| Internet-Draft | Revocation in OpenPGP | August 2023 |
| Gillmor | Expires 18 February 2024 | [Page] |
Cryptographic revocation is a hard problem. OpenPGP's revocation mechanisms are imperfect, not fully understood, and not as widely implemented as they could be. Additionally, some historical OpenPGP revocation mechanisms simply do not work in certain contexts. This document provides clarifying guidance on how OpenPGP revocation works, documents outstanding problems, and introduces a new mechanism for delegated revocations that improves on previous mechanism.¶
This note is to be removed before publishing as an RFC.¶
The latest revision of this draft can be found at https://dkg.gitlab.io/openpgp-revocation/. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-dkg-openpgp-revocation/.¶
Discussion of this document takes place on the OpenPGP Working Group mailing list (mailto:[email protected]), which is archived at https://mailarchive.ietf.org/arch/browse/openpgp/. Subscribe at https://www.ietf.org/mailman/listinfo/openpgp/.¶
Source for this draft and an issue tracker can be found at https://gitlab.com/dkg/openpgp-revocation.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 18 February 2024.¶
Copyright (c) 2023 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
OpenPGP revocation is complicated. This document attempts to clean it up and build a consensus around syntax, semantics, use cases, and workflows.¶
The only substantive wire format change is the introduction of the "Delegated Revoker" subpacket described in Section 8.¶
The term "OpenPGP Certificate" is used in this document interchangeably with "OpenPGP Transferable Public Key", as defined in Section 10.1 of [I-D.ietf-openpgp-crypto-refresh].¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
There are three kinds of signatures that do revocation: Key Revocation (0x20), Subkey Revocation (0x28), and Certification Revocation (0x30).¶
This document focuses on revoking full OpenPGP certificates (a.k.a. "Transferable Public Keys") using Key Revocation signatures (0x20).¶
This document also explicitly deprecates the remaining two signature types: Subkey Revocation (0x28) and Certification Revocation (0x30).¶
User ID self-certifications and direct-key self-signatures can be explicitly expired or replaced by the keyholder by issuing a superseding certification, so the only reason for a certification revocation is for third-party certifications.¶
When Alice revokes her certification over Bob's Primary Key and User ID, what is she saying specifically?¶
How does Alice's Certification Revocation signature packet get placed into Bob's certificate?¶
Why doesn't Alice just issue a superseding certification of her own over Bob's User ID instead of revoking it?¶
FIXME: Given an initial certification at time T, if the superseding certification has a timestamp of T+1, then will a verifier that cares about the certification still accept signatures from the key based on the User ID that were made exactly at time T?
Alternately, if the superseding certification has a timestamp of time T exactly, will verifiers prefer its expiration date or the initial certification's expiration date (or lack thereof)?¶
Why bother revoking a subkey? Why not just issue an superseding Subkey Binding Signature?¶
FIXME: One reason why revoking a subkey might be nice is if a subkey has been compromised, and multiple historical subkey bindings have been made. In that case, to have the same effect the keyholder would need to issue one superseding expiring Subkey Binding Signature for each known Subkey Binding Signature, which is kind of a mess.¶
What happens when a Subkey Binding Signature is revoked, and then later a new Subkey Binding Signature is made over the same subkey?¶
This section describes a number of outstanding challenges with implementing OpenPGP revocation in the state of the field before this document. Some of these problems are fixed by this document.¶
How does the user know that they have the correct revocation status? Where do they look for revocations from? With what frequency?¶
When the keyholder changes to a new key, how do they distribute revocations for older keys?¶
Given the chance to tamper with an OpenPGP certificate, the simplest thing that an adversary can do is to strip signature subpackets. Stripping a revocation signature subpacket is trivial, and the resulting certificate looks valid.¶
An OpenPGP implementation needs a reliable channel to fetch revocation signatures from, and a reliable and well-indexed storage mechanism to retain them safely to avoid using revoked certificates.¶
What if we find a Key Revocation signature made using SHA1 or MD5?¶
Should we consider the indicated key revoked?¶
Primary keys sign Subkey Binding Signatures. Signing-capable subkeys sign Primary Key Binding Signatures.¶
Is there ever a scenario where a signing-capable subkey might want to revoke its own Primary Key Binding Signature?¶
If so, how is that done?¶
