DMARC & Indirect Mail: Interoperability Issues between Domain-based Message Authentication, Reporting, and Conformance (DMARC) and Indirect Email Flows
Why strict DMARC policies break mailing lists, forwarders and other mail that does not travel straight from author to reader.
Martin, F., Ed., Lear, E., Ed., Draegen, T., Ed., Zwicky, E., Ed., and K. Andersen, Ed., "Interoperability Issues between Domain-based Message Authentication, Reporting, and Conformance (DMARC) and Indirect Email Flows", RFC 7960, DOI 10.17487/RFC7960, September 2016, <https://www.rfc-editor.org/info/rfc7960>.
@misc{rfc7960,
series = {Request for Comments},
number = 7960,
howpublished = {RFC 7960},
publisher = {RFC Editor},
doi = {10.17487/RFC7960},
url = {https://www.rfc-editor.org/info/rfc7960},
author = {F. Martin and E. Lear and T. Draegen and E. Zwicky and K. Andersen},
title = {{Interoperability Issues between Domain-based Message Authentication, Reporting, and Conformance (DMARC) and Indirect Email Flows}},
pagetotal = 27,
year = 2016,
month = sep,
} RFC text
Contents
Internet Engineering Task Force (IETF) F. Martin, Ed.
Request for Comments: 7960 LinkedIn
Category: Informational E. Lear, Ed.
ISSN: 2070-1721 Cisco Systems GmbH
T. Draegen, Ed.
dmarcian, inc.
E. Zwicky, Ed.
Yahoo
K. Andersen, Ed.
LinkedIn
September 2016
Interoperability Issues between Domain-based Message Authentication,
Reporting, and Conformance (DMARC) and Indirect Email Flows
Abstract
Domain-based Message Authentication, Reporting, and Conformance
(DMARC) introduces a mechanism for expressing domain-level policies
and preferences for email message validation, disposition, and
reporting. However, the DMARC mechanism enables potentially
disruptive interoperability issues when messages do not flow directly
from the author's administrative domain to the final Recipients.
Collectively, these email flows are referred to as "indirect email
flows". This document describes these interoperability issues and
presents possible methods for addressing them.
Status of This Memo
This document is not an Internet Standards Track specification; it is
published for informational purposes.
This document is a product of the Internet Engineering Task Force
(IETF). It represents the consensus of the IETF community. It has
received public review and has been approved for publication by the
Internet Engineering Steering Group (IESG). Not all documents
approved by the IESG are a candidate for any level of Internet
Standard; see Section 2 of RFC 7841.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
http://www.rfc-editor.org/info/rfc7960.
Copyright Notice
Copyright (c) 2016 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 Simplified BSD License.
Table of Contents
1. Introduction ....................................................3
1.1. Document Conventions .......................................4
2. Causes of Interoperability Issues ...............................4
2.1. Identifier Alignment .......................................4
2.1.1. DKIM Identifier(s) ..................................5
2.1.2. SPF Identifier(s) ...................................6
2.1.3. Multiple RFC5322.From Addresses .....................6
2.2. Message Forwarding .........................................6
2.3. Message Modification .......................................7
3. Internet Mail Architecture, DMARC, and Indirect Email Flows .....8
3.1. Message Handling System ....................................8
3.1.1. Message Submission Agents ...........................8
3.1.2. Message Transfer Agents .............................9
3.1.2.1. Message Encoding ...........................9
3.1.2.2. Header Standardization ....................10
3.1.2.3. Content Validation ........................10
3.1.3. Message Delivery Agents ............................10
3.2. Mediators .................................................11
3.2.1. Alias ..............................................11
3.2.2. ReSenders ..........................................12
3.2.3. Mailing Lists ......................................12
3.2.3.1. Mailing List Operational Effects ..........13
3.2.4. Gateways ...........................................13
3.2.5. Boundary Filters ...................................14
3.3. Combinations ..............................................15
4. Possible Mitigations of Interoperability Issues ................15
4.1. Mitigations in Current Use ................................16
4.1.1. Mitigations for Senders ............................16
4.1.1.1. Identifier Alignment ......................16
4.1.1.2. Message Modification ......................17
4.1.2. Mitigations for Receivers ..........................17
4.1.2.1. Identifier Alignment ......................17
4.1.2.2. Policy Override ...........................17
4.1.3. Mitigations for ReSenders ..........................18
4.1.3.1. Changes to the RFC5322.From ...............18
4.1.3.2. Avoiding Message Modification .............18
4.1.3.3. Mailing Lists .............................18
4.2. Proposed and In-Progress Mitigations ......................20
4.2.1. Getting More Radical: Requiring New
Communication Paths between MUAs ...................21
5. Security Considerations ........................................21
6. References .....................................................22
6.1. Normative References ......................................22
6.2. Informative References ....................................23
Appendix A. Example SPF Bounce ...................................24
A.1. Initial Message ...........................................24
A.2. Notification Message ......................................25
Acknowledgments ...................................................26
Authors' Addresses ................................................27
1. Introduction
DMARC [RFC7489] introduces a mechanism for expressing domain-level
policies and preferences for message validation, disposition, and
reporting. The DMARC mechanism, especially when employed with
restrictive policies, encounters several different types of
interoperability issues due to third-party message sourcing, message
transformation, or rerouting. In particular, DMARC with restrictive
policies causes problems for many Mailing Lists.
At the time of writing this document, the DMARC base specification
has been published as Informational RFC 7489 [RFC7489] and has seen
significant deployment within the email community.
Cases in which email does not flow directly from the author's
administrative domain to the recipient's domain(s) are collectively
referred to in this document as "indirect email flows". Due to
existing and increasing adoption of DMARC, the impact of DMARC-based
email rejection policies on indirect email flows can be significant
for a select subset of general email traffic.
Several known causes of interoperability issues are presented,
followed by a description of components within the Internet Mail
Architecture [RFC5598] where interoperability issues can arise.
Finally, known and possible methods for addressing interoperability
issues are presented. There are often multiple ways to address any
given interoperability issue. While this document strives to be
comprehensive in its review, it should not be treated as complete.
Note that some practices that are in use at the time of this document
may or may not be "best practices", especially as future standards
evolve.
1.1. Document Conventions
The notation used in this document for structured fields is taken
from [RFC5598], Section 1.3.
The term "notification message" ([RFC5321], Section 4.5.5) is used to
refer to a message with a null RFC5321.MailFrom.
The terms "Organizational Domain" and "Authenticated Identifiers" are
specified in DMARC ([RFC7489], Section 3).
All words that begin with capital letters take their formal meanings
from these references.
2. Causes of Interoperability Issues
Interoperability issues between DMARC and indirect email flows arise
when conformance to the DMARC specification leads a receiving
implementation to apply DMARC-based policy restrictions to messages
that are both compliant with the architecture as specified in
[RFC5598] and viewed as legitimate by the intended Recipient.
Note that domains that assert a "p=none" policy and email messages
that pass standard DMARC validation do not have any interoperability
issues.
Email messages that do not conform to IETF email specifications but
are considered legitimate by the intended Recipients are not
discussed in this document.
The rest of this section describes several conceptual causes of
interoperability issues.
2.1. Identifier Alignment
Note to operators and administrators: The identifiers that are used
by the Sender Policy Framework (SPF) are technical components of the
transport process for SMTP. They may or may not, as described below,
bear a meaningful relationship to the content or source of the
message itself. This "relationship by proximity" can be a point of
confusion for non-technical end users, either recipients or senders.
DMARC relies on DomainKeys Identified Mail (DKIM) [RFC6376] and SPF
[RFC7208] to perform message source validation. The DMARC [RFC7489]
specification refers to source domains that are validated by DKIM or
SPF as "Authenticated Identifiers". To be used by DMARC, an
"Authenticated Identifier" must also be related to the domain found
in the message's RFC5322.From header field [RFC5322]. This
relationship between an Authenticated Identifier's domain and the
domain of the RFC5322.From is referred to as "Identifier Alignment".
