SMTP security via opportunistic DANE TLS

This memo specifies a new connection security model for Message Transfer Agents (MTAs). This model is motivated by key features of inter-domain SMTP delivery, in particular the fact that the destination server is selected indirectly via DNS Mail Exchange (MX) records and that neither email addresses nor MX hostnames signal a requirement for either secure or cleartext transport. Therefore, aside from a few manually configured exceptions, SMTP transport security is of necessity opportunistic (for a definition of "Opportunistic Security" see ). This specification uses the presence of DANE TLSA records to securely signal TLS support and to publish the means by which SMTP clients can successfully authenticate legitimate SMTP servers. This becomes "opportunistic DANE TLS" and is resistant to downgrade and man-in-the-middle (MITM) attacks. It enables an incremental transition of the email backbone to authenticated TLS delivery, with increased global protection as adoption increases. With opportunistic DANE TLS, traffic from SMTP clients to domains that publish "usable" DANE TLSA records in accordance with this memo is authenticated and encrypted. Traffic from legacy clients or to domains that do not publish TLSA records will continue to be sent in the same manner as before, via manually configured security, (pre-DANE) opportunistic TLS or just cleartext SMTP. Problems with existing use of TLS in MTA to MTA SMTP that motivate this specification are described in . The specification itself follows in and which describe respectively how to locate and use DANE TLSA records with SMTP. In , we discuss application of DANE TLS to destinations for which channel integrity and confidentiality are mandatory. In we briefly comment on potential applicability of this specification to Message User Agents.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in . The following terms or concepts are used through the document: Active modification of network traffic by an adversary able to thereby compromise the confidentiality or integrity of the data. DNSSEC validation results, as defined in Section 4.3 of . Capabilities of the stub resolver in use as defined in ; note that this specification requires the use of a Security-Aware Stub Resolver. Best-effort use of TLS that is generally vulnerable to DNS forgery and STARTTLS downgrade attacks. When a TLS-encrypted communication channel is not available, message transmission takes place in the clear. MX record indirection generally precludes authentication even when TLS is available. Best-effort use of TLS, resistant to downgrade attacks for destinations with DNSSEC-validated TLSA records. When opportunistic DANE TLS is determined to be unavailable, clients should fall back to opportunistic TLS. Opportunistic DANE TLS requires support for DNSSEC, DANE and STARTTLS on the client side and STARTTLS plus a DNSSEC published TLSA record on the server side. (Special case of definition). One of the domain names associated by the SMTP client with the destination SMTP server for performing name checks on the server certificate. When name checks are applicable, at least one of the reference identifiers MUST match an DNS-ID (or if none are present the CN-ID) of the server certificate (see ). The RRDATA of an MX record consists of a 16 bit preference followed by a Mail Exchange domain name (see , Section 3.3.9). We will use the term "MX hostname" to refer to the latter, that is, the DNS domain name found after the preference value in an MX record. Thus an "MX hostname" is specifically a reference to a DNS domain name, rather than any host that bears that name. Email delivery is a multi-hop store & forward process. When an MTA is unable to forward a message that may become deliverable later the message is queued and delivery is retried periodically. Some MTAs may be configured with a fallback next-hop destination that handles messages that the MTA would otherwise queue and retry. When a fallback next-hop is configured, messages that would otherwise have to be delayed may be sent to the fallback next-hop destination instead. The fallback destination may itself be subject to opportunistic or mandatory DANE TLS () as though it were the original message destination. The logical destination for mail delivery. By default this is the domain portion of the recipient address, but MTAs may be configured to forward mail for some or all recipients via designated relays. The original next hop destination is, respectively, either the recipient domain or the associated configured relay. Message Transfer Agent (, Section 4.3.2). Message Submission Agent (, Section 4.3.1). Message User Agent (, Section 4.2.1). A DNS Resource Record as defined in , Section 3.6. An RRSet (, Section 5) is a group of DNS resource records that share the same label, class and type.

The Domain Name System Security Extensions (DNSSEC) add data origin authentication, data integrity and data non-existence proofs to the Domain Name System (DNS). DNSSEC is defined in , and . As described in the introduction of , TLS authentication via the existing public Certification Authority (CA) PKI suffers from an over-abundance of trusted parties capable of issuing certificates for any domain of their choice. DANE leverages the DNSSEC infrastructure to publish public keys and certificates for use with the Transport Layer Security (TLS) protocol via the "TLSA" DNS record type. With DNSSEC each domain can only vouch for the keys of its delegated sub-domains. The TLS protocol enables secure TCP communication. In the context of this memo, channel security is assumed to be provided by TLS. Used without authentication, TLS provides only privacy protection against eavesdropping attacks. Otherwise, TLS also provides data origin authentication to guard against MITM attacks.

With HTTPS, Transport Layer Security (TLS) employs X.509 certificates issued by one of the many Certification Authorities (CAs) bundled with popular web browsers to allow users to authenticate their "secure" websites. Before we specify a new DANE TLS security model for SMTP, we will explain why a new security model is needed. In the process, we will explain why the familiar HTTPS security model is inadequate to protect inter-domain SMTP traffic. The subsections below outline four key problems with applying traditional Web PKI to SMTP that are addressed by this specification. Since SMTP channel security policy is not explicitly specified in either the recipient address or the MX record, a new signaling mechanism is required to indicate when channel security is possible and should be used. The publication of TLSA records allows server operators to securely signal to SMTP clients that TLS is available and should be used. DANE TLSA makes it possible to simultaneously discover which destination domains support secure delivery via TLS and how to verify the authenticity of the associated SMTP services, providing a path forward to ubiquitous SMTP channel security.

The Simple Mail Transfer Protocol (SMTP) is a single-hop protocol in a multi-hop store & forward email delivery process. An SMTP envelope recipient address does not correspond to a specific transport-layer endpoint address, rather at each relay hop the transport-layer endpoint is the next-hop relay, while the envelope recipient address typically remains the same. Unlike the Hypertext Transfer Protocol (HTTP) and its corresponding secured version, HTTPS, where the use of TLS is signaled via the URI scheme, email recipient addresses do not directly signal transport security policy. Indeed, no such signaling could work well with SMTP since TLS encryption of SMTP protects email traffic on a hop-by-hop basis while email addresses could only express end-to-end policy. With no mechanism available to signal transport security policy, SMTP relays employ a best-effort "opportunistic" security model for TLS. A single SMTP server TCP listening endpoint can serve both TLS and non-TLS clients; the use of TLS is negotiated via the SMTP STARTTLS command (). The server signals TLS support to the client over a cleartext SMTP connection, and, if the client also supports TLS, it may negotiate a TLS encrypted channel to use for email transmission. The server's indication of TLS support can be easily suppressed by an MITM attacker. Thus pre-DANE SMTP TLS security can be subverted by simply downgrading a connection to cleartext. No TLS security feature can prevent this. The attacker can simply disable TLS.