You find a self-revoked primary key, and you find another OpenPGP certificate in the wild that uses the same key material (but maybe different creation date, or used as a subkey instead of as a primary key). Is it acceptable for use?¶
In certificate with primary key X, there is a revoked subkey Y (it was revoked by X issuing a valid Subkey Revocation signature). But the certificate with primary key Z, also has subkey Y. Is subkey Y valid for the Z certificate?¶
How should an implementation interpret a Key Revocation signature or Subkey Revocation signature with Reason for Revocation subpacket with ID 32 ("User ID information is no longer valid")?¶
How should an implementation interpret a Certification Revocation with a Reason for Revocation with, say, ID 1 ("Key is superseded")?¶
Do we just say these Revocation signatures are invalid? Do we ignore the Reasons for Revocation subpacket?¶
The Revocation Key Subpacket is deprecated because it suffers from several significant challenges in use. The Delegated Revoker mechanism (described in Section 8) is intended to avoid all of these problems.¶
The "Revocation Key" subpacket contains only a fingerprint. If a verifier observes such a packet, and a Key Revocation Signature that claims to be issued by the identified key, how does the verifier obtain the revoking key itself if they do not already have a copy of it?¶
What happens if the revocation key's public key packet is known, but it is not part of a certificate at all?¶
What if it is enclosed in a certificate, but that certificate indicates that it is expired, revoked, or otherwise invalid?¶
The following questions are based on the assumption that key A has pointed to key B in a "Revocation Key" subpacket.¶
What if B revokes both itself and key A: is A valid?¶
What if, instead, B indicates (via the deprecated "Revocation Key" subpacket) that key A is permitted to revoke key B? In this scenario, what happens when both A and B revoke each other?¶
What if A designates that B can revoke A, and B designates that C can revoke B? In that case, if C revokes B and then B revokes A, is A still valid?¶
Section 5.2.3.3 of [RFC4880] states:¶
Revoking a direct-key signature cancels that signature.¶
and¶
An implementation that encounters multiple self-signatures on the same object may resolve the ambiguity in any way it sees fit, but it is RECOMMENDED that priority be given to the most recent self-signature.¶
The revised version, Section 5.2.3.10 of [I-D.ietf-openpgp-crypto-refresh] makes this last sentence even stronger:¶
An implementation that encounters multiple self-signatures on the same object MUST select the most recent valid self-signature, and ignore all other self-signatures.¶
Consider the following sequence of events:¶
The verifier examines this sequence of packets: A, X, Y, Z.¶
X appears to have been superseded by Y. Should A be considered revoked?¶
But what if signature packet Y was a revocation signature instead:¶
Or, what if signature packet Y pointed to a different Revocation Key:¶
If, in any of these situations, the verifier does not consider A to be revoked by Z due to the presence of signature Y, then the mechanism does not work to protect the keyholder of A. An adversary who has taken control of A can create a signature packet Y to evade the third-party revocation capabilities that B is supposed to wield.¶
If delegating revocation power to a third party is intended to defend against an adversary, this implies that the guidance about superseding signatures cannot apply to signature packets that contain a Revocation Key. But then, if signature X is not revoked or superseded by signature Y (in whatever form Y takes), how should implementations deal with the other subpackets in signature X?¶
Surely it can issue a Key Revocation signature that covers the primary key itself.¶
But can it issue a Subkey Revocation signature on behalf of the primary key? Can it issue a Certification Revocation signature on behalf of the primary key?¶
FIXME: If a Revocation signature appears to have been made in the future, what should be done with it?¶
Implementations MUST NOT encrypt to a revoked certificate. Implementations MUST NOT accept a signature made by a revoked certificate as valid unless the revocation is "soft" (see Section 5) and the timestamp of the signature predates the timestamp of the revocation. Implementations MUST NOT use secret key material corresponding to a revoked certificate for signing, unless the secret key material also corresponds to a non-revoked certificate.¶
Implementations MAY use the secret key material corresponding to a revoked certificate.¶
Reasons for Revocation subpacket allows different values.¶
Some of them suggest that a verifier can still accept signatures from before the timestamp of the Revocation. These are "soft" revocations.¶
All the rest require that a verifier MUST treat the certificate as "hard" revoked, meaning that even signatures that have creation timestamps before the creation timestamp of the revocation signature should themselves be rejected.¶