DMARC allows for Identifier Alignment to be determined in two
different modes: strict and relaxed. Strict mode requires an exact
match between the domain of any of the Authenticated Identifiers and
the message's RFC5322.From header field [RFC5322]. Relaxed mode
allows for Identifier Alignment if Authenticated Identifiers and the
message's RFC5322.From header field [RFC5322] share the same
Organizational Domain. In general, DMARC interoperability issues are
the same for both strict and relaxed alignment, but strict alignment
constrains the possible solutions because of the more rigorous
matching requirement. Some of the mitigations described in this
document only work with the relaxed mode of Identifier Alignment.
2.1.1. DKIM Identifier(s)
DKIM provides a cryptographic means for one or more domain
identifiers to be associated with a particular message. As a
standalone technology, DKIM identifiers are not required to be
related to the source of the message's content. However, for a DKIM
identifier to align in DMARC, the signing domain of a valid signature
must be part of the same Organizational Domain as the domain in the
RFC5322.From header field [RFC5322].
In addition, DKIM allows for the possibility of multiple valid
signatures. These multiple signatures may be from the same or
different domains; there are no restrictions within the DKIM
specification. The DMARC mechanism will process Authenticated
Identifiers that are based on DKIM signatures until an aligned
Authenticated Identifier is found (if any). However, operational
experience has shown that some implementations have difficulty
processing multiple signatures. Implementations that cannot process
multiple DKIM signatures may incorrectly flag messages as "not
passing" (DMARC alignment) and erroneously apply DMARC-based policy
to otherwise conforming messages.
2.1.2. SPF Identifier(s)
The SPF specification [RFC7208] defines two Authenticated Identifiers
for each message. These identifiers derive from:
a. the RFC5321.MailFrom [RFC5321] domain, and
b. the RFC5321.HELO/.EHLO SMTP domain.
In the SPF specification, the RFC7208.MAILFROM [RFC7208] value is
defined to be based on RFC5321.MailFrom unless that value is absent
(as in the case of notification messages), in which case, the second
(RFC5321.HELO/.EHLO) identifier value is used. This "fallback"
definition has occasionally been misunderstood by operators of MTA
systems since notification messages are often an "automatic" feature
of MTA software. Some MTA software does not provide the ability to
apply a DKIM signature to such notification messages.
See Appendix A for an example treatment of this scenario.
For the purposes of DMARC validation/alignment, the hybrid
RFC7208.MAILFROM [RFC7208] identifier's domain is used if and only if
it is aligned with the RFC5322.From [RFC5322] domain. The alignment
of the validated domain is determined based on the DMARC record's
"strict" or "relaxed" designation as described above for the DKIM
identifiers and in [RFC7489].
2.1.3. Multiple RFC5322.From Addresses
[RFC5322] permits only one From header field, but it may contain
multiple mailboxes. Since this is an extremely rare usage, DMARC
specifies that the handling of this situation is implementation
dependent.
Because the presence of multiple domains can be used by an attacker
(an attacker could add their domain to the RFC5322.From field,
provide arbitrary new content, and sign the message), the DMARC
specification recommends that the strictest policy be applied to such
messages (Section 6.6.1 of [RFC7489]).
2.2. Message Forwarding
Section 3 describes forwarding behavior as it relates to the
components of the Internet Mail Architecture.
All forwarding operations involve the retransmission of email. As
discussed above, in order for SPF to yield an Authenticated
Identifier that is pertinent to DMARC, the domain of the
RFC7208.MAILFROM must be in alignment with the RFC5322.From header
field. Forwarding introduces specific issues to the availability of
SPF-based Authenticated Identifiers:
o If the RFC5321.MailFrom is present and the forwarder maintains the
original RFC5321.MailFrom, SPF validation will fail unless the
forwarder is an authorized part of the originator's email sending
infrastructure. If the forwarder replaces the RFC5321.MailFrom
with its own domain, SPF might pass, but Identifier Alignment with
the RFC5322.From header field will fail.
o If the RFC5321.MailFrom is empty (as in the case of Delivery
Status Notifications), the RFC5321.HELO/.EHLO domain of the
forwarder will likely be in a different Organizational Domain than
the original RFC5322.From header field's domain. SPF may pass,
but Identifier Alignment with the RFC5322.From header field will
fail.
In both cases, SPF cannot yield relevant Authenticated Identifiers,
and DKIM must be relied upon to produce results that are relevant to
DMARC.
2.3. Message Modification
Modification of email content invalidates most DKIM signatures, and
many message-forwarding systems modify email content. Mailing List
processors are a common example of such systems, but other forwarding
systems also make modifications.
Although DKIM provides a length flag so that content can be appended
without invalidating the signature, in practice, particularly with
MIME-encoded [RFC2045] messages, a Mailing List processor will do
more than simply append content (see Section 5.3 of [RFC5598] for
details). Furthermore, the length flag is seldom used due to
security issues (see Section 8.2 of [RFC6376] for additional security
considerations). Therefore, this method is only mentioned here for
completeness.
DKIM describes two canonicalizations for use when preparing the
header and body for DKIM processing: simple and relaxed. The latter
is designed to accommodate trivial modifications to whitespace and
folding that, at least in the header case, generally have no semantic
significance. However, the relaxed canonicalization is more
computationally intensive and may not have been preferred in the
early deployment of DKIM, leaving some deployments using the less
forgiving "simple" canonicalization. While the prevalence is
unknown, there are some DKIM verifiers that have problems evaluating
relaxed canonicalization correctly.
3. Internet Mail Architecture, DMARC, and Indirect Email Flows
This section describes components within the Internet Mail
Architecture [RFC5598] where interoperability issues between DMARC
and indirect email flows can be found.
3.1. Message Handling System
Section 4 of [RFC5598] describes six basic components that make up
the Message Handling System (MHS):
o Message
o Message User Agent (MUA)
o Message Submission Agent (MSA)
o Message Transfer Agent (MTA)
o Message Delivery Agent (MDA)
o Message Store (MS)
Of these components, MSA, MTA, and MDA are discussed in relation to
interoperability with DMARC.
Section 5 of [RFC5598] also defines a Mediator as a hybrid of several
component types. A Mediator is given special consideration in this
section due to the unique issues they face when attempting to
interoperate with DMARC.
3.1.1. Message Submission Agents
An MSA accepts messages submitted by a Message User Agent (MUA) and
enforces the policies of the hosting ADministrative Management Domain
(ADMD) and the requirements of Internet standards.
MSAs are split into two sub-components:
o Author-focused MSA functions (aMSA)
o MHS-focused MSA functions (hMSA)
MSA interoperability issues with DMARC begin when an aMSA accepts a
message where the RFC5322.From header field contains a domain that is
outside of the ADMD of the MSA. This situation manifests in one of
several ways, such as when someone uses a mail service with their own
domain but has failed to properly configure an SPF record or when an
MUA attempts to transmit mail as someone else. Examples of the
latter situation include "forward-to-friend" functionality commonly
found on news/article websites or "send-as" functionality present on
some MUAs.
When an hMSA takes responsibility for transit of a message containing
a domain in the RFC5322.From header field that is outside of the
hMSA's ADMD, the hMSA faces DMARC interoperability issues if the
domain publishes a DMARC policy of "quarantine" or "reject". These
issues are marked by the inherent difficulty of establishing
alignment with the domain present in a message's RFC5322.From header
field. Examples of this issue include:
o Partially open relays - a residential ISP that allows its
customers to relay non-local domains through its infrastructure.
o Embedded devices - cable/DSL modems, firewalls, wireless access
points, and printers that send email using hardcoded domains.
o Devices that send mail on behalf of a user - scanners, security
cameras, and alarms that send mail as their owner or a device
user.
o Email service providers - ESPs that service customers who are
using domains that publish a DMARC "reject" policy.
o Calendaring software - an invited member of an event modifies the
event, causing the calendaring software to emit an update that
claims to come from the creator of the event.
3.1.2. Message Transfer Agents
MTAs relay a message until the message reaches a destination MDA.
