This how-to is created by the Dutch Internet Standards Platform (the organization behind Internet.nl) in cooperation with industry experts (hosters and vendors) and is meant to provide practical information and guidance on implementing DANE for SMTP.
- DANE is a best-practice technology for securing the transfer of email (SMTP) between organizations across the public Internet.
- Successful DANE deployments require additional operational discipline.
- Automated monitoring of your own email servers and related DNS records is a must.
- Robust automation of coordinated DNS and email server certificate chain updates.
- These topics will be covered in more detail in this how-to.
- Please plan carefully and then deploy DANE for your email servers. Botched deployments not not only harm the domain in question, but also have a deterrent effect on adoption by others.
- For more information on adoption statistics and software support, take a look at: https://github.com/baknu/DANE-for-SMTP/wiki
- What is DANE?
- Why use DANE for SMTP?
- DANE TLSA record example
- Advantages of DANE explained by illustrations
- Reliable certificate rollover
- Tips, tricks and notices for implementation
- Inbound e-mail traffic (publishing DANE DNS records)
- Implementing DANE for SMTP on Postfix (inbound & outbound e-mail traffic)
- Implementing DANE for SMTP on Exim (inbound & outbound e-mail traffic)
- Implementing DANE for SMTP on Halon (inbound & outbound e-mail traffic)
- Implementing DANE for SMTP on Port25 PowerMTA (outbound e-mail traffic)
- Special thanks
Table of contents generated with markdown-toc
DANE is short for "DNS-based Authentication of Named Entities" and is described in RFC 6698 with "updates and operational guidance" in RFC 7671. DANE establishes a downgrade-resistant method to verify an SMTP servers identity before it starts to transport an email message over a STARTTLS encrypted layer. In order to achieve this it uses verification information retrieved from the recipients DNSSEC signed DNS zone. DANE does not rely on additional trusted parties outside the delegation chain in DNS.
DANE, as a method, has been designed to work with any TLS service. DANE for SMTP (which is described in RFC 7672) implements the DANE method for SMTP. It is used increasingly and adds active attack (man-in-the-middle) resistance to SMTP transport encryption RFC 7672 Section 1.3. DANE for SMTP uses the presence of DNS TLSA ressource records to securely signal TLS support and to publish the means by which SMTP clients can successfully authenticate legitimate SMTP servers. The result is called "opportunistic DANE TLS", and resists downgrade and man-in-the-middle (MITM) attacks when the destination domain and its MX hosts are DNSSEC signed, and TLSA records are published for each MX host. While possible, DANE for HTTP is not presently supported by the major browsers and so has seen little deployment.
At this time it is not possible to require mandatory transport encryption (TLS) in public mail transport. A mail server might not support transporting messages using encryption. Today only plaintext or opportunistic transport encryption are applicable – opportunistic because it is up to the receiving server to decide if it wants to and is able to send messages using TLS (via STARTTLS).
DANE offers several advantages by binding X.509 certificates to domains using DNSSEC. In an SMTP context this allows a) sending mail servers to securely signal TLS support by the receiving domain's mail server, and b) sending mail servers to verify the autenticity of the offered certificate by the receiving domain's mail server. This helps to address risks that occur when using oppurtunistic TLS connections between mail servers.
The way SMTP was designed including the usage of opportunistic TLS (via STARTTLS) is not without risks:
- A sending server has no means to determine beforehand if the receiving server supports encrypted transport. Only after the communication has begun the sending server may learn from the features the receiving server supports that it allows for encrypted transport.
- SMTP uses opportunistic TLS (via STARTTLS) as a mechanism to enable secure transport between mail servers. However, the fact that STARTTLS is opportunistic means that the initial connection from one mail server to another always starts unencrypted making it vulnerable to man in the middle attacks. If a mail server does not offer the 'STARTTLS capability' during the SMTP handshake (because it was stripped by an attacker), transport of mail occurs over an unencrypted connection.
- Forcing the use of TLS for all mail servers would break backwards compatibility and is therefore not considered a viable option since it could seriously disrupt mail delivery across the internet.