With SMTP, DNS Mail Exchange (MX) records abstract the next-hop transport endpoint and allow administrators to specify a set of target servers to which SMTP traffic should be directed for a given domain. A TLS client is vulnerable to MITM attacks unless it verifies that the server's certificate binds the public key to a name that matches one of the client's reference identifiers. A natural choice of reference identifier is the server's domain name. However, with SMTP, server names are not directly encoded in the recipient address, instead they are obtained indirectly via MX records. Without DNSSEC, the MX lookup is vulnerable to MITM and DNS cache poisoning attacks. Active attackers can forge DNS replies with fake MX records and can redirect email to servers with names of their choice. Therefore, secure verification of SMTP TLS certificates matching the server name is not possible without DNSSEC. One might try to harden TLS for SMTP against DNS attacks by using the envelope recipient domain as a reference identifier and by requiring each SMTP server to possess a trusted certificate for the envelope recipient domain rather than the MX hostname. Unfortunately, this is impractical as email for many domains is handled by third parties that are not in a position to obtain certificates for all the domains they serve. Deployment of the Server Name Indication (SNI) extension to TLS (see Section 3) is no panacea, since SNI key management is operationally challenging except when the email service provider is also the domain's registrar and its certificate issuer; this is rarely the case for email. Since the recipient domain name cannot be used as the SMTP server reference identifier, and neither can the MX hostname without DNSSEC, large-scale deployment of authenticated TLS for SMTP requires that the DNS be secure. Since SMTP security depends critically on DNSSEC, it is important to point out that consequently SMTP with DANE is the most conservative possible trust model. It trusts only what must be trusted and no more. Adding any other trusted actors to the mix can only reduce SMTP security. A sender may choose to further harden DNSSEC for selected high-value receiving domains by configuring explicit trust anchors for those domains instead of relying on the chain of trust from the root domain. However, detailed discussion of DNSSEC security practices is out of scope for this document.

Sending systems are in some cases explicitly configured to use TLS for mail sent to selected peer domains, but this requires configuring sending MTAs with appropriate subject names or certificate content digests from their peer domains. Due to the resulting administrative burden, such statically configured SMTP secure channels are used rarely (generally only between domains that make bilateral arrangements with their business partners). Internet email, on the other hand, requires regularly contacting new domains for which security configurations cannot be established in advance. The abstraction of the SMTP transport endpoint via DNS MX records, often across organization boundaries, limits the use of public CA PKI with SMTP to a small set of sender-configured peer domains. With little opportunity to use TLS authentication, sending MTAs are rarely configured with a comprehensive list of trusted CAs. SMTP services that support STARTTLS often deploy X.509 certificates that are self-signed or issued by a private CA.

Even if it were generally possible to determine a secure server name, the SMTP client would still need to verify that the server's certificate chain is issued by a trusted Certification Authority (a trust anchor). MTAs are not interactive applications where a human operator can make a decision (wisely or otherwise) to selectively disable TLS security policy when certificate chain verification fails. With no user to "click OK", the MTA's list of public CA trust anchors would need to be comprehensive in order to avoid bouncing mail addressed to sites that employ unknown Certification Authorities. On the other hand, each trusted CA can issue certificates for any domain. If even one of the configured CAs is compromised or operated by an adversary, it can subvert TLS security for all destinations. Any set of CAs is simultaneously both overly inclusive and not inclusive enough.

An SMTP client that implements opportunistic DANE TLS per this specification depends critically on the integrity of DNSSEC lookups, as discussed in . This section lists the DNS resolver requirements needed to avoid downgrade attacks when using opportunistic DANE TLS. A DNS lookup may signal an error or return a definitive answer. A security-aware resolver MUST be used for this specification. Security-aware resolvers will indicate the security status of a DNS RRSet with one of four possible values defined in Section 4.3 of : "secure", "insecure", "bogus" and "indeterminate". In the meaning of the "indeterminate" security status is:

An RRSet for which the resolver is not able to determine whether the RRSet should be signed, as the resolver is not able to obtain the necessary DNSSEC RRs. This can occur when the security-aware resolver is not able to contact security-aware name servers for the relevant zones. Note, the "indeterminate" security status has a conflicting definition in section 5 of .

There is no trust anchor that would indicate that a specific portion of the tree is secure. In this document the term "indeterminate" will be used exclusively in the sense. Therefore, obtaining "indeterminate" lookup results is a (transient) failure condition, namely, the inability to locate the relevant DNS records. DNS records that would be classified "indeterminate" in the sense of are simply classified as "insecure". We do not need to distinguish between zones that lack a suitable ancestor trust anchor, and delegations (ultimately) from a trust-anchor that designate a child zone as being "insecure". All "insecure" RRSets MUST be handled identically: in either case unvalidated data for the query domain is all that is and can be available, and authentication using the data is impossible. As the DNS root zone has been signed, we expect that validating resolvers used by Internet-facing MTAs will be configured with trust anchor data for the root zone, and that therefore domains with no ancestor trust anchor will not be possible in most deployments. As noted in section 4.3 of , a security-aware DNS resolver MUST be able to determine whether a given non-error DNS response is "secure", "insecure", "bogus" or "indeterminate". It is expected that most security-aware stub resolvers will not signal an "indeterminate" security status (in the sense of RFC4035) to the application, and will signal a "bogus" or error result instead. If a resolver does signal an RFC4035 "indeterminate" security status, this MUST be treated by the SMTP client as though a "bogus" or error result had been returned. An MTA making use of a non-validating security-aware stub resolver MAY use the stub resolver's ability, if available, to signal DNSSEC validation status based on information the stub resolver has learned from an upstream validating recursive resolver. Security-Oblivious stub-resolvers () MUST NOT be used. In accordance with section 4.9.3 of :

