Unfortunate History of Transient Numeric Identifiers

Issues with the Specification of Transient Numeric Identifiers While assessing IETF protocol specifications regarding the use of transient numeric identifiers, we have found that most of the issues discussed in this document arise as a result of one of the following conditions:

protocol specifications that under specify their transient numeric identifiers
protocol specifications that over specify their transient numeric identifiers
protocol implementations that simply fail to comply with the specified requirements

A number of IETF protocol specifications under specified their transient numeric identifiers, thus leading to implementations that were vulnerable to numerous off-path attacks. Examples of them are the specification of TCP local ports in or the specification of the DNS ID in . On the other hand, there are a number of IETF protocol specifications that over specify some of their associated transient numeric identifiers. For example, essentially overloads the semantics of IPv6 Interface Identifiers (IIDs) by embedding link-layer addresses in the IPv6 IIDs when the interoperability requirement of uniqueness could be achieved in other ways that do not result in negative security and privacy implications . Similarly, suggests the use of a global counter for the generation of Identification values when the interoperability requirement of uniqueness per {IPv6 Source Address, IPv6 Destination Address} could be achieved with other algorithms that do not result in negative security and privacy implications . Finally, there are protocol implementations that simply fail to comply with existing protocol specifications. For example, some popular operating systems still fail to implement transport-protocol ephemeral port randomization, as recommended in , or TCP Initial Sequence Number randomization, as recommended in . The following subsections document the timelines for a number of sample transient numeric identifiers that illustrate how the problem discussed in this document has affected protocols from different layers over time. These sample transient numeric identifiers have different interoperability requirements and failure severities (see ), and thus are considered to be representative of the problem being analyzed in this document.

IPv4/IPv6 Identification This section presents the timeline of the Identification field employed by IPv4 (in the base header) and IPv6 (in Fragment Headers). The reason for presenting both cases in the same section is to make it evident that, while the Identification value serves the same purpose in both protocols, the work and research done for the IPv4 case did not influence IPv6 specifications or implementations. The IPv4 Identification is specified in , which specifies the interoperability requirements for the Identification field, i.e., the sender must choose the Identification field to be unique for a given {Source Address, Destination Address, Protocol} for the time the datagram (or any fragment of it) could be alive in the Internet. It suggests that a sending protocol module may keep "a table of Identifiers, one entry for each destination it has communicated with in the last maximum packet lifetime for the [I]nternet", and it also suggests that "since the Identifier field allows 65,536 different values, hosts may be able to simply use unique identifiers independent of destination". The above has been interpreted numerous times as a suggestion to employ per-destination or global counters for the generation of Identification values. While does not suggest any flawed algorithm for the generation of Identification values, the specification omits a discussion of the security and privacy implications of predictable Identification values. This resulted in many IPv4 implementations generating predictable Identification values by means of a global counter, at least at some point in time. The IPv6 Identification was originally specified in . It serves the same purpose as its IPv4 counterpart, but rather than being part of the base header (as in the IPv4 case), it is part of the Fragment Header (which may or may not be present in an IPv6 packet). states that the Identification must be different than that of any other fragmented packet sent recently (within the maximum likely lifetime of a packet) with the same Source Address and Destination Address. Subsequently, it notes that this requirement can be met by means of a wrap-around 32-bit counter that is incremented each time a packet must be fragmented and that it is an implementation choice whether to use a global or a per-destination counter. Thus, the specification of the IPv6 Identification is similar to that of the IPv4 case, with the only difference that, in the IPv6 case, the suggestions to use simple counters is more explicit. is the first revision of the core IPv6 specification and maintains the same text for the specification of the IPv6 Identification field. , the second revision of the core IPv6 specification, removes the suggestion from to use a counter for the generation of IPv6 Identification values and points to for sample algorithms for their generation.

