 151 protocols that protect their header information and respect user 152 privacy. 154 2. Context and Rationale 156 The transport layer provides end-to-end interactions between 157 endpoints (processes) using an Internet path. Transport protocols 158 layer over the network-layer service, and are usually sent in the 159 payload of network-layer packets. They support end-to-end 160 communication between applications, using higher-layer protocols 161 running on the end systems (transport endpoints). 163 This simple architectural view does not present one of the core 164 functions of an Internet transport: to discover and adapt to the 165 network path that is currently being used. The design of Internet 166 transport protocols is as much about trying to avoid the unwanted 167 side effects of congestion on a flow and other capacity-sharing 168 flows, avoiding congestion collapse, adapting to changes in the path 169 characteristics, etc., as it is about end-to-end feature negotiation, 170 flow control, and optimising for performance of a specific 171 application. 173 Transport headers have end-to-end meaning, but have often been 174 observed by equipment within the network. Transport protocol 175 specifications have not tended to consider this, and have failed to 176 indicate what parts of the transport header are intended to be 177 invariant across protocol versions and visible to the network; what 178 parts of the header can be modified by the network to signal to the 179 transport, and in what way; and what parts of the header are private 180 and/or expected to change in future, and need to be protected for 181 privacy or to prevent protocol ossification. 183 Increasing concern about pervasive network monitoring 184 [RFC7258][RFC7624], and growing awareness of the problem of protocol 185 ossification caused by middlebox interference with Internet traffic, 186 has motivated a shift in transport protocol design. Recent transport 187 protocols, such as QUIC [I-D.ietf-quic-transport], encrypt the 188 majority of their transport headers to prevent observation and 189 protect against modification by the network, and to make explicit 190 their invariants and what is intended to be visible to the network. 192 Transport header encryption is expected to form a core part of future 193 transport protocol designs. It can help to protect against pervasive 194 monitoring, improve privacy, and reduce protocol ossification. 195 Transport protocols that use header encryption with secure key 196 distribution can provide confidentiality and protection for some, or 197 all, of the transport header information, controlling what is visible 198 to, and can be modified by, the network. 200 The increased use of transport header encryption has benefits, but 201 also has implications for the broader ecosystem. The transport 202 community has, to date, relied heavily on measurements and insights 203 from the network operations community to understand protocol 204 behaviour, and to inform the selection of appropriate mechanisms to 205 ensure a safe, reliable, and robust Internet. In turn, network 206 operators and access providers have relied upon being able to observe 207 traffic patterns and requirements, both in aggregate and at the flow 208 level, to help understand and optimise the behaviour of their 209 networks. Widespread use of transport header encryption could limit 210 such observations in future. It is important to understand how 211 transport header information is used in the network, to allow future 212 protocol designs to make an informed choice on what, if any, headers 213 to expose to the network. 215 2.1. Use of Transport Header Information in the Network 217 In-network measurement of transport flow characteristics can be used 218 to enhance performance, and control cost and service reliability. To 219 support network operations and enhance performance, some operators 220 have deployed functionality that utilises on-path observations of the 221 transport headers of packets passing through their network. 223 When network devices rely on the presence of a header field or the 224 semantics of specific header information, this can lead to 225 ossification where an endpoint has to supply a specific header to 226 receive the network service that it desires. 228 In some cases, network-layer use of transport header information can 229 be benign or advantageous to the protocol (e.g., recognising the 230 start of a TCP connection, providing header compression for a Secure 231 RTP flow, or explicitly using exposed protocol information to provide 232 consistent decisions by on-path devices). However, in other cases, 233 this can have unwanted outcomes, e.g., privacy impacts and 234 ossification. 236 Ossification can frustrate the evolution of a transport protocol. A 237 mechanism implemented in a network device, such as a firewall, that 238 requires a header field to have only a specific known set of values 239 can prevent the device from forwarding packets using a different 240 version of the protocol that introduces a feature that changes to a 241 new value for the observed field. 243 An example of ossification was observed in the development of 244 Transport Layer Security (TLS) 1.3 [RFC8446], where the design needed 245 to function in the presence of deployed middleboxes that relied on 246 the presence of certain header fields exposed in TLS 1.2. 248 The design of MPTCP also had to be revised to account for middleboxes 249 (known as "TCP Normalizers") that monitor the evolution of the window 250 advertised in the TCP header and then reset connections when the 251 window did not grow as expected. Similarly, issues have been 252 reported using TCP. For example, TCP Fast Open can experience 253 middleboxes that modify the transport header of packets by removing 254 "unknown" TCP options, segments with unrecognised TCP options can be 255 dropped, segments that contain data and set the SYN bit can be 256 dropped, or middleboxes that disrupt connections which send data 257 before completion of the three-way handshake. Other examples of 258 ossification have included middleboxes that rewrite TCP sequence and 259 acknowledgement numbers, but are unaware of the (newer) TCP selective 260 acknowledgement (SACK) Option and therefore fail to correctly rewrite 261 the selective acknowledgement header information to match the changes 262 that were made to the fixed TCP header, preventing SACK from 263 operating correctly. 265 In all these cases, middleboxes with a hard-coded understanding of 266 transport behaviour, interacted poorly with transport protocols after 267 the transport behaviour was changed. 269 In contrast, transport header encryption prevents an on-path device 270 from observing the transport headers, and therefore stops mechanisms 271 being built that directly rely on or infer semantics of the transport 272 header information. Encryption is normally combined with 273 authentication of the protected information. RFC 8546 summarises 274 this approach, stating that it is "The wire image, not the protocol's 275 specification, determines how third parties on the network paths 276 among protocol participants will interact with that protocol" 277 [RFC8546]. 279 While encryption can reduce ossification of the transport protocol, 280 it does not itself prevent ossification of the network service. 281 People seeking to understand network traffic could still come to rely 282 on pattern inferences and other heuristics or machine learning to 283 derive measurement data and as the basis for network forwarding 284 decisions. This can also create dependencies on the transport 285 protocol, or the patterns of traffic it can generate. 287 2.2. Authentication of Transport Header Information 289 The designers of a transport protocol decide whether to encrypt all, 290 or a part of, the transport header information. Section 4 of RFC8558 291 states: "Anything exposed to the path should be done with the intent 292 that it be used by the network elements on the path" [RFC8558]. New 293 protocol designs can decide not to encrypt certain transport header 294 fields, making those fields observable in the network. Where fields 295 are intended to immutable (i.e., observable but not modifiable by the 296 network), the endpoints are encouraged to use authentication to 297 provide a cryptographic integrity check that includes these immutable 298 fields to detect any manipulation by network devices. 300 Making part of a transport header observable can lead to ossification 301 of that part of a header, as middleboxes come to rely on observations 302 of the exposed fields. A protocol design that provides an observable 303 field might want to avoid inspection restricting the choice of usable 304 values in the field by intentionally varying the format and/or value 305 of the field to reduce the chance of ossification (see Section 4). 307 2.3. Observable Transport Header Fields 309 Transport headers fields have been observed within the network for a 310 variety of purposes. Some of these are related to network management 311 and operations. The lists below, and in the following section, seek 312 to identify some of these uses and the implications of the increased 313 use of transport header encryption. This analysis does not judge 314 whether specific practises are necessary, or endorse the use of any 315 specific approach. 317 Network Operations: Observable transport headers enable explicit 318 measurement and analysis of protocol performance, 319 network anomalies, and failure pathologies at any 320 point along the Internet path. In many cases, it 321 is important to relate observations to specific 322 equipment/configurations, to a specific network 323 segment, or sometimes to a specific protocol or 324 application. 326 When transport header information is not 327 observable, it cannot be used by network 328 operators. Operators might work without that 329 information, or they might turn to more ambitious 330 ways to collect, estimate, or infer this data. 331 (Operational practises aimed at guessing 332 transport parameters are out of scope for this 333 document, and are only mentioned here to 334 recognize that encryption does not stop operators 335 from attempting to apply practises that have been 336 used with unencrypted transport headers.) 338 See also Sections 3, 5, and 6.4. 340 Traffic Analysis: Observable transport headers have been used to 341 determine which transport protocols and features 342 are being used across a network segment, and to 343 measure trends in the pattern of usage. For some 344 use cases, end-to-end measurements/traces are 345 sufficient and can assist in developing and 346 debugging new transports and analysing their 347 deployment. In other uses, it is important to 348 relate observations to specific equipment/ 349 configurations or particular network segments. 351 This information can help anticipate the demand 352 for network upgrades and roll-out, or affect on- 353 going traffic engineering activities performed by 354 operators such as determining which parts of the 355 path contribute delay, jitter, or loss. 357 Tools that rely upon observing headers, could 358 fail to produce useful data when those headers 359 are encrypted. While this impact could, in many 360 cases, be small, there are scenarios where 361 operators have actively monitored and supported 362 particular services, e.g., to explore issues 363 relating to Quality of Service (QoS), to perform 364 fast re-routing of critical traffic, to mitigate 365 the characteristics of specific radio links, and 366 so on. 368 See also Sections 3.1-3.2, and 5. 370 Troubleshooting: Observable transport headers have been utilised 371 by operators as a part of network troubleshooting 372 and diagnostics. Metrics can help localise the 373 network segment introducing the loss or latency. 374 Effective troubleshooting often requires 375 understanding of transport behaviour. Flows 376 experiencing packet loss or jitter are hard to 377 distinguish from unaffected flows when only 378 observing network layer headers. 380 Observable transport feedback information (e.g., 381 RTP Control Protocol (RTCP) reception reports 382 [RFC3550]) can explicitly make loss metrics 383 visible to operators. Loss metrics can also be 384 deduced with more complexity from other header 385 information (e.g., by observing TCP SACK blocks). 386 When the transport header information is 387 encrypted, explicit observable fields could also 388 be made available at the network or transport 389 layers to provide these functions. 391 See also Section 3.3 and 5. 393 Network Protection: Observable transport headers currently provide 394 useful input to classify and detect anomalous 395 events, such as changes in application behaviour 396 or distributed denial of service attacks. 397 Operators often seek to uniquely disambiguate 398 unwanted traffic. 400 Where flows cannot be disambiguated based on 401 transport information, this could result in less- 402 efficient identification of unwanted traffic, the 403 introduction of rate limits for uncharacterised 404 traffic, or the use of heuristics to identify 405 anomalous flows. 