... a security-aware stub resolver MUST NOT place any reliance on signature validation allegedly performed on its behalf, except when the security-aware stub resolver obtained the data in question from a trusted security-aware recursive name server via a secure channel. To avoid much repetition in the text below, we will pause to explain the handling of "bogus" or "indeterminate" DNSSEC query responses. These are not necessarily the result of a malicious actor; they can, for example, occur when network packets are corrupted or lost in transit. Therefore, "bogus" or "indeterminate" replies are equated in this memo with lookup failure. There is an important non-failure condition we need to highlight in addition to the obvious case of the DNS client obtaining a non-empty "secure" or "insecure" RRSet of the requested type. Namely, it is not an error when either "secure" or "insecure" non-existence is determined for the requested data. When a DNSSEC response with a validation status that is either "secure" or "insecure" reports either no records of the requested type or non-existence of the query domain, the response is not a DNS error condition. The DNS client has not been left without an answer; it has learned that records of the requested type do not exist. Security-aware stub resolvers will, of course, also signal DNS lookup errors in other cases, for example when processing a "ServFail" RCODE, which will not have an associated DNSSEC status. All lookup errors are treated the same way by this specification, regardless of whether they are from a "bogus" or "indeterminate" DNSSEC status or from a more generic DNS error: the information that was requested cannot be obtained by the security-aware resolver at this time. A lookup error is thus a failure to obtain the relevant RRSet if it exists, or to determine that no such RRSet exists when it does not. In contrast to a "bogus" or an "indeterminate" response, an "insecure" DNSSEC response is not an error, rather, as explained above, it indicates that the target DNS zone is either delegated as an "insecure" child of a "secure" parent zone, or is not a descendant of any of the configured DNSSEC trust anchors in use by the SMTP client. "Insecure" results will leave the SMTP client with degraded channel security, but do not stand in the way of message delivery. See section for further details.

When a DNS lookup failure (error or "bogus" or "indeterminate" as defined above) prevents an SMTP client from determining which SMTP server or servers it should connect to, message delivery MUST be delayed. This naturally includes, for example, the case when a "bogus" or "indeterminate" response is encountered during MX resolution. When multiple MX hostnames are obtained from a successful MX lookup, but a later DNS lookup failure prevents network address resolution for a given MX hostname, delivery may proceed via any remaining MX hosts. When a particular SMTP server is securely identified as the delivery destination, a set of DNS lookups () MUST be performed to locate any related TLSA records. If any DNS queries used to locate TLSA records fail (be it due to "bogus" or "indeterminate" records, timeouts, malformed replies, ServFails, etc.), then the SMTP client MUST treat that server as unreachable and MUST NOT deliver the message via that server. If no servers are reachable, delivery is delayed. In what follows, we will only describe what happens when all relevant DNS queries succeed. If any DNS failure occurs, the SMTP client MUST behave as described in this section, by skipping the problem SMTP server, or the problem destination. Queries for candidate TLSA records are explicitly part of "all relevant DNS queries" and SMTP clients MUST NOT continue to connect to an SMTP server or destination whose TLSA record lookup fails.

SMTP clients that employ opportunistic DANE TLS to secure connections to SMTP servers MUST NOT use Security-Oblivious () stub-resolvers. A note about DNAME aliases: a query for a domain name whose ancestor domain is a DNAME alias returns the DNAME RR for the ancestor domain along with a CNAME that maps the query domain to the corresponding sub-domain of the target domain of the DNAME alias . Therefore, whenever we speak of CNAME aliases, we implicitly allow for the possibility that the alias in question is the result of an ancestor domain DNAME record. Consequently, no explicit support for DNAME records is needed in SMTP software; it is sufficient to process the resulting CNAME aliases. DNAME records only require special processing in the validating stub-resolver library that checks the integrity of the combined DNAME + CNAME reply. When DNSSEC validation is handled by a local caching resolver, rather than the MTA itself, even that part of the DNAME support logic is outside the MTA. When a stub resolver returns a response containing a CNAME alias that does not also contain the corresponding query results for the target of the alias, the SMTP client will need to repeat the query at the target of the alias, and should do so recursively up to some configured or implementation-dependent recursion limit. If at any stage of CNAME expansion an error is detected, the lookup of the original requested records MUST be considered to have failed. Whether a chain of CNAME records was returned in a single stub resolver response or via explicit recursion by the SMTP client, if at any stage of recursive expansion an "insecure" CNAME record is encountered, then it and all subsequent results (in particular, the final result) MUST be considered "insecure" regardless of whether any earlier CNAME records leading to the "insecure" record were "secure". Note that a security-aware non-validating stub resolver may return to the SMTP client an "insecure" reply received from a validating recursive resolver that contains a CNAME record along with additional answers recursively obtained starting at the target of the CNAME. In this case, the only possible conclusion is that some record in the set of records returned is "insecure", and it is in fact possible that the initial CNAME record and a subset of the subsequent records are "secure". If the SMTP client needs to determine the security status of the DNS zone containing the initial CNAME record, it may need to issue a separate query of type "CNAME" that returns only the initial CNAME record. In particular in when insecure A or AAAA records are found for an SMTP server via a CNAME alias, it may be necessary to perform an additional CNAME query to determine whether the DNS zone in which the alias is published is signed.