September 1981:: specifies the interoperability requirements for the IPv4 Identification but does not perform a vulnerability assessment of this transient numeric identifier.
December 1995:: , the first specification of the IPv6 protocol, is published. It suggests that a counter be used to generate the IPv6 Identification values and notes that it is an implementation choice whether to maintain a single counter for the node or multiple counters (e.g., one for each of the node's possible Source Addresses, or one for each active {Source Address, Destination Address} set).
December 1998:: finds that predictable IPv4 Identification values (as generated by most popular implementations) can be leveraged to count the number of packets sent by a target node. explains how to leverage the same vulnerability to implement a port-scanning technique known as "idle scan". A tool that implements this attack is publicly released.
December 1998:: , a revision of the IPv6 specification, is published, obsoleting . It maintains the same specification of the IPv6 Identification field as its predecessor .
December 1998:: OpenBSD implements randomization of the IPv4 Identification field .
November 1999:: discusses how to leverage predictable IPv4 Identification values to uncover the rules of a number of firewalls.
September 2002:: documents the implementation of the "idle scan" technique in the popular Network Mapper (nmap) tool.
November 2002:: explains how the IPv4 Identification field can be exploited to count the number of systems behind a NAT.
October 2003:: OpenBSD implements randomization of the IPv6 Identification field .
December 2003:: explains a technique to perform TCP data injection attacks based on predictable IPv4 Identification values, which requires less effort than TCP injection attacks performed with bare TCP packets.
January 2005:: discusses shortcomings in a number of techniques to mitigate predictable IPv4 Identification values.
October 2007:: describes a weakness in the pseudorandom number generator (PRNG) in use for the generation of IP Identification values by a number of operating systems.
June 2011:: describes how to perform idle scan attacks in IPv6.
November 2011:: Linux mitigates predictable IPv6 Identification values .
December 2011:: describes the security implications of predictable IPv6 Identification values and possible mitigations. This document has the intended status of "Standards Track", with the intention to formally update to introduce security and privacy requirements on the generation of IPv6 Identification values.
May 2012:: notes that some major IPv6 implementations still employ predictable IPv6 Identification values.
March 2013:: The 6man WG adopts but changes the track to "BCP" (while still formally updating ), posting the resulting document as .
June 2013:: A patch to incorporate support for IPv6-based idle scans in nmap is submitted .
December 2014:: The 6man WG changes the intended status of to "Informational" and posts it as . As a result, it no longer formally updates , and security and privacy requirements on the generation of IPv6 Identification values are eliminated.
June 2015:: notes that some popular host and router implementations still employ predictable IPv6 Identification values.
February 2016:: (based on ) analyzes the security and privacy implications of predictable IPv6 Identification values and provides guidance for selecting an algorithm to generate such values. However, being published as an "Informational" RFC, it does not formally update and does not introduce security and privacy requirements on the generation of IPv6 Identification values.
June 2016:: , a draft revision of , removes the suggestion from to use a counter for the generation of IPv6 Identification values but does not perform a vulnerability assessment of the generation of IPv6 Identification values and does not introduce security and privacy requirements on the generation of IPv6 Identification values.
July 2017:: is finally published as , obsoleting and pointing to for sample algorithms for the generation of IPv6 Identification values. However, it does not introduce security and privacy requirements on the generation of IPv6 Identification values.
October 2019:: notes that the IPv6 Identification generators of two popular operating systems are flawed.

TCP Initial Sequence Numbers (ISNs) suggests that the choice of the ISN of a connection is not arbitrary but aims to reduce the chances of a stale segment from being accepted by a new incarnation of a previous connection. suggests the use of a global 32-bit ISN generator that is incremented by 1 roughly every 4 microseconds. However, as a matter of fact, protection against stale segments from a previous incarnation of the connection is enforced by preventing the creation of a new incarnation of a previous connection before 2*MSL has passed since a segment corresponding to the old incarnation was last seen (where "MSL" is the "Maximum Segment Lifetime" ). This is accomplished by the TIME-WAIT state and TCP's "quiet time" concept (see ). Based on the assumption that ISNs are monotonically increasing across connections, many stacks (e.g., 4.2BSD-derived) use the ISN of an incoming SYN segment to perform "heuristics" that enable the creation of a new incarnation of a connection while the previous incarnation is still in the TIME-WAIT state (see p. 945 of ). This avoids an interoperability problem that may arise when a node establishes connections to a specific TCP end-point at a high rate . The interoperability requirements for TCP ISNs are probably not as clearly spelled out as one would expect. Furthermore, the suggestion of employing a global counter in negatively affects the security and privacy properties of the protocol.

