I remain concerned about the often-stated position that the DMARC
specification is OK if it only produces false PASS in rare cases.  False
PASS is never OK, and our goal should be to prevent the problem rather than
ignore it.

In RFC 7489, PASS is a consolidation of 8 different trust indicators, but
they do not all warrant the same level of trust.   We have not discussed
those differences, so I will restate them here:

Authentication Method:   DKIM vs SPF.  DKIM is a much stronger
authentication method than SPF.

SPF
In a multiple tenant environment, SPF assumes that the hosting service has
administrative controls in place to prevent one tenant from impersonating
the MAILFROM address of another tenant.   Therefore, if the IP address
authorized to send for the MAILFROM domain, then the message is
authorized.   This is a sufficiently weak assumption that some domain
owners do not publish SPF policies, because they do not want SPF to
authenticate for DMARC.

DKIM
In a multiple tenant environment, DKIM only assumes that the hosting
service provides standard operating system features to keep one tenants
private keys protected from use by other tenants

Authentication Relationship: nominally Strict vs Relaxed, but actually 4
different types
STRICT / EQUAL MATCH: The domain authenticates itself.  This is always
reliable.

PARENT-CHILD:  The parent domain validates a message which uses a child
domain.   This is consistent with the inherent powers associated with being
a DNS parent node, and with the inherent powers associated with upper
layers of a hierarchical organization.

CHILD-PARENT:  The child domain validates a message which uses a parent
domain.   Improper action of this type is expected to bring swift
retaliation from the parent, so the likelihood of misuse seems low.  I
think our collective experience with actual mail data is that nobody tries
to impersonate PSDs.

SIBLING: One part of an organization validates a message for a domain in a
different part of the organization.   This is a very weak authentication
method and has multiple risks.  The primary risks are created by private
registries, if the registry boundary is undetected.   For the PSL, the
threat comes from private registries that are not listed.  For the Tree
Walk, the threat comes from private registries that are not tagged.   The
secondary risk comes from assuming strong central control where it may not
exist, and where the impersonation may be the result of malware.
 Universities come to mind because we have discussed how they have
distributed control structures.   Is it wise for us to assume that a
message with MAIL FROM of [email protected] is authorized to send
messages with FROM [email protected], or
[email protected] can send messages for [email protected]?
 I don't think so.

My proposal:
Sibling authentication should be disabled by default, even for policies
that specify relaxed authentication.    Those organizations that want
sibling authentication should explicitly request it using a tag (to be
defined) on the Organizational Domain policy.  If the tag is not present,
relaxed authentication enables only exact, parent-child, and child-parent
relationships.

Doug Foster
_______________________________________________
dmarc mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dmarc

Reply via email to