Setting callbacks multiple times on a SslContextBuilder causes the previous callback
installed to leak, using replace_ex_data internally prevents that.
We also start using it in tokio-boring in with_ex_data_future, my understanding
is that the futures currently in use are never installed twice by that function
but that could change in the future with the addition of more async callbacks.
We introduce new methods replace_ex_data for both SslContextBuilder
and Ssl in case anyone is relying on the leaking behaviour of their
set_ex_data methods, but we do document that they leak now.
To handle lifetimes better and allow returning a &mut SslRef from
the client hello struct passed to the closure from
SslContextBuilder::set_select_certificate_callback, we make
the ClientHello struct itself own a reference to the FFI
client hello struct.
These two new kinds of methods immediately return a MidHandshakeSslStream
instead of actually initiating a handshake. This greatly simplifies
loops around MidHandshakeSslStream::WouldBlock.
Overwrite boringSSL's default key exchange preferences with safe
defaults using feature flags:
* "kx-pq-supported" enables support for PQ key exchange algorithms.
Classical key exchange is still preferred, but will be upgraded to PQ
if requested.
* "kx-pq-preferred" enables preference for PQ key exchange,
with fallback to classical key exchange if requested.
* "kx-nist-required" disables non-NIST key exchange.
Each feature implies "kx-safe-default". When this feature is enabled,
don't compile bindings for `SSL_CTX_set1_curves()` and `SslCurve`. This
is to prevent the feature flags from silently overriding curve
preferences chosen by the user.
Ideally we'd allow both: that is, use "kx-*" to set defaults, but still
allow the user to manually override them. However, this doesn't work
because by the time the `SSL_CTX` is constructed, we don't yet know
whether we're the client or server. (The "kx-*" features set different
preferences for each.) If "kx-sfe-default" is set, then the curve
preferences are set just before initiating a TLS handshake
(`SslStreamBuilder::connect()`) or waiting for a TLS handshake
(`SslStreamBuilder::accept()`).
According to [the docs](https://doc.rust-lang.org/stable/std/mem/fn.uninitialized.html),
> Calling this when the content is not yet fully initialized causes immediate undefined behavior.
> it [is] undefined behavior to have uninitialized data in a variable even if that variable has an integer type.
Using MaybeUninit instead, as recommended by the official documentation, avoids undefined behavior by not creating a `&mut` reference to uninitialized data.
* Add rerun-if-env-changed instructions for BORING_* variables
* Use X509_get0_notBefore() and X509_get0_notAfter() instead of X509_getm_notBefore() and X509_getm_notAfter().
According to
https://www.openssl.org/docs/man1.1.0/man3/X509_getm_notBefore.html,
"X509_getm_notBefore() and X509_getm_notAfter() are similar to
X509_get0_notBefore() and X509_get0_notAfter() except they return
non-constant mutable references to the associated date field of the
certificate".
* Only update boringssl submodule if BORING_BSSL_PATH not provided
* Allow BORING_BSSL_LIB_PATH to control link search
* Add fips feature
* Use X509_set_notAfter unconditionally for FIPS compatibility
This is equivalent according to
https://boringssl.googlesource.com/boringssl/+/c947efabcbc38dcf93e8ad0e6a76206cf0ec8072
The version of boringssl that's FIPS-certified doesn't have `X509_set1_notAfter`.
The only difference between that and `X509_set_notAfter` is whether they're const-correct,
which doesn't seem worth having two different code-paths.
* Check out fips commit automatically
* Verify the version of the compiler used for building boringssl
NIST specifies that it needs to be 7.0.1; I originally tried building with clang 10 and it failed.
Theoretically this should check the versions of Go and Ninja too, but they haven't given me trouble in practice.
Example error:
```
Compiling boring-sys v1.1.1 (/home/jnelson/work/boring/boring-sys)
error: failed to run custom build command for `boring-sys v1.1.1 (/home/jnelson/work/boring/boring-sys)`
Caused by:
process didn't exit successfully: `/home/jnelson/work/boring/target/debug/build/boring-sys-31b8ce53031cfd83/build-script-build` (exit status: 101)
--- stdout
cargo:rerun-if-env-changed=BORING_BSSL_PATH
--- stderr
warning: missing clang-7, trying other compilers: Permission denied (os error 13)
warning: FIPS requires clang version 7.0.1, skipping incompatible version "clang version 10.0.0-4ubuntu1 "
thread 'main' panicked at 'unsupported clang version "cc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0": FIPS requires clang 7.0.1', boring-sys/build.rs:216:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
```
* Add Github actions workflow testing FIPS
Co-authored-by: Joshua Nelson <jnelson@cloudflare.com>
In particular, this updates `foreign-types`, which had a lot of breaking changes.
- `ForeignType` is now an unsafe trait
- `*Ref` types no longer need a separate macro call, they're generated automatically
- Generated types now store `NonNull<T>` instead of `*mut T`
This ensures that all the Rust functions, types and constants
always match the actual BoringSSL definitions.
It also removes quite a lot of manually maintained code, as well
as the need for systest.
The value for `SslOptions::ALL`, for example, was wrong. On current
BoringSSL versions, this is a no-op, and is set to `0`.
Clearing it does nothing. So, the `clear_ctx_options` test, that
passed by accident, was adjusted to use a different option.
The `libc` crate is not required, as we only use it for types that
are already defined in the standard library. It was removed from
`boring-sys`. The same can be done to other crates later.
Anthony Ramine