| DomainKeys Identified Mail | T. Hansen |
| Internet Draft | AT&T Laboratories |
| Intended status: Informational | E. Siegel |
| Expires: July 31, 2010 | P. Hallam-Baker |
| Default Deny Security, Inc. | |
| D. Crocker | |
| Brandenburg InternetWorking | |
| January 27, 2010 |
DomainKeys Identified Mail (DKIM) Development, Deployment and Operations
draft-ietf-dkim-deployment-11
DomainKeys Identified Mail (DKIM) allows an organization to claim responsibility for transmitting a message, in a way that can be validated by a recipient. The organization can be the author's, the originating sending site, an intermediary, or one of their agents. A message can contain multiple signatures, from the same or different organizations involved with the message. DKIM defines a domain-level digital signature authentication framework for email, using public key cryptography, using the domain name service as its key server technology. This permits verification of a responsible organization, as well as the integrity of the message contents. DKIM will also provide a mechanism that permits potential email signers to publish information about their email signing practices; this will permit email receivers to make additional assessments about messages. DKIM's authentication of email identity can assist in the global control of "spam" and "phishing". This document provides implementation, deployment, operational and migration considerations for DKIM.
This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts.
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”.
The list of current Internet-Drafts can be accessed at <http://www.ietf.org/ietf/1id-abstracts.txt>.
The list of Internet-Draft Shadow Directories can be accessed at <http://www.ietf.org/shadow.html>.
This Internet-Draft will expire in July 31, 2010.
Copyright © 2010 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 (http://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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the BSD License.
This document may contain material from IETF Documents or IETF Contributions published or made publicly available before November 10, 2008. The person(s) controlling the copyright in some of this material may not have granted the IETF Trust the right to allow modifications of such material outside the IETF Standards Process. Without obtaining an adequate license from the person(s) controlling the copyright in such materials, this document may not be modified outside the IETF Standards Process, and derivative works of it may not be created outside the IETF Standards Process, except to format it for publication as an RFC or to translate it into languages other than English.
DomainKeys Identified Mail (DKIM) allows an organization to claim responsibility for transmitting a message, in a way that can be validated by a recipient. This document provides practical tips for: those who are developing DKIM software, mailing list managers, filtering strategies based on the output from DKIM verification, and DNS servers; those who are deploying DKIM software, keys, mailing list software, and migrating from DomainKeys [RFC4870]; and those who are responsible for the on-going operations of an email infrastructure that has deployed DKIM.
The document is organized around the key concepts related to DKIM. Within each section, additional considerations specific to development, deployment, or ongoing operations are highlighted where appropriate. The possibility of use of DKIM results as input to a local reputation database is also discussed.
DKIM participates in a trust-oriented enhancement to the Internet's email service, to facilitate message handling decisions, such as for delivery and for content display. Trust-oriented message handling has substantial differences from the more established approaches that consider messages in terms of risk and abuse. With trust, there is a collaborative exchange between a willing participant along the sending path and a willing participant at a recipient site. In contrast, the risk model entails independent, unilateral action by the recipient site, in the face of a potentially unknown, hostile and deceptive sender. This translates into a very basic technical difference: In the face of unilateral action by the recipient and even antagonistic efforts by the sender, risk-oriented mechanisms are be based on heuristics, that is, on guessing. Guessing produces statistical results with some false negatives and some false positives. For trust-based exchanges, the goal is the deterministic exchange of information. For DKIM, that information is the one identifier that represents a stream of mail for which an independent assessment is sought (by the signer.)
A trust-based service is built upon a validated Responsible Identifier that labels a stream of mail and is controlled by an identity (role, person or organization). The identity is acknowledging some degree of responsibility for the message stream. Given a basis for believing that an identifier is being used in an authorized manner, the recipient site can make and use an assessment of the associated identity. An identity can use different identifiers, on the assumption that the different streams might produce different assessments. For example, even the best-run marketing campaigns will tend to produce some complaints that can affect the reputation of the associated identifier, whereas a stream of transactional messages is likely to have a more pristine reputation.
Determining that the identifier's use is valid is quite different from determining that the content of a message is valid. The former means only that the identifier for the responsible role, person or organization has been legitimately associated with a message. The latter means that the content of the message can be believed and, typically, that the claimed author of the content is correct. DKIM validates only the presence of the identifier used to sign the message. Even when this identifier is validated, DKIM carries no implication that any of the message content, including the RFC5322.From field [RFC5322], is valid. Surprisingly, this limit to the semantics of a DKIM signature applies even when the validated signing identifier is the same domain name as is used in the RFC5322.From field! DKIM's only claim about message content is that the content cited in the DKIM-Signature: field's h= tag has been delivered without modification. That is, it asserts message content integrity -- between signing and verifying -- not message content validity.
As shown in Figure 1, this enhancement is a communication between a responsible role, person or organization that signs the message and a recipient organization that assesses its trust in the signer. The recipient then makes handling decisions based on a collection of assessments, of which the DKIM mechanism is only a part. In this model, as shown in Figure 1, validation is an intermediary step, having the sole task of passing a validated Responsible Identifier to the Identity Assessor. The communication is of a single Responsible Identifier that the Responsible Identity wishes to have used by the Identity Assessor. The Identifier is the sole, formal input and output value of DKIM signing. The Identity Assessor uses this single, provided Identifier for consulting whatever assessment data bases are deemed appropriate by the assessing entity. In turn, output from the Identity Assessor is fed into a Handling Filter engine that considers a range of factors, along with this single output value. The range of factors can include ancillary information from the DKIM validation.
Identity Assessment covers a range of possible functions. It can be as simple as determining whether the identifier is a member of some list, such as authorized operators or participants in a group that might be of interest for recipient assessment. Equally, it can indicate a degree of trust (reputation) that is to be afforded the actor using that identifier. The extent to which the assessment affects handling of the message is, of course, determined later, by the Handling Filter.
+------+------+ +------+------+
| Author | | Recipient |
+------+------+ +------+------+
| ^
| |
| +------+------+
| -->| Handling |<--
| -->| Filter |<--
| +-------------+
| ^
V Responsible |
+-------------+ Identifier +------+------+
| Responsible |. . . . . . . . . . .>| Identity |
| Identity | . . | Assessor |
+------+------+ . . +-------------+
| V . ^ ^
V . . | |
+------------------.-------.--------------------+ | |
| +------+------+ . . . > . +-------------+ | | | +-----------+
| | Identifier | | Identifier +--|--+ +--+ Assessment|
| | Signer +------------->| Validator | | | Databases |
| +-------------+ +-------------+ | +-----------+
| DKIM Service |
+-----------------------------------------------+
Figure 1: Actors in a Trust Sequence using DKIM
The signer of a message needs to be able to provide precise data and know what that data will mean upon delivery to the Assessor. If there is ambiguity in the choice that will be made on the receive side, then the sender cannot know what basis for assessment will be used. DKIM has three values that specify identification information and it is easy to confuse their use, although only one defines the formal input and output of DKIM, with the other two being used for internal protocol functioning and adjunct purposes, such as auditing and debugging.