As noted previously (in ), opportunistic TLS with SMTP servers that advertise TLS support via STARTTLS is subject to an MITM downgrade attack. Also some SMTP servers that are not, in fact, TLS capable erroneously advertise STARTTLS by default and clients need to be prepared to retry cleartext delivery after STARTTLS fails. In contrast, DNSSEC validated TLSA records MUST NOT be published for servers that do not support TLS. Clients can safely interpret their presence as a commitment by the server operator to implement TLS and STARTTLS. This memo defines four actions to be taken after the search for a TLSA record returns secure usable results, secure unusable results, insecure or no results or an error signal. The term "usable" in this context is in the sense of Section 4.1 of . Specifically, if the DNS lookup for a TLSA record returns: Any connection to the MTA MUST employ TLS encryption and MUST authenticate the SMTP server using the techniques discussed in the rest of this document. Failure to establish an authenticated TLS connection MUST result in falling back to the next SMTP server or delayed delivery. Any connection to the MTA MUST be made via TLS, but authentication is not required. Failure to establish an encrypted TLS connection MUST result in falling back to the next SMTP server or delayed delivery. A connection to the MTA SHOULD be made using (pre-DANE) opportunistic TLS, this includes using cleartext delivery when the remote SMTP server does not appear to support TLS. The MTA MAY retry in cleartext when delivery via TLS fails either during the handshake or even during data transfer. Lookup errors, including "bogus" and "indeterminate", as explained in MUST result in falling back to the next SMTP server or delayed delivery. An SMTP client MAY be configured to mandate DANE verified delivery for some destinations. With mandatory DANE TLS (), delivery proceeds only when "secure" TLSA records are used to establish an encrypted and authenticated TLS channel with the SMTP server. When the original next-hop destination is an address literal, rather than a DNS domain, DANE TLS does not apply. Delivery proceeds using any relevant security policy configured by the MTA administrator. Similarly, when an MX RRSet incorrectly lists a network address in lieu of an MX hostname, if an MTA chooses to connect to the network address in the non-conformant MX record, DANE TLSA does not apply for such a connection. In the subsections that follow we explain how to locate the SMTP servers and the associated TLSA records for a given next-hop destination domain. We also explain which name or names are to be used in identity checks of the SMTP server certificate.

In this section we consider next-hop domains that are subject to MX resolution and have MX records. The TLSA records and the associated base domain are derived separately for each MX hostname that is used to attempt message delivery. DANE TLS can authenticate message delivery to the intended next-hop domain only when the MX records are obtained securely via a DNSSEC validated lookup. MX records MUST be sorted by preference; an MX hostname with a worse (numerically higher) MX preference that has TLSA records MUST NOT preempt an MX hostname with a better (numerically lower) preference that has no TLSA records. In other words, prevention of delivery loops by obeying MX preferences MUST take precedence over channel security considerations. Even with two equal-preference MX records, an MTA is not obligated to choose the MX hostname that offers more security. Domains that want secure inbound mail delivery need to ensure that all their SMTP servers and MX records are configured accordingly. In the language of Section 5.1, the original next-hop domain is the "initial name". If the MX lookup of the initial name results in a CNAME alias, the MTA replaces the initial name with the resulting name and performs a new lookup with the new name. MTAs typically support recursion in CNAME expansion, so this replacement is performed repeatedly (up to the MTA's recursion limit) until the ultimate non-CNAME domain is found. If the MX RRSet (or any CNAME leading to it) is "insecure" (see ), DANE TLS need not apply, and delivery MAY proceed via pre-DANE opportunistic TLS. That said, the protocol in this memo is an "opportunistic security" protocol, meaning that it strives to communicate with each peer as securely as possible, while maintaining broad interoperability. Therefore, the SMTP client MAY proceed to use DANE TLS (as described in below) even with MX hosts obtained via an "insecure" MX RRSet. For example, when a hosting provider has a signed DNS zone and publishes TLSA records for its SMTP servers, hosted domains that are not signed may still benefit from the provider's TLSA records. Deliveries via the provider's SMTP servers will not be subject to active attacks when sending SMTP clients elect to make use of the provider's TLSA records. When the MX records are not (DNSSEC) signed, an active attacker can redirect SMTP clients to MX hosts of his choice. Such redirection is tamper-evident when SMTP servers found via "insecure" MX records are recorded as the next-hop relay in the MTA delivery logs in their original (rather than CNAME expanded) form. Sending MTAs SHOULD log unexpanded MX hostnames when these result from insecure MX lookups. Any successful authentication via an insecurely determined MX host MUST NOT be misrepresented in the mail logs as secure delivery to the intended next-hop domain. When DANE TLS is mandatory () for a given destination, delivery MUST be delayed when the MX RRSet is not "secure". If the MX RRset is not "insecure", then — assuming no DNS lookup errors () — the MX RRSet is "secure", and the SMTP client MUST treat each MX hostname as a separate non-MX destination for opportunistic DANE TLS (as described in ). When, for a given MX hostname, no TLSA records are found, or only "insecure" TLSA records are found, DANE TLSA is not applicable with the SMTP server in question and delivery proceeds to that host as with pre-DANE opportunistic TLS. To avoid downgrade attacks, any errors during TLSA lookups MUST, as explained in , cause the SMTP server in question to be treated as unreachable.

This section describes the algorithm used to locate the TLSA records and associated TLSA base domain for an input domain that is not subject to MX resolution or that lacks MX records. Such domains include: Any host configured by the sending MTA administrator as the next-hop relay for some or all domains, that is not subject to MX resolution. When a domain has MX records, we treat each MX host listed in those MX records as though it were a non-MX destination. That is, in the same way as we would treat an administrator-configured relay that handles mail for that domain. (Unlike administrator-specified relays, MTAs are not required to support CNAME expansion of next-hop names found via MX lookups). A next-hop destination domain subject to MX resolution that has no MX records. In this case the domain's name is implicitly also its sole SMTP server name. Note that DNS queries with type TLSA are mishandled by load balancing nameservers that serve the MX hostnames of some large email providers. The DNS zones served by these nameservers are not signed and contain no TLSA records, but queries for TLSA records fail, rather than returning the non-existence of the requested TLSA records. To avoid problems delivering mail to domains whose SMTP servers are served by the problem nameservers the SMTP client MUST perform any A and/or AAAA queries for the destination before attempting to locate the associated TLSA records. This lookup is needed in any case to determine whether the destination domain is reachable and the DNSSEC validation status of the chain of CNAME queries required to reach the ultimate address records. If no address records are found, the destination is unreachable. If address records are found, but the DNSSEC validation status of the first query response is "insecure" (see ), the SMTP client SHOULD NOT proceed to search for any associated TLSA records. With the problem domains, TLSA queries will lead to DNS lookup errors and cause messages to be consistently delayed and ultimately returned to the sender. We don't expect to find any "secure" TLSA records associated with a TLSA base domain that lies in an unsigned DNS zone. Therefore, skipping TLSA lookups in this case will also reduce latency with no detrimental impact on security. If the A and/or AAAA lookup of the "initial name" yields a CNAME, we replace it with the resulting name as if it were the initial name and perform a lookup again using the new name. This replacement is performed recursively (up to the MTA's recursion limit). We consider the following cases for handling a DNS response for an A or AAAA DNS lookup: When the DNS queries for A and/or AAAA records yield neither a list of addresses nor a CNAME (or CNAME expansion is not supported) the destination is unreachable. The answer is not a CNAME alias. If the address RRSet is "secure", TLSA lookups are performed as described in with the initial name as the candidate TLSA base domain. If no "secure" TLSA records are found, DANE TLS is not applicable and mail delivery proceeds with pre-DANE opportunistic TLS (which, being best-effort, degrades to cleartext delivery when STARTTLS is not available or the TLS handshake fails). The input domain is a CNAME alias, but the ultimate network address RRSet is "insecure" (see ). If the initial CNAME response is also "insecure", DANE TLS does not apply. Otherwise, this case is treated just like the non-CNAME case above, where a search is performed for a TLSA record with the original input domain as the candidate TLSA base domain. The input domain is a CNAME alias, and the ultimate network address RRSet is "secure" (see ). Two candidate TLSA base domains are tried: the fully CNAME-expanded initial name and, failing that, then the initial name itself. In summary, if it is possible to securely obtain the full, CNAME-expanded, DNSSEC-validated address records for the input domain, then that name is the preferred TLSA base domain. Otherwise, the unexpanded input-MX domain is the candidate TLSA base domain. When no "secure" TLSA records are found at either the CNAME-expanded or unexpanded domain, then DANE TLS does not apply for mail delivery via the input domain in question. And, as always, errors, bogus or indeterminate results for any query in the process MUST result in delaying or abandoning delivery.