Some common message-handling strategies break the integrity of DKIM
signatures. A restrictive DMARC policy along with a broken DKIM
signature causes latent interoperability problems to become overt
problems.
3.1.2.1. Message Encoding
An MTA may modify the message encoding, for instance, by converting
8-bit MIME sections to quoted-printable 7-bit sections. This
modification is outside the scope of DKIM canonicalization and will
invalidate DKIM signatures that include message content.
An MTA could also re-encode the message without changing the encoding
type: receiving a MIME-encoded message and producing a semantically
and semiotically equivalent MIME body that is not identical to the
original. This is characteristic of systems that use some other
message representation internally.
3.1.2.2. Header Standardization
An MTA may rewrite headers to bring them into compliance with
existing RFCs. For example, some common MTAs will correct
comprehensible but non-compliant date formats to compliant ones.
Header rewriting is outside the scope of DKIM canonicalization and
will invalidate DKIM signatures. All downstream DMARC processing
with be unable to utilize DKIM to yield Authenticated Identifiers due
to header rewriting.
Providing solutions for issues relating to non-RFC-compliant emails
is outside the scope of this document.
3.1.2.3. Content Validation
An MTA may also implement security-motivated changes to the content
of email messages, dropping or altering sections of messages, causing
breakage of DKIM signatures.
3.1.3. Message Delivery Agents
The MDA transfers a message from the MHS to a mailbox. Like the MSA,
the MDA consists of two sub-components:
o MHS-focused MDA functions (hMDA)
o Recipient-focused MDA functions (rMDA)
Both the hMDA and the rMDA can redirect a message to an alternative
address. DMARC interoperability issues related to redirecting of
messages are described in Section 3.2.
Sieve [RFC5228] functionality often lives in the rMDA sub-component
and can cause DMARC interoperability issues. The Sieve 'addheader'
and 'deleteheader' filtering actions can modify messages and
invalidate DKIM signatures, removing DKIM-supplied Authenticated
Identifiers as inputs to the DMARC mechanism. There are also Sieve
extensions [RFC5703] that modify the body. Sieve alterations may
only become an issue when the email is reintroduced into the
transport infrastructure.
3.2. Mediators
Mediators [RFC5598] forward messages through a re-posting process.
Mediators share some functionality with basic MTA relaying but have
greater flexibility in both addressing and content modifications.
DMARC interoperability issues are common within the context of
Mediators, which are often used precisely for their ability to modify
messages.
The DMARC design does not cope with some Mediator functionality such
as content modifications that invalidate DKIM signatures and
RFC5321.MailFrom rewriting to support SPF authentication of re-sent
mail when the new Recipient receives the message from the Mediator
rather than the initial organization.
3.2.1. Alias
An Alias is a simple re-addressing facility that provides one or more
new Internet Mail addresses, rather than a single, internal one. A
message continues through the transfer service for delivery to one or
more alternative addresses.
Aliases can be implemented by mailbox-level forwarding (e.g., through
"dot-forwarding"), Sieve-level forwarding (through the Sieve
"redirect" action), or other methods. When an Alias preserves
message content and does not make significant header changes, DKIM
signatures may remain valid. However, Aliases often extend the
delivery path outside of the scope covered by the originating ADMD's
SPF record(s).
Examples of Aliasing include:
o Forwarding email between free email (freemail) providers to try
different interfaces while maintaining an original email address;
o Consolidating many email addresses into a single account to
centralize processing; and
o Services that provide "activity-based", "role-based", "vanity", or
"temporary" email addresses such as universities and professional
associations. For instance, professional or alumni institutions
may offer their members an Alias for the duration of their
membership but may not want to deal with the long-term storage of
emails.
In most cases, the aMSA providing Alias services has no
administrative relationship to the ADMD of the originator or the
Final Recipient, so solutions to Alias-related DMARC failure should
not assume such a relationship.
3.2.2. ReSenders
ReSenders "splice" a message's addressing information to connect the
Author of the original message with the Recipient(s) of the new
message. The new Recipient sees the message as being from the
original Author, even if the Mediator adds commentary.
Without Authenticated Identifiers aligned with the Author's
RFC5322.From header field domain, the new Recipient has no way to
achieve a passing DMARC evaluation.
Examples of ReSenders include MUA-level forwarding by resending a
message to a new Recipient or by forwarding a message "inline" to a
new Recipient (this does not include forwarding a message "as an
attachment"). An additional example comes in the form of calendaring
software that allows a meeting attendee (not the meeting organizer)
to modify the content of an invite generating new invitations that
claim to be reissued from the meeting organizer.
3.2.3. Mailing Lists
A Mailing List receives messages as an explicit addressee and then
reposts them to a list of subscribed members. The Mailing List
performs a task that can be viewed as an elaboration of the ReSender
actions.
Mailing Lists share the same DMARC interoperability issues as
ReSenders (Section 3.2.2) and very commonly modify headers or message
content in ways that will cause DKIM to fail, including:
o prepending the RFC5322.Subject header field with a tag, to allow
the Recipient to easily identify the Mailing List within a subject
line listing;
o adding a footer to the email body to contain administrative
instructions;
o removing some MIME-parts from the email or converting the message
to text only;
o encrypting the body with the Recipient's key using PGP (Pretty
Good Privacy) or S/MIME;
o enforcing community standards by rewriting banned words; and
o allowing moderators to add arbitrary commentary to messages
(discussed in [RFC6377]).
Any such modifications would invalidate a DKIM signature.
Header and content modifications are common for many Mailing Lists
and are often central to present Mailing List functionality and
usage. Furthermore, MUAs have come to rely on Mailing List message
modifications to present messages to end users in expected ways.
3.2.3.1. Mailing List Operational Effects
Mailing Lists may also have the following DMARC interoperability
issues:
o Subscribed members may not receive email from members that post
using domains that publish a DMARC "p=reject" policy.
o Mailing Lists may interpret DMARC-related email rejections as an
inability to deliver email to the Recipients that are checking and
enforcing DMARC policy. This processing may cause subscribers
that are checking and enforcing DMARC policy to be inadvertently
suspended or removed from the Mailing List.
3.2.4. Gateways
A Gateway performs the basic routing and transfer work of message
relaying, but it is also permitted to modify content, structure,
addressing, and/or other attributes as needed to send the message
into a messaging environment that operates under different standards
or potentially incompatible policies.
Gateways share the same DMARC interoperability issues as ReSenders
(Section 3.2.2).
Gateways may also share the same DMARC interoperability issues as
MTAs (Section 3.1.2).
Receiver systems on the non-SMTP side of a protocol Gateway may be
unable to evaluate DKIM and SPF. If a message passes through a
second protocol Gateway back into the SMTP domain, the
transformations commonly break the original DKIM signature(s).
Gateway-level forwarding can introduce DMARC interoperability issues
if the Gateway is configured to rewrite the message into alternate
receiving domains. For example, an acquisition may lead an acquiring
company to decide to decommission the acquired company's domains by
rewriting messages to use the domain of the acquiring company. Since
the RFC5322.To header field is usually DKIM-signed, this kind of
rewriting will invalidate such DKIM signatures.
3.2.5. Boundary Filters
To enforce security boundaries, organizations can subject messages to
analysis for conformance with their safety policies. A filter might
alter the content to render it safe, such as by removing or otherwise
altering content deemed unacceptable.
Boundary Filters share the same DMARC interoperability issues as
ReSenders.
Issues may arise with SPF and DKIM evaluation if performed after
filter modifications.
Examples of Boundary Filters include:
o Malware scanning: To protect readers and its reputation, an MTA
that transfers a message may remove content believed to be harmful
from messages, reformulate content to canonical formats in order
to make them more trustworthy or easier to scan, and/or add text
in the body to indicate the message has been scanned. Any such
modifications would invalidate a DKIM signature.
o Spam filtering: To protect its reputation and assist other MTAs,
an MTA may modify a message to indicate its decision that the
message is likely to be unwanted and/or add text in the body to
indicate that such filtering has been done.
o Other text additions: An MTA may add an organizational disclaimer
or advertisement, for instance.
o URL alteration: Some systems will rewrite or alter embedded URLs
as a way to control the potential threat from malware.
o Secondary MX services: The secondary MX for an organization may be
external to the normal mail processing for the organization, and
it may queue and forward to the primary when it becomes available.