The salient values include the s=, d= and i= parameters in the DKIM-Signature: header field. In order to achieve the end-to-end determinism needed for this collaborative exchange from the signer to the assessor, the core model needs to specify what the signer is required to provide to the assessor. The Update to RFC4871 [RFC5672] now specifies:
The single, mandatory value that DKIM supplies as its output is:
The adjunct values are:
So, i= can have any of these properties:
This underscores why the tag needs to be treated as being opaque, since it can represent any semantics, known only to the signer.
Hence, i= serves well as a token that is usable like a Web cookie, for return to the signing ADMD -- such as for auditing and debugging. Of course in some scenarios the i= string might provide a useful adjunct value for additional (heuristic) processing by the Handling Filter.
A DKIM signing entity can serve different roles, such as being the author of content, or the operator of the mail service, or the operator of a reputation service that also provides signing services on behalf of its customers. In these different roles, the basis for distinguishing among portions of email traffic can vary. For an entity creating DKIM signatures it is likely that different portions of its mail will warrant different levels of trust. For example:
It is therefore likely to be useful for a signer to use different d= sub-domain names, for different message traffic streams, so that receivers can make differential assessments. However, too much differentiation -- that is, too fine a granularity of signing domains -- makes it difficult for the receiver to discern a sufficiently stable pattern of traffic for developing an accurate and reliable assessment. So the differentiation needs to achieve a balance. Generally in a trust system, legitimate signers have an incentive to pick a small stable set of identities, so that recipients and others can attribute reputations to them. The set of these identities a receiver trusts is likely to be quite a bit smaller than the set it views as risky.
The challenge in using additional layers of sub-domains is whether the extra granularity will be useful for the assessor. In fact, excessive levels invite ambiguity: if the assessor does not take advantage of the added granularity in the entire domain name that is provided, they might unilaterally decide to use only some rightmost part of the identifier. The signer cannot know what portion will be used. That ambiguity would move the use of DKIM back to the realm of heuristics, rather than the deterministic processing that is its goal.
Hence the challenge is to determine a useful scheme for labeling different traffic streams. The most obvious choices are among different types of content and/or different types of authors. Although stability is essential, it is likely that the choices will change, over time, so the scheme needs to be flexible.
For those originating message content, the most likely choice of sub-domain naming scheme will by based upon type of content, which can use content-oriented labels or service-oriented labels. For example:
transaction.example.com newsletter.example.com bugreport.example.com support.example.com sales.example.com marketing.example.com
where the choices are best dictated by whether they provide the Identity Assessor with the ability to discriminate usefully among streams of mail that demonstrate significantly different degrees of recipient acceptance or safety. Again, the danger in providing too fine a granularity is that related message streams that are labeled separately will not benefit from an aggregate reputation.
For those operating messaging services on behalf of a variety of customers, an obvious scheme to use has a different sub-domain label for each customer. For example:
widgetco.example.net moviestudio.example.net bigbank.example.net
However it can also be appropriate to label by the class of service or class of customer, such as:
premier.example.net free.example.net certified.example.net
Prior to using domain names for distinguishing among sources of data, IP Addresses have been the basis for distinction. Service operators typically have done this by dedicating specific outbound IP Addresses to specific mail streams -- typically to specific customers. For example, a university might want to distinguish mail from the Administration, versus mail from the student dorms. In order to make adoption of a DKIM-based service easier, it can be reasonable to translate the same partitioning of traffic, using domain names in place of the different IP Addresses.
DKIM gives the recipient site's Identity Assessor a verifiable identifier to use for analysis. Although the mechanism does not make claims that the signer is a Good Actor or a Bad Actor, it does make it possible to know that use of the identifier is valid. This is in marked contrast with schemes that do not have authentication. Without verification, it is not possible to know whether the identifier -- whether taken from the RFC5322.From field, RFC5321.MailFrom command, or the like -- is being used by an authorized agent. DKIM solves this problem. Hence with DKIM, the Assessor can know that two messages with the same DKIM d= identifier are, in fact, signed by the same person or organization. This permits a far more stable and accurate assessment of mail traffic using that identifier.
DKIM is distinctive, in that it provides an identifier which is not necessarily related to any other identifier in the message. Hence, the signer might be the author's ADMD, one of the operators along the transit path, or a reputation service being used by one of those handling services. In fact, a message can have multiple signatures, possibly by any number of these actors.
As discussed above, the choice of identifiers needs to be based on differences that the signer thinks will be useful for the recipient Assessor. Over time, industry practices establish norms for these choices.
A common view about a DKIM signature is that it carries a degree of assurance about some or all of the message contents, and in particular that the RFC5322.From field is likely to be valid. In fact, DKIM makes assurances only about the integrity of the data and not about its validity. Still, presumptions of RFC5322.From field validity remain a concern. Hence a signer using a domain name that is unrelated to the domain name in the RFC5322.From field can reasonably expect that the disparity will warrant some curiosity, at least until signing by independent operators has produced some established practice among recipient Assessors.
With the identifier(s) supplied by DKIM, the Assessor can consult an independent assessment service about the entity associated with the identifier(s). Another possibility is that the Assessor can develop its own reputation rating for the identifier(s). That is, over time, the Assessor can observe the stream of messages associated with the identifier(s) developing a reaction to associated content. For example, if there is a high percentage of user complaints regarding signed mail with a "d=" value of "widgetco.example.net", the Assessor might include that fact in the vector of data it provides to the Handling Filter. This is also discussed briefly in Section 5.4.
The assessment of the signing identifier is given to a Handling Filter that is defined by local policies, according to a potentially wide range of different factors and weightings. This section discusses some of the kinds of choices and weightings that are plausible, and the differential actions that might be performed. Because authenticated domain names represent a collaborative sequence between signer and assessor, actions can sometimes reasonably include contacting the signer.
The discussion focuses on variations in Organizational Trust versus Message Stream Risk, that is, the degree of positive assessment of a DKIM-signing organization, and the potential danger present in the message stream signed by that organization. While it might seem that higher trust automatically means lower risk, the experience with real-world operations provides examples of every combination of the two factors, as shown in Figure 2. For each axis, only three levels of granularity are listed, in order to keep discussion manageable. In real-world filtering engines, finer-grained distinctions are typically needed, and there typically are more axes. For example, there are different types of risk, so that an engine might distinguish between spam risk versus virus risk and take different actions based on which type of problematic content is present. For spam, the potential damage from a false negative is small, whereas the damage from a false positive is high. For a virus, the potential danger from a false negative is extremely high, while the likelihood of a false positive when using modern detection tools is extremely low. However for the discussion here, "risk" is taken as a single construct.
The DKIM d= identifier is independent of any other identifier in a message and can be a sub-domain of the name owned by the signer. This permits use of fine-grained and stable distinctions between different types of message streams, such as between transactional messages and marketing messages from the same organization. Hence, use of DKIM might permit a richer filtering model than has typically been possible for mail receiving engines.
Note that the realities of today's public Internet Mail environment necessitate having a baseline handling model that is quite suspicious. Hence, "strong" filtering rules really are the starting point, as indicated for the UNKNOWN cell.