Each candidate TLSA base domain (the original or fully CNAME-expanded name of a non-MX destination or a particular MX hostname of an MX destination) is in turn prefixed with service labels of the form "_<port>._tcp". The resulting domain name is used to issue a DNSSEC query with the query type set to TLSA ( Section 7.1). For SMTP, the destination TCP port is typically 25, but this may be different with custom routes specified by the MTA administrator in which case the SMTP client MUST use the appropriate number in the "_<port>" prefix in place of "_25". If, for example, the candidate base domain is "mx.example.com", and the SMTP connection is to port 25, the TLSA RRSet is obtained via a DNSSEC query of the form:

_25._tcp.mx.example.com. IN TLSA ? The query response may be a CNAME, or the actual TLSA RRSet. If the response is a CNAME, the SMTP client (through the use of its security-aware stub resolver) restarts the TLSA query at the target domain, following CNAMEs as appropriate and keeping track of whether the entire chain is "secure". If any "insecure" records are encountered, or the TLSA records don't exist, the next candidate TLSA base domain is tried instead. If the ultimate response is a "secure" TLSA RRSet, then the candidate TLSA base domain will be the actual TLSA base domain and the TLSA RRSet will constitute the TLSA records for the destination. If none of the candidate TLSA base domains yield "secure" TLSA records then delivery MAY proceed via pre-DANE opportunistic TLS. SMTP clients MAY elect to use "insecure" TLSA records to avoid STARTTLS downgrades or even to skip SMTP servers that fail authentication, but MUST NOT misrepresent authentication success as either a secure connection to the SMTP server or as a secure delivery to the intended next-hop domain. TLSA record publishers may leverage CNAMEs to reference a single authoritative TLSA RRSet specifying a common Certification Authority or a common end entity certificate to be used with multiple TLS services. Such CNAME expansion does not change the SMTP client's notion of the TLSA base domain; thus, when _25._tcp.mx.example.com is a CNAME, the base domain remains mx.example.com and this is still the reference identifier used together with the next-hop domain in peer certificate name checks. Note that shared end entity certificate associations expose the publishing domain to substitution attacks, where an MITM attacker can reroute traffic to a different server that shares the same end entity certificate. Such shared end entity TLSA records SHOULD be avoided unless the servers in question are functionally equivalent or employ mutually incompatible protocols (an active attacker gains nothing by diverting client traffic from one such server to another). A better example, employing a shared trust anchor rather than shared end-entity certificates, is illustrated by the DNSSEC validated records below:

This section describes which TLSA records are applicable to SMTP opportunistic DANE TLS and how to apply such records to authenticate the SMTP server. With opportunistic DANE TLS, both the TLS support implied by the presence of DANE TLSA records and the verification parameters necessary to authenticate the TLS peer are obtained together. In contrast to protocols where channel security policy is set exclusively by the client, authentication via this protocol is expected to be less prone to connection failure caused by incompatible configuration of the client and server.

The DANE TLSA specification defines multiple TLSA RR types via combinations of 3 numeric parameters. The numeric values of these parameters were later given symbolic names in . The rest of the TLSA record is the "certificate association data field", which specifies the full or digest value of a certificate or public key. Since opportunistic DANE TLS will be used by non-interactive MTAs, with no user to "press OK" when authentication fails, reliability of peer authentication is paramount. Server operators are advised to publish TLSA records that are least likely to fail authentication due to interoperability or operational problems. Because DANE TLS relies on coordinated changes to DNS and SMTP server settings, the best choice of records to publish will depend on site-specific practices. The certificate usage element of a TLSA record plays a critical role in determining how the corresponding certificate association data field is used to authenticate server's certificate chain. The next two subsections explain the process for certificate usages DANE-EE(3) and DANE-TA(2). The third subsection briefly explains why certificate usages PKIX-TA(0) and PKIX-EE(1) are not applicable with opportunistic DANE TLS. In summary, we RECOMMEND the use of either "DANE-EE(3) SPKI(1) SHA2-256(1)" or "DANE-TA(2) Cert(0) SHA2-256(1)" TLSA records depending on site needs. Other combinations of TLSA parameters are either explicitly unsupported, or offer little to recommend them over these two.

Authentication via certificate usage DANE-EE(3) TLSA records involves simply checking that the server's leaf certificate matches the TLSA record. In particular the binding of the server public key to its name is based entirely on the TLSA record association. The server MUST be considered authenticated even if none of the names in the certificate match the client's reference identity for the server. Similarly, the expiration date of the server certificate MUST be ignored, the validity period of the TLSA record key binding is determined by the validity interval of the TLSA record DNSSEC signature. With DANE-EE(3) servers need not employ SNI (may ignore the client's SNI message) even when the server is known under independent names that would otherwise require separate certificates. It is instead sufficient for the TLSA RRSets for all the domains in question to match the server's default certificate. Of course with SMTP servers it is simpler still to publish the same MX hostname for all the hosted domains. For domains where it is practical to make coordinated changes in DNS TLSA records during SMTP server key rotation, it is often best to publish end-entity DANE-EE(3) certificate associations. DANE-EE(3) certificates don't suddenly stop working when leaf or intermediate certificates expire, and don't fail when the server operator neglects to configure all the required issuer certificates in the server certificate chain. TLSA records published for SMTP servers SHOULD, in most cases, be "DANE-EE(3) SPKI(1) SHA2-256(1)" records. Since all DANE implementations are required to support SHA2-256, this record type works for all clients and need not change across certificate renewals with the same key.

