OpenBSD: Fix keepalive_ms and set_keepalive_ms. #99
Conversation
|
I've fixed things up so the |
Will do. Thank you! |
…BSD. OpenBSD (as of release 6.6) does not support the socket option TCP_KEEPIDLE. To be using SO_KEEPALIVE was not correct and was failing at runtime. NetBSD supports the option and it should therefore be treated like generic unix. The use of SO_KEEPALIVE as the KEEPALIVE_OPTION on NetBSD also does not seem to be correct.
|
@pfmooney: Thank you for the CI fixes! I stacked my patch on top of the new CI checks pass, and the fix worked in my testing in combination with the http client in The test program is the hyper::client example. |
| @@ -667,8 +667,9 @@ impl<T: AsRawSocket> AsSock for T { | |||
| cfg_if! { | |||
| if #[cfg(any(target_os = "macos", target_os = "ios"))] { | |||
| use libc::TCP_KEEPALIVE as KEEPALIVE_OPTION; | |||
| } else if #[cfg(any(target_os = "openbsd", target_os = "netbsd"))] { | |||
| use libc::SO_KEEPALIVE as KEEPALIVE_OPTION; | |||
| } else if #[cfg(target_os = "openbsd")] { | |||
There was a problem hiding this comment.
This is dropping the netbsd case. Is it safe to do so?
There was a problem hiding this comment.
Yes. NetBSD understands TCP_KEEPIDLE (tcp(4)), the cfg(unix) case just below this one applies to it correctly.
src/ext.rs
Outdated
| @@ -737,7 +738,25 @@ impl TcpStreamExt for TcpStream { | |||
| Ok(Some((secs as u32) * 1000)) | |||
| } | |||
|
|
|||
| #[cfg(unix)] | |||
| #[cfg(target_os = "openbsd")] | |||
There was a problem hiding this comment.
Style nit: Would you mind moving these two function definitions down to after their respective counterparts with the #[cfg(all(unix, not(target_os = "openbsd")))] gates? I think it would be nice list the specific openbsd case after the more general unix case.
OpenBSD (as of release 6.6) does not support the socket option
TCP_KEEPIDLE(tcp(4)). To be usingSO_KEEPALIVEinstead made it compile but was not correct and was failing at run-time.This pull request fixes #82. That github issue has a good description of the problem.
NetBSD supports the option (tcp(4)) and NetBSD should therefore be treated like generic unix. It therefore also did not seem right to use
SO_KEEPALIVEas theKEEPALIVE_OPTIONon NetBSD.I have tested the patched
net2as of f061e5e in combination with the http client inhyperversion=0.12.35. I have tested on OpenBSD 6.6 amd64 withrustc 1.38.0, on Linux x86_64 withrustc 1.43.1and on NetBSD 9.0 amd64 withrustc 1.42.0.