x/sys/unix: what if recvmsg syscall returns -1 and EWOULDBLOCK

[lwintermelon]

Below are codes of Recvmsg func from x/sys/unix .

// Recvmsg receives a message from a socket using the recvmsg system call. The
// received non-control data will be written to p, and any "out of band"
// control data will be written to oob. The flags are passed to recvmsg.
//
// The results are:
//   - n is the number of non-control data bytes read into p
//   - oobn is the number of control data bytes read into oob; this may be interpreted using [ParseSocketControlMessage]
//   - recvflags is flags returned by recvmsg
//   - from is the address of the sender
//
// If the underlying socket type is not SOCK_DGRAM, a received message
// containing oob data and a single '\0' of non-control data is treated as if
// the message contained only control data, i.e. n will be zero on return.
func Recvmsg(fd int, p, oob []byte, flags int) (n, oobn int, recvflags int, from Sockaddr, err error) {
	var iov [1]Iovec
	if len(p) > 0 {
		iov[0].Base = &p[0]
		iov[0].SetLen(len(p))
	}
	var rsa RawSockaddrAny
	n, oobn, recvflags, err = recvmsgRaw(fd, iov[:], oob, flags, &rsa) //// here recvmsgRaw may return n=-1 and err=EWOULDBLOCK 
	// source address is only specified if the socket is unconnected
	if rsa.Addr.Family != AF_UNSPEC {
		from, err = anyToSockaddr(fd, &rsa) //// later err may be overrided  to nil
	}
	return  //// In this case, it returns n=-1 but err=nil !
}

That is possible, right? In this case, unix.Recvmsg returns n=-1 but err=nil !

[gabyhelp]

Related Issues

• x/sys/unix: syscall_unix.go - Recvmsg reports empty/wrong error value #58898 (closed)

_{(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)}

[mknyszek]

CC @golang/runtime

[prattmic]

Is it possible for the kernel to write the sender into rsa and return an error?

On Linux, the core of recvmsg is https://elixir.bootlin.com/linux/v6.18.3/source/net/socket.c#L2824. The sender address is written in the if (uaddr != NULL) { block (L2856). This occurs only after sock_recvmsg succeeds, which I believe is where any EWOULDBLOCK would come from.

That said, RecvmsgBuffers takes the safer route of only attempting to set from if err != nil. I think it makes sense to do the same in Recvmsg. Feel free to send a CL.

[lwintermelon]

Is it possible for the kernel to write the sender into rsa and return an error?

For linux kernel, I believe it's possible, https://elixir.bootlin.com/linux/v6.18.3/source/net/socket.c#L2866 and L2874 are both possible to return an err after sender address is written.

For POSIX Programmer's Manual, it's undefined. The spec likely leaves these implementation details undefined.

I will prepare a CL to align Recvmsg with the safer logic used in RecvmsgBuffers.

[lwintermelon]

I have sent a CL https://go-review.googlesource.com/c/sys/+/734740