September 1981:: suggests the use of a global 32-bit ISN generator, whose lower bit is incremented roughly every 4 microseconds. However, such an ISN generator makes it trivial to predict the ISN that a TCP implementation will use for new connections, thus allowing a variety of attacks against TCP.
February 1985:: is the first to describe how to exploit predictable TCP ISNs for forging TCP connections that could then be leveraged for trust relationship exploitation.
April 1989:: discusses the security considerations for predictable ISNs (along with a range of other protocol-based vulnerabilities).
January 1995:: reports a real-world exploitation of the vulnerability described in ten years before (in 1985).
May 1996:: is the first IETF effort, authored by Steven Bellovin, to address predictable TCP ISNs. However, does not formally update . Note: The same concept specified in this document for TCP ISNs was later proposed for TCP ephemeral ports , TCP Timestamps, and eventually even IPv6 Interface Identifiers .
July 1996:: OpenBSD implements TCP ISN randomization based on random increments (please see ) .
December 2000:: OpenBSD implements TCP ISN randomization using simple randomization (please see ) .
March 2001:: provides a detailed analysis of statistical weaknesses in some TCP ISN generators and includes a survey of the algorithms in use by popular TCP implementations. Vulnerability advisories were released regarding statistical weaknesses in some TCP ISN generators, affecting popular TCP implementations. Other vulnerability advisories on the same vulnerability, such as , were published later on.
March 2002:: updates and complements . It concludes that "while some vendors [...] reacted promptly and tested their solutions properly, many still either ignored the issue and never evaluated their implementations, or implemented a flawed solution that apparently was not tested using a known approach" .
June 2007:: OpenBSD implements TCP ISN randomization based on the algorithm specified in (currently obsoleted and replaced by ) for the TCP endpoint that performs the active open while keeping the simple randomization scheme for the endpoint performing the passive open . This provides monotonically increasing ISNs for the "client side" (allowing the BSD heuristics to work as expected) while avoiding any patterns in the ISN generation for the "server side".
February 2012:: , published 27 years after Morris's original work , formally updates to mitigate predictable TCP ISNs.
August 2014:: The algorithm specified in becomes the recommended ("SHOULD") algorithm for TCP ISN generation in , an early revision of the core TCP specification .
August 2022:: , a revision of the core TCP specification, is published, adopting the algorithm specified in as the recommended ("SHOULD") algorithm for TCP ISN generation.

IPv6 Interface Identifiers (IIDs) IPv6 Interface Identifiers can be generated as a result of different mechanisms, including Stateless Address Autoconfiguration (SLAAC) , DHCPv6 , and manual configuration. This section focuses on Interface Identifiers resulting from SLAAC. The Interface Identifier of stable IPv6 addresses resulting from SLAAC originally resulted in the underlying link-layer address being embedded in the IID. At the time, employing the underlying link-layer address for the IID was seen as a convenient way to obtain a unique address. However, recent awareness about the security and privacy properties of this approach has led to the replacement of this flawed scheme with an alternative one that does not negatively affect the security and privacy properties of the protocol.

