Unintended Operational Issues With ULA

There are demonstrated and easily repeatible uses cases of ULA not being preferred in some OS and network equipment over legacy IPv4 that necessitate the immediate update to to better reflect the original intent of the RFC. As with most adjustments to standards, and using itself as a measurment, this update will likely take between 8-20 years to become common enough for relatively consistent behavior within operating systems. As a reference, as of the time of this writing, it has been 10 years since has been published but we continue to see existing commercial and open source operating systems exhibiting behavior. While it should be noted that defines a solution that is functional academically, operationally the solution of adjusting the address preference selection table is both operating system dependent and unable to be signalled by any network mechanism such as within a router advertisement, DHCPv6 option, or the like. This lack of an intra-protocol or network-based ability to adjust address selection preference, along with the inability to adjust a notable number of operating systems either programmatically or manually renders operational scalability of such a mechanism functionally untenable. It is anticipated that any update of would require an additional 8-20 years to be fully realized and properly implemented in a majority of network connected systems. In addition, in the current versions of Linux, the priority table (gai.conf) still makes reference to , further demonstrating the long timeframe to have updates reflected in a current, modern operating system. Examples of such out-of-date behavior can be found in printers, cameras, fixed devices, IoT sensors, and longer lifecycle equipment. It is especially important to note this behavior in the long lifecycle equipment that exists in industrial control and operational techology environments due to their very long mean time to replacement. The core issue is the stated interpretation from gai.conf that has the following default:

# Add another rule to the RFC 6724 scope table for IPv4 addresses. # By default the scope IDs described in section 3.2 in RFC 6724 are # used. Changing these defaults should hardly ever be necessary. # The defaults are equivalent to: # #scopev4 ::ffff:169.254.0.0/112 2 #scopev4 ::ffff:127.0.0.0/104 2 #scopev4 ::ffff:0.0.0.0/96 14 ]]> Notice that they are interpreting the legacy IPv4 address range as "scopev4" and the prefix ::ffff:0.0.0.0/96 which has a higher precedence (35) in then the ULA prefix of fc00::/7 (3). This results in legacy IPv4 being preferred over IPv6 ULA. The operational outcome is the move to dual-stack with ULA is inconsistent and imparts unnecessary difficulty for both troubleshooting and creating the baseline expected behavior which are both requirements for deployments. This results in operational and engineering teams not gaining IPv6 experience as limited traffic is actually using IPv6, and security baseline expectations are inconsistent at best and haphazard at worst. In practice, imposes several operational shortcomings preventing both consistent and desired behavior. If we define "desired behavior" as IPv6 preference over legacy IPv4 for address and protocol selection, then the resulting implemented behavior, based on , will fall short of that intent. Based on the current verbiage, dual-stacked hosts configured with both a legacy IPv4 address and an IPv6 ULA address, the resulting behavior will manifest as a host choosing IPv4 over ULA IPv6. This behavior deviates from the current goal of a host with legacy IPv4 address and also with an IPv6 GUA address preferring IPv6 over IPv4. Operationally and strategically, this manifests as an impediment to deployment of IPv6 for many non-service provider and mobile networks phasing in dual-stacked (both legacy IPv4 and IPv6) networking with the expectation of consistent behavior (alway use IPv6 before legacy IPv4). Other operational considerations are the use of the policy table detailed in section 2.1 of . While conceptually the intent was for a configurable, longest-match table to be adjusted as-needed. In practice, modifying the prefix policy table remains difficult across platforms, and in some cases impossible. Embedded, proprietary, closed source, and IoT devices are especially difficult to adjust and are, in many cases, incapable of any adjustment whatsoever. Large scale manipulation of the policy table also remains out of the realm of realistic support for small and medium scale operators due to lack of ability to manipulate all the hosts and systems, or a lack of tooling and access. Below is an example of a gai.conf file from a modern Linux installation as of 03 April 2022:

# If set to yes, each getaddrinfo(3) call will check whether this file # changed and if necessary reload. This option should not really be # used. There are possible runtime problems. The default is no. # # label # Add another rule to the RFC 3484 label table. See section 2.1 in # RFC 3484. The default is: # #label ::1/128 0 #label ::/0 1 #label 2002::/16 2 #label ::/96 3 #label ::ffff:0:0/96 4 #label fec0::/10 5 #label fc00::/7 6 #label 2001:0::/32 7 # # This default differs from the tables given in RFC 3484 by handling # (now obsolete) site-local IPv6 addresses and Unique Local Addresses. # The reason for this difference is that these addresses are never # NATed while IPv4 site-local addresses most probably are. Given # the precedence of IPv6 over IPv4 (see below) on machines having only # site-local IPv4 and IPv6 addresses a lookup for a global address would # see the IPv6 be preferred. The result is a long delay because the # site-local IPv6 addresses cannot be used while the IPv4 address is # (at least for the foreseeable future) NATed. We also treat Teredo # tunnels special. # # precedence # Add another rule to the RFC 3484 precedence table. See section 2.1 # and 10.3 in RFC 3484. The default is: # #precedence ::1/128 50 #precedence ::/0 40 #precedence 2002::/16 30 #precedence ::/96 20 #precedence ::ffff:0:0/96 10 # # For sites which prefer IPv4 connections change the last line to # #precedence ::ffff:0:0/96 100 # # scopev4 # Add another rule to the RFC 6724 scope table for IPv4 addresses. # By default the scope IDs described in section 3.2 in RFC 6724 are # used. Changing these defaults should hardly ever be necessary. # The defaults are equivalent to: # #scopev4 ::ffff:169.254.0.0/112 2 #scopev4 ::ffff:127.0.0.0/104 2 #scopev4 ::ffff:0.0.0.0/96 14 ]]> Several assumptions are made here and are largely based on interpretations of but are not operationally relevant in modern networks. As this file or an equivalent structure within a given operating system is referenced, it dictates the behavior of the getaddrinfo() or analogous process. More specifically, where getaddrinfo() or comparable API is used, the sorting behavior should take into account both the source address of the requesting host as well as the destination addresses returned and sort according to both source and destination addressing, i.e, when a ULA address is returned, the source address selection should return and use a ULA address if available. Similarly, if a GUA address is returned the source address selection should return a GUA source address if available. Here are some example failure modes: ULA per is less preferred (the Precedence value is lower) than all legacy IPv4 (represented by ::ffff:0:0/96 in the aforementioned table). Because of the lower Precedence value of fc00::/7, if a host has legacy IPv4 enabled, it will use legacy IPv4 before using ULA. A dual-stacked client will source the traffic from the legacy IPv4 address, meaning it will require a corresponding legacy IPv4 destination address. Per number 3, even a host choosing a destination with A and AAAA DNS records, the host in question will choose the A record to get an legacy IPv4 address for the destination, meaning ULA IPv6 is rendered completely unused. It is also notable that Happy Eyeballs ( ) will not change the source address selection process on a host. Happy Eyeballs will only modify the destination sorting process. As a direct result of the described failure modes, and in addition to the aforementioned operational implications, use of ULA is not a viable option for dual-stack \ networking transition planning, large scale network modeling, network lab environments or other modes of emulating a large scale networking that runs both IPv4 and IPv6 concurrently.