The table indicates differential handling for each combination, such as how aggressive or broad-based the filtering could be. Aggressiveness affects the types of incorrect assessments that are likely. So the table distinguishes various characteristics, including: 1) whether an organization is unknown, known to be good actors and known to be bad actors; and 2) the assessment of messages. It includes advice about the degree of filtering that might be done, and other message disposition. Perhaps unexpectedly, it also lists a case in which the receiving site might wish to deliver problematic mail, rather than redirecting or deleting it, but also contacting the signing organization and seeking
+-------------+-----------------------------------------------+ | S T R E A M * O R G A N I Z A T I O N A L T R U S T | | R I S K * Low Medium High | | +***************+***************+***************+ | Low * BENIGN: | DILIGENT: | PRISTINE | | * Moderate | Mild | Accept | | * filter | filter | | | +---------------+---------------+---------------+ | Medium * UNKNOWN: | TYPICAL: | PROTECTED: | | * Strong | Targeted | Accept & | | * filter | filter | Contact | | +---------------+---------------+---------------+ | High * MALICIOUS: | NEGLIGENT: | COMPROMISED: | | * Block & | Block | Block & | | * Counter | | Contact | +-------------+---------------+---------------+---------------+
Figure 2: Trust vs. Risk Handling Tradeoffs Example
[LEGEND]
By itself, verification of a digital signature only allows the verifier to conclude with a very high degree of certainty that the signature was created by a party with access to the corresponding private signing key. It follows that a verifier requires means to (1) obtain the public key for the purpose of verification and (2) infer useful attributes of the key holder.
In a traditional Public Key Infrastructure (PKI), the functions of key distribution and key accreditation are separated. In DKIM [RFC4871], these functions are both performed through the DNS.
In either case, the ability to infer semantics from a digital signature depends on the assumption that the corresponding private key is only accessible to a party with a particular set of attributes. In traditional PKI, a Trusted Third Party (TTP) vouches that the key holder has been validated with respect to a specified set of attributes. The range of attributes that can be attested in such a scheme is thus limited only to the type of attributes that a TTP can establish effective processes for validating. In DKIM, Trusted Third parties are not employed and the functions of key distribution and accreditation are combined.
Consequently there are only two types of inference that a signer can make from a key published in a DKIM Key Record:
The ability to draw any useful conclusion from verification of a digital signature relies on the assumption that the corresponding private key is only accessible to a party with a particular set of attributes. In the case of DKIM, this means that the party that created the corresponding DKIM key record in the specific zone intended to claim responsibility for the signed message.
Ideally we would like to draw a stronger conclusion, that if we obtain a DKIM key record from the DNS zone example.com, that the legitimate holder of the DNS zone example.com claims responsibility for the signed message. In order for this conclusion to be drawn it is necessary for the verifier to assume that the operational security of the DNS zone and corresponding private key are adequate.
Access to signing keys needs to be carefully managed to prevent use by unauthorized parties and to minimize the consequences if a compromise were to occur.
While a DKIM signing key is used to sign messages on behalf of many mail users, the signing key itself needs to be under direct control of as few key holders as possible. If a key holder were to leave the organization, all signing keys held by that key holder needs to be withdrawn from service and if appropriate, replaced.
If key management hardware support is available, it needs to be used. If keys are stored in software, appropriate file control protections needs to be employed, and any location in which the private key is stored in plaintext form needs to be excluded from regular backup processes and is best not accessible through any form of network including private local area networks. Auditing software needs to be used periodically to verify that the permissions on the private key files remain secure.
Wherever possible a signature key needs to exist in exactly one location and be erased when no longer used. Ideally a signature key pair needs to be generated as close to the signing point as possible and only the public key component transferred to another party. If this is not possible, the private key needs to be transported in an encrypted format that protects the confidentiality of the signing key. A shared directory on a local file system does not provide adequate security for distribution of signing keys in plaintext form.
Key escrow schemes are not necessary and are best not used. In the unlikely event of a signing key becoming lost, a new signature key pair can be generated as easily as recovery from a key escrow scheme.
To enable accountability and auditing:
Best practices for management of cryptographic keying material require keying material to be refreshed at regular intervals, particularly where key management is achieved through software. While this practice is highly desirable it is of considerably less importance than the requirement to maintain the secrecy of the corresponding private key. An operational practice in which the private key is stored in tamper proof hardware and changed once a year is considerably more desirable than one in which the signature key is changed on an hourly basis but maintained in software.
In order to use DKIM a DNS domain holder requires (1) the ability to create the necessary DKIM DNS records and (2) sufficient operational security controls to prevent insertion of spurious DNS records by an attacker.
DNS record management is often operated by an administrative staff that is different from those who operate an organization's email service. In order to ensure that DKIM DNS records are accurate, this imposes a requirement for careful coordination between the two operations groups. If the best practices for private key management described above are observed, such deployment is not a onetime event; DNS DKIM selectors will be changed over time signing keys are terminated and replaced.
At a minimum, a DNS server that handles queries for DKIM key records needs to allow the server administrators to add free-form TXT records. It would be better if the DKIM records could be entered using a structured form, supporting the DKIM-specific fields.
Ideally DNSSEC [RFC4034] needs to be employed in a configuration that provides protection against record insertion attacks and zone enumeration. In the case that NSEC3 [RFC5155] records are employed to prevent insertion attack, the OPT-OUT flag needs to be set clear.
Selectors are assigned according to the administrative needs of the signing domain, such as for rolling over to a new key or for delegating of the right to authenticate a portion of the namespace to a trusted third party. Examples include:
jun2005.eng._domainkey.example.com
widget.promotion._domainkey.example.com
It is intended that assessments of DKIM identities be based on the domain name, and not include the selector. While past practice of a signer can permit a verifier to infer additional properties of particular messages from the structure DKIM key selector, unannounced administrative changes such as a change of signing software can cause such heuristics to fail at any time.
While a signer can establish business rules, such as issue of individual signature keys for each end-user, DKIM makes no provision for communicating these to other parties. Out of band distribution of such business rules is outside the scope of DKIM. Consequently there is no means by which external parties can make use of such keys to attribute messages with any greater granularity than a DNS domain.
If per-user signing keys are assigned for internal purposes (e.g. authenticating messages sent to an MTA for distribution), the following issues need to be considered before using such signatures as an alternative to traditional edge signing at the outbound MTA:
A DKIM key record only asserts that the holder of the corresponding domain name makes a claim of responsibility for messages signed under the corresponding key. In some applications, such as bulk mail delivery, it is desirable to delegate use of the key. That is, to allow a third party to sign on behalf of the domain holder. The trust relationship is still established between the domain holder and the verifier but the private signature key is held by a third party.
Signature keys used by a third party signer needs to be kept entirely separate from those used by the domain holder and other third party signers. To limit potential exposure of the private key, the signature key pair needs to be generated by the third party signer and the public component of the key transmitted to the domain holder, rather than have the domain holder generate the key pair and transmit the private component to the third party signer.
Domain holders needs to adopt a least privilege approach and grant third party signers the minimum access necessary to perform the desired function. Limiting the access granted to Third Party Signers serves to protect the interests of both parties. The domain holder minimizes its security risk and the Trusted Third Party Signer avoids unnecessary liability.
In the most restrictive case a domain holder maintains full control over the creation of key records and employs appropriate key record restrictions to enforce restrictions on the messages for which the third party signer is able to sign. If such restrictions are impractical, the domain holder needs to delegate a DNS subzone for publishing key records to the third party signer. It is best that the domain holder NOT allow a third party signer unrestricted access to its DNS service for the purpose of publishing key records.