Expiration makes just as much sense as a soft revocation in many circumstances, and is typically better supported.¶
FIXME: describe the circumstances under which a soft revocation would be preferable over an expiration. If there are none, explicitly discourage soft revocation.¶
A revocation certificate indicates that a given primary key is revoked.¶
This can take two common forms. Each form is a sequence of OpenPGP packets:¶
Additionally, there is a deprecated form:¶
This document introduces a new form in Section 8:¶
When an implementation observes any of the above forms of revocation certificate for a certificate with primary key X, it should record it and indicate that X has been revoked and is no longer to be used, along with all of its User IDs and Subkeys.¶
FIXME: talk about interactions with HKP, VKS, WKD, OPENPGPKEY (DANE), or other key discovery methods?¶
An escrowed revocation certificate is just a valid revocation certificate that is not published. The parties who can retrieve or reassemble the escrowed revocation certificate can publish it to inform the rest of the world that the certificate has been revoked. It is described in Section 13.9 of [I-D.ietf-openpgp-crypto-refresh].¶
In what circumstances does escrowed revocation work? When is it inappropriate?¶
An escrowed hard revocation certificate covers the use case where where the keyholder has lost control of the secret key material, and someone besides the keyholder may have gotten access to the secret key material.¶
At key creation time, keyholder creates a hard revocation certificate. Optionally, they encrypt it to a set of trusted participants. The keyholder stores the revocation certificate somewhere they or one of the trusted participants will be able to access it.¶
If the keyholder sends it to any trusted participant immediately, that participant can trigger a revocation any time they like. In this case, the keyholder and the trusted participants should clarify between themselves what an appropriate signal should be for when the trusted participant should act¶
If physical access is retained by the keyholder, then the keyholder has to be capable of consenting for the revocation to be published.¶
Do regular updates of the escrowed revocation (e.g. after each signing). Store them somewhere safe?¶
FIXME: how to split an escrowed revocation certificate among N parties so that any K of them can reconstruct it.¶
[ See also https://gitlab.com/openpgp-wg/rfc4880bis/-/merge_requests/18 where this was originally specified, in a slightly different form ]¶
This document introduces a new mechanism for permitting a distinct key to revoke an OpenPGP certificate. It should effectively replace the deprecated "Revocation Key subpacket", and should resolve the concerns described in {#revocation-key-subpacket-challenges}.¶
It uses a novel signature type and a novel subpacket type, is self-contained, irrevocable, and can only be used to revoke the entire OpenPGP certificate rooted in the primary key it corresponds to.¶
This introduces a new Signature Type, "Delegated Revoker Signature" with type ID TBD.¶
This signature type is made over data structured in the same way as a Direct Key Signature, but it does not supersede or replace any other signature, and it cannot be revoked or superseded itself. It is a permanent delegation.¶
A Delegated Revoker Signature MUST be made over the primary key of a given certificate, and its hashed subpackets area MUST contain exactly one each of the following four subpackets:¶
The Issuer Fingerprint subpacket contains the fingerprint of the primary key that is asserting this delegation. The Revocable subpacket flag MUST be set to 0. All subpackets MUST be marked as Critical.¶
FIXME: What should a verifier do if the set of hashed subpackets does not match this list?¶
The Delegated Revoker Signature packet can be placed in an OpenPGP certificate immediately after the Primary Key packet, before any Key Revocation Signature or Direct Key Signature packet.¶
FIXME: test implementations with this placement strategy to see whether they choke on the certificate. If this does not work, consider recommending its inclusion in an unhashed Embedded Signature subpacket the relevant associated Key Revocation Signature packet (if it is travelling with the revocation), or in such a subpacket of a Direct Key Signature packet directly.¶
FIXME: how do we deal with avoiding a leak of the existence of this relationship (e.g., the "Sensitive" bit in the deprecated "Revocaton Key" subpacket)? Do we need to?¶
Someone concerned about the risk of social graph leakage (see Section 3.6.3) can simply mint a new secret key and encrypt its corresponding Secret Key packet to their preferred revoker.¶
Or should we add an "Exportable" subpacket to the list above and describe its syntax more explicitly?¶
