Table of Contents
- Particular sites’ opinions or configurations
- Site scanning or rating tools
- Java compatibility issue
Within certain limits, TLS server software can choose what kind of cryptography to use when a client connects. These choices can affect security, compatibility, and performance in complex ways. Most of these options are independent of a particular certificate. Certbot tries to provide defaults that we think are most useful to our users.
As described below, Certbot will default to modifying server software’s cryptographic settings to keep these up-to-date with what we think are appropriate defaults when new versions of the Certbot are installed (for example, by an operating system package manager).
When this feature is implemented, this document will be updated to describe how to disable these automatic changes.
Software that uses cryptography must inevitably make choices about what kind of cryptography to use and how. These choices entail assumptions about how well particular cryptographic mechanisms resist attack, and what trade-offs are available and appropriate. The choices are constrained by compatibility issues (in order to interoperate with other software, an implementation must agree to use cryptographic mechanisms that the other side also supports) and protocol issues (cryptographic mechanisms must be specified in protocols and there must be a way to agree to use them in a particular context).
The best choices for a particular application may change over time in response to new research, new standardization events, changes in computer hardware, and changes in the prevalence of legacy software. Much important research on cryptanalysis and cryptographic vulnerabilities is unpublished because many researchers have been working in the interest of improving some entities’ communications security while weakening, or failing to improve, others’ security. But important information that improves our understanding of the state of the art is published regularly.
When enabling TLS support in a compatible web server (which is a separate step from obtaining a certificate), Certbot has the ability to update that web server’s TLS configuration. Again, this is different from the cryptographic particulars of the certificate itself; the certificate as of the initial release will be RSA-signed using one of Let’s Encrypt’s 2048-bit RSA keys, and will describe the subscriber’s RSA public key (“subject public key”) of at least 2048 bits, which is used for key establishment.
Note that the subscriber’s RSA public key can be used in a wide variety of key establishment methods, most of which do not use RSA directly for key exchange, but only for authenticating the server! For example, in DHE and ECDHE key exchanges, the subject public key is just used to sign other parameters for authentication. You do not have to “use RSA” for other purposes just because you’re using an RSA key for authentication.
The certificate doesn’t specify other cryptographic or ciphersuite particulars; for example, it doesn’t say whether or not parties should use a particular symmetric algorithm like 3DES, or what cipher modes they should use. All of these details are negotiated between client and server independent of the content of the ciphersuite. The Let’s Encrypt project hopes to provide useful defaults that reflect good security choices with respect to the publicly-known state of the art. However, the Let’s Encrypt certificate authority does not dictate end-users’ security policy, and any site is welcome to change its preferences in accordance with its own policy or its administrators’ preferences, and use different cryptographic mechanisms or parameters, or a different priority order, than the defaults provided by Certbot.
If you don’t use Certbot to configure your server directly, because the client doesn’t integrate with your server software or because you chose not to use this integration, then the cryptographic defaults haven’t been modified, and the cryptography chosen by the server will still be whatever the default for your software was. For example, if you obtain a certificate using standalone mode and then manually install it in an IMAP or LDAP server, your cryptographic settings will not be modified by the client in any way.
Initially, Certbot will configure users’ servers to use the cryptographic defaults recommended by the Mozilla project. These settings are well-reasoned recommendations that carefully consider client software compatibility. They are described at
and the version implemented by Certbot will be the version that was most current as of the release date of each client version. Mozilla offers three separate sets of cryptographic options, which trade off security and compatibility differently. These are referred to as the “Modern”, “Intermediate”, and “Old” configurations (in order from most secure to least secure, and least-backwards compatible to most-backwards compatible). The client will follow the Mozilla defaults for the Intermediate configuration by default, at least with regards to ciphersuites and TLS versions. Mozilla’s web site describes which client software will be compatible with each configuration. You can also use the Qualys SSL Labs site to test your server and see whether it will be compatible with particular software versions.
The Let’s Encrypt project expects to follow the Mozilla recommendations
in the future as those recommendations are updated. (For example, some
users have proposed prioritizing a new ciphersuite known as
which uses the ChaCha and Poly1305 algorithms, and which is already
implemented by the Chrome browser. Mozilla has delayed recommending
0xcc13 over compatibility and standardization concerns, but is likely
to recommend it in the future once these concerns have been addressed. At
that point, Certbot would likely follow the Mozilla recommendations and favor
the use of this ciphersuite as well.)
The Let’s Encrypt project may deviate from the Mozilla recommendations in the future if good cause is shown and we believe our users’ priorities would be well-served by doing so. In general, please address relevant proposals for changing priorities to the Mozilla security team first, before asking the Certbot developers to change Certbot’s priorities. The Mozilla security team is likely to have more resources and expertise to bring to bear on evaluating reasons why its recommendations should be updated.
The Let’s Encrypt project will entertain proposals to create a very small number of alternative configurations (apart from Modern, Intermediate, and Old) that there’s reason to believe would be widely used by sysadmins; this would usually be a preferable course to modifying an existing configuration. For example, if many sysadmins want their servers configured to track a different expert recommendation, Certbot could add an option to do so.
