RPKI Manifest Number Handling

RPKI Manifest Number Handling Asia Pacific Network Information Centre

6 Cordelia St South Brisbane 4101 Australia QLD tomh@apnic.net

Asia-Pacific Network Information Centre

6 Cordelia St South Brisbane QLD 4101 Australia ggm@apnic.net

Fastly

Amsterdam Netherlands job@fastly.com

General Internet Engineering Task Force template The Resource Public Key Infrastructure (RPKI) makes use of signed objects called manifests. A manifest lists each file that a publisher intends to include within an RPKI repository, and can be used to detect certain forms of attack against a repository. Manifests include a "manifest number" (manifestNumber), which the publisher must increment whenever it issues a new manifest, and Relying Parties (RPs) are required to verify that a newly-retrieved manifest for a given Certification Authority (CA) has a higher manifestNumber than the previously-validated manifest. However, the manifestNumber field is 20 octets in length (i.e. not unbounded), and no behaviour is specified for when a manifestNumber reaches the largest possible value. This document specifies publisher and RP behaviour for this scenario.

The Resource Public Key Infrastructure (RPKI) makes use of signed objects called manifests . A manifest lists each file that a publisher intends to include within an RPKI repository , and can be used to detect certain forms of attack against a repository. Manifests include a "manifest number" (manifestNumber), which the publisher must increment whenever it issues a new manifest, and Relying Parties (RPs) are required to verify that a newly-retrieved manifest for a given Certification Authority (CA) has a higher manifestNumber than the previously-validated manifest (see section 4.2.1 of ). However, the manifestNumber field is 20 octets in length (i.e. not unbounded), and no behaviour is specified for when a manifestNumber reaches the largest possible value (2^159-1), which means that a publisher can no longer make use of a given CA certificate when that value is reached. (For the purposes of , a "CA" is represented by a CA certificate with a stable location and a stable private key. Reissuing a CA certificate with changed resources or a changed expiry date does not change the identity of the CA such that the stored manifestNumber for the CA is reset, for example.) While it is practically impossible for a publisher to reach the largest possible value under normal operating conditions (it would require that the publisher issue one manifest per second for 23,171,956,451,847,141,650,870 quintillion years), there is a chance that it could be reached due to bugs in the issuance or publication systems or incorrect/inadvertent use of those systems. For example: occur: Incrementing by large values when issuing manifests, such that the time to reach that largest value is reduced. Reissuing new manifests in an infinite delay-free loop, such that the manifestNumber increases by a large value in a comparatively short period of time. Inadvertently setting the manifestNumber to the largest possible value, such that the publisher will no longer be able to publish usable manifests for that repository. These scenarios might also arise in combination and be more severe as a result: for example, a large manifest number increment bug in conjunction with a manifest reissuance loop problem. For a subordinate CA, the risk of repository invalidation due to this problem can be addressed by the publisher simply using the key rollover process () to get a new Certification Authority (CA) certificate. RPs will treat this new certificate as though it represents a distinct CA, and the manifestNumber can be reset at that point. However, this option is not available for RPKI Trust Anchors (TAs). If a TA publishes a manifest with the largest-possible manifestNumber value, then it is not possible to make use of the TA after that point, because the certificate location (stored in the associated Trust Anchor Locator (TAL) ) and its private key cannot be changed. Issuing a new TA and distributing the associated TAL to clients would involve a large amount of work for TA operators and RPs, and there would be a limited degree of RPKI protection by way of that TA for the time between the issuance of the problematic manifest and the installation of the new TAL for a given client. In order to avoid these problems, this document defines how publishers and RPs can handle this scenario in order to facilitate ongoing use of an affected repository.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in .