407 See also Sections 6.2 and 6.3. 409 Verifiable Data: Observable transport headers can provide open and 410 verifiable measurements to support operations, 411 research, and protocol development. The ability 412 of multiple stake holders to review transport 413 header traces helps develop insight into 414 performance and traffic contribution of specific 415 variants of a protocol. Independently observed 416 data is important to help ensure the health of 417 the research and development communities. 419 When transport header information can not be 420 observed, this can reduce the range of actors 421 that can observe data. This limits the 422 information sources available to the Internet 423 community to understand the operation of new 424 transport protocols, reducing information to 425 inform design decisions and standardisation of 426 the new protocols and related operational 427 practises 428 See also Section 6. 430 SLA Compliance: Observable transport headers coupled with 431 published transport specifications allow 432 operators and regulators to explore the 433 compliance with Service Level Agreements (SLAs). 435 When transport header information can not be 436 observed, other methods have to be found to 437 confirm that the traffic produced conforms to the 438 expectations of the operator or developer. 440 Independently verifiable performance metrics can 441 be utilised to demonstrate regulatory compliance 442 in some jurisdictions, and as a basis for 443 informing design decisions. This can bring 444 assurance to those operating networks, often 445 avoiding deployment of complex techniques that 446 routinely monitor and manage Internet traffic 447 flows (e.g., avoiding the capital and operational 448 costs of deploying flow rate-limiting and network 449 circuit-breaker methods [RFC8084]). 451 See also Sections 5 and 6.1-6.3. 453 Note, again, that this lists uses that have been made of transport 454 header information, and does not necessarily endorse any particular 455 approach. 457 3. Current uses of Transport Headers within the Network 459 In response to pervasive monitoring [RFC7624] revelations and the 460 IETF consensus that "Pervasive Monitoring is an Attack" [RFC7258], 461 efforts are underway to increase encryption of Internet traffic. 462 Applying confidentiality to transport header fields affects how 463 protocol information is used [RFC8404], requiring consideration of 464 the trade-offs discussed in Section 2.3. 466 There are architectural challenges and considerations in the way 467 transport protocols are designed, and the ability to characterise and 468 compare different transport solutions [Measure]. The decision about 469 which transport headers fields are made observable offers trade-offs 470 around header confidentiality versus header observability (including 471 non-encrypted but authenticated header fields) for network operations 472 and management, and the implications for ossification and user 473 privacy. Different parties will view the relative importance of 474 these differently. For some, the benefits of encrypting all 475 transport headers outweigh the impact of doing so; others might 476 analyse the security, privacy and ossification impacts and arrive at 477 a different trade-off. 479 To understand the implications, it is necessary to understand how 480 transport layer headers are currently observed and/or modified by 481 middleboxes within the network. This section therefore reviews 482 examples of current usage. It does not consider the intentional 483 modification of transport headers by middleboxes (such as in Network 484 Address Translation, NAT, or Firewalls). Common issues concerning IP 485 address sharing are described in [RFC6269]. 487 3.1. Observing Transport Information in the Network 489 In-network observation of transport protocol headers requires 490 knowledge of the format of the transport header: 492 o Flows have to be identified at the level where observation is 493 performed. This implies visibility of the protocol and version of 494 the header, e.g., by defining the wire image [RFC8546]. As 495 protocols evolve over time, new transport headers could be 496 introduced. Detecting this could require interpretation of 497 protocol version information or connection setup information; 499 o Observing transport information depends on knowing the location 500 and syntax of the observed transport headers. IETF transport 501 protocols can specify this information. 503 The following subsections describe various ways that observable 504 transport information has been utilised. 506 3.1.1. Flow Identification Using Transport Layer Headers 508 Flow/Session identification [RFC8558] is a common function. For 509 example, performed by measurement activities, QoS classification, 510 firewalls, Denial of Service, DOS, prevention. 512 Observable transport header information, together with information in 513 the network header, has been used to identify flows and their 514 connection state, together with the set of protocol options being 515 used. Transport protocols, such as TCP and the Stream Control 516 Transport Protocol (SCTP), specify a standard base header that 517 includes sequence number information and other data. They also have 518 the possibility to negotiate additional headers at connection setup, 519 identified by an option number in the transport header. 521 In some uses, a low-numbered (well-known) transport port number can 522 identify the protocol. However, port information alone is not 523 sufficient to guarantee identification. Applications can use 524 arbitrary ports, multiple sessions can be multiplexed on a single 525 port, and ports can be re-used by subsequent sessions. UDP-based 526 protocols often do not use well-known port numbers. Some flows can 527 be identified by observing signalling protocol data (e.g., [RFC3261], 528 [I-D.ietf-rtcweb-overview]) or through the use of magic numbers 529 placed in the first byte(s) of the datagram payload [RFC7983]. 531 When transport header information can not be observed, this removes 532 information that could be used to classify flows by passive observers 533 along the path. More ambitious ways could be used to collect, 534 estimate, or infer flow information, including heuristics based on 535 the analysis of traffic patterns. For example, an operator that 536 cannot access the Session Description Protocol (SDP) session 537 descriptions to classify a flow as audio traffic, might instead use 538 (possibly less-reliable) heuristics to infer that short UDP packets 539 with regular spacing carry audio traffic. Operational practises 540 aimed at inferring transport parameters are out of scope for this 541 document, and are only mentioned here to recognize that encryption 542 does not prevent operators from attempting to apply practises that 543 were used with unencrypted transport headers. 545 3.1.2. Metrics derived from Transport Layer Headers 547 Observable transport headers enable explicit measurement and analysis 548 of protocol performance, network anomalies, and failure pathologies 549 at any point along the Internet path. Some operators use passive 550 monitoring to manage their portion of the Internet by characterizing 551 the performance of link/network segments. Inferences from transport 552 headers are used to derive performance metrics. A variety of open 553 source and commercial tools have been deployed that utilise transport 554 header information in this way to derive the following metrics: 556 Traffic Rate and Volume: Protocol sequence number and packet size 557 could be used to derive volume measures per-application, to 558 characterise the traffic that uses a network segment or the 559 pattern of network usage. Measurements can be per endpoint or for 560 an endpoint aggregate (e.g., to assess subscriber usage). 561 Measurements can also be used to trigger traffic shaping, and to 562 associate QoS support within the network and lower layers. Volume 563 measures can also be valuable for capacity planning and providing 564 detail of trends in usage. The traffic rate and volume can be 565 determined providing that the packets belonging to individual 566 flows can be identified, but there might be no additional 567 information about a flow when the transport headers cannot be 568 observed. 570 Loss Rate and Loss Pattern: Flow loss rate can be derived (e.g., 571 from transport sequence numbers or inferred from observing 572 transport protocol interactions) and has been used as a metric for 573 performance assessment and to characterise transport behaviour. 574 Understanding the location and root cause of loss can help an 575 operator determine whether this requires corrective action. 576 Network operators have used the variation in patterns of loss as a 577 key performance metric, utilising this to detect changes in the 578 offered service. 580 There are various causes of loss, including: corruption of link 581 frames (e.g., due to interference on a radio link), buffering loss 582 (e.g., overflow due to congestion, Active Queue Management, AQM 583 [RFC7567], or inadequate provision following traffic pre-emption), 584 and policing (traffic management). Understanding flow loss rates 585 requires either observing sequence numbers in network or transport 586 headers, or maintaining per-flow packet counters (flow 587 identification often requires transport header information). Per- 588 hop loss can also sometimes be monitored at the interface level by 589 devices in the network. 591 Losses can often occur as bursts, randomly-timed events, etc. The 592 pattern of loss can provide insight into the cause of loss. It 593 can also be valuable to understand the conditions under which loss 594 occurs, which usually requires relating loss to the traffic 595 flowing on the network node/segment at the time of loss. This can 596 also help identify cases where loss could have been wrongly 597 identified, or where the transport did not require transmission of 598 a lost packet. 600 Throughput and Goodput: Throughput is the amount of payload data 601 sent by a flow per time interval. Goodput [RFC7928] is a measure 602 of useful data exchanged (the ratio of useful data to total volume 603 of traffic sent by a flow). The throughput of a flow can be 604 determined in the absence of transport header information, 605 providing that the individual flow can be identified, and the 606 overhead known. Goodput requires ability to differentiate loss 607 and retransmission of packets, for example by observing packet 608 sequence numbers in the TCP or the Real-time Transport Protocol 609 (RTP) headers [RFC3550]. 611 Latency: Latency is a key performance metric that impacts 612 application and user-perceived response times. It often 613 indirectly impacts throughput and flow completion time. This 614 determines the reaction time of the transport protocol itself, 615 impacting flow setup, congestion control, loss recovery, and other 616 transport mechanisms. The observed latency can have many 617 components [Latency]. Of these, unnecessary/unwanted queuing in 618 network buffers has often been observed as a significant factor 620 [bufferbloat]. Once the cause of unwanted latency has been 621 identified, this can often be eliminated. 623 To measure latency across a part of a path, an observation point 624 [RFC7799] can measure the experienced round trip time (RTT) using 625 packet sequence numbers and acknowledgements, or by observing 626 header timestamp information. Such information allows an 627 observation point in the network to determine not only the path 628 RTT, but also allows measurement of the upstream and downstream 629 contribution to the RTT. This could be used to locate a source of 630 latency, e.g., by observing cases where the median RTT is much 631 greater than the minimum RTT for a part of a path. 633 The service offered by network operators can benefit from latency 634 information to understand the impact of configuration changes and 635 to tune deployed services. Latency metrics are key to evaluating 636 and deploying AQM [RFC7567], DiffServ [RFC2474], and Explicit 637 Congestion Notification (ECN) [RFC3168] [RFC8087]. Measurements 638 could identify excessively large buffers, indicating where to 639 deploy or configure AQM. An AQM method is often deployed in 640 combination with other techniques, such as scheduling [RFC7567] 641 [RFC8290] and although parameter-less methods are desired 642 [RFC7567], current methods often require tuning [RFC8290] 643 [RFC8289] [RFC8033] because they cannot scale across all possible 644 deployment scenarios. 646 Variation in delay: Some network applications are sensitive to 647 (small) changes in packet timing (jitter). Short and long-term 648 delay variation can impact on the latency of a flow and hence the 649 perceived quality of applications using the network. For example, 650 jitter metrics are often cited when characterising paths 651 supporting real-time traffic. The expected performance of such 652 applications, can be inferred from a measure the variation in 653 delay observed along a portion of the path [RFC3393] [RFC5481]. 