Organisation: | Copyright (C) 2020-2023 Olivier Boudeville |
---|---|
Contact: | about (dash) leec (at) esperide (dot) com |
Creation date: | Wednesday, November 11, 2020 |
Lastly updated: | Sunday, March 19, 2023 |
Version: | 1.1.9 |
Status: | Stable |
Dedication: | Users and maintainers of the LEEC library |
Abstract: | The role of the LEEC library is to interact from Erlang/OTP with servers implementing the ACME protocol - specifically Let's Encrypt servers, mostly in order to generate X.509 certificates. |
The latest version of this documentation is to be found at the official LEEC website (http://leec.esperide.org).
This LEEC documentation is also available in the PDF format (see Ceylan-LEEC-technical-manual-english.pdf), and mirrored here.
Table of Contents
The LEEC library is the Ceylan fork of the original and much appreciated letsencrypt-erlang, which is a Let's Encrypt client library for Erlang, whose author is Guillaume Bour.
LEEC's purpose is to obtain proper X.509 security certificates from Let's Encrypt - or more generally ACME servers (version 2), typically in order to secure one's webservers so that they can offer a solid HTTPS connectivity, which is pretty much standard nowadays.
Each certificate covers a single domain or subdomain (possibly a virtual host, and is obtained by proving its ownership by the requester.
At least currently no wildcard certificate is supported (more complex challenges would have to be implemented).
These certificates last for 90 days, so they shall be renewed regularly.
LEEC is notably used in the context of US-Web.
Compared to the original letsencrypt-erlang library, the main differences introduced by LEEC are:
So, even if LEEC can be seen mostly as a "reckless" fork (in the sense that it became quickly obvious that retaining upstream compatibility could hardly be achieved) - with so many source-level differences (in terms of conventions, Myriad integration, whitespace cleanup) that a pull request can difficultly be considered - yet, in spite of the appearances, it remained quite close to the original (mainly differences of form) and followed the same structure.
By some ways, this LEEC fork is safer and more robust than the original, by others not (e.g. test coverage, autonomous use, continuous integration). A key goal was to make it easier to understand and maintain.
Most of the elements of this pull request from Marc Worrell have also been integrated.
The general dependencies are:
The LEEC-specific ones, which are automatically managed by rebar3 if opting for a rebar-based build, are:
If wanting to switch from the default jsx to jiffy, following files shall be updated:
(none in Myriad)
A potential dependency problem is that many Erlang-based webservers are powered by Cowboy (thus Cowlib) whereas LEEC used to rely necessarily on Shotgun, thus on Gun (and thus Cowlib) as well. Most of the time this implied different (potentially incompatible) versions of Cowlib, whereas only up to one should exist in the code path at any time.
We prefer sticking to the Cowlib version that is induced by Cowboy. At the time of this writing, the latest Cowboy stable version (the one that webserver projects such as US-Web want) is 2.8.0 and relies on Cowlib 2.9.1, whereas the latest Shotgun stable version, 0.5.0, is lagging behind, relying on Gun 1.3.1, itself relying on Cowlib 2.6.0 (too old).
An attempt of solution was to remove the dependency of LEEC onto Shotgun (as it induced a dependency on an older Cowlib) but to use Gun instead, which is lower-level yet might be chosen in order to rely on the target Cowlib version. However we did not found a suitable Gun version for that (1.3 being too old, 2.0.* not ready).
So a last-resort solution has been to rely instead on the even lower-level Erlang-native httpc client module (involving inets and ssl). The result, although based only on HTTP/1.1 with no connection-reuse, proved satisfactory right from the start and thus is provided as an alternate way of using LEEC, without involving any extra dependency.
This allows embedding LEEC with only one dependency onto Myriad and one onto a JSON parser (either jsx or jiffy) - and no other one (top-level or induced).
Two build procedures can be used (from the root of LEEC), and are now mostly the same:.
$ rebar3 upgrade --all $ rebar3 compile
This last procedure is the one that we prefer and use routinely (see the US-Web native deployment script as an example thereof).