Some domains may prefer to avoid the operational complexity of publishing unique TLSA RRs for each TLS service. If the domain employs a common issuing Certification Authority to create certificates for multiple TLS services, it may be simpler to publish the issuing authority as a trust anchor (TA) for the certificate chains of all relevant services. The TLSA query domain (TLSA base domain with port and protocol prefix labels) for each service issued by the same TA may then be set to a CNAME alias that points to a common TLSA RRSet that matches the TA. For example:

example.com. IN MX 0 mx1.example.com. example.com. IN MX 0 mx2.example.com. _25._tcp.mx1.example.com. IN CNAME tlsa201._dane.example.com. _25._tcp.mx2.example.com. IN CNAME tlsa201._dane.example.com. tlsa201._dane.example.com. IN TLSA 2 0 1 e3b0c44298fc1c14.... With usage DANE-TA(2) the server certificates will need to have names that match one of the client's reference identifiers (see ). The server MAY employ SNI to select the appropriate certificate to present to the client. SMTP servers that rely on certificate usage DANE-TA(2) TLSA records for TLS authentication MUST include the TA certificate as part of the certificate chain presented in the TLS handshake server certificate message even when it is a self-signed root certificate. Many SMTP servers are not configured with a comprehensive list of trust anchors, nor are they expected to at any point in the future. Some MTAs will ignore all locally trusted certificates when processing usage DANE-TA(2) TLSA records. Thus even when the TA happens to be a public Certification Authority known to the SMTP client, authentication is likely to fail unless the TA certificate is included in the TLS server certificate message. With some SMTP server software it is not possible to configure the server to include self-signed (root) CA certificates in the server certificate chain. Such servers MUST either publish DANE-TA(2) records for an intermediate certificate or MUST instead use DANE-EE(3) TLSA records. TLSA records with matching type Full(0) are discouraged. While these potentially obviate the need to transmit the TA certificate in the TLS server certificate message, client implementations may not be able to augment the server certificate chain with the data obtained from DNS, especially when the TLSA record supplies a bare key (selector SPKI(1)). Since the server will need to transmit the TA certificate in any case, server operators SHOULD publish TLSA records with a matching type other than Full(0) and avoid potential interoperability issues with large TLSA records containing full certificates or keys. TLSA Publishers employing DANE-TA(2) records SHOULD publish records with a selector of Cert(0). Such TLSA records are associated with the whole trust anchor certificate, not just with the trust anchor public key. In particular, the SMTP client SHOULD then apply any relevant constraints from the trust anchor certificate, such as, for example, path length constraints. While a selector of SPKI(1) may also be employed, the resulting TLSA record will not specify the full trust anchor certificate content, and elements of the trust anchor certificate other than the public key become mutable. This may, for example, allow a subsidiary CA to issue a chain that violates the trust anchor's path length or name constraints.

Note, this section applies to MTA-to-MTA SMTP on port 25. That is, to servers that are the SMTP servers for one or more destination domains. Other uses of SMTP, such as in MUA-to-MSA submission on ports 587 or 465 are out of scope for this document. Where those other uses also employ TLS opportunistically and/or depend on DNSSEC as a result of DNS-based discovery of service location, the relevant specifications should, as appropriate, arrive at similar conclusions. As noted in and , sending MTAs cannot, without relying on DNSSEC for secure MX records and DANE for STARTTLS support signaling, perform server identity verification or prevent STARTTLS downgrade attacks. The use of PKIX CAs offers no added security since an attacker capable of compromising DNSSEC is free to replace any PKIX-TA(0) or PKIX-EE(1) TLSA records with records bearing any convenient non-PKIX certificate usage. Finally, as explained in , there is no list of trusted CAs agreed by all MTAs, and no user to "click OK" when a server's CA is not trusted by a client. Therefore, TLSA records for the port 25 SMTP service used by client MTAs SHOULD NOT include TLSA RRs with certificate usage PKIX-TA(0) or PKIX-EE(1). SMTP client MTAs cannot be expected to be configured with a suitably complete set of trusted public CAs. Lacking a complete set of public CAs, MTA clients would not be able to verify the certificates of SMTP servers whose issuing root CAs are not trusted by the client. Opportunistic DANE TLS needs to interoperate without bilateral coordination of security settings between client and server systems. Therefore, parameter choices that are fragile in the absence of bilateral coordination are unsupported. Nothing is lost since the PKIX certificate usages cannot aid SMTP TLS security, they can only impede SMTP TLS interoperability. SMTP client treatment of TLSA RRs with certificate usages PKIX-TA(0) or PKIX-EE(1) is undefined. As with any other unsupported certificate usage, SMTP clients MAY treat such records as "unusable"

When at least one usable "secure" TLSA record is found, the SMTP client MUST use TLSA records to authenticate the SMTP server. Messages MUST NOT be delivered via the SMTP server if authentication fails, otherwise the SMTP client is vulnerable to MITM attacks.

The SMTP client MUST NOT perform certificate name checks with certificate usage DANE-EE(3); see above.