654 The requirements resemble those for the measurement of latency. 656 Flow Reordering: Significant packet reordering within a flow can 657 impact time-critical applications and can be interpreted as loss 658 by reliable transports. Many transport protocol techniques are 659 impacted by reordering (e.g., triggering TCP retransmission or re- 660 buffering of real-time applications). Packet reordering can occur 661 for many reasons, from equipment design to misconfiguration of 662 forwarding rules. Network tools can detect and measure unwanted/ 663 excessive reordering, and the impact on transport performance. 665 There have been initiatives in the IETF transport area to reduce 666 the impact of reordering within a transport flow, possibly leading 667 to a reduction in the requirements for preserving ordering. These 668 have potential to simplify network equipment design as well as the 669 potential to improve robustness of the transport service. 670 Measurements of reordering can help understand the present level 671 of reordering within deployed infrastructure, and inform decisions 672 about how to progress such mechanisms. Key performance indicators 673 are retransmission rate, packet drop rate, sector utilisation 674 level, a measure of reordering, peak rate, the ECN congestion 675 experienced (CE) marking rate, etc. 677 Metrics have been defined that evaluate whether a network has 678 maintained packet order on a packet-by-packet basis [RFC4737] 679 [RFC5236]. 681 Techniques for measuring reordering typically observe packet 682 sequence numbers. Some protocols provide in-built monitoring and 683 reporting functions. Transport fields in the RTP header [RFC3550] 684 [RFC4585] can be observed to derive traffic volume measurements 685 and provide information on the progress and quality of a session 686 using RTP. As with other measurement, metadata assist in 687 understanding the context under which the data was collected, 688 including the time, observation point [RFC7799], and way in which 689 metrics were accumulated. The RTCP protocol directly reports some 690 of this information in a form that can be directly visible in the 691 network. A user of summary measurement data has to trust the 692 source of this data and the method used to generate the summary 693 information. 695 These metrics can support network operations, inform capacity 696 planning, and assist in determining the demand for equipment and/or 697 configuration changes by network operators. They can also inform 698 Internet engineering activities by informing the development of new 699 protocols, methodologies, and procedures. 701 In some cases, measurements could involve active injection of test 702 traffic to perform a measurement (see section 3.4 of [RFC7799]). 703 However, most operators do not have access to user equipment, 704 therefore the point of test is normally different from the transport 705 endpoint. Injection of test traffic can incur an additional cost in 706 running such tests (e.g., the implications of capacity tests in a 707 mobile network are obvious). Some active measurements [RFC7799] 708 (e.g., response under load or particular workloads) perturb other 709 traffic, and could require dedicated access to the network segment. 711 Passive measurements (see section 3.6 of [RFC7799]) can have 712 advantages in terms of eliminating unproductive test traffic, 713 reducing the influence of test traffic on the overall traffic mix, 714 and the ability to choose the point of observation (see 715 Section 3.2.1). Measurements can rely on observing packet headers, 716 which is not possible if those headers are encrypted, but could 717 utilise information about traffic volumes or patterns of interaction 718 to deduce metrics. 720 An alternative approach is to use in-network techniques add and 721 observe packet headers to facilitate measurements while traffic 722 traverses an operational network. This approach does not require the 723 cooperation of an endpoint. 725 3.1.3. Transport use of Network Layer Header Fields 727 Information from the transport protocol is used by a multi-field 728 classifier as a part of policy framework. Policies are commonly used 729 for management of the QoS or Quality of Experience (QoE) in resource- 730 constrained networks, and by firewalls to implement access rules (see 731 also section 2.2.2 of [RFC8404]). Network-layer classification 732 methods that rely on a multi-field classifier (e.g., inferring QoS 733 from the 5-tuple or choice of application protocol) are incompatible 734 with transport protocols that encrypt the transport information. 735 Traffic that cannot be classified typically receives a default 736 treatment. 738 Transport information can also be explicitly set in network-layer 739 header fields that are not encrypted, serving as a replacement/ 740 addition to the exposed transport information [RFC8558]. This 741 information can enable a different forwarding treatment by the 742 network, even when a transport employs encryption to protect other 743 header information. 745 The user of a transport that multiplexes multiple sub-flows might 746 want to obscure the presence and characteristics of these sub-flows. 747 On the other hand, an encrypted transport could set the network-layer 748 information to indicate the presence of sub-flows, and to reflect the 749 service requirements of individual sub-flows. There are several ways 750 this could be done: 752 IP Address: Applications normally expose the addresses used by 753 endpoints, and this is used in the forwarding decisions in network 754 devices. Address and other protocol information can be used by a 755 Multi-Field (MF) classifier to determine how traffic is treated 756 [RFC2475], and hence the quality of experience for a flow. 758 Using the IPv6 Network-Layer Flow Label: A number of Standards Track 759 and Best Current Practice RFCs (e.g., [RFC8085], [RFC6437], 760 [RFC6438]) encourage endpoints to set the IPv6 Flow label field of 761 the network-layer header. IPv6 "source nodes SHOULD assign each 762 unrelated transport connection and application data stream to a 763 new flow" [RFC6437]. A multiplexing transport could choose to use 764 multiple Flow labels to allow the network to independently forward 765 sub-flows. RFC6437 provides further guidance on choosing a flow 766 label value, stating these "should be chosen such that their bits 767 exhibit a high degree of variability", and chosen so that "third 768 parties should be unlikely to be able to guess the next value that 769 a source of flow labels will choose". 771 Once set, a flow label can provide information that can help 772 inform network-layer queuing and forwarding [RFC6438], for example 773 with Equal Cost Multi-Path routing and Link Aggregation [RFC6294]. 774 Considerations when using IPsec are further described in 775 [RFC6438]. 777 The choice of how to assign a Flow Label needs to avoid 778 introducing linkability that a network device could observe. 779 Inappropriate use by the transport can have privacy implications 780 (e.g., assigning the same label to two independent flows that 781 ought not to be classified the same). 783 Using the Network-Layer Differentiated Services Code Point: 784 Applications can expose their delivery expectations to the network 785 by setting the Differentiated Services Code Point (DSCP) field of 786 IPv4 and IPv6 packets [RFC2474]. For example, WebRTC applications 787 identify different forwarding treatments for individual sub-flows 788 (audio vs. video) based on the value of the DSCP field 789 [I-D.ietf-tsvwg-rtcweb-qos]). This provides explicit information 790 to inform network-layer queuing and forwarding, rather than an 791 operator inferring traffic requirements from transport and 792 application headers via a multi-field classifier. Inappropriate 793 use by the transport can have privacy implications (e.g., 794 assigning a different DSCP to a subflow could assist in a network 795 device discovering the traffic pattern used by an application, 796 assigning the same label to two independent flows that ought not 797 to be classified the same). The field is mutable, i.e., some 798 network devices can be expected to change this field (use of each 799 DSCP value is defined by an RFC). 801 Since the DSCP value can impact the quality of experience for a 802 flow, observations of service performance has to consider this 803 field when a network path supports differentiated service 804 treatment. 806 Using Explicit Congestion Marking: ECN [RFC3168] is a transport 807 mechanism that utilises the ECN field in the network-layer header. 808 Use of ECN explicitly informs the network-layer that a transport 809 is ECN-capable, and requests ECN treatment of the flow. An ECN- 810 capable transport can offer benefits when used over a path with 811 equipment that implements an AQM method with CE marking of IP 812 packets [RFC8087], since it can react to congestion without also 813 having to recover from lost packets. 815 ECN exposes the presence of congestion. The reception of CE- 816 marked packets can be used to estimate the level of incipient 817 congestion on the upstream portion of the path from the point of 818 observation (Section 2.5 of [RFC8087]). Interpreting the marking 819 behaviour (i.e., assessing congestion and diagnosing faults) 820 requires context from the transport layer, such as path RTT. 822 AQM and ECN offer a range of algorithms and configuration options. 823 Tools therefore have to be available to network operators and 824 researchers to understand the implication of configuration choices 825 and transport behaviour as the use of ECN increases and new 826 methods emerge [RFC7567]. 828 When transport headers cannot be observed, operators are unable to 829 use this information directly. Careful use of the network layer 830 features can help provide similar information in the case where the 831 network is unable to inspect transport protocol headers. 832 Section Section 5 describes use of network extension headers. 834 3.2. Transport Measurement 836 The common language between network operators and application/content 837 providers/users is packet transfer performance at a layer that all 838 can view and analyse. For most packets, this has been the transport 839 layer, until the emergence of transport protocols performing header 840 encryption, with the obvious exception of VPNs and IPsec. 842 When encryption hides more layers in each packet, people seeking 843 understanding of the network operation rely more on pattern inference 844 and other heuristics. It remains to be seen whether more complex 845 inferences can be mastered to produce the same monitoring accuracy 846 (see section 2.1.1 of [RFC8404]). 848 When measurement datasets are made available by servers or client 849 endpoints, additional metadata, such as the state of the network, is 850 often necessary to interpret this data to answer questions about 851 network performance or understand a pathology. Collecting and 852 coordinating such metadata is more difficult when the observation 853 point is at a different location to the bottleneck/device under 854 evaluation [RFC7799]. 856 Packet sampling techniques are used to scale the processing involved 857 in observing packets on high rate links. This exports only the 858 packet header information of (randomly) selected packets. The 859 utility of these measurements depends on the type of bearer and 860 number of mechanisms used by network devices. Simple routers are 861 relatively easy to manage, a device with more complexity demands 862 understanding of the choice of many system parameters. This level of 863 complexity exists when several network methods are combined. 865 This section discusses topics concerning observation of transport 866 flows, with a focus on transport measurement. 868 3.2.1. Point of Observation 870 On-path measurements are particularly useful for locating the source 871 of problems, or to assess the performance of a network segment or a 872 particular device configuration. Often issues can only be understood 873 in the context of the other flows that share a particular path, 874 common network device, interface port, etc. A simple example is 875 monitoring of a network device that uses a scheduler or active queue 876 management technique [RFC7567], where it could be desirable to 877 understand whether the algorithms are correctly controlling latency, 878 or if overload protection is working. This understanding implies 879 knowledge of how traffic is assigned to any sub-queues used for flow 880 scheduling, but can also require information about how the traffic 881 dynamics impact active queue management, starvation prevention 882 mechanisms, and circuit-breakers. 884 Sometimes multiple on-path observation points have to be used. By 885 correlating observations of headers at multiple points along the path 886 (e.g., at the ingress and egress of a network segment), an observer 887 can determine the contribution of a portion of the path to an 888 observed metric, to locate a source of delay, jitter, loss, 889 reordering, congestion marking, etc. 891 3.2.2. Use by Operators to Plan and Provision Networks 893 Traffic rate and volume measurements are used by operators to help 894 plan deployment of new equipment and configuration in their networks. 