]> Secure SMTP/TLS SRV Announcement
steffen@sdaoden.eu
 General Internet Engineering Task Force SMTP TLS Secure SMTP STARTTLS SRV This specification defines a DNS (RFC 1035) SRV (RFC 2782) record that announces TLS (RFC 9325) secured SMTP (RFC 5321, RFC 3207), optionally including Implicit TLS.
Introduction defines a widely adopted DNS-based service discovery protocol. is a specification of SRV records for the email protocols IMAP, POP3, and SUBMISSION. This includes DNS service names for Implicit TLS protocol variants. SMTP connections to MTAs () are yet excluded from their share on the widely supported SRV record, at least from the IETF point of view. (According usage however exists for many years, see for example the German De-Mail definition law; and in recent years notable FOSS MTA implementations gained support for SRV lookups.) Moreover, no Implicit TLS SMTP protocol variant has ever been specified by the IETF, even though the achievable non-batchable packet roundtrip savings for the SMTP protocol are immense, and despite the fact that many (most FOSS) SMTP implementations already have, or can easily implement, built-in support for this mode. This specification adds a SMTP/TLS service name for SRV records.
Conventions and Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here. The term "Implicit TLS" refers to the automatic negotiation of TLS whenever a TCP connection is made on a particular TCP port that is used exclusively by that server for TLS connections. The term "Implicit TLS" is intended to contrast with the use of the STARTTLS command in SMTP that is used by the client and the server to explicitly negotiate TLS on an established cleartext TCP connection. The term "FOSS" refers to Free and Open Source Software.
SMTP/TLS SRV Service Name The service name for TLS enabled Secure SMTP is smtp-tls, the resulting DNS label _smtp-tls.
STARTTLS
Whenever a domain publishes an according DNS SRV resource record it asserts availability of Secure SMTP, that is, of the STARTTLS SMTP service extension on the normal SMTP port, specified by IANA as port 25. The port number MUST be given as 25.
Implicit TLS
If the port number of the published SRV resource record does not equal the IANA port 25,0 the domain announces support for Implicit TLS on the given port in addition to STARTTLS support on the IANA port. Servers SHOULD NOT announce STARTTLS in the EHLO command response of an Implicit TLS connection. If a client issues a STARTTLS extension command during an Implicit TLS connection, the SMTP reply code 554 SHOULD be returned, if enhanced status codes are used, 5.5.1 SHOULD be used.
DNS SRV RRs MUST take priority over MX ones: no MX lookup SHOULD be performed, and no otherwise available MX RR MUST be used in the presence of a SRV RR. After a successful _smtp-tls DNS SRV lookup cleartext communication SHOULD NOT be used. Servers and clients which make use of the _smtp-tls DNS SRV SHOULD follow the guidelines of TLS.
Examples
STARTTLS This example announces support for the STARTTLS SMTP service extension.
Implicit TLS This example specifies port number 842 to indicate an Implicit TLS port in addition to the STARTTLS SMTP service extension.
Prioritized Server Selection A multi-server scenario where the main server supports Implicit TLS, and the backup server only supports STARTTLS.
Guidance for MTAs If, after a successful _smtp-tls DNS SRV lookup that announces Implicit TLS support, connecting to the announced port fails, a connection to the IANA SMTP port 25 SHOULD be established which MUST use the STARTTLS SMTP extension. If none of the shortcomings described next apply, the fallback connection MAY be omitted.
Informative remark: This is meant to overcome two shortcomings: first the given port may be blocked along the network path; it may take time until an IANA registration takes places, and/or the network adapts; whereas expected to be a rare event for the System Ports range (), defining a recover strategy seems useful. Second DNS SRV lookups could return results unprotected by DNSSEC, or without perceived knowledge of whether DNSSEC was actually used, for example, when the DNS is accessed via some kind of furtherly unspecified intermediate proxy that needs to be trusted: in either case the possibility of DNS forge attacks exist; if the STARTTLS secured connection to the IANA SMTP port fails, the DNS result SHOULD be treated with maximum suspicion, if at all applicable, for example by treating it as a negative result (). The mail log record may give useful insight. If the STARTTLS secured connection succeeds, it may make sense to add a caching mechanism in order to avoid retrying Implicit TLS in situations where connections repeatedly cannot be established.
The section of named "Guidance for MUAs" (section 4) in parts also applies to this specification.
Guidance for Service Providers The equally named section of (section 5) also applies to this specification.
