The vmm-sys-util main branch causes a "No buffer space available" error in the vhost test.

[uran0sH]

Use the main branch of vmm-sys-util, and test on vhost, run: cargo test --workspace --no-default-features --features vhost-user. It will cause the problem below.

thread 'vhost_user::connection::tests::send_fd' panicked at vhost/src/vhost_user/connection.rs:763:50:
called `Result::unwrap()` on an `Err` value: SocketRetry(Os { code: 105, kind: Uncategorized, message: "No buffer space available" })
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

And I revert to 02321df. It can pass the test.

[uran0sH]

@roypat @stefano-garzarella @germag

[stefano-garzarella]

Use the main branch of vmm-sys-util, and test on vhost, run: cargo test --workspace --no-default-features --features vhost-user. It will cause the problem below.

thread 'vhost_user::connection::tests::send_fd' panicked at vhost/src/vhost_user/connection.rs:763:50:
called `Result::unwrap()` on an `Err` value: SocketRetry(Os { code: 105, kind: Uncategorized, message: "No buffer space available" })
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

@uran0sH on Linux or macOs or both?

And I revert to 02321df. It can pass the test.

mmm, that commit is just a changelog update, maybe you pasted the wrong one.

[roypat]

Copy my comment from rust-vmm/vhost#308 (comment), I think this is an issue with the test that only got exposed because we started handling MSG_CTRUNC in the case of in_fds == []:

I think this is expected. Looking at the test, it's a negative test:

   // Following communication pattern should not work:
   // Sending side: data(header, body) with fds
   // Receiving side: data(header), data(body) with fds

and indeed, doing recv_with_fds with 0 fds when actually there are fds, causes linux to set MSG_CTRUNC and thus after rust-vmm/vhost#252 that is converted into an ENOBUFS (although on MacOS it'll work).

[stefano-garzarella]

@roypat oh I see, @uran0sH can you try to update the test in vhost?

[uran0sH]

I update the test in this commit: ff22032e1a371b92cf11c7ec8526da5709e7ed5b.
Could you please help me check if these changes are correct? For Linux, we check whether it is an error. For mac, the check for received data has been modified. @roypat @stefano-garzarella