This will not invalidate DKIM but will prevent the primary from
validating SPF normally. In this case, however, it is
inappropriate for a primary MX server to perform an SPF check
against its own secondaries. Rather, the secondary MX should
perform this function and employ some trusted mechanism to
communicate the results of the SPF, DKIM, and DMARC evaluation(s)
to the primary MX server.
3.3. Combinations
Indirect email flows can be combined. For example, a university
student may subscribe to a Mailing List (using his university email
address) while this university email address is configured to forward
all emails to a freemail or a post-education corporate account
provider where a more permanent email address for this student
exists.
Within an organization, the message may pass through various MTAs
(Section 3.1.2), each of which performs a different function
(authentication, filtering, distribution, etc.).
4. Possible Mitigations of Interoperability Issues
Solutions to interoperability issues between DMARC and indirect email
flows vary widely in their scope and implications. They range from
improvements to underlying processing, such as proper handling of
multiple DKIM signatures, to more radical changes to the messaging
architecture. This section describes possible ways to address
interoperability issues. Note that these particular mechanisms may
not be considered "best practices" and may, in some cases, violate
various conventions or expectations.
Receivers sometimes need to deliver email messages that do not
conform to any standard or protocol, but are otherwise desired by end
users. Mitigating the impact of DMARC on indirect email flows is
especially important to receivers that operate services where ease of
use and compatibility with existing email flows is a priority.
DMARC provides a mechanism (local policy) for receivers to make
decisions about identity alignment acceptability based on information
outside DMARC and communicate those decisions as "overrides" to the
sender. This facility can be used to ease some interoperability
issues, although care is needed to ensure that this does not create
loopholes for abuse.
To further complicate the usage of mitigations, mitigation may not be
desired if the email in question is of a certain category of high
value or high risk (security-related) transactional messages (dealing
with financial transactions or medical records, for example). In
these cases, mitigating the impact of DMARC due to indirect email
flows may not be desirable (counterproductive or allowing for abuse).
As a final note, mail systems are diverse and widely deployed.
Systems of various ages and capabilities are expected to preserve
interoperability with the rest of the SMTP ecosystem. For instance,
Qmail is still used, although the base code has not been updated
since 1998. ezmlm, a once popular MLM, is still deployed but has not
been updated since 1997, although a new version (ezmlm-idx) exists.
Old versions of other open- and closed-source MTAs are still commonly
in operation. When dealing with aging or unsupported systems, some
solutions may be time-consuming and/or disruptive to implement.
4.1. Mitigations in Current Use
Because DMARC is already widely deployed, many operators already have
mitigations in use. These mitigations vary in their effectiveness
and side effects but have the advantage that they are currently
available.
4.1.1. Mitigations for Senders
4.1.1.1. Identifier Alignment
o MTAs handling multiple domains may choose to change
RFC5321.MailFrom to align with RFC5322.From to improve SPF
usability for DMARC.
o MTAs handling multiple domains may also choose to align
RFC5321.HELO/.EHLO to RFC5322.From, particularly when sending
notification messages. Dynamically adjusting the
RFC5321.HELO/.EHLO based on the RFC5322.From may not be possible
for some MTA software.
o MTAs may choose to DKIM-sign notification messages with an aligned
domain to allow a DKIM-based DMARC pass.
o MTAs sending email on behalf of multiple domains may require
Domain Owners to provide DKIM keys to use DKIM to avoid SPF
validation issues, given the requirement for DMARC alignment with
the RFC5322.From header field. Managing DKIM keys with a third
party has security risks that should be carefully managed (see
also [RFC6376], Section 8). Methods involving CNAMEs and/or
subdomains may alleviate some risks.
o Senders who are sending on behalf of users in other administrative
domains may choose to use an RFC5322.From under the sender's
control. The new From can be either a forwarding address in a
domain controlled by the Sender or a placeholder address, with the
original user's address in an RFC5322.Reply-to header field.
However, performing this modification may cause the Recipient's
MUA to deviate from customary behavior.
o When implementing "forward-to-friend" functionality, one approach
to avoid DMARC failures is to pass a well-formed message to the
user's MUA so that it may fill in an appropriate identity and
submit through its own MSA.
o Senders can use domains with distinct DMARC policies for email
sent directly and email known to use indirect mail flows.
However, for most well-known brands, all active domains are likely
to be targeted equally by abusers.
4.1.1.2. Message Modification
o Senders can maximize survivability of DKIM signatures by limiting
the header fields they sign and using relaxed canonicalization.
Using the DKIM length tag to allow appended signatures is
discouraged due to the security risk created by allowing arbitrary
content to be appended to legitimate email.
o Senders can also maximize survivability by starting with RFC-
compliant headers and common body formats.
o In order to minimize transport-based conversions, Senders can
convert messages to a lowest denominator MIME content-transfer
encoding such as quoted-printable or base64 before signing
([RFC6376], Section 5.3).
4.1.2. Mitigations for Receivers
4.1.2.1. Identifier Alignment
o Receivers should update DKIM handling libraries to ensure that
they process all valid DKIM signatures and check each signature
for alignment.
4.1.2.2. Policy Override
o Receivers can amalgamate data from their user base to create lists
of forwarders and use such lists to inform DMARC local policy
overrides. This process may be easier for large receivers where
data and resources to create such lists are more readily available
than at smaller sites, where there are fewer accounts that receive
forwarded mail and other resources may be scarce.
4.1.3. Mitigations for ReSenders
4.1.3.1. Changes to the RFC5322.From
Many ReSender issues can be avoided by using an RFC5322.From header
field under the ReSender's control, instead of the initial
RFC5322.From. This will correct Identifier Alignment issues and
allow arbitrary message modification as long as the ReSender signs
the message with an aligned domain signature. When ReSenders change
the RFC5322.From, it is desirable to preserve the information about
the original initiator of the message.
A first option is to use the Original-From [RFC5703] (or X-Original-
From) header field for this purpose in various contexts (X- header
field names are discouraged by [RFC6648]). However, handling of
Original-From (or X-Original-From) is not defined anywhere. It is
not currently used consistently or displayed to the user, and in any
situation where it is used, it is a new unauthenticated identifier
available for exploitation unless included within the scope of the
new DKIM signature(s).
Another option for ReSenders is to rewrite the RFC5322.From header
field address to a locally controlled address that will be forwarded
back to the original sender (subject to its own ReSender forwarding
mitigations).
4.1.3.2. Avoiding Message Modification
o Forwarders can choose to add email header fields instead of
modifying existing headers or bodies, for instance, to indicate a
message may be spam.
o Forwarders can minimize the circumstances in which they choose to
fix messages, preferring to preserve non-compliant headers to
creating DKIM failures.
o Forwarders can choose to reject messages with suspect or harmful
content instead of modifying them.
4.1.3.3. Mailing Lists
[RFC6377] provides some guidance on using DKIM with Mailing lists.
The following mitigation techniques can be used to ease
interoperability issues with DMARC and Mailing lists:
o Configuring the Mailing List Manager (MLM) to alter the
RFC5322.From header field to use the domain of the MLM is a
mitigation policy that is now present in several different Mailing
List software distributions. Since most list subscribers prefer
to know the identity of the Author of the original message,
typically this information may be provided in the display name
part of the RFC5322.From header field. This display name needs to
be carefully crafted so as to not collide with the original
display name of the Author, nor contain something that looks like
an email address or domain name. These modifications may to some
extent defeat the purpose of DMARC itself. It may make it
difficult to ensure that users of all email clients can easily
reply to the Author, the list, or all using the email client
features provided for that purpose. Use of the RFC5322.Reply-To
header field can alleviate this problem depending on whether the
Mailing List is configured to reply-to-list, reply-to-author, or
reply-to-fixed-address; however, it is important to note that this
header field can take multiple email addresses. When altering the
RFC5322.From, there are three possibilities:
1. change it to put the Mailing List email address,
2. change it to a locally defined address that will be forwarded
back to the original sender, or
3. "break" the address by modifying the domain to a non-existent
domain (such as by adding a suffix like ".invalid").