To match a server via a TLSA record with certificate usage DANE-TA(2), the client MUST perform name checks to ensure that it has reached the correct server. In all DANE-TA(2) cases the SMTP client MUST employ the TLSA base domain as the primary reference identifier for matching the server certificate. If the TLSA base domain was obtained indirectly via a "secure" MX lookup (including any CNAME-expanded name of an MX hostname), then the original next-hop domain used in the MX lookup MUST be included as as a second reference identifier. The CNAME-expanded original next-hop domain MUST be included as a third reference identifier if different from the original next-hop domain. When the client MTA is employing DANE TLS security despite "insecure" MX redirection the MX hostname is the only reference identifier. If MX records were not used (e.g., if none exist) and the TLSA base domain is the CNAME-expanded original next-hop domain, then the original next-hop domain MUST be included as a second reference identifier. Accepting certificates with the original next-hop domain in addition to the MX hostname allows a domain with multiple MX hostnames to field a single certificate bearing a single domain name (i.e., the email domain) across all the SMTP servers. This also aids interoperability with pre-DANE SMTP clients that are configured to look for the email domain name in server certificates. For example, with "secure" DNS records as below:

exchange.example.org. IN CNAME mail.example.org. mail.example.org. IN CNAME example.com. example.com. IN MX 10 mx10.example.com. example.com. IN MX 15 mx15.example.com. example.com. IN MX 20 mx20.example.com. ; mx10.example.com. IN A 192.0.2.10 _25._tcp.mx10.example.com. IN TLSA 2 0 1 ... ; mx15.example.com. IN CNAME mxbackup.example.com. mxbackup.example.com. IN A 192.0.2.15 ; _25._tcp.mxbackup.example.com. IN TLSA ? (NXDOMAIN) _25._tcp.mx15.example.com. IN TLSA 2 0 1 ... ; mx20.example.com. IN CNAME mxbackup.example.net. mxbackup.example.net. IN A 198.51.100.20 _25._tcp.mxbackup.example.net. IN TLSA 2 0 1 ... Certificate name checks for delivery of mail to exchange.example.org via any of the associated SMTP servers MUST accept at least the names "exchange.example.org" and "example.com", which are respectively the original and fully expanded next-hop domain. When the SMTP server is mx10.example.com, name checks MUST accept the TLSA base domain "mx10.example.com". If, despite the fact that MX hostnames are required to not be aliases, the MTA supports delivery via "mx15.example.com" or "mx20.example.com" then name checks MUST accept the respective TLSA base domains "mx15.example.com" and "mxbackup.example.net".

When name checks are applicable (certificate usage DANE-TA(2)), if the server certificate contains a Subject Alternative Name extension (), with at least one DNS-ID () then only the DNS-IDs are matched against the client's reference identifiers. The CN-ID () is only considered when no DNS-IDs are present. The server certificate is considered matched when one of its presented identifiers () matches any of the client's reference identifiers. Wildcards are valid in either DNS-IDs or the CN-ID when applicable. The wildcard character must be the entire first label of the DNS-ID or CN-ID. Thus, "*.example.com" is valid, while "smtp*.example.com" and "*smtp.example.com" are not. SMTP clients MUST support wildcards that match the first label of the reference identifier, with the remaining labels matching verbatim. For example, the DNS-ID "*.example.com" matches the reference identifier "mx1.example.com". SMTP clients MAY, subject to local policy allow wildcards to match multiple reference identifier labels, but servers cannot expect broad support for such a policy. Therefore any wildcards in server certificates SHOULD match exactly one label in either the TLSA base domain or the next-hop domain.

Two TLSA records MUST be published before employing a new EE or TA public key or certificate, one matching the currently deployed key and the other matching the new key scheduled to replace it. Once sufficient time has elapsed for all DNS caches to expire the previous TLSA RRSet and related signature RRsets, servers may be configured to use the new EE private key and associated public key certificate or may employ certificates signed by the new trust anchor. Once the new public key or certificate is in use, the TLSA RR that matches the retired key can be removed from DNS, leaving only RRs that match keys or certificates in active use. As described in , when server certificates are validated via a DANE-TA(2) trust anchor, and CNAME records are employed to store the TA association data at a single location, the responsibility of updating the TLSA RRSet shifts to the operator of the trust anchor. Before a new trust anchor is used to sign any new server certificates, its certificate (digest) is added to the relevant TLSA RRSet. After enough time elapses for the original TLSA RRSet to age out of DNS caches, the new trust anchor can start issuing new server certificates. Once all certificates issued under the previous trust anchor have expired, its associated RRs can be removed from the TLSA RRSet. In the DANE-TA(2) key management model server operators do not generally need to update DNS TLSA records after initially creating a CNAME record that references the centrally operated DANE-TA(2) RRSet. If a particular server's key is compromised, its TLSA CNAME SHOULD be replaced with a DANE-EE(3) association until the certificate for the compromised key expires, at which point it can return to using a CNAME record. If the central trust anchor is compromised, all servers need to be issued new keys by a new TA, and an updated DANE-TA(2) TLSA RRSet needs to be published containing just the new TA. SMTP servers cannot expect broad CRL or OCSP support from SMTP clients. As outlined above, with DANE, compromised server or trust anchor keys can be "revoked" by removing them from the DNS without the need for client-side support for OCSP or CRLs.

While specifies multiple digest algorithms, it does not specify a protocol by which the SMTP client and TLSA record publisher can agree on the strongest shared algorithm. Such a protocol would allow the client and server to avoid exposure to any deprecated weaker algorithms that are published for compatibility with less capable clients, but should be ignored when possible. Such a protocol is specified in . SMTP clients and servers that implement this specification MUST comply with the requirements outlined under "Digest Algorithm Agility" in .

An MTA implementing this protocol may require a stronger security assurance when sending email to selected destinations. The sending organization may need to send sensitive email and/or may have regulatory obligations to protect its content. This protocol is not in conflict with such a requirement, and in fact can often simplify authenticated delivery to such destinations. Specifically, with domains that publish DANE TLSA records for their MX hostnames, a sending MTA can be configured to use the receiving domains's DANE TLSA records to authenticate the corresponding SMTP server. Authentication via DANE TLSA records is easier to manage, as changes in the receiver's expected certificate properties are made on the receiver end and don't require manually communicated configuration changes. With mandatory DANE TLS, when no usable TLSA records are found, message delivery is delayed. Thus, mail is only sent when an authenticated TLS channel is established to the remote SMTP server. Administrators of mail servers that employ mandatory DANE TLS, need to carefully monitor their mail logs and queues. If a partner domain unwittingly misconfigures their TLSA records, disables DNSSEC, or misconfigures SMTP server certificate chains, mail will be delayed and may bounce if the issue is not resolved in a timely manner.

We note that the SMTP protocol is also used between Message User Agents (MUAs) and Message Submission Agents (MSAs) . In a protocol is specified that enables an MUA to dynamically locate the MSA based on the user's email address. SMTP connection security considerations for MUAs implementing are largely analogous to connection security requirements for MTAs, and this specification could be applied largely verbatim with DNS MX records replaced by corresponding DNS Service (SRV) records . However, until MUAs begin to adopt the dynamic configuration mechanisms of they are adequately served by more traditional static TLS security policies. Specification of DANE TLS for Message User Agent (MUA) to Message Submission Agent (MSA) SMTP is left to future documents that focus specifically on SMTP security between MUAs and MSAs.