Deployments need to establish, document and observe processes for managing the entire lifecycle of an asymmetric key pair.
When it is determined that a new key pair is required:
When it is determined that a private signature key is no longer required:
Creating messages that have one or more DKIM signatures, requires support in only two outbound email service components:
If the module creates more than one signature, there needs to be the appropriate means of telling it which one(s) to use. If a large number of names are used for signing, it will help to have the administrative tool support a batch processing mode.
A receiver attempting to verify a DKIM signature obtains the public key that is associated with the signature for that message. The DKIM-Signature: header in the message contains the d= tag with the basic domain name doing the signing and serving as output to the Identity Assessor, and the s= tag with the selector that is added to the name, for finding the specific public key. Hence, the relevant <selector>._domainkey.<domain-name> DNS record needs to contain a DKIM-related RR that provides the public key information.
The administrator of the zone containing the relevant domain name adds this information. Initial DKIM DNS information is contained within TXT RRs. DNS administrative software varies considerably in its abilities to support DKIM names, such as with underscores, and to add new types of DNS information.
The module doing signing can be placed anywhere within an organization's trusted Administrative Management Domain (ADMD); obvious choices include department-level posting agents, as well as outbound boundary MTAs to the open Internet. However any other module, including the author's MUA, is potentially acceptable, as long as the signature survives any remaining handling within the ADMD. Hence the choice among the modules depends upon software development, administrative overhead, security exposures and transit handling tradeoffs. One perspective that helps to resolve this choice is the difference between the increased flexibility, from placement at (or close to) the MUA, versus the streamlined administration and operation, that is more easily obtained by implementing the mechanism "deeper" into the organization's email infrastructure, such as at its boundary MTA.
Note the discussion in Section 2.2, concerning use of the i= tag.
The signing module uses the appropriate private key to create one or more signatures. (See Section 6.5 for a discussion of multiple signatures.) The means by which the signing module obtains the private key(s) is not specified by DKIM. Given that DKIM is intended for use during email transit, rather than for long-term storage, it is expected that keys will be changed regularly. For administrative convenience, it is best not to hard-code key information into software.
Every organization (ADMD) will have its own policies and practices for deciding when to sign messages (message stream) and with what domain name, selector and key. Examples of particular message streams include all mail sent from the ADMD, versus mail from particular types of user accounts, versus mail having particular types of content. Given this variability, and the likelihood that signing practices will change over time, it will be useful to have these decisions represented through run-time configuration information, rather than being hard-coded into the signing software.
As noted in Section 2.3, the choice of signing name granularity requires balancing administrative convenience and utility for recipients. Too much granularity is higher administrative overhead and well might attempt to impose more differential analysis on the recipient than they wish to support. In such cases, they are likely to use only a super-name -- right-hand substring -- of the signing name. When this occurs, the signer will not know what portion is being used; this then moves DKIM back to the non-deterministic world of heuristics, rather than the mechanistic world of signer/recipient collaboration that DKIM seeks.
A message recipient can verify a DKIM signature to determine if a claim of responsibility has been made for the message by a trusted domain.
Access control requires two components: authentication and authorization. By design, verification of a DKIM signature only provides the authentication component of an access control decision and needs to be combined with additional sources of information such as reputation data to arrive at an access control decision.
DKIM requires that a message with a signature that is found to be invalid is to be treated as if the message had not been signed at all.
If a DKIM signature fails to verify, it is entirely possible that the message is valid and that either there is a configuration error in the signer's system (e.g. a missing key record) or that the message was inadvertently modified in transit. It is thus undesirable for mail infrastructure to treat messages with invalid signatures less favorably than those with no signatures whatsoever. Contrariwise, creation of an invalid signature requires a trivial amount of effort on the part of an attacker. If messages with invalid signatures were to be treated preferentially to messages with no signatures whatsoever, attackers will simply add invalid signature blocks to gain the preferential treatment. It follows that messages with invalid signatures need to be treated no better and no worse than those with no signature at all.
As with any other digital signature scheme, verifiers need to consider only the part of the message that is inside the scope of the message as being authenticated by the signature.
For example, if the l= option is employed to specify a content length for the scope of the signature, only the part of the message that is within the scope of the content signature would be considered authentic.
Public Key cryptography provides an exceptionally high degree of assurance, bordering on absolute certainty, that the party that created a valid digital signature had access to the private key corresponding to the public key indicated in the signature.
In order to make useful conclusions from the verification of a valid digital signature, the verifier is obliged to make assumptions that fall far short of absolute certainty. Consequently, mere validation of a DKIM signature does not represent proof positive that a valid claim of responsibility was made for it by the indicated party, that the message is authentic, or that the message is not abusive. In particular:
In practice these limitations have little or no impact on the field of use for which DKIM is designed but can have a bearing if use is made of the DKIM message signature format or key retrieval mechanism in other specifications.
In particular the DKIM key retrieval mechanism is designed for ease of use and deployment rather than to provide a high assurance Public Key Infrastructure suitable for purposes that require robust non-repudiation such as establishing legally binding contracts. Developers seeking to extend DKIM beyond its design application needs to consider replacing or supplementing the DNS key retrieval mechanism with one that is designed to meet the intended purposes.
DKIM is frequently employed in a mail filtering strategy to avoid performing content analysis on email originating from trusted sources. Messages that carry a valid DKIM signature from a trusted source can be whitelisted, avoiding the need to perform computation and hence energy intensive content analysis to determine the disposition of the message.
Mail sources can be determined to be trusted by means of previously observed behavior and/or reference to external reputation or accreditation services. The precise means by which this is accomplished is outside the scope of DKIM.
Adaptive (or learning) spam filtering mechanisms that are not capable of verifying DKIM signatures need to, at minimum, be configured to ignore DKIM header data entirely.
Intermediaries such as mailing lists pose a particular challenge for DKIM implementations, as the message processing steps performed by the intermediary can cause the message content to change in ways that prevent the signature passing verification.
Such intermediaries are strongly encouraged to deploy DKIM signing so that a verifiable claim of responsibility remains available to parties attempting to verify the modified message.
In many deployments it is desirable to separate signature verification from the application relying on the verification. A system can choose to relay information indicating the results of its message authentication efforts using various means; adding a "results header" to the message is one such mechanism. [RFC5451] For example, consider the cases where:
In such cases it is important that the communication link between the signature verifier and the relying application be sufficiently secure to prevent insertion of a message that carries a bogus results header.
An intermediary that generates results headers need to ensure that relying applications are able to distinguish valid results headers issued by the intermediary from those introduced by an attacker. For example, this can be accomplished by signing the results header. At a minimum, results headers on incoming messages need to be removed if they purport to have been issued by the intermediary but cannot be verified as authentic.
Further discussion on trusting the results as relayed from a verifier to something downstream can be found in [RFC5451]
As described in section Section 2.1, a DKIM signature tells the signature verifier that the owner of a particular domain name accepts some responsibility for the message. It does not, in and of itself, provide any information about the trustworthiness or behavior of that identity. What it does provide is a verified identity to which such behavioral information can be associated, so that those who collect and use such information can be assured that it truly pertains to the identity in question.