- By default mail servers do not validate the authenticity of another mail server's certificate; any random certificate is accepted (see RFC 3207).
- It was unclear which CAs to trust when validating the certificate for a given destination.
- In MTA-to-MTA SMTP, server hostnames for the destination domain are obtained indirectly via DNS MX lookups, but, without DNSSEC, these names cannot be trusted. As a result, it was unclear which names to verify in the certificate.
As as result, even when STARTTLS is used, a man in the middle attacker:
- can strip the STARTTLS capability send by the receiving mail server because this communication takes place without the use of encryption.
- can insert any certificate of his choice in the in the unencrypted communication and use it to intercept the traffic.
The risks of SMTP with opportunistic TLS can be mitigated by using DANE:
- The presence of a DNS TLSA record for the receiving domain's mail server, should be interpreted by the sending mail server as the capability to use TLS. The use of TLS can therefore be forced when communicating with the receiving domain's mail server.
- The operator of the receiving domain's mail server is obligated to ensure that any published TLSA records at all times match the server's certificate chain, and that STARTTLS is enabled and working. This makes it possible for the sending domain's mail server to validate the certificate offered by the receiving domain's mail server.
In short: DANE allows sending mail servers to unconditionally require STARTTLS with a matching certificate chain. Otherwise, the sending mail server aborts the connection and tries another server or defers the message. Receiving servers with published TLSA records, are therefore no longer vulnerable to the afore mentioned man in the middle attacks.
Internet.nl currently does not include MTA-STS in its tests. This paragraph explains why this is the case.
First you need to understand that, as explained on our website, the selection and development of tests performed by internet.nl is primairily based on:
- the ‘comply-or-explain’ list of the Dutch Standardisation Forum which is mandatory for all Dutch government agencies;
- security advices of the Dutch NCSC;
- relevant RFC’s of IETF.
Final decisions in this area are made by our steering committee.
MTA-STS is a fairly new standard for enforcing authenticated transport encryption between mail servers. The standard was developed primarily by the big cloud mail providers and seems to especially meet their specific needs. We have seen quite some uptake of DANE (e.g. by Comcast, Protonmail and Cisco), but until now not by the major cloud mail providers. MTA-STS is relatively complex because it needs an extra HTTPS interface (including certificate validation) to function, and it is also considered to be less secure than DANE due to trust on first use and caching.
MTA-STS relies on trust on first use (TOFU) and policy caching. So the initial connection is just STARTTLS (without authentication of the receiving mail server and without enforcing encryption). During that first connection the MTA-STS policy will be cached by the sending mail server and subsequent encrypted connections will be enforced and authenticated (until the policy cache time expires). This caching mechanism means the larger and more up up-to-date the cache, the better MTA-STS will work. Basically big (cloud) mail providers (that contact about every mail server worldwide each day) will have a large up-to-date policy cache, but smaller mail providers will not and therefore will run more often into trust on first use (i.e. initial less secure connection). So basically MTA-STS seems to offer better security to big players than to the smaller ones. In fact, the MTA-STS RFC states that DANE offers better downgrade resistance.
In view of the foregoing and considering the facts that the Dutch NCSC advises to use DANE for securing mail transport, and DNSSEC adoption in the Netherlands (and some other countries like .SE, .NO and .CZ) is quite high and keeps increasing, DANE and not MTA-STS is on the 'comply or explain' list and subsequently included on internet.nl.
Note that MTA-STA and DANE can co-exists next to each other. They intentionally do not interfere.
Usage: says something about the type of certificate that is used for this TLSA record.
2: intermediate / root certificate
3: end-entity certificate (also called 'host certificate' or 'server certificate')
Selector: this is about the scope of the fingerprint regarding this TLSA record.
0: fingerprint with regard to the full certificate
1: fingerprint with regard to the public key
Matching-Type: information about the hashing mechanism used for fingeeprint regarding this TLSA record.
0: no hasing, full information
1: SHA2-256 hash
2: SHA2-512 hash
The illustration below shows two TLS capable mail servers without using DANE. This scenario exposes the mail transport to the risks described above.