To ensure that the server sends the right certificate chain, the SMTP client MUST send the TLS SNI extension containing the TLSA base domain. This precludes the use of the SSL 2.0 compatible SSL HELLO by the SMTP client. Each SMTP server MUST present a certificate chain (see Section 7.4.2) that matches at least one of the TLSA records. The server MAY rely on SNI to determine which certificate chain to present to the client. Clients that don't send SNI information may not see the expected certificate chain. If the server's TLSA records match the server's default certificate chain, the server need not support SNI. In either case, the server need not include the SNI extension in its TLS HELLO as simply returning a matching certificate chain is sufficient. Servers MUST NOT enforce the use of SNI by clients, as the client may be using unauthenticated opportunistic TLS and may not expect any particular certificate from the server. If the client sends no SNI extension, or sends an SNI extension for an unsupported domain, the server MUST simply send some fallback certificate chain of its choice. The reason for not enforcing strict matching of the requested SNI hostname is that DANE TLS clients are typically willing to accept multiple server names, but can only send one name in the SNI extension. The server's fallback certificate may match a different name acceptable to the client, e.g., the original next-hop domain.

Since many SMTP servers either do not support or do not enable any anonymous TLS cipher suites, SMTP client TLS HELLO messages SHOULD offer to negotiate a typical set of non-anonymous cipher suites required for interoperability with such servers. An SMTP client employing pre-DANE opportunistic TLS MAY in addition include one or more anonymous TLS cipher suites in its TLS HELLO. SMTP servers, that need to interoperate with opportunistic TLS clients SHOULD be prepared to interoperate with such clients by either always selecting a mutually supported non-anonymous cipher suite or by correctly handling client connections that negotiate anonymous cipher suites. Note that while SMTP server operators are under no obligation to enable anonymous cipher suites, no security is gained by sending certificates to clients that will ignore them. Indeed support for anonymous cipher suites in the server makes audit trails more informative. Log entries that record connections that employed an anonymous cipher suite record the fact that the clients did not care to authenticate the server.

An operational error on the sending or receiving side that cannot be corrected in a timely manner may, at times, lead to consistent failure to deliver time-sensitive email. The sending MTA administrator may have to choose between letting email queue until the error is resolved and disabling opportunistic or mandatory DANE TLS () for one or more destinations. The choice to disable DANE TLS security should not be made lightly. Every reasonable effort should be made to determine that problems with mail delivery are the result of an operational error, and not an attack. A fallback strategy may be to configure explicit out-of-band TLS security settings if supported by the sending MTA. SMTP clients may deploy opportunistic DANE TLS incrementally by enabling it only for selected sites, or may occasionally need to disable opportunistic DANE TLS for peers that fail to interoperate due to misconfiguration or software defects on either end. Some implementations MAY support DANE TLS in an "audit only" mode in which failure to achieve the requisite security level is logged as a warning and delivery proceeds at a reduced security level. Unless local policy specifies "audit only" or that opportunistic DANE TLS is not to be used for a particular destination, an SMTP client MUST NOT deliver mail via a server whose certificate chain fails to match at least one TLSA record when usable TLSA records are found for that server.

Some MTAs enable STARTTLS selectively. For example they might only support STARTTLS with clients that have previously demonstrated "proper MTA behavior", for example by retrying the delivery of deferred messages (greylisting). If such an MTA publishes DANE TLSA records, sending MTAs that implement this specification will not attempt the initial cleartext SMTP transaction needed to establish the "proper MTA behavior", because they cannot establish the required channel security. Server operators MUST NOT implement selective STARTTLS if they also want to support DANE TLSA. TLSA Publishers MUST follow the guidelines in the "TLSA Publisher Requirements" section of . TLSA Publishers SHOULD follow the TLSA publication size guidance found in under "DANE DNS Record Size Guidelines". TLSA Publishers SHOULD follow the TLSA record TTL and signature lifetime recommendations found in under "Operational Considerations".

This protocol leverages DANE TLSA records to implement MITM resistant opportunistic security () for SMTP. For destination domains that sign their MX records and publish signed TLSA records for their MX hostnames, this protocol allows sending MTAs to securely discover both the availability of TLS and how to authenticate the destination. This protocol does not aim to secure all SMTP traffic, as that is not practical until DNSSEC and DANE adoption are universal. The incremental deployment provided by following this specification is a best possible path for securing SMTP. This protocol coexists and interoperates with the existing insecure Internet email backbone. The protocol does not preclude existing non-opportunistic SMTP TLS security arrangements, which can continue to be used as before via manual configuration with negotiated out-of-band key and TLS configuration exchanges. Opportunistic SMTP TLS depends critically on DNSSEC for downgrade resistance and secure resolution of the destination name. If DNSSEC is compromised, it is not possible to fall back on the public CA PKI to prevent MITM attacks. A successful breach of DNSSEC enables the attacker to publish TLSA usage 3 certificate associations, and thereby bypass any security benefit the legitimate domain owner might hope to gain by publishing usage 0 or 1 TLSA RRs. Given the lack of public CA PKI support in existing MTA deployments, avoiding certificate usages 0 and 1 simplifies implementation and deployment with no adverse security consequences. Implementations must strictly follow the portions of this specification that indicate when it is appropriate to initiate a non-authenticated connection or cleartext connection to a SMTP server. Specifically, in order to prevent downgrade attacks on this protocol, implementation must not initiate a connection when this specification indicates a particular SMTP server must be considered unreachable.

This specification requires no support from IANA.

The authors would like to extend great thanks to Tony Finch, who started the original version of a DANE SMTP document. His work is greatly appreciated and has been incorporated into this document. The authors would like to additionally thank Phil Pennock for his comments and advice on this document. Acknowledgments from Viktor: Thanks to Paul Hoffman who motivated me to begin work on this memo and provided feedback on early drafts. Thanks to Patrick Koetter, Perry Metzger and Nico Williams for valuable review comments. Thanks also to Wietse Venema who created Postfix, and whose advice and feedback were essential to the development of the Postfix DANE implementation.