895 Data is also valuable to equipment vendors who want to understand 896 traffic trends and patterns of usage as inputs to decisions about 897 planning products and provisioning for new deployments. This 898 measurement information can also be correlated with billing 899 information when this is also collected by an operator. 901 Trends in aggregate traffic can be observed and can be related to the 902 endpoint addresses being used, but when transport information is not 903 observable, it might be impossible to correlate patterns in 904 measurements with changes in transport protocols. This increases the 905 dependency on other indirect sources of information to inform 906 planning and provisioning. 908 3.2.3. Service Performance Measurement 910 Performance measurements (e.g., throughput, loss, latency) can be 911 used by various actors to analyse the service offered to the users of 912 a network segment, and to inform operational practice. 914 3.2.4. Measuring Transport to Support Network Operations 916 The traffic that can be observed by on-path network devices (the 917 "wire image") is a function of transport protocol design/options, 918 network use, applications, and user characteristics. In general, 919 when only a small proportion of the traffic has a specific 920 (different) characteristic, such traffic seldom leads to operational 921 concern, although the ability to measure and monitor it is less. The 922 desire to understand the traffic and protocol interactions typically 923 grows as the proportion of traffic increases in volume. The 924 challenges increase when multiple instances of an evolving protocol 925 contribute to the traffic that share network capacity. 927 Operators can manage traffic load (e.g., when the network is severely 928 overloaded) by deploying rate-limiters, traffic shaping, or network 929 transport circuit breakers [RFC8084]. The information provided by 930 observing transport headers is a source of data that can help to 931 inform such mechanisms. 933 Congestion Control Compliance of Traffic: Congestion control is a 934 key transport function [RFC2914]. Many network operators 935 implicitly accept that TCP traffic complies with a behaviour that 936 is acceptable for the shared Internet. TCP algorithms have been 937 continuously improved over decades, and have reached a level of 938 efficiency and correctness that is difficult to match in custom 939 application-layer mechanisms [RFC8085]. 941 A standards-compliant TCP stack provides congestion control that 942 is judged safe for use across the Internet. Applications 943 developed on top of well-designed transports can be expected to 944 appropriately control their network usage, reacting when the 945 network experiences congestion, by back-off and reduce the load 946 placed on the network. This is the normal expected behaviour for 947 IETF-specified transports (e.g., TCP and SCTP). 949 However, when anomalies are detected, tools can interpret the 950 transport protocol header information to help understand the 951 impact of specific transport protocols (or protocol mechanisms) on 952 the other traffic that shares a network. An observation in the 953 network can gain an understanding of the dynamics of a flow and 954 its congestion control behaviour. Analysing observed flows can 955 help to build confidence that an application flow backs-off its 956 share of the network load under persistent congestion, and hence 957 to understand whether the behaviour is appropriate for sharing 958 limited network capacity. For example, it is common to visualise 959 plots of TCP sequence numbers versus time for a flow to understand 960 how a flow shares available capacity, deduce its dynamics in 961 response to congestion, etc. 963 The ability to identify sources that contribute to persistent 964 congestion is important to the safe operation of network 965 infrastructure, and can inform configuration of network devices to 966 complement the endpoint congestion avoidance mechanisms [RFC7567] 967 [RFC8084] to avoid a portion of the network being driven into 968 congestion collapse [RFC2914]. 970 Congestion Control Compliance for UDP traffic: UDP provides a 971 minimal message-passing datagram transport that has no inherent 972 congestion control mechanisms. Because congestion control is 973 critical to the stable operation of the Internet, applications and 974 other protocols that choose to use UDP as a transport have to 975 employ mechanisms to prevent collapse, avoid unacceptable 976 contributions to jitter/latency, and to establish an acceptable 977 share of capacity with concurrent traffic [RFC8085]. 979 A network operator can observe the headers of transport protocols 980 layered above UDP to understand if the datagram flows comply with 981 congestion control expectations. This can help inform a decision 982 on whether it might be appropriate to deploy methods such as rate- 983 limiters to enforce acceptable usage. 985 UDP flows that expose a well-known header can be observed to gain 986 understanding of the dynamics of a flow and its congestion control 987 behaviour. For example, tools exist to monitor various aspects of 988 RTP header information and RTCP reports for real-time flows (see 989 Section 3.1.2). The Secure RTP and RTCP extensions [RFC3711] were 990 explicitly designed to expose some header information to enable 991 such observation, while protecting the payload data. 993 3.3. Use for Network Diagnostics and Troubleshooting 995 Transport header information can be utilised for a variety of 996 operational tasks [RFC8404]: to diagnose network problems, assess 997 network provider performance, evaluate equipment or protocol 998 performance, capacity planning, management of security threats 999 (including denial of service), and responding to user performance 1000 questions. Section 3.1.2 and Section 5 of [RFC8404] provide further 1001 examples. 1003 Operators can monitor the health of a portion of the Internet, to 1004 provide early warning and trigger action. Traffic and performance 1005 measurements can assist in setting buffer sizes, debugging and 1006 diagnosing the root causes of faults that concern a particular user's 1007 traffic. They can also be used to support post-mortem investigation 1008 after an anomaly to determine the root cause of a problem. 1010 In other cases, measurement involves dissecting network traffic 1011 flows. Observed transport header information can help identify 1012 whether link/network tuning is effective and alert to potential 1013 problems that can be hard to derive from link or device measurements 1014 alone. 1016 An alternative could rely on access to endpoint diagnostic tools or 1017 user involvement in diagnosing and troubleshooting unusual use cases 1018 or to troubleshoot non-trivial problems. 1020 Another approach is to use traffic pattern analysis. Such tools can 1021 provide useful information during network anomalies (e.g., detecting 1022 significant reordering, high or intermittent loss), however indirect 1023 measurements would need to be carefully designed to provide reliable 1024 signals for diagnostics and troubleshooting. 1026 The design trade-offs for radio networks are often very different 1027 from those of wired networks. A radio-based network (e.g., cellular 1028 mobile, enterprise WiFi, satellite access/back-haul, point-to-point 1029 radio) has the complexity of a subsystem that performs radio resource 1030 management, with direct impact on the available capacity, and 1031 potentially loss/reordering of packets. The impact of the pattern of 1032 loss and congestion, differs for different traffic types, correlation 1033 with propagation and interference can all have significant impact on 1034 the cost and performance of a provided service. For radio links, the 1035 use for this type of information is expected to increase as operators 1036 bring together heterogeneous types of network equipment and seek to 1037 deploy opportunistic methods to access radio spectrum. 1039 Lack of tools and resulting information can reduce the ability of an 1040 operator to observe transport performance and could limit the ability 1041 of network operators to trace problems, make appropriate QoS 1042 decisions, or respond to other queries about the network service. 1044 A network operator supporting traffic that uses transport header 1045 encryption is unable to use tools that rely on transport protocol 1046 information. However, the use of encryption has the desirable effect 1047 of preventing unintended observation of the payload data and these 1048 tools seldom seek to observe the payload, or other application 1049 details. A flow that hides its transport header information could 1050 imply "don't touch" to some operators. This might limit a trouble- 1051 shooting response to "can't help, no trouble found". 1053 3.4. Header Compression 1055 Header compression saves link capacity by compressing network and 1056 transport protocol headers on a per-hop basis. It was widely used 1057 with low bandwidth dial-up access links, and still finds application 1058 on wireless links that are subject to capacity constraints. Header 1059 compression has been specified for use with TCP/IP and RTP/UDP/IP 1060 flows [RFC2507], [RFC2508], [RFC4995]. 1062 While it is possible to compress only the network layer headers, 1063 significant savings can be made if both the network and transport 1064 layer headers are compressed together as a single unit. The Secure 1065 RTP extensions [RFC3711] were explicitly designed to leave the 1066 transport protocol headers unencrypted, but authenticated, since 1067 support for header compression was considered important. Encrypting 1068 the transport protocol headers does not break such header 1069 compression, but does cause a fall back to compressing only the 1070 network layer headers, with a significant reduction in efficiency. 1072 4. Encryption and Authentication of Transport Headers 1074 End-to-end encryption can be applied at various protocol layers. It 1075 can be applied above the transport to encrypt the transport payload 1076 (e.g., using TLS). This can hide information from an eavesdropper in 1077 the network. It can also help protect the privacy of a user, by 1078 hiding data relating to user/device identity or location. 1080 4.1. Motivation 1082 There are several motivations for encryption: 1084 o One motive to encrypt transport headers is in response to 1085 perceptions that the network has become ossified, since traffic 1086 inspecting middleboxes prevent new protocols and mechanisms from 1087 being deployed. This has lead to a perception that there is too 1088 much "manipulation" of protocol headers within the network, and 1089 that designing to deploy in such networks is preventing transport 1090 evolution. One benefit of encrypting transport headers is that it 1091 can help improve the pace of transport development by eliminating 1092 interference by deployed middleboxes. 1094 o Another motivation stems from increased concerns about privacy and 1095 surveillance. Users value the ability to protect their identity 1096 and location, and defend against traffic analysis. Revelations 1097 about the use of pervasive surveillance [RFC7624] have, to some 1098 extent, eroded trust in the service offered by network operators 1099 and have led to an increased use of encryption to avoid unwanted 1100 eavesdropping on communications. Concerns have also been voiced 1101 about the addition of information to packets by third parties to 1102 provide analytics, customization, advertising, cross-site tracking 1103 of users, to bill the customer, or to selectively allow or block 1104 content. Whatever the reasons, the IETF is designing protocols 1105 that include transport header encryption (e.g., QUIC 1106 [I-D.ietf-quic-transport]) to supplement the already widespread 1107 payload encryption, and to further limit exposure of transport 1108 metadata to the network. 1110 The use of transport header authentication and encryption exposes a 1111 tussle between middlebox vendors, operators, applications developers 1112 and users: 1114 o On the one hand, future Internet protocols that support transport 1115 header encryption assist in the restoration of the end-to-end 1116 nature of the Internet by returning complex processing to the 1117 endpoints, since middleboxes cannot modify what they cannot see, 1118 and can improve privacy by reducing leakage of transport metadata. 1120 o On the other hand, encryption of transport layer header 1121 information has implications for people who are responsible for 1122 operating networks, and researchers and analysts seeking to 1123 understand the dynamics of protocols and traffic patterns. 1125 A decision to use transport header encryption can improve user 1126 privacy, and can reduce protocol ossification and help the evolution 1127 of the transport protocol stack, but is also has implications for 1128 network operations and management. 