The latter modification may create issues because it is an invalid
domain name, and some MTAs may pay particular attention to the
validity of email addresses in RFC5322.From and the reputation of
the domains present there.
o Configuring the MLM to "wrap" the message in a MIME message/rfc822
part and to send as the Mailing List email address. Many email
clients (as of the publication of this document), especially
mobile clients, have difficulty reading such messages, and this is
not expected to change soon.
o Configuring the MLM to not modify the message so that the DKIM
signature remains valid. Some Mailing Lists are set up this way
and require few additional changes to ensure the DKIM signature is
preserved. Moving lists that currently modify mail to a policy
like this may be too much of a change for the members of such
lists.
o Rejecting posts or membership requests from domains with a DMARC
policy other than "p=none". However, members or potential members
of such Mailing Lists may complain of unfair exclusion.
o To alleviate unsubscribes to the Mailing List due to the messages
bouncing because of DMARC, the MLM needs to not act on
notification messages due to message authentication issues.
[RFC3463] specifies Enhanced Mail System Status Codes, which help
differentiate between various failure conditions. Correctly
interpreting Extended SMTP error messages is useful in this case.
In particular, extended status codes for SPF and DKIM causes are
defined in [RFC7372], and DMARC-related failure indications are
discussed in DMARC ([RFC7489], Section 10.3).
All these techniques may provide some specific challenges to MUAs and
different operational usages for end users (like rewriting filters to
sort emails in folders). There will be some time before all
implications are understood and accommodated.
4.2. Proposed and In-Progress Mitigations
The following mitigations are based on Internet-Drafts (I-Ds) that
are still in process. They are described here to offer an
exploratory path for solutions. These solutions should not be used
in a production environment. Because of the transient nature of
I-Ds, specific citations are not included because a number of them
will inevitably become obsolete, and those that gain consensus in the
community will become RFCs and should be discovered as such.
o Third-party authorization schemes provide ways to extend
Identifier Alignment under control of the Domain Owner.
o Ways to canonicalize messages that transit Mailing Lists so that
their alterations can be isolated from the original signed
content.
o Mechanisms to record message transformations applied at each hop
so they can be reversed and the original signed content recovered.
o "Conditional" DKIM signatures, whereby the Author domain indicates
its signature is only good if accompanied by a signature from an
expected downstream relay.
o Mechanisms to extend Authentication-Results [RFC7601] to multiple
hops, creating a provable chain of custody as well as a view of
message authentication results at each handling step.
4.2.1. Getting More Radical: Requiring New Communication Paths between
MUAs
In practice, a number of operators are using strict alignment mode in
DMARC in order to avoid receiving new and innovative forms of
unwanted and unauthentic email through systems purporting to be
Mailing List handlers. The receiving ADMD has no knowledge of which
lists the user has subscribed to and which they have not. One avenue
of exploration would be for the user to authorize Mailing Lists as
proxies for authentication, at which point the receiving ADMD would
be vesting some trust in the Mailing List service. The creators of
DKIM foresaw precisely this possibility at the time by not tightly
binding any semantics to the RFC5322.From header field. Some
experimental work has taken place in this area, as mentioned above.
Additional work might examine a new communication path to the user to
authorize some form of transitive trust.
5. Security Considerations
This document is an analysis of DMARC's impact on indirect email
flows. It describes the possibility of accidental denial of service
that can be created by rejections of messages by DMARC-aware mail
receivers.
Section 4.1.1.1 discusses the importance of appropriate DKIM key
management vis-a-vis third-party email senders.
Section 4.1.3.3 warns that rewriting the RFC5322.From header field to
create an artificial domain name should not be done with any domain.
6. References
6.1. Normative References
[RFC2045] Freed, N. and N. Borenstein, "Multipurpose Internet Mail
Extensions (MIME) Part One: Format of Internet Message
Bodies", RFC 2045, DOI 10.17487/RFC2045, November 1996,
<http://www.rfc-editor.org/info/rfc2045>.
[RFC3463] Vaudreuil, G., "Enhanced Mail System Status Codes",
RFC 3463, DOI 10.17487/RFC3463, January 2003,
<http://www.rfc-editor.org/info/rfc3463>.
[RFC5228] Guenther, P., Ed. and T. Showalter, Ed., "Sieve: An Email
Filtering Language", RFC 5228, DOI 10.17487/RFC5228,
January 2008, <http://www.rfc-editor.org/info/rfc5228>.
[RFC5321] Klensin, J., "Simple Mail Transfer Protocol", RFC 5321,
DOI 10.17487/RFC5321, October 2008,
<http://www.rfc-editor.org/info/rfc5321>.
[RFC5322] Resnick, P., Ed., "Internet Message Format", RFC 5322,
DOI 10.17487/RFC5322, October 2008,
<http://www.rfc-editor.org/info/rfc5322>.
[RFC5598] Crocker, D., "Internet Mail Architecture", RFC 5598,
DOI 10.17487/RFC5598, July 2009,
<http://www.rfc-editor.org/info/rfc5598>.
[RFC5703] Hansen, T. and C. Daboo, "Sieve Email Filtering: MIME Part
Tests, Iteration, Extraction, Replacement, and Enclosure",
RFC 5703, DOI 10.17487/RFC5703, October 2009,
<http://www.rfc-editor.org/info/rfc5703>.
[RFC6376] Crocker, D., Ed., Hansen, T., Ed., and M. Kucherawy, Ed.,
"DomainKeys Identified Mail (DKIM) Signatures", STD 76,
RFC 6376, DOI 10.17487/RFC6376, September 2011,
<http://www.rfc-editor.org/info/rfc6376>.
[RFC6377] Kucherawy, M., "DomainKeys Identified Mail (DKIM) and
Mailing Lists", BCP 167, RFC 6377, DOI 10.17487/RFC6377,
September 2011, <http://www.rfc-editor.org/info/rfc6377>.
[RFC6648] Saint-Andre, P., Crocker, D., and M. Nottingham,
"Deprecating the "X-" Prefix and Similar Constructs in
Application Protocols", BCP 178, RFC 6648,
DOI 10.17487/RFC6648, June 2012,
<http://www.rfc-editor.org/info/rfc6648>.
[RFC7208] Kitterman, S., "Sender Policy Framework (SPF) for
Authorizing Use of Domains in Email, Version 1", RFC 7208,
DOI 10.17487/RFC7208, April 2014,
<http://www.rfc-editor.org/info/rfc7208>.
[RFC7372] Kucherawy, M., "Email Authentication Status Codes",
RFC 7372, DOI 10.17487/RFC7372, September 2014,
<http://www.rfc-editor.org/info/rfc7372>.
6.2. Informative References
[RFC7489] Kucherawy, M., Ed. and E. Zwicky, Ed., "Domain-based
Message Authentication, Reporting, and Conformance
(DMARC)", RFC 7489, DOI 10.17487/RFC7489, March 2015,
<http://www.rfc-editor.org/info/rfc7489>.
[RFC7601] Kucherawy, M., "Message Header Field for Indicating
Message Authentication Status", RFC 7601,
DOI 10.17487/RFC7601, August 2015,
<http://www.rfc-editor.org/info/rfc7601>.
Appendix A. Example SPF Bounce
This example illustrates a notification message "bounce".