January 1997:: specifies the syntax of IPv6 global addresses (referred to as "An IPv6 Provider-Based Unicast Address Format" at the time), which is consistent with the IPv6 addressing architecture specified in . Hosts are recommended to "generate addresses using link-specific addresses as Interface ID such as 48 bit IEEE-802 MAC addresses".
July 1998:: specifies "An IPv6 Aggregatable Global Unicast Address Format" (obsoleting ), changing the size of the IID to 64 bits, and specifies that IIDs must be constructed in IEEE 64-bit Extended Unique Identifier (EUI-64) format. How such identifiers are constructed is specified in the corresponding "IPv6 over <link>" specifications, such as "IPv6 over Ethernet".
January 2001:: recognizes the problem of IPv6 network activity correlation and specifies IPv6 temporary addresses. Temporary addresses are to be used along with stable addresses.
August 2003:: obsoletes , making the Top-Level Aggregator (TLA) / Next-Level Aggregator (NLA) structure historic, though the syntax and recommendations for the stable IIDs remain unchanged.
February 2006:: is published as the latest "IP Version 6 Addressing Architecture", requiring the IIDs of "all unicast addresses, except those that start with the binary value 000" to employ the Modified EUI-64 format. The details of constructing such interface identifiers are defined in the corresponding "IPv6 over <link>" specifications.
March 2008:: provides hints regarding how patterns in IPv6 addresses could be leveraged for the purpose of address scanning.
December 2011:: notes that the original scheme for generating stable addresses allows for IPv6 address scanning and for active host tracking (even when IPv6 temporary addresses are employed). It also specifies an alternative algorithm meant to replace IIDs based on Modified EUI-64 format identifiers.
November 2012:: The 6man WG adopts as a working group item (as ). However, the document no longer formally updates ; therefore, the specified algorithm no longer formally replaces the Modified EUI-64 format identifiers.
February 2013:: An address-scanning tool (scan6 of ) that leverages IPv6 address patterns is released .
July 2013:: elaborates on the security and privacy properties of all known algorithms for generating IPv6 IIDs.
January 2014:: The 6man WG posts ("Recommendation on Stable IPv6 Interface Identifiers"), recommending for the generation of stable addresses.
April 2014:: (formerly ) is published, specifying "A Method for Generating Semantically Opaque Interface Identifiers with IPv6 Stateless Address Autoconfiguration (SLAAC)" as an alternative to (but not replacement of) Modified EUI-64 format IIDs.
March 2016:: (formerly and later ), about "Network Reconnaissance in IPv6 Networks", is published.
March 2016:: (formerly and later ), about "Security and Privacy Considerations for IPv6 Address Generation Mechanisms", is published.
May 2016:: is posted, with the goal of specifying requirements for non-stable addresses and updating such that use of only temporary addresses is allowed.
May 2016:: is posted, providing an analysis of how different aspects on an address (from stability to usage mode) affect their corresponding security and privacy properties and meaning to eventually provide advice in this area.
February 2017:: , produced by the 6man WG, is published as ("Recommendation on Stable IPv6 Interface Identifiers"), with requirements for stable addresses and a recommendation to employ for the generation of stable addresses. It formally updates a large number of RFCs.
March 2018:: is posted (as suggested by the 6man WG) to address flaws in by revising it (as an alternative to the effort, posted in March 2016).
July 2018:: is adopted (as ) as a WG item of the 6man WG.
December 2020:: is approved by the IESG for publication as an RFC.
February 2021:: is finally published as .

NTP Reference IDs (REFIDs) The NTP Reference ID is a 32-bit code identifying the particular server or reference clock. Above stratum 1 (secondary servers and clients), this value can be employed to avoid degree-one timing loops, that is, scenarios where two NTP peers are (mutually) the time source of each other. If using the IPv4 address family, the identifier is the four-octet IPv4 address. If using the IPv6 address family, it is the first four octets of the MD5 hash of the IPv6 address.

June 2010:: ("Network Time Protocol Version 4: Protocol and Algorithms Specification") is published. It specifies that, for NTP peers with stratum higher than 1, the REFID embeds the IPv4 address of the time source or the first four octets of the MD5 hash of the IPv6 address of the time source.
July 2016:: is posted, describing the information leakage produced via the NTP REFID. It proposes that NTP returns a special REFID when a packet employs an IP Source Address that is not believed to be a current NTP peer but otherwise generates and returns the common REFID. It is subsequently adopted by the NTP WG as .
April 2019:: notes that the proposed fix specified in is, at the very least, sub-optimal. As a result of a lack of WG support, the effort is eventually abandoned.