IANA Considerations The author does not expect IANA to add a dedicated port for "Implicit TLS SMTP", given that it has registered according ones for other email protocols, for example POP3S, no sooner but 1999, a quarter of a century ago. The author nonetheless asks for assigning a dedicated port for Implicit TLS SMTP, and suggests port 842 in the System Ports range. The author wants to point out that the contra arguments given in section 7 of that created according POP3S and IMAPS assignments in 1999 are contradicted by operational reality in the internet, and here that includes the IETF by means of . A dedicated port enables administrators to apply strict policies, for example in firewalls.
Security Considerations First of all, the equally named section of (section 6) also applies to this specification. Due to fact that one and a half decade passed since RFC 6186 it follows a reiteration of the reasoning. This specification avoids downgrade attacks on the opportunistic approach of STARTTLS, accomplished via the mechanism used for many other IETF standardized protocols, most notably (IMAP, POP3, SUBMISSION). With its Implicit TLS capability it grants the SMTP protocol the same level of confidentiality through TLS as is already standardized for the other email protocols; this is considered a value by itself, even for the possibly lesser sensitive MTA-to-MTA communication. Implicit TLS reduces the number of packet roundtrips, that at a protocol stage where the widely used command pipelining performance improvement extension cannot be used; it must be said that today these roundtrips are anachronistic (also environmentally): for example the about 4.2 million known DANE for SMTP enabled domains at the time of this writing alone, which must use TLS by standard definition, likely generate (several) billion(s) of useless sequential and blocking roundtrip packets each and every day. The security of DNS is out of scope for this specification, but DNSSEC and secure DNS transport etc exists. Selection of the appropriate transport layer security protocol is out of scope for this specification, please see for example TLS.
References Normative References Informative References
Rationale Plenty of methods have been standardized by the IETF to perform service TLS capability discovery for SMTP. They have in common that they represent SMTP specific solutions to a problem that other protocols address by means this document thus reiterates for SMTP. They represent further development effort and error surfaces. They also impose increased permanent administration effort, and, due to this, are inaccessible to a large number, if not the majority, of private email server operators: either through additional costs, the impossibility to set up a dedicated name server, if so required, and, also as a superset of the former, restrictive web interfaces that support only a minimal set of DNS resource record types. Regardless the fact that several of these extensions switch the SMTP protocol to a MUST usage of TLS, they still require the additional STARTTLS plus EHLO roundtrips that for example disregards; and even though that RFC explicitly refers to DANE for SMTP, giving no reason for why SMTP "requires a different approach", except for the TLS certificate discovery that could and should be hoped for as a general solution, including for clients for any sort of network protocol, DANE falls in this category, too: what this specification tries to achieve is therefore not counteracting , but accompanying it; the author wants to state again that according mechanisms already exist in active use, and that many SMTP client/server software implementations already support SRV RR lookups: activating this existing code base by standard means is only a consequence. The other extension that mentions, MTA-STS, requires SMTP operators to install and maintain an additional HTTP protocol server, accessible on the usual HTTP/S ports and the same domain as the email service. Dependent on the software this may require non-trivial and error prone path access policy server rules. It is an approach that is complicated to implement, and that brings the complexity of another network protocol in the SMTP path; given that several different HTTP protocol exist, which in turn base upon several different transport protocols, the complexity and protocol is alien to SMTP. The SVCB DNS approach would not bring additional value to the problem in question, but on the other hand requires additional implementation efforts as email protocols do not make any use of it as of today. Coming back to DANE for SMTP it must be said that unfortunately DNSSEC is still consciously unsupported by major players as of today. Furthermore provider web interfaces do often not allow users creation of the necessary DANE TLSA resource records: not always can the answer be "run your own name server", also letting aside the administrative effort of doing so. However, discovering an Implicit TLS enabled email service via the _smtp-tls._tcp SRV record, thereafter checking for applicability of DNS-Based Authentication of Named Entities (DANE), possibly even Using Raw Public Keys in Transport Layer Security (TLS) would be a bright future that addresses possibly all of today's issues.
Acknowledgements Thanks to Jan Ingvoldstad for remarks on the missing rationale. Thanks to Jeremy Harris for spending time and revealing the many problems of early draft variants, as well as comments on how to do it better; very special thanks to him for a kickstart implementation of this very draft in the widely used FOSS MTA Exim. Jeremy Harris, Viktor Dukhovni and Wietse Venema commented on the initial odd usage of port 0 for the STARTTLS discovery case. Thanks to Alex Brotman for hinting on the fallback strategy. Many thanks to Jonas Stalder for dozens of corrections and suggestions, and also for finding IETF tooling bugs.