For a given CA, an RP MUST NOT reject a new manifest issued by that CA on the grounds of it not having a higher manifestNumber than a previously-validated manifest if the new manifest has a different filename from that of the previously-validated manifest. In other words, an RP MUST reset its stored manifestNumber for a given CA if the CA changes the filename of its manifest. With this behaviour, it is possible for a CA to be configured such that any time it issues a new manifest, it uses a new filename for that manifest. If a CA were configured in this way, the manifestNumber validation set out in section 4.2.1 of would have no purpose. To avoid this outcome, CAs SHOULD NOT use new filenames for manifests except in situations where it is necessary to ensure the ongoing validity of the CA or its repository. Similarly, RP software SHOULD alert its operators when a manifest filename changes for a given CA. To avoid certain forms of replay attack, the RP MUST verify that the URI in the accessLocation in one of the id-ad-signedObject accessMethods in the manifest's Subject Information Access (SIA) extension exactly matches the URI presented in the RPKI Repository Delta Protocol (RRDP) "publish" element or the path presented by remote rsync servers. Section 2.2 of contains non-normative guidance for the naming of manifest files in repositories. While a CA that supports the behaviour described in this section cannot preserve the exact filename suggested by that text (per section 2.1 of ), the CA SHOULD still ensure that the filename is a value derived from the public key of the CA, per the more general guidance in that section. Note that the approach set out in this section is different from that described in .

The previous section contains a specific update for the handling of manifest numbers, in order to address one potential permanent invalidity scenario. RPs that encounter other permanent invalidity scenarios SHOULD also consider how those can be addressed such that the scenario does not require the relevant CA or TA to perform a key rollover operation. For example, in the event that an RP recognises that a permanent invalidity scenario has occurred, the RP could alert the operator and provide an option to the operator to stop relying on cached data for the affected repository, so that the CA can rectify the problem.

CA software may opt to support the manifest number reset functionality in various ways. For example, it could change the manifest filename when the manifestNumber reaches a certain threshold, or it could alert the operator in this scenario and request confirmation that the filename should be changed.

N/A

Implementation status This section records the status of known implementations of the protocol defined by this specification at the time of posting of this Internet-Draft, and is based on a proposal described in . The description of implementations in this section is intended to assist the IETF in its decision processes in progressing drafts to RFCs. Please note that the listing of any individual implementation here does not imply endorsement by the IETF. Furthermore, no effort has been spent to verify the information presented here that was supplied by IETF contributors. This is not intended as, and must not be construed to be, a catalog of available implementations or their features. Readers are advised to note that other implementations may exist. According to , "this will allow reviewers and working groups to assign due consideration to documents that have the benefit of running code, which may serve as evidence of valuable experimentation and feedback that have made the implemented protocols more mature. It is up to the individual working groups to use this information as they see fit".

OpenBSD

The authors would like to thank for their review and feedback on this document.

&RFC2119; &RFC4387; &RFC6488; &RFC8174; &RFC8182; &RFC9286; &RFC1982; &RFC5280; &RFC6480; &RFC6481; &RFC6489; &RFC7942; &RFC8630; &RFC8488; rpki-client OpenBSD Project

Serial number arithmetic is an approach that has been used in the DNS context (among others) to permit the indefinite use of a finite number space. At least in theory, it would be possible to use a similar approach with the manifestNumber field as well. However, unlike the corresponding DNS context with Start of Authority (SOA) resource records, an RPKI CA does not have visibility into or control over RPKI RPs generally. This means that it is not possible to select an updated manifestNumber value or to manage the relevant state transitions so as to guarantee that all RPs will have valid state at the end of the process. The approach proposed in does not have this problem.

The thisUpdate field in the manifest object is of type GeneralizedTime, defined in section 4.1.2.5.2 of . This type has a maximum value of 99991231235959Z (i.e. 31 December 9999 23:59:59 GMT). Section 4.2.1 of requires that "[e]ach RP MUST verify that this field value is greater (more recent) than the most recent manifest it has validated", so it would appear to be subject to the same problem as for manifest numbers. However, during validation, if the RP detects that the current time is not between the manifest thisUpdate and nextUpdate values, the RP must treat the fetch as a failed fetch. Therefore, the RP will not cache a manifest with a current date far in the future, and the CA can rectify the problem here by reissuing the relevant manifest with the correct date.