In the course of considering how to handle this issue, we received recommendations with sources of expert guidance on ciphersuites and other cryptographic parameters. We’re grateful to everyone who contributed suggestions. The recommendations we received are available under Feedback.
Certbot users are welcome to review these authorities to better inform their own cryptographic parameter choices. We also welcome suggestions of other resources to add to this list. Please keep in mind that different recommendations may reflect different priorities or evaluations of trade-offs, especially related to compatibility!
We receive lots of feedback on the type of ciphersuites that Let’s Encrypt supports and list some collated feedback below. This section aims to track suggestions and references that people have offered or identified to improve the ciphersuites that Let’s Encrypt enables when configuring TLS on servers.
Because of the Chatham House Rule applicable to some of the discussions, people are not individually credited for their suggestions, but most suggestions here were made or found by other people, and I thank them for their contributions.
Some people provided rationale information mostly having to do with compatibility of particular user-agents (especially UAs that don’t support ECC, or that don’t support DH groups > 1024 bits). Some ciphersuite configurations have been chosen to try to increase compatibility with older UAs while allowing newer UAs to negotiate stronger crypto. For example, some configurations forego forward secrecy entirely for connections from old UAs, like by offering ECDHE and RSA key exchange, but no DHE at all. (There are UAs that can fail the negotiation completely if a DHE ciphersuite with prime > 1024 bits is offered.)
IETF has published a BCP document, RFC 7525, “Recommendations for Secure Use of Transport Layer Security (TLS) and Datagram Transport Layer Security (DTLS)”
BetterCrypto.org, a collaboration of mostly European IT security experts, has published a draft paper, “Applied Crypto Hardening”
Gillmor’s Internet-Draft “Negotiated Discrete Log Diffie-Hellman Ephemeral Parameters for TLS” is being developed at the IETF TLS WG. It advocates using standardized DH groups in all cases, not individually-chosen ones (mostly because of the Triple Handshake attack which can involve maliciously choosing invalid DH groups). The draft provides a list of recommended groups, with primes beginning at 2048 bits and going up from there. It also has a new protocol mechanism for agreeing to use these groups, with the possibility of backwards compatibility (and use of weaker DH groups) for older clients and servers that don’t know about this mechanism.
Mozilla’s general server configuration guidance is available at https://wiki.mozilla.org/Security/Server_Side_TLS
Mozilla has also produced a configuration generator: https://ssl-config.mozilla.org
The Dutch National Cyber Security Centre has published guidance on “ICT-beveiligingsrichtlijnen voor Transport Layer Security (TLS)” (“IT Security Guidelines for Transport Layer Security (TLS)”). These are available only in Dutch at
I have access to an English-language summary of the recommendations.
Damien Giry collects recommendations by academic researchers and standards organizations about keylengths for particular cryptoperiods, years, or security levels. The keylength recommendations of the various sources are summarized in a chart. This site has been updated over time and includes expert guidance from eight sources published between 2000 and 2017.
NISA published its “NIST Special Publication 800-52 Revision 1: Guidelines for the Selection, Configuration, and Use of Transport Layer Security (TLS) Implementations”
and its “NIST Special Publication 800-57: Recommendation for Key Management – Part 1: General (Revision 3)”
ENISA published its “Algorithms, Key Sizes and Parameters Report - 2013”
The WeakDH/Logjam research has thrown into question the safety of some existing practice using DH ciphersuites, especially the use of standardized groups with a prime ≤ 1024 bits. The authors provided detailed guidance, including ciphersuite lists, at
These lists may have been derived from Mozilla’s recommendations. One of the authors clarified his view of the priorities for various changes as a result of the research at
In particular, he supports ECDHE and also supports the use of the standardized groups in the FF-DHE Internet-Draft mentioned above (which isn’t clear from the group’s original recommendations).
Amazon ELB explains its current ciphersuite choices at
The 18F site (https://18f.gsa.gov/) is using
ssl_ciphers 'kEECDH+ECDSA+AES128 kEECDH+ECDSA+AES256 kEECDH+AES128 kEECDH+AES256 kEDH+AES128 kEDH+AES256 DES-CBC3-SHA +SHA !aNULL !eNULL !LOW !MD5 !EXP !DSS !PSK !SRP !kECDH !CAMELLIA !RC4 !SEED';
Qualys offers the best-known TLS security scanner, maintained by Ivan Ristić.
A lot of backward-compatibility concerns have to do with Java hard-coding DHE primes to a 1024-bit limit, accepting DHE ciphersuites in negotiation, and then aborting the connection entirely if a prime > 1024 bits is presented. The simple summary is that servers offering a Java-compatible DHE ciphersuite in preference to other Java-compatible ciphersuites, and then presenting a DH group with a prime > 1024 bits, will be completely incompatible with clients running some versions of Java. (This may also be the case with very old MSIE versions…?) There are various strategies for dealing with this, and maybe we can document the options here.