1130 4.2. Approaches to Transport Header Protection 1132 The following briefly reviews some security design options for 1133 transport protocols. A Survey of Transport Security Protocols 1134 [I-D.ietf-taps-transport-security] provides more details concerning 1135 commonly used encryption methods at the transport layer. 1137 Authenticating the Transport Protocol Header: Transport layer header 1138 information can be authenticated. An integrity check that 1139 protects the immutable transport header fields, but can still 1140 expose the transport protocol header information in the clear, 1141 allows in-network devices to observe these fields. An integrity 1142 check is not able to prevent in-network modification, but can 1143 prevent a receiving from accepting changes and avoid impact on the 1144 transport protocol operation. 1146 An example transport authentication mechanism is TCP- 1147 Authentication (TCP-AO) [RFC5925]. This TCP option authenticates 1148 the IP pseudo header, TCP header, and TCP data. TCP-AO protects 1149 the transport layer, preventing attacks from disabling the TCP 1150 connection itself and provides replay protection. TCP-AO might 1151 interact with middleboxes, depending on their behaviour [RFC3234]. 1153 The IPsec Authentication Header (AH) [RFC4302] was designed to 1154 work at the network layer and authenticate the IP payload. This 1155 approach authenticates all transport headers, and verifies their 1156 integrity at the receiver, preventing in-network modification. 1157 Secure RTP [RFC3711] is another example of a transport protocol 1158 that allows header authentication. 1160 Greasing: Protocols often provide extensibility features, reserving 1161 fields or values for use by future versions of a specification. 1162 The specification of receivers has traditionally ignored 1163 unspecified values, however in-network devices have emerged that 1164 ossify to require a certain value in a field, or re-use a field 1165 for another purpose. When the specification is later updated, it 1166 is impossible to deploy the new use of the field, and forwarding 1167 of the protocol could even become conditional on a specific header 1168 field value. 1170 A protocol can intentionally vary the value, format, and/or 1171 presence of observable transport header fields. This behaviour, 1172 known as GREASE (Generate Random Extensions And Sustain 1173 Extensibility) is designed to avoid a network device ossifying the 1174 use of a specific observable field. Greasing seeks to ease 1175 deployment of new methods. It can also prevent in-network devices 1176 utilising the information in a transport header, or can make an 1177 observation robust to a set of changing values, rather than a 1178 specific set of values 1180 Selectively Encrypting Transport Headers and Payload: A transport 1181 protocol design can encrypt selected header fields, while also 1182 choosing to authenticate the entire transport header. This allows 1183 specific transport header fields to be made observable by network 1184 devices. End-to end integrity checks can prevent an endpoint from 1185 undetected modification of the immutable transport headers. 1187 Mutable fields in the transport header provide opportunities for 1188 middleboxes to modify the transport behaviour (e.g., the extended 1189 headers described in [I-D.trammell-plus-abstract-mech]). This 1190 considers only immutable fields in the transport headers, that is, 1191 fields that can be authenticated End-to-End across a path. 1193 An example of a method that encrypts some, but not all, transport 1194 information is GRE-in-UDP [RFC8086] when used with GRE encryption. 1196 Optional Encryption of Header Information: There are implications to 1197 the use of optional header encryption in the design of a transport 1198 protocol, where support of optional mechanisms can increase the 1199 complexity of the protocol and its implementation, and in the 1200 management decisions that are have to be made to use variable 1201 format fields. Instead, fields of a specific type ought to always 1202 be sent with the same level of confidentiality or integrity 1203 protection. 1205 As seen, different transports use encryption to protect their header 1206 information to varying degrees. The trend is towards increased 1207 protection. 1209 5. Addition of Transport Information to Network-Layer Headers 1211 An on-path device can make measurements by utilising additional 1212 protocol headers carrying operations, administration and management 1213 (OAM) information in an additional packet header. Using network- 1214 layer approaches to reveal information has the potential that the 1215 same method (and hence same observation and analysis tools) can be 1216 consistently used by multiple transport protocols [RFC8558]. There 1217 could also be less desirable implications of separating the operation 1218 of the transport protocol from the measurement framework. 1220 5.1. Use of OAM within a Maintenance Domain 1222 OAM information can be added at the ingress to a maintenance domain 1223 (e.g., an Ethernet protocol header with timestamps and sequence 1224 number information using a method such as 802.11ag or in-situ OAM 1225 [I-D.ietf-ippm-ioam-data], or as a part of encapsulation protocol). 1226 The additional header information is typically removed the at the 1227 egress of the maintenance domain. 1229 Although some types of measurements are supported, this approach does 1230 not cover the entire range of measurements described in this 1231 document. In some cases, it can be difficult to position measurement 1232 tools at the appropriate segments/nodes and there can be challenges 1233 in correlating the downstream/upstream information when in-band OAM 1234 data is inserted by an on-path device. 1236 5.2. Use of OAM across Multiple Maintenance Domains 1238 OAM information can also be added at the network layer as an IPv6 1239 extension header or an IPv4 option. This information can be used 1240 across multiple network segments, or between the transport endpoints. 1242 One example is the IPv6 Performance and Diagnostic Metrics (PDM) 1243 Destination Option [RFC8250]. This allows a sender to optionally 1244 include a destination option that caries header fields that can be 1245 used to observe timestamps and packet sequence numbers. This 1246 information could be authenticated by receiving transport endpoints 1247 when the information is added at the sender and visible at the 1248 receiving endpoint, although methods to do this have not currently 1249 been proposed. This method has to be explicitly enabled at the 1250 sender. 1252 Current measurement results suggest that it could currently be 1253 undesirable to rely on methods requiring end-to-end support of 1254 network options or extension headers across the Internet. IPv4 1255 network options are often not supported (or are carried on a slower 1256 processing path) and some IPv6 networks have been observed to drop 1257 packets that set an IPv6 header extension (e.g., results from 2016 in 1258 [RFC7872]). 1260 Protocols can be designed to expose header information separately to 1261 the (hidden) fields used by the protocol state machine. On the one 1262 hand, such approaches can simplify tools by exposing the relevant 1263 metrics (loss, latency, etc), rather having to derive this from other 1264 fields. This also permits the protocol to evolve independently of 1265 the ossified observable header [RFC8558]. On the other hand, 1266 protocols do not necessarily have an incentive to expose the actual 1267 information that is utilised by the protocol itself and could 1268 therefore manipulate the exposed header information to gain an 1269 advantage from the network. Where the information is provided by an 1270 endpoint, the incentive to reflect actual transport information has 1271 to be considered when proposing a method. 1273 6. Implications of Protecting the Transport Headers 1275 The choice of which transport header fields to expose and which to 1276 encrypt is a design decision for the transport protocol. Selective 1277 encryption requires trading conflicting goals of observability and 1278 network support, privacy, and risk of ossification, to decide what 1279 header fields to protect and which to make visible. 1281 Security work typically employs a design technique that seeks to 1282 expose only what is needed. This approach provides incentives to not 1283 reveal any information that is not necessary for the end-to-end 1284 communication. However, there can be performance and operational 1285 benefits in exposing selected information to network tools. 1287 This section explores key implications of working with encrypted 1288 transport protocols. 1290 6.1. Independent Measurement 1292 Independent observation by multiple actors is important if the 1293 transport community is to maintain an accurate understanding of the 1294 network. Encrypting transport header encryption changes the ability 1295 to collect and independently analyse data. Internet transport 1296 protocols employ a set of mechanisms. Some of these have to work in 1297 cooperation with the network layer for loss detection and recovery, 1298 congestion detection and control. Others have to work only end-to- 1299 end (e.g., parameter negotiation, flow-control). 1301 The majority of present Internet applications use two well-known 1302 transport protocols, TCP and UDP. Although TCP represents the 1303 majority of current traffic, many real-time applications use UDP, and 1304 much of this traffic utilises RTP format headers in the payload of 1305 the UDP datagram. Since these protocol headers have been fixed for 1306 decades, a range of tools and analysis methods have became common and 1307 well-understood. 1309 Protocols that expose the state information used by the transport 1310 protocol in their header information (e.g., timestamps used to 1311 calculate the RTT, packet numbers used to asses congestion and 1312 requests for retransmission) provide an incentive for the sending 1313 endpoint to provide correct information, since the protocol will not 1314 work otherwise. This increases confidence that the observer 1315 understands the transport interaction with the network. For example, 1316 when TCP is used over an unencrypted network path (i.e., one that 1317 does not use IPsec or other encryption below the transport), it 1318 implicitly exposes header information that can be used for 1319 measurement at any point along the path. This information is 1320 necessary for the protocol's correct operation, therefore there is no 1321 incentive for a TCP or RTP implementation to put incorrect 1322 information in this transport header. A network device can have 1323 confidence that the well-known (and ossified) transport information 1324 represents the actual state of the endpoints. 1326 When encryption is used to hide some or all of the transport headers, 1327 the transport protocol chooses which information to reveal to the 1328 network about its internal state, what information to leave 1329 encrypted, and what fields to grease to protect against future 1330 ossification. Such a transport could be designed, for example, to 1331 provide summary data regarding its performance, congestion control 1332 state, etc., or to make an explicit measurement signal available. 1333 For example, a QUIC endpoint can optionally set the spin bit to 1334 reflect to explicitly reveal the RTT of an encrypted transport 1335 session to the on-path network devices [I-D.ietf-quic-transport]). 1337 When providing or using such information, it is important to consider 1338 the privacy of the user and their incentive for providing accurate 1339 and detailed information. Protocols that selectively reveal some 1340 transport state or measurement signals are choosing to establish a 1341 trust relationship with the network operators. There is no protocol 1342 mechanism that can guarantee that the information provided represents 1343 the actual transport state of the endpoints, since those endpoints 1344 can always send additional information in the encrypted part of the 1345 header, to update or replace whatever they reveal. This reduces the 1346 ability to independently measure and verify that a protocol is 1347 behaving as expected. For some operational uses, the information has 1348 to contain sufficient detail to understand, and possibly reconstruct, 1349 the network traffic pattern for further testing. In this case, 1350 operators have to gain the trust of transport protocol implementers 1351 if the transport headers are to correctly reveal such information. 1353 Operations, Administration, and Maintenance (OAM) data records 1354 [I-D.ietf-ippm-ioam-data] could be embedded into a variety of 1355 encapsulation methods at different layers to support the goals of a 1356 specific operational domain. OAM-related metadata can support 1357 functions such as performance evaluation, path-tracing, path 1358 verification information, classification and a diversity of other 1359 uses. When encryption is used to hide some or all of the transport 1360 headers, analysis requires coordination between actors at different 1361 layers to successfully characterise flows and correlate the 1362 performance or behaviour of a specific mechanism with the 1363 configuration and traffic using operational equipment (e.g., 1364 combining transport and network measurements to explore congestion 1365 control dynamics, the implications of designs for active queue 1366 management or circuit breakers). 