Add support for sendmmsg(2) on linux #1171
Conversation
colinmarc
force-pushed
the
sendmmsg
branch
3 times, most recently
from
10452dc to
779b2b7
Compare
This comment was marked as resolved.
This comment was marked as resolved.
colinmarc
force-pushed
the
sendmmsg
branch
2 times, most recently
from
e6b2394 to
6534026
Compare
src/net/send_recv/msg.rs
Outdated
| #[cfg(target_os = "linux")] | ||
| impl<'a> MMsgHdr<'a> { | ||
| /// Constructs a new message with no destination address. | ||
| pub fn new(iov: &[IoSlice<'a>], control: &mut SendAncillaryBuffer<'_, '_, '_>) -> Self { |
There was a problem hiding this comment.
Can you comment on why control needs to be a mut reference here?
There was a problem hiding this comment.
The dumb answer is "because sendmsg requires a mut reference" (as does with_noaddr_msghdr and friends). I assumed some random API modified the cmsg on send, but I don't know which one.
|
I think I addressed all your comments. |
src/net/send_recv/msg.rs
Outdated
| iov: &[IoSlice<'a>], | ||
| control: &mut SendAncillaryBuffer<'_, '_, '_>, | ||
| ) -> Self { | ||
| with_v4_msghdr(addr, iov, control, Self::wrap) |
There was a problem hiding this comment.
Isn't this unsound? with_v4_msghdr allocates the V4 socket address on the stack and then passes that stack address in the msghdr in Self::wrap. So when we eventually call into the system call it is pointing to invalid stack memory.
There was a problem hiding this comment.
Yes, you're right. Maybe SockAddrStorage can be used for this?
There was a problem hiding this comment.
In order to avoid passing the length in here, we'll need to add a RawSockAddr type, I think, created with SockAddrV4::as_raw etc. It'll be slightly unergonomic, as the borrow checker won't let you do this:
MMsgHdr::new_with_addr(my_v4_addr.as_raw(), ...)Since that would create a temporary value which is dropped. But it should work fine otherwise.
Then the signature becomes:
pub fn new_with_addr(
addr: &'a RawSockAddr,
iov: &[IoSlice<'a>],
control: &mut SendAncillaryBuffer<'_, '_, '_>,
) -> Self {
// ...
}I think that helps with the recvmmsg(2) case, as well.
There was a problem hiding this comment.
Add a RawSocketAddr type in my latest push (it's a separate commit, so can be reviewed independently).
colinmarc
force-pushed
the
sendmmsg
branch
3 times, most recently
from
afe3c24 to
695e084
Compare
It wraps SockAddrStorage, but includes a length, for when a sockaddr has to be stored for a longer lifetime.
https://man7.org/linux/man-pages/man2/sendmmsg.2.html Partially addresses bytecodealliance#1156. Signed-off-by: Colin Marc <hi@colinmarc.com>
|
Hi, anything needed from me on this? |
| @@ -107,6 +126,35 @@ impl SocketAddrAny { | |||
| } | |||
| } | |||
|
|
|||
| /// A raw sockaddr and its length. | |||
| #[repr(C)] | |||
There was a problem hiding this comment.
Is this repr(C) needed? Is there anywhere we pass the full RawSocketAddr to an ffi call?
| @@ -781,6 +826,58 @@ pub fn sendmsg_any( | |||
| } | |||
| } | |||
|
|
|||
| /// `sendmsg(msghdr)`—Sends a message on a socket to a specific address. | |||
There was a problem hiding this comment.
| /// `sendmsg(msghdr)`—Sends a message on a socket to a specific address. | |
| /// `sendmsg(msghdr)`—Sends a message on a socket to a specific raw socket address. |
| target_os = "redox", | ||
| target_os = "vita", | ||
| target_os = "wasi" | ||
| )))] |
There was a problem hiding this comment.
This cfg attribute is unneeded because we're in the linux_raw backend here.
| control: &mut SendAncillaryBuffer<'_, '_, '_>, | ||
| flags: SendFlags, | ||
| ) -> io::Result<usize> { | ||
| match addr { |
There was a problem hiding this comment.
Instead of matching addr here, would it work to have the backend sendmsg_raw and with_raw_msghdr take an Option<&RawSocketAddr> as well, and then in with_raw_msghdr convert the Option into a pointer for msg_name using crate::utils::option_as_ptr?
https://man7.org/linux/man-pages/man2/sendmmsg.2.html
Partially addresses #1156. I would've liked to add
recvmmsgin the same PR, but it's actually much more complicated.