The main example of LEEC in action can be found in link with US-Web, whose sources can be found here; see notably class_USCertificateManager.erl and us_web_letsencrypt_handler.erl [1].
[1] | By the way, if you are currently reading LEEC's official website, then your browser already checked LEEC-obtained certificates! |
This mode of operation, described in this section, is typical of the use case where an Erlang-based webserver (in this case based on Cowboy) has to renew certificates corresponding to the various virtual hosts (possibly dispatched under various domains) that it is making available.
A first part is to create as many LEEC FSMs as domains of interest, which will connect to the target ACME servers (most probably Let's Encrypt ones). Each FSM is a LEEC agent that will generate its own (strong) RSA key, create automatically its throwaway ACME account on the server, secure properly the communication (with TLS signatures, nonces, etc.), and wait for further user request regarding its domain of interest (e.g. foobar.org).
Such a request is bound to ask the ACME server to generate (as a Certificate Authority) a X.509 certificate covering, thanks to SAN, a set of subdomains (FQDN) to secure (e.g. hello.foobar.org, hurricane.foobar.org) - knowing that no wildcard certificate can be obtained with the http-01 challenge being used here. The ACME server will send challenges to LEEC so that it can prove that it controls indeed all these subdomains.
A second part of the LEEC action is to ensure that these answers are available indeed, as tokens. In practice the ACME server will attempt to read them at specific URLs (prefixed with .well-known/acme-challenge/) expected to be served for these subdomains (most probably thanks to virtual hosting). If the ACME server is able to query and read, directly from a domain, the right tokens corresponding to the challenges it sent for this domain, then the proof of actual control by the requester is established, and the ACME server can thus issue a corresponding certificate and transmit it appropriately to LEEC.
The overall webserver of the user shall thus track the transitions of these FSMs until (hopefully) they successfully complete their procedure and obtain from their ACME server the corresponding certificate. Then only the user webserver will be able to fire its https support with these brand new certificates [2].
[2] | Before, even if suitable certificates were pre-existing, at least the ACME URL prefix was to remain over http instead of being automatically promoted to https as all others. |
Finally, a task scheduler may be used to trigger renewals on time (not too soon, not too late, as ACME rules apply and, of course, each FQDN shall be covered by a valid certificate at any time), and a task ring may be used to (paradoxically) ensure that the webserver as a whole does not interact too much in parallel (through its various LEEC FSMs) with the ACME server (despite hosting potentially a large number of FQDNs), knowing that severe rate limits (example in production) apply.
LEEC does its best to go through this procedure, validating as much as possible each of these steps for a better reliability/control, and reporting outcome for tracability and error management.
In practice, the user code is expected:
For US-Web, (1), (2) and (4) are managed by class_USCertificateManager.erl (see respectively init_leec/5, request_certificate/1 and the onCertificateRequestOutcome/2 callback). (3) is taken in charge by us_web_letsencrypt_handler.erl (see init/2).
At least the ACME servers from Let's Encrypt enforce various rate limits (both in production environment and in staging one) that are fairly low, which leads to preferring requesting certificates only on a per-domain basis (e.g. globally for foobar.org) rather than on a per-hostname host basis (e.g. one for baz.foobar.org, another one for hurrican.foobar.org, etc., these hosts being virtual ones or not), as such requests would quickly become too numerous to respect these rate thresholds.
A per-domain certificate should then include directly its various hostnames as Subject Alternative Names (SAN entries).
With the http-01 challenge type, no wildcard for such SAN hosts (e.g. *.foobar.org) can be specified, so all the wanted ones have to be explicitly listed [3].
[3] | As a result, the certificate may disclose virtual hosts that would be otherwise invisible from the Internet (as not even declared in the DNS entries for that domain that would act as wildcard name resolvers). |
So for example, with LEEC, the certificate for foobar.org (that would be managed by a dedicated LEEC agent) should list following SAN entries: baz.foobar.org, hurrican.foobar.org, etc.