1368 Some measurements could be completed by utilising endpoint-based 1369 logging (e.g., based on Quic-Trace [Quic-Trace]). Such information 1370 has a diversity of uses, including developers wishing to debug/ 1371 understand the transport/application protocols with which they work, 1372 researchers seeking to spot trends and anomalies, and to characterise 1373 variants of protocols. A standard format for endpoint logging could 1374 allow these to be shared (after appropriate anonymisation) to 1375 understand performance and pathologies. Measurements based on 1376 logging have to establish the validity and provenance of the logged 1377 information to establish how and when traces were captured. 1379 Despite being applicable in some scenarios, endpoint logs do not 1380 provide equivalent information to in-network measurements. In 1381 particular, endpoint logs contain only a part of the information to 1382 understand the operation of network devices and identify issues such 1383 as link performance or capacity sharing between multiple flows. 1384 Additional information has to be combined to determine which 1385 equipment/links are used and the configuration of equipment along the 1386 network paths being measured. 1388 6.2. Characterising "Unknown" Network Traffic 1390 The patterns and types of traffic that share Internet capacity change 1391 over time as networked applications, usage patterns and protocols 1392 continue to evolve. 1394 If "unknown" or "uncharacterised" traffic patterns form a small part 1395 of the traffic aggregate passing through a network device or segment 1396 of the network the path, the dynamics of the uncharacterised traffic 1397 might not have a significant collateral impact on the performance of 1398 other traffic that shares this network segment. Once the proportion 1399 of this traffic increases, monitoring the traffic can determine if 1400 appropriate safety measures have to be put in place. 1402 Tracking the impact of new mechanisms and protocols requires traffic 1403 volume to be measured and new transport behaviours to be identified. 1404 This is especially true of protocols operating over a UDP substrate. 1405 The level and style of encryption has to be considered in determining 1406 how this activity is performed. On a shorter timescale, information 1407 could also have to be collected to manage denial of service attacks 1408 against the infrastructure. 1410 6.3. Accountability and Internet Transport Protocols 1412 Information provided by tools observing transport headers can be used 1413 to classify traffic, and to limit the network capacity used by 1414 certain flows, as discussed in Section 3.2.4). Equally, operators 1415 could use analysis of transport headers and transport flow state to 1416 demonstrate that they are not providing differential treatment to 1417 certain flows. Obfuscating or hiding this information using 1418 encryption could lead operators and maintainers of middleboxes 1419 (firewalls, etc.) to seek other methods to classify, and potentially 1420 other mechanisms to condition, network traffic. 1422 A lack of data that reduces the level of precision with which flows 1423 can be classified also reduces the design space for conditioning 1424 mechanisms (e.g., rate limiting, circuit breaker techniques 1425 [RFC8084], or blocking of uncharacterised traffic), and this has to 1426 be considered when evaluating the impact of designs for transport 1427 encryption [RFC5218]. 1429 6.4. Impact on Operational Cost 1431 Some network operators currently use observed transport header 1432 information as a part of their operational practice, and have 1433 developed tools and techniques that use information observed in 1434 currently deployed transports and their applications. A variety of 1435 open source and proprietary tools have been deployed that use this 1436 information for a variety of short and long term measurements. 1437 Encryption of the transport information prevents tooling from 1438 observing the header information, limiting its utility. 1440 Alternative diagnostic and troubleshooting tools would have to be 1441 developed and deployed is transport header encryption is widely 1442 deployed. Introducing a new protocol or application might then 1443 require these tool chains and practises to be updated, and could in 1444 turn impact operational mechanisms, and policies. Each change can 1445 introduce associated costs, including the cost of collecting data, 1446 and the tooling to handle multiple formats (possibly as these co- 1447 exist in the network, when measurements span time periods during 1448 which changes are deployed, or to compare with historical data). 1449 These costs are incurred by an operator to manage the service and 1450 debug network issues. 1452 At the time of writing, the additional operational cost of using 1453 encrypted transports is not yet well understood. Design trade-offs 1454 could mitigate these costs by explicitly choosing to expose selected 1455 information (e.g., header invariants and the spin-bit in QUIC 1456 [I-D.ietf-quic-transport]), the specification of common log formats, 1457 and development of alternative approaches. 1459 6.5. Impact on Research, Development and Deployment 1461 Transport protocol evolution, and the ability to measure and 1462 understand the impact of protocol changes, have to proceed hand-in- 1463 hand. Observable transport headers can provide open and verifiable 1464 measurement data. Observation of pathologies has a critical role in 1465 the design of transport protocol mechanisms and development of new 1466 mechanisms and protocols. This helps understanding the interactions 1467 between cooperating protocols and network mechanism, the implications 1468 of sharing capacity with other traffic and the impact of different 1469 patterns of usage. The ability of other stake holders to review 1470 transport header traces helps develop insight into performance and 1471 traffic contribution of specific variants of a protocol. 1473 Development of new transport protocol mechanisms has to consider the 1474 scale of deployment and the range of environments in which the 1475 transport is used. Experience has shown that it is often difficult 1476 to correctly implement new mechanisms [RFC8085], and that mechanisms 1477 often evolve as a protocol matures, or in response to changes in 1478 network conditions, changes in network traffic, or changes to 1479 application usage. Analysis is especially valuable when based on the 1480 behaviour experienced across a range of topologies, vendor equipment, 1481 and traffic patterns. 1483 New transport protocol formats are expected to facilitate an 1484 increased pace of transport evolution, and with it the possibility to 1485 experiment with and deploy a wide range of protocol mechanisms. 1486 There has been recent interest in a wide range of new transport 1487 methods, e.g., Larger Initial Window, Proportional Rate Reduction 1488 (PRR), congestion control methods based on measuring bottleneck 1489 bandwidth and round-trip propagation time, the introduction of AQM 1490 techniques and new forms of ECN response (e.g., Data Centre TCP, 1491 DCTP, and methods proposed for L4S).The growth and diversity of 1492 applications and protocols using the Internet also continues to 1493 expand. For each new method or application it is desirable to build 1494 a body of data reflecting its behaviour under a wide range of 1495 deployment scenarios, traffic load, and interactions with other 1496 deployed/candidate methods. 1498 Encryption of transport header information could reduce the range of 1499 actors that can observe useful data. This would limit the 1500 information sources available to the Internet community to understand 1501 the operation of new transport protocols, reducing information to 1502 inform design decisions and standardisation of the new protocols and 1503 related operational practises. The cooperating dependence of 1504 network, application, and host to provide communication performance 1505 on the Internet is uncertain when only endpoints (i.e., at user 1506 devices and within service platforms) can observe performance, and 1507 when performance cannot be independently verified by all parties. 1509 Independently observed data is also important to ensure the health of 1510 the research and development communities and can help promote 1511 acceptance of proposed specifications by the wider community (e.g., 1512 as a method to judge the safety for Internet deployment) and provides 1513 valuable input during standardisation. Open standards motivate a 1514 desire to include independent observation and evaluation of 1515 performance data, which in turn demands control over where and when 1516 measurement samples are collected. This requires consideration of 1517 the methods used to observe data and the appropriate balance between 1518 encrypting all and no transport information. 1520 7. Conclusions 1522 Header encryption and strong integrity checks are being incorporated 1523 into new transport protocols and have important benefits. The pace 1524 of development of transports using the WebRTC data channel, and the 1525 rapid deployment of the QUIC transport protocol, can both be 1526 attributed to using the combination of UDP as a substrate while 1527 providing confidentiality and authentication of the encapsulated 1528 transport headers and payload. 1530 This document has described some current practises, and the 1531 implications for some stakeholders, when transport layer header 1532 encryption is used. It does not judge whether these practises are 1533 necessary, or endorse the use of any specific practise. Rather, the 1534 intent is to highlight operational tools and practises to consider 1535 when designing transport protocols, so protocol designers can make 1536 informed choice about what transport header fields to encrypt, and 1537 whether it might be beneficial to make an explicit choice to expose 1538 certain fields to the network. In making such a decision, it is 1539 important to balance: 1541 o User Privacy: The less transport header information that is 1542 exposed to the network, the lower the risk of leaking metadata 1543 that might have privacy implications for the users. Transports 1544 that chose to expose some header fields need to make a privacy 1545 assessment to understand the privacy cost versus benefit trade-off 1546 in making that information available. The process used to define 1547 and expose the QUIC spin bit to the network is an example of such 1548 an analysis. 1550 o Protocol Ossification: Unencrypted transport header fields are 1551 likely to ossify rapidly, as middleboxes come to rely on their 1552 presence, making it difficult to change the transport in future. 1553 This argues that the choice to expose information to the network 1554 is made deliberately and with care, since it is essentially 1555 defining a stable interface between the transport and the network. 1556 Some protocols will want to make that interface as limited as 1557 possible; other protocols might find value in exposing certain 1558 information to signal to the network, or in allowing the network 1559 to change certain header fields as signals to the transport. The 1560 visible wire image of a protocol should be explicitly designed. 1562 o Impact on Operational Practice: The network operations community 1563 has long relied on being able to understand Internet traffic 1564 patterns, both in aggregate and at the flow level, to support 1565 network management, traffic engineering, and troubleshooting. 1566 Operational practice has developed based on the information 1567 available from unencrypted transport headers. The IETF has 1568 supported this practice by developing operations and management 1569 specifications, interface specifications, and associated Best 1570 Current Practises. Widespread deployment of transport protocols 1571 that encrypt their header information might impact network 1572 operations, unless operators can develop alternative practises 1573 that work without access to the transport header information. 1575 o Pace of Evolution: Removing obstacles to change can enable an 1576 increased pace of evolution. If a protocol changes its transport 1577 header format (wire image) or their transport behaviour, this can 1578 result in the currently deployed tools and methods becoming no 1579 longer relevant. Where this needs to be accompanied by 1580 development of appropriate operational support functions and 1581 procedures, it can incur a cost in new tooling to catch-up with 1582 each change. Protocols that consistently expose observable data 1583 do not require such development, but can suffer from ossification 1584 and need to consider if the exposed protocol metadata has privacy 1585 implications, There is no single deployment context, and therefore 1586 designers need to consider the diversity of operational networks 1587 (ISPs, enterprises, DDoS mitigation and firewall maintainers, 1588 etc.). 