This section lays out a taxonomy of some of the different identities, or combinations of identities, that might usefully be represented by a DKIM signature.
Perhaps the simplest case is when an organization signs its own outbound email using its own domain in the SDID [RFC5672] of the signature. For example, Company A would sign the outbound mail from its employees with d=companyA.example.
In the most straightforward configuration, the addresses in the RFC5322.From would also be in the companyA.example domain, but that direct correlation is not required.
A special case of the Single Domain Signature is an Author Signature as defined by the Author Domain Signing Practices specification [RFC5617]. Author signatures are signatures from an author's organization that have an SDID value that matches that of an RFC5322.From address of the signed message.
Although an author signature might in some cases be proof against spoofing the domain name of the RFC5322.From address, it is important to note that the DKIM and ADSP validation apply only to the exact address string and not to look-alike addresses nor to the human-friendly "display-name" or names and addresses used within the body of the message. That is, it protects only against the misuse of a precise address string within the RFC5322.From field and nothing else. For example, a message from bob@domain.example with a valid signature where d=d0main.example would fail an ADSP check because the signature domain, however similar, is distinct; however a message from bob@d0main.example with a valid signature where d=d0main.example would pass an ADSP check, even though to a human it might be obvious that d0main.example is likely a malicious attempt to spoof the domain domain.example. This example highlights that ADSP, like DKIM, is only able to validate a signing identifier: it still requires some external process to attach a meaningful reputation to that identifier.
Another approach that might be taken by an organization with multiple active subdomains is to apply the same (single) signature domain to mail from all subdomains. In this case, the signature chosen would usually be the signature of a parent domain common to all subdomains. For example, mail from marketing.domain.example, sales.domain.example, and engineering.domain.example might all use a signature where d=domain.example.
This approach has the virtue of simplicity, but it is important to consider the implications of such a choice. As discussed in Section 2.3, if the type of mail sent from the different subdomains is significantly different or if there is reason to believe that the reputation of the subdomains would differ, then it can be a good idea to acknowledge this and provide distinct signatures for each of the subdomains (d=marketing.domain.example, sales.domain.example, etc.). However, if the mail and reputations are likely to be similar, then the simpler approach of using a single common parent domain in the signature can work well.
Another approach to distinguishing the streams using a single DKIM key would be to leverage the AUID [RFC5672] (i= tag) in the DKIM signature to differentiate the mail streams. For example, marketing email would be signed with i=@marketing.domain.example and d=domain.example.
It's important to remember, however, that under core DKIM semantics the AUID is opaque to receivers. That means that it will only be an effective differentiator if there is an out of band agreement about the i= semantics.
A signature whose domain does not match the domain of the RFC5322.From address is sometimes referred to as a third party signature. In certain cases even the parent domain signature described above would be considered a third party signature because it would not be an exact match for the domain in the RFC5322.From address.
Although there is often heated debate about the value of third party signatures, it is important to note that the DKIM specification attaches no particular significance to the identity in a DKIM signature. The identity specified within the signature is the identity that is taking responsibility for the message, and it is only the interpretation of a given receiver that gives one identity more or less significance than another. In particular, most independent reputation services assign trust based on the specific identifier string, not its "role": in general they make no distinction between, for example, an author signature and a third party signature.
For some, a signature unrelated to the author domain (the domain in the RFC5322.From address) is less valuable because there is an assumption that the presence of an author signature guarantees that the use of the address in the RFC5322.From header is authorized.
For others, that relevance is tied strictly to the recorded behavioral data assigned to the identity in question, i.e. its trust assessment or reputation. The reasoning here is that an identity with a good reputation is unlikely to maintain that good reputation if it is in the habit of vouching for messages that are unwanted or abusive; in fact, doing so will rapidly degrade its reputation so that future messages will no longer benefit from it. It is therefore low risk to facilitate the delivery of messages that contain a valid signature of a domain with a strong positive reputation, independent of whether or not that domain is associated with the address in the RFC5322.From header field of the message.
Third party signatures encompass a wide range of identities. Some of the more common are:
For most of the cases described so far, there has been an assumption that the signing agent was responsible for creating and maintaining its own DKIM signing infrastructure, including its own keys, and signing with its own identity.
A different model arises when an organization uses a trusted third party sender for certain key business functions, but still wants that email to benefit from the organization's own identity and reputation: in other words, the mail would come out of the trusted third party's mail servers, but the signature applied would be that of the controlling organization.
This can be done by having the third party generate a key pair that is designated uniquely for use by that trusted third party and publishing the public key in the controlling organization's DNS domain, thus enabling the third party to sign mail using the signature of the controlling organization. For example, if Company A outsources its employee benefits to a third party, it can use a special key pair that enables the benefits company to sign mail as "companyA.example". Because the key pair is unique to that trusted third party, it is easy for Company A to revoke the authorization if necessary by simply removing the public key from the companyA.example DNS.
A more cautious approach would be to create a dedicated subdomain (e.g. benefits.companyA.example) to segment the outsourced mail stream, and to publish the public key there; the signature would then use d=benefits.companyA.example.
Another possibility for configuring trusted third party access, as discussed in section 3.4, is to have Company A use DNS delegation and have the designated subdomain managed directly by the trusted third party. In this case, Company A would create a subdomain benefits.companya.example, and delegate the DNS management of that subdomain to the benefits company so it could maintain its own key records. When revocation becomes necessary, Company A could simply remove the DNS delegation record.
A simple configuration for DKIM-signed mail is to have a single signature on a given message. This works well for domains that manage and send all of their own email from single sources, or for cases where multiple email streams exist but each has its own unique key pair. It also represents the case in which only one of the participants in an email sequence is able to sign, no matter whether it represents the author or one of the operators.
The examples thus far have considered the implications of using different identities in DKIM signatures, but have used only one such identity for any given message. In some cases, it can make sense to have more than one identity claiming responsibility for the same message.
There are a number of situations where applying more than one DKIM signature to the same message might make sense. A few examples are:
One important caveat to the use of multiple signatures is that there is currently no clear consensus among receivers on how they plan to handle them. The opinions range from ignoring all but one signature (and the specification of which of them is verified differs from receiver to receiver), to verifying all signatures present and applying a weighted blend of the trust assessments for those identifiers, to verifying all signatures present and simply using the identifier that represents the most positive trust assessment. It is likely that the industry will evolve to accept multiple signatures using either the second or third of these, but it can take some time before one approach becomes pervasive.
Signatures are created by different types of email actors, based on different criteria, such as where the actor operates in the sequence from author to recipient, whether they want different messages to be evaluated under the same reputation or a different one, and so on. This section provides some examples of usage scenarios for DKIM deployments; the selection is not intended to be exhaustive, but to illustrate a set of key deployment considerations.
The simplest DKIM configuration is to have some mail from a given organization (Company A) be signed with the same d= value (e.g. d=companya.example). If there is a desire to associate additional information, the AUID [RFC5672] value can become uniqueID@companya.example, or @uniqueID.companya.example.
In this scenario, Company A need only generate a single signing key and publish it under their top level domain (companya.example); the signing module would then tailor the AUID value as needed at signing time.