A.1. Initial Message
Here is the message as it exits the Origin MTA (segv.d1.example):
Return-Path: <jqd@d1.example>
Received: from [10.10.10.131] (w-x-y-z.dsl.static.isp.com [w.x.y.z])
(authenticated bits=0)
by segv.d1.example with ESMTP id t0FN4a8O084569;
Thu, 14 Jan 2015 15:00:01 -0800 (PST)
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=d1.example;
s=20130426; t=1421363082;
bh=EoJqaaRvhrngQxmQ3VnRIIMRBgecuKf1pdkxtfGyWaU=;
h=Message-ID:Date:From:MIME-Version:To:CC:Subject:Content-Type:
Content-Transfer-Encoding;
b=HxsvPubDE+R96v9dM9Y7V3dJUXvajd6rvF5ec5BPe/vpVBRJnD4I2weEIyYijrvQw
bv9uUA1t94kMN0Q+haFo6hiQPnkuDxku5+oxyZWOqtNH7CTMgcBWWTp4QD4Gd3TRJl
gotsX4RkbNcUhlfnoQ0p+CywWjieI8aR6eof6WDQ=
Message-ID: <54B84785.1060301@d1.example>
Date: Thu, 14 Jan 2015 15:00:01 -0800
From: John Q Doe <jqd@d1.example>
To: no-recipient@dmarc.org
Subject: Example 1
Hey gang,
This is a test message.
--J.
A.2. Notification Message
When dmarc.org rejects the message without a DKIM signature, it
specifies the RFC5321.HELO/.EHLO domain as dmarc.org.local, which has
no SPF record. dmarc.org has a reject policy in place for such non-
passing cases. Since there is no DKIM signature on the notification
message, the failed SPF lookup results in a dmarc=fail, and
d1.example could be expected to discard the notification message
itself:
Return-Path: <>
Received: from dmarc.org.local (mail.dmarc.org. [192.0.2.1])
by mx.d1.example with ESMTPS id Lkm25302jJR5
for <jqd@d1.example>
(version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128);
Thu, 14 Jan 2015 15:00:24 -0800 (PST)
Authentication-Results: mx.d1.example;
spf=none (d1.example: dmarc.org.local does not designate
permitted sender hosts) smtp.mail=;
dmarc=fail (p=REJECT dis=NONE) header.from=dmarc.org
MIME-Version: 1.0
Received: from segv.d1.example (segv.d1.example [198.51.100.1])
by 192.0.2.2 with SMTP id u67mr102828634qge33; Thu,
14 Jan 2015 15:00:24 -0800 (PST)
From: Mail Delivery Subsystem <mailer-daemon@dmarc.org>
To: jqd@d1.example
Subject: Delivery Status Notification (Failure)
Message-ID: <001a11c16e6a9ead220528df294a@dmarc.org>
Date: Thu, 14 Jan 2016 23:00:24 +0000
Content-Type: text/plain; charset=UTF-8
This is an automatically generated Delivery Status Notification
Delivery to the following recipient failed permanently:
no-recipient@dmarc.org
Technical details of permanent failure:
Your message was rejected by the server for the recipient domain
dmarc.org by mail.dmarc.org [192.0.2.1].
The error that the other server returned was:
550 5.1.1 <no-recipient@dmarc.org>... User unknown
----- Original message -----
Return-Path: <jqd@d1.example>
Received: from [203.252.0.131] (131-0-252-203-dsl.static.example.com
[203.252.0.131]) (authenticated bits=0)
by segv.d1.example with ESMTP id t0FN4a8O084569;
Thu, 14 Jan 2015 15:00:01 -0800 (PST)
(envelope-from jqd@d1.example)
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=d1.example;
s=20130426; t=1421363082;
bh=EoJqaaRvhrngQxmQ3VnRIIMRBgecuKf1pdkxtfGyWaU=;
h=Message-ID:Date:From:MIME-Version:To:CC:Subject:Content-Type:
Content-Transfer-Encoding;
b=HxsvPubDE+R96v9dM9Y7V3dJUXvajd6rvF5ec5BPe/vpVBRJnD4I2weEIyYijrvQw
bv9uUA1t94kMN0Q+haFo6hiQPnkuDxku5+oxyZWOqtNH7CTMgcBWWTp4QD4Gd3TRJl
gotsX4RkbNcUhlfnoQ0p+CywWjieI8aR6eof6WDQ=
Message-ID: <54B84785.1060301@d1.example>
Date: Thu, 14 Jan 2015 15:00:01 -0800
From: John Q Doe <jqd@d1.example>
To: no-recipient@dmarc.org
Subject: Example 1
Hey gang,
This is a test message.
--J.
Acknowledgments
Miles Fidelman, John Levine, David Crocker, Stephen J. Turnbull, Rolf
E. Sonneveld, Tim Draegen, and Franck Martin contributed to the IETF
DMARC Working Group's wiki page listing all known interoperability
issues with DMARC and indirect email flows.
Tim Draegen created the first draft of this document from these
contributions and by ham-fistedly mapping contributions into the
language of [RFC5598].
Authors' Addresses
Franck Martin (editor)
LinkedIn
Mountain View, CA
United States of America
Email: fmartin@linkedin.com
Eliot Lear (editor)
Cisco Systems GmbH
Richtistrasse 7
Wallisellen, ZH CH-8304
Switzerland
Phone: +41 44 878 9200
Email: lear@cisco.com
Tim Draegen (editor)
dmarcian, inc.
PO Box 1007
Brevard, NC 28712
United States of America
Email: tim@dmarcian.com
Elizabeth Zwicky (editor)
Yahoo
Sunnyvale, CA
United States of America
Email: zwicky@yahoo-inc.com
Kurt Andersen (editor)
LinkedIn
2029 Stierlin Court
Mountain View, CA 94043
United States of America
Email: kandersen@linkedin.com
ISO Plain Language
On this page
§1The Problem in One Page
DMARC, defined in RFC 7489, introduces a mechanism for expressing domain-level policies and preferences for message validation, disposition and reporting. The mechanism, especially when employed with restrictive policies, encounters several different types of interoperability issue because of third-party message sourcing, message transformation or rerouting. In particular, DMARC with restrictive policies causes problems for many mailing lists.
At the time of writing, the DMARC base specification had been published as Informational RFC 7489 and had seen significant deployment within the email community. Cases in which email does not flow directly from the author's administrative domain to the recipient's domains are collectively called indirect email flows. Because of existing and increasing adoption of DMARC, the impact of DMARC-based email rejection policies on indirect email flows can be significant for a select subset of general email traffic.
The issues arise when conformance to the DMARC specification leads a receiving implementation to apply DMARC-based policy restrictions to messages that are both compliant with the architecture as specified in RFC 5598 and viewed as legitimate by the intended recipient. Domains that assert a p=none policy, and messages that pass standard DMARC validation, have no interoperability issues. Messages that do not conform to IETF email specifications but are considered legitimate are not discussed.
The document presents several known causes of interoperability issues, then describes the components of the Internet Mail Architecture where they can arise, and finally presents known and possible methods for addressing them. There are often multiple ways to address any given issue, and while the document strives to be comprehensive it should not be treated as complete. It also notes that some practices in use at the time may or may not be best practices, especially as future standards evolve.
A domain can ask receivers to reject its mail if it fails a security check. The check compares the domain that signed the message with the domain in the From line. Forwarded or rewritten mail breaks that comparison, so the message gets blocked even though the person who sent it would not object.
§2Why the Check Fails
DMARC relies on DomainKeys Identified Mail in RFC 6376 and SPF in RFC 7208 to perform message source validation, and calls the source domains they validate the Authenticated Identifiers. To be used by DMARC, an Authenticated Identifier must also be related to the domain found in the message's From header field, and that relationship is called Identifier Alignment. DMARC allows alignment to be determined in two modes. Strict mode requires an exact match between the domain of any Authenticated Identifier and the From header field. Relaxed mode allows alignment when the Authenticated Identifier and the From header field share the same Organizational Domain. The interoperability issues are generally the same for both, but strict alignment constrains the possible solutions because of the more rigorous matching requirement, and some mitigations only work with relaxed alignment.
With DKIM, a signing domain must be part of the same Organizational Domain as the domain in the From header field for the identifier to align. DKIM also allows multiple valid signatures, from the same or different domains, and DMARC processes Authenticated Identifiers based on DKIM signatures until an aligned one is found. Operational experience has shown that some implementations have difficulty processing multiple signatures, and implementations that cannot process them may incorrectly flag messages as not passing alignment and apply policy to otherwise conforming messages.