Transport-Protocol Ephemeral Port Numbers Most (if not all) transport protocols employ "port numbers" to demultiplex packets to the corresponding transport-protocol instances. "Ephemeral ports" refer to the local ports employed in active OPEN requests, that is, typically the local port numbers employed on the side initiating the communication.

August 1980:: notes that the UDP source port is optional and identifies the port of the sending process. It does not specify interoperability requirements for source port selection, nor does it suggest possible ways to select port numbers. Most popular implementations end up selecting source ports from a system-wide global counter.
September 1981:: (the TCP specification) essentially describes the use of port numbers and specifies that port numbers should result in a unique socket pair {local address, local port, remote address, remote port}. How ephemeral ports are selected and the port range from which they are selected are left unspecified.
July 1996:: OpenBSD implements ephemeral port randomization .
July 2008:: The CERT Coordination Center publishes details of what became known as the "Kaminsky Attack" on the DNS. The attack exploits the lack of ephemeral port randomization and DNS ID randomization in many major DNS implementations to perform cache poisoning in an effective and practical manner.
January 2009:: mandates the use of port randomization for DNS resolvers and mandates that implementations must randomize ports from the range of available ports (53 or 1024 and above) that is as large as possible and practicable. It does not recommend possible algorithms for port randomization, although the document specifically targets DNS resolvers, for which a simple port randomization suffices (e.g., Algorithm 1 of ). This document led to the implementation of port randomization in the DNS resolvers themselves, rather than in the underlying transport protocols.
January 2011:: notes that many TCP and UDP implementations result in predictable ephemeral port numbers and also notes that many implementations select port numbers from a small portion of the whole port number space. It recommends the implementation and use of ephemeral port randomization, proposes a number of possible algorithms for port randomization, and also recommends to randomize port numbers over the range 1024-65535.
March 2016:: reports a non-normal distribution of the ephemeral port numbers employed by the NTP clients of an Internet Time Service.
April 2019:: notes that some NTP implementations employ the NTP service port (123) as the local port for nonsymmetric modes and aims to update the NTP specification to recommend port randomization in such cases, which is in line with . The proposal experiences some pushback in the relevant working group (NTP WG) but is finally adopted as a working group item as .
August 2021:: is finally published as .

DNS ID The DNS ID can be employed to match DNS replies to outstanding DNS queries.

November 1987:: specifies that the DNS ID is a 16-bit identifier assigned by the program that generates any kind of query and that this identifier is copied in the corresponding reply and can be used by the requester to match up replies to outstanding queries. It does not specify the interoperability requirements for this numeric identifier, nor does it suggest an algorithm for generating it.
August 1993:: describes DNS cache poisoning attacks that require the attacker to guess the DNS ID.
June 1995:: suggests that both the UDP source port and the DNS ID of query packets should be randomized, although that might not provide enough entropy to prevent an attacker from guessing these values.
April 1997:: finds that implementations employ predictable UDP source ports and predictable DNS IDs and argues that both should be randomized.
November 2002:: finds that, by spoofing multiple requests for the same domain name from different IP addresses, an attacker may guess the DNS ID employed for a victim with a high probability of success, thus allowing for DNS cache poisoning attacks.
March 2007:: finds that the Microsoft Windows DNS server generates predictable DNS ID values.
July 2007:: finds that a popular DNS server software (BIND 9) that randomizes the DNS ID is still subject to DNS cache poisoning attacks by forging a large number of queries and leveraging the birthday paradox.
October 2007:: finds that OpenBSD's DNS software (based on the BIND DNS server of the Internet Systems Consortium (ISC)) generates predictable DNS ID values.
January 2009:: is published, requiring resolvers to randomize the DNS ID of queries and to verify that the DNS ID of a reply matches that of the DNS query as part of the DNS reply validation process.
May 2010:: finds that the Windows SMTP Service implements its own DNS resolver that results in predictable DNS ID values. Additionally, it fails to validate that the DNS ID of a reply matches that of the DNS query that supposedly elicited it.