A slight variation of the one signature case is where Company A signs some of its mail, but it wants to differentiate different categories of its outbound mail by using different identifiers. For example, it might choose to distinguish marketing mail, billing or transactional mail, and individual corporate email into marketing.companya.example, billing.companya.example, and companya.example, where each category is assigned a unique subdomain and unique signing keys.
Some domains might decide to sign all of their outgoing mail. If all of the legitimate mail for a domain is signed, recipients can be more aggressive in their filtering of mail that uses the domain but does not have a valid signature from the domain; in such a configuration, the absence of a signature would be more significant than for the general case. It might be desirable for such domains to be able to advertise their intent to other receivers: this is the topic of Author Domain Signing Practices (ADSP).
Note that ADSP is not for everyone. Sending domains that do not control all legitimate outbound mail purporting to be from their domain (i.e., with an RFC5322.From address in their domain) are likely to experience delivery problems with some percentage of that mail. Administrators evaluating ADSP for their domains needs to carefully weigh the risk of phishing attacks against the likelihood of undelivered mail.
This section covers some examples of ADSP usage: for the complete specification, see [RFC5617]
In the ADSP specification, an address in the RFC5322.From header field of a message is defined as an "Author Address", and an "Author Domain" is defined as anything to the right of the '@' in an Author Address.
An "Author Signature" is thus any valid signature where the value of the SDID matches an Author Domain in the message.
It is important to note that unlike the DKIM specification which makes no correlation between the signature domain and any message headers, the ADSP specification applies only to the author domain. In essence, under ADSP, any non-author signatures are ignored (treated as if they are not present).
Signers wishing to publish an Author Domain Signing Practices (ADSP) [RFC5617] record describing their signing practices will thus want to include an author signature on their outbound mail to avoid ADSP verification failures.
An organization (Company A) can specify its signing practices by publishing an ADSP record with "dkim=all" or "dkim=discardable". In order to avoid misdelivery of its mail at receivers that are validating ADSP, Company A needs to first have done an exhaustive analysis to determine all sources of outbound mail from its domain (companyA.example) and ensure that they all have valid author signatures from that domain.
For example, email with an RFC5322.From address of bob@companyA.example needs to have an author signature where the SDID value is "companyA.example" or it will fail an ADSP validation.
Note that once an organization publishes an ADSP record using dkim=all or dkim=discardable, any email with an RFC5322.From address that uses the domain where the ADSP record is published that does not have a valid author signature is at risk of being misdelivered or discarded. For example, if a message with an RFC5322.From address of newsletter@companyA.example has a signature with d=marketing.companyA.example, that message will fail the ADSP check because the signature would not be considered a valid author signature.
Because the semantics of an ADSP author signature are more constrained than the semantics of a "pure" DKIM signature, it is important to make sure the nuances are well understood before deploying an ADSP record. The ADSP specification [RFC5617] provides some fairly extensive lookup examples (in Appendix A) and usage examples (in Appendix B).
In particular, in order to prevent mail from being negatively impacted or even discarded at the receiver, it is essential to perform a thorough survey of outbound mail from a domain before publishing an ADSP policy of anything stronger than "unknown". This includes mail that might be sent from external sources that might not be authorized to use the domain signature, as well as mail that risks modification in transit that might invalidate an otherwise valid author signature (e.g. mailing lists, courtesy forwarders, and other paths that could add or modify headers, or modify the message body).
An organization might choose to outsource certain key services to an independent company. For example, Company A might outsource its benefits management, or Organization B might outsource its marketing email.
If Company A wants to ensure that all of the mail sent on its behalf through the benefits providers email servers shares the Company A reputation, as discussed in Section 6.4 it can either publish keys designated for the use of the benefits provider under companyA.example (preferably under a designated subdomain of companyA.example), or it can delegate a subdomain (e.g. benefits.companyA.example) to the provider and enable the provider to generate the keys and manage the DNS for the designated subdomain.
In both of these cases, mail would be physically going out of the benefit provider's mail servers with a signature of e.g. d=benefits.companya.example. Note that the RFC5322.From address is not constrained: it could either be affiliated with the benefits company (e.g. benefits-admin@benefitprovider.example, or benefits-provider@benefits.companya.example), or with the companyA domain.
Note that in both of the above scenarios, as discussed in Section 3.4, security concerns dictate that the keys be generated by the organization that plans to do the signing so that there is no need to transfer the private key. In other words, the benefits provider would generate keys for both of the above scenarios.
Another way to manage the service provider configuration would be to have the service provider sign the outgoing mail on behalf of its client Company A with its own (provider) identifier. For example, an Email Service Provider (ESP A) might want to share its own mailing reputation with its clients, and might sign all outgoing mail from its clients with its own d= domain (e.g. d=espa.example).
When the ESP want to distinguish among its clients, it has two options:
Note that this scenario and the delegation scenario are not mutually exclusive: in some cases, it can be desirable to sign the same message with both the ESP and the ESP client identities.
An ISP (ISP A) might want to assign signatures to outbound mail from its users according to each user's past sending behavior (reputation). In other words, the ISP would segment its outbound traffic according to its own assessment of message quality, to aid recipients in differentiating among these different streams. Since the semantics of behavioral assessments are not valid AUID values, ISP A (ispa.example) can configure subdomains corresponding to the assessment categories (e.g. good.ispa.example, neutral.ispa.example, bad.ispa.example), and use these subdomains in the d= value of the signature.
The signing module can also set the AUID value to have a unique user id (distinct from the local-part of the user's email address), for example user3456@neutral.domain.example. Using a userid that is distinct from a given email alias is useful in environments where a single user might register multiple email aliases.
Note that in this case the AUID values are only partially stable. They are stable in the sense that a given i= value will always represent the same identity, but they are unstable in the sense that a given user can migrate among the assessment subdomains depending on their sending behavior (i.e., the same user might have multiple AUID values over the lifetime of a single account).
In this scenario, ISP A can generate as many keys as there are assessment subdomains (SDID values), so that each assessment subdomain has its own key. The signing module would then choose its signing key based on the assessment of the user whose mail was being signed, and if desired include the user id in the AUID of the signature. As discussed earlier, the per-user granularity of the AUID can be ignored by verifiers; so organizations choosing to use it ought not rely on its use for receiver side filtering results. However, some organizations might also find the information useful for their own purposes in processing bounces or abuse reports.
Another scenario is that of an agent, usually a re-mailer of some kind, that signs on behalf of the service or organization that it represents. Some examples of agents might be a mailing list manager, or the "forward article to a friend" service that many online publications offer. In most of these cases, the signature is asserting that the message originated with, or was relayed by, the service asserting responsibility. In general, if the service is configured in such a way that its forwarding would break existing DKIM signatures, it needs to always add its own signature.
The robustness of DKIM's verification mechanism is based on the fact that only authorized signing modules have access to the designated private key. This has the side effect that email submission and delivery scenarios that originate or relay messages from outside the domain of the authorized signing module will not have access to that protected private key, and thus will be unable to attach the expected domain signature to those messages. Such scenarios include mailing lists, courtesy forwarders, MTAs at hotels, hotspot networks used by travelling users, and other paths that could add or modify headers, or modify the message body.