SPF defines two Authenticated Identifiers per message, derived from the envelope sender domain and from the HELO or EHLO SMTP domain. The MAILFROM value is based on the envelope sender unless that value is absent, as in the case of notification messages, in which case the HELO identifier is used. This fallback definition has occasionally been misunderstood by operators, since notification messages are often an automatic feature of MTA software, and some MTA software cannot apply a DKIM signature to them.
The message format permits only one From header field, but it may contain multiple mailboxes. Because an attacker could add their domain to the From field, provide arbitrary new content and sign the message, DMARC recommends that the strictest policy be applied to such messages.
Forwarding involves the retransmission of email and introduces specific issues for SPF. If the envelope sender is present and the forwarder maintains it, SPF validation fails unless the forwarder is an authorised part of the originator's sending infrastructure; if the forwarder replaces the envelope sender with its own domain, SPF might pass but alignment with the From header field will fail. If the envelope sender is empty, as in the case of delivery status notifications, the HELO domain of the forwarder will likely be in a different Organizational Domain from the From header field's domain, so SPF may pass but alignment will fail. In both cases SPF cannot yield relevant Authenticated Identifiers, and DKIM must be relied upon.
Modification of email content invalidates most DKIM signatures, and many forwarding systems modify content. Mailing list processors are the common example. DKIM provides a length flag so that content can be appended without invalidating the signature, but in practice, particularly with MIME-encoded messages, a mailing list processor does more than simply append content, and the length flag is seldom used because of security issues. DKIM also describes two canonicalizations for preparing the header and body: simple and relaxed. The relaxed form is designed to accommodate trivial changes to whitespace and folding that generally have no semantic significance in the header, but it is more computationally intensive and may not have been preferred in the early deployment of DKIM, leaving some deployments using the less forgiving simple canonicalization. While the prevalence is unknown, there are some DKIM verifiers with problems evaluating relaxed canonicalization correctly.
There are three ways to go wrong. The signing domain may not match the From line, the message may have been forwarded so the envelope no longer points at the sender, or the message may have been edited so the signature no longer matches. Any one of them is enough for a strict policy to reject the mail.
| Case | Effect on SPF | Effect on DKIM |
|---|---|---|
| The forwarder keeps the original envelope sender | Fails, unless the forwarder is an authorised part of the originator's sending infrastructure. | Signatures survive, since the signed content is unchanged. |
| The forwarder replaces the envelope sender with its own domain | May pass, but the domain no longer aligns with the From header field. | Signatures survive, since the signed content is unchanged. |
| A notification message with an empty envelope sender | Uses the HELO domain of the forwarder, which is usually in a different Organizational Domain. | Survives only if the software can sign notification messages at all. |
| The message content or headers are modified | Unaffected. | Invalidated, because the signed content no longer matches. |
§3Where in the Mail System This Happens
The Internet Mail Architecture defines six basic components that make up the Message Handling System: the Message, the Message User Agent, the Message Submission Agent, the Message Transfer Agent, the Message Delivery Agent and the Message Store. It also defines a Mediator as a hybrid of several component types, which the document gives special consideration because of the unique issues they face when trying to interoperate with DMARC.
An MSA accepts messages submitted by a Message User Agent and enforces the policies of the hosting administrative domain and the requirements of Internet standards. It is split into author-focused functions and MHS-focused functions. Interoperability issues begin when an author-focused MSA accepts a message whose From header field contains a domain outside the MSA's administrative domain, which happens when someone uses a mail service with their own domain but has not configured an SPF record properly, or when a user agent tries to transmit mail as someone else, such as the forward-to-friend functionality on news sites or send-as functionality in some mail clients. When an MHS-focused MSA takes responsibility for transit of a message whose From domain is outside its administrative domain, it faces issues if that domain publishes a quarantine or reject policy. The examples given are partially open relays, embedded devices such as cable and DSL modems, firewalls, wireless access points and printers that send email using hardcoded domains, devices that send mail on behalf of a user such as scanners, security cameras and alarms, email service providers servicing customers whose domains publish a reject policy, and calendaring software where an invited attendee modifies an event and causes an update to be sent claiming to come from the creator of the event.
An MTA relays a message until it reaches a destination delivery agent, and some common message-handling strategies break the integrity of DKIM signatures, so a restrictive DMARC policy turns latent problems into overt ones. An MTA may modify the message encoding, for instance by converting 8-bit MIME sections to quoted-printable 7-bit sections, which is outside the scope of DKIM canonicalization and invalidates signatures that include message content. An MTA could also re-encode a message without changing the encoding type, producing a semantically equivalent MIME body that is not identical to the original, which is characteristic of systems that use another message representation internally. An MTA may rewrite headers to bring them into compliance with existing RFCs, for example correcting a comprehensible but non-compliant date format, which is outside the scope of DKIM canonicalization and invalidates signatures, leaving downstream DMARC processing unable to use DKIM. An MTA may also implement security-motivated changes to the content of messages, dropping or altering sections, which also breaks signatures.
An MDA transfers a message to a mailbox and, like the MSA, has MHS-focused and recipient-focused sub-components, both of which can redirect a message to an alternative address. Sieve filtering often lives in the recipient-focused part, and its addheader and deleteheader actions can modify messages and invalidate DKIM signatures; there are also Sieve extensions that modify the body, and such alterations only become an issue when the email is reintroduced into the transport infrastructure.
Mediators forward messages through a re-posting process. They share some functionality with basic relaying but have greater flexibility in addressing and content modifications, and DMARC interoperability issues are common within them, because they are often used precisely for their ability to modify messages. The DMARC design does not cope with some mediator functionality, such as content modifications that invalidate DKIM signatures and rewriting of the envelope sender to support SPF authentication of re-sent mail.
An Alias is a simple re-addressing facility that provides one or more new Internet Mail addresses rather than a single internal one, and the message continues through the transfer service to one or more alternative addresses. Aliases can be implemented by mailbox-level forwarding, by Sieve's redirect action, or by other methods. When an alias preserves message content and makes no significant header changes, DKIM signatures may remain valid, but aliases often extend the delivery path outside the scope covered by the originating administrative domain's SPF records. Examples include forwarding email between free email providers while keeping an original address, consolidating many addresses into a single account, and services offering activity-based, role-based, vanity or temporary addresses, such as universities offering an alias for the duration of a membership.
A ReSender splices a message's addressing information to connect the author of the original message with the recipients of the new message, so the new recipient sees the message as coming from the original author even if the mediator adds commentary. Without Authenticated Identifiers aligned with the author's From domain, the new recipient has no way to achieve a passing DMARC evaluation. Examples are user-agent forwarding by resending a message to a new recipient or forwarding it inline, and calendaring software that lets an attendee modify an invite so the new invitations claim to be reissued from the meeting organizer.
A Mailing List receives messages as an explicit addressee and reposts them to subscribed members, doing what can be viewed as an elaboration of the ReSender actions, and sharing the same interoperability issues. It very commonly modifies headers or content in ways that will cause DKIM to fail: prepending the Subject header with a tag, adding a footer to the body, removing MIME parts or converting the message to text only, encrypting the body with the recipient's key using PGP or S/MIME, rewriting banned words to enforce community standards, and letting moderators add arbitrary commentary. Such modifications are often central to present mailing list functionality, and mail clients have come to rely on them to present messages to users in expected ways. Mailing lists also have operational effects: subscribed members may not receive email from members who post using domains that publish a reject policy, and a list may interpret a DMARC-related rejection as an inability to deliver mail to recipients that are checking DMARC policy, which can cause subscribers that enforce DMARC to be inadvertently suspended or removed from the list.
Gateways perform the basic routing and transfer work of relaying but are also permitted to modify content, structure, addressing and other attributes to send the message into a messaging environment under different standards or incompatible policies. They share the issues of ReSenders and MTAs. Receiver systems on the non-SMTP side of a gateway may be unable to evaluate DKIM and SPF at all, and if a message passes through a second gateway back into the SMTP domain, the transformations commonly break the original signatures. Gateway-level forwarding can also introduce issues when the gateway rewrites the message into alternate receiving domains, for example when an acquiring company decides to decommission the acquired company's domains by rewriting messages to use its own domain, which invalidates signatures since the To header field is usually signed.
