RFC9000 says: endpoints MUST treat any packet ending in a valid
stateless reset token as a Stateless Reset
Previously, we did not detect stateless resets that appeared to be
unprotected packets (e.g. Retry or Version Negotiation) or which were
successfully decrypted (astronomically unlikely with TLS, but possible
with custom cryptographic layers).
(cherry picked from commit 7f26029284)
Linux 5.15 was observed to emit an EINVAL when attempting to transmit
UDP with GSO over IPv6 on a virtio net interface. 6.1 was also
observed to emit EINVAL, though only after EIO. In both cases, EINVAL
was suppressed by disabling GSO.
(cherry picked from commit a0dd5ff2fa8c5ba6f64c20d6f8a5f21424b8c62b)
Basically all of the `actions-rs/*` actions are unmaintained. See
<https://github.com/actions-rs/toolchain/issues/216> for more
information. Due to their age they generate several warnings in
CI runs.
To get rid of those warnings the occurrences of
`actions-rs/toolchain` are replaced by `dtolnay/rust-toolchain`,
and the occurrences of `actions-rs/cargo` are replaced by direct
invocations of `cargo`.
This commit adds a new function that refuses new connections without
impacting existing connections. Internally, this just sets the
connection limit to 0, which causes incoming connections to be rejected.
This is the same approach that was taken in 0.8.5 when `Incoming` was
dropped.
This was previously used by the `quinn` layer to stop accepting
connections when the `Incoming` stream was dropped. That stream has
since been replaced by an async function. Users who still want this
can use set_server_config directly.
This commit updates the existing Quinn sub-crate dependencies to remove
the explicit `webpki` v0.22 dependency. It's not used by directly
anywhere and can be a transitive dependency through rustls as required.
Nazar Mokrynskyi