LEEC implemented independent (gen_statem) FSMs to allow typically for concurrent certificate renewals to be triggered (thanks to autonomous LEEC agents, per-FSM connection pools, separate keys, etc.).
A drawback of the aforementioned Let's Encrypt rate limits is that, while a given FSM is to remain below said thresholds, a set of parallel ones may not.
Should this issue arise, an option is to use a single FSM and to trigger certificate requests in turn. Another one is to rely on a task ring in order to avoid by design that such FSMs overlap.
Currently LEEC creates automatically throwaway ACME accounts, which is convenient yet may prevent the use if CAA (Certificate Authority Authorization).
If using LEEC to generate a certificate for a baz.foobar.org host, the following three files shall be obtained from the Let's Encrypt ACME server:
To get information about this certificate:
$ openssl x509 -text -noout -in baz.foobar.org.crt Certificate: Data: Version: 3 (0x2) Serial Number: 04:34:17:fd:ee:9b:bd:6b:c2:02:b1:c0:84:62:ed:a6:88:5c Signature Algorithm: sha256WithRSAEncryption Issuer: C = US, O = Let's Encrypt, CN = R3 Validity Not Before: Dec 27 08:21:38 2020 GMT Not After : Mar 27 08:21:38 2021 GMT Subject: CN = baz.foobar.org Subject Public Key Info: Public Key Algorithm: rsaEncryption RSA Public-Key: (2048 bit) Modulus: [...] Exponent: 65537 (0x10001) X509v3 extensions: X509v3 Key Usage: critical Digital Signature, Key Encipherment X509v3 Extended Key Usage: TLS Web Server Authentication, TLS Web Client Authentication X509v3 Basic Constraints: critical CA:FALSE X509v3 Subject Key Identifier: [...] X509v3 Authority Key Identifier: keyid:C0:CC:03:46:B9:58:20:CC:5C:72:70:F3:E1:2E:CB:20:B6:F5:68:3A Authority Information Access: OCSP - URI:http://ocsp.stg-int-x1.letsencrypt.org CA Issuers - URI:http://cert.stg-int-x1.letsencrypt.org/ X509v3 Subject Alternative Name: DNS:hello.baz.foobar.org.crt, DNS:world.foobar.org.crt, DNS:somesite.foobar.org.crt X509v3 Certificate Policies: Policy: 2.23.140.1.2.1 Policy: 1.3.6.1.4.1.44947.1.1.1 CPS: http://cps.letsencrypt.org CT Precertificate SCTs: Signed Certificate Timestamp: Version : v1 (0x0) Log ID : [...] Timestamp : Jan 2 09:23:20.310 2021 GMT Extensions: none Signature : ecdsa-with-SHA256 Signed Certificate Timestamp: Version : v1 (0x0) Log ID : [...] Timestamp : Jan 2 09:23:20.320 2021 GMT Extensions: none Signature : ecdsa-with-SHA256 [...] Signature Algorithm: sha256WithRSAEncryption [...]
A *.key (e.g. my-foobar-leec-agent-private.key) file is a (PEM, strong enough) RSA private key generated by LEEC so that its agent can safely authenticate to the ACME servers it is interacting with.
lets-encrypt-r3-cross-signed.pem is the (PEM) certificate associated to the Certificate Authority (Let's Encrypt here). It is automatically downloaded by LEEC if not already available.
The dh-params.pem file contains the parameters generated by LEEC in order to allow for safer Ephemeral Diffie-Helman key exchanges that is used to provide Forward Secrecy with TLS (see this article for further information). Its generation may take quite some time.
Ceylan-LEEC is distributed under the APACHE 2.0 licence, like the original work that it derives from.
Bugs, questions, remarks, patches, requests for enhancements, etc. are to be sent through the project interface (typically issues), or directly at the email address mentioned at the beginning of this document.
If you have information more detailed or more recent than those presented in this document, if you noticed errors, neglects or points insufficiently discussed, drop us a line! (for that, follow the Support guidelines).
One may also be interested in our mini-HOWTO regarding cybersecurity.