Boundary Filters subject messages to analysis for conformance with safety policies and may alter the content to render it safe, such as by removing unacceptable content. They share the ReSender issues, and problems arise if SPF and DKIM evaluation is performed after filter modifications. The examples are malware scanning, where an MTA removes content believed to be harmful, reformats content to canonical formats or adds text to the body noting the message has been scanned; spam filtering, where an MTA modifies a message to indicate it is likely unwanted; other text additions such as an organisational disclaimer or advertisement; URL alteration, where systems rewrite embedded URLs to control the threat from malware; and secondary MX services, where a secondary mail server outside the normal processing queues and forwards to the primary when it becomes available, which will not invalidate DKIM but will prevent the primary from validating SPF normally, so it is inappropriate for a primary to run an SPF check against its own secondaries and the secondary should instead perform it and communicate the results by a trusted mechanism.
The document walks through the mail system looking for the places that break a signature. Mail servers that re-encode or tidy headers, filters that add a footer, mailing lists that rewrite the subject line, devices that send mail under a hardcoded domain, and gateways that translate between protocols all sit on the path and can each break it.
- Message Handling System
- Message The content itself.
- Message User Agent Where a person reads and writes mail.
- Message Submission Agent Author-focused and MHS-focused functions.
- Message Transfer Agent Relays the message to the destination.
- Message Delivery Agent Moves the message into a mailbox; may apply Sieve rules.
- Message Store Where the delivered message is kept.
- Mediators Hybrids that re-post messages, with more freedom to change them.
- Alias
- ReSender
- Mailing List
- Gateway
- Boundary Filter
§4What Operators Are Doing About It
Because DMARC is already widely deployed, many operators already have mitigations in use. They vary in effectiveness and side effects but have the advantage of being available now. The document notes that mitigations may not be wanted for email of a certain category of high value or high risk, such as security-related transactional messages dealing with financial transactions or medical records, where mitigating the impact of DMARC may be counterproductive or allow abuse. Mail systems are also diverse and widely deployed, with systems of various ages and capabilities expected to preserve interoperability with the rest of the SMTP ecosystem: Qmail is still used although the base code has not been updated since 1998, and ezmlm, a once popular mailing list manager, is still deployed but has not been updated since 1997. When dealing with aging or unsupported systems, some solutions may be time-consuming or disruptive to implement.
For senders. MTAs handling multiple domains may change the envelope sender to align with the From header to improve SPF usability, and may align the HELO or EHLO with the From header, particularly when sending notification messages, though dynamically adjusting the HELO based on the From header may not be possible for some software. MTAs may sign notification messages with an aligned domain to allow a DKIM-based pass. MTAs sending on behalf of multiple domains may require domain owners to provide DKIM keys, since managing DKIM keys with a third party has security risks, though methods involving CNAMEs and subdomains may alleviate some of them. Senders acting for users in other administrative domains may use a From address under the sender's own control, either a forwarding address in a domain the sender controls or a placeholder address, with the original user's address in a Reply-To header, though this may cause the recipient's mail client to deviate from customary behaviour. For forward-to-friend functionality, one approach is to pass a well-formed message to the user's mail client so that it can fill in an appropriate identity and submit through its own submission agent. Senders can use domains with distinct DMARC policies for mail sent directly and mail known to use indirect flows, though for well-known brands all active domains are likely to be targeted equally by abusers.
Senders can also maximise the survivability of DKIM signatures by limiting the header fields they sign and using relaxed canonicalization, and by starting with compliant headers and common body formats. Using the DKIM length tag to allow appended signatures is discouraged because of the security risk of allowing arbitrary content to be appended to legitimate mail. To minimise transport-based conversions, senders can convert messages to a lowest-denominator content-transfer encoding such as quoted-printable or base64 before signing.
For receivers. Receivers should update their DKIM handling libraries to ensure they process all valid signatures and check each one for alignment. Receivers can also amalgamate data from their user base to create lists of forwarders and use those lists to inform DMARC local policy overrides, which may be easier for large receivers where data and resources are more readily available than at smaller sites.
For re-senders. Many issues can be avoided by using a From header under the re-sender's control instead of the original one, which corrects alignment and allows arbitrary modification as long as the re-sender signs the message with an aligned domain signature. When the From is changed it is desirable to preserve information about the original initiator, for instance with an Original-From header, although handling of that field is not defined anywhere, it is not used consistently or displayed to the user, and where it is used it is a new unauthenticated identifier available for exploitation unless it is included within the scope of the new signatures. Another option is to rewrite the From address to a locally controlled address that will be forwarded back to the original sender. Forwarders can also add header fields instead of modifying existing headers or bodies, can minimise the circumstances in which they fix messages, preferring to preserve non-compliant headers rather than create DKIM failures, and can choose to reject messages with suspect or harmful content instead of modifying them.
For mailing lists. Configuring the list manager to alter the From header to use the domain of the mailing list is a mitigation now present in several list software distributions. Since most subscribers prefer to know the identity of the original author, that information is typically provided in the display name part of the From header, which needs to be crafted so as not to collide with the original author's display name nor contain something that looks like an email address or domain name. These modifications may to some extent defeat the purpose of DMARC and may make it hard for users of all mail clients to reply to the author, the list or all of them; the Reply-To header field can alleviate this depending on whether the list is configured to reply to the list, to the author or to a fixed address, though that field can hold multiple addresses. When altering the From header there are three possibilities: put the mailing list's own address in it, change it to a locally defined address that forwards back to the original sender, or break the address by modifying the domain to a non-existent one such as by adding a suffix like .invalid. The last may create issues because it is an invalid domain name and some MTAs pay particular attention to the validity of addresses in the From header and to the reputation of the domains there.
Other list options are to configure the list manager to wrap the message in a MIME message/rfc822 part and send as the list's own address, though many mail clients, especially mobile ones, have difficulty reading such messages and this is not expected to change soon; to configure the manager not to modify the message so the signature remains valid, which some lists already do and which requires few additional changes, though moving lists that currently modify mail to such a policy may be too much of a change for their members; and to reject posts or membership requests from domains with a policy other than p=none, which members may complain about as unfair exclusion. To alleviate unsubscribes caused by messages bouncing because of DMARC, the list manager needs not to act on notification messages due to message authentication issues, and correctly interpreting extended SMTP error messages is useful, since enhanced status codes for SPF and DKIM causes are defined in RFC 7372 and DMARC-related failure indications are discussed in the DMARC specification. All these techniques may create specific challenges for mail clients and for end-user workflows such as rules that sort mail into folders, and there will be some time before all the implications are understood.
The fixes are mostly practical: sign the message with a domain the list controls, do not touch the message if you can avoid it, and if you must change it, sign it again. The document warns that some of these choices partly defeat the point of DMARC, and that the trade-off should be thought about for important mail such as bank or medical messages.
- For sendersAlign the envelope sender or HELO with the From header, sign notification messages, use relaxed canonicalization, and sign in a lowest-common-denominator encoding before sending.
- For receiversProcess all valid DKIM signatures and check each for alignment, and build lists of known forwarders to drive local policy overrides.
- For re-sendersUse a From address under your own control and sign it, add header fields rather than editing the message, and reject suspect mail rather than rewriting it.
- For mailing listsChange the From to the list's own domain, or leave the message untouched so the signature survives, or reject posts from domains with a policy other than p=none.
§5Security Considerations
The document is an analysis of DMARC's impact on indirect email flows, and describes the possibility of accidental denial of service that can be created by rejections of messages by DMARC-aware mail receivers. It points back to the discussion of appropriate DKIM key management in the face of third-party senders, and warns that rewriting the From header field to create an artificial domain name should not be done with any domain.
Two cautions. Strict checking can block mail by accident, which is a denial of service in effect. And a list that invents a fake From domain is a bad idea.