For example, assume Joe works for Company A and has an email address joe@companya.example. Joe also has an ISP-1 account joe@isp1.example.com, and he uses ISP-1's multiple address feature to attach his work email address, joe@companya.example, to email from his ISP-1 account. When Joe sends email from his ISP-1 account and uses joe@companya.example as his designated RFC5322.From address, that email cannot have a signature with d=companya.example because the ISP-1 servers have no access to Company A's private key. In ISP-1's case it will have an ISP-1 signature, but for some other mail clients offering the same multiple address feature there might be no signature at all on the message.
Another example might be the use of a forward article to a friend service. Most instances of these services today allow someone to send an article with their email address in the RFC5322.From to their designated recipient. If Joe used either of his two addresses (joe@companya.example or joe@isp1.example.com), the forwarder would be equally unable to sign with a corresponding domain. As in the mail client case, the forwarder can either sign as its own domain, or can put no signature on the message.
A third example is the use of privately configured forwarding. Assume that Joe has another account at ISP-2, joe@isp-2.example.com, but he'd prefer to read his ISP-2 mail from his ISP-1 account. He sets up his ISP-2 account to forward all incoming mail to joe@isp1.example.com. Assume alice@companyb.example sends joe@isp-2.example.com an email. Depending on how companyb.example configured its signature, and depending on whether or not ISP-2 modifies messages that it forwards, it is possible that when Alice's message is received in Joe's ISP-1 account the original signature fails verification.
One identity is particularly amenable to easy and accurate assessment: the organization's own identity. Members of an organization tend to trust messages that purport to be from within that organization. However Internet Mail does not provide a straightforward means of determining whether such mail is, in fact, from within the organization. DKIM can be used to remedy this exposure. If the organization signs all of its mail, then its boundary MTAs can look for mail purporting to be from the organization that does not contain a verifiable signature.
Such mail can in most cases be presumed to be spurious. However, domain managers are advised to consider the ways that mail processing can modify messages in ways that will invalidate an existing DKIM signature: mailing lists, courtesy forwarders, and other paths that could add or modify headers or modify the message body (e.g. MTAs at hotels, hotspot networks used by travelling users, and other scenarios described in the previous section). Such breakage is particularly relevant in the presence of Author Domain Signing Practices.
Although DKIM's use of domain names is optimized for a scope of organization-level signing, it is possible to administer sub-domains or otherwise adjust signatures in a way that supports per-user identification. This user level granularity can be specified in two ways: either by sharing the signing identity and specifying an extension to the i= value that has a per-user granularity, or by creating and signing with unique per-user keys.
A subdomain or local part in the i= tag needs to be treated as an opaque identifier and thus need not correspond directly to a DNS subdomain or be a specific user address.
The primary way to sign with per-user keys requires each user to have a distinct DNS (sub)domain, where each distinct d= value has a key published. (It is possible, although not advised, to publish the same key in more than one distinct domain.)
It is technically possible to publish per-user keys within a single domain or subdomain by utilizing different selector values. This is not advised and is unlikely to be treated uniquely by Assessors: the primary purpose of selectors is to facilitate key management, and the DKIM specification recommends against using them in determining or assessing identities.
In most cases, it would be impractical to sign email on a per-user granularity. Such an approach would be
As mentioned before, what might make sense, however, is to use the infrastructure that enables finer granularity in signatures to identify segments smaller than a domain but much larger than a per-user segmentation. For example, a university might want to segment student, staff, and faculty mail into three distinct streams with differing reputations. This can be done by creating separate sub-domains for the desired segments, and either specifying the subdomains in the i= tag of the DKIM Signature or by adding subdomains to the d= tag and assigning and signing with different keys for each subdomain.
For those who choose to represent user level granularity in signatures, the performance and management considerations above suggest that it would be more effective to do it by specifying a local part or subdomain extension in the i= tag rather than by extending the d= domain and publishing individual keys.
It is expected that the most common venue for a DKIM implementation will be within the infrastructure of an organization's email service, such as a department or a boundary MTA. What follows are some general recommendations for the Email Infrastructure.
The DKIM specification is expected to be used primarily between Boundary MTAs, or other infrastructure components of the originating and receiving ADMDs. However there is nothing in DKIM that is specific to those venues. In particular, MUAs can also support DKIM signing and verifying directly.
It is common for components of an ADMD's email infrastructure to do violence to a message, such that a DKIM signature might be rendered invalid. Hence, users of MUAs that support DKIM signing and/or verifying need a basis for knowing that their associated email infrastructure will not break a signature.
The security considerations of the DKIM protocol are described in the DKIM base specification [RFC4871].
This document has no considerations for IANA.
The effort of the DKIM Working Group is gratefully acknowledged.
| [RFC4871] | Allman, E., Callas, J., Delany, M., Libbey, M., Fenton, J., and M. Thomas, “DomainKeys Identified Mail (DKIM) Signatures”, RFC 4871, May 2007. |
| [RFC5322] | Resnick, P., Ed., “Internet Message Format”, RFC 5322, October 2008. |
| [RFC5451] | Kucherawy, M., “Message Header Field for Indicating Message Authentication Status”, RFC 5451, April 2009. |
| [RFC5585] | Hansen, T., Crocker, D., and P. Hallam-Baker, “DomainKeys Identified Mail (DKIM) Service Overview”, RFC 5585, July 2009. |
| [RFC5617] | Allman, E., Fenton, J., Delany, M., and J. Levine, “DomainKeys Identified Mail (DKIM) Author Domain Signing Practices (ADSP)”, RFC 5617, August 2009. |
| [RFC5672] | Crocker, D., “RFC 4871 DomainKeys Identified Mail (DKIM) Signatures -- Update”, RFC 5672, August 2009. |
| [RFC4034] | Arends, R., Austein, R., Larson, M., Massey, D., and S. Rose, “Resource Records for the DNS Security Extensions”, RFC 4034, March 2005. |
| [RFC4870] | Delany, M., “Domain-Based Email Authentication Using Public Keys Advertised in the DNS (DomainKeys)”, RFC 4870, May 2007. |
| [RFC5155] | Laurie, B., Sisson, G., Arends, R., and D. Blacka, “DNS Security (DNSSEC) Hashed Authenticated Denial of Existence”, RFC 5155, March 2008. |
There are three migration occasions worth noting in particular for DKIM:
The case of deploying a new key selector record is described elsewhere (Section 3.5).
As with any migration, the steps required will be determined by who is doing the migration and their assessment of
Signers and verifiers have different considerations.
DKIM replaces the earlier DomainKeys (DK) specification. Selector files are mostly compatible between the two specifications.
A signer that currently signs with DK will go through various stages as it migrates to using DKIM, not all of which are required for all signers. The real questions that a signer needs to ask are:
If no one is looking at the DK signature any more, then it's no longer necessary to sign with DK. Or if all "large players" are looking at DKIM in addition to or instead of DK, a signer can choose to stop signing with DK.
With respect to signing policies, a reasonable, initial approach is to use DKIM signatures in the same way as DomainKeys signatures are already being used. In particular, the same selectors and DNS Key Records can be used for both, after verifying that they are compatible as discussed below.
Each secondary step in all of the following scenarios is to be prefaced with the gating factor "test, then when comfortable with the previous step's results, continue".
One migration strategy is to:
Another migration strategy is to:
A combined migration strategy is to:
Another migration strategy is to:
Another migration strategy is to:
Note that when you have separate key records for DK and DKIM, you can use the same public key for both.