The illustration below shows what happens when an attacker performs a man in the middle (MITM) attack and forces an unsecure connection by stripping the TLS capability from the receiving e-mail server.
- When an attacker controls the network communication between the sending and receiving server it may downgrade the session by removing the information that indicates the receiving server supports encrypted transport. The sending server would proceed and transport the message unencrypted making all data contained in the message visible to the attacker.
The illustration below shows what happens when an attacker performs a man in the middle (MITM) attack and inserts its own certificate into the connection process.
- Traditional SMTP protocol design does not offer any means for a sending server to automatically verify it communicates with the right server. As a result an attacker may fool the sending server to hand over the message to a server under the attackers control. This is called a man-in-the-middle attack, because the attacker places the bad server between the sending and the good receiving server.
- To accieve this the attacker would configure the bad server to identify as the good receiving server. It would even use a TLS certificate that identifies with the name of the good server. The attacker would then poison the sending servers DNS telling it to go to the bad server whenever it wants to send messages to the good one. The fooled server would hand over all messages to the man-in-the-middle server.
The illustration below shows how the use of DANE can protect against man in the middle (MITM) attacks by addressing the shortcomings of TLS without DANE.
- The operator of the receiving server publishes a TLSA record in its DNSSEC signed zone. The mere existence of the record indicates to the sending server that the recieving server must offer STARTTLS. Otherwise transport must not take place.
- The receiving server's TLSA record contains policy information and a string that identifies the correct certificate. Only the good servers certiciate will match these verification criteria. Unless the attacker is also in possesion of the good servers certificate it cannot disguise as good and the man-in-the-middle server will be detected. It the verification data does not match the servers certificate the sending server will not transport the message.
Although guaranteeing reliable DNS resolving is actually an advantage of DNSSEC, it is still worth mentioning here. Notice that in the example above (TLS with DANE) the lack of DNSSEC would make it possible for an attacker to alter DNS responses (2 and 4). Such an attack can be used to trick the sender into sending e-mail to a rogue e-mail server.
A TLSA record identifies a certificate. If the certificate is exchanged for a new one also the associated TLSA record needs to be updated. Otherwise there will be a mismatch, verification will fail and DANE aware servers will not send messages to the receivers domain.
Usually this creates the intent not to exchange the certificate at all. But this collides with best-practices to replace crytographic material – certificates and keys – from time to time.
To satisfy both requirements establish these maintenance procedures:
- DANE allows to publish multiple TLSA records. As long as one matches the receving servers certificates everything is fine. Publish two sets of TLSA records like this:
- One TLSA record matching the certificate currently in use.
- A second TLSA record matching a fallback or future certificate to be used.
- Wait two rounds of TTL before you start using a new certicate. This allows clients to clear their DNS cache and pick up the new TLSA record.
- Once the new certificate has been deployed, tested and all is well, the legacy TLSA records may be dropped. Notice that removing the legacy TLSA record too soon can cause an outage.
Two ways of handling certificate rollover are known to work well, in combination with automated monitoring to ensure that the TLSA records and certificates are always current and correct.
- Current + next. This roll-over scheme always publishes two TLSA records per server certificate.
- One with the SHA2-256 fingerprint of the mail server's current public key (a "3 1 1" record).
- And a second with the SHA2-256 fingerprint of the mail server's next public key (also a "3 1 1" record).
- Current + issuer. This roll-over scheme always publishes two TLSA records per mail server certificate.
- One with the SHA2-256 fingerprint of the mail server's current public key (3 1 1)
- And a second with the SHA2-256 fingerprint of the public key of an issuing CA that directly or indirectly signed the server certificate (2 1 1). This need not be (and typically is not) a root CA.
- With the "current + next" approach, the second key pair (of which the public key can be used for the next TLSA "3 1 1" fingerprint) can be created / known in advance of obtaining the corresponding certificate for that key. In particular, if keys are rotated often enough (every 30 to 90 days or so), the next key can be pre-generated as soon as the previous key and certificate are deployed. This allows plenty of time to publish the corresponding next "3 1 1" TLSA record to replace the legacy record for the decommissioned key. The process begins again with another "next" key generated right away.
