refactor(s2n-quic-platform): move socket config to separate module #1737
Conversation
camshaft
force-pushed
the
camshaft/platform-syscalls
branch
2 times, most recently
from
cba69de
to
f1beb3d
Compare
camshaft
force-pushed
the
camshaft/platform-syscalls
branch
from
f1beb3d
to
6022be0
Compare
| success |= libc!(setsockopt( | ||
| rx_socket.as_raw_fd(), | ||
| libc::IPPROTO_IP, | ||
| libc::IP_PKTINFO, | ||
| &enabled as *const _ as _, | ||
| core::mem::size_of_val(&enabled) as _, | ||
| )) | ||
| .is_ok(); | ||
|
|
||
| success |= libc!(setsockopt( | ||
| rx_socket.as_raw_fd(), | ||
| libc::IPPROTO_IPV6, | ||
| libc::IPV6_RECVPKTINFO, | ||
| &enabled as *const _ as _, | ||
| core::mem::size_of_val(&enabled) as _, | ||
| )) | ||
| .is_ok(); |
There was a problem hiding this comment.
this previously had the check for if rx_addr.is_ipv4(). Why did that get removed?
There was a problem hiding this comment.
It's not really needed to guard against. If the call fails because it's only a single IP-version stack, then that single call with fail. In this case, we consider the configuration successful if either/both of these calls succeed.
| use std::os::unix::io::AsRawFd; | ||
| let enabled: libc::c_int = 1; | ||
|
|
||
| success |= libc!(setsockopt( |
There was a problem hiding this comment.
this previously had the check for if rx_addr.is_ipv4() || !cfg!(any(target_os = "macos", target_os = "ios")). Why is that removed?
There was a problem hiding this comment.
we're no longer bailing on errors; just returning if they succeeded or not. this new approach is more robust since we try our best to enable everything we can and fall back if it wasn't successful.
Description of changes:
While implementing the AF_XDP IO provider, I need a way to create a configured UDP socket outside of the tokio IO provider to reserve the port for the endpoint.
This change moves all of the syscalls out of the tokio IO provider into a separate module.
Call-outs:
While I was here, I realized that we were allowing MTU probing to occur even on platforms where we can't guarantee the DF bit is set. I've gone ahead and added logic to set the max MTU to the minimum value in these cases:
s2n-quic/quic/s2n-quic-platform/src/io/tokio.rs
Lines 139 to 142 in f1beb3d
Testing:
Since this is just extracting function calls into a separate module, the existing tests should cover the differences. I've opened an issue to implement a test that the MTU sizes are correct on non-linux platforms: #1739.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.