The first step in some of the above scenarios is ensuring that the selector DNS key records are compatible for both DK and DKIM. The format of the DNS key record was intentionally meant to be backwardly compatible between the two systems, but not necessarily upwardly compatible. DKIM has enhanced the DK DNS key record format by adding several optional parameters, which DK needs to ignore. However, there is one critical difference between DK and DKIM DNS key records: the definitions of the "g" fields:
The principal use of DomainKeys is at Boundary MTAs. Because no operational transition is ever instantaneous, it is advisable to continue performing DomainKeys signing until it is determined that DomainKeys receive-side support is no longer used, or is sufficiently reduced. That is, a signer needs to add a DKIM signature to a message that also has a DomainKeys signature and keep it there until they decide it is deemed no longer useful. The signer can do its transitions in a straightforward manner, or more gradually. Note that because digital signatures are not free, there is a cost to performing both signing algorithms, so signing with both algorithms ought not be needlessly prolonged.
The tricky part is deciding when DK signatures are no longer necessary. The real questions are: how many DomainKeys verifiers are there that do *not* also do DKIM verification, which of those are important, and how can you track their usage? Most of the early adopters of DK verification have added DKIM verification, but not all yet. If a verifier finds a message with both DK and DKIM, it can choose to verify both signatures, or just one or the other.
Many DNS services offer tracking statistics so it can be determined how often a DNS record has been accessed. By using separate DNS selector key records for your signatures, you can chart the usage of your records over time, and watch the trends. An additional distinguishing factor to track would take into account the verifiers that verify both the DK and DKIM signatures, and discount those from counts of DK selector usage. When the number for DK selector access reaches a low-enough level, that's the time to consider discontinuing signing with DK.
Note, this level of rigor is not required. It is perfectly reasonable for a DK signer to decide to follow the "flash cut" scenario described above.
As a verifier, several issues needs to be considered:
At the time of writing, there is still a significant number of sites that are only producing DK signatures. Over time, it is expected that this number will go to zero, but it might take several years. So it would be prudent for the foreseeable future for a verifier to look for and verify both DKIM and DK signatures.
For a period of time, there will be sites that sign with both DK and DKIM. A verifier receiving a message that has both types of signatures can verify both signatures, or just one. One disadvantage of verifying both signatures is that signers will have a more difficult time deciding how many verifiers are still using their DK selectors. One transition strategy is to verify the DKIM signature, then only verify the DK signature if the DKIM verification fails.
The format of the DNS key record was intentionally meant to be backwardly compatible between DK and DKIM, but not necessarily upwardly compatible. DKIM has enhanced the DK DNS key record format by adding several optional parameters, which DK needs to ignore. However, there is one key difference between DK and DKIM DNS key records: the definitions of the g fields:
If a DKIM verifier finds a selector record that has an empty "g" field ("g=;") and it does not have a "v" field ("v=DKIM1;") at its beginning, it is faced with deciding if this record was
[RFC4871] defines the use of two hash algorithms, SHA-1 and SHA-256. The security of all hash algorithms is constantly under attack, and SHA-1 has already shown weaknesses as of this writing. Migrating from SHA-1 to SHA-256 is not an issue, because all verifiers are already required to support SHA-256. But when it becomes necessary to replace SHA-256 with a more secure algorithm, there will be a migratory period. In the following, "NEWHASH" is used to represent a new hash algorithm. Section 4.1 of [RFC4871] briefly discusses this scenario.
As with migrating from DK to DKIM, migrating hash algorithms is dependent on the signer's best guess as to the utility of continuing to sign with the older algorithms and the expected support for the newer algorithm by verifiers. The utility of continuing to sign with the older algorithms is also based on how broken the existing hash algorithms are considered and how important that is to the signers.
One strategy is to wait until it's determined that there is a large enough base of verifiers available that support NEWHASH, and then flash cut to the new algorithm.
Another strategy is to sign with both the old and new hash algorithms for a period of time. This is particularly useful for testing the new code to support the new hash algorithm, as verifiers will continue to accept the signature for the older hash algorithm and ought to ignore any signature that fails because the code is slightly wrong. Once the signer has determined that the new code is correct AND it's determined that there is a large enough base of verifiers available that support NEWHASH, the signer can flash cut to the new algorithm.
One advantage migrating hash algorithms has is that the selector can be completely compatible for all hash algorithms. The key selector has an optional "h=" field that can be used to list the hash algorithms being used; it also is used to limit the algorithms that a verifier will accept. If the signer is not currently using the key-selector "h=" field, no change is required. If the signer is currently using the key-selector "h=" field, NEWHASH will need to be added to the list, as in "h=sha256:NEWHASH;". (When the signer is no longer using sha256, it can be removed from the "h=" list.)
When a new hash algorithm becomes standardized, it is best for a verifier to start supporting it as quickly as possible.
[RFC4871] defines the use of the RSA signing algorithm. Similar to hashes, signing algorithms are constantly under attack, and when it becomes necessary to replace RSA with a newer signing algorithm, there will be a migratory period. In the following, "NEWALG" is used to represent a new signing algorithm.
As with the other migration issues discussed above, migrating signing algorithms is dependent on the signer's best guess as to the utility of continuing to sign with the older algorithms and the expected support for the newer algorithm by verifiers. The utility of continuing to sign with the older algorithms is also based on how broken the existing signing algorithms are considered and how important that is to the signers.
As before, the two basic strategies are to 1) wait until there is sufficient base of verifiers available that support NEWALG and then do a flash cut to NEWALG, and 2) using a phased approach by signing with both the old and new algorithms before removing support for the old algorithm.
It is unlikely that a new algorithm would be able to use the same public key as "rsa", so using the same selector DNS record for both algorithms' keys is ruled out. Therefore, in order to use the new algorithm, a new DNS selector record would need to be deployed in parallel with the existing DNS selector record for the existing algorithm. The new DNS selector record would specify a different "k=" value to reflect the use of NEWALG.
When a new hash algorithm becomes standardized, it is best for a verifier to start supporting it as quickly as possible.
NOTE: This section could possibly be changed into a reference to something else, such as another rfc.
Correct implementation of a cryptographic algorithm is a necessary but not a sufficient condition for the coding of cryptographic applications. Coding of cryptographic libraries requires close attention to security considerations that are unique to cryptographic applications.
In addition to the usual security coding considerations, such as avoiding buffer or integer overflow and underflow, implementers need to pay close attention to management of cryptographic private keys and session keys, ensuring that these are correctly initialized and disposed of.
Operating system mechanisms that permit the confidentiality of private keys to be protected against other processes ought to be used when available. In particular, great care needs to be taken when releasing memory pages to the operating system to ensure that private key information is not disclosed to other processes.
Certain implementations of public key algorithms such as RSA can be vulnerable to a timing analysis attack.
Support for cryptographic hardware providing key management capabilities is strongly encouraged. In addition to offering performance benefits, many cryptographic hardware devices provide robust and verifiable management of private keys.
Fortunately appropriately designed and coded cryptographic libraries are available for most operating system platforms under license terms compatible with commercial, open source and free software license terms. Use of standard cryptographic libraries is strongly encouraged. These have been extensively tested, reduce development time and support a wide range of cryptographic hardware.