- Deployment of a new certificate and key must be predicated (automated check) on the corresponding TLSA "3 1 1" record being in place for some time, not only on the primary DNS nameserver, but also on all secondary nameservers. Explicit queries against all the servers are to check for this are highly recommended.
- Some servers have keys and certificates for multiple public key algorithms (e.g. both RSA and ECDSA). In that case, not all clients will negotiate the same algorithm and see the same key. This means that a single "3 1 1" record cannot match the server's currently deployed certificate chains. Consequently, for such servers the "3 1 1" current + "3 1 1" next TLSA records need to be provisioned separately for each algorithm. Failure to do that can result in hard to debug connectivity problems with some sending systems and not others.
- Use of the same key (and perhaps wildcard certificate) across all of a domain's SMTP servers (all MX hosts) is not recommended. Such keys and certificates tend to be rotated across all the servers at the same time, and any deployment mistakes then lead to an outage of inbound email. Large sites with proper monitoring and carefully designed and automated rollover processes can make wildcard certificates work, but if in doubt, don't overestimate your team's ability to execute this flawlessly.
- When monitoring your systems, test every IPv4 and IPv6 address of each MX host, not all clients will be able connect to every address, and none should encounter incorrect TLSA records, neglected certificates, or even non-working STARTTLS. Also test each public key algorithm, or stick to just one. All DANE-capable SMTP servers support both RSA and ECDSA P-256, so you can pick just RSA (2048-bit key) or ECDSA (P-256).
- Make sure that your servers support TLS 1.2, and offer STARTTLS to all clients, even those that have not sent you email in the past. Denying STARTTLS to clients with no IP "reputation" would lock them out indefinitely if they support DANE, since they then can never send any initial mail in the clear to establish their reputation.
This section describes several pionts for attention when implementing DANE for SMTP.
- DANE is meant to be used for the MX domain. So if you are using another domain's mail servers, make sure to ask the administrator of that domain (your mail provider) to support DANE by setting up a TLSA record.
- When implementing DANE we advise to first publish DANE records on your MX domains, and then enable DANE verification on your sending mail servers.
- DANE is backwards compatible. So if your mail server supports DANE and a connecting mail server does not support it yet, usually STARTTLS or plain text is used.
- DANE relies on the security that is provided by DNSSEC. Make your primary domain and MX domain support DNSSEC before implementing DANE. It is important that DNSSEC is implemented properly. A lot of DANE breakage stems from receiving/recipient domains with broken DNSSEC implementations.
- Purchasing of expensive certificates for mail servers has no to little added value for the confidentiality since mail servers don't validate certificates by default. Depending on the context there can be other advantages which makes organizations decide to use specific certificates.
- It is possible to use self-signed certificates.
- Section 3.2 of RFC 7672 states that SMTP clients must not perform certificate name checks when using an end-entity certificate (usage type 3). However, it also states that SMTP clients must perform certificate name checks when using an intermediate or root certificate (usage type 2). The latter is necessary to prevent attackers from abusing a random intermediate or root certificate to falsely validate their evil mail servers.
- Section 3.1 of RFC 7672 states that the expiration date of the end-entity certificate MUST be ignored.
- It is recommended to use a certificates public key for generating a TLSA signature (selector type "1") instead of the full certificate (selector type "0"), because this enables the reuse of key materials. Alghough it is wise te refresh your key material once in a while, note that the use of Forward Secrecy decreases the need to use a new key-pair on every occasion.
- An issuer certificate (usage type "2") validates only when the full certificate chain is offered by the receiving mail server.
- Mail servers by default don't validate certificates and therefore don't have their own certificate store. That's why DANE for SMTP only supports usage type "2" (DANE-TA) and usage type "3" (DANE-EE). Usage type "0" (PKIX-TA) and usage type "1" (PKIX-EE) are not supported.
- Make sure the TTL (time-to-live) of your TLSA records is not too high. This makes it possible to apply changes relatively fast in case of problems. A TTL between 30 minutes (1800) and 1 hour (3600) is recommended.
- The refresh value of your full DNS zone should be in accordance with the TTL setting of your TLSA record, to make sure all name