Alternately: what if we said that this is simply always treated as sensitive, in that without explicitly being part of the described workflow, it must not be transmitted except in the presence of a valid revocation? This puts the burden on the holder of the secret key to keep a copy of the delegation lying around, which is a novel use case.¶
The Delegated Revoker Subpacket has type ID 36.¶
It is only valid in the hashed subpackets section of a Delegated Revoker Signature (see Section 8.1) over a primary key. It MUST be marked as Critical.¶
The contents of this subpacket are a full Public Key packet, see Section 5.5.1.1 of [I-D.ietf-openpgp-crypto-refresh].¶
The embedded Public Key packet MUST be signing-capable.¶
Unlike other OpenPGP Signature packets, a Delegated Revoker Signature cannot be revoked, and is not superseded by any other Signature packet, including other Delegated Revoker Signature packets. If multiple valid Delegated Revoker Signatures are issued by a primary key X, they are all capable of issuing Key Revocation signatures over X on behalf of X.¶
This Public Key MAY be the same Public Key packet that is the root of a larger OpenPGP certificate, but it need not be. In the Delegated Revoker context, this Public Key packet is used on its own, regardless of the status of any matching OpenPGP certificate.¶
If an OpenPGP certificate with primary key X has an associated Delegated Revoker containing Public Key Y, that only indicates that the Y MAY make a valid Key Revocation signature on behalf of X, revoking X itself.¶
The Delegated Revoker Public Key (Y in the example above) MUST NOT be used to validate any other type of signature associated with that OpenPGP certificate.¶
FIXME: should we constrain the types of Key Revocations it can issue (i.e., the contents of any Reason for Revocation subpackets, or "hard" or "soft" choices)?¶
FIXME: What if the Delegated Revoker Signature is made over a digest algorithm that is deemed unsafe in the future? FIXME: What if the embedded Public Key Packet is known to be weak or compromised?¶
The Delegated Revoker mechanism is only useful for a potential scenario where the keyholder has lost control of the primary secret key. Otherwise, the keyholder could simply issue the desired Key Revocation signature (type 0x20) directly.¶
If the keyholder needs a hard revocation, and they have access to an escrowed Key Revocation signature, they can use that.¶
So the circumstances where a Delegated Revoker is relevant¶
Keyholder needs to choose who they think will be responsible for handling the delegated responsibility of revoking when the time is needed. This could be another individual, or it could be a machine that has distinct security and operational characteristics from the machine that holds the primary key.¶
FIXME: should this document outline how a group of trusted parties could effectively revoke a given certificate that authorized them to do so?¶
This draft asks IANA to do several things, all within the OpenPGP protocol group.¶
Add an entry "Delegated Revoker" to OpenPGP subpackets, type 36, referencing this document, Section 8.2.¶
The "Reasons for Revocation Code" registry should add a column to indicate "Hard/Soft". Only "Key is Superseded" and "Key is retired and no longer used" are marked "Soft". All other values should be treated as "Hard".¶
Add an entry "Delegated Revoker Signature" to OpenPGP Signature Types, type ID TBD, referencing this document, Section 8.1.¶
"Signature Types" registry entries should have References updated:¶
The "Reason for Revocation" entry in the "Subpacket Types" registry should have its References column updated to point to this document.¶
This document describes ways that the OpenPGP ecosystem deals with revoked certificates. Revocation is a security measure because it is a method of last resort for dealing with cryptographic credentials that are known to have failed for one reason or another.¶
The entire document is therefore focused on security. Implementers who ignore this guidance may either allow known-bad key material to be used (by ignoring a valid revocation signature) or may issue revocation signatures that other implementations are likely to ignore.¶
FIXME: Can all of the plausible workflows described in this document be done with the Stateless OpenPGP Interface? If not, what is missing?¶
FIXME: This document should include several different valid OpenPGP Revocation Certificates.¶
Phil Zimmermann motivated the Delegated Revoker work.¶
RFC Editor Note: Please delete this section before publication.¶
3.6.3. Social Graph Leakage
If it's possible to find a certificate containing the matching fingerprint in a deprecated "Revocation Key" subpacket, then an observer can learn who the keyholder trusts even when no revocation is needed.¶
An attacker that wants to target Alice, for example, can observe that Alice has indicated Bob seems trustworthy if Alice has pointed to Bob's key's fingerprint with a deprecated "Revocation Key" subpacket. The attacker might then go after Bob as a way to get to Alice.¶