Walkthrough of the rpki-client implementation This section describes the implementation with regard to handling manifest numbers. The process is composed of multiple stages:

Fetching the manifests and acquiring referenced files
Preprocessing of the manifests
Selecting the first candidate manifest
Matching file names and hashes
Optionally selecting the second candidate manifest

Stage: Fetching the Manifests and Acquiring Referenced Files The RP follows rpkiNotify or caRepository pointers in the SubjectInfoAccess extension of valid CA certificates to queue up synchronization tasks. At the end of this stage the RP has zero, one, or two manifests for a given caRepository. Depending on the validation status, the RP stores files into two locations: DIR_VALID or DIR_TEMP. DIR_VALID contains objects which were found to be valid (current, not revoked, not expired) during the previous validation run, the DIR_TEMP location contains files retrieved via RRDP or rsync which have not yet been validated, or were rejected by the validation process. If the remote publication point is unreachable on both RRDP and rsync, no purported "new" manifest file will be stored in DIR_TEMP. It is possible that the DIR_VALID location contains a locally cached version of the object from a previous validation run.

Stage: preprocessing of the manifests Constructing the path and filename based on the rpkiManifest of the CA certificate, the RP attempts to open what purportedly are two version of the same .mft file in DIR_TEMP and DIR_VALID, respectively. For brevity's sake, the version in DIR_TEMP is associated with a data structure named mft1, the version in DIR_VALID is associated with a data structure named mft2. Assuming two files exist in the DIR_TEMP and DIR_VALID locations, both files are run through a series of checks. If any check fails, that file will be considered ineligible.

Can the file be opened?
Can the content of the file be decoded as DER?
Can the DER-content be parsed as CMS ContentInfo?
Is the CMS self-signage correct?
Can exactly one CMS SignerInfo be extracted?
Is the ContentInfo of the right version?
Is the SignerInfo of the right version?
Does the SignerInfo have the correct signed attributes?
Does the SignerInfo have the correct digest and signature algorithms?
Does the ContentInfo have the right type of embedded content?
Does the eContentType match the Content-Type?
Does the CMS contain zero CRLs?
Can exactly one X.509 cert be extracted from the SignerInfo?
Can the notBefore field be extracted from the X.509 cert?
Can the notAfter field be extracted from the X.509 cert?
Does the X.509 cert's SKI match the SignerInfo's SignerIdentifier?
Can the AIA be extracted from the X.509 EE?
Can the AKI be extracted from the X.509 EE?
Can the SIA be extracted from the X.509 EE?
Does the SIA Signed Object pathname match the pathname presented by the publication point?
Can the SKI be extracted from the X.509 EE?
Are the X509 EE's RFC 3779 extensions set to inherit?
Can the eContent be parsed according to the ASN.1 formal syntax?
Is the Manifest eContent of the right version?
Can the manifestNumber be extracted?
Is the CMS signing-time before the Manifest nextUpdate time?
Is 'now' not before the Manifest thisUpdate?
Is 'now' not after the Manifest nextUpdate?
Is the Manifest nextUpdate not before the Manifest thisUpdate?
Does a valid certification path from a TA to this EE cert exist?

Through the above checks, the mft1 and mft2 data structures are populated, or marked ineligible.

Stage: selecting the first candidate manifest Assuming both mft1 and mft2 successfully passed through stage 2, a comparison can be made between mft1 and mft2 to select the candidate mft for the next stage. The RP checks whether the locally cached version mft2 (from DIR_VALID) is older in the sense that was issued earlier than mft1 (from DIR_TEMP) by comparing the Manifest thisUpdate timestamp, and has a smaller manifestNumber. If both conditions are true, the RP will select mft1 as candidate for stage stage 4. If there was some kind of issue with mft1(such as it being older than or has the same thisUpdate as mft2, or it having a manifestNumber which is lower than or equal to mft2), the RP proceeds with stage 5.

Stage: matching file names and hashes for mft1 The RP will now verify the hash value of each file listed in manifest mft1 matches the value obtained by hashing the file acquired from the publication point. If the computed hash value of a file listed on the manifest does not match the hash value contained in the manifest, then a failed fetch occurred and the RP proceeds to stage 5. If all the files and hashes matched, mft1 and its associated files are moved from DIR_TEMP to DIR_VALID. The manifest handling procedure now ends.

Optional Stage: matching file names and hashes for mft2 This stage is only reached if there was an issue with mft1. The RP will now verify the hash value of each file listed in manifest mft2 matches the value obtained by hashing the file acquired from the publication point. If the computed hash value of a file listed on the manifest does not match the hash value contained in the manifest, then the caRepository is busted.