1590 o Supporting Common Specifications: Common, open, specifications can 1591 stimulate engagement by developers, users, researchers, and the 1592 broader community. Increased protocol diversity can be beneficial 1593 in meeting new requirements, but the ability to innovate without 1594 public scrutiny risks point solutions that optimise for specific 1595 cases, but that can accidentally disrupt operations of/in 1596 different parts of the network. The social contract that 1597 maintains the stability of the Internet relies on accepting common 1598 interworking specifications, and on it being possible to detect 1599 violations. It is important to find new ways of maintaining that 1600 community trust as increased use of transport header encryption 1601 limits visibility into transport behaviour. 1603 o Impact on Benchmarking and Understanding Feature Interactions: An 1604 appropriate vantage point for observation, coupled with timing 1605 information about traffic flows, provides a valuable tool for 1606 benchmarking network devices, endpoint stacks, functions, and/or 1607 configurations. This can also help with understanding complex 1608 feature interactions. An inability to observe transport layer 1609 header information can make it harder to diagnose and explore 1610 interactions between features at different protocol layers, a 1611 side-effect of not allowing a choice of vantage point from which 1612 this information is observed. New approaches might have to be 1613 developed. 1615 o Impact on Research and Development: Hiding transport information 1616 can impede independent research into new mechanisms, measurement 1617 of behaviour, and development initiatives. Experience shows that 1618 transport protocols are complicated to design and complex to 1619 deploy, and that individual mechanisms have to be evaluated while 1620 considering other mechanisms, across a broad range of network 1621 topologies and with attention to the impact on traffic sharing the 1622 capacity. If increased use of transport header encryption results 1623 in reduced availability of open data, it could eliminate the 1624 independent self-checks to the standardisation process that have 1625 previously been in place from research and academic contributors 1626 (e.g., the role of the IRTF Internet Congestion Control Research 1627 Group (ICCRG) and research publications in reviewing new transport 1628 mechanisms and assessing the impact of their deployment). 1630 Observable transport information information might be useful to 1631 various stakeholders. Other stakeholders have incentives to limit 1632 what can be observed. This document does not make recommendations 1633 about what information ought to be exposed, to whom it ought to be 1634 observable, or how this will be achieved. There are also design 1635 choices about where observable fields are placed. For example, one 1636 location could be a part of the transport header outside of the 1637 encryption envelope, another alternative is to carry the information 1638 in a network-layer extension header. New transport protocol designs 1639 ought to explicitly identify any fields that are intended to be 1640 observed, consider if there are alternative ways of providing the 1641 information, and reflect on the implications of observable fields 1642 being used by in-network devices, and how this might impact user 1643 privacy and protocol evolution when these fields become ossified. 1645 As [RFC7258] notes, "Making networks unmanageable to mitigate 1646 (pervasive monitoring) is not an acceptable outcome, but ignoring 1647 (pervasive monitoring) would go against the consensus documented 1648 here." Providing explicit information can help avoid traffic being 1649 inappropriately classified, impacting application performance. An 1650 appropriate balance will emerge over time as real instances of this 1651 tension are analysed [RFC7258]. This balance between information 1652 exposed and information hidden ought to be carefully considered when 1653 specifying new transport protocols. 1655 8. Security Considerations 1657 This document is about design and deployment considerations for 1658 transport protocols. Issues relating to security are discussed 1659 throughout this document. 1661 Authentication, confidentiality protection, and integrity protection 1662 are identified as Transport Features by [RFC8095]. As currently 1663 deployed in the Internet, these features are generally provided by a 1664 protocol or layer on top of the transport protocol 1665 [I-D.ietf-taps-transport-security]. 1667 Confidentiality and strong integrity checks have properties that can 1668 also be incorporated into the design of a transport protocol. 1669 Integrity checks can protect an endpoint from undetected modification 1670 of protocol fields by network devices, whereas encryption and 1671 obfuscation or greasing can further prevent these headers being 1672 utilised by network devices. Preventing observation of headers 1673 provides an opportunity for greater freedom to update the protocols 1674 and can ease experimentation with new techniques and their final 1675 deployment in endpoints. A protocol specification needs to weigh the 1676 costs of ossifying common headers, versus the potential benefits of 1677 exposing specific information that could be observed along the 1678 network path to provide tools to manage new variants of protocols. 1680 A protocol design that uses header encryption can provide 1681 confidentiality of some or all of the protocol header information. 1682 This prevents an on-path device from knowledge of the header field. 1683 It therefore prevents mechanisms being built that directly rely on 1684 the information or seeks to infer semantics of an exposed header 1685 field. Reduces visibility into transport metadata can limit the 1686 ability to measure and characterise traffic. It can also provide 1687 privacy benefits in some cases. 1689 Extending the transport payload security context to also include the 1690 transport protocol header protects both information with the same 1691 key. A privacy concern would arise if this key was shared with a 1692 third party, e.g., providing access to transport header information 1693 to debug a performance issue, would also result in exposing the 1694 transport payload data to the same third party. A layered security 1695 design that separates network data from payload data would avoid such 1696 risks. 1698 Exposed transport headers are sometimes utilised as a part of the 1699 information to detect anomalies in network traffic. "While PM is an 1700 attack, other forms of monitoring that might fit the definition of PM 1701 can be beneficial and not part of any attack, e.g., network 1702 management functions monitor packets or flows and anti-spam 1703 mechanisms need to see mail message content." [RFC7258]. This can 1704 be used as the first line of defence to identify potential threats 1705 from DOS or malware and redirect suspect traffic to dedicated nodes 1706 responsible for DOS analysis, malware detection, or to perform packet 1707 "scrubbing" (the normalization of packets so that there are no 1708 ambiguities in interpretation by the ultimate destination of the 1709 packet). These techniques are currently used by some operators to 1710 also defend from distributed DOS attacks. 1712 Exposed transport header fields are sometimes also utilised as a part 1713 of the information used by the receiver of a transport protocol to 1714 protect the transport layer from data injection by an attacker. In 1715 evaluating this use of exposed header information, it is important to 1716 consider whether it introduces a significant DOS threat. For 1717 example, an attacker could construct a DOS attack by sending packets 1718 with a sequence number that falls within the currently accepted range 1719 of sequence numbers at the receiving endpoint, this would then 1720 introduce additional work at the receiving endpoint, even though the 1721 data in the attacking packet might not finally be delivered by the 1722 transport layer. This is sometimes known as a "shadowing attack". 1723 An attack can, for example, disrupt receiver processing, trigger loss 1724 and retransmission, or make a receiving endpoint perform unproductive 1725 decryption of packets that cannot be successfully decrypted (forcing 1726 a receiver to commit decryption resources, or to update and then 1727 restore protocol state). 1729 One mitigation to off-path attack is to deny knowledge of what header 1730 information is accepted by a receiver or obfuscate the accepted 1731 header information, e.g., setting a non-predictable initial value for 1732 a sequence number during a protocol handshake, as in [RFC3550] and 1733 [RFC6056], or a port value that can not be predicted (see section 5.1 1734 of [RFC8085]). A receiver could also require additional information 1735 to be used as a part of a validation check before accepting packets 1736 at the transport layer (e.g., utilising a part of the sequence number 1737 space that is encrypted; or by verifying an encrypted token not 1738 visible to an attacker). This would also mitigate against on-path 1739 attacks. An additional processing cost can be incurred when 1740 decryption has to be attempted before a receiver is able to discard 1741 injected packets. 1743 Open standards motivate a desire for this evaluation to include 1744 independent observation and evaluation of performance data, which in 1745 turn suggests control over where and when measurement samples are 1746 collected. This requires consideration of the appropriate balance 1747 between encrypting all and no transport information. Open data, and 1748 accessibility to tools that can help understand trends in application 1749 deployment, network traffic and usage patterns can all contribute to 1750 understanding security challenges. 1752 The Security and Privacy Considerations in the Framework for Large- 1753 Scale Measurement of Broadband Performance (LMAP) [RFC7594] contain 1754 considerations for Active and Passive measurement techniques and 1755 supporting material on measurement context. 1757 9. IANA Considerations 1759 XX RFC ED - PLEASE REMOVE THIS SECTION XXX 1761 This memo includes no request to IANA. 1763 10. Acknowledgements 1765 The authors would like to thank Mohamed Boucadair, Spencer Dawkins, 1766 Tom Herbert, Jana Iyengar, Mirja Kuehlewind, Kyle Rose, Kathleen 1767 Moriarty, Al Morton, Chris Seal, Joe Touch, Brian Trammell, Chris 1768 Wood, Thomas Fossati, and other members of the TSVWG for their 1769 comments and feedback. 1771 This work has received funding from the European Union's Horizon 2020 1772 research and innovation programme under grant agreement No 688421, 1773 and the EU Stand ICT Call 4. The opinions expressed and arguments 1774 employed reflect only the authors' view. The European Commission is 1775 not responsible for any use that might be made of that information. 1777 This work has received funding from the UK Engineering and Physical 1778 Sciences Research Council under grant EP/R04144X/1. 1780 11. Informative References 1782 [bufferbloat] 1783 Gettys, J. and K. Nichols, "Bufferbloat: dark buffers in 1784 the Internet. Communications of the ACM, 55(1):57-65", 1785 January 2012. 1787 [I-D.ietf-ippm-ioam-data] 1788 Brockners, F., Bhandari, S., Pignataro, C., Gredler, H., 1789 Leddy, J., Youell, S., Mizrahi, T., Mozes, D., Lapukhov, 1790 P., Chang, R., daniel.bernier@bell.ca, d., and J. Lemon, 1791 "Data Fields for In-situ OAM", draft-ietf-ippm-ioam- 1792 data-06 (work in progress), July 2019. 1794 [I-D.ietf-quic-transport] 1795 Iyengar, J. and M. Thomson, "QUIC: A UDP-Based Multiplexed 1796 and Secure Transport", draft-ietf-quic-transport-22 (work 1797 in progress), July 2019. 1799 [I-D.ietf-rtcweb-overview] 1800 Alvestrand, H., "Overview: Real Time Protocols for 1801 Browser-based Applications", draft-ietf-rtcweb-overview-19 1802 (work in progress), November 2017. 1804 [I-D.ietf-taps-transport-security] 1805 Wood, C., Enghardt, T., Pauly, T., Perkins, C., and K. 1806 Rose, "A Survey of Transport Security Protocols", draft- 1807 ietf-taps-transport-security-08 (work in progress), August 1808 2019. 1810 [I-D.ietf-tsvwg-rtcweb-qos] 1811 Jones, P., Dhesikan, S., Jennings, C., and D. Druta, "DSCP 1812 Packet Markings for WebRTC QoS", draft-ietf-tsvwg-rtcweb- 1813 qos-18 (work in progress), August 2016. 1815 [I-D.trammell-plus-abstract-mech] 1816 Trammell, B., "Abstract Mechanisms for a Cooperative Path 1817 Layer under Endpoint Control", draft-trammell-plus- 1818 abstract-mech-00 (work in progress), September 2016. 1820 [Latency] Briscoe, B., "Reducing Internet Latency: A Survey of 1821 Techniques and Their Merits, IEEE Comm. Surveys & 1822 Tutorials. 26;18(3) p2149-2196", November 2014. 