1 | /*! # Customising private key usage |
2 | |
3 | By default rustls supports PKCS#8-format[^1] RSA or ECDSA keys, plus PKCS#1-format RSA keys. |
4 | |
5 | However, if your private key resides in a HSM, or in another process, or perhaps |
6 | another machine, rustls has some extension points to support this: |
7 | |
8 | The main trait you must implement is [`sign::SigningKey`][signing_key]. The primary method here |
9 | is [`choose_scheme`][choose_scheme] where you are given a set of [`SignatureScheme`s][sig_scheme] the client says |
10 | it supports: you must choose one (or return `None` -- this aborts the handshake). Having |
11 | done that, you return an implementation of the [`sign::Signer`][signer] trait. |
12 | The [`sign()`][sign_method] performs the signature and returns it. |
13 | |
14 | (Unfortunately this is currently designed for keys with low latency access, like in a |
15 | PKCS#11 provider, Microsoft CryptoAPI, etc. so is blocking rather than asynchronous. |
16 | It's a TODO to make these and other extension points async.) |
17 | |
18 | Once you have these two pieces, configuring a server to use them involves, briefly: |
19 | |
20 | - packaging your `sign::SigningKey` with the matching certificate chain into a [`sign::CertifiedKey`][certified_key] |
21 | - making a [`ResolvesServerCertUsingSni`][cert_using_sni] and feeding in your `sign::CertifiedKey` for all SNI hostnames you want to use it for, |
22 | - setting that as your `ServerConfig`'s [`cert_resolver`][cert_resolver] |
23 | |
24 | [signing_key]: ../../sign/trait.SigningKey.html |
25 | [choose_scheme]: ../../sign/trait.SigningKey.html#tymethod.choose_scheme |
26 | [sig_scheme]: ../../enum.SignatureScheme.html |
27 | [signer]: ../../sign/trait.Signer.html |
28 | [sign_method]: ../../sign/trait.Signer.html#tymethod.sign |
29 | [certified_key]: ../../sign/struct.CertifiedKey.html |
30 | [cert_using_sni]: ../../struct.ResolvesServerCertUsingSni.html |
31 | [cert_resolver]: ../../struct.ServerConfig.html#structfield.cert_resolver |
32 | |
33 | [^1]: For PKCS#8 it does not support password encryption -- there's not a meaningful threat |
34 | model addressed by this, and the encryption supported is typically extremely poor. |
35 | |
36 | # Unexpected EOF |
37 | |
38 | TLS has a `close_notify` mechanism to prevent truncation attacks[^2]. |
39 | According to the TLS RFCs, each party is required to send a `close_notify` message before |
40 | closing the write side of the connection. However, some implementations don't send it. |
41 | So long as the application layer protocol (for instance HTTP/2) has message length framing |
42 | and can reject truncated messages, this is not a security problem. |
43 | |
44 | Rustls treats an EOF without `close_notify` as an error of type `std::io::Error` with |
45 | `ErrorKind::UnexpectedEof`. In some situations it's appropriate for the application to handle |
46 | this error the same way it would handle a normal EOF (a read returning `Ok(0)`). In particular |
47 | if `UnexpectedEof` occurs on an idle connection it is appropriate to treat it the same way as a |
48 | clean shutdown. And if an application always uses messages with length framing (in other words, |
49 | messages are never delimited by the close of the TCP connection), it can unconditionally |
50 | ignore `UnexpectedEof` errors from rustls. |
51 | |
52 | [^2]: <https://datatracker.ietf.org/doc/html/rfc8446#section-6.1> |
53 | */ |
54 | |