1824 [Measure] Fairhurst, G., Kuehlewind, M., and D. Lopez, "Measurement- 1825 based Protocol Design, Eur. Conf. on Networks and 1826 Communications, Oulu, Finland.", June 2017. 1828 [Quic-Trace] 1829 "https:QUIC trace utilities //github.com/google/quic- 1830 trace". 1832 [RFC2474] Nichols, K., Blake, S., Baker, F., and D. Black, 1833 "Definition of the Differentiated Services Field (DS 1834 Field) in the IPv4 and IPv6 Headers", RFC 2474, 1835 DOI 10.17487/RFC2474, December 1998, 1836 . 1838 [RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z., 1839 and W. Weiss, "An Architecture for Differentiated 1840 Services", RFC 2475, DOI 10.17487/RFC2475, December 1998, 1841 . 1843 [RFC2507] Degermark, M., Nordgren, B., and S. Pink, "IP Header 1844 Compression", RFC 2507, DOI 10.17487/RFC2507, February 1845 1999, . 1847 [RFC2508] Casner, S. and V. Jacobson, "Compressing IP/UDP/RTP 1848 Headers for Low-Speed Serial Links", RFC 2508, 1849 DOI 10.17487/RFC2508, February 1999, 1850 . 1852 [RFC2914] Floyd, S., "Congestion Control Principles", BCP 41, 1853 RFC 2914, DOI 10.17487/RFC2914, September 2000, 1854 . 1856 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 1857 of Explicit Congestion Notification (ECN) to IP", 1858 RFC 3168, DOI 10.17487/RFC3168, September 2001, 1859 . 1861 [RFC3234] Carpenter, B. and S. Brim, "Middleboxes: Taxonomy and 1862 Issues", RFC 3234, DOI 10.17487/RFC3234, February 2002, 1863 . 1865 [RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, 1866 A., Peterson, J., Sparks, R., Handley, M., and E. 1867 Schooler, "SIP: Session Initiation Protocol", RFC 3261, 1868 DOI 10.17487/RFC3261, June 2002, 1869 . 1871 [RFC3393] Demichelis, C. and P. Chimento, "IP Packet Delay Variation 1872 Metric for IP Performance Metrics (IPPM)", RFC 3393, 1873 DOI 10.17487/RFC3393, November 2002, 1874 . 1876 [RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V. 1877 Jacobson, "RTP: A Transport Protocol for Real-Time 1878 Applications", STD 64, RFC 3550, DOI 10.17487/RFC3550, 1879 July 2003, . 1881 [RFC3711] Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K. 1882 Norrman, "The Secure Real-time Transport Protocol (SRTP)", 1883 RFC 3711, DOI 10.17487/RFC3711, March 2004, 1884 . 1886 [RFC4302] Kent, S., "IP Authentication Header", RFC 4302, 1887 DOI 10.17487/RFC4302, December 2005, 1888 . 1890 [RFC4303] Kent, S., "IP Encapsulating Security Payload (ESP)", 1891 RFC 4303, DOI 10.17487/RFC4303, December 2005, 1892 . 1894 [RFC4585] Ott, J., Wenger, S., Sato, N., Burmeister, C., and J. Rey, 1895 "Extended RTP Profile for Real-time Transport Control 1896 Protocol (RTCP)-Based Feedback (RTP/AVPF)", RFC 4585, 1897 DOI 10.17487/RFC4585, July 2006, 1898 . 1900 [RFC4737] Morton, A., Ciavattone, L., Ramachandran, G., Shalunov, 1901 S., and J. Perser, "Packet Reordering Metrics", RFC 4737, 1902 DOI 10.17487/RFC4737, November 2006, 1903 . 1905 [RFC4995] Jonsson, L-E., Pelletier, G., and K. Sandlund, "The RObust 1906 Header Compression (ROHC) Framework", RFC 4995, 1907 DOI 10.17487/RFC4995, July 2007, 1908 . 1910 [RFC5218] Thaler, D. and B. Aboba, "What Makes for a Successful 1911 Protocol?", RFC 5218, DOI 10.17487/RFC5218, July 2008, 1912 . 1914 [RFC5236] Jayasumana, A., Piratla, N., Banka, T., Bare, A., and R. 1915 Whitner, "Improved Packet Reordering Metrics", RFC 5236, 1916 DOI 10.17487/RFC5236, June 2008, 1917 . 1919 [RFC5481] Morton, A. and B. Claise, "Packet Delay Variation 1920 Applicability Statement", RFC 5481, DOI 10.17487/RFC5481, 1921 March 2009, . 1923 [RFC5925] Touch, J., Mankin, A., and R. Bonica, "The TCP 1924 Authentication Option", RFC 5925, DOI 10.17487/RFC5925, 1925 June 2010, . 1927 [RFC6056] Larsen, M. and F. Gont, "Recommendations for Transport- 1928 Protocol Port Randomization", BCP 156, RFC 6056, 1929 DOI 10.17487/RFC6056, January 2011, 1930 . 1932 [RFC6269] Ford, M., Ed., Boucadair, M., Durand, A., Levis, P., and 1933 P. Roberts, "Issues with IP Address Sharing", RFC 6269, 1934 DOI 10.17487/RFC6269, June 2011, 1935 . 1937 [RFC6294] Hu, Q. and B. Carpenter, "Survey of Proposed Use Cases for 1938 the IPv6 Flow Label", RFC 6294, DOI 10.17487/RFC6294, June 1939 2011, . 1941 [RFC6347] Rescorla, E. and N. Modadugu, "Datagram Transport Layer 1942 Security Version 1.2", RFC 6347, DOI 10.17487/RFC6347, 1943 January 2012, . 1945 [RFC6437] Amante, S., Carpenter, B., Jiang, S., and J. Rajahalme, 1946 "IPv6 Flow Label Specification", RFC 6437, 1947 DOI 10.17487/RFC6437, November 2011, 1948 . 1950 [RFC6438] Carpenter, B. and S. Amante, "Using the IPv6 Flow Label 1951 for Equal Cost Multipath Routing and Link Aggregation in 1952 Tunnels", RFC 6438, DOI 10.17487/RFC6438, November 2011, 1953 . 1955 [RFC7258] Farrell, S. and H. Tschofenig, "Pervasive Monitoring Is an 1956 Attack", BCP 188, RFC 7258, DOI 10.17487/RFC7258, May 1957 2014, . 1959 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 1960 Recommendations Regarding Active Queue Management", 1961 BCP 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 1962 . 1964 [RFC7594] Eardley, P., Morton, A., Bagnulo, M., Burbridge, T., 1965 Aitken, P., and A. Akhter, "A Framework for Large-Scale 1966 Measurement of Broadband Performance (LMAP)", RFC 7594, 1967 DOI 10.17487/RFC7594, September 2015, 1968 . 1970 [RFC7624] Barnes, R., Schneier, B., Jennings, C., Hardie, T., 1971 Trammell, B., Huitema, C., and D. Borkmann, 1972 "Confidentiality in the Face of Pervasive Surveillance: A 1973 Threat Model and Problem Statement", RFC 7624, 1974 DOI 10.17487/RFC7624, August 2015, 1975 . 1977 [RFC7799] Morton, A., "Active and Passive Metrics and Methods (with 1978 Hybrid Types In-Between)", RFC 7799, DOI 10.17487/RFC7799, 1979 May 2016, . 1981 [RFC7872] Gont, F., Linkova, J., Chown, T., and W. Liu, 1982 "Observations on the Dropping of Packets with IPv6 1983 Extension Headers in the Real World", RFC 7872, 1984 DOI 10.17487/RFC7872, June 2016, 1985 . 1987 [RFC7928] Kuhn, N., Ed., Natarajan, P., Ed., Khademi, N., Ed., and 1988 D. Ros, "Characterization Guidelines for Active Queue 1989 Management (AQM)", RFC 7928, DOI 10.17487/RFC7928, July 1990 2016, . 1992 [RFC7983] Petit-Huguenin, M. and G. Salgueiro, "Multiplexing Scheme 1993 Updates for Secure Real-time Transport Protocol (SRTP) 1994 Extension for Datagram Transport Layer Security (DTLS)", 1995 RFC 7983, DOI 10.17487/RFC7983, September 2016, 1996 . 1998 [RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White, 1999 "Proportional Integral Controller Enhanced (PIE): A 2000 Lightweight Control Scheme to Address the Bufferbloat 2001 Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017, 2002 . 2004 [RFC8084] Fairhurst, G., "Network Transport Circuit Breakers", 2005 BCP 208, RFC 8084, DOI 10.17487/RFC8084, March 2017, 2006 . 2008 [RFC8085] Eggert, L., Fairhurst, G., and G. Shepherd, "UDP Usage 2009 Guidelines", BCP 145, RFC 8085, DOI 10.17487/RFC8085, 2010 March 2017, . 2012 [RFC8086] Yong, L., Ed., Crabbe, E., Xu, X., and T. Herbert, "GRE- 2013 in-UDP Encapsulation", RFC 8086, DOI 10.17487/RFC8086, 2014 March 2017, . 2016 [RFC8087] Fairhurst, G. and M. Welzl, "The Benefits of Using 2017 Explicit Congestion Notification (ECN)", RFC 8087, 2018 DOI 10.17487/RFC8087, March 2017, 2019 . 2021 [RFC8095] Fairhurst, G., Ed., Trammell, B., Ed., and M. Kuehlewind, 2022 Ed., "Services Provided by IETF Transport Protocols and 2023 Congestion Control Mechanisms", RFC 8095, 2024 DOI 10.17487/RFC8095, March 2017, 2025 . 2027 [RFC8250] Elkins, N., Hamilton, R., and M. Ackermann, "IPv6 2028 Performance and Diagnostic Metrics (PDM) Destination 2029 Option", RFC 8250, DOI 10.17487/RFC8250, September 2017, 2030 . 2032 [RFC8289] Nichols, K., Jacobson, V., McGregor, A., Ed., and J. 2033 Iyengar, Ed., "Controlled Delay Active Queue Management", 2034 RFC 8289, DOI 10.17487/RFC8289, January 2018, 2035 . 2037 [RFC8290] Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, 2038 J., and E. Dumazet, "The Flow Queue CoDel Packet Scheduler 2039 and Active Queue Management Algorithm", RFC 8290, 2040 DOI 10.17487/RFC8290, January 2018, 2041 . 2043 [RFC8404] Moriarty, K., Ed. and A. Morton, Ed., "Effects of 2044 Pervasive Encryption on Operators", RFC 8404, 2045 DOI 10.17487/RFC8404, July 2018, 2046 . 2048 [RFC8446] Rescorla, E., "The Transport Layer Security (TLS) Protocol 2049 Version 1.3", RFC 8446, DOI 10.17487/RFC8446, August 2018, 2050 . 2052 [RFC8546] Trammell, B. and M. Kuehlewind, "The Wire Image of a 2053 Network Protocol", RFC 8546, DOI 10.17487/RFC8546, April 2054 2019, . 2056 [RFC8548] Bittau, A., Giffin, D., Handley, M., Mazieres, D., Slack, 2057 Q., and E. Smith, "Cryptographic Protection of TCP Streams 2058 (tcpcrypt)", RFC 8548, DOI 10.17487/RFC8548, May 2019, 2059 . 2061 [RFC8558] Hardie, T., Ed., "Transport Protocol Path Signals", 2062 RFC 8558, DOI 10.17487/RFC8558, April 2019, 2063 . 2065 Appendix A. Revision information 2067 -00 This is an individual draft for the IETF community. 2069 -01 This draft was a result of walking away from the text for a few 2070 days and then reorganising the content. 2072 -02 This draft fixes textual errors. 2074 -03 This draft follows feedback from people reading this draft. 2076 -04 This adds an additional contributor and includes significant 2077 reworking to ready this for review by the wider IETF community Colin 2078 Perkins joined the author list. 2080 Comments from the community are welcome on the text and 2081 recommendations. 2083 -05 Corrections received and helpful inputs from Mohamed Boucadair. 2085 -06 Updated following comments from Stephen Farrell, and feedback via 2086 email. Added a draft conclusion section to sketch some strawman 2087 scenarios that could emerge. 2089 -07 Updated following comments from Al Morton, Chris Seal, and other 2090 feedback via email. 2092 -08 Updated to address comments sent to the TSVWG mailing list by 2093 Kathleen Moriarty (on 08/05/2018 and 17/05/2018), Joe Touch on 2094 11/05/2018, and Spencer Dawkins. 2096 -09 Updated security considerations. 2098 -10 Updated references, split the Introduction, and added a paragraph 2099 giving some examples of why ossification has been an issue. 2101 -01 This resolved some reference issues. Updated section on 2102 observation by devices on the path. 2104 -02 Comments received from Kyle Rose, Spencer Dawkins and Tom 2105 Herbert. The network-layer information has also been re-organised 2106 after comments at IETF-103. 2108 -03 Added a section on header compression and rewriting of sections 2109 referring to RTP transport. This version contains author editorial 2110 work and removed duplicate section. 2112 -04 Revised following SecDir Review 2113 o Added some text on TLS story (additional input sought on relevant 2114 considerations). 2116 o Section 2, paragraph 8 - changed to be clearer, in particular, 2117 added "Encryption with secure key distribution prevents" 2119 o Flow label description rewritten based on PS/BCP RFCs. 2121 o Clarify requirements from RFCs concerning the IPv6 flow label and 2122 highlight ways it can be used with encryption. (section 3.1.3) 2124 o Add text on the explicit spin-bit work in the QUIC DT. Added 2125 greasing of spin-bit. (Section 6.1) 2127 o Updated section 6 and added more explanation of impact on 2128 operators. 2130 o Other comments addressed. 2132 -05 Editorial pass and minor corrections noted on TSVWG list. 2134 -06 Updated conclusions and minor corrections. Responded to request 2135 to add OAM discussion to Section 6.1. 2137 -07 Addressed feedback from Ruediger and Thomas. 2139 Section 2 deserved some work to make it easier to read and avoid 2140 repetition. This edit finally gets to this, and eliminates some 2141 duplication. This also moves some of the material from section 2 to 2142 reform a clearer conclusion. The scope remains focussed on the usage 2143 of transport headers and the implications of encryption - not on 2144 proposals for new techniques/specifications to be developed. 2146 -08 Addressed feedback and completed editorial work, including 2147 updating the text referring to RFC7872, in preparation for a WGLC. 2149 -09 Updated following WGLC. In particular, thanks to Joe Touch 2150 (specific comments and commentary on style and tone); Dimitri Tikonov 2151 (editorial); Christian Huitema (various); David Black (various). 2152 Amended privacy considerations based on SECDIR review. Emile Stephan 2153 (inputs on operations measurement); Various others. 2155 Added summary text and refs to key sections. Note to editors: The 2156 section numbers are hard-linked. 2158 -10 Updated following additional feedback from 1st WGLC. Comments 2159 from David Black; Tommy Pauly; Ian Swett; Mirja Kuehlewind; Peter 2160 Gutmann; Ekr; and many others via the TSVWG list. Some people 2161 thought that "needed" and "need" could represent requirements in the 2162 document, etc. this has been clarified. 2164 Authors' Addresses 2166 Godred Fairhurst 2167 University of Aberdeen 2168 Department of Engineering 2169 Fraser Noble Building 2170 Aberdeen AB24 3UE 2171 Scotland 2173 EMail: gorry@erg.abdn.ac.uk 2174 URI: http://www.erg.abdn.ac.uk/ 2176 Colin Perkins 2177 University of Glasgow 2178 School of Computing Science 2179 Glasgow G12 8QQ 2180 Scotland 2182 EMail: csp@csperkins.org 2183 URI: https://csperkins.org/